National Academies Press: OpenBook

Criminal Careers and "Career Criminals,": Volume I (1986)

Chapter: Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness

« Previous: Appendix A: Participation in Criminal Careers
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 292
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 293
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 294
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 295
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 296
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 297
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 298
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 299
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 300
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 301
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 302
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 303
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 304
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 305
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 306
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 307
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 308
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 309
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 310
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 311
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 312
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 313
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 314
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 315
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 316
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 317
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 318
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 319
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 320
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 321
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 322
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 323
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 324
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 325
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 326
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 327
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 328
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 329
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 330
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 331
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 332
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 333
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 334
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 335
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 336
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 337
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 338
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 339
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 340
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 341
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 342
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 343
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 344
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 345
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 346
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 347
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 348
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 349
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 350
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 351
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 352
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 353
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 354
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 355
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 356
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 357
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 358
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 359
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 360
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 361
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 362
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 363
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 364
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 365
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 366
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 367
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 368
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 369
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 370
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 371
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 372
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 373
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 374
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 375
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 376
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 377
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 378
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 379
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 380
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 381
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 382
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 383
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 384
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 385
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 386
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 387
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 388
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 389
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 390
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 391
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 392
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 393
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 394
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 395
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 396
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 397
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 398
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 399
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 400
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 401
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 402
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 403
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 404
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 405
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 406
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 407
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 408
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 409
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 410
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 411
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 412
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 413
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 414
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 415
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 416
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 417
Suggested Citation:"Appendix B: Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume I. Washington, DC: The National Academies Press. doi: 10.17226/922.
×
Page 418

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Appendix B Research on Criminal Careers: Individual Frequency Rates and Offense Seriousness Jacqueline Cohen INTRODUCTION The level of crime experienced in a soci- ety varies with both the participation by individuals (b or d) in that society and the frequency of offending by active offenders (A). Increases in crime may be clue to in- creases in either the participation rate or the frequency of offending. Distinguishing among the different dimensions of criminal career's has implications both for our under- stancling ofthe factors contributing to crime and for efforts to control crime. The characterization of criminal careers invoked here assumes that offending is not pervasive throughout a population, but rather is generally restricted to a subset of individuals who are actively committing crimes cluring some period oftime. It is also assumed that the constituents of the subset of active offenders vary with time as some inclividuals become criminally active (onset of careers) and others terminate their crim- inal activity. Uncler this characterization, The author would like to thank Arnold Barnett, Alfred Blumstein, David Farrington, and Jeffrey Roth for their helpful comments on an earlier ver- sion of this paper. 292 the defining attribute of offenders is com- mission of at least one crime. Participation, the subject of Appendix A, refers to the size of the criminally active offender subset cluring some observation pe- riod. This subset of active offenders includes both new offenders (first offense occurs dur- ing the observation period) and persisting offenders (criminal activity began in an ear- lier period and continues into the current observation period). Participation rates dur- ing any observation period will thus depend on the number of individuals who become offenders and how long they remain crimi- nally active. The longer criminal careers are, the greater will be the contribution of persist- ers to participation in any observation period. The subset of active offenders in any observation period is distinguished by hav- ing a positive frequency of committing crimes (e.g., five crimes per year per active offender). Beyond the requirement of at least one offense for active offenders, fre- quency rates may vary substantially across active offenders, with some offenders hav- ing very high rates and others low rates of offending. Frequencies may also vary over time for an individual. Individual offenders who have the highest frequencies will con- tribute most to total crimes.

APPENDIX B: RESEARCH ON CRIMINAL CAREERS Many different offense types may contrib- ute to an individual's frequency. Individual offenders, for example, may vary in the scope of their offending, from "specialists" (who engage predominantly in only one type of offense or one group of closely related offenses) to "generalists" (who en- gage in a wide variety of offense types). The degree of specialization may also vary across offense types; some offense types may be committed exclusively by special- ists, while others are routinely committed as part of an offender's varied mixture of offense types. The mix of offenses commit- ted by any offender may also vary as of- fending continues individual offenders may become either more or less special- ized, or increase or decrease the serious- ness of their offending. If there are consis- tent patterns of change in the mix of offenses, then commission of serious of- fenses may be characteristic of certain peri- ods during criminal careers (e.g., later ca- reers may be periods of more serious criminal activity). The various aspects of individual crimi- nality participation, career length, fre- quency, and crime mix will affect the con- tribution of individual offenders to the total volume of crime experienced at any time. Offending may be widespread, with many offenders each committing crimes at rela- tively low rates; in this event, individual offenders contribute very little to the total volume of crime. Altematively, individual 293 same individuals commit crimes over longer periods of time, and these persisters are major contributors to total crime. This appendix provides a critical review of the emerging body of research that em- pirically characterizes various dimensions of individual offending. Because of its scope and volume, the full range of the literature is beyond the reach of a single paper. Nar- rowing the focus of this review builds on a natural partition of the various dimensions of criminal careers. Participation delimits the subset of active offenders in a popula- tion; this dimension of criminal careers is addressed in Appendix A. This appendix focuses on the progress, or course, of indi- vidual offending during criminal careers, as measured by frequency rates and offense seriousness. Frequency rates are addressed first, fol- lowed by offense seriousness. In reviewing the research findings, special attention is given to their validity in light of various methodological concerns. In many in- stances, frequencies or offense seriousness are not addressed directly in the reported results, and whenever possible, available data have been reanalyzed in order to pre- sent results on frequency rates and offense seriousness in comparable terms. INDIVIDUAL OFFENDING FREQUENCIES FOR ACTIVE OFFENDERS , , frequencies may be high and participationIndividual offending frequencies, A, are a low; individual offenders would then befundamental feature of individual criminal responsible for a larger portion of total crimes. Career lengths may be short or lone. If careers are characteristically short, then there is likely to be a large turnover of active offenders as individuals quickly ter minate careers and new individuals be come criminally active. In this event, new offenders would be major contributors to crime. Also, with short careers, current par ticipation levels may be relatively low, while cumulative participation (all individ uals who were ever criminally active) is more widespread in the population. If crim inal careers are characteristically long, the careers. Despite the importance of A in estimating the magnitude of offending dur- ing~criminal careers, research that statisti- cally characterizes the intensity of offend- ing for large numbers of ordinary offenders is relatively recent. Much of the early re- search on individual criminal careers con- sisted of biographical or autobiographical studies. While such case studies provided interesting and often insightful reports on the individuals studied, there was little in ~Some of the classics among these studies are Booth (1929), Shaw (1930, 1931), Sutherland (1937), and Martin (19521.

294 dication that the individuals were represen- ~tive of a larger group of offenders. Indeed, We subjects were more likely chosen for Weir fascinating uniqueness than for Weir representativeness. More recently, a large body of research has examined the attributes of large sam- ples of offenders. This research includes both studies of self-reported delinquency and studies using official records, such as arrest histones.2 Because this research has been largely motivated by interest in the causes and prevention of crime, it has fo- cused on identifying the correlates social, economic, psychological, and o~erwise- of offending. This research has typically developed estimates of participation (i.e., We prevalence of offenders) or of continued offending in different population subgroups. Estimates of the intensity of offending by identified offenders, A, are rarely provided. A related body of literature attempts to de- velop topologies of offenders win similar social or psychological at~ibutes.3 2The self-report literature is extensive and in- cludes over 100 studies. A partial bibliography is available in the review of the National Council on Crime and Delinquency (1970~. A critical review of much of this research is found in Reiss (1973) and Hindelang, Hirschi, and Weis (19791. The following represent only a small sample of the available re- search in this area: Reiss and Rhodes (1959), Hirschi (1969), Gold (1970), Waldo and Chiricos (1972), Williams and Gold (1972), Elliott and Voss (1974), Elliott and Ageton (1980), Hindelang, Hirschi, and Weis (1981), Elliott et al. (1983~. A recent review of participation measures, including those based on self-reports, is available in Visher and Roth (Appen- dix A). Analyses of official records typically involve lon- gitudinal analysis of large samples of criminal rec- ords. Among such studies are Glueck and Glueck (1937, 1940), McCord and McCord (1959), Robins (1966), Wolfgang, Figlio, and Sellin (1972), West and Farrington (1973, 1977), Robins, West, and Heganic (1975), Robins and Wish (1977), McCord (1978), Farrington and West (1981), Hindelang, Hirschi, and Weis (1981), Famngton (1983b, 1984~. 3See Warren (1971) and Gibbon,s (1975) for re- views of the topology literature. Examples of typol- ogy research are found in Kinch (1962), Gibbons (1965), Hurwitz (1965), Roebuck and Quinney (1967), and Davies (1969~. CRIMINAL CAREERS AND CAREER CRIMINALS Recent interest in the crime control ef- fects of incapacitation has underscored the importance of developing estimates of A. Recognizing the impact of variability in A on estimates of incapacitative effects, the National Research Council Panel on Deter- rent and Incapacitative Effects (Blumstein, Cohen, and Nagin, 1978:80) made the fol- lowing recommendation: Empirical investigation should also be directed at estimating the parameters measuring the level of individual criminal activity, especially the indivicI~al grime rates ... and career lengths.... Furthermore since estimates of the incapacita- tive effect are sensitive to variations in these parameters, these estimates should not be re- stricted to highly aggregated population aver- ages. They should be disaggregated by crime type and demographic group and should reflect the statistical distribution of the parameters. Recent studies in two research pro- grams~ne at the Rand Corporation and the other at Carnegie-Mellon University have begun to provide explicit, disaggre- gated estimates of A. That research is re- viewed in this section, in particular the very different approaches used and the resulting estimates of A. A number of other studies provide estimates of participation rates and aggregate incidence rates for a study popu- lation. These data provide a basis for devel- oping estimates of A for the studied popula- tions. The results of these new analyses are also reported below. Throughout this review of estimates of A, various methodological issues in the mea- surement of A are discussed and suggestions are made for further research in this area. The section begins with a discussion of the distinction between A, the main interest here, and more commonly available esti- mates of aggregate incidence rates. Distinguishing Individual Frequency Rates from Aggregate Incidence Rates Individual frequency rates, A, apply only to active offenders. This restriction distin- guishes A from the more commonly avail- able measure of aggregate incidence rates, which reflect the frequency of offenses, or arrests, in the general population. Aggre

APPENDIX B.: RESEARCH ON CRIMINAL CA0ERS gate incidence rates are exemplified by the annual crime rates and arrest rates reported by the Federal Bureau of Investigation. The key feature distinguishing A from aggregate incidence rates is the population base on which the estimates are calculated. In calculating A, only individuals with at least one offense, or arrest, are included in the population base. Estimates of A thus reflect the average frequency of offending for individuals who are actively committing crimes. Aggregate incidence rates, by con- trast, apply to a total population. The popu- lation at risk includes offenders and nonof- fenders alike. Aggregate incidence rates reflect the com- bined contribution of participation rates for offenders in a population, ~ or b, and individ- ual frequency rates, A or A, for active of- fenders. Consider, for example, estimates of aggregate arrest rates for some population i: Aggregate = Number of arrests of persons arrest in population i rate for population i Number of persons in population i This aggregate measure can be partitioned between the participation rate for offenders (~) and the frequency rate for those of- fenders (,u): Aggregate Number of persons arrest arrested in population i rate for Number of persons in population i population i Number of arrests of persons in population i Number of persons arrested in population i = Participation rated x Frequency rated (1) The conceptual distinction between par- ticipation rates and individual frequency rates has important implications for the evaluation of incapacitative effects. The crime control potential of incapacitation hinges on the magnitude of an individual's offending frequency, A. This is the expected 295 number of crimes averted by incapacitating an offender. Aggregate incidence rates in- clude rates of zero for nonoffenders, who are not vulnerable to incarceration, except in the rare cases of wrongful conviction. Aggregate incidence rates, therefore, would seriously underestimate the crime reduc- tion achieved by incapacitation. Likewise, the impact of incapacitation policies on prison populations depends on the partici- pation rates of offenders in a population. The more widespread that offenders are in a population, the greater will be the potential increases in prison populations as a result of incapacitation strategies. To the extent that A exceeds one offense per offender, aggre- gate incidence rates will overstate the prev- alence of offenders. In this event, use of aggregate incidence rates in place of partic- ipation rates would lead to overestimates of the potential impact of increased incapaci- tation on prison populations. Accurate esti- mates of the tradeoffs between increases in prison population and reductions in crime through alternative incapacitation policies depend critically on having separate esti- mates of participation and frequency rates. The partition of aggregate incidence rates into participation rates, on the one hand, and individual frequency rates for active offenders, on the other, may also be useful in evaluating the effectiveness of other crime conko1 policies. To date, evaluations of deterrent and rehabilitative effects have relied almost exclusively on aggregate out- come measures.4 To the extent that partici 4Recidivism rates are a special variant of aggre- ~ate incidence rates. While restricted to a popula- tion of identified offenders, the prospective per- foll~ance of this population is the combined result of the level of continued participation by active offenders and the magnitude of individual fre- quency rates for those who remain criminally active. In particular, failure to recidivate during a follow-up period may occur because some offenders end their criminal careers altogether, or because some offend- ers who do remain criminally active do so at low frequency rates. In the latter case, extending the length ofthe follow-up period increases the likelihood of observing eventual recidivism; in the former case, recidivism will never occur no matter how long the follow-up period.

296 pation rates and frequency rates are dif- ferentially affected by deterrence or reha- bilitation policies, important effects on these component parts may be obscured in the aggregate measures. Analyses of effects on the partitioned measures may provide valuable insights for improving the crime control effectiveness of deterrence and re- habilitation policies. It may be, for example, that different strategies will be more effec- tive if they are targeted on selected popula- tion subgroups. Review of Estimates of Individual Frequency Rates The main interest in this subsection is empirically based estimates of A. Relatively few studies provide explicit estimates of A, and they are limited to samples of serious adult offenders. Three such studies are re- viewed in this section. Indirect estimates of A, derived expressly for the panel from other published data on current participa- tion rates and aggregate incidence rates, are also reviewed. The indirect estimates have been developed for a wider variety of study populations. The studies that provide explicit esti- mates of A are summarized in Table 1. Separate rates are generally estimated for individual offense types, and total rates are presented for larger offense classes. Of- fense-specific frequencies reflect the aver- age number of offenses committed when that offender is active in that offense type. Active offenders are distinguished by hav- ing at least one offense (or one arrest) for a crime type. Individuals with no offenses of a particular type during the observation period are excluded in the computation of rates for that offense type. By this criterion, the earliest self-report survey, the Rand survey of 49 prison in- mates (Petersilia, Greenwood, and Lavin, 1977), is properly excluded from consider- ation. The offending rates reported in that study apply to the total sample of 49 of- fenders; rates of zero for inmates who re- ported no offenses of a given type are in- cluded in the reported rates. As the di- rect precursor of the two later Rand inmate CRIMINAL CAREERS AND CAREER CRIMINALS surveys, however, the study is included here. All of the studies summarized in Table 1 are based on samples of adult offenders. The samples are also generally restricted to more serious, or more criminally active, subsets of adult offenders. The three self- report surveys, all by the Rand Corporation, are based on surveys of inmates serving sentences in state prisons and, in one study, inmates in local jails. These inmate samples are thus restricted to offenders whose cur- rent offense or prior criminal record was serious enough to have warranted a sen- tence of incarceration. The two studies using official arrest rec- ords are based on samples of adult ar- restees. While potentially including a broader range of offenders, these studies also focus on offending by a more serious subset of offenders. To enter the sample, an offender must have had at least one arrest for a serious index offense (murder, rape, robbery, aggravated assault, burglary, or auto theft) during the sampling period. This criterion excluded offenders who engaged exclusively in minor offenses or who were never arrested for a serious index offense. By focusing on subsets of offenders who are active in serious offense types, studies of frequency rates can develop estimates of A in those serious offense types. Because they are generally quite rare, these more serious offense types have often been excluded from surveys of general population samples. The studies that provide estimates of A for serious offenders are in direct contrast to the much larger body of research on participa- tion in offending, which is typically based on self- or official reports of offending and deviance for juveniles sampled from a gen- eral population, and which is therefore dominated by the more common minor of- fenses, such as vandalism or simple assault. This difference in design reflects the dif- ferent focus of the studies, in the first in- stance, the intensity of serious offending by more continuously active offenders, and in the second, the scope of deviance found in a broad population. In addition to focusing on active offend- ers, the studies in Table 1 also restrict the

297 ~n ~0 . ~ 4' ~ c ~) ~ e ~=a ~ ~ .- s4, ~ C ~ 4J . - 01 C ~ · - ~ ~C q~ - - 0 0 JJ JJ Ll .- ~ C S JJ ~ ~0 0 ~ ~ O ~ 1 .- O ~ ~ .- - - ~ C ~ ~ s ~ ~ ~ ~ s s e 1 ~ ~ ~ 0 - - 0 ~ ~ ~ ~ ~ ~ - - ~ 0 ~ ~ ~ c arl 3 c 4J c ~ al ~ ~a ~ ~ . ~ ~4 ~ ~ c 4) ' '8 q4 a,~ ~ ~ Q m0 ~ o ~ ~s O ~O ~ O O ~ ~, C Y q~l 1 Q' ~ ~S .- V ~ · ~ ~ O ~ ~ ~ O 4, ~ ~0 0 ~ q ~ 1 0 ~ ~ O.^ O O eq C) 0 ~ -~ S "r, ~ ~ S O 14 C) Ll P4 4) o Y O 0 4) ~ ~ C ~ a~ C) C) tn %4 ~ ~ 0 q~ 0 ~0 c 0 q~ c ~ o ~ ~/ ~ ~3 a, C :- o s c ~ ~ o ~ :, ~ c t' 0 ~ · ~ o c' 0~ 4) q~ ~ ~ O ~54 ~ 0 0 ~ ~s: ~ ~ ~ ~ ·- ~ ~a ~ ~ ~ ~ ~ c C)~ X ~ ~ ~ ~ s ~ ~ ~ ~ ~ ~ ~ ~ ~ o ~ ~ O ~4, o ~ rn C ~ O ~ ~o ~ ~ C ~ o ~ ~ ~ ~ 0 ~ ~ s ~ O ~ C ~ ~ 4~5 - 0 ~ ~ 0 ~ ~S ~ ~ ~ 3 ~ ~ ~ ~ O o ~ O ~ ~ O o ~ O ~ C~ ~ JJ ~ ~ >1 ~ ~ ~ ~ 0 0 ~ C ~r ~ ~ ~ ~ ~ C ~ ~n . 0 C O O ~ O · ~ 4' ~ O.Q C) 0 ~ 0 0 ~ 0 ~ ~ ~ a' a, O ~ ~ 0 ~ Q ~ ~ ~ C ~O ~ ~ C ~ ~ mO ~ ~ mC ~ 1 ~O ~ ~ O ~ q ~ O P4 dP ~ O ~ .- 0 0 - - ~ O ~ O ~ O ~ ~ ~ 0 0 ~ ~ -- .- C7, ~ ~ ~ ~ ~ ~ ~ .- ~ ~ ~ O ~ .- 0 C ~ ~ O O ~: 4: ~ 0 ~ 0 ~ 0 ~ - ~ ~ 0 ~ ~ ~ 0 a a ~ ~ om oq~ o~ ~ c ~0 0 D O ~ 0 ~ o. ~ ~a, eJ ~ ~ ~ 0 4, ~ ~ C ~ 0 ~ ~ ~0 0 c ~ ~.- 0 ~ ~ ~ a' ~c cn 18 m-- ~ ~ ~ ~ ~ O X ~ r~ a: ~ ~ ~0 ~ ~ ~ ~ ~ ~c ~ ~ ~ ~ ·- _ ~ 'O ' ~ C) s~ Ql ~ ~ ~ ~ U] ~ JJ ~ JJ ~ ~ ~ O ~ ~ ~ 0 0-- ~ ~ ~ 0 C ·,1 c, ~ O ~ ~ ~ ~ O ~ ~ ~ ~ ~ ~ ·- c _ c q3 s a~ ~ 4~ ~ c~ ~ ~ C C) ~ ~0 Q 4 ~C ~ 0 · - )~ (d )4 tU 4) C C ~1 :C q3 o ~ ~ C C ~ ~ ~ 0 ~ ~ ~ ~ C ~ O tr 1 C ~ ~ a~ aJ ~ ~ _d 0 ~ ::~ q~ ~ ~ ~ c ~) ~C ~ ~ q~ P~ 0 ~ ~ ~ ~ S ~ q~ ~3 · q~ q~ O ~1 ~I Y C ~ ~ CD · ~ ~ JJ ~ 0 0 ~ ~ 1 · - q~ 0 0 0 C 4~ ~ 0 0 ~ O :^ Ll ~ C O ~ ~ O O ~ 04 0 (0 80 ~ O JJ D ~l ~ 4) a, C a O ~ o' ~ ~ 0 JJ a, 0 ~ s C C C 4, 4~ ~ C ~ ~ ~ 0 q4 - - JJ q3 ~ ~ O ~ ~ O . - O 0 ~ q~ ~ ~ ~ ~ 14 c c ~. ~ ~ c 0 ~ ~ 0 q~ a, ~ 0 0 ~ ~ c ~ ~ E3 0 ~ c ~ ~ 0 ~ s ~ q~ 4J ~ ~ c s ~ ~a ~ ~ q~ ~ ~ q3 a, 0 v 4 ~ 0 ~ s 4) ~ 0 s ~ ~ Ql JJ P. S Q w4 a~ ~ 4~ 0 JJ a 0 ~ E! ~ 0 C ~E~ C.) ~ q~ C ~ ~ v P4O ~ ~ C) O q~ ~ :1 ~ ~ g O O q ~C .- q~ O ~ O ~ a' c s ~ s o [Q 0 a' ~ ~ ~ 0 E! q~ O ~ ~ O ~ 0 ~ :. ~ '- ~ ~ :e (1) .- 1 a, ~a' _ ~ ~ ~0 q ~ 1 4~Q ~dP ~m ·v 4, c ~ ~ ~c ~ ~0 c c oo ~0 4, c c a, O L. ~ ~0 1 a, ·a ~ c ·- 0 ~a, ~ ~ 0 0 ~ ~ ~ 0 . ~ ~ ~ c ~c s~ 0 ~ O ~ ~ m >~ ~n c ~ c) s 0 ~ ~ -4 0 ~ :> ~ O E: o a~ ~ ~ 4J 4~ F ~ Y .- ~ .- ~ ~r, u ~ ~ ~ ~ ~ o. 0 ~ ~ ~ JJ _ ~ ~ ~ ~ ~ 0 0 ~ O. ~ ~ 0 s 0 ~ ~ ~ ~ ~ ~ o c 3 ~a~ ~ ~ ~ ~ ~ ~ E~ ~ ~ Q ~ O ~ ~ a~ Q ~ S C C,) O ~S ~ C ·- 0 0 ~ ~ ·rl C U] ~ a' _ ~ O ~ ~ ~ ~ s O a, ~ ~ 4) c 3 ~ ~ ~ ~ 0 ~ a, _ 0 ~ ~ ~ ~ · V ~ ~ ~ ~ . ~ ~ oP a : ~ ~ 0 ~ ~ c q~ c q~ ~ 0 q~ :~, O ~0 :~ C ~ q~ C C S ~ O O q~ ~ C q~ Ll -= ~: ~ ~ ~ Q. 0 0 ·- 3 0 0 0 ~ ~ ~ q~ a s~ ~4J ~ e ~ ~ ~ ~ ~ ~ ~ ~ ., :^ 0 0 ~ ~ 4J _ ~ ~ 0 o U~ ~ ~ ~ tV r~: dJ q~ tD ~ 0 D ~ ~ ~'t S . D ;;. S a" ~ ~ ~ ~ ~ ~ ~ ~ ~ O ~ ~ ~ O ~ ~ ~ 0 GS 0 0 ~ 0 ~ . - ~ O ~ ~ ~ ~ . - O ~ ~ D _ ~·- ~J 4~ 0 ~ X ·- O O O ~ O ~ ~ ~ ~ 0 ~ O _' ~P~ q~ ~ ~ ~ C W ~ Ch C) ~ -' ~ C ~ ~ ~ O ~ C t15 o ~ ~ ~ ~ ~ o ~ ~ ~ ~ ~ ~ 0 ~ c ~ ~ s ~ o: ~Z O ~ 3 s~ ~s~ ~ ~ ~ s ~ ~ 0 ~ ~- ~o:1 "1 4, ~x Y ~ ~ 0 s -,' ~a,~ ~ ~a, U}0 L. ~ ~a s - c JJ a, o H '4: 3 4~ ~·. ~ 0 O ~ ~a e ~ ~ o.~- _ ~ e ~ ~ ~JJ :, q~ 4 L E ~ .- ~ 4 ~u~ - C) ~ O ~4, ~ c~ ~ P. · c a ~e 0 ~ e.- oc ~ e m~- ·- co m eo 43·- · . · JJ 0) c O ~ ~ 4, ·., S~ ~C q~ ~s O ~ 0 Ll ~ qo4 ~ Ll ~o ~ ~ C ~ ~ ~ a' 4, ~ 0 ~ CJ .^ C ~ ~ ~ ~ 0 id ~ ~ ~ V~ 0 .- ~ 0 ~ ~C ~ ~ 0 ~ C) E~ ~ q~ 0 0 ~ ~ ~ ~O ~ C) a~ _I 1 a ~0 0 1 C ~o mm o C) ~- ^ ~ ro ~ ~ ~ ~ 0 ~ ~ 0 ~ ~ :>. .~1 O ~ 0 ~ ~ '- ~ ~ O ~ ~ ~-O ~0, o Q ~ ~ ~ ~ co ~O ~ ~a C c .- s~ 0 ~ ~ . q ~-q~ -~ q~ c) O .,' 4) r"a, ~.rl ~ ~ C S~ Ll ~ ~ ~ ~ C C a~ ~ ~ a~ 0 c~ a~ ~ c c o~ a~ ~rl ~ ·~ F ~ ~ ~ ' ~ ~ ~ ~ - ~ S 0 . - O ~ O ~ 0 ~ ~ S ~ C _1 ~ ~ ~ 0 ~ ~ W id ~ 4J ta c ~ 0 ~ ra 4J a' C v EG "o _ ~ ~ O 0 ~ ~ ~ ~ c c ~ ~ ~ .- ~ . ~ ~ 0 C, c c ~ Q~, 0 .- c 0 ~l a~ c ·- ~ a~ ~ ~ ~ w ~ ~ a ~1 ~r' ~ ~ 41 ~ O :' ~ ~ ·- C ~ n: · - ~ 4, ~ ·a O :> ~n Ll m u ~ v, ~ a) ~ ~ 0 t) s 4) C _ ~O 1 C ·- = - c~ ~ O C ~ ~ ~ ~ ~ ~ C 0 ~ w 4J ~ ~ ~ · ~ 0 ~ c ~ · 0 ~ ~ ~ . 0 c _ 00 JJ , - W C L. C ~ ~ ~ · - ~ Q ~ ~ _ ~ C ~ ~ O ~ ~ ~ "a ~4' 4, W 0 ~ 0 =- - w E . - C ao c a.' ·~- - ~ E a, O c ~2: cn O ~ C) ~·-q~ ~- - - w~-l ~ rl ~W ~ t~~ - 16 ~ ~' ~ C ~O =0 C o C ~5 0 1 ~O O ~ '- ~ JJ.- 10.' U] ~C ~ ~ · - C E ~ ~ ~ C ~q~ 0 0 C ~ S C-- 4, 0 ~ C 0 0 ~ ~ ·- 0-- $4 · - O ~ 0 e `' c 4, ·- ~ ~ 0 3-- ~ Q 1 ~O ~ £ ~ Q' a, m P4 .~ ~ t-° ~ t, ·-C O ~ ~ Q~ ~ E c ~I e ~ 0~ O U ~ ~ O ~ 3 ·- m ~4, 0 ~ .a o. - ~ ~ ~ 0 E 3 $:1: ~o c ~ ~ ~ ~ c ~ ~ ~ ~ ~ -. ~Ll ~ C ~ ~ ~ O O a, a' ~ 0 C" U] P C) ~:1: ·- E ~ ~ O ~ ~ 4~ ~ ~

298 m E~ C) o 4, 4~ Pu a, D: ~n cu ~a Cc, a, 4) c .,' c cu q~ -0 q' a, Ul cn Y c' a 4~ .,, Q `0 .,, JJ a) ·,' E~ -0 ~n O h4 O 0 .,. s~ cu ,O-c v O tn ~Q ~O .~1 8 (U CR C ~a U] c IL ~n ~o C cu ~a cu t8 c ~ c) f: · - O , :~ ~n 0 :~ ~n Q) a' t) c, ~: a ~ ~ ~ ~U .4 m C ~ 4, C 0 .,, ~U ~n ~a ~u cu 1 cu tn 0 ~ 4~ ~a 3 1 c c ~n mO s c oo ·~ ~ 0 ~ Q ~ O c cu ~n 0 ~ q~ ~ ~n ~ c q~ ~, a~ ~ O -= Q c 0 ~ ~a O ~ ~ ~ q~ ~ ' ES >' ~ 0 a, C ~U 0 1 (U C . - q' a a 4~ x ut ~ 1 o q~ m ~U C u "a n ~u : u c cu ~n cu c 4, q~ _ O . m C 4J C: CU ~a mm ~n ~ q) ~ c P. Q C JJ :~ ~ V" CU . - JJ CD S CD C b3 CD O 0 0 ~n.- Y. ~ C, s ~ ~n CD C) CD ~ W ~ `0 CU q~ ~ OS~ CD ~l 3 CU ~ Q CU CU C C) .,' ~, S3 CD 0e 0 S 1 C O ~ C CU O ~ o Q £ ~ ~s cu ~ ~ u, C Y C CD '< 0 ~ ~n ~ cu s ~ ~ ~o ~ ~ · - c c~ ~ ~ w . - CD ~ O CD · - CD V £ 3 CD C ~n ~n s I ~n c. 1 H C CU CU ~n CD CD ~, Ul O C 3 · e RS cu ~n 0 C) c w >, CD C} --l ~ W ~n ~ O 0 ~ cu c c CU ~ 1 C I CD U] CU O C C) CU ~n ~x C CD CD CU S CU tU ~o- ~ a CD q~ CU CU .,1 0 ~ 3 ul ~ C C C CU CD _I . - : ~CU l~ :~ S q~ ~ __ ~n ~ 4J · O c ,' c.) c) ~ ~ ~ ~ -~ · - CD ~I C C C' S CD-I CU CD W Ll CD CU ~ ~ C) JJ ~ CU 04 q, ~ ·^ 115 ~ JJ 0 3 C ~ 'S ~ O X ·- ~n 3 CU ~ ' C S ~ . - ~ ~ CD ~ O CU J~ JJ U) k ~ 54 ~ 1 0 · - td CD O >1 0 o~ 3 ~ C) ~ q'l CU Sv 4J o ~ e ~o s C 43 oa tcou CD ". ~ ~n :~ 54 ~ CD CU CD C 0 S CD C Cp W CD _. W W S O W o CU S 1 V Q. 4J JJ CU C (D~ CU 05 C £ ~ ~u, .,' . -O C ~n ~n ~ . ~ c,) cu CD CU~n cu ~ JJ ~ ~ ~ ~ ~ · - :>, c CU 115 0 C O CU ~ CD Q cu ,£, s q~ £ q~ ~C ~ == ~ ' O ~ CU CU ~n . u' a' ~ c q~ ~n u, ~ CD ~ -I C lU t14 -I CU q~ CD O ~ - O ~ ~ ~ O ~ CD q~ 1d ' :~ :~ C .^ 4~ ·a ~ ~n JJ ~5 JJ O ~ CD ¢' 0 ~ CD (U CU CU ~ ~ ~ CD CD C ~ ~ In ~ P4 ~ -- J~ _' 0 4J ~ C 1 CD ~ ~ ~ (n ~n c CD I ~ CDU) O e e ~ -~m CD ~ C ~ ~ ~U~ C~ y. ~- ~n ~ 0 0 CD ~n tt CD Ul S ~I CD £ O _ c c . - 1 O CD ~J~ 43 CD ~ ~0 Ll. - ~ CU O C V 43 CD C CD 0 s ,a ~ 3 C CU CUi S Ll O O ~ ~ ~.. ~ CD ~. CU cu ~ e Cu e _ ~_ .. C m q' U c) c' ~ ~ q4 CU ~ :^ q4 CU 0 C JJ 0 ~ ~ ~ ~_ t1JJ CD ~C t:J ~ Q Oo4 S a a cn y' O '~, 1 1 · ~n ~ O 4> cu (D ~ Ll e C cu o ,' ~ ~n t: ~ JJ c u ~ cu · - q~ O ~ ~n C ~ O ~ CD 16 0 Ch U~ C C~ ~ . - CD ~ J~ CD 4~ ~ O ~ ~ ~ ~ ' - C O O CU ~ CD ~Y. ~ ~ ~ CD CD O O q.l £ u 0 CD Q o C~ ~U e 3 ~D Q ~n CD ~ C :D Q 114-I CD 14 0 fa C CU _~ CD U ·- JJ Q S :>' ~ N ~d 1 _I O C ~ Ll 1 . - C CD ~J O O ~ ~a C ~ q~ CD J) CD ~n ~ ~ ~ u, . 4) ~ £ ~4 CD C~ 3 "0 0 CD · ~ c ~ q~l ~ s c CU C U C ~l 4~ CD q~ ~ CD CU O ~ C q~ S S a~ _I O ~ O C) ~ 4~ ~ ~ O 1 4J 1 -I 0 CU ~I CDCD -~ O cu cu £ s ~a1 S ~ ~ U) >, ·- ~ Ul C CD ~ ~ Q 4 ~cO td 4 ~O )-I ~ ~ O Ul O ~ ~ O OI CD ~ (J C CD -1 0 O ~ ~ . - CD C CU - `, ~ C CD -~ ~ ~ S ~ O C CD CD ,1 q3 C I nS CD UY_145E~- CD me ~ 3 CD CCU Ll~ eCU~ CU X Ul CU CD > ns ~ - "l ~ ~ O q~ ~ ~ ~ :> O JJ CD '- '- ~ 0 4~ CD ~.a J~ C tn ·- c ~ ~ ~ CD ~n ~ ,~ G5 ~ >, I . ~ a, 3 on CD ~ ~ CD ~n O tn ~n ~ ~ ~ ~q - , ~ y' ,~ ~ ~ ra CD U ~ CU CD Q ~ U CD CU ~ J~ O :~' CD C '0 ~ ~ ~ C £ O U 3 ~ J~ ~ ~ U) tO CD >, CD O C ~ 10 ~ CD C ~ ~ U] S _I ~ ~CD ~ U ~ ~ ~ >' CD tt '~ ~ ~ ~ CD C CD ~ .^ ~ (n ~, - · ~ CD ' - ~ u Cu e C ~ "' cu, - ~ ~ ~D · ~, - q~ C C >, CD 3, - . - C N . - . - ~ '~ ~ ·' ~ ~ _ O ~ O O ~ C ~ ~ ~ .- ~ ~ ~a 'a C C C,a ~ C ~n C .- ~ ~ I 'o ~ C r~l C O O ·-l CD V ~ ~ S r~ ·^ dJ 4 J ~ ~ O '- CD CD a' e cn cu · - ~ c ~ . - ~ CD ~ r~ e ~ ~ Q CD ~ e e ~ 0 ~ e ~ ~ ~ ~ Q '~ S C ~ ~ ~ ~ ~ y. ~ ~ '- CD ' - ' - ~ q~ '^ rd CD CD CD ~ CU ~ ~ ~ CD CD :>1 C CD ~ S E~ ~- 0 ~ ~ ~ e ~ ~ e 0 ~u ~ u ~ ~~- ~ c e ~ CD O ~ ~: ~ ~n c u~ CD ~ ~ 1~5 S V ~ C S 4~ eq 'C ~ · U, CD ~ ~ ,.4 O 0 4) C ,~ ~ ~ .~1 ' - C O CU C 0 0 CD ~ ~ CU '^ S O eq q~ CD CD S~ ~ J~ V ~ CD I ~ ~ ~ ~ ~, - CD ,l >~ 4 CD CU ' - [.) (U o ~ ,0 ~ ,~ u tU CD ~ CU S CD CD CD ~ q~ ~I CD CD > J~ 04 :- ~ O '° £ s ~ ~ 0 u] I ·~' U .^ 4J CD ·' CD tn 0 cu ~c JJ c JJ ~ 00 cu CD CU Ll ~ q~ ~ ,~1 CD '- '0 ~ '0 CD S ~ ~ ~ S ~ ~ CD ~ ~ 0: E' ~ U) O ~ S S ~ S O 1 CU O C 'O 00 U ~ CU ~ 0 O ~ ~ ~ - CU 'O :' a, ~ ~ ~ CU 'O CD ~ ~ _ Q O ~ ~ 00 U U Q B) Se o0 cu ~ s :3 ~ 0 ~ ~ 0 CU 0 U ~ 0 ~a ~ cu _ CD ~ C CU -I ~ 'O C q, 00- U C_ C CD ~ 0 CD CU CD ~,' ~ CD ~ ~ ~ q4 _ CJ, O ~ O 0 e ~ ~ ~' £ I ~, - I ro 0 o CU CD ~ CU e ~ ~ ~n 0 CU Q C Y Y' CU 0 CD ~1 q~ 'O O C ~o , - 0 41 CD S ~l O J~ ~ CU C ' Ll .~ to C) O S C ~ 0 y, CU C C CD CU V4 _ (D 4 q~ O JJ q~ i.4 ~ ' - O CD ~n oq JJ e Q CD ~ 0 £ CD e ~ 0 0 C :, ~ £ u O c `4 ~ _ CD CD ~ I cu ~ C ~n ~ 3 0 0 ~ r~l CD eq dJ-- ~ o 0 ld CD CD 0 ~ C CD Q 0 £ Q ~ CD £ CD :D U q4 C ~ Q ~0 ~ q~ s 0 ~ ;' O u ~ £ J~ :- CD . - ~ S ~ al ~ CD ~ J) O _ ~ 0 S ~ C CU CU E" C :>, 'O O ~ - ' O U CU C 0 3 CD ~ CD 4J ~ ~ ~ · ~, - ·- O CU r~ ~ ~ '0 CD CD ~ 3 `8 CU b! CD S S 1 - 0 . - ~ O e ~JJ ·s~q4 ~ CD CD% ~ CD CD C U 0 £ c CD ^ CU CD >, I ~ ~'~ O CD S J) I ~ CU O -/ r ~O C U~ ~ r ~r~l C CD ~q4 tU CD ~ fa 0 3 ~ ~ 05 '0 Cl. 3 '0 ~C ~CO a' Q 0tU S ~ 03 C ~ C · ,~1 CD C tn ~ Q C )~ X c 4~ q ~n 11 e ~. 0 e ~ -4J ~o~ ~ O O tn ~ :> 0 CD ~n c y' 0 ~ CD CD C CD ql C O C C ~CU O CD ' C ~O ~eq CD ~, ~I JJ ·,4 C ~ . - .,4 ~ ~ ' - CD ~' - r~ L, CD O Z CD Ul CD )-I 0 ~' , - ·^ ~ 0 ~c C ~ e 0 ~ t0 JJ O ,0 0 4J ~ ,., ~ ~ ~ ~CD 04 '0 C (D Ut CU , - CD 4 ~C (D CU O ~ ' - C ~ C: . - 04 0 ~ 0 ~ C tU e CD '0 0 ~S 'O e ~ '^ O ~ ~ r~ ~ CU Q 3 C 0 3 ~ 11 ~ '0 ~ C ~ CD CD ~ £ =,- ~ ~ CD ~ C ~ CD ~ ' CD ~ CU C C 0 ·- ~ ~ qS s s q~ CD ·^ 'O ~ CD C ~ 0 U ~O ~ ~ ·- ~ (D '- ~ C ~ ~ C ~ 0 O CD CU tIJ ~ y. C ~ JJ, - : - '~ ~ ~ o Z ~ 'O O '~ .,4 CD rd ~ C ~ CD ~ Y. 'O X ~ ~ C .^ CD ~ O 'O ~ .,' U ~ it ~ ,1 .^ CD ~ ~ (D ~ .^ ~.rl F e ~ c 0 s , - 0 c CD ~, 'O C CU ~n ~ ~ ~ c CD CD . ,~1 ~ 85 t~ ~ ~i ~ S ~O ~ S U) CD ~ ~ u] 'O £ . - ~-l . - ~ c ~ ~, ~ · - ~ c 0 ~n ~ '8 4J ~ en s U CD ~J :~ 05 ~l '- O ~ ~ CD ·- ~ ~ r~ O O CD O ~ C 'O CD CD e ~ cu ~ ~,- CD CD O4 0 ~ C 54 ta Ll JJ U Q ~Ll CD c,) U · ~ U CD ~ S Gl ~ C ~ ~ C ~ O ~ ~ CD oq o 0 ~ O . JJ C (D ~,g CU ~ , ~0 ~ ~ CD ~ CD 3 0 C ~ ~ Q ~ tn ~ P~ . CD CD CD e I >, CD `u (D O -~ C ~ O? Ll ~ 0 CU ' - RS >, (D ~ CD ~ 0 U U ~ CD O CD k '~ C CD CD 0 ~ ~ {n ~ c 4J, - ~ q~ ~ CD ~ CU ~ ~ ~ ~ ~ ^ CU S L' S (D C ' - `0 ~ ' - 'O ~ ~ ~ ~ . - ~, - C ~ ~ C ~ O 0 C ~4 (u ~t ,4 ~ ~ ~ CD ~ ~ ~ ~ ~ O ~ ~ ~ O ~ ~ O CD ~ ~ C ~ ~ ' - CD S O O ~ ~ ~ (D ~ ~ CD O (D ~, - e, - ~ ~ ~ ~ U ~ CD U U ~ ~ ~ ~ Y. ~ Z ~ C ~ C ~n, - 3 ~ .^ ~ ~ CD e ~, - u ~ ~o . - ~ 3 ~ ~ C ~ a~ ~ CD CU CU O S CU tU ·t d' ~ 'q~ CD q~ ~ CD O ~ O '- CD 40 0 ·^ ~ CD CD ~ ~ ~ ·,4 C, ~ (D ~ ~ ~0 e ~ ~ ~- ~ 3 Q r~ ~ Q ~n- ~O ~O un q ~u 0 ~ ~ ~ C) 3: 3 C24 ~0 0 cn O Q ~O V ~

299 0 1 ' 4, ~C :^ V s_ ms ~ ~_ ~ ~ C 3,Q 1 C ~q~ · e m_ os ·- ~ Llo ~0 ~ ^~ - e ~ · o ~ ~ ~ ~ 0 ~ ~ O ~1 0 4, ~ ~ ~ ~.` ~ ~ ~ ~.- ~ ~ ~ C O ~ ~ ~ ~ 0 _ s 00 ' 4, C s 0 ~ 0 ~ ~ ~P. Ql ~ ~ q~ O :~ ~C.)S~q' a'== Q 4' ~ :>' e-~- c os ~ ~o~ -= ~ 0) ~ -= 4~- e ~ 0 '~ ~ -R JJ 4J 4) ~ c) ~ ~ ~ )~ s ~a -~ ~ ~ ~l Ll ~ _ ~ 14 ~ a~ ~1 C ~ C ~ ~ ~ ~ C ~ ~ ~ ~ ~ ~ ~ ~ C ~ C O C Q Ll 4, ~ O ~- ~ _ ~ ~ _ ~ ~ ~O ~ ~ ~ O 0 ~ s ~a >1 ~ ~ ~_ > ~ 4~ ~ q ~Oo ~ 4~ _ ~ c 4J ~ JJ c ~ c IG ~ ~ ~ ~ ~ c 0 ~ ~ ~ q' s JJ 4J ~o _ Q. ~ ~a ~a t) 0 ~ c 41 c ~s c rl ·a ~ ~ 0 ~o ~ 0- ~ 10 ~ ~ C ~ V :> ~ C ~. - . - o) ~ 4) ~ _ tU JJ 4) ~ Y ~ _ ~ _ ~ C ~ ~ C _ 0 ~ ~ tn q~ n Q ~ O ~ q ~ ~ ~ ~ ~ ~ ~ al ~ · ~ ~ a) C) ~ ~ ~ t~ O ~ c JJ 4~ ~4 ·a c Q O ·^ oo O O x 4, ~ ~ ~ m ns ~ ~n 1 4, ~ m~ 41 4, ~ e ~ c) c ~ ~ a, · 0 e ~ c~ ~ :> _ ·. ~ ~ ~ c ~ ~ ~ ~ c s Oo tn ~ c ~ ~ ~ o. :- 0 43~ - c ~n ~ q4 :. ~".r1 ~n 0 ~c ~ - 0 0 0 ·- ~ 13 4) a ~0 q~ :^ 4, "rl 1 1 0 ~1 1 ~1 ~1 _tO 4~ ~ JJ 1 1 1~ 54 ~0 a, s 4J o`_ ~ ~ q~ ~ oo cn o4 ~ ~ ~ ~Oq~ ~C ~ C ~~S ~ ~ C ~- ~~ ~ e · ~ C ~ ~ 0 ~ _m >_ ~ 0 ~ ~ ~ 0 ~ ~- q4 1 ·. O · 4J 43 1 0 C a) ~a c (o :b O :~ c u' 0 ~o ~ a, L~ ~00 ~ ~ C tn O ~ ~u G ~c ~ ~ ~q~ 3 ~ ~ ~ ~ ·- 6, ~ ~ . - aQ ~ ' ~J as a, c 0 ~4' 4, c 4J :, ~ ~ ,a u~ JI ~ a, ~ .- s C a ~ ~ ~ ~ ~ c V1 JJ 4J ~ s ~ ~ ~ a) ~q~ ~ ~ JJ ~ s c ~ ~ c ~_ JJ m c 0 tn ~o a) ~ JJ Y ·- ~n c . 0 ~ ~ 43 4~ . . - .- u~ 4) ~ c Q) _ s~ ~ ~ ~ ~ C ~ O >- ~ JJ _ ~ ~ ~ ~ ~C ~ s ~, c ~ ~ s ~ ~ c a, ~ a' ~ ~ · ~ ~ ~ c ~ ~ a, ~ o' 0 81 41 ~ 3 e 3 ~ q~ ~·- p, Q. ~ ~ ~ C O ~ S O C O ~ ~ - >, E O q ~ · e ~ ~ q~ cn c) ~ o ~ Q S ~ oq C ~ u) C O ' C O S . - u) ~ ~d . - 0 ~ ~ q~ . - q~ q~ c ~ ~ 1 3 O ~ ~ Q O - :~ ~ JJ ~ G, ~ - ~ ~ ~ O S ~ ~ O ~ ~ I O c ·- ~4 ~ a) 0 ~ s q ~ ~rl ~ eq ~ Q~ 64 0 q ~ g] t ~n a, s a~ ~ s ~ 0 C ~ t) ~ ~ ~ ~ ~ ~ ~ ~ 1 c ~ 4J4~ e ~ ~ - ~n c 4J 4) ~ ~ ~ c ~ ~= 3 o O · ~ )~ ~ ' C ~ 1~ ~ _ 3 C ~ ~4 1 4' E ~ ~ JJ ~0 O :' ~G C G, ~ ~] C ~ ~ ~ e~ ~ C)~ ~ 0 C O~ a) ~ ~ c4~ ~C.~ a)q~ e~o ~_ >~_ o P4 ~ ~ ~ °q e m ~ 4~ · m ~ 4J ~ 0 a~ 4~ ~ ~ ~ ~ 4~ c c O ~ 0 ~ ~ q~ ~ e ~ ~ c ~ ~ ~ ~ - ~ ~ ~ ~ ~ c ~ ~ c - ~ Om ~-O~ eo ~ - U1 o~ ~ ~ ~ 0 ~ ~ ~ 4) a) ~ 0 ~ c q~ ~ ~ c . ~ C ~ _ ~ q~ ~ 0 c ~ 0 ~ ~ ~ 4, ~ ~ q~ ~ ~n ~ 0 c, ~ C ~n ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ c c ~ :> ~ ~ s q~ ~ o _ ~o ~ ~ 0 ~n ~ q~ . ~ a, e O ~ ~ ~ ~ 0 ~ o~ ~ ~ o' ~ ~ v, 0 e ~ ~ x xc osso cq~ ~ ~s Oq~ c ~ o~ ss~ ~ 0 ~ e o ~ c C~-~= C)~5 O0-== C) O ~ mq~o 3= 0 ~e 0 ~n os - mCJE~ JJ ~ ~a O ~ ~ JJ >' ~ >t L' ~-l ~ al ~ t) Q O~ C aJ aml C u,~ ~ al ~ a 4~ C · 4) V Ll t~ ~ O a' L' a' I c~o~e - - O ~ a~ al ~ ~ a' . ~a C ~ ~ ~ c a, C 0 Ll ~ c 0 a al JJ Ll {J Ll t) a) c O~' ~ ~ ~ C O C ~ C O q5 C o ~ X O C ~al ~ o3 a. O ~ ~ C ~Y ~ ~ c, ~ a, C) O a' O a' aJ q, O a~ a, ~·a s a ~ s L. n ~L. .a ~ a C : ~ C - ~L' O Ll r) O o- - O X ~ m a, Y 4J a' al e ~ _ a, ~ 3 L' ~ a' Le O ~ 4' ~ a, Ll N ol) L, ~ ~ ~ L' ~a C) ~ e ~4) c ~a -~ al ~ 4) JJ 1 .,' e ~ O ~ O L' · - S E~ s ~ >, J~ · q;5 I L' ~ O a' a, · 1 .CI =, L' c JJ _, a~ ~ 8 L. C 4, ~ C ~--01 C~ L L~ ~ 40 q~ C 1 ~r C~I O 4~ ~ e Ll ~ C _~O ·- 3 4, e:, ~ G, O al .^ c LlL1 4J C L' ~ t) L JJ ~ ~ L u~ a ~c _ ~a~ :~ C ~ _- _~ q~ · Ll · O ~ O C al c O ~ S-I . - V ·- ~ Ll ~ L' ~ C '4 · - t ql L 4' 0 0 ~ J~ P' C) e c, a~ · q;5 1 I L' ~ O q3 C a' a, ~ ~ ~ ~.^ · 1 .CI =~ L~ C~ ~1 0 ~ c O _I a~ a, ~a .,' ~ ~ O ~ ~ ~ ~ 8 L. C J~ `5 ~ O ~ al JJ ~ (V C ~ · - _' ~ L' U) C ~ ~ ~ C~ C · ~ L~ ~ 40 q~ ~ ~ ~ ~ L q~ 3 ~ :> O L~ >~ Ll L 3 ~ L' L. ~ 05 4, ~ a, q~ ~ ~ c al.a C al ~ ~0 :> ~ ~ ~a al ~ ~ ~u :> ~ CO ~ JJ (U dJ _ ~ ~ L~ t_ Gl C C _~ ~) L ~ ~ a, a' s~ ~ s0 ~ ~C aldP ~1 ~ al JJ~ ~r C q~ ~ ~ 1 C.) O O L~ O ~1) _I tJ ,' ~L. ~a C 4J L ~>~ ~ :~ ~I t) o O J) o J~ _ ~. - c ~ ' a, ~ ~ ~ ~ ~ L. ~ C L. ·r' ~ ·~1 4J 411 t.) 0 1 4J C? Q ~ 3 X ~a~ ~al 34 ~ ~ a, =~1 C ~ ~ :, ~ s~ c ~ ~ ~r v GJ - al ~ L' 1~1 4J ~C -I L' ~ al ~JJ · - L' ns (~5 O al ~c :^ al .a, c ~ C.)rO ~ s ~4~ a, ~ a) c L~ 4~ 4J ~I JJ G~ J~ Le · - as O q ~O C ~ 40 ~a 3 ~ O v ~ ~ ~ a~ ~a S L' ~a n ~ L. C ~ S ~ 4~ ~ :~ 3 ~ q~ L' >, O ~ 154 L' -I ~a L' t0 · L. ~ D1 ~ ~ ,a a, 0m C _. a' ~al I I · ~ ~J ^: I L ~- ~ ~ L~ ~ ~1l a n5 JJ O O ~ L. ~ ~ ~ ~ C l l Ll S e-~- S cO ~c ~d an `,C ~, z,, ~ -~0 =~= .,l O ~JJ 3 ~ Q O) E ~ C ~- · c2, ~ a, Ll C L. · Y C ~ .~1 0 - Z - O ~ ~ ~ O O ~ al C 0 C ~ ~ al t) c _ ~.- al q~ ~ ~ u,~ ~ ~, c s ~ s u, q~ ~ v 4~ O O ~ L~ · L~ O ~5 - u~ - C al o - V~ ~ O - ~ ~ ~ a~ o - a ~s ~ E C ~ r~ ~ 0 t° ·a ~° 4~ ,~ E ~ L~ O O ~ 0 JJ al C al . q: ~4 v J ~C, C ~ O ~ ~ ~ a' -' X C °J aJ t0 ~I Z J.) L. (P 0 ~ C a, ~ J., O `0 L. C-) n - ~-e ~ 11 G) s 3 al ro al q~ - 4~ N L~ ·- ~·-~ E ~4 al L~ ~ E~ ~ ~t~ U] ' - ~ .~1 0 ~1 C (V ~C ~ ~ C Z 11 =- - Z .^ ~ ~ ~ C · ~ a.) L. C ~ al q~ O O ~C) ~ ~ '- '-- ~-eq ~ ·a C 'l a~ C q~ a, ,4~ 4J ~ o4 {V ).4 O C Ll 0 ~ O G' ~ Z C ~ C a, ~ 3 ~ . - e ~ ~ O L' 13 S al ~4 ~ 4, ~ C) C al O C ~ al-~ c ~ al L' -' V O a, E ~ a, V~ '-~ O 04= 0 ~ ~4, 1 a~ C :3 a~ c o, ·a c C C al ~ c) c JJ a~ ~ q: ~ ~ O ~ E! C :> C ~Y al ~ a, O JJ O a'-- ~ O a' O 0 ~ ~ C 0 al ~ C du ~ ~ - ~- E - a~ ~ 4~ ~ ~ al ~ s-_ al S ~ ~ ~ ~ a~ ~ Q. ~ ~ 3 OJ Ll ~ al ~ ~ :, .,. ~e O E C L~ . ~ ·~1 ~ ~ ~ ·- ~ ~ V E . - -- E w O O aJ ~U ~ ~ ~ L ~S L~ ¢0 O t7J C) L~ ~ ·- ~ ~ Ll t~d ~ O .- tU L~ ~ ~ ~ ~ ~ ~ ~ 0 ~ s 0 ~c' c' u, al a' C) Q, ~ ~ ~ ~ ~ ~ ~ al O ~ ~ al O ~

300 a, E~ 4) ~o C SS E ~ CRE~ ~- ~1 O s q~ e ~ 0u ~ 4~ al . s ~n ~: ~ ~ ,~ _' ~ C) ~ ~ Q. J · - 4' · - ~ ~ ~ ~ C ~ mm e~ ~ ~ O O 4) C O ~ 4) q4 ~o ~ ~ O q~ ~ a, q4 O4 0 q~ O ~ ~ C ~ ~ ~ t) O ~ ~ ~ O :> ~ P4- - V q~ 4 4) ~ ~ ~ ~ ~ ~ ·e ~ ~ ~ ~ -~ ~ 1 0 _- - c: 1 ~ ~ ~5 4~ ~ S ~ ~ ~ ;~ C,) 4) ~ O ~ ~ ~ q4 0 4' C, ~ O O ~ ~ ~ ~ ~ = O ~ ~ O v 4' ~ O ~ q~ $~ ·r4 a' C) S ~ JJ ~ H4ar1 ~ ~ . U 4' ~a ~ 4, u, ~ 3 t11 1 C' U) 4, ~ U) ~n 4) ~ O 4, S ~ ~ ~ ~ ~ O ~ O <: O q~ ~ ~ ~ C ~ . : 4 ~C _ ~Q ~ ~ . - ~ ~ ~ . - ~ a ~ ~· 0 ~ ~ 4) O- - i4 q4 O ~ · ~ ~ ~ ~ U] q~ O O ~ JJ - ~ ql 4' ~ ~ ~ O ~ ~ .,, ~ ~ ~ ~ _, .,~ ·- : 14 ~ 4, ,1 ·o ~t U) :^ ~ S U) U) O m O S ~4 O .~ - C U) ~ O 1 ~ ~ ~ C . - ~ c c q~ a. a, ~ ~ ~ q4 4J C ~ ~ O ~ ~ ~ O ~n 4~ ~ ~ q~ a~ a~ Ll O ~ p~ o O C ~ L~ :>,.C O q~ q-l ~ 1~ 4J Ul ~ 4, a, O JJ 4~ 4~ ~ ~n o. S ~ O C S q~ d0 ~ ~ ~ O C la1 I tr ut ~ ~ O O q4 ~4 ~ ~ ~ 4, ~ £~^ ~ C C. - ~1 41 .~4 o 4J.~1 3 C ~ ~ ~ £ ~ [Q ~ O . - q4 ~o C ~ 14 0 n ~ o. tJ -d, JJ u) q' o c) a) p: c) .~' q4 o Ll a~ 4~ ~ os c - ~4e ~s ~ ~ o ~ ~ ~ . ·~ ~4 ~ ~ c 4) ·,~ ~ ~ oo t) o go 4J s ) ~= ~n .,, _ ~n l C, ~ U, 43 ~ ~ o 4J ~ c~ e~ ~ 0 4, ~ 0 s q~ ~ ~ s 0 ~ $4 0 0 C) 0 ~ q~ JJ tll 1 U] £ ~on ~ · - ~ 3 ~ g ~·^ ~ · q~ a~ 0 a, s q~ £ JJ O ~ O C) ~ ~ ~ a Ll q~ C O · - ·^ C) ro ~ ~4 ~4 c ~ ~ a, .,, ~ ~ c, 4) ~ Y C £ ~ ~ O ·~ X-~ E~ O ~ O ~ 4J ~a L. · - U) ~ t0 >, ~ ~D _ ~ ~q ~ q4 U) O JJ C 64 o, O c c ~ a, .,- 4,, .Y ~.e L' JJ C U) U) ~rl 4J ~Q 4~ a~ IB · ~ c n ~ q' a O ~ ~ c 0 3 q~ ~ q~ 0 ~ ~ ~ ~ ~ q~- ~ C D E O ~ a, 4) .C ~a U s ~ 3 4~ ~n rn~ E C = O ~ sUn ~ ~ .- a JJ~ ~ ~ O ~ ~ L' ~ S ~ C ~ ~ ~ c m 4J c o' ~ ~ ~ w ~ ~ ~ ~ ~ o ~ q~ :3 ~ ·- IJ U) O = - c ~ ~ s ~ da ~ ~ 3 ~ ~ O eq ~ o eq ~OmS ~ .,, · -4, ~ ~- 1 S ~ ~ ~ ~Q. 0 ~ ~ a~ ~o ~ ~l ~£ ~ £ :, ~ ~ ~ 0 C 0 s ul O ~ ·- ~S ~ ~O ~ ~ m . - ~ ~ ~ q~ c ~ q ~·^ IU _ ss ~ ~ 0 4) v E~ ~ ·^ ~ r0 0 ~ ~ ' ·,1 U) ~ ~ eq a a, ~ ~ · ~ Q' ~ c u ~ ~ m" ~ ~ ~ O C ~ ~ ~ ~ D ~ ~ S~ ~ ~ ~ ~ ~ ·- ~ C X · - O ~ O UJ RJ D ~ ~ · - ~ ~ ~ ~ ~ C ~O ~ £ o,~ dJ ,~ ~1 o4 ~ ~ S ·,O ~ 54 £ 4~ ~ O o 14 C) :~ £ ~1 1O E~ ~ :~ ~ ~ ~ ~ ~ ~ C ~ ~ ~ C 4, 0 0 X a: ~- a 3 ·- ~ a~ 3t ~ ~:1,= v~ a u, ~ c 0 4~ eq JJ ' 1 o S ~ O O ~ ~ ~ ~ . 3 ~ C =~ a ~ ~ ~ == ~ ~ c c ~a ta ~ Q ~o ~ <~ ~ ,~ ~ ~ g u~ ~ ~ c~ ~ 3 3 , ~- c %4 0 w ' 0 0 54 ~ `4 ~ q~ ~ ~ ~ . ~ ~ =m 0 Lq g c JJ ~ C~ ~ ' ~ 11 0 g n ~ ~ ~ c ~ a) O ~ ·-l 00 O O ·S ~1 ~ ~ :^ Z ~ ~ 1 a) ~ eq ~ Ll ~ {0 _1 ~ ~ O eq ~ O ~ C ~ _t ~ fo L~ ~, ~ ~ 3- - '~ ~q~ C O C O d-= ~ ~ ~ ~ S~ ~ O ~ C U)-· - ~ ~ ~ ~ ~ JJ ~4 W O . ~q4 m 0 ~ ~ ~ w C m 14 0 41 ~o C 64 ~ O ~ ~ ~ 4, ~ ~ ~ 0 u s C m ~ ~ 0 C: 4, mm ~ q~ ~ cR o ~ ~ 4, ~4 m ~ 3t ~ :~.O ~ q4 ~0 0 ~ ~ ~ P' 4, ~ 4) 4' ~ ~ ~ 10 m ~ 3 0 . ~ C ~ ~ ~ 1 0 ~ ~C ~ :, ~: ~ C 0 m ~ m ~ =_ 0 4, ~ ~ ~:, C ~ ~ - 1 ~ ~ m ~ :> m O 4~ ~ O g Ll 4~ 0 ~ 3-I (p .,. ~1 ~1 u4 Ll O ~ C 1!u 0 ~ 4, P. O ~ ~ 0 ~_ q:, 4, ~ C 0 · - O C Oq JJ 4, ~ ~ C) Ll ~ JJ 0 4, ~ ~ ~ 3 ~ u, m C g C ~ O ,~ - q4 0 :~ £ 4' ~ ~ O _ ~ Ll ~ C Ll a, ~ JJ .,., 0 ~ C ~ C 4, S 4, C' 64 ~ D ~· - q ~4.] es 4, o ~ ·- ~ o ~ o ~ :' ~ ~ o =- w ~ ! u) JJ u) c u~ a ~ ~ o ~ q~ o l ~ c o ' oo C 11 I ~ ~ q~ ~ ~ ~ z q~ ~ 3 _ O , ~ 4J q ~ O a) .- 1 v.a ~ ~ ~ U] · - · - 1 ~ U~ r0 J~ ·~ ~ C ~ C ~4 3 ~a ~ ·-

JJ o a' ~ q~ 4, ~ c ~ ~ o ~ome C _ 4, 4) s C C ~ ~ v~ cr 4 .,. ~ ~ ec 4) ~ ~ 4' ~ ~ ~ C q~ m ·^ 3 ~4 q~ 1 V O ~ c ~n 54 00 4' 4, ~ O '- c, ~ os e~ u4 ~ 0 - C) ~o ~3 U) C 4., q~ o , - ~ q~ C a) oq ~n ~ 1 t) ~ C In a) C :1 4' q~ ,~ o q4 U] a' . - o 4, ·^ 0 ~ . - c s QJ ~n 4, ~ Ll q~ C C o ~3 C ~ q4 U] O .Q C eq 4~ c 0 ~n ~ ~ s 4' ,,. JJ o ~ - .~4 ~JJ ~ ~ · e -1 0 ~_ ~n ~ ~ 4) m eq a' ~ c c, ~ C ~ 4, JJ 4) ~ o~ ~o ~ O c tr ~ s ~ ~ q4 ~ )4 0 ~ ~ q~ ~ O 43 ~ oq ~ s ~ c ~ O o o 4\ ~ :- 3 E ~ 4, ~ e JJ . - ~n Ll C ~ 41 Ql P' ~n . - S~ . . - 4, ~ q~ a, s 0 s ~ . - 43 ~ ~ a, q~ 4, 0. - ~ ° t) ~ Ll C q4 ~ 0 ~ ~ . - ~ C a) ~C ~ . - .,. ~ ~n :> 4) E c) o~ a' X ~ ~ E~ ~ ~ O I. .,, 0 JJ s ~n .,. s 0 JJ (Q ~ C~ JJ , - F :' O ~a ~S ~ ~n c a' _ a, :' ~3 a, ~ 40 ~ ~ G) s O ~ ~: = C u, c .- ~: q4 ~ u' O a' a' ~ ~ 4~ ~ 0 0 P4 ~ a, F CJ~ 54 J~ ~ ~ ~ a, u,-ma JJ {Q 1 ~ 4' 0 0 ~ C) :> · ~o C m . Y c) s 64 ~o , q~ ' - C Ll c) ~ :, ~ ~ O c m ~ ~ u~ I ~ 4, m 4' c u, 0 C ~ OQ ~ a, ~ JJ C . - ~ q~ ~ L1 4, s ~ ~ ~ O ~ 3 ~: O q~ s JJ G, a, ~ c = _' C u~ oo ~n C 4) q~ a~ s q~ Cn O ~ C S 4) 4J 0J 3 4J O q4 a, c .,' 0 a, .,, .,, F :> ~q - C O a .,' q4 O~ q, O L' . - to )~ JJ ~ ~n O. GJ ~i U] ~Q L' Q q~ O Ll U} 1 a~ q~ ~q 41 e ~JJ . - ~ ~ JJ F 0 C C 41 0 . - a eq a' Ul a, 4) ~ ~ ~ ~ ~ - a, q~ ~ ~ 3 a) 0 JJ s ~ ~ ul 43 c ~ ~r a s~ ~ ~ q~ Ll C ~ q~ ~ ~ ~ O al q3 ms c C E~ a, ~ ~ O x ~ . u, . - E ~ ~ C, ~a a~ c ~ a U] 00 ~rl ~ ~5 a~ :~ ~ 3 s ~ ~ ~ c 54 v c ~ - - 3 ~ ~ p4 - O ~ E~ JJ C' O ~ ~q 1 L. . - . - ~n 54 ~ L~ ~ ~ a) ~ s o' u, ~ Q. q~ aJ O O q~ 54 o' ~a $4 ~a o' ~ E a) a' ,1 . C o :>. ~s C JJ C E ~ O . - ~ . - S 4) ~ Ll O ~ ~q o s ~ , 4J ~ O 4' s o Ll U] .,, 0 a :' ~ C t) ~ U) 0 un a, eq ~ ' ~oCO o tt )" o 11 U] a' 'rl Z U) - 4d q~ ~ V, ~ ~ <: s 53 C Ll q4 4J q~ 3 0 O c s 3 0 o 54 q~ 4, ~5 ~ c a' ~ JJ ·u eq q~ a O o r~ ~ O c ~P o E O a ~n Q C~ ~r O s 3 u, c u~ a, a) 0 s u,n c ~n ~ ~ u' 4~ ~ a' o c s ~ q' O ~ c 301 C) ~ oo 4, 4' 4, oB N U) C a q~ 4, 4, ~ a, ~ ~ O ~ fo ~ ~ e~ · m 00 ~ ~ U) 0 :-s ~c) 0 ~ ~ C t) ~ ~ ~, ~ ~ C c a, :' ~ ~ ~ O ~ ~ C O -~ 0 q~ ~ ~ JJ cn ~ I =. - s c ~ C~ O ~ ~ 0 = . - ~ c a, 54 a' ~a ~ ~ s U) a, 41 ~ ~ ~ C ms U] ~ 0 0 ~, ~ C' c ~ ~a 0 ~ n. ~ 3 ~ ~ ~ JJ 0, 0 10 q5 ~ Q ,~ dJ ~ C U) O eq O ' - C 4) ~ E ~ c, 4' ~ ~ ~ ~ :^ S 4,J Ll q~ rl ~ ~ ~ V O JJ O q~ ~ Q. ~ e s U) 0 ~q q4a) ~ C eq ~ . - F 16 S 10 ~ O X :~ c~ a U) o ~ 0o 0 = - O o? U) C 0 ~ a, m0 ~ q~ q~ C O U) C 11 a, a~ a~ q~ a, z ~4 ~ 3 _ O :' . ~ 0 ~ ~ ~ Q . - ~ C ~ C q~ 3 ~ ~a . - a' ~O . o u, ~_ u, a\ . - o E o q~ JJ U) ~o q~ o .,' . - o a, s v o ~n E .,' ~q a .,, .,' Q. U) a, U) c a, q~ q~ o a q3 a 3 :z ut a .,, a) . - C a) c S~ a~ 84 a' a q~ ~ c 4: ~1 c 4, q~ o .

302 calculation of frequency rates to periods when the offenders were at risk of commit- ting offenses in the community. Time when the offender was incapacitated through in- carceration or long-term hospitalization (e.g., more than a 1-month stay) was ex- cluded from the time at risk. The resulting frequencies reflect the intensity of offend- ing while an offender is criminally active in the community, A, the rate of offending that would be expected ' if the offender were never incarcerated. The estimates of A are to be distinguished from the effective rates for offenders, A*. In estimating the reduction in crime associated with different periods of incarceration, the appropriate quantity is the individual's active offending frequency. This is the rate at which crimes would be committed if the offender were not incarcer- ated. Any time spent incarcerated will re- duce the annual active rate to yield the effective rate for offenders. For example, an offender may commit crimes at a rate of 10 per year while he is free in the community. If this offender is incarcerated for 6 months, however, he can only commit crimes at rate 10 for the 6 months he is free. His active rate is 10 offenses per year, but his effective rate during the entire year is only 5, since he was only actively committing crimes in the community for half of the year. The effective offending rate reflects the reduction in the potential level of crime as a result of current incapacitation policies. Be- cause effective rates are already discounted by current incapacitation levels, using the effective rate instead of the active rate would lead to underestimates of the total crime reduction associated with increases in incapacitation. When the effective annual offending rate is used, incarceration during all of the following year would be incor- rectly estimated to avert only five offenses. This fails to include the additional five of- fenses that would have occurred had the offender not been incarcerated for one-half year. On the basis of the offender's offend- ing frequency while free, incarceration for a full year can be expected to avert 10 of- fenses. CRIMINAL CAREERS AND CAREER CRIMINALS Estimating A from Self-Reports: The Rand Inmate Surveys Survey of Habitual Offenders. The study of 49 habitual offenders by Petersilia, Greenwood, and Lavin (1977) laid the groundwork for later Rand surveys of larger samples of inmates. The original 49 inmates were chosen as exemplars of serious, recid- ivistic offenders. To be included in the sample, an inmate had to be currently serv- ing a sentence for at least one armed rob- bery conviction and have at least one prior sentence of incarceration. Through per- sonal interviews, the inmates were asked about their frequency of offending and prior criminal record (arrests, convictions, and incarcerations) for nine offense types, as juveniles, as young adults (before their first incarceration as adults), and prior to the start of the current sentence. In addition, they were asked about other aspects oftheir personal histories, including family circum- stances, school and employment experi- ences, drug and alcohol use, personal moti- vations for crime, and styles of committing crimes (e.g., the amount of planning and preparation, use of accomplices). The findings on average levels of offend- ing over time for the 49 offenders, including offenders who were active in an offense and those who were not, are summarized in Table 2. As one reads down the table, the offense classes become more inclusive; the total rate includes offending in any of the nine offense types surveyed. Except for vi- olent offenses, the reported monthly rates declined markedly as offenders got older. The anomalous slight increase in monthly rates with age for violent offenses is attrib- uted by the authors to the sampling crite- rion that required an atoned robbery prior to the current incarceration, which marks the end of the adult period (Petersilia, Green- wood, and Lavin:27~. As indicated above, all 49 offenders are included in a rate, whether or not they were active in that offense type. The opposite trends across age observed for violent of- fenses and all other offense types illustrate

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS TABLE B-2 Aggregate Offense Rates per Month of Street Time for 49 Habitual Offenders 303 Juvenile Young Adult Adult Offense Class Period Period Period Total Violent 0.1O O.16 0.20 0.15 Safety 1.15 0.43 0.24 0.49 Nondr~ ~2.37 0.92 0.38 0.99 Total- 3.28 1.52- 0.64 1.51 Mean number of offense types reported 2.50 2.50 1.85 4.00 NOTE: The reported rates are aggregate incidence rates for the sample and not frequency rates for active offenders only. Rates were obtained by dividing all offenses reported by the total number of months at risk for the entire sample, whether active in an offense or not. Violent offenses include robbery, aggravated assault, and rape. ~ afety offenses include burglary in addition to the violent offenses. CNondrug offenses include auto theft, purse snatching, theft over S50, and forgery in addition to safety offenses. dTotal offenses include drug sales in addition to nondrug offenses listed above. SOURCE: Petersilia, Greenwood, and Lavin (1977). the important contribution of participation in aggregate incidence rates. As more of- fense types are included in successive rows of the table, the aggregate rates decline more markedly with age. As indicated by the last line of Table 2, the decline with age is influenced by the decline in the number of offense types reported by the offenders. By contrast, because of the sampling crite- rion used, aggregate rates increase for the class of violent offenses as the number of offenders active in robbery increases in suc . cess~ve age perioc s. The 49 habitual offenders surveyed were serious habitual offenders. Together, they averaged 21 years of criminal activity from the time oftheir first reported offense to the current incarceration. Almost 50 percent of that time was spent incarcerated. Despite their strong commitment to crime, these chronic offenders averaged only 18 offenses per year of street time. Because these rates fail to exclude offenders who were not crim- inally active in an age period, however, the reported rates underestimate individual of- fending frequencies for active offenders. One of the more interesting Endings of this study is the offenders' extensive in- volvement in employment. Two-thirds of the offenders reported that after age 18 they were working at least 75 percent of the time they were not incarcerated. Three-quarters of the respondents indicated that while em- ployed, as adults, they were working full

304 time. This high participation in legitimate employment by chronic offenders about 56 percent, on average, were employed full time in any month free (.75 x .75 = .561 matches that found in nationwide surveys of prison inmates, in which about 60 percent of all inmates in 1974 and in 1979 were employed full time in the month prior to their arrest (Bureau of Justice Statistics, 1979b:73, 1981c:2831. While employment was widespread among the 49 offenders, the jobs held tended to be economically marginal. Only half of the respondents re- ported that a job was their usual source of income as an adult. First Inmate Survey. Peterson and Braiker (1980) reported the results of the first large-scale survey of self-reported of- fending by prison inmates. The original sample was drawn randomly from the in- mate population at five California prisons in 1976. Usable survey responses were ob- tained from 624 inmates (47 percent of the original sample). The low response rate raises some concern about possible re- sponse biases in the final estimates. A comparison of the respondent sample with the total inmate population revealed no significant differences in terms of con- victed offense type, age, or race-ethnicity. Inmates with more extensive prior records, however, were overrepresented among re- sponclents. The inmate respondents might thus be expected to have higher individual offense rates than inmates generally. In ad- dition, by relying on a sample drawn from the resident population, with their longer prison terms, the offense rates of the inmate sample will not apply directly to either a cohort of incoming prisoners or the more general population of active offenders who are not incarcerated. The survey asked respondents to indicate the total number of times they had commit- ted each of 11 offense types while not incar- cerated during the 3 years preceding the start of the current incarceration. The fre- quency of offending was obtained from re- sponses to fixed categories: 0, 1-2, 3~, ~10, and more than 10 offenses. If the frequency was more than 10, respondents CRIMINAL CAREERS AND CAREER CRIMINALS were asked to indicate the total number of offenses they committed. For frequencies of 10 or less, researchers used the midpoint of a category to estimate rates. Respondent reports of time incarcerated or hospitalized were used to estimate street time during the 3-year observation period. Offense-specific individual frequency rates reflecting the number of offenses committed per year of street time were computed only for of- fenders who reported that they were active in- an offense type. Using estimates of the probability of arrest and incarceration for a crime, and the expected time served in prison, the researchers adjusted the esti- mates of it for the inmate sample to reflect the rates expected for a sample of incoming prisoners and for a sample of"street of- fenders" (i.e., active offenders who are at some risk of incarceration after an offense). Estimates of ~ and A for the three groups of offenders are shown in Table 3. As was expected, the As reported by the resident inmate sample are higher than those esti- mated for a sample of street offenders. The difference reflects the greater risk of longer terms in prison faced by high-rate offenders. Even if high-rate offenders are no more vulnerable to incarceration for any single offense than are other offenders, their higher As increase the chance that they will be incarcerated during some observation period. Except for drug sales, street offend- ers averaged under five offenses per year free for several serious offense types. A number of factors raise concerns about the reliability of the estimates in Table 3. First is the imprecision of the frequency categories used in the survey. As designed, the frequency categories provide the great- est resolution for low frequencies; they are far less precise for larger numbers. Aside from the error resulting from using the mid- point of a frequency category to estimate A, the uneven treatment of categories may have implicitly suggested to respondents that low frequencies were more sensible responses. This has been found in other contexts to bias responses downward (Locander and Burton, 1976; Bradburn and Sudman, 1979~. Second, further unreliability is likely to

30S o _ _ At: U] · - U] I: a' a) ~ a' a' to to . - a' 0 o ~ ~ o · - P: ~ 1 · - ~ U] U] ~ o. - ~ P. =1 - a) ~ o ~ o .U . - ~ U] ~ · - C) ~ ' · - U] tU o so a) PI tU ~ 1 0 a] m a) ~ m~ ~ E~ ~ ~ ~1 U1 S~ a a o 4J a S~ cn ~1 UO S~ a o U] .,, P4 .-, 8 H U2 a o .,' S~ a) .,. U] vl ~ ~ :^ q:5 ·- ~ -~1 ~ O a, ~5 ~ a H O ~, ·-=l a' ~ 0 ~1 ~ . - ~1 ~1 C) v1 ~ ~ v · - a H O ~' ~ ~QI a) ~ 0 ~1 ~ ~J P4 vl :, ~ V O a ~1 (L) ~ ~ S~ H O ¢. .-I V ~ ~ O S-l ~ . - S~ Ll ~ Q] a U) O E~ 00 (D CO ~O ~ ~ ~ C~ U~ In ~0 a~ ~ vg u~ O · ~ ~ ~ ~ u ~n u~ ~° IO ~ V dJ O O s s a u~ E~ m tD CO 0D ~ O ~n ~ ~ ~ 0 ~ 0 ~ ~r 0 ~ o' ~ ~r · ~ ~ ~ ~ ~ d4 ~ ~ ~ 0 ~ U~ ~ ~ ~ ~ ~ ·n u' ~a, a) ~a a ~ ~ ~ ~ s >' {s Q ~ ~ ~ ~ ~·- ~ ~ t15 U] _I 4) V ~ ~4 a) O ~ ~Q ~ ~ 0 ~ u,. - ~ ~ ~ ~ ~ O O O P: ~ ~ a~ ~ a) 0 S ~S 3 n O c: U] Q, ~ cn JJ ~ ~a ~ ~ ~ o ~ .- ~ ~1 0 4 O? a) ~ ~ O = O ~ V -~ a. u' ~ ~ ~a X a, O ~U] a) a) ~ u' ~ s ~a s .,' .o ~ ~, O V O ~ 3 n5 -1 ~a)u2 U) ~ ~ ~O~ O O 3 ~ ~.- JJ ~ ~U] c~ ~ a, (D 54 ~ ~ O · ~,~ 11 0 ~ ~ U] ~a ~ V 4 Z _ ~ ~ ~ ~ · O O u, ~ a) ~ · ~, ~ ~n u, ~ ~ a~ ~ ~ Q ~ _ -t ~ ~ ~ O O O a' ~ a) ~ u, V ~ a) ~ u, O s a' ~V a) a1 V U] ~ ~U] a, ~ ~ ~a .- ~ U] 0 ~ ~- - · - ~ ~ 0 ~ ~ ~ ~ ~ ~ ~ 0 · - V ·- ~ a u, :> a) ~ ~ a) ~ a) u, ~ ~ O u, {5 a, ~- a) ~ s=_~ O v ~ v ~ a O ~ ~ ~ ~ ~ O Q ~ a) .-, a a) ~ ~o O ~ ~ ~ ~ ~, a ~ ~ 0 - u, ~ a) 0 .,- ~ ~ ~ V- - ~ a) ~ ·- ~ ~ ~ ~ Q) s ~ ~ a) a) n' E~ Q ~:: ~5 ~: ~ ~ ~: m1 ~ o ~ OQI `4 ol Q u, a) tQ a O -' ~ ~ U2 Q ~0 ~ O . O s: .,' ~4 o .,' h a, ~q Q o o u, a Q ~a a) E~ > ~·e 1 o a t) a Y O u' m a . c o U] U] a P4 a a, o ~.. s C) 4 ~0 ._ ~U) o v

306 have been introduced by the use of a 3-year observation period. Except for highly sa- lient events or regularly repeated events, the recall problems inherent in reporting total frequencies for a period as long as 3 years are likely to be substantial. The recall problem is exacerbated by the delay intro- duced by the length oftime served between the end of the observation period and the date of the survey. Peterson and Braiker (1980:16) note that, "in most cases, the re- spondent did not have to reconstruct events occurring long ago." As evidence for this claim, they indicate that half the respon- dents had served less than 2 years at the time of survey, and three-quarters had served 3 years or less. This means, how- ever, that half the respondents were being asked to report on the frequency of their criminal activities that occurred 5 years or more before the survey. The frequency es- timates are thus vulnerable to large report- ing errors. Third, if respondents reported commit- ting an offense, they were assumed to be active in that offense throughout the 3-year observation period. Ignoring possible initi- ation or termination of careers in that of- fense type sometime during the 3-year observation period would understate of- fending frequencies during truly active pe- riods within those 3 years. In a later section on estimating As from participation rates and aggregate incidence rates, however, it will be demonstrated that this bias is less serious when As are large. Fourth, the survey was anonymous and so provided only limited opportunities for internal checks on the quality of responses. On the basis of checks that were made, there is reason to suspect that the As were underestimated. As many as 6 percent ofthe respondents denied committing any type of offense, including the one for which they were incarcerated. Those respondents were either all falsely convicted or, what is more likely, many of them underreported their offenses. Another internal reliability check was available in the form of a question that asked the frequency with which burglaries were committed in a "typical month." CRIMINAL CAREERS AND CAREER CRIMINALS Based on responses for a typical month, the median frequency was four burglaries per month active; based on the total number of burglaries committed during reported street months in the 3-year observation period, the median monthly frequency rate while active was only .24. This 16-fold difference in reported rates led the authors to express uncertainty about the accuracy of the abso- lute magnitude of rates reported by offend- ers (1980:26~. The difference may reflect the fact that while estimates of A based on reports for the total observation period may be underestimated, estimates based on re- ports for a "typical month" are vulnerable to overestimates arising from erroneously ap- plying short-term spurts in criminal activity during that month to the entire observation period. When the two frequency estimates are compared, there is much better agree- ment (correlation of.79) on relative magni- tudes using categories of none, low (for those below the median), and high (for those above the median). Thus, the analysis of correlates later in this appendix is based on these grosser characterizations of offense rates; absolute rates are presented only to illustrate significant differences that are found. Estimates of A for the entire observation period reveal certain general features of individual offending. First, individual fre- quency rates are highly skewed; most ofthe offenders reported that they committed zero or only one crime within each offense type, but a few offenders reported very high of- fending frequencies (Table 4~. This same general distribution of frequencies was ob- served for burglary when either total num- ber of offenses reported for the observation period or the rates based on a typical month were used. Second, individuals also indicated con- siderable variability in their types of of- fending. Very few offenders (18 percent) reported committing only one type of of- fense exclusively; 49 percent reported com- mitting four or more types of offenses dur- ing the observation period.5 5Peterson and Braiker (1980:Table 17~. These percentages are based on the 543 respondents who

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS TABLE B-4 Distribution of Number of Offenses Reported, First Rand Inmate Survey (N = 624) 307 Percentage of Respondents Reporting Each Number of Offenses in 3-Year Observation Period Offense Type 0 1-23-5 6-10 10+ Armed robbery 66 197 5 4 Burglary 54 1611 8 11 Car theft 73 17- 5 3 3 Aggravated assault Beating 65 256 3 1 Cut/shot 80 154 1 1 Threat 64 227 4 3 Forgery 68 158 5 4 Rape 94 40.6 0.3 0.3 Attempted murder/ murder 82 134 1 0.3 Cons 44 1812 16 10 Drug sales] 61 155 7 11 Thor drug sales, the frequency categories are 0, 1-10, 10-50, 50-100, and more than 100. SOURCE: Peterson and Braiker (1980:Table 7). Third, variability in the number of of- fense types in which an offender was in- volved was a major factor in the decline in total offense rates observed for older of- fenders. As seen in Table 5, average total offense scores were lower for older offend- ers. These scores, however, reflect the com- bination of participation in an offense type and offending frequencies by those active in that offense. There is considerable stabil- ity across age when considering only aver- age intensity scores for active offenders (the middle column of Table 5~. The decrease in total offense scores arises primarily from a decrease in the number of offense types reported by older offenders (last column of reported committing any of the 11 surveyed offense types. Another 81 respondents denied committing any of the 11 offense types; roughly half of the 81 (6 percent of all respondents) were convicted of a surveyed offense on their current incarceration. Table 5~. The offending frequencies for ac- tive offenders reveal little evidence of sys- tematic trends with age for most offense types (Table 6~. Only the rates for aggra- vated assault (and drug sales after age 21) declined significantly with the age of the offender. (Because of the small number of offenders and the very low rates, the de- cline for homicide with age is not statisti- cally significant.) The results for age do not necessarily represent aging effects for individual of- fenders because they are based on rates observed in an age cross-section (age in the 3-year observation period). The same of- fenders were not compared over different ages. Thus, the age differences reflect a combination of aging effects for individuals and cohort effects across offenders. Separat- ing the aging effects from the cohort effects would require longitudinal data on frequen- cies for several cohorts and assumptions

308 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-5 Variations in Individual Offending Frequencies (I) by Age, First Rand Inmate Survey Total Intensity Number of Active Aged Offense Scoreb Scale ScoreC Offense Type Under 21 0.64 1.44 4.61 21-25 0.57 1.42 4.09 26-30 0.49 1.44 3.29 Over 30 0.35 1.33- 2.70 Wage was taken at the midpoint of the 3-year observation period. bFor each of the 11 offense types studied, respondents were assigned a score that reflected their level of activity in that offense. If the respondent reported that he never committed the offense, he was assigned a score of 0 for that offense. Respondents committing the offense at a rate below the median were assigned a score of 1; if their rate was above the median, they were assigned a score of 2. Scores were averaged over the 11 offense types in the study to yield a respondent's "total offense score." Individual scores were then averaged across respondents. CFor each offense type studied, respondents who reported commit- ting that offense were assigned a score indicating whether their offense rate for that offense type was below the median (score = 1) or above the median (score = 2). Scores were averaged over all offenses in which the respondent was active, to yield his "intensity scale scorg. n These scores were then averaged across respondents. -Mean number of separate offense types respondents reported committing out of 11 offenses studied. SOURCE: Peterson and Braiker (1980:Tables 27-29). about the nature of the relationships among age, cohort, and period effects. Fourth, As varied systematically with other attributes of the offenders. As indi- cated in Table 7, both the intensity score for active offenders and the number of active offense types were lower for blacks than whites. Offenders with more extensive prior records reported being active in more offense types and had a higher average intensity score within active offense types than other offenders. Likewise, self-report- ed drug users reported higher numbers of active offense types and a greater intensity of offending within active offense types than nondrug users. Finally, offenders reporting extensive ju- venile crime reported active involvement in more offense types and a greater intensity of offending in active types than other offend- ers. These attributes generally maintained significant independent contributions in multivariate analyses to distinguish high- rate offenders. Drug use, however, was sig- nificant for property crimes but not violent crimes in multivariate analyses. Second Inmate Survey. Rand's second survey of inmates, described in Marquis (1981), Chaiken and Chaiken (1982a), and Peterson et al. (1982), is the most compre- hensive of the Rand surveys; it includes

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS inmates from prisons and jails in three states (California, Michigan, and Texas). Use of multiple jurisdictions and inclusion of the presumably less serious category of offend- ers jail inmates provided the opportu- nity to assess the variability in A among different inmate samples. As in the first Rand inmate survey, re- sponse rates in the second survey were generally low, averaging from 49 to 71.5 percent among jail and prison inmates in California and Michigan (Table 81. Re- sponse rates were lowest for inmates at state prisons. The notable exception was the 82 percent response rate among Texas prison inmates. There was also considerable vari- ability in response rates across institutions. (The Texas jail inmates were dropped from the analysis because they were found to include many offenders awaiting transfer to state prison and hence were much like prison inmates.) Because ofthe low and varying response rates, important differences in offender at- tributes were found between the originally designated sample and the initial respon- dent sample in all three states. Respondent representativeness was improved in Texas by weighting respondents by the response 309 rates found in different institutions. In Cal- ifornia and Michigan, responses by replace- ment respondents (who were selected at the same time the original sample was drawn) were included. This improved re- spondent representativeness on most char- acteristics. Nevertheless, younger inmates (under age 25) were overrepresented in the final sample and inmates with low reading levels were underrepresented among re- spondents in both states. In California, His- panics were also underrepresented among respondents. Building on the experience gained from the first inmate survey, the researchers in- cluded some design changes in the second survey. First, rather than draw an equal probability sample of Me resident popula- tion of state prisons, they drew a sample of prison inmates to simulate a cohort of in- coming prisoners. This was done to facili- tate analysis of the implications of sentenc- ing policies. A cohort of incoming prisoners is closer in characteristics to offenders con- victed ant! sentenced to incarceration than is the resident population, which includes many more individuals sentenced to long terms than does an incoming cohort. The sample of prison inmates was drawn by TABLE B-6 Variations in Individual Offending Frequencies (I) with Age for "Street Offenders, n Based on Self-Reported Offenses in First Rand Inmate Survey Age of Inmate in 3-Year Observation Period Offense Type Under 2121-2526-30 Over 30 Armed robbery 1.9 2.12.5 1.0 Burglary 3.9 13.98.3 5.3 Auto theft 4.3 2.31.3 5.2 Aggravated assault 3.1 2.8 2.6 1.4 Forgery 1.2 3.3 7.5 2.6 Rape 0.9 0.9 1.9 0.8 Homicide 0.3 0.2 0.1 0.05 Cons 5.1 8.4 8.1 9.6 Drug sales 62.0 237.0 112.0 91.0 SOURCE: Peterson and Braiker (1980:Table 30).

310 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-7 Variations in Individual Offending Frequencies (I) by Attributes of Offenders, First Rand Inmate Survey TotalIntensity Number of OffenseScale Active Offense Attribute ScoreaScor b Typed: Race White 0.641.50 4.22 Black 0.48-1.39 3.56 Hispanic 0.561.38 3.90 Prior criminal record ****** *** No felonies 0.361.28 2.92 Prior felonies 0.541.42 3.91 Prior prison 0.641.52 4.19 Drug use ****** *** Drug user 0.691.54 4.56 Not drug user 0.461.37 3.39 Juvenile crime ****** *** None 0.301.29 2.32 Not serious 0.501.40 3.63 Infrequent, serious 0.651.48 4.31 Frequent, serious 0.811.64 5.15 Thor each of the 11 offense types studied, respondents were assigned a score reflecting their level of activity in that offense. If the respondent reported that he never committed the offense, he was assigned a score of 0 for that offense. Respondents committing the offense at a rate below the median were assigned a score of 1; if their rate was above the median, they were assigned a score of 2. Scores were averaged over the 11 offense types in the study to yield a respondent's "total offense score." Individual scores were then averaged across respondents. -For each offense type studied, respondents who reported committing that offense were assigned a score indicating whether their offense rate for that offense type was below the median (score = 1) or above the median (score = 2). Scores were averaged over all offenses a respondent was active in to yield his n intensity scale score. n These scores were then averaged across respondents. Mean number of separate offense types respondents reported committing out of 11 offenses studied. *Significant at the .05 level. **Significant at the .01 level. ***Significant at the .001 level. SOURCE: Peterson and Braiker (1980:Tables 27-29, 34, 82, and 84).

APPENDIX B: RESEARCH ON CRIMINAL CAREERS TABLE B-8 Response Rates to Second Rand Inmate Survey Final Percent of Original Sample Sample Sample Respondinga Size California Prisons 49.4 (N = 307) 357 Jails 66.4 (N = 437) 437 Michigan Prisons 49.0 (N = 335) 422 Jails 71.5 (N = 373) 373 Texas Prisons 82.2 (N = 601) 601 figures do not include an even lower response rate of 34-35 percent for replacement respondents in California and Michigan prisons. SOURCE: Peterson et al. (1982:Tables 2, 10, and 11) selecting offenders with probabilities in- versely proportional to the total time to be served on their sentence. Because of the generally shorter terms served in local jails and, thus, lesser bias toward long terms in the resident population, the sample of jail inmates was drawn randomly from, or in- cluded all, convicted inmates residing in the selected jails on the sampling date. In comparisons of the attributes of the originally designated sample with those of an incoming cohort and resident inmates, the sample's characteristics were found to be generally closer to those of the incoming cohort than to those of the resident inmates in Texas and Michigan. This was not so in California, where only imprecise estimates of the total time to be served were available for prison inmates (Peterson et al., 1982:56 60~. There were some noteworthy differences between the simulated incoming cohort sample and an actual incoming cohort in all three states. Younger offenders were under- represented in all three scheduled sam- ples. Overrepresentation of young inmates among actual respondents in California and 3~] . Michigan, as noted above, decreased the differences in age between the actual re- spondents and an incoming cohort. Inmates with no prior prison terms were slightly overrepresented in the actual samples in California (67 percent vs. 63 percent in an incoming cohort) and Michigan (63 percent vs. 60 percent in an incoming cohort). In- mates convicted of robbery were over- represented in the actual sample in Califor- nia (38 percent vs. 30 percent in an incoming cohort), as were inmates con- victed of assault in the Michigan sample (15 percent vs. 10 percent in an incoming co- hort). Time served was apparently underes- timated for these offense types, which in- creased their representation in the sample.6 6This imprecision in time-served estimates, and especially the underestimate of time served by rob- bery inmates, for California inmates might have implications for the crime-reduction benefits esti- mated for selective incapacitation policies applied to robbery inmates (Greenwood, 1982~. Use of these underestimates of time served for policies would understate the crime reduction already achieved and overstate the reduction possible from increases

312 Because ofthese differences, the results for the inmate samples may not apply precisely to a cohort of incoming prisoners. Also, unlike in the first inmate sample, no attempt was made to adjust the results from the sample to estimate A for a sample of street offenders. Given the greater risk of imprisonment for high-rate offenders, the rates in the inmate sample will be higher than the rates that would be expected in a more representative sample of street offenders. A second and major change in the second survey was in the design of survey items eliciting frequency of offending. Instead of a uniform 3-year observation period, respon- dents were asked about 2 calendar years preceding their current incarceration. The arrest leading to the current incarceration identified the second calendar year and marked the end of the observation period, which varied from 13 months (for inmates arrested in January of the second calendar year) to 24 months (for inmates arrested in December of the second calendar year). Inmates were given a calendar on which to identify the exact months of the observation period. As in the first survey, months spent incarcerated or hospitalized during the ob- servation period were reported by the in- mate and subtracted to yield time at risk, or street months, during the observation pe- riod. The question eliciting counts of crimes committed was also much more sensitive to high rates of offending. For total counts of 10 or less, the respondent was asked to indicate the exact number of offenses com- mitted during his street months. For counts greater than 10, respondents were asked first to choose a category that best described the smallest unit of time in which offenses were committed (e.g., monthly, weekly, daily) and then to indicate the exact number of offenses committed in that unit of time. They were also asked to indicate the num in time served. However, this is apparently not a problem; the selective incapacitation analysis ap- pears to be based on the respondents' self-reports of their expected total time served and not on aggre- gate, imprecise averages available from the Depart ment of Corrections. CRIMINAL CAREERS AND CAREER CRIMINALS her of months during the observation period in which they committed offenses at that rate. While potentially enhancing the resolu- tion of rate estimates, the increased com- plexity of the frequency items greatly in- creased respondent problems in answering those items. The computation of individual rates, in tum, depended on the respon- dents' reports of street months, number of crimes committed, and number of months in which crimes were committed. The ex- tent of respondent confusion is reflected in the number of respondents who gave am- biguous answers to at least one of the three components. On the basis of a reanalysis of the data from the second inmate survey, Visher (Volume II:Table 11) indicates that 35 to 40 percent ofthe responses by inmates active in robbery or burglary were ambigu- ous. This widespread ambiguity in re- sponses was likely one of the reasons why minimum and maximum rates were com- puted for each respondent in the original analysis. Finally, a number of internal and external reliability checks were used in the second survey to assess the quality of the self- reports. Internal quality was judged by the consistency of responses to redundant items in the survey. External reliability was as- sessed only for prison inmates by compar- ing self-reported arrests and convictions with official-record data. Eleven percent of the respondents erred on more than 20 percent of the possible 27 intemal-qualit,v indicators. Almost 60 percent erred on more than 20 percent ofthe 14 extemal-reliability indicators (Chaiken and Chaiken, 1982a: Tables B.3 and B.7~. The indicators of internal and external quality were compared with various other self-reported attributes of the respondents (Chaiken and Chaiken, 1982a:22~226), and most attributes were found to be unre- lated to response quality. There were three notable exceptions, however: · Prisoners convicted of burglary and those with a self-image of being a "family man," "working man," or "straight" had good reliability.

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS · The responses of older respondents were of better quality than those of younger respondents, but older respondents tended to leave more questions blank. · For the most part, there were no differ- ences in response quality by race; black respondents, however, had substantially worse internal quality, especially on indica- tors of confusion and inconsistency. To assure that none of the results was driven by poor-quality responses, the re- searchers estimated frequency rates for both the total sample and a sample that excluded the 42 percent of respondents with poor-quality responses.7 There were no systematic patterns in the resulting fre- quency rates, which usually varied well within a factor of two in the two estimates. Estimates of ct and A for active offenders in the prison and jail inmate samples are shown in Table 9. As found in the first inmate survey, the As are highly skewed; most offenders reported committing no or only a few offenses and a small number reported very high frequencies-consider- ably higher than those reported in the first inmate survey. In the first survey, the 90th percentile rate for armed robbery was under 20 offenses per year of street time for active offenders. This contrasts with a 90th per- centile rate for robbery among California prisoners of 155 offenses per year in the second survey. The mean As in the second survey are seven and eight times larger than those reported in the first survey. These much higher As, based on offenses reported for smaller units of time (months, weeks, or even days in some cases), were foreshad- owed by the results in the first survey which yielded much higher As when esti- mates were based on the number of crimes committed in a "typical month." It is also noteworthy that A varied be- tween prison and jail inmates and across states. For serious offenses, A and ~ were generally higher for prison inmates than for 7The cutoff for poor-quality responses was set such that about 20 percent of Me respondents were excluded on the internal-quality indicators and 20 percent excluded on the external-quality indicators. 3~3 jail inmates. This suggests that judges were effectively distinguishing between high- and low-rate offenders in their sentencing decisions. Those decisions, however, are strongly influenced by an offender's crimi- nal record; according to official records, first offenders are more likely to be sentenced to jail than to prison (Blumstein et al., 1983~. To the extent that these first offenders in official records are actually just beginning their offending careers, their retrospective reports~in the survey will not reveal many offenses. If followed into the future, how- ever, they may reveal much higher rates of offending. Their retrospective rates, based on the survey, may be deflated partly by their being considered criminally active for the entire observation period. A prospective study of offending frequencies is required both to assess the stability over time of frequencies that have been estimated retro- spectively and to assess the degree to which current sentencing patterns distinguish among future high- and low-rate offend- ers. The differences in A across states parallel the differences in reported index crime rates (crime per 100,000 population) in the three states, which were highest in Califor- nia and lowest in Texas (Table 10~. These differences across states may reflect differ- ences in offending behavior across states. They may also reflect differences in the thresholds for incarceration in the different states, with incarceration being applied broadly in Michigan and Texas and re- stricted more narrowly to serious offenders in California. This latter interpretation is consistent with the differences in incarcer- ation rates across the states (Table 10~. In- carceration, especially in state prisons, was far less frequent in California than in either Michigan or New York. The negative rela- tionship between incarceration rates and As is also consistent with deterrent effects, i.e., individuals commit crimes at lower rates where sanction rates are higher. Distin- guishing between these competing hypoth- eses requires data on the relative participa- tion rates of high-rate offenders among inmate and street-offender populations in each state and on the offense-specific incar

314 a) H at: o a) V] - - U] a) ·,' a) ·,' a) o ·,' ·,' H ~1 - o ·,1 ·,' V ·,. Pi a' m ~ a) m ~ En up ~1 0 U) .,, P. ~5 a) X .,, En : i> is: ·~' o . C) ·,1 ton . - .,, u°' .,, _ _ _ Us Or O CO 0 11· ~11 o ~11 1 · ~ ~ 1· ~· 1· to Z ~ ~ ~ Z · ~ 0 UP Z . . ~ -~ _1 Dee ~ - ~ Cal - ~Cal - ~ - ~_ _ GO O ~in ° 11 · ~a' 11 0 A) ~11 o 1 . a' ~ 1 · a' ~ 1 . ~· t- ~ z · ~ a' ~ z· r~ ~ Z · ~ ~ ~ _ ~ c~ OD ~ _ ~ a~ - ~c~ _ _I _I d~ _ _ . u~ aD ~ 0 c~0 0 In 11 ·-IL9 11 · -I tD 11 o 1 · C~ o 1· ~1· (X)· Z · o ~ Z · ~ ~ ~ Z · ~ ~ d~ -~ d~ aD ~ _ tn ~(~} -C~ _~ ~r _ _ _ ~0 0 co 0 c~ o ~0 a, 11 . a' ~11 . un e, 11 o 1 · ~ a' 1 · 10 co 1 · CO · o Z · 0 ~C~ z · ~ ~1 ~ Z · _ ~ _I ~ _ [.n _I ~c~ _ ~ r_ _ _ _ 0 O c~ 11 · 1 ~11 . t~ ~11 o 1 · c° ~ `9 · o un 1 · ~· _I · oo ~ 0 z · un ~ Z · ao In _ ~ ~ ~ -oo ~·~ ~ ~ _I t~ ~a) .,. . - · - a) ~a, ~54 41 P. ~·= S ~·= ~ ~'= :~: .~' ~ o ~ =1 ·e ~ ° ~=1 ·e ~ ° U] ~q a) >d >, C O G) 0) tJ1 Q 5) ~1 ~I Q- ~ QJ U] O E ~

315 _ ~ oo o o o ~C ~ 0 ~0 ~0 ~co a, ~J4 11 · 1 a:' 11 0 o ~11. ~1 U ~11· d. ~11 o 1 · ~ 1- ~· O . ~1 . u' o 1 · ~ o 1 · In d. 2 · ~ '~ a:~ Z · 0 1 el' Z · 0 d ~_I z · _ U~ ~ ~-~ - ~ _' ~_' - d' _1 ~ -~ ' - ~ -~ '~ U~ ^ ^ o~ u ~O ID ~_ 0 0 ~O ~r 11. ~G0 tl o CO ~11. I1 - 11 0 ~kD 11 u-) I · o ~ 1· ~· 1. ~ t- ~· ~· 1· o ° Z · cc ~ u~ Z· ~ C ~ Z . ~ a' ~ z . ~ ~ z ~ ~ _ tg ~ aD _' -~ ~ ~_I - U~ ~ ~ _ ~ ~- ~ _ ~ ~ t- ~U~ tD ~CO ~1- a ~vo ~kg _ O ~o C ~o ~o O - 11 · ~11 · 1 _' 11 · 1 ~11 · ~d' 11 _1 1 · o Lc, 1 · 0 ~ d. · ~ ~ U~ · In ~ 1 · ~ ~ d' Z . a, t- ~ z . _, ~ z · ~ _' ~ z · d. ~-~ Z C~ ' ~ -~ ~ ~ -d' q~ ~_I -~ c~ ~ -~ ~ o0 ~ -~ ~ a' ~0 ~_ _ O co ~0 ~n a ~a _I ~\9 ~_I ~_1 1 O ~o C~ a) 11 . 1 ~11 o u ~a' 11 · 1 - ~11· ~° 11 lD ~ · O ~ ~· ~1 · 1 · ~ t- ~- · U~ a, 1 · ~ ~ o ; · ~ ~ O z . ~ a' U~ Z . c~ 0 u~ z · ~ 0 LO Z 0 _ a' u~ ~c~ ~ ~ L~ ~. - ~ Ln ~ ~ ~ ~ ~ ~ - C ~^ ^ ~L5 r ~a, ~0 as c ~a, ~O · - O ~O o C~ 11 ° · 1 ~11 0 - ~11 · 1 ~11 · ~u ~11 _1 1 · · er u ~. o · 1 · a, 0 kD · a, r~ 1 · kD o t~ Z ~ ~ ~ Z · ~ ~ Z · ~ ~ z · a' d~ _ ~ r_ ~c~ -kD a, ~·- -~o ~ ~ -d~ .- ~In -~ d ~a) ~a) ~ .,, .-, . - rl a S ~54 · - s ~· - s ~· - s ~ s ~· - s .. a) O a) ·e ~ O ~·e a) O ~·e ~ O ~·e ~ O ~51-~: a ~=1 ~ ~ ~=1^ ~ ~1- ~m ~=1- ~ £ s~ ^ a s O _ s ~5 ~ a s O a, ~ u s ~O

316 . - o 1 m A: ED x .,, .,, 0 s {Q C, ·,' as :~: ~ .- .^ 0 0 Cal P' In ha O ED or 11 1 · ~ ~ Z ~ O - - o ED CD 11 Cal 1 · ~ UP or :Z ~ ~ Go ~ - _ _ ao up cat ~Go Cal ~O us 11 1 co 11 o 1 · ao ~· cot a, Z ~ ~ Z ~- - -al ~- ~ cot d' - Cut ~ ~O to 11 1 co . ~ ~. as is ret d' ~ ~ -~ ~ Cut ~ - a, 11 1 · UP Z - - C~ O ~ UP or 11 1 · ~ ~ Z ~ Go us (D -d' 01 a, . - ·,l ,~ ·e ~ o ~ =1 - ~: ~ ~: ~l ~rl c) ~ x ~ ~ o.~ a ~ - - ~ o o E ~E~ - D 11 o 1 ~ o ~ z o -m - - ~ c~ ~ - ~ `9 u ~ u~ ~ 11 o 1 · ~ d' ~ z ~ 'm ao ~ ~ ~ ~ - o c~ ~ a, 11 0 1 . u~ o ~ ~ z ~ -m co - ~ ~ ~ ·e a, c) .m ~ o ~ ~: · 1 ~) S ~ :~ a, ) c ~ ~ ms ~ o ~m= - a) c) CQ Q. s ~ cn a tu s ~. - ee aJ ~ ~o ~ N O ~C) ~ ~0 ~' - - [Q O - ~O a' ~ ~ 0 ~ O ~ ~ ~ Q aJ -~ 3 ~0 G5 ~ tQ a) ·^ 0 s.^ 0 - S a) ~ ~bq ~ ~ =0 0 S ~O ~ ~tQ ~ O J ~4~ t) ·- 3) 3 a 3 ~n ~ ~ ~ ~ . - ~0 ~ O ~_ C) ~ ~a ~ o ~ ~1 ~0 ~(IJ tU IJ _I [Q O 0 3 S X · - ~ ~1 =e,1 0 ~0 0 0 ~L) U] t) ~ ~ Q O O ~ O ~ .- ~0 P ~e-1 ~ 1~S tJ tQ `: O ~ ~ ~ S ·,~ s ~a ~0 ~ ~ ~ ~ P' S 0 ~ ~ ~Q dV 81 ~J) `1 0 ' - Q ~ ~ ~ - ~ . - ~S ~ ~ ~·e J~ ~ .^ t) a, )-I ~ ,~, ~ to ~S ~ (L) 4,) ~co ~S ~n ~a) 0 ~ ~ ~ 0 _ ~u~ ~ a~ - a, - . ~ ~l t" Q ~ ~ · ~ ~ -~ O G) ~ = -a~ ~ 0 3 ~1 ~ (1)~ ~ aJ a, ~.~1 a) 0 s0 ~ ~ OQ ~ ~ ~ ~ ~ ~S Q~ ~ ~Q ~n c a O ~ E~ ~ ~ ~ ~ ~ S~ c ~ ~n~ ~ ~. a, Q ~a, ~ ~ `: · .L, m 0 ~ ~n ~ ~ ~ ~ ~ s Y O S ~ ~- O ~ c- - ~ ~ ~ ~ 0 - ~ O ~ ~ O S S ~ ~ s Q ~, ~ ~ ~c~ E~ 3 ~ a) ~o. ~q Q. . ~ ~ 0 ~ - a) ~ ~ s~ ~: a~ ~ ~ c, [,Q ·e .. a, u, ~ ~ ' - u'= E~ ~ ' ~ <' S ~ ~ ~ ~ 0 ~ os ~ ~ ~ ~ s ~O Z _~ E~ · ·\ U] QJ ~{lLI ~J a,1 ~ ~ ~ ~ S ~ ~.- ~ O S E~ U] ~O ~l ~

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS TABLE B-10 Variations in Levels of Crime and Sanctions Across Survey States 3~7 Indicator California Michigan Texas Median individual offending frequencies Robbery Prison8.0 5.7 3.2 Jail5.5 - 4.8 N.A. Burglary Prison9.8 6.2 3.6 Jail6.3 4.9 N.A. Reported index crime rate in 1976 (crimes per 100,000 population)b7,204 1978 incarceration rate 6,478 5,407 (sentenced inmates per 100,000 population) PrisonE 88 162 189 Jaild 58 28 10 2Chaiken and Chaiken (1982a:Tables A.3 and A.6). bFederal Bureau of Investigation (1977). CBureau of Justice Statistics (1980b). dThe total population and sentenced population in jails in each state are available in Bureau of Justice Statistics (1981a). The proportion sentenced was applied to the total rate of jail inmates per 100,000 population provided in Bureau of Justice Statistics (1979a). ceration risk per conviction or per arrest in each state. In identifying the correlates of A, the Rand researchers first characterized offend ers by the cluster of different offense types that they reported committing in the obser vation period. Excluding the 13 percent of respondents who reported that they did not commit any of the eight offense types stud ied, almost 90 percent of the remaining respondents fell within 10 clusters (Table 111. These clusters ("criminal varieties"), based only on the types of offenses commit ted, tend also to distinguish offenders by their offending frequencies. The more ver satile offenders, who committed multiple self-reports, the violent predators were offense types, also committed those of- distinguished from other olienders by Senses and others at very high rates. As their reported in Table 11, violent predators- who were distinguished by their reports of committing robbery, assault, and drug deals in the observation period not only com- mitted those offenses at higher rates than other offenders, but also reported commit- ting the most burglaries and had high rates for thefts. Official-record data of arrests or convic- tions for robbery, assault, and drug deals did little to distinguish the violent pre- dators from other olienders. Offenders who reported committing those offenses often had no official record for one or more of the offense takes. On the basis of Weir

318 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE Bell Variations in High-Rate Offending by Offender Types, Second Rand Inmate Survey Offender Type Violent predators (robbery, assault, drug dealing) Robber-assaulters Robber-dealers Low-level robbers Mere assaulters Burglar-dealers Low-level burglars Property and drug offenders Low-level property offenders Drug dealers Total Percent of Respondents in Offender Type (N = 1,777) 15.0 7.8 9.2 11.8 5.1 9.8 8.4 6.3 8.2 5.5 87.1 Annual A, 90th Percentile Robbery Burglary Assault Theft 135 65 41 10 516 315 377 206 __ 18 517 14 726 407 189 113 6 406 105 -- 97 663 9 __ 560 NOTE: The 13 percent of offenders who did not report committing any of the offense types studied are excluded from the total number of respondents reported here. SOURCE: Chaiken and Chaiken (1982a:Tables 2.5 and 2.20) · youthfulness, · extensive self-reported juvenile crimi- nal activity and drug use, · irregular employment, · use of multiple drugs as adults, and · lack of family obligations. In contrast to the violent predators, high- rate offenders among the less serious prop- erty offenders (fraud, forgery, credit card crimes) were distinguished by their · higher education levels, · being married, and · being recently unemployed. To assess the accuracy of prediction mod- els, the inmate sample was split, and one . half of it was used to estimate the model. The estimated model was then applied to the remaining half of the sample. On the basis of the results for the validation sam- ple, low-rate offenders can be predicted with high accuracy by using prediction models. The models did uniformly poorly, however, in predicting high-rate offenders. Well over half of those predicted to be high-rate offenders were in fact low-rate offenders, and a substantial portion re- ported no crimes at all of the predicted offense type (Chaiken and Chaiken, 1982a: Table 3.7~. Such high error rates are com- mon when attempting to predict such rare events as high-rate offending. (This is often referred to as the problem of low base rates.)

APPENDIX B: RESEARCH ON CRIMINAL CAREERS Some caution is also needed in general- izing results of predictive factors based on inmate samples to broader groups of of- fenders. As noted above, the distinguishing features of the violent predators included extensive involvement in drug use and se- rious crimes as juveniles and youthfulness (compared with other inmates in the sam- ple); The seriousness of juvenile involve- ment in crime and drugs for these young adult offenders may well be a distinguish- ing attribute of violent predators in prison, but be less decisive in identifying violent predators in a convicted sample. For youth- ful violent predators found among inmates, it is likely that their serious juvenile history was a key factor in the decision to incarcer- ate them (see, for example, Greenwood, Abrahamse, and Zimring, 1984~.8 Other young violent predators who do not have a serious juvenile history might be less likely to be incarcerated and, therefore, might be found in a convicted sample but less so in an inmate sample. This would reduce the role of a serious juvenile history as a distin- guishing factor for violent predators in a convicted sample. In general, to the extent that variables found to distinguish among inmates also influence the selection process that generates inmates, those variables may be less reliable as predictors of high-rate offending in non-inmate samples. As indicated earlier, because of ambigu- ous responses, the Rand researchers esti- mated a minimum and a maximum A for each offender. In the prediction analysis, the minimum and maximum estimates were averaged to yield a single estimate for the respondent. When the responses were un- ambiguous, the minimum and maximum estimates were identical. The reanalysis of sit is also noteworthy that the analysis of sentenc- ing patterns for young adults in Greenwood, Abr`ah`amse, and Zimr~ng (1984) did not find evi- dence that young adults are Heated more leniently than older offenders who are convicted of the s`ame offense and have similar records. This result con~a- dicts Me widely held belief that young adults win serious juvenile records are Heated win unwar- ranted leniency because juvenile criminal histories `are not considered at sentencing. 3~9 the inmate survey data by Visher (Volume II) indicates that these minimum and max- imum estimates represent extreme upper and lower bounds on A. For example, if an offender indicated he committed between 1 and 10 offenses in his calendar period, but failed to indicate the exact number of of- fenses, he was assigned 1 as the minimum A and 10 as the maximum. If he failed to indicate the number of months in which he committed a crime, he was assigned 1 as the minimum and his reported number of street months was used as the maximum. In cal- culating the minimum offense rate, the min- imum number of crimes committed was divided by the maximum number of street months. For the maximum offense rates, the maximum number of crimes committed was divided by the minimum number of street months. These estimating strategies tend to move the estimates to the most extreme pos- sible values, rather than to provide a reason- able estimate for the individual. Just as the mean of all offenders is sensitive to the pres- ence of a few individuals with very high rates, so the average of the minimum and maximum for a single individual is sensitive to very high maximum rates. Taking the average of these extreme rates is likely to overstate the true rate for an individual. Visher (Volume II) used an alternative strategy for dealing with the ambiguous responses. Rather than develop the extreme values possible for the estimate, she devel- oped a "best estimate" for each individual. For example, individuals who indicated 1 to 10 offenses but did not report the exact number committed were assigned a value from 1 to 10 based on the distribution of those values for unambiguous respondents. The particular value assigned to an offender could often be further pinpointed within the range 1 to 10 by the respondent's an- swer to a later categorical question that asked whether the inmate committed 0, 1-2, 3 5, ~10, or more than 10 offenses in the observation period. On the basis of this response, the ambiguous respondent was assigned an exact value within the selected category according to the distribution in that category found among unambiguous respondents.

320 Using the minimum-maximum strategy, the Rand estimate yields an average of 5.5 crimes committed t(1 + 10~/2] for all ambig- uous respondents in the 1-10 category. This frequency of 5.5 is one-third to two-thirds higher than the average number reported by unambiguous respondents. When the distribution-matching strategy is used, the average number of crimes committed for ambiguous respondents matches the lower average that is found among unambiguous respondents (Visher, Volume II:Table 8~. Similar estimating strategies were used for other ambiguous responses. The original Rand estimates and the estimates resulting from Visher's reanalysis for burglary and robbery are presented in Table 12. Visher's estimates are always much closer to the minimum estimate. In a highly skewed dis CRIMINAL CAREERS AND CAREER CRIMINALS tribution like that found for A, any individ- ual with an unknown rate is much more likely to have a rate that is lower, rather than higher, than the mean. Relative to the distribution for unambiguous responses, av- eraging the minimum and maximum esti- mates for each ambiguous individual overweights the higher rate and under- weights the more likely lower rate for the individual. Visher's "best estimate" ap- proach is likely to provide a more reason- able estimate of A for ambiguous respon- dents. Also shown in Table 12 are the offense rates found by Visher when only unambig- uous respondents are included in the anal- ysis. The clis~ibution is still highly skewed, with most active offenders reporting rela- tively low annual As. The distributions of TABLE B-12 Alternative Estimates of Individual Offending Frequencies (I) from Second Rand Inmate Survey Rand Survey Minimum ~ Maximum Visher Reanalysis Unambiguous All Cases, ~Cases, ~ Robbery 25th percentile 1.82.3 1.5 1.1 50th percentile 3.66.0 3.8 2.8 75th percentile 12.021.5 12.4 6.9 90th percentile 68.0100.5 71.6 43.2 Mean 40.662.2 43.4 36.5 (N = 594) (N = 294) Burglary 25th percentile2.42.82.01.8 50th percentile4.86.04.74.3 75th percentile23.335.023.417.5 90th percentile196.0265.0195.9158.4 Mean75.8118.679.055.7 (N = 824) (N = 451) NOTE: Frequencies are for active offenders in California, Michigan, and Texas combined. SOURCE: Visher (Volume II:Tables 9 and 11).

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS TABLE B-13 Alternative Estimates of Individual Offending Frequencies (A) from Two Surveys of Inmates in California Prisons 32] First Inmate Survey (1976) Second Inmate Reanalysis Incoming Survey of Second Cohorta (1978-1979)b Inmate SurveyC Robbery :1 Median(2.0)U 8.0 -5.1 Mean7.0e 49-74f 42.4 Burglary ha Median(3.0)" 9.8 6.0 Mean15.3 116-204 98.8 . . Three-year observation period (Peterson and Braiker, 1980:Tables 2 and 10a). bless than 2-year observation period (Chaiken and Chaiken, 1982a:Tables A.3 and A.6). CDerived from Visher (Volume II) reanalysis of original Rand data. These rates include both prison and jail inmates and so they are not directly comparable to the other rates for prison inmates only reported in this table. ~ edians are not reported for incoming prisoners in the original report. The medians are estimated here using the ratio of the mean to the median reported for the actual respondent sample in Peterson and Braiker (1980:Table 10a). eThe robbery rate reported in the first survey is for armed robbery only (mean of 4.61). Applying the proportion of armed robberies found among all reported robberies reported in national statistics--65.8 percent--(Federal Bureau of Investigation, 1974), the total robbery rate is estimated as 7.0. fThe range for the mean reflects the minimum and maximum estimates for each respondent. The median is based on the average of the minimum and maximum estimates for each respondent. rates at or below the median are quite sim- ilar. The rates for high-rate olienders and the mean of all offenders, however, are lower among the unambiguous respon- dents. This reflects the greater complexity, and thus greater risk of confusion or errors, in providing responses for frequencies greater than 10. Since ambiguous responses are more likely to occur for high-rate of- fenders, excluding ambiguous responses entirely would underestimate As. The various estimates of A are compared in Table 13 for California prison inmates in the two Rand surveys who were active in robbery and burglary. The rates for the second survey, based on minimum-max- imum estimates, are an order of magnitude higher than the rates in the first survey. Even using Visher's "best estimate" ap- proach, the mean rates from the second survey are five to seven times higher than the mean rates reported only 2 years earlier by prison inmates in California. Several factors may account for the siz

322 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-14 Distribution of Responses in Frequency Categories by California Prisoners in the Two Rand Inmate Surveys Offense Survey Type Sample . . . Percentage of Respondents in Frequency Category 0 1-2 3-5 6-10 10+ Burglary I 54 16 118 11 II 46 15 137 19 Car theft I 73 17 -53 3 II 75 11 65 3 Cons I 44 18 1216 11 II 44 17 116 21 Drug sales I 61 15 67 11 II 49 10 106 26 Forgery I 68 15 85 4 II 73 10 82 7 SOURCE: Chaiken and Rolph (1983:Table 5). able difference estimates from _ in the magnitude of the the two inmate surveys. First, the original inmate survey used a 3-year observation period (versus 13 to 24 months in the second survey). Providing an accurate count of total frequencies over such a long period was no doubt difficult for respondents, and many of the more distant offenses undoubtedly went unreported in the first survey. Second, recall problems were aggravated by the length of time in- mates had served between the observation period and administration of the first sur- vey.9 In addition to its more complex fre- quency items, the second survey also used the original categorical response set of the first survey. The responses to these repli- cated items provide one indication of the 9Recall that half of the respondents had served at least 2 years at the time of the first survey. More than half of the respondents in the second survey had served less than 1 year at the time of the survey. underreporting problem in the first survey. Despite the fact that the observation period in the first survey (median of 28 months of street time) was twice the length of the observation period in the second survey (median of 14 months of street time), the distributions over the frequency categories tor the two surveys were strikingly close (Table 14~. Reported offending increased markedly for only burglary and drug sales. The correlation between the distributions for the two surveys is .96. For most offense types, offenders reported the same relative frequencies, regardless of the length of the observation period. To the extent that these reported frequencies reflect only offenders' most recent experiences, more offenses will go unreported and frequency rates based on the entire observation period will be under- estimated as the length of the observation . . perloc . Increases. Differences in the length of the observa- tion periods also contribute to differences in the computed rates in another way. Rates

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS for active offenders were based on the re- ported frequency of offenses by offenders who reported committing at least one of- fense in the observation period. In the com- putations the offender was assumed to be active in a reported offense type over the entire observation period. This fails to ac- count for initiation and termination of activ- ity in an offense that may occur any time during the observation period. Failure to exclude these periods of inactivity will bias the rates for active offenders downward. This bias toward underestimates becomes more severe as the observation period gets longer. Rates based on the longer 3-Year observation period in the first survey are thus more likely to be underestimated than are rates in the second survey, which had a shorter observation period. All the above factors would lead to under- estimates of offense rates in the first survey. Correspondingly, there are reasons to ex- pect that the offense rates from the second survey are overestimated. As already dis . cussec , using a minimum-maximum strat- egy to estimate individual rates is likely to lead to inflated estimates. Even the lower rates obtained by the "best estimate" ap- proach of Visher (Volume II), however, are vulnerable to overestimation. One factor potentially leading to overestimates in the second inmate survey is the shorter obser- vation period combined with the reauire- ment that inmates had to report committing at least one crime of a given offense type to be considered active in that offense cate- gory. The estimated frequencies for active offenders were thus conditional on at least one reported offense; their unconditional rates while active would be lower. The overestimates in A arising from the required one offense become larger as the length of the observation period gets shorter. A minimum of one offense in a 3-year observation period contributes a minimum of.333 to the estimated annual rate. In a 1-year observation period, the estimated annual rate has a minimum an- nual frequency of 1. Restricting the re- quired offense for active offenders to avail- able street time, one required offense in only 1 available month of street time yields 323 a minimum annual frequency of 12. This overestimate, however, becomes less seri- ous as the number of crimes committed increases and, thus, is less likely to have a serious effect on the mean frequency rate. An estimated rate of 1.5 offenses in a 1-year observation period is clearly more vulnera- ble to overestimation than is an estimated rate of 10 offenses in the same 1-year obser- vation period. Certain design features in the items elic- iting frequencies on the second survey are more important contributors to overesti- mates of A. The second survey used fre- quency reports for smaller units of time for high-rate offenders to provide much greater resolution among rates at the high end of the frequency continuum. Reliance on smaller time intervals, however, may well lead to overestimates of rates when offend- ing is irregular over the entire observation periocl, i.e., spurts of high levels of activity are interspersed with periods of low levels of activity. Evidence of such spurts in of- fending behavior was found in self-reports to the National Youth Survey; for example, an annualized rate of 130 shoplifting of- fenses, based on a rate or ~ to ~ onenses per week, was found to represent only 20 shop- lifting offenses, all committed during 2 months in the summer (Elliott et al., 1983:117~. The higher rates in the second inmate survey were also foreshadowed in the first survey: the responses for burglaries committed in a "typical month" yielded a median frequency rate of 4 burglaries per month active whereas the responses for to- tal burglaries during all reported street months within the 3-year observation pe- riod yielded a median rate of only .24 bur- glaries per month active (Peterson and Braiker, 1980:26~. To the extent that the frequency re- sponses for short time intervals represent rates in periods of peak activity and not an average of high- and low-rate periods, ex- trapolating from peak rates in short periods may seriously overestimate annual rates. The magnitude of the bias will depend on the difference between high and low rates and on the relative proportion of total time free devoted to peak-rate activities. The

324 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-15 Adjusting Offending Frequencies (A) from the Second Rand Inmate Survey for Spurts in Activity Adjusted Mean Offense Type Offending Frequenc Change from Original Frequency Estimateb Minimum Mean Percent Change Frequency Maximum Mean Percent Frequency Change Robbery 3949 -20.674-47.4 Assault 7 57.1 +5.67.6-1.3 Burglary 115116 --1.2204-43.8 Theft (other than vehicle) 160185 -13.7326-51.0 Motor vehicle theft 2938 -24.5102-71.9 Forgery 5362 -15.394-44.1 Fraud 48156 -69.2202-76.2 Drug deals 850927 -8.31,681-49.4 Lithe mean rates are adjusted downward to reflect the offender's frequency over spurts in activity and quiescent periods (Chaiken and Ralph, 1985:Appendix). ~ inimum and maximum frequencies were estimated for each active offender in a sample reflecting a cohort of incoming inmates to California prisons (Chaiken and Chaiken, 1982a:Tables A.3-A.14). larger the difference between high and low rates and the smaller the fraction of peak periods in total time free, the larger will be the overestimate. Chaiken and Rolph (1985:Chapter 2) ex- amined the elect of offending spurts on A in the second survey and found that much of the criminal activity reported by the respon- dents occurred near the time of their cur- rent incarceration. Moreover, the duration of the periods of criminal activity and the counts of crimes reported did not increase as the duration of street time available to respondents increased. Frequency rates for offenders with very short street times were thus especially vulnerable to overestima- tion because of spurts in activity during those shorter observation periods and the absence of corresponding data for quiescent periods. To account for these spurts of ac- tivity, Chaiken and Rolph adjusted the esti- mated rates downward to reflect an estimate of A averaged over active and quiescent periods. As indicated in Table 15, the ad- justments resulted in about a 15 to 25 per- cent reduction in the original minimum frequency estimates for most offense types and a 40 to 80 percent reduction in the original maximum frequency estimates. Mean As for California prison inmates, for example, decreased from a maximum of 204 burglaries per year free to a mean annual rate of 115 burglaries; mean annual robbery rates decreased from a maximum of 74 to 39 after adjusting for offending spurts. Only the mean As for assault were unaffected by the adjustment for short-term spurts in ac- tivity. Even with the adjustment for spurts, a few individuals were estimated to commit crimes at very high rates an average of one or more crimes a day. The mean A is very sensitive to these few very high-rate of- fenders and is thus vulnerable to serious overestimation arising from errors in the estimated rates for these individuals. To reduce the impact of these high-rate of- fenders, mean As can be reestimated by identifying an upper threshold value for individual frequencies and assigning that value to all offenders whose estimated rates exceed it. The frequency estimates from the

APPENDIX B: RESEARCH ON CRIMINAL CAREERS Visher (Volume II) reanalysis ofthe original inmate survey data can be used to illustrate the change in the mean A when different thresholds are used (Table 16~. When the 90th percentile is used as a threshold for robbery, for example, the upper threshold value is 107.1 robberies per year for Califor- nia inmates. The 10 percent of active rob- bers with As above that threshold each com- mittec] an estimated average of 300.4 robberies per year. When the 90th percen- tile value of 107.1 is assigned to these high- rate robbers, the mean A decreases from 42.4 to 21.8. When the 90th percentile is used for burglary in California, the mean A is similarly reduced by half, from 98.8 to 44.6 burglaries annually. Although various features of the estima- tion procedures used in the two surveys contribute to underestimates of frequency rates for the first survey and overestimates for the second, it is also possible that the relative differences in rates between the two surveys reflect real differences in the population bases used in the two estimates. In the first survey, participation is based on reporting at least one offense in a 3-year observation period. This contrasts with at least one offense in half the time in the second survey. The longer time window of the first survey increases the likelihood that low-rate offenders were included among active offenders in an offense type. Low- rate offenders were less likely to be de- tected as active during the shorter observa- tion period of the second survey. The different lengths of the observation period thus may have affected sample composition for active offenders in the two surveys. A greater representation of low-rate offenders in the first survey would lower the mean offense rate from that survey compared with the mean obtained from the second survey. Estimating Individual Frequency Rates from Arrest Histories In addition to self-reports of crimes, offi- cial arrest histories have also been used to estimate offending frequencies. Two such estimates are reviewed here. One is based on the longitudinal arrest histories of all 325 adults arrested for murder, rape, robbery, aggravated assault, burglary, or auto theft in Washington, D.C., during 1973 (Blumstein and Cohen, 1979~; the other, on adult ar- restees in the Detroit Standard Metropoli- tan Statistical Area (SMSA) cluring 197~1977 (Cohen, 1981, 1983~. (Data for both studies were drawn from computer- ized criminal history files maintained by the Federal Bureau of Investigation.) In Table 17 the demographic attributes of arrestees in the two study sites are compared with those of persons arrested for index offenses (excluding larceny) in other U.S. cities. The two arrestee populations differ markedly with respect to race. The Washington, D.C., arrestees, who mirror the unique racial composition ofthat city in the early 1970s,~° are not representative of arrestees in other cities, but the arrestees in the Detroit SMSA match more closely the racial composition of urban arrestees found nationally. Be- cause the study populations are restricted to adult arrestees in each jurisdiction, juve- niles (under age 18) are unclerrepresented in the two arrestee populations. Among adults only, the arrestees are closer in age to arrestees in other U.S. cities, although those aged 1~24 are somewhat underrepre- sented.~ Wig respect to sex, the arrestees are almost exclusively males. The differ- ences between Me two jurisdictions pro- vicle a basis for assessing the degree to which individual arrest and offending fre . . . . . quencles vary across JU0SC beckons. The original populations in bow study 10In the 1970 census, the population of Washing ton, D.C., was 71.1 percent black, compared with 12.3 percent black for the total urban population of the United States (Bureau of the Census, 1981:Ta- bles 15 and 24~. iiThis difference for age is not accounted for by similar differences in the general population. In the 1970 census, for example, about 20 percent of the adult population (age 18 or over) in Washington, D.C., and in all urban areas of the United States were 18-24 years old. Analysis of the arrestee data, however, indicates that multiple arrests in a year are more common for young adults than older adults. This would inflate the representation of young adults in the count of arrests in the national data for cities.

326 REV o s U] a) ~4 EN a, · - U) Q H a) ~Q ·~, REV at: ~ ^ ~ REV U] ·,' a, ~ k ~ O H U] a) P ~ .,' O ' - ~ Al U] REV O ~ o to t - S.l Q. U. - U] _ O v at: 1 A :t ED h4 O 1 m ~Q E3 m Em ~ O m~ ~ ~ · · · . · . In ~ ~ 0 co CO ~ ~ ~ ~ a, 0 ~ao ~ ~ ~ ~ 0 · · · · · · · · · · · · ~ tD US ~ O O tD at ~ 0 a, ~ tD LO ~ ~0 ~ ~ ~ U~ ~ U~ ~ - 00 a) ~ ~ a) a . - . - H a, ~ ~ ~ ~ v ~ ~ ' - ~ ~ ~ ~ ~ L4 a, P4 ~ P~ P4 a) ~ ,` S S S S S U] :' ~ a' ~ 0 ~ 0 r1 <: ~ a' a' oo CD i~ ~ 0 co ~ ~ ~ _I· e · · · ~ U1 ~) ~ t- ~ d' (~ ~ X 11 a) z E-'- O ~C~ . ao Z ~ ~ . - ~ c: ~_ _ u ~ ~ ~ 0 cn 0 kD O ~ ~ ~ ~ ~ ~ ~ ~ ~ O ~ U~ ~ CD - - 11 d' ~ Z o_ ~~ ~ ~ ~ ~ ~ ~ ~ == · . ~ . ~ . . cn a' ~ ~ ~ ~r ~ ~ u~ ~ ~d. Q G) 11 Qo S Z a: E-'- r~ co ~ ~ cn 0 c~ ~ ~ ~ 0 0 · ~ ~ ~ ~ ~ ~· ~ ~ ~ ~ ~ O ~ m0 ~ O ~q ~ ~ ~ _I a~ ~ 11 ~ Z E~- _ tD ,. S 11 Z - .,' _ O ~ 11 _I ~ Z C.)- U] ,~ _ U] a) 11 ~ S Z a, E~- ~ ~ O O ~ ~ · · · · · - O U, ~ `9 u~ a~ ~ ~ 0 · · · · · ~ ~ ~ ~r co 0 cn ~ CO a' C~ ~ tn a 0 .,' ~a ao ~ · . . .~ . ,. t~ ao ~ ~ C~l d' d' ~ ~·, U, O S O a' ~ ~ 0 a' cn .... ~ . ,' ~1 0 ~ _I tQ CD ~ ~ d' C~ C~ · · · · · ~ ~ ~ ~ ~ O U~ ~ ~ U~ ID ~r ~ 0 ~ ~ ~ ~ a' · . · · · ~ ~ d. O ~ ~ U~ 0 0 c~ ·n . . . . · . . . · . u~ ~ ~ ~O a~ a~ e ao ~ u~ ~ a, O d. ~ ~kD ao u~ a) a, a a) ~ ~ ~ v v v v v v L4 a) ~ ~ a' ~ ,< P4 P4 ~ ~ ~ ~ s ~ s s s s s s JJ a~ a~ u~ O u~ O u~ ~ ~ a' ~ ~ a' s E~ - H g _ ·,' :> S ,' G) ~J O aJ ~ a _~ ~1 . - .,1 ~ . - Ql a) ~ ~ ~ J~ ,, a) ~ ~ ~ ~ ~ ,~ V V V C) V V _ s~ ~ ~ ~ 54 ~ - a.~ a) a~ a~ ~ a ''' V P4 P4 n. ~, ~ P] ~C ~ S~ S JsJ S S S a~ ~ 0 ~ 0 ln as a~ o~ 0 00 ~ ~5 o a) .,' s~ a C] ·. E~ o z

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS 327 TABLE B-17 Comparison of Attributes of Arrestees Nationwide with Arrestees in Washington, D.C., and Detroit, 1973-1977 1973 UCR Index Arrests 1973 1974-1977 for Cities Washington, D.C., Detroit SMSA (excluding larceny)a Adult Arresteesb Adult Arresteesb Total Persons ~ 18 (N = 5,338) (N = 18,635) Attribute (percent) (percent) (percent) (percent) Race Sex Age White55.3 8.1 - 56.8 Nonwhite44.7 91.8 43.2 Male92.5 89.7 95.5 Female7.5 10.3 4.5 <1842.1 0.1 21.0C 18-2017.2 29.718.6 23.6 21-2414.9 25.724.4 23.1 25-299.9 17.119.9 15.1 30-345.7 9.812.3 7.3 35-393.5 6.08.4 3.8 40-442.5 4.35.0 2.3 45-491.7 2.94.6 1.7 >502.5 4.36.7 2.1 National data for cities are based on arrests and not arrestees. Persons with more than one arrest in a year are counted more than once in the arrest data (Federal Bureau of Investigation, 1974). tBlumstein and Cohen (1979:Table 1) and Cohen (1981). Sage in 1974 is reported here. The large number of arrestees under age 18 includes many persons who enter the data because of later arrests as adults during the period 1975-1977. sites numbered several thousand 5,338 in Washington, D.C., and 18,635 in the Detroit SMSA. The analyses of offending frequen- cies, however, focus on selected cohorts within these cross-sections of arrestees. Co- hort subsamples were used so that changes in frequency rates could be analyzed over time for the same arrestees rather than a changing mix of arrestees in the cross- sections.~2 Cohorts were defined to include Comparisons over age, using the cross-section of arrestees, for example, include different subsets of arrestees at different ages. Frequency rates at age 18 are based on the histories of both young and old individuals from the original study popula- tions who reached age 18 in the same year and whose first arrest as adults occurred at ages 1~20. These criteria ensured that co- hort members were the same age and that they were active in criminal careers as adults before age 21. The resulting fre- quency rate estimates were thus based on the arrest experiences of offenders who had arrestees in the sampling year, while frequency rates at age 30 are restricted to older arrestees in the sampling year; arrestees under age 30 in the sam- pling year will not enter frequency rate estimates for age 30.

328 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-18 Sample Sizes for Cohorts of Arrestees Used to Estimate Frequency Rates from Official Arrest Histories Detroit SMSA Washington, D.C., Adult Arrestees, 1974-1977 Offense Type Adult Arrestees, 1973 Whites Blacks Robbery181 44 120 Aggravated assault ;72 75 65 Burglary 153 - 126 133 Larceny 167 123 146 Auto theft 100 64 64 Weapons 80 48 67 Drugs 110 69 78 All others 303 258 233 NOTE: Four separate cohorts were identified in each study site: those aged 18 in each of the years 1963-1966 among Washington, D.C., arresters, and those 18 in 1964-1967 among Detroit SMSA arresters. The numbers reported in this table are for the four cohorts combined; the numbers in each cohort are smaller. SOURCE: Blumstein and Cohen (1979:Table 7) for Washington, D.C.; sample sizes for the Detroit SMSA are available from the private files of the author. at least two arrests, one in the sampling year and one at ages 18, 19, or 20. This restric- tion, combined with the further require- ment that the arrest in the sampling year be for an index offense other than larceny, limits the analysis to frequency rates for reasonably serious adult arresters. As is evident in Table 18, the focus on cohorts within an original annual cross-section of arrestees considerably reduces the sample size available for analysis. When cohorts are combined, offenders active in individual offense types (i.e., at least one arrest for that offense during their arrest histories) num- ber under 200. Sample sizes are even smaller for individual cohorts. Estimates of ,u. Individual arrest fre- quencies, ,u (mean arrests per offender an- nually), were estimated for the cohorts in Washington, D.C., and in the Detroit SMSA. Only the period from age 21 (after the first arrest as an adult) to before the sampling year was used in the analysis. The requirement of an arrest prior to and then another arrest after the estimation period was intended to ensure- with reasonable certainty that arrestees were criminally active throughout the estimation period. The required arrests at either end of the estimation period were excluded from the analysis, and time spent incarcerated was excluded from the time at risk of arrest in the estimation periocl. The mean Us estimated for adult arrestees in their 20s between 1966 and 1973 are reported for the two jurisdictions in Table 19. Arrest frequencies were lowest for auto theft and aggravated assault; offenders who were active in those offenses averaged about one arrest every ~7 years of street time. The highest arrest frequencies were found for drug offenses and the residual category of"all others"; while free in the

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS TABLE B-l9 Mean Individual Arrest Frequencies (I) from Official Arrest Histories 329 Black to Washington, D.C., Detroit SMSA Adults White Offense Type Adults Whites Blacks Ration Robbery 0.23 0.13 0.23 1.74 Aggravated assault 0.19 0.18 0.18 0.97 Burglary 0.26 0.-18 0.22 1.19 Larceny 0.27 0.13 0.29 2.13 Auto theft 0.14 0.13 0.15 1.15 Weapons 0.22 0.22 0.19 0.89 Drugs 0.32 0.31 0.24 0.77 All others 0.40 0.38 0.38 0.79 Any index 0.5 ~0.25 0.39 1.54 Index 0.4 ~0.24 0.30 1.24 (excluding larceny) Total 1.0~: 0.55 0.56 1.01 (excluding traffic) NOTE: Arrest rates, arrests per offender per year free, were estimated by offense type for only those offenders with at least one arrest for that offense some time during their arrest histories. Only arrests prior to the sampling years and after the first arrest as an adult were considered. When computing ps for cohorts among 1973 arresters in Washington, D.C., only the most serious charge was considered (Blumstein and Cohen, 1979); for cohorts among 1974-1977 arresters in the Detroit SMSA, all charges recorded for an arrest were counted (Cohen, 1981). This will inflate the rates for Detroit arresters somewhat compared with those for Washington, D.C., arresters. However, since only 10 percent of Detroit arrests involve more than one charge, the rates are reasonably comparable. Ache ratios were computed before ps were rounded to the two significant figures reported in this table. ~ Crests for murder and rape are excluded from index rates in Washington, D.C. The resulting underestimates of index frequencies are likely to be small since in 1973 murder and rape constituted only 7.3 percent of all adult arrests for index offenses. Also, reported index rates in Washington, D.C., are simple averages of index rates found for the five offender types in this table who have at least one index arrest (i.e., robbers, aggravated assaulters, burglars, larcenists, and auto thieves). For index rates excluding larceny, both arrests for larceny and larcenists are excluded from the simple average. SReported ~total" rates for Washington, D.C., are simple averages of the "total" rates found for each of the eight offender types identified in the table.

330 community, these active offenders averaged one arrest every 2.5 to 3.5 years. Offenders active in robbery and burglary averaged one arrest every 4 years; white offenders in the Detroit SMSA had somewhat lower fre- quencies for these offense types. There are striking similarities in offense- specific Us, both between races in the same jurisdiction and across jurisdictions. The largest differences are between whites and blacks, but even these are small. The ratio of black to white frequencies in the Detroit SMSA is very close to 1 for most offense types. The largest differences, of about 2:1, are found for robbery and larceny. Also, frequency rates for Washington, D.C., ar- restees who were predominantly black are very similar to the arrest frequencies of black offenders in the Detroit SMSA. These racial differences in arrest frequencies, however, do not come close to the ratios of 10:1 or even 5:1 observed for robbery and other violent offenses in aggregate inci- dence rates of arrests per capita (see Table 1-2, this volume). The substantial racial differences found in aggregate incidence rates appear to reflect primarily the differ- ences between the races in participation in crime. Among active offenders, frequency rates are more similar for black and white offenders. Mean arrest frequencies were also ana- lyzed for variations with the offender's age, number of prior arrests, and cohort mem- bership. Individual arrest histories were broken down into separate observation years, each characterized by the attributes of offenders at the start of that year. Mean arrest frequencies were then estimated by aggregating all observation years with the same attributes. Multivariate regression analysis on the resulting mean Us did not generally reveal significant trends in of- fense-specific rates, either as offenders got oilier or as they accumulated additional ar- rests. There was, however, some indication of a cohort effect among Washington, D.C., arrestees, i.e., ,u was higher for more recent cohorts, but a similar cohort effect was not found in the Detroit SMSA data. The general stability over time within cohorts contrasts with other observations of CRIMINAL CAREERS AND CAREER CRIMINALS strong effects in the full cross-section of arresters: arrest frequencies systematically declined with age and increased as arrests accumulated for adult offenders. These op- posite results suggest that the variability in frequency rates observed in a cross-section arises predominantly from heterogeneity in rates for the different mixes of offenders compared within a cross-section. Arrest fre- quencies appear to be more stable over time within the same cohort. The results within cohorts, however, must be regarded as pre- liminary because of the small samples of arrestees on which they are based and be- cause of the limited number of years ex- tending only through the 20s for adult of- fenders that were examined. An important factor potentially affecting the accuracy of estimates of ,u is incomplete arrest history data. Failure to record large numbers of arrests in centrally maintained criminal history repositories, for example, could lead to serious biases in estimates of arrest frequencies. This possibility was highlighted by the findings of recent audits of criminal history files. In checking sam- ples of arrest records maintained by various local police deponents and courts in Michigan, the Michigan State Police (1983a, b) found that 21 to 47 percent of felony arrests were not recorded in the cen- tral criminal history files. A similar audit by researchers at the Police Foundation (Sher- man and Click, 1984) found both under- and overcounts in arrest histories for arrests appearing in local records. Nonrecording rates during the sampled years in the Detroit SMSA and Washington, D.C., studies varied substantially (Table 20~. In the Detroit SMSA, under half of the arrests reported in local police statistics were recorded in the central arrest history data, but in Washington, D.C., recording in arrest histories was almost complete. The low recording rate in the Detroit SMSA was due mainly to a police policy of forwarding to the central repository only arrest reports with known dispositions. Other factors that may affect police reporting to central repos- itories are the extent to which local depart- meets are linked routinely to the central repository. Lack of geographic proximity,

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS TABLE B-20 Extent of Incomplete Arrest History Data: in Arrest Histories to Arrests in Police Reports 33] Ratio of Arrests Offense Type Washington, D.C., Adults in 1973 Detroit SMSA Adults in 1974-1977a Robbery .97 .43 Aggravated assault .91 .40 Burglary .94 47 Auto theft .99 - .44 NOTE: The complete census of arrests for adults available in the FBI's computerized arrest history data for Washington, D.C., and the Detroit SMSA for the sampling years was compared with local police reports of arrests in the same jurisdictions and years in State of Michigan 1975 Uniform Crime Report and State of Michigan 1977 Uniform Crime Report for the Detroit SMSA, and in the Annual Report of 1973 and Annual Report of 1974 of the Metropolitan Police Department in Washington, D.C. lithe numbers of adult (aged 17 or older) arrests for the Detroit SMSA are estimated separately by crime type by applying the adult proportion of arrests reported statewide to local statistics. absence of administrative ties, and heavy reliance on locally maintained records could all contribute to reduced reporting to a central repository. Incomplete arrest history data can lead to complex biases in estimates of,u. When substantial numbers of arrests are not re- corded in arrest histories, the number of arrestees found in the arrest history data for a sampling year will be understated, as will the total number of arrests that year. Some individuals with only one arrest in the sam- pling year will be missing from the arrest history data, as will some individuals who had several arrests that year. Since failure to record arresters is more likely for offenders with only one arrest, low-rate arrestees, with their smaller expected number of ar- rests, are also more likely to be missing from samples based on incomplete arrest history data. This would lead to overestimates of,u. But incomplete recording also understates the number of arrests for arrestees who are sampled, which contributes to underesti mates of ,u. The ratios in Table 20 reflect both nonrecording of arrests and under- counting of arrestees in the arrest history data. The accuracy of arrest-frequency esti- mates will therefore depend on the relative strengths of these two opposing potential sources of bias. Aside from the potential errors arising from incomplete arrest history data, ar- rests even when fully recorded are only a sample of all crimes committed by of- fenders, and so arrest frequencies will in- clude only a portion of total offending fre- quencies. Nevertheless, estimates off, like those in Table 19, can be used to generate estimates of A. If the probability of arrest for a crime, q, is independent of A, then the mean ,u is just the product of the mean individual crime rate and the risk of ar- rest per crime, ,u = Aq. Using this relation- ship with estimates of the probability of arrest per crime, ,u can be transformed into A. Assuming a single arrest risk for all of

332 fenders in an offense type, Blumstein and Cohen (1979) and Cohen (1981) used aggre- gate data on the number of reported arrests, Ai, divided by the number of reported crimes for offense type i, Ci, to develop offense-specific estimates of the probability of arrest for a crime. This simple ratio was then adjusted by the offense-specific rate at which victims report crimes to the police, ri, to account for unreported crimes among crimes committed. Another offense-specific adjustment was macle to account for arrests of multiple offenders for the same crime incident, Oi, that are included among re- ported arrests. The final estimate of the probability of arrest per crime for offense · . type ~ Is qi= (A,,/Oi)/(C'/ri). Reported arrests and crimes, Ai and Ci, are generally available in local police re- ports. Victim reporting rates, ri, are some- times available for specific jurisdictions; if not, national data are available annually (e.g., Bureau of Justice Statistics, 1982a). Estimates of the average number of offend- ers per crime incident, Oi, are available from national criminal victimization data in Reiss (1980b).~3 Estimates of the arrest risk per crime for different offenses in Washington, D.C., and in the Detroit SMSA are reported in Table 21. The probability of arrest per crime is highest for aggravated assault, which prob- ably reflects the direct confrontation be- tween offender and victim in this offense and the high proportion of offenders who are known to victims~6.5 percent in 1980 13The bias due to including multiple offenders per crime in reported arrests was first pointed out in Shinnar and Shinnar (1975~. The average number of offenders per crime incident, Oi, is derived from crime incidents in which the number of offenders is known and is reported by victims in the victimiza- tion surveys. The adjustment used here thus in- volves the assumption that the number of offenders per crime incident is not substantially different for crime incidents in which the number of offenders is not known. The adjustment will overstate Oi and understate qi if the number of offenders is more likely to be known in crime incidents involving multiple offenders. CRIMINAL CAREERS AND CAREER CRIMINALS (Bureau of Justice Statistics, 1982a:Table 51~. Arrest risks per crime for other offense types are generally under 5 percent. These estimates of q by offense type, which are based on aggregate data, are similar in mag- nitude to estimates developed from self- reports of arrests and crimes by prison in- mates (Table 3-2; Peterson and Braiker, 1980:Table 2; and Petersilia, 1983:Table 4.41. - Estimates of A. The estimates of q in Table 21 were combined with the estimates of ,u in Table 19 to obtain estimates of A (Table 22~. Within any single offense type and jurisdiction, the same arrest risk, q, was applied uniformly to all active offenders. As indicated in Table 22, mean As for adult arrestees in Washington, D.C., and the De- troit SMSA were generally similar in mag- nitude for most offense types. Frequencies were lowest for offenses involving actual or threatened violence; offenders active in ag- gravated assault were estimated to commit an average of two to three of these crimes annually, and offenders active in robbery were estimated to commit three to five rob- beries per year free. Individual frequencies for property crimes were generally higher at five or more crimes per year free. The largest difference in As was found for auto Den; A, on average, was three auto thefts per year free in Washington, D.C., com- pared win nine in the Detroit SMSA.~4 Within suggests an interesting hypothesis about the potential influence of different criminal oppor- tunities reflected by differences between jurisdic:- tions in the availability of targets-on A. Like the differences in individual auto theft frequencies for active offenders, reported auto thefts per 100,000 population were also lower in Washington, D.C., than in the Detroit SMSA. While the reported index crime rate in Washington, D.C., for 1973 (6,949 index crimes per 100,000 population) was 94 per- cent of the corresponding index crime rate in the Detroit SMSA for the years 1974-1977 (7,411 index crimes per 100,000 population), the reported auto theft rate in Washington, D.C. (642), was only 63 percent of the rate in the Detroit SMSA (1,014~. Exploring this hypothesis further would require, at a minimum, data on the numbers of registered and newly manufactured vehicles in each jurisdiction.

APPENDIX B: RESEARCH ON CRIMINAL CAREERS TABLE B-21 Estimates of the Arrest Risk per Crime (I) by Offense Type 333 Offense Type Washington, D.C.a Detroit SMSAt Robbery .069 .043 Aggravated assault .111 .062 Burglary .049 .038 Larceny .026 .030 Auto theft .047 .015 Weapons: (.056) - (.051) Drugs (-040) (.068) All others (.038) (.028) ~Blumstein and Cohen (1979). tCohen (1981). -The probability of arrest per crime, a, is only roughly approximated for the predominantly victimless offenses of "weapons, n Drugs, and "all other" for which the number of reported crimes, C;. is very close to the number of reported arrests, A.. _~ Commission of these crimes typically goes unreported unless they~are discovered by the police, and when discovered by the police, they usually result in arrest. The rate of reporting these offenses to the police, pi, is arbitrarily set at one-quarter of the average reporting rate for all offense types in the victimization surveys, and the average number of offenders per crime incident for all offenses in the victimization surveys is used for Of. When various offense types are com- bined, active offenders are estimated to commit 9 to 13 index offenses and 15 to 26 total offenses annually. The As for individ- ual offense types are similar in the two jurisdictions, but total frequencies are higher in Washington, D.C., than in the Detroit SMSA. This reflects official records that include more diverse offending in the former than the latter. When 12 major of- fense categories are considered, arrestees in Washington, D.C., averaged 2.7 crime types in their arrest histories, compared with 2.1 crime types for both black and white ar- restees in the Detroit SMSA.~5 Several factors can affect the accuracy of A when it is estimated from arrest history data. i5The 12 offense categories are murder, rape, robbery, aggravated assault, burglary, larceny, auto theft, weapons, drugs, stolen property, fraud, and a residual category for all other offense types. In addition to the problems of incomplete arrest data, discussed above, biases may arise from potential errors in the offense- specific estimates of the arrest risk per crime, qi. Reported crimes, Ci, and reported arrests, Ai, are subject to nonrecording er- rors. Likewise, both the reporting rate by victims, ri, and the multiple-offender rate for crime incidents, Oi, are subject to recall and other response errors by respondents to victimization surveys. The multiple-offend- er ratio is also vulnerable to errors if the number of offenders varies systematically between crime incidents for which the vic- tim knows (and reports on victimization surveys) the number of offenders and those for which the number of offenders is not known or not reported. In view ofthe various potential sources of errors in estimates of q developed from aggregate data, it is advisable to explore the sensitivity of estimates of A to reasonable

334 CRIMINAL, CAREERS AND CAREER CRIMINALS TABLE B-22 Mean Individual Offending Frequencies (I) Derived from Official Arrest Histories Washington, D.C., Detroit Offense TypeAdultsa SMSA Adult Robbery3.4 4~7 Aggravated assault1.7 2.9 Burglary5.7 5~3 Larceny10.9 7.3 Auto theft3.0 9.3 Any index13.2C 8.7d Index (excluding larceny)7.4c 6.7d Total (excluding traffic)25.9c 14.6d - anlumstein and Cohen (1979:Table 19). erived from data in Cohen (1983:Table 16). CIn Washington, D.C., rates for index offenses do not include murder and rape. Also, the n index" rates reported here are simple averages of the "index" frequencies found in Blumstein and Cohen (1979:Table 19) for the five offender types with at least one index arrest (i.e., robbers, burglars, aggravated assaulters, larcenists, and auto thieves). Similarly, total frequencies are a simple average of n total" rates found for all offender types examined in the same table. ~ Order and rape are included in index frequencies for the Detroit SMSA. In computing ~ for "index" and "total" offense types, the arrest probabilities for individual offense types are weighted by the distribution of offense types found in the aggregate. variations in q. First, a generous range on the possible values of the various compo- nents that enter the estimates of q must be established (Table 231. Using a typical ar- rest frequency value, ,u = 0.2, the variations in estimates of q and in the associated esti- mates of A can then be explored. Invoking logical and empirical constraints on the var- ious component values, the ranges explored are .1 to .3 for the ratio of A to C, .25 to .75 for r, and 1.25 to 2.5 for O. The resulting range in the estimates of q is large, .010 to .180 (a low of 1 arrest for every 100 crimes and a high of 1 arrest for every 5.6 crimes). The q estimates vary sixfold as O and r are varied across the columns of Table 23, and threefold as the ratio A:C is varied across the rows. The impact of these variations in q on estimates of A can be substantial. At the highest value of q, A is just over 1 crime committed annu- ally; at the lowest value of q, by contrast, A averages 20 crimes committed annually. For the most typical value of q available in current estimates, .05, the range of A is narrower from 1.7 crimes annually at the maximum value of q to 10 crimes per year at the minimum value of q. The actual value of A is likely to be found in an even narrower range between these worst-case extremes. Thus, the variation in estimates of A as a result of errors in estimating q is likely to be within a reasonable range. Another factor potentially distorting the estimates of A derived from Us is the reli- ance in existing estimates on a single of- fense-specific qi for all offenders in that offense. In these estimates, all arrestees for

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS TABLE B-23 Sensitivity of Offending Frequencies (a) to Errors in Estimating the Arrest Risk per Crime (hi) 335 Typical Range for Ratio of Reported Arrests to Current 5 Estimates (O = 2.0, r = .5) Minimum Estimates (O = 2.5, r = .25) Maximum g Estimates (O = 1.25, r = .75) Reported Crimes ~ (A:C) ~lb ~A ~A .1 .025 8.0 .010 20.0 .060 3.3 .2 .050 4.0 .020 10.0 .120 1.7 .3 .075 2.7 .030 6.7 .180 1.1 AProbability of arrest per crime is estimated as ~ = (A/O)/(C/r). _Mean offending frequency is estimated as A = u/a for a tvoical value of ~ = 0.2. an offense type are assumed to be equally vulnerable to arrest for their crimes. To the extent that offenders actually differ in their arrest risk per crime, applying the same offense-specific qi uniformly to all offenders not only distorts the variations in hi esti- mated for different offender subgroups, but also introduces errors into the mean value of hi estimated for the entire population. Variations in q for any offense type may result from differential enforcement prac- tices that increase arrest vulnerability for some offenders compared with others. Such differential treatment for the saline offense type is especially worrisome when examin- ing differences in A across demographic subgroups; concerns have been raised, for example, about the possibility that police exercise greater discretion in arresting fe- males, juveniles, and whites. A lower qi for these subgroups would lead to their under- representation among arrestees and to a corresponding underestimate of their hi if the same q was applied unifo~ly to the Us of all subgroups in offense-specific analy- ses. Some preliminary evidence suggests, however, that biases resulting from differ- ences in q may not be a serious problem for demographic subgroups. In particular, rea- sonably close consistency has been found between the attributes of arrestees (espe- cially sex, age, and race) for an offense type r ~ ,~ ~ ~ ,~ ~ and those of offenders in the same offense type described by victims in victimization surveys (Hindelang, 1978a, 1980; Hin- delang, Hirschi, and Weis, 1979~.~6 Another potential source of variation in offense-specific qs is variation in enforce- ment practices that are associated with of- fending behavior. On the one hand, high- rate offenders in an offense type may be especially skillful at avoiding detection and arrest, which would result in a negative relationship between A and q in that of- fense. On the other hand, their high rate of activity may increase their vulnerability to arrest per crime as they become known to the police and are targeted as suspects, which would contribute to a positive rela- tionship between A and q in that offense. Failure to address subgroup differences in q that vary systematically with differ- ences in A can lead to biases in the estimates of A derived from A, even in offense-specific thin relying on victim reports, the results for offender attributes are limited to crime incidents in which victims can describe the offender, either because of direct observation of the offender or because of a subsequent arrest. The results will apply to offenders generally if the subset of crime incidents for which offender attributes are known is adequately representative of crime incidents gener- ally.

336 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-24 Example of Biases in Estimates of Individual Offending Frequencies (I) Arising from Use of a Single Probability of Arrest per Crime (a) for an Offense Type Offender Subpopulations 1 2 3 4 Proportion of offenders in each subpopulation .75 .15.09.01 1.0 Individual annual offending frequency (I) 1.0 5.010.0100.0 3.4 Case 1: Negative relationship between ~ and ~ Probability of arrest per crime (a) .06 .05.04.03 .056 Individual annual arrest frequency (a) .06 .25.403.00 .149 Individual annual offending frequency (A) estimated using population average ~ = 0.056 Case 2: Positive relationship between A and g Probability of arrest per crime (a) Individual annual arrest frequency (a) Individual annual offending frequency (A) estimated using population average = 0.034 Weighted Total Population Mean 1.1 4.5 7.1 53.6 2.7 .03 .04 .05 .03 .20 .06 .034 .506.00 .158 .9 5.9 14.7 176.5 4.6 analyses. Hypothetical data can be used to explore the biases in A that arise when q and A vary systematically with one another for an offense type. Say there are four offender subgroups (see Table 24) and that, consis- tent with the observations of highly skewed distributions of A, most offenders commit crimes at rates below the mean frequency of 3.4 set for the entire offender population, ant! a small fraction (1 percent in this exam ple) commit crimes at very high rates. Case 1 (see table) explores the biases that arise when A and q are negatively related, that is, q is lower for offenders with higher As. In this case, applying the population average value of q uniformly to all offender sub- groups leads to an underestimate of A for that population. Conversely, if A and q are positively related, as in Case 2 in Table 24, use of the population average for q in all

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS subgroups leads to an overestimate of A for the population. The potential for biases in A from use of a single value of q for all of- fenders in an offense type highlights the importance of empirically investigating the nature of any further variations in q after controlling for differences among offense types. i Some preliminary evidence is available on the relationship between A and q. Anal- ysis of experiences in 2 early years of a 5-year survey of offending by a sample of U.S. youths revealed a tendency for the arrest risk per offender to increase with increases in the number of self-reported offenses (Dunford and Elliott, 1984:Table 7~. Less than 1 percent ofthe offenders who reported committing only 1 or 2 offenses in 2 years were arrested during those years, but 19 percent of those who reported com- mitting more than 200 offenses during that period were arrested. These data on of- fenders and arrests provide a basis for cal- culating the arrest risk per crime, q, for different offender subgroups.~7 The results, reported in Table 25, are rather striking: the estimated values of q are reasonably close over wide ranges of A. Nevertheless, there is a downward trend in q as A increases, especially at lower values of A. Beyond 20 crimes in 2 years, the arrest probability q does not vary substantially with further in- creases in the number of crimes committed. The research available to date generally fails to find systematic or substantial varia- tions in q. With the exception of higher arrest risks for low-rate offenders, there is little variation in q with differences in A or in the demographic attributes of offenders. While still only preliminary, these results suggest that the errors may be small when a single q is used to derive A from A. i7These estimates of q were provided by Blumstein (1985, personal communication) and are calculated as follows: if q is the probability of arrest per crime, and p = 1 - q is the probability of not being arrested for a crime, then the probability of no arrests for persons committing n offenses is pn and the probability of at least one arrest for these n offenses is 1 _ pn. 337 Alternative Estimates of Individual Offending Frequencies The explicit estimates of offending fre- quencies derived from the analyses of self- reports by inmates and arrest histories re- viewed above are limited to serious adult offenders. A number of other studies pro- vide estimates of participation rates by of- fenders, b, and aggregate incidence rates, A, for the same sampled population. Those studies-provide an opportunity to develop frequency estimates for a wider array of offenders, particularly juvenile offenders. Aggregate incidence rates reflect the num- ber of offenses per sampled individual, in- cluding both active offenders and nonoffend- ers. When combined with b, however, A provides a basis for estimating ,u. Relying on the relationship in Equation 1, the gen- eral strategy is to divide A by b to yield arrest frequency rates, ,u, for the active sub- set within a sample. If,u and b vary together systematically across offender subgroups, the resulting es- timates of ,u will be subject to the same biases illustrated in Table 24 (replacing q with b and A with ,u in the example). The estimate offs derived in this manner is most accurate when ,u and b vary independently across offenders. A more pervasive problem is that, taken directly, the estimates of raw Us are artificially inflated by the require- ment that all active offenders must have at least one event to enter the participation measure and by the fact that raw frequen- cies can never be smaller than one. The rates can be adjusted, however, to eliminate the bias from the required event in the measurement period and to produce uncon- ditional frequency rates for active offenders. The basic approach to adjusting raw fre- quencies is to restrict the estimate of fre- quency rates to the period after the re- quired first event. Simply, the one required event is subtracted from the raw frequency rate and then the length of the measure- ment period is adjusted to also exclude the time to this first event. As a reasonable approximation, and assuming that the time between offenses is distributed exponen- tially with mean 1/A, the expected time to

338 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-25 Estimation of Probability of Arrest per Crime (if) from Data on Offenders and Arrestees in National Youth Survey Number of Standard Self-Reported Midpoint Probability Deviation Offenses in Number of Number of Fraction of Arrest per of ~ 1976 and 1978 Offenses Offenders Arreste ~Crime (I)- Estimate 1-21.5149 .0067 .004479 .004467 3-54151 .0199 .005004 .002871 .6-10 ~181 .0110 .001388 .000978 11-2015207 .0290 .001959 .000793 21-5035233 .0300 .000871 .000327 51-10075131 .0382 .000519 .000230 101-200150109 .0734 .000508 .000176 201+25090 .1889 .000837 .000193 Stanford and Elliott (1984:Table 7). bIf g is the probability of arrest per crime and ~ = 1 - ~ is the probability of not being arrested for a crime, then the probability of no arrests for persons committing n crimes is it- and the fraction ever arrested is just 1 - ~ . The midpoint value for the range of crimes committed is used for n to estimate 5. The results,however,are roughly comparable within the entire range. For the 11-20 group, for example, g = .002670 for n = 11 and 5 = .001470 for n = 20, compared with the midpoint value of ~ = .001959 for n = 15. CThe standard deviation for the estimate of ~ is estimated from: S = ~[q (1 - q)]/~(number of persons)(offending rate)]. - In the 11-20 example, S = ~ [(.001959)(.998041)]/~(207)(15)] = .000793. the first offense-conditioned on the fact that there is at least one offense in a mea- surement period of length t is given by J X&e AX OX E(T~Ti ~ t) = l - e-At A ~ em:" (2) The requirement of at least one offense in t reduces the unconditional mean time between offenses, 1/A, by the quantity te-At/(1 - e-At). If the raw frequency is given by N of- fenses in a measurement period of length t (where N- 1 for each offender), the ad- justed unconditional frequency after the first offense can be approximated by N - 1 N 1 se-At t t- + A 1- e-&' or A N 1 - e At t . (3 )

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS Given the value of the raw frequency rate, Nlt, and the length of the measurement period, t, it is possible to solve Equation 3 for the desired adjusted frequency, A, using iterative numerical methods. Equation 3 indicates that for large values of A or long t, the upward bias in Nlt estimated directly from the data is small, and the adjustment is less important. In these cases Nlt is very close in value to the desired unconditional A. The estimates of frequency rates ob- tained directly from inmate self-reports (Peterson and Braiker, 1980; Chaiken and Chaiken, 1982a) and from arrest histories (Blumstein and Cohen, 1979; Cohen, 1981, 1983) are vulnerable to a similar upward bias. In both types of estimates, frequency rates for an offense type are estimated only for offenders who have at least one offense (or one arrest) for that offense type. In the arrest histories, the required arrest can oc- cur any time in a career; in the inmate self-reports, the required crime must occur during the observation period prior to the current incarceration. As indicated above, however, the upward bias in frequency rates is small when the length of time dur- ing which the required event may occur is long-as in the arrest histories, or when frequency rates are large as in the inmate self-reports. For typical raw offending frequencies of five crimes or more per year found in the inmate self-reports for example, A is overestimated by less than 1 percent. For an annual raw frequency of 5, the adjusted unconditional frequency is 4.97. The reasonableness ofthe approximation in Equation 3 was assessed using data on the juvenile arrest experiences (as mea- sured by police contacts) of a cohort of boys born in 1945 and residing in Philadelphia between the ages of 10 and 18. In an anal- ysis done expressly for the panel, the au- thors ofthe original study (Wolfgang, Figlio, and Sellin, 1972) provided annual estimates of individual arrest rates for active offenders at each age from 13 to 17 and separately by the offender's age at first arrest. For each crime category examined, the number of active ofFenHer~ in a year (i.e., juveniles 339 with at least one arrest for that offense type in that year) was identified, along with the offenders' total arrests for the same offense type in that year, which yielded an annual raw arrest frequency. Forty-four juveniles first arrested for a property index offense (burglary, larceny, or auto theft) at age 15, for example, were arrested again for a similar offense at age 16. These 44 active property offenders experi- enced a total of 50 property arrests at age 16. The resulting average annual ,u of 1.14 property arrests is inflated by the require- ment of one property arrest that year for these active property offenders. Using Equation 3, the adjusted unconditional ,u for property offenders is reduced to .28 arrest annually. To test the adequacy of the adjustment, a separate estimate was generated from the raw arrest history data for these juveniles. Using data on the dates of arrests for indi- viduals, the one required arrest for each active offender was eliminated and the ac- tual time to this required arrest was ex- cluded from the time at risk of a property arrest during that year. The resulting empir- ically based ,u for these 16-year-old active offenders was almost identical to the ap- proximation. Considering only the actual arrest experiences for each individual after the required arrest at age 16, the average annual ,u is also estimated to be .28 property arrest per year per active offender. The adjusted estimates of ,u based on Equation 3 are compared in Table 26 with the estimates derived directly from the ar- rest history data. Averaging across the indi- vidual ages, the adjusted Us in the various offense categories are strikingly similar to the estimates derived directly from the ar- rest history data. This same strong corre- spondence is also found within the separate estimates by age at current arrest and age at first arrest; the correlations between the two alternative estimates exceed .96 for each offense category. The adjustment in Equa- tion 3 has the advantage that it relies on aggregate data on incidence rates and par- ticipation rates for a sample and does not require individual-level data on the actual timing of offenses or arrests. Nevertheless,

340 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-26 Assessment of Adequacy of Approximation of Arrest Frequencies (a) (juvenile offenders in Philadelphia) . Arrest History Raw Annual Adjusted Estimate of Offense Type and Arrest Annual Arrest Annual Arrest Offender Race Frequencya Frequencyb Frequency_ All offenders Robbery 1.112 0.232 0.244 Violent indeed 1.061 0.131 0.124 Property index 1.210 0.410 0.401 Non-index- 1.331 0.611 0.639 Total 1.469 0.839 0.879 White offenders Robbery 1.133 0.273 0.295 Violent index 1.049 0.109 0.092 Property index 1.172 0.342 0.331 Non-index 1.239 0.459 0.479 Total 1.317 0.587 0.620 Nonwhite offenders Robbery 1.109 0.229 0.238 Violent index 1.064 0.144 0.131 Property index 1.239 0.459 0.453 Non-index 1.439 0.789 0.812 Total 1.632 1.082 1.133 NOTE: Annual arrest frequencies, a, were estimated from the number of arresters and their associated number of arrests each year for a cohort of boys born in 1945 and residing in Philadelphia from ages 10-18. The estimates were developed expressly for the Panel on Research on Criminal Careers by the authors of the original study (Wolfgang, Figlio, and Sellin, 1972). aThe raw frequency rate is calculated from the ratio of arrests to active arrestees in each year. When individuals must have at least one arrest in the measurement period to be counted among active offenders, the raw frequency rate is inflated by this required event. DRelying on only the raw frequency rate and the length of the measurement period, the adjustment in equation (3) is used to remove the upward bias introduced by the required one event.for active offenders. CRelying on individual-level data on the actual timing of arrests for active offenders, the arrest history data provide direct estimates of arrest frequencies after the first arrest in each year.

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS the approximation appears to be quite ade- quate for providing estimates offs that re- move the upward bias introduced by the requirement that active offenders have at least one event in the measurement period. This approximation will be used below to derive frequency rates for active offenders from published data for various offender samples. Wherever possible, these esti- mates will be disaggregated to explore the variations in frequency rates with offender attributes. (See Table 27 for a description of the various data sets that are used.) The adjusted Us reported in Table 26 for juvenile offenders in Philadelphia are strik- ingly consistent with the Us reported earlier for adult offenders (Table 191. Not only are offense-specific Us similar in magnitude for adult and juvenile offenders, similar pat- tems are also observed between racial groups. As for white and black adult ar- restees in the Detroit SMSA, the differences are small between the adjusted Us for white and nonwhite juvenile offenders. The larg- est nonwhite-to-white ratio for juveniles is the value of 1.8 for total offenses. Further confirmation of the adequacy of the adjusted estimates offs is found in data from the arrest histories of adult (age 18 or older) men arrested and arraigned in Brooklyn, New York, during the summer of 1979. McGahey (1982) reports that during the 2 years preceding the sampled arrest, the average monthly arrest rate for this sam- ple while free was .049. This rate, equiva- lent to 1.176 arrests while free in the full 2-year period, applies to the total sample, including sample members who had no arrests in the measurement period. Com- bining this aggregate incidence rate for the total sample with the participation rate of 44 percent ever arrested during the measure- ment period yields an average ,u of 2.673 (1.176/.44) arrests in 2 years free for of- fenders who are active in the measurement period (i.e., those with at least one arrest). Adjusting this rate, using Equation 3 to eliminate the one required arrest, the an- nual ,u is estimated to be 1.22. This esti- mate, an average of one arrest every 10 months, is similar in magnitude to the total Us found for active offenders among Phila 34] delphia juveniles (.84 arrest per offender per year) and among Washington, D.C., adults (1.1 arrests per offender per year free). The variation in ,u by age at current arrest and age at first arrest for juvenile offenders in Philadelphia is shown in Table 28. Arrest frequencies for active juvenile offenders generally do not display the consistent sharp increases with age that are observed in aggregate population arrest rates (see, for example, Figure 1-2 in this volume). Except for robbery, which rises sharply at ages 15 and 16, arrest frequencies for active of- fenders do not display strong systematic increases with age. Indeed, ,u decreases over age for property offenses. The variation in ," as age at first arrest increases is also shown in Table 28. Once again, frequencies for property offenses tend to decline as age of initiation increases. An even stronger decline with age of initi- ation is evident for total offenses. The Us are more stable over different starting ages for robbery and violent index offenses. In Table 28, age at current arrest and age at first arrest are treated separately. Because these variables vary together, i.e., older starters are only available for arrest at older current ages, these separate analyses could well mask opposite eliects of these vari- ables on ,u. Multivariate regression analysis was used to explore this possibility (Table 29~. To allow for the tendency of ,u to peak at age 16, age at current arrest was parti- tioned at age 16 to permit different age trends up to age 16 and between ages 16 and 17. Separate regressions were run for each offense category. The dominant effect on ,u was a consistently significant negative effect for age at first arrest in all ollense categories: older starters had lower Us. This decline for older starters is not due to the reduction in time at risk of offending before age 18 for offenders first arrested at older ages. For each age at first arrest and age at current arrest, ,u was estimated only for offenders who were active that year (as indicated by at least one arrest that year). All estimates were thus restricted to a com- mon exposure time. There were fewer significant trends in ,u

342 U] a) a) .,, . ~ ~5 H Q) ·,4 At: a) ·,1 Q o U] ~q .,' Q o In 1 at: En ~ to a, .,1 ~ U] en O · - · - ~ a, O 00 Q ~ O O a) U] :' U] a, o~ ~ ~ ~- -A Sit ~ U1 ~a, ~ · - ~{Q ~ :^ {Q ~ ~ _ ~. - S ~ Q ~ ~0 3 ~Q ~ ~ == ~ ~ ~ ~ S ~3 ~ ~ ~ ~ ~O ·- ~ ~ ns · - _' ~ ~ ~a . n, ~ ~ 0 ~ u ~ ~ ~ ~ ~ a, ~ ~ ~ U] eq ~ ~ cn ~a, a, a) a, ~ ~ ~ ~ ~ ~ ~ O s .' ~ ~ ~ ~ ~· - O~ 0 ·- ~ ~ ~ ~ ~ ~ ~ ~ ~ e.- ~ ~ ~ ~ ·-l a) ~ ~i-. ~ ra s4 ~ a) u' ~ ~ a) ~ ~ ~ ~ ~ ~ 0 u, a) ~ · - 54 ~ ~ ~ a, :^ ~ ~ a) · - ~. - u' m~ ~ ~ ~ ~ ~ 0 ~ ~ - ~n ·^ ~ ~ ~ ~ v ~ ~ ~ ~ ~ ~ O ~ ~ ~ ~ O ~ ~ ~ ~>~ ~ ~ ~ ~ ~ ~ ~ ~ o-- ~ ·- k4 t~ 0 ~ ~ G) _I ~ · - 0 ·~1 ~ Q-- ~ S~ ~1 ~54 07 ~ ~: ~3 ' 05 ~ ~ U] ~S -'- a) 14 ~ a) 54 U2 ~ u'~ ~ ~ ~ ~ . - U)= ~ · - U] c~ ~ ~O ~ ~ ~ - ~' o,~ ~ m a, ~ ~ ~ ~o ~ ~ ~ ~ 0 3 ^ CJ1 ~ Q. C) O ~ 1Q l~ ^ ~ S ~ O O O ~ O ~ ~ - ~ ~ ~ V~ ~a ~ 3 ~ ~ ~ ~ ~ ~ ~ ~ U2 ~a,- u~ eq u~ O ~ a~ a~ eQ u~ u~ cn cn ~ ~ ·- ~ CQ o~ ~ ~ ~ ~ ~o ~ · - a, ~ ~ ~ a, 0 ~ ~ ~ S" ~ ~ ~ a, ~ ~ ~ ~ ~ a, ~ ~ ~ ~ ~ ~ ~ a, Q ~ ~ ~ ~ c: a~ ~ ~ ~ ~ ~O ~ ~ ~ u~ ~ ~ ~ ~ ~ ~ o ~ ~l a~ s ~ ~ 3 ~ ~ - u! a) ~ ~ 3 ~ ~ ~ ~ ~ ~ ~ u, ~ ~ ~ ~ ~ - ~ ~ a z ~z ~ ~ ~- mm a) ~ ~ ~ _ a) s: ~ _ u, ~ a) a, rrJ 3 ~ U] J~ (1) C) ~ ~:^ Lt, ~ ~ ~ 01 U] S ns O, 0 ~ a) aD · - 01 Q _I ' ~ ~ G) ~ ~ ttS 0 a) U1 · - ~ U2 ~ ~ ~ ~ ~ S ~ ~ ~ ~ ~ ~ O 0 a, ~ ~ ~· - ~ 0 ~ ~ ~ ~ ~ ~ ~ a) ~ ~ 0 ~ ~ ~ ~ ~ · - ~· - ~ ~a' ~ a, ~ a' 0 ~0 a) 0 u~ 0 3 a) ~ G) ~ ~-~ _~ 0 rn ~ ~ U' ~ ~ ~ Q ~ ~ ~ u' ~ a, ~n ~ ~u' a' ~ u' s~ ~ u' a, · - ~5 ~ u' ~ s~ Q >, u' ~. - . - ~ , ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ 0= 0 O O ~ :~ ~O ~ O - Q ~O ~O~ U~ k O Q, ~ ~ ~V a,-co · ~ ~a, ~ 05 0 ~ ~ ~ ~ ~ O ~ ~ ~ ~ ~ a, X ~ ~ ~ ~ ~ ~ ~ ~·.' ' ' ' ' [,q ~ ~ a' ~ ~ ~ ~ o ~O ~ s ' ~ -- - ~ ~ ~ ~ ~- - ' ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ · - ~ ~11 11 11 11 11 ·~ a) u~ ~ ~"l ~ o4 ·- u~ a) aJ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~V ~ ~Q ~ ~ ~I ~ Z ZZ ~ Z -- ~ 0 o~ a~ ~-1 ~ ~ ~ ~ ~ 0 ~ 0 to ~ ~- ~ ~ 00 ~ U] ~ ~ dJ _I ~ O ~1 =1 ~a O O O U] ~0 a ~a ~ - 1 u, U2 ~0 a) ~ JJ u ~ a, eq ~ ~ ~ 1 cr ~: u ' ' - ~ ~ O ~ ~ O ' a ~ ~ _~ Y cs ~t ~u ~0 ns ^ ~ ~ 0 ~t~ S O o o ~ ~ ~o ~ ~ ~ ~ o ~ ~ ~ ~ ~ ~ ~ mm o == ~m ~ m~ o -- · - ~ a ~ a, U] t ~ 0 0 ~ ~ u~ ~ 0~ eQ 0 ~ a' 0 ~ 0 ~ ~Q S s a) ~ ~l u~ U] ~ s a,~ ~ ~ ~ ~ ~ ~ 0 81 .- a' 0 a) ~ ~ ~ 0 ~ .~1 Q ~ ' Q~ _' C) c: ,= _I ~ O ~ ~ ~ ~ ~O O O ~ ~m ~ ~ ~ ~ ~ O ~ ~- ~ ~ 0'- ~ ~ ~ O ~ ~ ~ S O · - ~ · - l-e ~ a) ~ _ ~ ~ ~ ~ ~ ~ ~ ~. ~ ~ ~ _ ~ ~ ~ ~u' a ·,l ~ {d co u~ ~ a,~ ,l ~ ~ ~ {U ~ · c~ ° Q ~ ~ _d ~ ~ ~ ~ · Q ~ ~ ~ ~ ~ ~ O ~ Pd ~ 1 o ~ ~ eq L4 >4 ~ ~ ·- U~) o o ,= o ~5 ~ o ~ ~ · ~) O O ~: .- ~ Q ~ ~ ~ ~ ~ ~ ~ Z ~ Q Q~ P4 ~ O ·- ~ 0 u, a' ~a, ~ ~r: ~Z _ _ u~ - ~0 a ~cx) a' ~a ~a' ~5 _ _~ ~ _ _ ~ O ~O - C:. - ~>, ~ cn ~ ~ ~s ~ ~ ~ ~s ~ co ~ m ~ ~ ~ 0 a' tu. - ~ ~ O ~ ~ r4 Cn c:~ ~ ,~ , ~ _ O 3

343 _ ~ I ~ ~a) ° ' A' to ~ ~ ~ ~ 0 a) ~ ~ ~ ~ ·- ~ ~ ~ ~ s t, ~ ~ ~ ~ Us -.- ~ ~ ~ ins . - ~ . - o ~ ~ ~ ~ G. ~ ~ ~ 3 0 ' O ~·- ~ ~ ~ ~ O ·-~ 65 0 ~ ' ~ 41) dJ O ·,1 1 C).- ~ ~ ~ ~ ~ ~ ~ ~ 3 · - ~ ~ ~ ~ Q Jo ~ ~ ~ · - ~ ~ ~ O O N en ~ O., to C) .~1 I ~ a-1 ~ p`- - O O ' a, ~ . - ~ ~ ~ ~ ~ ~ ~ ~ s Jo ~ a) c' ~ ~ ~ a~ O ~ O ~ ~ · - ~ ' ~ ~ ' - ~ ~ ' to) ~ I' ~ Jo ~ t) O ~ ~ ~ ~ ~ ~ 3 a~-~ ~ ~ u~ c: a~ ~ c: ~ ~ ~ ~ ~ ~ ~ ~ e~ ~ erl ~ ' ' ~ ~ . - m<1 u, o. ~ ~ ~ 0 In ~ t) t~l~lc ~ v a ~tn ~ ~ ~ :' ~ ~ ~ '^ ~ u'= cn u,~a ~ ~ ~m u,~ ~> ~ a {Q 'l ~ s ~ ~ ~ ~ ~ 0 · - ~ a, C.) ~ ~ ~ ~ ~ ~ ~ a) ~ - s 3 ~ ~ ~ O ~ ~ ~ ~ ~ ~ ~ ~ - = ~a ~4 ~10 ~ ~ ~R eq =. - a, ~ c ca :~ a) ~ ~ 0 0 ~ 54 . - 3 ~ ~ eq CD .,' ~ ~ '^ C c: ~0 0 - O {Q a U] ~ ~ ~ ~) ~ ~ £ -~ L4 O 0 C ~eQ ~ ~ 0 ~ ~ O ~ ~ ~ 3 O ~ ~ ~£ ,~, ,1 ·- O ~ C ~, ~ m a- - ~ C O ~V C u~ a, 3 ~-'- ~' ~, O C , ~ a, O ~ ~) 0 ~ c 0 ~ e ~ = ~ C) 0 0 ~u' O ~n 54 ~ ~ ~ C ~ ~ C a) 03 ~4 Q =-- u ~o >, ~ ~ ~ ~ ~S~= ~ ~ ~ ~ ~ ~ ~ a, ·o N ~ ~I £ 0~4 ~I o .,, ~ ~ O ~1 .- N ·- _I ~ 54 u ~_I ~, a) 41) o U] U] u, 1 a ~n C ~ C ~ O^ ~ a) ~C ,1 ·~1 N ~O ~- C c- - ~ ·'.^ O C ~ ~ £ · - a) ~ · 54 ~s~ 0 cn 0 · - Y £ ~ C >4 a) ~° ·-~ .Q ~ a) C 0 ~ Q4. - . ~ a' s ~ 1 `~5 C) ~ =- - ~ ~ ~ ~ Z C ~ ~ V ~ ~ ~ 3 C 'o · - ~ ~ C ' 0- - ~ 0 ~ ~ C a) ~ 0 ~ ~ ~ Y ~ 0 ~ ft JJ ~ - ~ S tit ~11 ~ y ~ -~ ~ O 4) >t 03) Z G) O ·r1 ~ ' O ~54 0 =- - ~ ~ m- - ~ · - m- - c~ a) cr ~ ~ J_~ ~N 3 ~ ~N a) ~ 0 ~ c ~ Y O ~ ~ C a, ,0 ~ .,~ ·~1 t_ ~O O ~ ~O ~ ~ ~ ~ N ~ ~ ~ ~ ~. - ~ . - ~ S ~ ~ O ,~ ~O -~ F a~ ~ ~ 0 0 ~ O ~ ~- 11 a, C _' ~ 3 ~ · - a ~ ~ c~ ~ ~ 0 ~ ~ ~ O ~° c 0 ~ ~ ~ ~ Z ~ ~ ~ z ·- ~ ~ - ~ - 3 ~ E ~u ~U] a, · a' _ ~_ _ · - U] O 0 s ~0 S S: ~· - S _ a) 0 c C~D ~H

344 CRlMlNAL CAREERS AND CAREER CRlMlNALS TABLE B-28 Variation in Annual Arrest Frequency (a) by Age at Current Arrest and Age at First Arrest (juvenile offenders in Philadelphia) Violent Property Robbery Indexa Indexb TotalC Age at current arrest 13 .12 N.A.d .48 .63 14 .12 .08 .46 .72 15 .39 .12 .38 .87 16 .36 .19 .39 .94 17 .16 .08 .29 .83 Age at first arrest <13 .19 .14 .51 1.07 14 .10 .20 .25 .69 15 .61 N.A. .25 .74 16 .20 N.A. .36 .57 17 N.A. .19 .11 .36 NOTE: Annual arrest frequencies, i, were estimated from the number of arrestees and their associated number of arrests each year. These estimates, developed expressly for the Panel on Research on Criminal Careers by the authors of the original study (Wolfgang, Figlio, and Sellin, 1972), were then adjusted using equation (3) to eliminate the effect of the one required arrest each year. Violent index offenses include arrests for murder, rape, and aggravated assault. Robbery, which is usually included among violent offenses in the FBI's annual Uniform Crime Report, is treated as a separate crime category here. bProperty index offenses include arrests for burglary, larceny, and auto theft. CTotal offenses include all arrests for any type of offense except traffic and juvenile status offenses (i.e., runaway, truancy, and Incorrigibility). -In some age categories, the arrest data were insufficient to provide reliable adjusted arrest frequency estimates.

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS TABLE B-29 Regression Results Exploring Effect of Age at First Arrest and Age at Current Arrest on Arrest Frequencies (a) of Juvenile Offenders in Philadelphia (significant coefficient) 345 Age at Age at Current Arrest Offense Type First Arrest 13-16 16-17 R2 SW injurya -* NS NS .37 SW theftb -(+) NS NS .37 Property index: -(+) NS NS .33 SW non-index) -** +*** NS .71 Non-indexl -** +** NS .71 Total offensesf -** +*** NS .73 NOTE: Annual arrest frequencies, i, were estimated by age at current arrest and age at first arrest from the number of arresters and their associated number of arrests each year. These estimates, developed expressly for the Panel on Research on Criminal Careers by the authors of the original study (Wolfgang, Figlio, and Sellin, 1972), were then adjusted using equation (3) to eliminate the effect of the one required arrest each year. lithe Sellin-Wolfgang (SW) injury offenses include all offenses involving injury to victims. bThe Sellin-Wolfgang tsw) theft offenses include all offenses involving loss of property. CProperty index offenses include burglary, larceny, and auto theft. ~ elfin-Wolfgang (SW) non-index offenses include all offenses not involving injury to victims or loss or damage to property (excluding traffic and juvenile status offenses). Non-index offenses include all offenses (other than juvenile status offenses and traffic offenses) not included among the FBI's index offenses of murder, rape, robbery, aggravated assault, burglary, larceny, or auto theft. fTota1 offenses include all offenses except traffic and juvenile status offenses (i.e., runaway, truancy, and incorrigibility). NS = Coefficient not significant. +Significant at .10 level. *Significant at .05 level. **Significant at .01 level. ***Significant at .001 level or better.

346 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-30 Variation in Arrest Frequencies (a) with Arrest Record ( juvenile offenders in Philadelphia) Number of Arrests So Far Number of Offenders Subsequent Mean Annual Arrest Rate While Free (all offenses) - 31,033 1.15 4708 ~ 1.43 5534 1.52 6386 1.79 7303 1.92 8224 2.08 9175 2.25 SOURCE: Barnett and Lofaso (1985:Table II). over age at current arrest. In the more ag- gregate categories of non-index offenses and total offenses, which include many dis- tinct offense types, ,u increased and reached a peak at age 16. This same increase was not observed in the less aggregate offense cate- gories. These results suggest that the in- creases with age at current arrest may re- flect the effect of varying numbers of active crime types within the aggregate offense categories. While ,u may be relatively stable over age within any single offense type, older offenders may be active in more of- fense types, thus increasing their total ,u in the aggregate offense categories. Using a slightly different estimation tech- nique on the arrest data for Philadelphia juveniles, Bamett and Lofaso (1985) as- sessed the relationship between past and future Us. Without exception, there was a clear upward trend: offenders who had more arrests were also subsequently ar- rested at higher rates (Table 30~. Bamett and Lofaso explored the possibil- ity that the variation observed with prior arrests was attributable to the fact that those with fewer prior arrests included many of- fenders who terminated their criminal ac- tivity while still juveniles. Failure to ex- clude this time after career termination from the time at risk of subsequent arrests would lead to underestimates of arrest fre- quencies for offenders who remain active. They found that for each arrest after the third, about 28 percent of offenders do not have another arrest as juveniles. Because the arrest data are truncated at age 18, how- ever, it is possible that many of these appar- ently ended careers actually continue into adulthood. Assuming that no o~enclers ac- tually terminated their careers during their juvenile years, and using the estimated ,us for juvenile offenclers, Bamett and Lofaso (1985) estimated the expected number of offenders who will have an arrest-free pe- riod to age 18, even though they remain active in criminal careers. This expected number was not significantly different from the observed number; thus, career terrnina- tion does not appear to be a substantial

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS factor in the variation in arrest frequencies observed for juvenile offenders. The authors then explored the role of prior arrest frequencies in the observed relationship. Within the limited time avail- able until age 18, offenders who accumu- lated large numbers of arrests must have done so at higher frequencies than those with only a few arrests. Thus, arrest record may be a reasonable proxy for prior arrest frequencies. Examining both the number of prior arrests and prior arrest frequencies together, they found arrest frequency in the past to be predictive of subsequent arrest frequencies for juveniles who remained ac- tive, but the relationship between the num- ber of prior arrests and subsequent arrest frequencies disappeared. The data for juveniles in Philadelphia provide estimates of ,u based on official police and court records. Actual offending frequencies, As, for juveniles based on self-reports of crimes committed were de- rived from published data from annual sur- veys of a national sample of U.S. youths. From 1976 to 1980, the same national sam- ple of youths (males and females), aged 11-17 in 1976, were surveyed annually about the crimes they committed in the preceding year (Elliott et al., 1983~. The number of respondents declined somewhat in subsequent years, but there was little loss from the original respondent sample of 1,725 adolescents in 1976. Elliott et al. (1983:Part II) report exten- sive data on aggregate incidence rates (C, crimes per capita) for all sampled youths, and participation rates, d. These rates were reported separately for each survey year, by offense type, and by various attributes ofthe respondents. Dividing the incidence rates by the corresponding participation rates yields estimates of A (see Equation 11. Equation 3 is then used to eliminate the upward bias introduced by the requirement that all active offenders in a year must have at least one offense during that year. The final adjusted estimates of A (aver- aged over the 5 survey years) are reported in Table 31. Overall, youthful offenders active in serious offenses reported commit- ting an annual average of 8 robberies, 7 347 felony thefts, 4 felony assaults, and 7 index offenses. Combining all offense types, these active offenders are estimated to have com- mitted 37 crimes annually. These frequen- cies are remarkably similar in magnitude to the average offending frequencies esti- mated from arrest histories of adult ar- restees (Table 221. Despite differences in data sources, estimation techniques, and attributes of the offender samples (includ- ing, predominantly, urban black males among adult arrestees>, a strong conver- gence is emerging in the estimates of A. In part, this convergence reflects the generally small differences in A observed for different offender attributes (Table 31~. While As for male offenders are always larger than those estimated for female offenders, the male-to- female ratios are generally under 2:1. This is substantially smaller than the differences usually observed for the sexes in aggregate incidence rates (e.g., Table 1-2, this vol- ume). The differences in A by race are similarly small compared with the large differences observed in aggregate inci- dence rates (e.g., Table 1-2, this volume). There is also a general absence of sharp increases with age among juveniles (e.g., Figure 1-2, this volume). The notable ex- ception in Table 31 is the consistent in- crease in total A with age. A similar age pattern was observed among adult arrestees and adult inmates, as well as among juvenile arrestees in Phila- delphia. In all instances, frequencies for any single offense type were relatively sta- ble over age, while frequencies for aggre- gate offense categories, which include many different offense types, increased with age among juveniles and decreased with age among adults. This pattern sug- gests that the age effect observed in aggre- ~ate offense categories may be due to changes in the number of active offense types for offenders, i.e., offending first be- comes more diverse during the juvenile years as new offense types are added and then becomes more specialized in the adult years as the number of active offense types declines. Like the pattern observed for other demo- graphic attributes, there was also very little

348 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-31 Variation in Annual Offending Frequency (A) in National Youth Survey OffenderFelony Felony AttributeRobbery Thefts Assaul ~Index: Totald All offenders 7.56.5 3.9 6.8 36.6 Sex Male 8.47.1 4.4 7.6 42.2 Female 3.34.4 2.6 3.7 26.7 Race White 9.65.8 3.4 6.4 34.9 Black 4.67.4 3.9 6.3 39.6 Hispanic 3.118.4 4.3 7.3 42.2 Age 13 6.33.6 4.0 6.0 21.4 14 5.46.5 3.4 5.6 25.0 15 15.3e6.2 5.5 10.0e -28.4 16 6.74.5 3.3 5.2 31.8 7 6.58.7 3.4 7.1 47.9 18 5.66.3 3.7 5.3 45.0 19 5.28.8 6.6 7.9 57.3 Employment Unemployed 5.48.4 4.1 6.8 23.7 Part time 8.35.5 3.6 6.3 30.5 Full time 9.07.3 4.5 7.7 66.4 NOTE: Annual offending frequencies, X, were estimated expressly for the Panel on Research on Criminal Careers from data available in Elliott et al. (1983:Part lI) reporting aggregate incidence rates (C, crimes per capita in the sample) and annual participation rates (d, fraction of sample with at least one crime) for each offense type. The resulting raw frequency rates were large enough that no adjustment was needed for the required one crime per year. Annual frequencies, A, are averaged over the 5 survey years. felony theft includes self-reports that the respondent stole a motor vehicle, stole something worth more than $50, broke into a builbding or vehicle, or bought stolen goods. -Felony assault includes self-reports that the respondent committed aggravated assault, sexual assault, or participated in gang fights. CIndex offenses include self-reports of offenses involving robbery, felony assault, or felony theft (excluding the purchase of stolden goods). -In addition to robbery, felony assault, and felony theft, total offenses include minor assaults, minor thefts, drug sales, prostitution, carrying a weapon, and certain public order violations (panhandling and disorderly conduct). lithe frequency rates for 15-year-olds are inflated by a very high number of robberies reported\by ls-year-olds in 1977. When this unusual rate is excluded from the average, the annual frequency estimates are reduced to 5.8 robberies and 5.4 index offenses.

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS 349 TABLE B-32 Estimates of Annual Offending Frequencies (a) Based on Arrest Histories for Addicts in Drug Treatment Programs Individual Arrest Probability of Individual Offending Offense Rates, ~ Arrest per Crime, Ma Rates, ~ Type Santa Clara Brooklyn Santa Clara Brooklyn Santa Clara Brooklyn . Property .91 .61 .043 .026 21.2 23.3 (N = 45) (N = 41) Robbery .77 .36 .091 .0S9 8.4 6.1 (N = 5) (N = 9) Assault .65 .42 .194 .133 3.3 3.1 (N = 22) (N = 16) NOTE: Derived from data on aggregate incidence rates, A, and annual participation rates, b, in Sechrest (1979:Tables 4 and 6). Rates reported are a simple average of annual rates estimated for active offenders in the 2 years before program entry and 2 years immediately after program entry for 277 addicts in 1970-1971 in Santa Clara, Calif., and 473 addicts in 1969-1970 in Brooklyn, N.Y. No adjustments were made for any time not at risk because of incarceration or hospital confinement. Average number of active offenders per year for the 4 years of data is in parentheses. 9The probability of arrest per reported crime, A, is based on the ratio of reported arrests to reported crimes in Santa Clara for 1970 (California Department of Justice, 1980) and in New York City for 1971 (Vera Institute of Justice, 1977). These raw ratios were adjusted for nonreporting by victims to the police using reporting rates found in victim surveys for 1974 in California (Bureau of Justice Statistics, 1981b) and in New York State (Bureau of Justice Statistics, 1980a). A second adjustment was made for the average number of offenders per crime incident, O. The same adjustment factor, based on national data (Reiss, 1980b), was applied in both states. variation in A with the employment status of these youths. Whether unemployed or working, offenders active in serious offense types committed those crimes at similar rates. Total offenses, by contrast, which in- clude many less serious offense types, did vary with employment status. During the adolescent and young adult years sur- veyed when youths are normally expected to be attending school increased time that active offenders spent working was associ- ated with increased As. This increase in offending with increased work was due pri- marily to higher As in less serious offense types. Once a youth was active in the more serious offense types, neither employment status nor other demographic attribute strongly influenced A in those offense types. In contrast to the limited effects observed for employment and other demographic at- tributes, offending frequencies vary sub- stantially with drug use by active offenders. Three studies provide data on aggregate incidence rates and participation rates in crime for drug users. These data have been used to derive estimates of offending fre- quencies for active offenders among drug users (Tables 32-34~. Table 32 reports annual arrest frequen- cies, A, and offending frequencies, A, for participants in two methadone drug treat- ment programs, one in Brooklyn, New York, and the other in Santa Clara, California. In

350 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-33 Estimates of Annual Offending Frequencies (I) from Self-Reported Offenses in Weekly Interviews with Heroin Users a Heroin Use Offense Type Irregular Regular Daily Total Any nondrug offense 133.3 172.7 215.2 174.8 (N = 53) (N = 73) (N = 60) (N = 186) Robbery 8.9 16.7 26.5 20.4 (N = 12) (N = 18) (N = 27) (N = 57) Burglary 12.8 35.9 60.5 41.2 (N = 19) (N = 33) (N = 35) (N = 87) Shoplifting 67.8 74.7 105.7 84.5 (for resale) (N = 31) (N = 48) (N = 42) (N = 121) Other larceny 14.8 40.4 32.4 32.1 (N = 22) (N = 41) (N = 37) (N = 100) Forgery 6.8 11.4 18.4 13.4 (N = 6) (N = 7) (N = 8) (N = 21) Con games 138.1 125.2 85.9 113.8 (N = 16) (N = 21) (N = 21) (N = 58) NOTE: Derived from data in Johnson et al. (1983:Tables VI.1 and VIII.2). Aggregate incidence rates (C, crimes per capita in the sample) were computed for a 2-month period from annualized rates provided in Table VIII.2. These were divided by participation rates (d, fraction of sample reporting at least one crime) for the 2-month reporting period in Table VI.1 to yield raw frequency rates for active offenders in the average 57-day reporting period available for each respondent. The raw frequencies for the reporting period were then adjusted to eliminate the one required offense in that reporting period using equation (3), and the annualized adjusted rates are reported here. The number of active offenders generating each frequency estimate is in parentheses. aIrregular heroin users reported using heroin 0, 1, or 2 days per week. Regular users reported use on 3-5 days per week. Daily users reported use on 6-7 days per week. both study sites, official arrest records were obtained for all program participants for the 2 years preceding and 2 years following entry into the program. The resulting an- nual inciclence rates of arrests per capita, A, found in the sample and the arrest partici pation rates, b, are reported separately for each year by Sechrest (1979:Tables 4 and 6~. These data were used to derive ,u for the subsets of active offenders, using Equations 1 and 3. Averaging over the 4 years, active offenders among drug addicts averaged one

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS TABLE B-34 Estimates of Annual Offending Frequencies (A) from Self-Reports by Heroin Users Offense Type Males Females Robbery Assault Burglary Vehicle theft Theft from vehicle Shoplifting Other theft Forgery/counterfeiting Con games Drug sales 29.7 3.3 24.8 7.4 12.5 68.2 12.0 17.8 17.6 186.7 28.7 (2.7)a 4.9 (2.2)a 8.3 63.1 7.6 25.4 12.6 118.8 NOTE: No adjustments were made for any time not at risk due to incarceration or hospital confinement. Orate is based on fewer than 20 active offenders and thus is subject to more sampling variation than other rates in this table. SOURCE: Derived from data on total offenses and participation rates in Inciardi (1979:Tables 4 and 5). arrest every 1.5 to 3 years for serious crimes in Brooklyn, and one arrest every 1.1 to 1.5 years in Santa Clara. These arrest frequen- cies for drug addicts are at least twice as high as those found for comparable offense types among adult arrestees generally (Ta- ble 19). Separate estimates of q were developed for Santa Clara and Brooklyn by use of aggregate data on A and C reported in local police statistics, reporting rates by victims, r, in California and in New York State, and national estimates of the number of offend- ers per crime incident, O. Even relying on local data for different jurisdictions, the re- sulting arrest probabilities for separate of- fense types (Table 32) are remarkably sim- ilar to the estimates for Washington, D.C., and for the Detroit SMSA reported earlier (Table 21~. In each jurisdiction, q is highest for assault at .10 to .20, followed by robbery at rates of about .05 to .10 per crime, and then by property offenses at rates under .05. 35] Combining q with ,u yielded almost iden- tical estimates of A for drug addicts in the two study sites (Table 32~. The higher arrest frequencies in Santa Clara compared with Brooklyn appear to result from higher arrest probabilities per crime in Santa Clara. Ac- tive offenders among drug addicts in treat- ment programs are estimated to commit crimes at about twice the annual rate found for adult arrestees generally (Table 22~. The As estimated from official arrest his- tories for active offenders in drug treatment programs are comparable in magnitude to the As estimated from the self-reported crimes of irregular heroin users in another sample of drug users (Table 33~. Using weekly self-reports of daily drug use and crimes, Johnson et al. (1983) characterized a sample of active heroin users in terms of their intensity of heroin use. Irregular users reported using drugs less than 3 days a week, regular users on 3 to 5 days a week, and daily users on 6 or 7 days a week. On

352 the basis of self-reported crimes during an average 57-day reporting period for each respondent, Johnson et al. (1983) estimated ct for the 2-month reporting period and an- nualized incidence rates, C. These rates were used to develop the estimates of A reported in Table 33. Offending frequencies generally in- creased as drug use increased. The As for daily drug users in Table 33 are at least six times higher than the As estimated for of- fenders generally (e.g., Tables 22 and 311. Similarly high As were estimated from inci- dence and participation rates reported in Inciardi (1979) for active heroin users (Ta- ble 341. It is also noteworthy that among active offenders in this sample of heroin users, males and females reported commit- ting crimes at very similar rates for most offense types. The higher As estimated for daily heroin users when compared with other offenders are consistent with the large differentials in As that have been observed for drug users between periods of heavy drug use and periods of no drug use (McGlothlin, Anglin, and Wilson, 1978; Ball et al., 1981, 1983; Gropper, 19851. Summary Despite differences in observation tech- niques (self-reports or arrest records), in the samples used (inmate, arrestee, or general population samples), and in the jurisdic- tions examined (with their differences in population characteristics and criminal jus- tice practices), considerable convergence is emerging in mean A estimates for specific offense types. Frequency rates, however, do vary across offense types; mean As within violent crimes are lower than within prop- erty crimes: While they are free in the community i8Burglary is a notable exception; male frequen- cies for burglary are about five times higher than female rates. This difference for burglary may be partially offset by a difference in A of similar magni- tude in the opposite direction for pickpocketing, which is not included in Table 34 because of the small numbers of offenders active in this offense type. CRIMINAL CAREERS AND CAREER CRIMINALS · active violent offenders are estimated to commit an average of 2 to 4 serious assaults per year, and · active property offenders are estimated to commit an annual average of 5 to 10 crimes for each of the property crimes they commit. Mean frequencies for inmates are higher than those found among offenders in the community. On the basis of inmate self- re-ports in California and Michigan, when they are not incarcerated, these inmates are estimated to commit · an annual average of 15 to 20 robberies per offender active in robbery, and · an annual average of 45 to 50 burglaries per offender active in burglary. These higher rates for inmates are to be expected because high-rate offenders are more likely to be found in a prison popula- tion, both because their high frequency of crimes increases their exposure to arrest, and because selectivity in criminal justice decision making increases their risk of in- carceration if arrested. Frequencies esti- mated for Texas inmates, however, do not follow this pattem; they more closely re- semble As found for offenders in the com- munity in other jurisdictions. Individual frequencies vary considerably across offenders. The distribution of A is highly skewed: the median offender in any offense type engages in under 10 crimes, but the 10 percent of offenders with the highest rates exceed 100 crimes each annu- ally. Finding the factors that distinguish these highest-rate offenders is especially important in developing policies intended to reduce crime. In contrast to patterns observed in aggre- gate population arrest rates and in participa- tion rates, mean As do not vary substantially with the demographic attributes of sex, age, or race. Differences in mean As are ob- served, however, with age of onset of ca- reers, drug use, employment, and prior criminal involvement. Active offenders who begin criminal activity at young ages, use drugs heavily, are unemployed for long pe- riods of time as adults, and have extensive

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS records of criminal activity generally com- mit crimes at higher rates than other of- fenders. Methodological Issues in Estimating Individual Frequency Rates A number of methodological problems are apparent in reviewing empirical esti- mates of individual frequency rates. These relate primarily to the representativeness of samples used to generate estimates and var- ious problems in measuring rates within those samples. This section discusses the problems and various proposals for address- ing them in future research. Biases in Offender Samples: Differences in Sampling Probabilities Problems of sample representativeness are common to estimates of individual fre- quency rates. The c~ro.~.~-.cenhon .~nmnl~.c used to estimate offending frequencies are generally not representative of the popula- tion of offenders actually active in crime at any time. Offenders are distinguished from nonoffenders by their commission of at least one crime. In self-report samples, the active offenders are identified from their self- reports of offenses committed. In official- record samples, only active offenders iden- tified by the criminal justice system through arrest, conviction, or incarceration enter the analysis. The sampling bias is introduced by the requirement that sampled offenders mat have at least one criminal event e.g., a self-reported offense or an arrest during some fixed sampling period. To the extent that A, and the associated ,u, varies in mag- nitude across offenders, all active offenders are not equally likely to enter the sample. In particular, offenders with a higher A, or a higher individual ,u in arrestee samples, will be more likely than other offenders to meet the sampling criterion of at least one criminal event in the sampling window. As a result, these higher rate offenders will be overrepresented in offender samples. The nature of this sampling bias can be illustrated using a simple characterization 3S3 of individual offending. For purposes ofthis example, each offender is assumed to have a constant A. That rate, however, varies across individuals. As was found in the Rand sur- veys, the distribution is highly skewed: many offenders have a low rate, and a small number of offenders have very high rates. In this example the mean rate is five of- fenses and the median rate is less than one offense. The distribution of offenders in the total active population in this example is shown in Table 35. The probability that an offender will have at least one offense in a sampling period is also presented for sev- eral values of A and for different sampling periods. Regardless of the length of the sampling period, high-rate offenders are al- ways more likely to have at least one offense and thus to be included in a sample of active offenders. The bias against including low-rate offenders is most severe in very short sampling periods. In a 1-day sample, the highest rate offenders in the table are more than 200 times more likely to be sampled than the lowest rate offenders. In the 1-year sample, the disproportionality of sampling reduces to 10.S between the high- est and lowest rate offenders. Despite the higher sampling probabili- ties for individual high-rate offenders, the contribution of those offenders to the of- fender sample will depend on their relative representation in the total offender popula- tion and on the length of the sampling period. The composition of the total of- fender population and of various offender samples for different-length sampling peri- ods in this example is shown in Table 36. The overrepresentation of high-rate offend- ers in cross-section offender samples is greatest for the short sampling periods. In a 1-week sample, for example, only 1 percent of the total population have As greater than 50, but those high-rate offenders represent 7.2 percent of the offender sample. As the length of the sampling period increases, the offender sample becomes more representa- tive of the total offender population. For the parameter values used in this illustration, a sampling period of 3 years provides a rea- sonably representative sample of the total offender population.

354 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-35 Variations in Sampling Probabilities for Offenders with Different Individual Offense Rates (A): Proportion of Offender Annual Population Individual with _ Offense Frequencies 1 Rate, ~<I- - An Illustration _ ~· A _ _ _ ~ ~ Prohabilitv of At Least One Event In sampling Period 3 5 Years Year S Day Week 1 3 6 1 Month Months Months Year 0.1 .05 -- .002 .0~8.025 .049.095.259.393 0.5 .26 .001 .010 .041.118 .221.393.777.918 1 .52 .003 .019 .080.221 .393.632.950.993 3 .67 .008 .056 .221.528 .778.9501.0001.000 5 .74 .013 .092 .341.713 .918.9931.0001.000 10 .85 .027 .175 .565.918 .9931.0001.0001.000 20 .93 .053 .319 .811.993 1.0001.0001.0001.000 50 .99 .126 .617 .9841.000 1.0001.0001.0001.000 100 .9995 .237 .853 1.0001.000 1.0001.0001.0001.000 Assumes ~ is distributed according to a gamma distribution: flea, b) = ret ~ ha le by. with shape parameter a = .25 and scale parameter b = .05 and Eta) the gamma function. The expected value of A is given by a/b = 5 and the variance of is ~2 = 100. Assumes ~ is constant for individual offenders. The probability of at least one event in an interval of length t is given by 1 - e it. While oversampling high-rate offenders is a problem in both self-report and official- record studies, the bias toward high-rate offenders is especially severe in samples based on official contacts with the criminal justice system. This problem was drama- tized in several studies that included both self-report and official-contact measures of offending for general population samples. The Cambridge study (West and Far- rington, 1973, 1977) prospectively followed a cohort of London boys. The primary mea- sure of offending was convictions in court. At various intervals during the observation period, the youths were also asked to report the frequency with which they committed offenses. West and Farrington (1973:165) report that "official delinquents twith con- victions] had committed more delinquent acts according to their own admission than non-delinquents." This higher rate of self- reported delinquency for official delin- quents is illustrated in Table 37. The National Youth Survey (Elliott et al., 1983) prospectively followed a representa- tive national sample of U.S. youths aged 11-17 in 1976. These youths were inter- viewed annually and asked about the fre- quency of their offending in the previous year. In addition, data on their recorded police contacts through 1980 were obtained from local police authorities, and data on arrest records through 1978 were analyzed. Based on their self-reported offenses in the 3 years 197~1978, respondents were clas- sified into offender categories that reflected the intensity and seriousness of their self- reported offending (Dunford and Elliott, 1984~. "Career offenders," for example, in- clude persons who reported committing 12

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS or more total offenses per year or 3 or more index offenses per year in 2 or more consec- utive years. In a comparison of self-reported offending patterns for youths who were ar- rested with those not arrested by 1978, youths who were arrested were also more likely to report committing a greater num- ber of offenses or more serious offenses (i.e., they were more likely to be "career of- fenders") than nonarrested youths (Table 38~. It is also noteworthy that only 6 percent of the surveyed youths had arrest records by 1978. This arrest participation rate is quite low compared with the rates of 35 and 33 percent found among juveniles in the two Philadelphia birth cohorts (see Appendix A), and the conviction participation rate of 20 percent found among boys in London 355 (West and Farrington, 1973~. Several factors contribute to the lower arrest participation through 1978 in the National Youth Survey. First, many of the youths were still under age 15 by the 1978 cutoff date. The number of youths arrested can be expected to in- crease as arrests through 1980 are included. Also, females made up about one-half of the National Youth Survey sample, which low- ers arrest participation compared with the all-male samples in Philadelphia and Lon- don. Finally, the National Youth Survey is based on a representative national sample and includes about 30 percent of youths from rural areas. This broader geographic representation can be expected to lower arrest participation compared with the sam- ples from large urban areas in Philadelphia and London. TABLE B-36 Variations in Distribution of Sampled Offenders for Different Sampling Periods: An Illustration Proportion Annual of Offender Proportion of Offenders with Individual Population with Freauencies ~ A in Sample of Active Offendersb Offense Frequencies <Aa Rate, ~ Day Week Month Months Months Year Years Years 0.1 .05 -- .001 .001 .003 .004 .007 .014 .021 0.5 .26 .011 .020 .032 .056 .083 .121 .195 .224 1 .52 .050 .064 .106 .181 .256 .348 .469 .495 3 .67 .110 .139 .223 .352 .453 .545 .635 .653 5 .74 .156 .197 .308 .460 .562 .641 .713 .727 10 .85 .292 .369 .528 .678 .747 .793 .834 .843 20 .93 .506 .597 .757 .850 .882 .904 .923 .927 50 .99 .887 .928 .966 .979 .984 .987 .990 .990 100 .9995 ·9999 ·9999 ·9999 ·9999 ·9999 9999 9999 9999 - ~Assumes A is distributed according to a gamma distribution: a f(AIa, b) = b Aa~le~bA with shape parameter a = .25 and scale parameter b = .05 and p(a) the gamma function. The expected value of ~ is seven by a/b = 5 and the variance of is=2 = 100. -active offenders nave at least one ottense in the sampling period. The example assumes A is constant for individual offenders. The distribution of offenders in a sample is obtained by applying the probability of at least one event in the sampling period for an offender from Table 35 to a discrete approximation (using only the values of A in this table) of the continuous gamma distribution of the offender population and renormalizing the resulting distribution.

/ 356 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-37 Relationship Between Official Delinquency (Conviction) and Self-Reported Offenses in the Cambridge Study of London Boys Official Record of Convictions Convicted before ages 14-15 First convicted after ages 14-1561 Not convicted297 Average Number of Delinquent Number of Youths Acts Admitted at Ages 14-15 - 47 15.5 11.6 8.3 Ache measure reported is the number of different offense types admitted; it is only a rough approximation of variations in frequency rates over different delinquency statuses. SOURCE: West and Farrington (1973:165). TABLE B-38 Comparison of Official Record of Arrests with Self-Reported Offending for 1976-1978 in National Youth Survey Percentage in Each Self-Reported Offender Category Noncareer Career Arrest Status Nonoffendersa Offendersb OffendersC Totald Individuals 13 37 50 100 with arrests (N = 9) (N = 26) (N = 35) (N = 70) by 1978 Individuals 39 44 17 100 without arrests (N = 461) (N = 516) (N = 207) (N = 1,184) by 1978 Total N = 470 N = 542 N = 242 N = 1,254 a"Nonoffenders" report three or fewer delinquent acts per year and report no index offenses. bnNoncareer offenders" report any combination of annual rates except those classifying nonottenuers or career orrenaers. CnCareer offenders" report 12 or more total offenses per year or 3 or more index offenses per year in 2 or more consecutive years. dOnly subjects participating in all five annual surveys, consenting to a police record search, and having that search completed through 1978 were included, i.e., 73 percent of the 1,725 youth participating in the study and 53 percent of the 2,360 in the originally designated sample. SOURCE: Dunford and Elliott (1984:Table 6).

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS TABLE B-39 Distribution of Self-Reported Offending Groups by Official Contact 357 Percentage of Respondents in Each Self-Reported Offender Category: Community Sample Appeared in Institu Picked up Juvenile Placed on Institu- tionalized by Police Court Probation tionalized Sample (N = 942) (N = 64) (N = 73) (N = 35) (N = 20) (N = 245) Self-Reported Offender Total Categories Nonoffendera 9 Low frequency, minor High frequency, minors Low frequency, majord High frequency, majore Total 100 35 24 19 14 o 17 25 23 1 O O 21 17 10 25 22 3432 100100 o 2 17 29 37 100 15 30 45 100 7 80 100 Non offender: no reported offenses in Previous Year. a_ _ Plow frequency, minor: fewer than 48 minor offenses (the median reported) and no major offenses in the previous year. CHigh frequency, minor: 48 or more minor offenses and no major offenses in the previous year. -Low frequency, major: fewer than five major offenses (the median reported) in the previous year. Major offenses include motor vehicle theft, grand theft, aggravated assault, selling hard drugs, rape, robbery, and breaking and entering. thigh frequency, major: five or more major offenses in previous year. SOURCE: Cernkovich, Giordano, and Pugh (1983:Tables II and III). A similar pattern of more frequent high- rate offending for offenders with official rec- ords is reported in Cernkovich, Giordano, and Pugh (1983~. These researchers com- pared self-reported offending with self- reports of official contacts in a stratified population sample of youths aged 1~19 residing in a north central U.S. city. Self- reported offending in this sample was also compared with reports in a sample of insti- tutionalized juveniles. As indicated in Ta- ble 39, high-frequency, serious offenders were more likely to be found among re- spondents who reported official criminal justice contacts. Less than 15 percent ofthe total community sample were high-frequen- cy major offenders, but 32 to 45 percent of respondents with official criminal justice contacts were high-frequency serious of- fenders. This increased representation may result partly from a greater willingness of some respondents to report offenses and official contacts. In another sample of insti- tutionalized youths, however, in which knowledge of official contacts did not de- pend on self-reports, 80 percent were high- frequency, serious offenders, based on their self-reported offenses. The frequency of offending alone did not distinguish re- spondents with official contacts from other respondents. High-frequency, minor of- fenders were generally less prevalent among respondents with official contacts. Respondents with official contacts were dis- tinguished primarily by their higher fre- quency of major offenses (Table 39~. The

358 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-40 Percentage of Community Sample Reporting Criminal Justice Sanctions in Previous Year a Self-Renorted Offender Cateqories~ Low High Self-Reported Sanctions Nonoffender Frequency, Frequency, Minor Minor Picked up by police Appeared in juvenile court Placed on probation Institutionalized Total Low High Frequency, Frequency, Major Major 0.0 1.2 0.0 0.0 100 3.3 - 7.2 8.4 17.2 4.5 1.8 0.6 100 8.1 2.7 1.4 100 100 Resee definitions of offender categories in Table 39. SOURCE: Cernkovich, Giordano, and Pugh (1983:Table IT). greater likelihood of sanctions for these more serious offenders is reported in Table 40. The various studies reviewed above indi- cate the overrepresentation of high-rate of- fenders, especially in more serious offense types, in samples based on official criminal justice contacts. The much greater over- representation in these samples, compared with samples of self-reported offenders, re- sults from the generally lower rates for the criteria for entering a sample. A sample of arrestees, for example, includes individuals with at least one arrest in the sampling period. Similarly, active offenders in a gen- eral population sample include individuals who report committing at least one offense in the sampling period. While 32.5 percent of the community respondents reported committing at least one major offense in the previous year, only 7 percent reported be- ing arrested and 2 percent reported being institutionalized (Table 39~. As indicated in Table 35, the problem of oversampling high-rate offenders is more severe when rates for the criterion event are low. Look- ing at the column for a 1-year sampling period, for example, at low rates for A of one 9.0 5.6 3.4 8.0 0.2 7.0 00 offense or fewer per year, which are com- monly found for individual arrest frequen- cies, oversampling of offenders with ,u equals 1 is considerable compared with offenders with lower rates. For offenders with rates of one, the probability of entering the sample is .632; for offenders with rates of .5 or lower, the probability is no more than .393. The differences in sampling probabilities are less severe for rates higher than one. In general, estimates of frequency rates derived from cross-section official-record samples are not generalizable to all offend- ers. The estimated frequency rates are those of the more active and more serious of- fenders who come to the attention of the criminal justice system, and thus they con- stitute an inflated estimate of mean offend- ing rates for all offenders. Nevertheless, the distribution of frequency rates found in these unrepresentative, more serious sam- ples are often of greatest interest and value from the perspective of policy concerns, such as identifying effective means of crime control. The illustration of sampling probabilities in Table 35 indicates that one way to recluce

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS the severe overrepresentation of high-rate offenders in these official record samples is to increase the length of the sampling pe- riod. The length of the sampling period needed will depend on the anticipated rates for the sampling event. The lower the mean rate at which sampling events occur (e.g., one arrest or fewer per year per active offender), the longer the sampling period that is needed to improve the representa- tion of offenders with low arrest rates. For q, the arrest risk per crime, an individual ar- rest rate, A, corresponds to offense rate A = ,u~q. Except when A and q are highly nega- tively correlated, increasing the representa- tion of low ,u offenders in an official-record sample will also increase the representation of low A offenders, and arrestees will be- come a more representative sample of all offenders. Another strategy for increasing the repre- sentativeness of offender samples is to in- clude only offenders whose first event oc- curs in the sampling period (e.g., first arrest or first reported offense). This sample of starting offenders mirrors the distribution of offending rates in the total offender popula- tion. Excluding the recidivists in the sam- pling period eliminates the overrepresenta- tion of high-rate offenders. However, because of the overrepresentation of recid- ivists in offender samples, a larger sampling frame may be required to ensure an ade- quate sample for analysis when the sample is restricted to only first-time offenders in a sampling period. Also, the analysis of fre- quency rates for first-time offenders must be based on prospective data; the sampling criterion ensures that all offenders were inactive before the sampling period. Biases in Offender Samples: The Problem of Low Response Rates to Self Report Surveys While official-record samples are biased by the overrepresentation of high-rate of- fenders, there is some evidence to suggest that high-rate offenders are underrepre- sented in self-report samples. This is be- cause surveys of self-reported offending are generally characterized by reasonably high 359 nonresponse rates. In a follow-up survey of a 10 percent random sample of the Philadel- phia cohort (Wolfgang and Collins, 1978), for example, only 58 percent of the desig- nated sample were actually interviewed. The Rand inmate surveys involved re- sponse rates of 47 to 72 percent among California and Michigan inmates (Peterson and Braiker, 1980:6; Peterson et al., 1982:Tables 10 and 11~. In the National Youth Survey, 27 percent of the original sample did not participate (Elliott et al., 1983:11~117~. Participation dropped fur- ther as informed consent was requested of sample members so that official-record in- formation could be obtained. Farrington (1984) noted that "the boys [in the Cambridge study] from the most unco- operative families were significantly more likely to be convicted than the remainder, and the convicted boys tended to be more uncooperative themselves to the research." This same problem was noted in the fol- low-up of the Philadelphia birth cohort. Official delinquents were less likely to be interviewed (51 percent) than nondelin- quents (64 percent). Response rates also de- clined with increasing seriousness of of- fending: 63 percent one-time offenders were interviewed, compared with 53 percent for "recidivists" (two to four arrests), and only 38 percent for "chronics" (five or more arrests) (Wolfgang and Collins, 1978:32~33~. Nonresponse in self-report samples is more characteristic of more serious offend- ers. Exclusion of these more serious of- fenders from self-report ciata will contribute to underestimates of offense rates. Obtaining Valid Frequency Estimates from Self-Reports Assuming the problems of sample repre- sentativeness are resolved either by im- proved sampling techniques to increase representativeness or by properly narrow- ing the scope of application of sample re- sults there are other measurement prob- lems in developing rate estimates for the sample itself. In self-report surveys, these problems relate to obtaining accurate fre- quency estimates from individuals.

360 Obtaining accurate estimates of the fre- quency of offending is extremely proble- matic in retrospective self-report surveys. The earliest sel£report studies, relying on gross categories like "never," "occasion- ally," and "often," provided only crude es- timates offrequency. Such response catego- ries are likely to be highly variable in their meaning, both for different offense types and for different respondents. As a result, they cannot be translated into numerical frequencies with any precision. As interest in obtaining explicit fre- quency measures increased, the survey items also increased in numerical precision. Alternative strategies included use of nu- merical categories or reliance on open- ended questions the respondent completed by indicating an exact number of offenses. The most complex survey designs involve a mixture of categories and open-ended re- sponses. The respondent is first guided through a series of categorical choices to pick the unit of time that best describes his rate (e.g., yearly, monthly, weekly, daily). The respondent then supplies in an open- ended way the typical number of offenses in that unit of time. Studies that have in- cluded alternative frequency items in the saline survey instrument report very high variability in the As derived from responses to different item types. Individual offense rates in the first Rand survey of prison inmates (Peterson and Brai- ker, 1980:26), for example, were based pri- marily on respondent choices among sev- eral numerical categories that described the number of offenses committed in the 3 years prior to their current incarceration: 0, 1-2, 3-5, 6-10, or more than 10. As reported earlier, respondents who indicated more than 10 offenses were asked to supply an exact number in an open-ended question. In another part of the survey, an open- ended question asked respondents how many burglaries they committed in a typical month. The median rate of 4 burglaries reported for a typical month was substan- tially higher than the median monthly rate of only .24 derived from the total number of burglaries reported for the 3-year period. This extreme difference foreshadowed the CRIMINAL CAREERS AND CAREER CRIMINALS much higher estimates of A obtained in the second inmate survey, in which smaller units of time were used to report frequencies over 10 in the 2-year observation period. Similarly, the National Youth Survey (E1- liott et al., 1983:15) asked an open-ended question on the number of offenses commit- ted in the previous year. If the reported number of offenses was 10 or more, the respondents were also asked to indicate the frequency category (e.g., "once a month," "once a week," "~3 times a day") that best described their offending. The midpoint of the category was used to generate alterna- tive estimates of A. While the two responses were in general agreement, Elliott et al. (1983:117) reported that, "At the upper end of the frequency continuum, estimates based on the midpoint of the category are substantially higher than the frequency re- sponses given directly." In sum, estimates of A are extremely sen- sitive to the design of the survey item. Estimates for high-rate offenders are espe- cially vulnerable to design effects. The ac- curacy of various estimates of A thus hinges on the relative accuracy of alternative sur- vey items. There are a priori reasons for believing that As based on smaller units of time may be overestimated, and that As based on total offenses reported for a longer time period may be underestimated. Smaller, "typical" time units (e.g., monthly frequencies) are particularly vulnerable to overestimates when offending is intermittent, e.g., there are short periods of high offending. In this event, it is doubtful that respondents aver- age across periods of high and low activity to give the typical frequency of offending. It is more likely that frequencies during the more salient high-rate periods are reported `` . ... as typical. A subsequent analysis of responses to the second Rand inmate survey (Chaiken and Rolph, 1985) explored the intermittent char- acter of offending. The inmates reported more criminal activity for the period just prior to the current incarceration, regardless of the length of the available observation period. In this event, short observation pe- riods for some inmates would be especially

APPENDIX B.: ~SEaRCH ON CRIMINAL CANERS vulnerable to overestimates of frequency rates because those observation periods are more likely to include periods of spurts in activity and to exclude quiescent periods. Adjusting for spurts in activity did reduce the frequency rate estimates (see Table 15~. Use of longer time periods has the poten- tial advantage of smoothing out periods of high and low offending. As the observation periods get longer and more distant, how- ever, underestimates associated with mem- ory lapses are likely to become an increas- ing problem. Empirical research assessing the relative accuracy of alternative self-report measure- ment strategies is essential. This research should examine existing strategies to pro- vide a basis for calibrating existing esti- mates of A and explore the efficacy of alter- native techniques intended to improve the accuracy of estimates. A program of basic research that may produce results applica- ble to this problem is outlined in a report from the National Research Council on cog- nitive aspects of survey methodology (Jabine et al., 19841. Efforts to obtain accurate self-reports of A suffer from many of the same problems confronted in surveys of the frequency of crime victimizations. Considerable atten- tion has been devoted to identifying and resolving measurement problems in victim surveys. The design of improved self-report instruments for offenders may benefit from the accumulating store of knowledge on the design of victim surveys (see Penick and Owens, 1976~. One potential solution to improved self- reported frequencies may lie in a strategy of repeated interviews of the same persons that ask about their experiences in recent and short time intervals, like "last week" or even "yesterday." A strategy of repeated short-term interviews (four interviews in about 1 month) was used in research on noninstitutionalized opiate users in neigh- borhoods of Harlem in New York City (Johnson et al., 1983:9~. Annualizing fre- quencies from a short observation period, however, is vulnerable to biases arising from observations generated during spurts, or lulls, in individual activity. To avoid such 36' biases, the short-term interviews should be repeated at random intervals over a longer observation period, perhaps a year or more. Such a scheme might involve four or six interviews about the previous week's activ- ities, spread randomly over an 18-month period. Repeating the interviews for ran- domly selected weeks increases the chances that the weeks surveyed are a rep- resentative mix of spurt and lull periods for an individual. Spacing these interviews over a longer total observation period (e.g., 18 months) also reduces the testing effects of the survey on the behavior itself. The research on New York City drug users came close to this design through the use of subsequent 28-day interview cycles conducted 3 to 6 months after the first inter- view cycle. These later interview cycles, however, were only available for about one- third of the sample. Also, the results in this study may be limited by the small number of interview cycles per person and the rel- atively short total observation period of only 6 months. All respondents provided a min- imum of four weekly interviews. A more representative picture of the frequency of offending and drug use might have been obtained by scattering these interviews over the full observation period, rather than bunching interviews into cycles of 4 con- secutive weeks. Despite their limitations, the data avail- able for the small number of respondents interviewed in more than one cycle in the New York City study provide a basis for beginning to assess the merits of a repeated, short-term interview strategy. In particular, they provide a preliminary basis for exam- ining the extent to which intermittent spurts and lulls in offending are really a problem in frequency estimation. If evi- dence of spurts is found, the data can be used to provide preliminary estimates ofthe magnitude of the differences in offense rates in high- and low-rate periods as well as estimates of the usual duration of the spurts and lulls. From a practical point of view, use of a strategy of repeated interviews with the same persons is vulnerable to sample attri- tion problems. Also, reliance on multiple

362 interviews will limit the size of the sample of offenders that can be interviewed. Sup- pose, for example, that resources are avail- able for 1,000 interviews. Interviewing each person four times will reduce the po- tential sample size fourfold. Because of the smaller sample size, it then becomes espe- cially important to select for interview indi- viduals who are, in fact, more likely to be active offenders. Selection might be accom- plished through a screening interview to identify offenders in a much larger general population sample. This screening inter- view might simply ask if the individual had ever committed any offenses, or whether he or she had committed any offenses in the past year or two. The "ever committed" option is more likely to yield a more repre- sentative sample of high- and low-rate of- fenders. The longer prior observation pe- riod of this option, however, increases the risk that individuals passing through the screen will no longer be active offenders at the time of the screening interview. The shorter and more recent screening period (1 to 2 years) of the second option is more likely to yield currently active offenders. There will, however, be some underrepre- sentation of low-rate offenders passing through the screen. As indicated in the illustration in Table 36, for example, of- fenders with As of one or less represent about 35 percent of a 1-year sample com- pared with 52 percent of all active offend ers. The foregoing discussion focused princi- pally on alternative sampling and interview strategies for estimating individual fre- quency rates from self-reports. The discus- sion, however, also raised a more general issue in the design of research on individual offending pattems. It is clear from the con- sideration ofthe pros and cons of alternative strategies that no one strategy is well suited to measuring all aspects of offending behav- ior. Measures of participation require general population samples. Unless the samples are very large, however, they will not provide adequate numbers of active offenders for estimating frequency rates. The strategy of repeated interviews is certainly not feasible CRIMINAL CAREERS AND CAREER CRIMINALS for very large population samples. A multi- stage sampling strategy that addresses only a limited set of questions at each sampling stage would seem to be the best strategy. Administering a relatively brief screening interview to large population samples should be adequate to estimate participa- tion in crime. More extensive surveys ofthe details of offending could then be adminis- tered to the smaller sample of active of- fenders identified by the screening inter vlew. Obtaining Valid Frequency Estimates from Official Records Reliance on arrest history data in devel- oping estimates of frequency rates also in- volves measurement problems. These re- late primarily to the reliability of arrest records and the adjustments used to esti- mate crimes committed from arrests re- corded. Obtaining accurate estimates of offense rates from arrest histories rests fundamen- tally on official arrest histories. Of principal concern are record completeness and accu- racy. The analyses reviewed in this paper relied on centralized, computerized crimi- nal history files maintained by the FBI. Similar computerized criminal history files are increasingly being developed by state agencies (Bureau of Justice Statistics, 19851. With respect to estimating A, the problem of record completeness refers principally to the extent to which all arrests occurring at the 1~1 leve] are in fact included in the centralized files. Failure to include arrests in individual histories could lead to biases in estimates of A. Missing arrest reports are an especially severe problem for less seri- ous offenses. Entries in the FBI history files, for example, are usually triggered by submission to the FBI of a fingerprint record associated with an arrest or admis- sion to a detention or correctional facility. Many arrests for less serious offenses (e.g., for nuisance offenses like vagrancy, drunk- enness, disorderly conduct, trespassing) generally fall below the threshold for sub- mission of a fingerprint record and thus will not be recorded in the FBI's files.

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS For the very same reason, however, ar- rests for more serious charges, especially felonies, are likely to be far more reliably recorded in official-record data. Neverthe- less, an audit of the central criminal history records in Michigan found nontrivial, nonrecording of arrests, even for serious offense types (Michigan State Police, 1983a, by. This nonrecording was attributed to in- creasing use of pretrial release for offenders arrested on serious charges and to charge reductions at early case screening by pros- ecutors, both factors that reduce the likeli- hood that fingerprints will be obtained and forwarded to centralized repositories. Some portion of nonrecorded events can also be attributed to differences between local agencies and auditors in the classifica- tion of events, especially the classification of offense types. Such classification differ- ences are most likely to occur when locally recorded criminal events, based on crime categories found in local statutes, must be converted to some other crime classification scheme. Ambiguities in this conversion will likely result in undercounts of some offense types and overcounts of others. Such differ- ences in classification are likely to be rea- sonably uniform within a jurisdiction, i.e., the same "misclassification" will occur con- sistently. Inconsistencies in classifications, however, can introduce more serious distor- tions in comparisons across jurisdictions, since the same offense label may be applied to different criminal behaviors. Thus, a higher arrest frequency for some crime type in one jurisdiction might reflect a difference in classification rather than a difference in actual offending. Comparisons of offenders whose records are largely confined to a single jurisdiction are likely to be less vul- nerable to classification differences. Nonrecording of some events leads to complex biases in estimates of ,u derived from officially recorded criminal histories. These errors include not only underesti- mates from missing arrests for arrestees who are included in the history data, but also overestimates from the failure to include some low-rate offenders whose one arrest in the sampling year is not recorded in the arrest history data. The accuracy of the esti 363 mates computed directly from arrest history data thus depends on the relative strengths ofthese two opposing biases. To adjust ,u for arrestees on whom data are entered re- quires an estimate of the undercount of arrests forjustthose arresters. Extrapolating these adjusted rates to all arrestees then requires a further assessment of the degree to which ,u for arrestees who are missing entirely from the history data differs from that of included arresters. Research empir- ically investigating the relative magnitudes of these two potential sources of bias is needed, especially in jurisdictions where nonrecording in the central criminal history repository is widespread. Arrest history records are, of course, also subject to the variety of data transformation and data entry errors associated with creat- ing computerized files. These include er- rors in identification of the individual and coding problems in converting local de- scriptions of events into a uniform coding scheme. If they are not systematic and large, such errors are part of the normal variability in measurement. They should not seriously bias estimates derived by ag- gregating sampled individuals. They pose a more serious threat, however, to estimates for particular individuals. Another cause for concern in relying on officially recorded arrests to estimate ,u is the possible confounding of enforcement practices with offending behavior. The ar- restrate, ,u, reflects the contributions of both A and q, that is, ,uj = Aj X qj for any offender j, and high values for ,u may therefore be due to high values for A or q. This confound- ing of enforcement practices with offending behavior is especially problematic when comparing ,u across crime types or across jurisdictions, both situations in which en- forcement patterns are likely to vary. It may also be a problem in comparisons among offender subgroups if q varies systemati- cally with offender attributes. One solution to the distortions intro- duced by the arrest process is to invoke explicit estimates of q in order to estimate underlying offense rates from available es- timates of ,u. This procedure has been used in a preliminary way, for example, by

364 Blumstein and Cohen (1979~. To accommo- date the main sources of variability, esti- mates of q are developed separately by crime type and for the specific jurisdictions being studied. Nevertheless, these results are potentially limited by applying a single arrest probability to all offenders within any crime type and jurisdiction. Such estimates fail to address systematic variability in q with offender attributes and will distort comparisons of A across the groups. A rela- tionship between individual offense rates, hi, and arrest probabilities, qi, would be especially problematic. The results available to date generally fail to find systematic variations in q, especially with demographic attributes. Such results are consistent with the practice of using a single estimate of q to derive A from A. While encouraging, further research empir- ically investigating the patterns of variation in q is needed, especially systematic varia- tions with A and with other offender at- tributes. These analyses will benefit from combining self-report data on crimes com- mitted with data on arrests from self-reports and official records for the same individuals, as was done in the second Rand inmate survey (Peterson et al., 1982) and the Na- tional Youth Survey (Dunford and Elliott, 1984~. Such data provide an opportunity for developing disaggregated estimates of q that will permit more refined adjustments to estimates of ,u derived from arrest histories. CRIMINAL CAREERS AND CAREER CRIMINALS served to commit exactly three offenses in a year need not have the same underlying rates. When offending is probabilistic rather than deterministic, an offender with an an nual frequency equal to three need not commit exactly three offenses each year. Instead, his observed number of offenses will vary from year to year, but in the long run if he is observed over a long enough period, it is expected that his observed mean annual rate will be three. Thus, be cause of the probabilistic nature of offend ing, there is some likelihood that offenders whose A is 10 or 20, for example, will have a realization of only three offenses in a year, as will some offenders with A of only 1. Relying on only the observed realization of offenses (or arrests) for each individual to estimate his frequency is thus vulnerable to error. Instead, estimates for individuals should be developed using statistical esti mation techniques like those proposed in Rolph, Chaiken, and Houchens (1981) and Lehoczky (Volume II). Those techniques better approximate the likely underlying frequency rate for an individual by using information on the distribution of offending in the total sample to adjust individual rates estimated from the observed number of offenses for each individual. A Sampling Strategy for Detecting High-Rate Offenders This review has highlighted some impor tant difficulties in measuring offense rates Measurement Problems In Deve70p~ng for individuals. In addition to these direct Individual Frequency Estimates measurement problems, other difficulties plague efforts to identity differences in ~ using more readily available and directly measurable attributes of offenders (see, for example, the discussion of prediction and classification in Chapter 6 and in Gottfredson and Gottfredson, Volume II). Nevertheless, for important policy and re search reasons, there is increasing interest in being able to identify high-rate offenders within the general offender population. The earlier discussion of sampling biases highlighted the problems in obtaining rep resentative samples of offenders. Precisely those features of sampling processes that Most estimates of frequency rates aggre- gate data over individuals to generate esti- mates of the mean or median rate for a sample. However, there are occasions when distinguishing among offenders who have different frequency rates is important, so that estimates for particular individuals are needed. The data available for an individual rep- resent the particular realization of an of- fense history that is observed for an individ- ual and are distinct from the underlying rates that gave rise to that realization. For example, two offenders who are both ob

APPENDIX B: RESEARCH ON CRIMINAL CAREERS contribute to sampling bias problems- par- ticularly the requirement of an event during the sampling period can be turned to ad- vantage to identify high-rate offenders with reasonable accuracy. Capitalizing on the overrepresentation of high-rate offenders, samples drawn in short time periods, say a week or a month, will include dispropor- tionately large numbers of high-rate offend- ers. The shorter the sampling period, the lower the false-positive errors of including low-rate offenders. In the illustration in Ta- ble 36, for example, only 25.3 percent of offenders in the 6-month sample have an 1 ~· . ~. ~. ~ 365 on the adequacy of these usually unstated assumptions. Because the available observable data are only indirect indicators of actual crimes committed, improving the precision oftech- niques for measuring these observable data can only go so far in improving the accuracy of estimates for the underlying, but unob- served, crime process. Improved knowl- edge about unobserved crimes committed requires explicit behavioral models that link the unobserved crime process with the observed data. With explicit models, the adequacy of estimates can be measured nua1 frequencies greater than 1U; in the 1-week sample of the same population, those high-rate offenders increase to 63.1 percent. A further refinement ofthe selection strat- egy might involve drawing a 1-week sample (e.g., those arrested in 1 week, or those admitting to crimes during the past week in a general population survey) and then look- ing at the offending in that sample in an- other week, say 1 or 2 months later. The persons identified as criminally active in both samples are very likely to be high-rate offenders. Such a strategy has the advantage that it relies on only observed offending or arrest experiences to identify high-rate of- fenders and does not invoke other poten- tially controversial variables, such as vari- ous personal attributes of offenders. Explicit Behavioral Models of Offending Various models of individual offending are implicit in virtually all estimates of fre- quency rates, regardless of the data source used. Because of the inherent difficulties in obtaining direct observations of crimes committed by individual offenders, esti- mates of frequencies rest on other observ- able data, like arrests and self-reported crimes, as indirect indicators of the under- lying crime process of interest. The various estimation strategies that are applied to these indirect data rest fundamentally on models characterizing both individual of- fending and the processes that give rise to the observable data. The accuracy of the frequency estimates that emerge depends against tne adequacy of the model's as- sumptions. Models of individual offending have moved from undifferentiated treatments of offending, like those underlying such tradi- tional aggregate measures as per capita crime rates and recidivism rates, to differ- entiated characterizations that partition of- fending levels among the various aspects of a criminal career, especially participation, frequency, crime seriousness, and career length. In their early formulations, these differentiated models of criminal careers relied on a number of simplifying assump- tions, principally that individual careers are stationary over time and often invariant across offenders. This simple characteriza- tion of careers underlies most currently available estimates of frequency rates. More recent developments have begun to enrich the basic model to better accommo- date the complexities of real criminal ca- reers (Chaiken and Rolph, 1985; Flinn, Vol- ume II; Lehoczky, Volume II). One issue of concern has been possible nonstationarities in A during individual careers. Two forms of nonstationarity have been addressed: spurts in criminal activity as offenders move be- tween active and quiescent periods, and changes in frequencies as offenders age. These more recent model developments represent considerable conceptual ad- vances over the very simple model of crim- inal careers that underlies most available estimates of the various career dimensions. As the models of criminal careers continue to reflect more fully the underlying behav- ioral processes, there will be increased con

366 CRIMINAL CAREERS AND CAREER CRIMINALS fidence in the credibility of empirical esti- mates derived from those models. OFFENSE SERIOUSNESS DURING CRIMINAL CAREERS The study of frequency rates reflects a terns ot ottense seriousness has obvious implications for crime control policies. Some crime control policies may be con cemed primarily with reducing serious, predatory crime. Knowing more about pat tems of individual offending may be useful in focusing these crime control efforts on concern for the intensity of individual of- offenders most likely to be engaging in fending how often offenders commit those offenses. If there is specialization in crimes. The analysis of offense seriousness, ~ ~' in contrast, is concerned with the pattern of offense types that are committed. At a static level, the offense mix for different offenders might be compared to assess differences in the distribution of offense types. Identify ing offenders who are more likely to engage in serious, predatory crimes is of particular policy interest. At a dynamic level, analyses of offense seriousness focus on the changes in offense types during criminal careers. Two issues of concern are the extent of "specialization" and the extent of"escala tion" in offense types as careers progress. Specialization refers to the tendency of in dividual offenders to repeat the same of fense type as offending continues. Escala tion, by contrast, is a tendency to move to more serious offense types. The opposite pattern is also possible successive of fenses decrease in seriousness. In both in stances, the emphasis is on changes in the mix of offense types committed, not on the frequency of offending in each offense type. These alternative patterns are found in commonly held conceptions of criminality. Criminals are sometimes pictured as spe- cialists who may try several different of- fense types until they find the particular offense that best suits their skills and oppor- tunities, which they then adopt as their specialty. Probably the most commonly held view of criminality is one of escalation, i.e., individual offenders who remain crim- inally active become gradually more hard- ened and dangerous in their commitment to crime and move to increasingly more seri- ous offense types. An alternative scenario is one of de-escalation, e.g., the serious youth- ful offender who continues offending moves to less serious offenses and becomes a chronic, nuisance offender as he gets older. Accurately identifying prevailing pat those offense types, tor example, then ot- fenders who remain active and who have records in those offenses are prime candi- dates for committing future offenses of the same type. If offending escalates in serious- ness for offenders who remain active, then crime control efforts are most useful later in an offender's career; if offending de-esca- lates in seriousness, they are most useful early in an offender's career. A number of studies that have empirically investigated offense seriousness are re- viewed in this section, in particular their research approaches and reported findings. Whenever possible, data from the various reported sources have been reanalyzed to address a common set of issues, using the same statistical techniques on each data set. Following that, some of the methodological issues in research on offense seriousness are discussed. Review of Empirical Research The study designs and reported results of the studies included in this review are sum- marized in Table 41. The studies all rely on official-record data. Some use the broad cat- egory of police contacts, which may include police stops for questioning that do not lead to arrest or any formal charges but which are recorded in police files. Others rely on ar- rests or convictions, which more directly link an individual to a particular charged offense. The only explicit criterion for in- cluding studies in this review was that they examine empirical data on offense types for some population of offenders. The studies included are not meant to be exhaustive. , However, they do cover a variety or ap- proaches to the topic. . . ~ Research into offense switching relies on official-record data because those represent

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS the only reliable source of information on the sequence of different types of offenses in criminal careers. Data on actual offenses committed are usually obtained from self- reports by offenders. Short of having of- fenders keep daily journals oftheir criminal activities, or obtaining self-reports of of- fenses through frequent and regular inter- views (e.g., daily or weekly), accurate chronicles of the sequence of all offenses committed by individual offenders are not available. The usual self-report interval of a year or more is too long to obtain reliable information on the sequence of offenses during the reporting period for individuals. Repeated surveys ofthe same sample, how- ever, could provide self-report data on the mix ot crime types In successive periods.~9 It is important to stress that analyses that rely on official data reflect offense serious- ness for only official contacts. The re.c,~lt.c are not likely to be representative of of- fenses actually committed. The distortion in offense types between actual offenses com- mitted and official contacts arises from dif- ferences among offense types in the chance that an offense will result in an official contact. Offense types with a high arrest probability, like murder and aggravated as- sault, are more likely to appear in official- record data. Correspondingly, offense types with a low arrest probability, like larceny or drug law violations, will be underrep- resented in official records. Differences across offense types in the time interval between a crime and an arrest for that crime may also distort the ordering of offense types. While commission of a robbery may precede a larceny, a shorter time to arrest, if one occurs at all, for larceny than for rob i9The research on the criminal activities and drug use for a sample of New York City offenders (Johnson et al., 1985) is unique in its use of almost daily interviews. However, no analysis of Me se- quence of offenses is reported. The observation period (averaging only 2 months) and the sample size bust over 200) may be too small to yield a large enough sample of offenses for detailed analysis of sequences. The National Your Survey (Elliott et al., 1983) used repeated annual surveys win We same sample, and these data could be analyzed for yearly changes in offense seriousness. 367 bery might result in a larceny arrest preced- ing a robbery arrest. Because of differences in the probability and timing of arrests for different offense types, the results of analy- ses of patterns of offense seriousness based on official contacts cannot be generalized to crimes committed. Because the studies included in this re- view rely on official-record data of offense seriousness, they are necessarily based on samples of offenders known to the criminal justice system through some form of official contact. In some studies, the basis for enter- ing the research sample was an official- record event during some sampling period of a year or more. Those studies involve cross-section samples of arrestees (studies 3 to 5 in Table 41), of juveniles processed or adjudicated in juvenile court (studies 7 and 8), or of inmates (studies 2 and 9~. By virtue of the sampling event, all members of the cross-section sample are presumed to be criminally active during a common time period. Two studies both of juveniles- rely on birth cohorts from the general pop- ulation (studies 1 and 6~. Analysis of offense seriousness in these studies focuses on the subsample of offenders within a cohort who have at least one police contact. All the studies in Table 41 involve either retrospective or prospective longitudinal data on sequences of offense types for indi- vidual offenders. Such data are essential for analyzing switching patterns among offense types in successive events. Dynamic pro- cesses like "specialization" or"escalation" make no sense except in the context of longitudinal data on individuals' event his- tories. Even in studies in which samples of offenders were drawn from cross-sections of active offenders in some observation pe- riod, the analysis is based on longitudinal data for the cross-section samples of of- fenders. Some studies not included in this review report only results on the offense mix for a sample of offenders. These one-time distri- butions over offense types provide some insight into the frequency of different of- Sense types, highlighting which types are rare and which commonplace in a sample of offenders, but they provide no sequence

368 A A U] o . - U] a) Q a) o o s u to u . - s~ · - . o U] m to ED U] to {Q 0 ·r1 0 ~ U] {Q . - a, {Q ~ c: a) ¢¢ 0 0 · - 0 · - Id a) U] Q o a, · - en U] A :, #A U] ,' ~n _ O ~ t O tJ) O ~ O .,1 ~Q U] ~S ~ ~ C o' - U ~ ~ =' - ~ ~ ~ ' - ~O ~S ~ . - · - (L) O U _I ~L) O 0 53 (V (L) ·^~ O (V u a. c: ~ n' eG tU ~ £ ~ ~ ~ c: ~ ~ ~ · eq CQ ~ ~ U ~ - O ~ · - ~ ~ ' - - - u ~ ~o ~ ~ ~ ~ ~ ~ ~ u, O ~ O u, ao O ~ eq ~ U ~1 ~ O ~ ~ ~ O O · - ~ U U ~0 o ~ ~ ~ ~o ~=- - Q ~ u u u ~ ~ · - ~ ~a' ~ 0 ~ ~ ~ ~ ~ ~ ~ ~ ~ u - O ~ ~ ~ O U] ~ ~ ~ ~ ~ · - ~ ~ ~ O ~ · - ~ ~ ' ~ ~ ~ ~ ·e U ~ U ~ ~ ~ ~ ·-l U ~ eq ~ ~ ~ ~ ~ ~ O ~ U ~ ~ ~ ~ ~ ~ ~ O · - ~ eq ~ o ~ U] o ·^ - eQ ·- ~ ~ O S~ ~ U ~ ~ U ~ ~ ~ ~ ~ · - ~ ~ ~ s U ~ ~ o U ~ ~ ~ U] ~ o ~. - ~ ~ o ~ o ~ ~ ~ ~ ~ ~ X ~ ~ U ~O ~ s ~ O ~ ~ ~ ~- - o ~ ~ O O U, ~ tU ~ O t' ~ ~Ce ~ O - 3 o .^ o 1 0' a' 1 U -c ~ ~·^ O ~r O ~ O ~ ~ ~5 U~l ~ ~' ' - ~ ~ o ~· - ~ tU ~ ~ ~ O ~ ~ ^ ~ ~O Q ~ ~ ~ ~ O 0 a) ~ ~ ~ ~0 ~ · - ~ ~ (Q , S U ~ 11 ~ ~ 0 11 ~S ~ ~ V ~ o-- U] ~ ~ o ~ c~ ~ U] CQ . - =~ - ~ ~ ~ z ~ ~· - a u 3 ~ ~ - ~ ~ ~ - "' a, 3 Ll o ~ s~ ~ ·- O a) 0 ~ a) ~ ~ 0 0 u a, ~ u, [Q ~ ~ ~ ~ 11 ~ ~ e ~ c~ ~ ~ ~ O ~ 1 ·,, 0 ~ ~ a~ 0 u~ z ~ 0 ~u ~u ~ ~ ~ ~n 53 - ~ s ~ ~ O 3 ~ O u ~ tn ·- ~ 0 0 ~Q ~ ~ ~ u a) a)~- ~ u u a) ~ O u 0 ~[Q ~ ~ ~ ~ u Q. U ~ O ~ ~ ~ ~ ~ ~O .- ~ ~ ~ O ~ u' ~ ~ ~ ~ ~ ~ O ~: ~ O £ ~ ~ u O _~= ~ ~ ~ S ~ ~ 1 ~ ~ U ~ O ~ O ~ ·- ~ mm O O ~ mm ~ ~ ~ m~ O ~ O P,-~ ·~ "~ ·~ ~ ~ U ~ ~ ~ ~ ~ ~ ~ U eq · ' 0 ~ ~ ~ tD · - 3 ~ '= o £ ~ _ ' s ~ ~ 'm 0 - ~ ~ a' U ~ 0 11 ~ ~ · - 11 ~O s ~ U ~ Z · - Z - ~ ~ ~ - 11 U .,' tR U, a, Q a) ~ C~ Z Un to .,' d. _ ~1 S~ 0 a, ~ ~ - m ~ sO s ~ o ~ ~ ~O C) E~ O Q-~, ~c) oo a' _ ~a, O CD - ~ _ s U] CQ 1 O ~ 3 S-^ O ~ 1 :3 3 · - ~ ~ v ~ ~ O ~O O >, ·^ ~ ~U ~ U] o O. S~ ~ ~ ~ O ~ ·e O S~ O S V ~ CO [Q (Ll U ·- ~ ~ O · - U ~ · - ~ ~ ~ ~ ~ ~ u ~ ~ · - 3 ~ ~ ~1 ~ o 0) ~d ~ U Q 3 · - O ~ c- - ~ ~ ~ ~ ~· - ~ ·,1 o- - o ~ · - ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ V ~ ~ ~ ~ · - - u 3 ~Q u, a

369 ,- ~o ' · - Mu In Qu, ~ ~ ~I ~ ~· - ·,~ '~ .s: I ~ 3 ~lo, :: al 3 so ~ I S ~ o ~ ~ u, ~ ~u, so ~ up ~ up a: c: ~ O In ~ en ~U] ~0 ~ x ~ ~ ~ 0 ~ 0 ~u, ~ ~ ~ a, ~ ~ ~ ~ ~a) v ~ ~ ~ :> ~ a) ~ ~ ~ s 0 ~ C: 0 ~ ~ [Q C: a) ~ up U] O O (Q O ~ ~ O ~ CQ ~ ~ O ~[Q ~ O · - ~ ~ ~ S ~ ~ ~ ~ · - u, ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ 3 ~ ~ ~ ~ o ~ ~ ~ ~ ~O S ' al up -- ~a, (~) C,) .~ a) ~ al ~ ~ ~ 0 a) ~ ~tQ ~ ~, ~ ~ ~ ~ o ~ ~ ·- ~ ·- 0~ ~ ~ ~ ~ ~ ^o ~ ~ O · - ~ ~ · - >- 0 ~ ~ ~ ~ ~ ~ O ~ ~ ~ a) ~ ~ · - s ·^ ~ ~ ~ ~ 0 [Q ~ ~ · - s C) O {,) ·- N _I ~ ~ U] ~J O <1) ~ ~ ~ ~ ~ ~ ~ ~ O ~ ~ ~ ~ ~3 ~ ~ ~- - ~ ~ O ~· - ~ ~ W O ~ =- - ~ ~ ~ Z. - a) ~ 0 ~ ~s ~ ~ u ~u~ ~ a ~- ~U] ~ ~J ~ ~ 0 u~ ~ ~ O ~ ~ ~ m~ ~ ~ ~ ~ ~ ~o ~ O ~- - · - 0 ·- u] J~ O] u] ~GS ~ 03 O G~ tu 0 ~ _I 0 tu 0 0 0Sat ~t~) 4 ~3 ~J ~ · - ~ · - ~ ~ ~ ~ ~ ~ ~N ~ ~ ~ ~· - ~ ~ ~O Ed ~ Q. S ~ _GS O GS E~ O ~·- O ~ ~ ~ O ~O O O ~=: ~ oo U] · - {Q Q r~ a) 3 a, ~ ~ 11 U] ~ a' 1- .. ~ ~ ~ ~ ~ 0 a, Z ~ O ~m a) ~ {Q al O ~ ~ ~ S ~ ~ S ~ ~ O a) t) ~ ~ ~ v a) ~ ~ ~ s: O ~ O ~ ~ - ~- ~ ~ ,' ~ ~ ~ ~ ~ 0 3 ~ ~ s,. u, ~ ~ a) ~ ~ ~ O . - ~ ~ ~ ~ 0 ~ O 0 ao S ~ s ~ ~ ~ a) ~ 3 ~ u~ ~ a) ~ ~ 0 eQ a' ~ W ° ~ ~ ~ ~ ~n O ~ ~n ~ ~ ~ ~ ~ 0 ~ O a) o ~ ~ ~ =~ s~ I ~ a) tQ ) ~1 0 0) 0 ~_1 ~ ~ H ~ ~ ~ ~ ~ ~ s~ ~rs ~ ~ ~ m a, ~ ~ ~ ~ O :^ a: P4 ~O - 3 ~ ~: P~ ~ O ~D ~ cn ~ ' _ ~ .,' I ~I ~ ~ a' ~ ~ ~ a) . - ~ ~0 ~n 0 ~ tn - e,l ~ ~ ~ O a) ~ ~ 0 ~ Q ~ O ~ ~ t~ ~ ~ ~ s s~ ~s O P4 11 ~·d -~ ~a~ 11 ~ . a) ~ ~ ~ ~ ~ ~ u, ~ ~ ~ ~ ~ u, ~ ~ ~ z ~ ~a s O Z U] O ' - ~ ~O- Ll ~ ~ ~ U] .,, _ ~a, ~ s~ 0 ~ ~. - 0 a, 0 ~ c: =: s~ ~ ~ u, ~ ~ ~ 3~- t) u, ou ~ O ~ rs1 3 ~n ~ 3 s ~ ~ ~ O ul ~ O ~ O c: c: 3 u~ ~ a l ~ ~ u~ ~ ~ s~ O ~I ~ ~ :> u, ~ ~ a, .- ~ ~ ~ ~ ~u, ~ ~n ~ ,~ ~ ,,,, u, ~ ~ ' ~ ~ ~ 0~ ~u, ~ a) s~ ~ tu 0 ~ . ~ ~ a) ~ ~0 ~ ~ ~ ~ ~ ~ c' ~ ~ O ~ O ~ O ~ N ~ ~ ~ ~ ~ ~ ~ S C-) td ~ '- ~ ~ ~ ~ ~ ~ O ~ ~ ~ ~ ~ Q a' ' ~ Z ~ ~ _ ~_ O e0 ~n ~\9 ~ CD a' . - ~ ~ ~ p,, ~ _ .,' _ a, 0 P. ~

370 U] U] a) U] o . - 0 a) U] in . - ~ U] U] o o ." a) o . S~ 1Q Q o S~ ~a a) a,1 o C) m E~ ~rl U] CQ V .,' cn S ~Q a) ~n a ~ ~ ~ U] ~ Q _I ~ CQ .,~ S~ =~- ~ ~ O ~ ·,1 U] - ~ ~ ~ O L0 ~ oo O E - ~U] N _I = n5 ~ ~ ~ O a, ~ ~ v _ ca ~ 0 ~ ~ ~ 0 v~ U] ~ ~ ~ ~ a~ ~ ~ O _ O V ~ - = a,1 = ~-I GS ' ~ ~ ~ 0m ~ ~ ~ ~ ~ V ,1 0 ,.o ~U] S O ~ ~ X ~, ~ ll'- ~ v ~ 0 -a, - O ~ ~ U] ~) ~ 0 1 a' 00 ~Q 11 '- ·- U] ~ 3 ·,1 0 S o-1 ~ ~ ~ ~ ~ ~ V ~ ~ ~ a) s a) ~ ~ ~ - ~ ~ O a) Q' ~ ~ ~ n' £ ~ = - ~ ~ ~ ~ ~ .,, ~ U] O ~ S~ ~, ~ U) ~ ^ ~ ~ ~ -_ O ~ ~ £ ~ ~ > == v "l s, ~ _I ~ ~ ~ cn ~ ~ .,' O a, ~ ~ 0 ~ s ~ , - ~ ~ ~ ~ 1 ~ ~ ~ ~ tn a) ~ ~ ~ ' ~ v ~ ~ 0 cn O ~ ~ ~) 1Q ~ ~ O O ~ ~ V ~ 1 u' _I ~ s a) a, ~ tR 11 ~ ~ .- S~ ~ ~ ~ O ~ O ~ ~q U, 0) ~ ~ 0 e-~ ~ ~ s t) ~ ~z; ~ ~ O Q 0 a; P. 115 ~ 3 ~ o ~ a o- ~ o4 0 ~ I ~ 0 a, a, ~ ~ ~ _ ~ ~ ~ ~ ~ ~ 11 ~ s ~ ~ ~ ee ~ ~ ~ ~ · - ~ ~ ~ a) Q~ ~ ~ ~ z Q O ~ ~ ~ ~ ~ ~_ - ~ ~ ' ~ ~ ~ ~ ~ ~ O ~ s ~ ~ ~O ~ S ~ .~ O ~ ~ $~ ~ .¢ ~ ~ o ~ ~ ~Q ~ ~ ~ .- .,, 'm ~ ~U) ~ ~ ~ -~ ~ ~ s ~q ~ ~ Q s ~ ~ ~ :~: 11 ~ ~ ~ ~ 0 ~ 3 v ~ ~ ~ u' O tn ~ ~ ~m ~ ~s ~ ~ - = ' ' ~ Z ~O ~ 3 0 es ~ s~ :^ ~ O ~ ^ ~ ~ - ' Q ~ ~ o =-- ~) 0 a) m ~ ~ ~ -m ~Q n5 ~ ~ ~ ~ 0 ~ s ~ ~ 0 ~ ~ ~ S ~ ~ u, Q ~ ~ ~ ~ ~ ~ ~ ~ ~O - - ~ mo o O ~ s-' Q u, ~ ' O ~ u, ~ ~s ~ a, ' ~ ~ 0 0 es S~ a, a, aml O £ ~ ~O ~ ~ `: m c' ~: ~ ~ ~ O ~ ~ ~ U] cn ~ ~ O\e,1 ~ ~- U, a, ~ 0 ~ _ a, ~ - ~ - .^ r~ U) S CQ ~ 0 ~ ~ ~ ~ ILl U) - ~0 m :~

v be ~ O A) 1 C>, ~Of ~tR 3 U] . - ~ ~ ~O 4, ~v ~al a) 0 ~ ~ O to a) u, O v O ~ a, ~· - (Ill ~ ~ ~Q. ~ ~ ~'m O ~ ~ ' - en V ~ ~ P. O ~ ~ ~ ~ ~ ~O ~ ~ ~ O ~ ' - S ~ ~ ~ ~ ~ ~'- ~O ~ ~ ~ ~ O a~ o O -~ ~ 41) ~ ~ ~ ~ ~ 0 0 ~.- 0 0 O -l ~ X ~ (L) ~ O V ~-I -I Q4_I a) a) ~ ~ ~ ~0 ~ ~ ~ - ~ ~ ~ v ~U] O U] ~ ~ ~ V ~ ~ ~ ~N U] (L) ~ ~ ~ ~ ~ ~ V ~O O O ~ ~ ~ ~ ~ ~ ~ V ' ~ ~ ~ ~ ~ ~ - = ~ m^ ~ ~ - == ~ m Io ·-1 U] ~ ~ U] ~ ~ ~ ~ ~ ~ ~ S ~ ~ V ~ ·- ~n ~ ~ ~ ~ ~ O ~ ~ 0 V C O 3 ~ ~ ~ ~ ~ ·e ~ ~ ~ H O ~v ~ ~ ~ ~ ~ O n5 ~ ~ eQ ~ ~ V ~ V ~ ~ ~ ~ ~ O ~ ~ ~ ~ ~ Q ~ ~ ~ ~ ~ ~ V O ~ ~ O V ~ ~ ~S ~-~ ~ ~ ~ O O O O O~ ~ ~ ~ ~ - ~ Z 3 V ~ O ~ ~ ~ ~ ~O o~ ~1 ~ ~ ~ c' - ~ ~ ~O U] V ~ U] O ~ Q ~U] te o ~ ~ - O ·-l U] ~ O tQ ·e ~ ~n ~n ~, 0 ·- v ~ ~ ~0 s ~Q 3 O ~1 1 ~ O ~ 0 3 _ 3 tO o O O ~aJ V ~ ~ ~ ~ V ~ ~ O O ' o ~ ' V - ' - _~ ,1 C~ n~ oo ~ S ~ Sm ~ m~ ~ = ~V t-~ ~ 3 a) ~ 4) ~ O~ ~l eq -I ~ ~ ' ~ ~ ' O ~ ' 3 c`] ~ ' _I - O _1 ~ "l ~ ·' V eq ~ ~ ~ ~ ~ ~ ~O o ~ . ~ ~S ~ == -~ _I ~ O _' ~ O O co O ~ ~ ~ S O O U, ~ =3 ~ 3 ~ o tn 1 ~q 0 o C) ~ O `: . - . - ,' o :~ ~i - ~1 O V U] - o ~ ~, eq _I ~ _ m s .,. o s ~ s . - v ~ .,1 0 ~ s o ~ ~1 v ~ ~l o ~ ~ r eq o. o ~ arl ] ~ ~ v a,1 vs ~ a, ~ ~ ~ o 3 C) 0 oS tQ O ' ~ O _ ~ . ~ § . - .,, ~Q . - ·'S O _I ~ 3 ~ O ~ V ~ ~ .,1 0 V S O U, C) ~ -I - t: O C) ' ~ · ~= :' O _I ~ S ~) H ~ 3 371 ~Q ~n a, Q]

372 . - o c) Hi' a ha a o . - O a) U) SIR ~ a,) to en a) at, o o . - Pi o . - a) a Q o a) A. en G) 1 U] O tU ~as ~ O so ~ ~ ~ en 0 u, ~a, ~S ~ ~ ~ 3 ~ ~= a) "l ~0 ~ ~ ~ s O ~ ~ ~ ~ == , ~to\ ~} ~- ~ ~ ~ O ~ ~ ~ ~ ~ ~ O A, ~ ~ en ~ O ~ 3 ~ ~ o so ~ ~ .- ~5 ln ~ (Q al ~ ~ ~ ~ a, ~ ~, ~ ~ ~ .^ ~ u, ~ u, ~ O ~Q' 0 ~a, ~ ~ 0 3 ~- 0 `: ~ · - ~ ~ s~ n5 ~ ~ dP ~ ~ I _ N >t U] U] td ~ ~ ~ ~ ~ ~ a~ ~ o~ ~ ~ ~ ·- . - ~ tQ S~ ~ o~ ~ ~ _ ~ ~ a) ~ ~ ~ ~ a, {Q ~ ~ ~ ~- 0 ~ 0 U] ·e ~ ~ ~ ~ ~ a~ ~ ~ ~ O ~ ~ O .- ~ u, 0 a _1 a, u ~O GS ~ 0 ~1 0 0 0 _I U] -1 00 ~ 0 G5 ~ a) ~ ~ 0 ~ ~ ~ ~ ~ ~ 0 :^ a, ~a, ~ ~tU 3 X Q. ~ ~S O O ~ O O ~ ~ ~: ~Q ~ ~ ~O ~ ~ ~ ·- O ~ ~ ~ ~ ~ V ~ O · - ~q eq ~ a) eQ a, o a) . - . - O O o ·. - ~Q 1 u, ~Q o O a a _~ ·, u' ~ r~, m0^ 1,q .- a' a, ~ ~ .,_, ~ _ ~O u' O · - o , - O. tQ u, ,' 3 tQ .= Q ~n I ~rl e: :^ . ~ U] tQ C, ~o U] a, o U] . - tQ ~ . - ~ :^ a o C) £ ,: o4 ~ ~ .,' X ' Eli N '- O $¢ o U] U] S~ CQ ~ O - . - t: ·,' i, U] 411 eQ ~ S ~ - 3 u, u, O X ~ El3 ''t - ~u s~ o a) ~5 ~Q O O S 3 11 z , O ~ O ~n tQ ~ ~ ~ 0 o. - ~ ~, o . - C) ~ tQ O u, aJ ~ C: u' ' 3 0 ,' O U] (Q ~ ~ ~n O O ~ . - P~ GS S-l Q U] I o ·. o S~ o I ~ . - C) ~ U] O C) .,1 J~ O t) U] ~ 3 . - ~Q .,' tn

373 0 In ~ a eQ · - 1 ~O c: ~ .- ~ o s . - ~ ~o ~ ~s o ~.,' ·,' ~ ~ I: ~ ~o o V ~ ~ ~ ~ Cal · - o ~ ~ ~ ~ ~ ~ ~ ~ ~ ' - "= ~ ~ o ' ~ O ~ CQV ~ O Cal ~ ~ ~ ~ ~ o - ~ ~ s ~ ~ ~ ~ ~ ~ o As ~ ~ ~ ~ ~ ~o ~1 ' - t) ~ ~ ~ ~ N _ _ ~ ~ ~ ·-~ ~ ~ ~ ~ · - ·- 0 ~ a' 0 ~ ~ ~ ~ ~ ~ ~ ~ ~ 0 3 ho' to ~ Al es ° m-- ~ ~ ~ O ~ ~ O ~ o] ~ ~ ~ ~ ~ ·- <' ~ a, o - - ·- s~ ~Ce O aJ (L) ~5 dJ O O 0) U] Q ~ S ~J ~ ~ 03 ~ aJ ~ u] O Q~ ~ Q ~ ~ S Z O U] C,Q ~ ~ fIS ~N ~Q. U] ·~1 0 · - ~1 ~ <1) p 05 0 Q O ·- O U] O ~3 0 0 · - C) Q U) o o ·rl C) cn tQ tQ o C) · - O ~ S~ dP {Q ~ ~ O (Q a ~n ~ ~ ~ '- - a) ~ . s~ s~ O >-^ U ~ ~ ~ ~ ~ X [Q ~ ·' ~ ~ [Q O ~ ~ r, ~ ~ · - V ~ 0-^ · - O t) . ~ ~ a) ~ ~ 0 O ~ ~ ~ - a, 1 ~1 ~a U] U] <51 ~ 3 ~ a) _t 0 ~t)~ ~ ~ a, ~ 1 -1 ~:^ _I ~ ~ ~ ~J ·- O O ~ U] O ~ ~Q GS V O · ~(1) U O ~ =a,~-I ~ O - ~Q ~Q ~ ~ ~ ~ ~ ~ ~ ~ ~ O .- a) ~ ~ 0 ~ 0 m _ t~ ~ C) Q ~ <1) ~ ~ ~ ~ ~ ~ ~ O ~ S ~ ~ ~ ~ O V ~ ~ ~ ' - ~ ~' - ~ ~ O ~ · - ~ I_ . - U] ~ ~ ~ ~ ~ ~ ~ O ~ ~ ~ Q ~ O U ~ ~ ~ ~ G5 ~ . - ~ ~ Q. `:: tQ <1) _~ O -~ t~S O ~ ~ ~ ~ ~ O O 41) ~ tB ~ Q ~ O ~ ~ ~ O ~ .^ -I ~ ~ ~ ' - O ~ ~ ~ ~ _ _ t) ~ 3 o ~' ~ CQ ~ ~ ~ ~ ~ ~ u, O O ~ ~ ~ ~ ~ ~ ~ V Q' OZ ~ 3 - O ~ ~ ~ ~ = - S ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ V ~ O ~ ~ ~ ~ - to ~ ~ ~ ~ ~.^ ~ ~ ~ ~ ~ · - ~ ~ ~ ~ ~ ~ U] ·~- - ~ Q _' Q U] ~ GS ~- - ~ ~ · - ~ ~ O GS - ~ S d' S ' - ~ ·,1 U] '- U] S . - ' (L~ O {Q {Q ~ cn a) 0 0 t ~Q Q . ~o'~ S N O a) ~ a) ~ [Q O S ~Q o ~q O . ~a dJ o O ·- CQ Q, Q Q. 1 ~ O O ~ O . - S 3 ~q n ~ 0e o . ~n o^ Q. u] ~q o s o s - CQ Q S O O S U ~ ~ O (1) ~ O ~ a~ S ''t i ~JJ ~J tQ ~ ~ ~Q O 0 ~ >1 ~ O Q O m O ~ ~ O a~ ~S ~ =~ - 0 0 0 ~1 ~ ~ ~C ~) ~ ~ ~ e~ - 1 1 _' ~1 ~Q1 1 · 0 a~ Q~ P' '- ~.Ll

374 information. Some of these studies report offense mix at several different observation times. These multiple glimpses provide some insight into aggregate trends in of- fense mix (for example, increases or de- creases in the representation of particular offense types over time), but they do not reflect the dynamics of change for individ- ual offenders. Because they rely on aggre- gate data, they cannot separate changes in offense types over time for individuals from changes in the mix of offenders. An increase in the representation of serious offenses, for example, might result from movement toward increasingly more serious offenses by active offenders or from the attrition of offenders who engaged predominantly in less serious offense types. In the latter in- stance only the more serious offenders would remain in the sample and there would be little or no change in offense mix for those who are still criminally active. Data on offense mix alone without data on individual sequences cannot distinguish these different processes. . , . Early U.S. Studies Not surprisingly, the two oldest studies reviewed are also the crudest methodologi- cally (Frum, 1958; Peterson, Pitl~an, and O'Neal, 1962~. Relying on small samples of offenders, the studies focused on uniquely characterizing the criminal history of each sample member. Concerned with escala- tion, Frum (1958) formed criminal profiles of each sample member that documented the exact time sequence of offense types in that individual's criminal history. These were then sorted to form subsets that exhib- ited the same general pattem. Peterson, Pittman, and O'Neal (1962) were interested in specialization. On the basis of the mix of offense types they found in individuals' complete criminal histories, they character- ized the individuals as having "assaultive," "nonassaultive," "neither," or"both" types of arrests in their criminal histories. The relative frequency of the pure types, "as- saultive" and "nonassaultive," among sam- ple members was used as an indication of the extent of specialization. CRIMINAL CAREERS AND CAREER CRIMINALS On the basis of his analysis of criminal history profiles tracking offense types from the first juvenile offense to the most recent offense leading to incarceration Frum reported a general pattern of escala- tion in seriousness and concluded that the study "offers partial confirmation of the popular view that children who commit minor delinquencies and who persist in crime progress into areas of more and more serious crime" (1958:49~. Notably, there were also a substantial number of offenders (53/148 = 36 percent) whose histories showed persistence in serious property fel- onies from the juvenile to the adult period. Frum's conclusions about offense-switch- ing pattems, however, are likely to be seri- ously biased by his use of a sample of prison inmates. Offenders sentenced to prison are more likely to have been convicted of more serious offenses and to have a larger num- ber of convictions in their records. Thus there will be a bias toward more serious offense types at the end of the inmates criminal histories. This selection bias could account for much of the escalation to, and specialization in, serious offense types ob- served by Frum. Offenders who do persist, but in less serious offense types, are less likely to be in prison and would not have entered his sample. According to Peterson, Pit~nan, ant! O'Neal (1962), "stable deviance," or spe- cialization, exists when a prior arrest history (i.e., excluding the arrest that led to sam- pling the offencler) contains arrests for ei- ~er assaultive offenses or nonassaultive of- fenses, but not both.20 Uncler this definition, a single nonassaultive arrest in a long his- tor,v of assaultive arrests, or vice versa, is 20In Peterson, Pittman, and O'Neal (1962:44), assaultive offenses are crimes of violence against persons, including criminal homicide, suspicion of homicide, aggravated and simple assault, and con- cealed weapons offenses. Nonassaultive offenses involve theft of property, including burglary, lar- ceny, auto theft, forgery, embezzlement, and trading in stolen property. Rape and robbery were explicitly excluded because of their ambiguity with respect to violence- rape includes statutory rape and robbery involves both the threat or use of force against persons and theft of property.

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS enough to violate the criteria of specializa- tion. Using this apparently restrictive defi- nition, they report surprisingly high rates of stable deviance: 91 percent stable in the total sample of 88 men and 86 percent stable when offenders with "neither" of- fense type are excluded. Both figures, how- ever, overstate stability, in one case by count- ing offenders with no prior assaultive or nonassaultive arrests among stable deviants, and in the other by excluding those offenders entirely. These residual offenders, whose his- tories may contain a wide range of offense types other than assaultive or nonassaultive, are more appropriately treated as unstable. If one requires at least one prior assaultive or nonassaultive arrest and applies the au- thors' criteria for offense specialization, 58 percent (51/88) of the offenders displayed stable deviance 20 percent with only property offenses and 38 percent with only offenses against persons in their arrest his- tories. Another 33 percent (29/88) of offenders had no prior assaultive or nonassaultive ar- rests, and 9 percent (8/88) had "mixed" histo- ries, including both assaultive and nonas- saultive offenses throughout their histories. Even after adjustment for histories with no prior assaultive or nonassaultive arrests, the resulting specialization rate may still be inflated by offenders who have only one prior arrest in their histories. A single arrest of some type is not sufficient to indicate a tendency to repeated offending in the same type. Unfortunately, the distribution over prior arrests was not reported. While the median number of arrests per person in the full sample was reported to be 11, which suggests that multiple arrests per person were quite common, there also appear to be systematic differences in the number of prior arrests across different offender sub- groups. The "mixed" offenders were re- ported to have very high numbers of prior arrests a median of 52.3 arrests per person. It is thus possible that the specialists have fewer prior arrests and include a dispropor- tionate share of offenders with only one prior arrest. Perhaps the best indicator ofthe extent of specialization is the number of offenders who exhibited stable deviance in both the 375 early adult period, before age 30, and the later adult period, at or after age 30. Accord- ing to the numbers reported in Table 1 of the study, 28 percent (25/88) of the offend- ers were specialists, i.e., had prior arrests of only one type in both the early and late adult periods 18 percent in assaultive crimes and 10 percent in the nonassaultive, "theft" crimes. Thus specialization, which here required long-term activity restricted within two major offense classes, was rea- sonably~common. It must be noted that these results are based on a sample of older arrestees men aged 40 or over who were arrested for crim- inal homicide or suspicion of homicide, aggravated assault, burglary, or larceny in St. Louis during 1958. Half of the sample was 48 or older in 1958. The minimum age of 40 was used "so that individuals would have had an opportunity to stabilize their criminal patterns" (p. 45~. If such stabiliza- tion does in fact increase as offenders get older, then the level of specialization ob- served in this study is likely to be higher than would be found in a more representa- tive sample of active, and younger, offend- ers. Certain age dependencies in specializa- tion are suggested by comparing arrest activity in early (before age 30) and late adult periods. The later an offender's first arrest for assaultive or nonassaultive of- fenses, the more likely that offender will be a specialist in crimes against persons; there were 72 percent assaultive specialists among those first arrested at 30 or older, but only 40 percent among those first arrested before age 30. This suggests some tendency for older offenders to engage in violent offenses rather than property offenses. In contrast to the small samples surveyed in earlier studies, Shannon (1981) used lon- gitudinal data on 4,079 youths (male and female) in three birth cohorts (1942, 1949, and 1955) "residing continuously" in Racine, Wisconsin.2i Police contacts of this "Continuous residence" is defined as having no more than 3 years' absence from Racine, Wiscon- sin, during the follow-up periods from age 6 to age 32 for the 1942 cohort, to age 25 for the 1949 cohort, and to age 21 for the 1955 cohort.

376 sample were examined for escalation in seriousness and for any clustering of con- tacts in related offense types. Several dif- ferent approaches to the measurement of escalation were employed, including aver- age seriousness scores on successive con- tacts and by age at contact, and the propor- tion of serious contacts or of persons with serious contacts by age at contact. Shannon reported that almost without exception no consistent trends in seriousness of offenses were observed across the various measures. Average seriousness for all contacts does exhibit an interesting pattern over age at contact, however. For both the 1942 and 1949 cohorts, average seriousness was rela- tively stable through age 15. After age 15, average seriousness declined to a new, sta- ble level of seriousness. For the 1955 co- hort, average seriousness remained at a higher stable level through age 21. In these results, average seriousness was based on all contacts, including traffic offenses and police stops for investigation or suspicion. To the extent that these less serious of- fenses characteristically occurred only after age 15 and were more common in the ear- lier cohorts, including them in the analysis might explain the drop in seriousness after age 15 in the first two cohorts. In fact, Shannon (1981:95-100) reported that aver- age seriousness was remarkably flat when contacts for suspicion, investigation, and traffic were excluded. Clustering of police contacts in related offense types was assessed using a tech- nique of"geometric scaling," introduced earlier by the author (Shannon, 1968~. This technique scores an individual's offense history based on the types of offenses found; the sequence in which different types appear is ignored.22 The technique involves finding the distribution of offend 22The details of the scoring procedure are not provided in Shannon (1968) or Shannon (19811. From Me brief description of the technique that is available, it appears that individual histories are scored using a series of ones and zeroes to indicate either the presence or absence of each offense type. The most frequently appearing series are then iden- tified. CRIMINAL CAREERS AND CAREER CRIMINALS ers over observed combinations of offenses. The usefulness of the scale is assessed par- tially by the number of offense combina- tions needed to represent the great bulk of observed offense histories. The fewer the combinations, the more useful the scale is in characterizing a large number of offense histories. In both the original 1968 article reporting data from Madison, Wisconsin, and the more recent analysis of birth co- horts in Racine, Shannon (1981:48) noted that "the recorded contacts of most offend- ers are of a random nature and most combi- nations of contacts are not meaningful in that they do not involve related activities." As Shannon pointer] out, the attempts to identify meaningful clusters of related of- fense types are seriously hampered by the overwhelming predominance of single con- tacts and the high frequency of minor vio- lations for vagrancy, disorderly conduct, and traffic offenses found in the data. These contacts swamp any potential relationships among more serious offenses in the data. Excluding single contacts and minor nui- sance violations from the analysis might result in more conceptually useful clusters for the more serious recidivists among of- fenders. Analyses Using Transition Matrices The remaining studies in this review (Ta- ble 41, numbers 4 to 9) are the most com- prehensive. The studies examine an array of issues concerning changes in offense seri- ousness based on large samples of offenders and their arrests (or police contacts) for a wide range of offense types. All of the stud- ies analyze transition matrices that reflect the chance of a next arrest for type j after an arrest for type i. Both juveniles ant! adults are represented, but no one study includes continuous transitions from juvenile years into adulthood. Nevertheless, the separate analyses of juveniles and adults provide some tentative insights into similarities and differences in offense switching during these two periods. The large samples of offenders in these studies also permit exam- ination of differences in switching among demographic groups.

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS The earliest of the studies, Wolfgang, Figlio, and Sellin (1972), analyzed offense switching for offenders found in a birth cohort. The study sample included all males born in 1945 and residing in Phila- delphia from age 10 to age 18. The full cohort of 9,945 boys consisted of 7,043 whites and 2,902 nonwhites. Ofthese, 3,475 (2,017 whites and 1,458 nonwhites) were considered delinquent by virtue of at least one recorded contact with the Philadelphia police. The analysis of switching included act police contacts (other than traffic of- fenses) in Philadelphia before age 18. The offense types used in the study were characterized by the nature of the harm done and not by standard criminal code categories. The offense categories used and the overall distribution of police contacts in each category are presented in Table 42. Offenses were distinguished primarily by whether they involved injury to a person, damage to property, or theft of property. Offenses that involved more than one of those elements (e.g., robbery) were in- cluded in the "combination" category. Of- fenses including none of those elements were classed as "non-index." The least se- rious non-index offenses were by far the most common offense type (63.2 percent). The most common serious offense type was theft (16 percent of all police contacts). Offenses involving violence, either against persons or property, accounted for 13.3 per- cent of all police contacts. The three other studies focusing on juve- niles (Bursik, 1980; Rojek and Erikson, 1982; Smith and Smith, 1984) relied on cross-section samples of juveniles proc- essed, adjudicated delinquent by the juve- nile court, or incarcerated during the sam- pling period. Since the juveniles in these samples had reached juvenile court at least once, they are likely to represent more se- rious subsamples of juvenile offenders than the juveniles studied by Wolfgang, Figlio, and Sellin (1972~. Bursik (1980) used a random sample of juveniles who were adjudicated delinquent in Cook County (Chicago), Illinois, and had reached age 17 by the time of data collec- tion. The sampling years were not speci . ~. . 377 fled. The sample included 134 whites and 355 nonwhites and, although not indicated, apparently both males and females. Evi- dence of offense switching found in official records, including all police contacts and court appearances through their 17th birth- day, was analyzed.23 The sampled event was included in the analysis. Offenses were grouped in this study on the basis of criminal code categories to form four offense categories: personal injury, per- sonal property (e.g., robbery), impersonal property (e.g., theft), and all others. It is not clear whether less serious juvenile status offenses and traffic offenses were included in the analysis. The analysis was limited to the first four transitions for juveniles who had at least five contacts. This further re- stricts the results to offense seriousness for more active juvenile offenders. The more se- rious nature ofthe sample, compared with the Philadelphia cohort, is evident in the smaller representation of "other" offenses and the considerably higher representation of theft- type offenses included in the impersonal property category (see Table 42~. Rojek and Erikson (1982) analyzed data for a random sample of 1,180 juveniles (male and female) processed by the juvenile court in Pima County, Arizona, during a 3-year period. The analysis was restricted to juveniles who had at least one prior arrest when sampled and who resided in Pima County for at least 2 years. Official offense histories for the sampled juveniles were obtained from all police agencies operating in Pima County. The vast majority of the sample was described as "at risk" (of arrest) from age 8 to age 18. It appears that the analysis included all arrests in a history, 23It is not clear in Bursik (1980) whether police contacts and court appearances are treated as sepa- rate events in the analysis. When court appearances result from a police contact and thus relate to the same triggering offense, only one of the official contacts should be counted. Only when court ap- pearances are not the result of a police contact, as sometimes happens for juvenile status offenses about which a parent or school authority files a complaint directly in juvenile court, should court appearances be treated as distinct events.

378 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-42 Distribution of Offense Types among Juvenile Offenders in Several Jurisdictions (All Offenders, All Contacts) Study Offense Type Percentage of Official Contacts Philadelphia birth cohort (Wolfgang, Figlio, and Sellin, 1972)a Cook County, Ill., juvenile court sample (Bursik, 1980)b Pima County, Ariz., juvenile court sample (Rojek and Erikson, 1982)C New Jersey juvenile correctional facilities sample (Smith and Smith, 1984)d Injury Theft Damage Combination (robbery) Non-index Personal injury Personal property (robbery) Impersonal property Other Persons Property Other crimes Runaway Other status Injury Robbery Property Damage Non-index 8.4 16.0 4.9 7.4 63.2 11.6 11.5 45.6 31.3 3.7 36.0 11.0 29.0 20.3 12.6 5.1 48.8 4.1 29.3 derived from data provided in Wolfgang, Figlio, and Sellin (1972:Matrices 11.1-11.9 and Tables 11.2-11.6). Data include up to 9 police contacts for each individual, which represented 92 percent (9,384/10,214) of all police contacts (traffic offenses excluded). bDerived from data provided in Bursik (1980:Tables 2a and 2b). Data include all contacts up to the fifth for those juveniles who have at least 5 contacts, for a total of 2,345 contacts. Distribution of offense types excludes first contacts that were not reported in the original study. SDerived from data provided in Rojek and Erikson (1982:Table 2). The data include up to 5 contacts for each individual, for a total of 3,545 contacts. ~ erived from data on 9,000 arrests for 767 incarcerated juveniles, available in official court records (Smith and Smith, 1984:Table 1).

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS including the one that led to sampling the offender. ~ ~ This sample is distinguished from the other studies reviewed here by its large representation of females, 37 percent of the sampled juveniles. The analysis also fo- cused explicitly on juvenile status offenses (including runaway), 49.3 percent of all of- fenses (Table 42~. Thirty-nine percent of all contacts for males were for runaway and other status offenses; for females, the two categories accounted for 72 percent of all contacts (55 percent for runaway alone). Smith and Smith (1984) obtained data from official court records for all juveniles incarcerated in New Jersey state correc- tional facilities any time between October 1977 and December 1978. Because of their small number in the original sample, fe- males were eliminated, and the analysis focused on the previous arrests, including the one leading to the current incarceration, of 767 males aged 13-18 when sampled. These incarcerated youths averaged 11.7 arrests each, 70 percent of which were ar- rests for FBI index crimes. Also, 25 percent ofthe sample had previously been incarcer- ated. The serious nature of this sample is reflected in the mix of offenses in their records (Table 42~. Property crimes Just under 50 percent) and injury offenses (12.6 percent) represented major portions of all arrests in their records. The remaining two studies reviewed here analyze offense-switching patterns for cross-section samples of adult arrestees from Washington, D.C. (Moitra, 1981), and from the Detroit SMSA and a collection of smaller SMSAs in southern Michigan (Blumstein, Cohen, and Das, 1985~. Both studies examined the arrest histories of adults arrested for the serious index of- fenses of murder, rape, robbery, aggravated assault, burglary, and auto theft during the sampling period. For the Washington, D.C., sample of 5,338 offenders, the sampling period was a single year (1973~. For the Detroit SMSA (N = 18,635) and southern Michigan (N = 13,562) samples, a 4-year sampling period (1974-1977) was used. The sampling criterion based on offense type narrows the analysis to a sample presum 379 ably arrested for more serious offenses all nau at least one arrest tor one of the serious criterion offenses. In the analysis of changes in offense seriousness, however, virtually all previous offense types were consid- ered.24 The analysis stopped at the arrest prior to the sampled arrest, thus avoiding a bias of switching toward the more narrow subset of serious offenses found among sampled arrests. Females were not explicitly excluded from the adult samples, but they were rep- resented in very small numbers in all sam- ples: 10.3 percent in Washington, D.C., 4.5 percent in Detroit, and 6.3 percent in south- ern Michigan. The samples were domi- nated by males, and separate analyses by sex were not provided. Arrestees in Wash- ington, D.C., in 1973 were predominantly black (92 percent), as was the general pop- ulation of Mat city (71 percent black in the 1970 census). The Michigan samples were more racially diverse- 43 percent and 37 percent black in Me Detroit SMSA and southern Michigan, respectively. Separate analyses of offense seriousness by race were provided for the Michigan samples. Offense categories were defined in terms of criminal code classifications. Only ar- restees who had at least two arrests prior to Me sampled arrest were included in Me analyses. Because of the larger sample sizes, the ollense types were clisaggregated more finely than in any of the juvenile studies (Table 431. For comparison win the juvenile samples, offense types were also aggregated to include violent offenses, property offenses, robbery, drug offenses, and all other offenses. The more serious nature of these adult 24The analysis of Michigan arrestees by Blumstein, Cohen, and Das (1985) excluded the least serious offense types that make up the "public order" and "other" offense categories. This ex- cluded nuisance offenses like drunkenness, va- grancy, disorderly conduct, trespassing, and ob- scene behavior, which accounted for 35.7 percent of all arrests in the Michigan jurisdictions. These same offense types represent 26.3 percent of total arrests in the Washington, D.C., sample and 29.8 percent of arrests prior to the sampling period.

380 U] al a) o - U] At U] a, U] A: .,, U] a) Q. En U] o o o ,' Q .,, U] ·,' 1 _ m u, ~n a m ~ S4 E~ ~ a) Q E~ U] o o U] U] a o V s~ ,. Sv ,' ~: a S~ ~s o ~n ~! Q g - ~o U] a) 11 ,' r: Z ~- 11 - U] V m - U] a, .,, s :~ o U] Y V 11 Z m- ~ ~ ~0 a, e, · · ·· · - O ~ U~00 0 0 ~ ~ 0D ~ · · · ~ 1 1 1 1 ~ a' oo~ tD CO ~ ~ o C~ · · ·· · · · · · ~ , - _1, -~1 oo O un ~ t- tS) 1 1 1 1 Mo ~ u, u~ · · · · · · · · · ~ o ~kD t~ t- ~ t ~1 1 -~ ~ ~1 1 ~ ~ ~ ~ ~ ~ ~ u~ ~ · · · · · · · · · ~ _I ~1 ~a~ 0 ~ u~ CO ~ It~ 1 1 ~ ~1 1 oD ~r ~ - ~ ~- ~ ~ - - - - O O ~Ll~ C~ ~ 1 11 . · . · . . . . . c~ ~ u ~ ~ u~ ~ ~ ~ ~ Z - - ~ - ~ ~ - - - 1 1 _ - - - - 1 1 u~ o 11 ~n Q, O E~ ~ ~r CDO ~ ~ o0 U~ ·n ~ ~ ~ · ·· · · · · · · · · ~ O ~ ~ O d. ~ ~ ~ 1 - - a) ~ O U] V a.) ~ ~ u, OQ ~ ~ ~ V O Q, ~ ~ ~ "s ~ ~ ~ ~ ~S Q S O ~ ~ ~ ~ ~ ~ ~ ~ ~ ~: =~: ~ ~ m ~: ~P~ 0

381 · · . ~ d' ~ UP 0 1 ~1 ~ ~1 1 oo Go ~ ~ · · . ~ 1 - _I ~ 1 Lo _1 1 CO ~ · · . ~ 1 ~1 1 o ~ ~ UP · · . ~ ~1 ~ CO 1 ~1 ~ _1 1 ~ _ ~ ~ as up ~ ~ · · e ~ Lo can ~1 - 1 _ _ _ 1 U] a) O al U] ~ ~ ~ a) a) s: ~ ~ m1 [Q ~ a, ~ a, Q. :~ - P~Q E~ O O Q ~S ~1 ~ ~ O ~ ~ -1 ~ P. ~ ~ O o C) ~ ~ CO U~ ~ · · e · e O ~ O ~ ~ o en a) ~4 ~Q S .~' C) .~t ~ tQ O ·,1 a) ~ Q Q o o o Q., ·-' ~ s~ U] o~ U2 ~ ~n O ~ 3 ..O ~ ~ U. .,, o ·,' Q .,, u] ·~' a) N ,' S O O E~ 3 a .. tQ 33 Z a ut o . - .-, ~ eq 0 ~· - P~ s o o c) . - Q .,! ~5 C) X a '~ ~ ~ a) CQ a) ~: U] S: ~ O O ~? :' 3 ~ ' U] a a S~ ~ O a ' :^ a S~ ~ S~ ~ O a, 54 · - Q _1 Q a ~ ~ a) t) ~ · - ~ "1 Q .,4 ~ U] O U] a) u' ~n ~ ~ U] O O e ~ ~ O U] aJ a, ~ ~ U] -I P,. (V ~ a) 0 s s · - ~ ~ ;~8 a, ·,' S · - · - e _I tn a U] a, o a E~ ·e ao - ~4 · - o S-l ~ ^ U~ a, - ·rl U] · ~ · S C) ~ ' O ~ .~1 ·,1 ~ a) .~1 ut S [Q 3 ~ o En i4 C) o3 U] a) ·,4 a

382 in a, Decreasing Seriousness CRIMINAL CAREERS AND CAREER CRIMINALS Offense Type of k + 1st Arrest \ \ it' \ Moves to Less Serious Offense Types (De-escalation) \~ Moves to More Serious Offense Types (Escalation) ~\ Diagonal Elements \ {specialization) Decreasing Ser lousness ! \\ J N p;; = 1 for all i j=1 FIGURE B-1 Transition matrix for o~ense-type switches between successive arrests. arrestee samples is evident in the higher representation of violent offenses, espe- cially in Washington, D.C., for which the category of less serious "other" offenses is included. There are also important differ- ences among the adult samples. Arrests for property offenses were especially prevalent in the Michigan samples due primarily to the higher representation of burglary and larceny arrests. Comparing whites and blacks in the two Michigan samples, rob- bery and larceny were more prevalent among black offenders, and burglary was more prevalent among white offenders. For robbery and larceny, this is consistent with differences in arrest frequencies reported for the same samples in Table 19. Transition matrices are particularly well suited to analyses of changes in offense seriousness. They have the advantage of accommodating differences in the mix of offense types and the sequence in which offense types occur. The matrices are made up of individual transition probabilities, Pit, which reflect the frequency with which an offense of type i is followed on the next arrest by an offense of type j. These proba- bilities are estimated from the number of arrest sequences (or police contacts for ju- veniles) from type i to type j (nit) observed among all arrest sequences that start with type i (nit = ~j=~N nit for N offense types) with Pij = ntylni. The structure of an offense transition ma- trix is illustrated in Figure 1. Offense types ofthe kth arrest in arrest sequences form the rows of the matrix, and offense types of the next arrest in a sequence form the columns. All switches from arrests of offense type i are distributed over the elements of row i. Thus, the transition probabilities of each row ofthe matrix sum to 1. Specialization in offense types is indicated by the size of the diagonal elements, Pii, of a transition matrix (Figure 1~. The diagonal indicates the rela- tive propensity to repeat the same offense type on successive arrests. Escalation (or de-escalation), by contrast, is indicated by

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS the relative size of nondiagonal elements. The offense types in Figure 1 are ordered from the most serious to least serious. In this case, the tendency to switch to more serious offense types on successive arrests is represented by the transition crobabili- ties falling below the diagonal. Switches to less serious offense types on successive ar- rests are represented by the transition prob- abilities above the diagonal. If there is es- calation, switches to more serious offense types will be more frequent than switches to less serious offense types, and the fre- quency of switches below the diagonal will be greater than above. The reverse will hold if there is de-escalation on successive arrests. Clustering of related offense types would also be evident in the nondiagonal ele- ments of the transition matrix. Clusters of offense types are formed when there is a greater tendency for switching among of- fense types within a cluster and a lower tendency to switch to offenses outside a cluster. In the context of offense-type switching, there is particular interest in de- te~ining whether violent offenses and property offenses form distinctive clusters. As was evident in Table 42, these aggrega- tions of offense types are often used on the basis of a priori conceptual similarities in the offenses. It would be useful to know whether these clusters can also be observed empirically among offense types. Typically, a separate matrix is estimated for each transition to permit examination of changes in offense-switching patterns on successive arrests. Transition processes that vary on successive transitions are nonsta- tionary, and separate matrices for each tran- sition should be preserved because they contain valuable information about offense- switching pattems. A particularly interest- ing form of nonstationarity is trends across transitions that indicate increases or de- creases in specialization or in escalation. Nonstationarity might also arise from changes in the sample of offenders at suc- cessive transitions; while offenders with only a few arrests enter early transition matrices, the later matrices are limited to offenders who have a sufficiently large 383 number of arrests. When there are no im- portant differences across transition matri- ces, the transition process is stationary. In this event, the chance of moving from type i to type j is the same regardless of the tran- sition number. Stationary transition pro- cesses can be represented by a single tran- sition matrix estimated by combining all pairs of successive arrests. In addition to varying across transitions, offense-switching patterns may also vary for different subgroups of offenders. The varia- tion across groups can be partially tested by estimating separate transition matrices for different subgroups of the population. Pop- ulation differences will be evident in differ- ences in the estimates. Transition matrices can also be assessed to determine whether offense-type switch- ing between arrests is a Markov process. In a first-order Markov process, the offense type of the next arrest depends, at most, on the offense type of the immediately preced- ing arrest and is independent of all other prior offense types. To the extent that of- fense switching is indeed Markovian (or can be made Markovian by some modification of offense-type categories), the extensively developed theory of Markov chains can be brought to bear to explore the dyna- mic and equilibrium properties of offense- type switching between arrests, including such considerations as the expected num- ber of times the same offense will be re- peated, or the expected distribution of of- fense types on any particular arrest, or the likelihood of switching to some offense type permanently (see Cox and Miller, 1965; Kemeny, Snell, and Thompson, 1966; Karlin and Taylor, 1975~. Note that offense switching need not be Markovian for anal- yses of changes in offense seriousness. The simple transition probabilities can be esti- mated for each arrest, whether Markovian or not. Also, the analyses of stationarity, homogeneity, specialization, and escalation do not invoke, or in any way depend on, Markov properties for the transition matri- ces. A full treatment of offense switching us- ing transition matrices includes consider- ation of stationarity, specialization, escala

384 CRIMINAL CAREERS AND CAREER CRIMINALS A Offense Type of k + 1st Arrest B Offense Type of Next Arrest . · · · j · · · e ~j ~e ~ 1 1 r. . . njj(1) e ~ e a, ~I - S ye o a) Q 1 a) a) o . cat co IL o ·_ O) 6, ~e Q Q O ~9 Q O ~ In .4J a) ~0 e e e e ~ . njj(9) FIGURE B-2 Contingency table analysis of offense switching. A: Frequencies of offense switches be- tween adjacent arrests. B: Frequencies of offense switches from type i for different offender subgroups. lion, homogeneity across population sub- groups, and the Markov property. These issues are examined to varying degrees in the studies reviewed here. Where not re- ported in the original study, sufficient data were often provided to permit supplemen- tary analyses. The results of reanalysis or supplementary analysis of the data for this appendix are summarized below. Of partic- ular interest are consistencies or differences in results across different samples, espe- cially comparisons of results for juvenile and adult offenders. The general method for analyzing transi- tion matrices derives from the analysis of contingency tables introduced by Goodman (1962, 1968) and later extended by Haber- man (1973, 1979~. These techniques were applied in varying degrees to transition ma- trices for offense switching in Wolfgang, Figlio, and Sellin (1972), Bursik (1980), back into the original contingency tables of nit reflecting the number of arrests of type i followed on the next arrest by offense type j. Standard x2 tests are then applied to various forms of these contingency tables, testing the observed frequency distribution against the null hypothesis of independence between the rows and columns of the table. Under independence, column entries do not depend on their location in any particular row. Subgroup Differences. The contingency table method is illustrated here to examine differences in offense switching across sub- groups of offenders. The general structure of the contingency table tor ottense switches from one arrest to the next is pre- sented in Figure 2A. A separate table like this is available for each offender subgroup. To test for homogeneity across these sub ~,% ~ groups, the individual tables of offense Rojek and Erikson (1982), and Smith and ~ Smith (1984~. Similar analyses were possi ble using the original data from Moitra (1981) and Blumstein, Cohen, and Das (1985~. The basic approach is to transform matrices of transition probabilities, / N Pie = nip / ~ nip, / j=1 switching tor each subgroup are recom- bined to form new tables, as in part B of Figure 2. By taking the corresponding rows for switches from a specific offense type i from all the separate subgroup tables, the rows of the new table represent switches from type i for each offender subgroup. A separate table like the one in Figure 2B is formed for each offense type i. Homogeneity across subgroups is evalu- ated by using a standard x2 test on each of

APPENDIX B: RESEARCH ON CRIMINAL CAREERS the newly formed tables (Goodman, 1962, 1968~. Typically, offenders are distin- guished by race-ethnicity, sex, or age, and the test assesses whether the frequency of switches from type i to type j is indepen- dent of these offender subgroups.25 The results of tests for homogeneity in offense switching across racial/ethnic groups in var- ious studies are summarized in Table 44. Significant differences between races were found in every sample examined except the study of juveniles in Pima County, Arizona. The particular offense switches contribut- ing to significant differences across sub- groups can be assessed using a statistical test applied to individual entries in a fre- quency table. As discussed in Haberman (1973, 1979), a standardized normal deviate, called the adjusted standardized residual (ASR), can be formed from the deviation between observed and expected frequen- cies for each table entry.26 Since the ASR is 25The same general method described for homo- geneity was also used to assess stationarity in of- fense-switching patterns on successive arrests. All switches from a specific offense type i for successive transition matrices (i.e., the first to second arrest, the second to third arrest, and so on) were combined in the same frequency table, and a x2 test was per- formed to assess whether the frequency of switches from type i to type j was independent of the partic- ular transition in which the switch occurred. By use ofthis test switching was found to be stationary over , successive transitions for juveniles in Wolfgang, Figlio, and Sellin (1972), Bur- sik (1980), and Rojek and Erikson (19821. On the basis of the x2 test de- scribed here, combined with analysis of probability plots for successive transitions (Forbes, 1971), general stationarity was also observed for adults in Washings ton, D.C. (Moitra, 1981), and in two Michigan jurisdic- tions (Blumstein, Cohen, and Das, 1985~. In view of this widespread stationarity, the separate transition matrices for successive arrests are appropriately aggre- gated to form a single summary transition matrix for each offender subgroup, and these summary matrices are used in the various analyses reported here. 26The adjusted standardized residual (ASR) for switches between row i and column j in a complete frequency table is given by: ASRu = ell/Sij, where en is the standardized residual; eu = (nix-nit X n~ln..) 1~. and sit is the estimate of the asymptotic standard deviation, 385 distributed as a standard normal variable, a simple test of the significance of departures from independence is available for individ- ual observed frequencies in a table. Statis- tically significant values ofthe ASR indicate that individual offense switches are signifi- cantly more (or less) likely than would be expected if switching was independent of the row variable in a transition matrix. A consistent pattern of significant ASRs is observed for the Detroit SMSA and Phila- delchia samples. As illustrated in Tables 45 and 46, nonwhite offenders were generally more likely Man white offenders to switch to serious offenses, especially those involv- ing violence or robbery. For example, the likelihood of switching from violent of- fenses to robbery for black offenders in the Detroit SMSA was .16. This is more than twice the same likelihood found for white offenders (.07) and results in a significant ASR value of 5.6.27 Nonwhites were less sir = [~l - ni.ln..~l-non..]'. Before adjustment by sit, the standardized residual is just the contribution of an individual table entry to a x2 statistic before squaring, where nit is the observed frequency count in row i and column j; c n is the row marginal = ~ n, j=1 R nj is the column marginal = ~ i=1 R C n.. is the table total = ~ ~ n', i=l j=1 for C total columns; for R total rows; and 27With only two categories of offenders (white vs. black, or white vs. nonwhite), the ASRs for white offenders are just the negative of the ASR values reported for nonwhite or black offenders in Tables 45 and 46. So, for example, the ASR for transitions from violent offenses to robbery for black offenders in the Detroit SMSA is 5.6 (Table 46~; for white offenders in the same sample, the corresponding ASR is -5.6. In this example, black offenders are signifi- cantly more likely, and white offenders significantly less likely, to switch from a violent offense to a robbery on the next arrest than would be expected if switching from violent offenses was independent of race.

386 TABLE B-44 CRIMINAL CAREERS AND CAREER CRIMINALS Test of Homogeneity of Offense Switching Across Rac ial-Ethnic Subgroups for D if f Brent Of f ender Samples ADULT SAMPLES Detroit SMSAa Southern Michigan Countiesa (without Resistance) (without Resistance) _ Offense x2 Type White/Black Individual Offense Type sb Murder 20.8C Rape 8.6 Robbery 30~0* Aggravated assault Drugs Burglary Larceny Auto theft Weapons Fraud Total 36.7* 47.8* 58.3* 77.6* 31.9* 39.0* 19.5* 370.2* (90 d.f.) Collapsed Offense Typesd Violente54~0* Property_56.1* Robbery6.0* Drugs41.2* Total157.3* (12 d.f.) JUVENILE SAMPLES Philadelphia] (with Resistance) x2 White/Black 3.8 13.6 22.5* 15.9 33.6* 52.5* 47.0* 40.8* 21.5* 37.8* 289.0* (90 d.f.) 21.6* 92.2* 13.6* 22.2* 149.6* (12 d.f.) Cook Countyh (without Resistance)_ Pima County1 (with Resistance) Wh i te/ Wh i te/ Wh i te/Black/ Offense NonwhiteOffense Nonwhite Offense Hispanic Type (5 d.f.)Type (3 d.f.) Type (10 d.f.) Non-index 182.9*Personal Persons 17.1 Injury 31.1*injury 1.0 Property 17.5 Theft 39.2*Personal Other Damage 68.6*property 0.5 crimes 5.4 Combination 15.8*Impersonal Runaway 6.7 (robbery) property 13.4* Other Total 337.6*Other 24.4* status 14.6 (25 d.f.)Total 39.3* Total 61.3 (12 d.f.) (50 d.f.)

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS TABLE B- 4 4 Cont inued derived from data on race-specific offense switching provided by the authors of Blumstein, Cohen, and Das (1985). Thor individual offense types, x2 was calculated with 9 degrees of freedom. 2 2 CThe X value is unreliable. While the -value for X is <0.05 or better, the contingency table contains a number of low expected values (n < 5). Those will tend to inflate the x2 value and introduce errors in the p-vaJue. 2 -For collapsed offense types, X was calculated with 3 degrees of freedom. ~ efIncludes murder, rape, aggravated assault, and weapons. -Includes burglary, larceny, auto theft, and fraud. derived from data on race-specific offense switching available in Wolfgang, Figlio, and Sellin (1972:Matrices 11.11 to 11.26 and Tables 11.27 and 11.28). ~ erived from data on race-specific offense switching available in Bursik (1980:Tables 2a and 2b). iDerived from data on race/ethnic-specific offense switching available in Rojek and Erikson (1982:Table 7). *x2 significant at the .05 level or better. likely to switch to less serious drug offenses or to desist from offending. Rates of switch- ing to offenses involving theft of property . . were slm1 ar across races. In Cook County, nonwhite juveniles were more likely than white juveniles to move to the violent offenses of personal injury or personal property (e.g., robbery), and less likely to move to the least serious offenses in the catch-all "others" category. Black adult offenders in both the Detroit SMSA and southern Michigan were more likely than white adult offenders to move to robbery and larceny and less likely to move among the property offenses of burglary, larceny, auto theft, and fraud. Black offend- ers in Detroit were also less likely than white offenders to move to drugs from other offense types, and black offenders in south- ern Michigan were more likely than white offenders to move to the violent offenses of murder, rape, aggravated assault, and weap- ons. When the offense types are collapsed for the Michigan samples, the higher likeli- hood of violent offenses among black of- fenders is also observed in the Detroit sam- ple. The Pima County sample of juveniles is 387 the only study in which there was no evi- dence of differences in offense switching between racial-ethnic subgroups. This sam- ple is distinguished from the other samples by its much lower representation of blacks (6 percent of the sample), by its larger rep- resentation of females (37.5 percent of the sample), and by its explicit consideration of juvenile status offenses (59.3 percent of all official contacts in the sample). All these factors contribute to a greater representa- tion of less serious offense types in the sample, and they may obscure racial-ethnic differences among more serious offense types. The Pima County sample is the only one that includes enough females to test differ- ences in offense switching by sex. As might be expected, offense switching is signifi- cantly different for male and female of- fenders in Pima County. Female offenders were much more likely than males to desist or to move to a runaway offense. Switches to these two categories together accounted for 73 percent of the switches by females, com- pared with only 36 percent by males. It is not possible with these published data to restrict the analysis of switching patterns of

388 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-45 Differences in Offense Switching by Race of Offender for Adults in the Detroit SMSA: Transition Probabilities and Adjusted Standardized Residuals (ASRs) Offense Type on kth Arrest Offense Type on _ + 1st Arrest Violent Property Robbery Drugs Violent Black.39.37.16.08 White.34- .44.07.15 (ASR)(1.8)(-2.6)(5.6)(-4.8) +******** Property Black.12.72.10.06 White.10.73.06.10 (ASR)(1.8) (4~8)(-5.9) + ***** * Robbery Black.22.39.31.07 White.16.40.35.08 (ASR)(2.4) * Drugs Black.13.44.09.34 White.09·3705·50 (ASR)(2.6)(2.7)(3.4)(-5.9) ********** NOTES: Derived from data on race-specific offense switching provided by Blumstein, Cohen, and Das (1985). Only ASRS significant at the .10 level or better (two-tailed test using standard normal distribution) are reported in parentheses. ASRs for black offender S are reported in the table. With only two categories of offenders compared, the ASRs for white offenders are just the negative of those reported for black offenders. +Significant at the .10 level. *Significant at the .05 level. **Significant at the .01 level. ***Significant at the .001 level. males and females to the more serious, nonstatus offenses. Two studies examined differences in of- fense switching by age. Rojek and Erikson (1982) computed separate transition matri- ces for three age-of-onset categories (12-13, 1~15, 1~17 years old at first arrest). These three starting groups could have been com- parec3 directly to one another using contin- gency tables in which each row was drawn from a separate age-of-onset matrix. Rather than use this direct test for differences

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS across different stardng age groups, Rojek and Erickson compared the matrix for each starting age to the matrices for the first four transitions based on the entire sample. No significant x2 values were found in any of these comparisons, and the authors con- cluded that "none of the age-specific matri 389 ces differs from the transition probabilities [for the first four transitions]" (Rojek and Erikson, 1982:18~. This information, how- ever, is not very useful in assessing differ- ences in offense switching across the dif- ferent age-of-onset groups. Unfortunately, Me transition matrices by age of onset were TABLE B-46 Differences in Offense Switching by Race of Offender for Juveniles in Philadelphia Birth Cohort. Transition Probabilities and Adjusted Standardized Residuals (ASRs) . Offense Type on kth Offense Type on k + 1st Police Contact Contact Non-index Injury Theft Damage . Combination Desist Non-index Nonwhite.4709.12.03.06.24 White.43.04.09.01.04.39 (ASR)(3.1)(7.1)(3.9)(3.2)(3.1)(-12.0) *************** Injury Nonwhite.42.10.09.03.07.29 White.38.07.08.003.02.46 (ASR) ( 2.4)(3.2)(-4.4) ****** Theft Nonwhite.42.06.21.03.11.17 White.39.04.22.02.06.28 (ASR) (2.4) (3~8)(-5.0) * ****** Damage Nonwhite.49.11.18.04.04.13 White·3704070903 (ASR)(2.9)(3.1)(4.0)(-2.2) (-7.0) ******** *** Combination (robbery) Nonwhite.39.09.14.02.14.22 White.40.02.13.01.14.30 (ASR) ( 3 3) (-2.3) *** * NOTES: Derived from data on race-specific offense switching available in Wolfgang, Figlio, and Sellin (1972:Matrices 11.11 to 11.26 and Tables 11.27 and 11.28). Only ASRS significant at the .10 level or better (two-tailed test using standard normal distribution) are reported in parentheses. ASRS for nonwhite offenders are reported in the table. With only two categories of offenders compared, the ASRS for white offenders are just the negative of those reported for nonwhite offenders. *Significant at the .05 level. **Significant at the .01 level. ***Significant at the .001 level.

390 not published and so reanalysis of the data was not possible. Wolfgang, Figlio, and Sellin (1972) exam- ined offense switching over 16 periods (6 months each) from ages 10 to 18. Because of the fine disaggregation by age, the of- fense types were aggregated to form three categories: non-index, index, and no of- fense. In each age matrix, an offender was characterized by his state at the end of the age period. An offender with no police con- tacts during an age period was character- ized by the "no offense" category for that period. Likewise, if the last contact during an age period was an index offense, the offender was characterized as having an index offense in that period. Separate tran- sition matrices were estimated for all adja- cent age periods. Variations in offense switching over age were assessed by pair-wise comparisons of adjacent transition matrices. In general, the X2 tests of equality between pairs of adja- cent matrices were not significant, which indicates stationarity in switching across age. This pattern was broken at three points. First, the transitions between matri- ces for ages 1~13.5 and ages 13.~14.0 were significantly different; there was a reduction in the probability of remaining a nonoffender. There was another major re- duction in the "no offense" category at ages 14.~15 versus 1~15.5. Third, the new pat- tem continued to ages 16.~17 versus 17-17.5 and 17-17.5 versus 17.~18, when the likelihood of moving to the "no offense" category increased again. The analysis by age accommodated ab- sence of a police contact within any age period through the "no offense" category. This "no offense" category dominated the transitions in all age periods examined. Also, because of the small number of Police contacts in any 6-month period, offense types were highly aggregated as index or non-index offenses. Thus, the dominance of the "no offense" category and the limited variability in offense types examined tend to obscure any patterns of offense switching over age that might be occurring. An alternative strategy for examining age differences would have been to limit the CRIMINAL CAREERS AND CAREER CRIMINALS analysis to recidivists who had at least two police contacts and to characterize each offense transition at successive police con- tacts by age ofthe offender at that transition. A separate matrix could then be estimated for each age at transition. This would focus the analysis of age differences on actual offense switching at successive police con- tacts. Other analyses of subgroup differ- ences might examine how patterns of of- fense switching vary across offenders with different career lengths, or with variations in the length of time between successive arrests. Summarizing, the same general differ- ence in offense switching was observed across races in several jurisdictions and for both juveniles and adults. Nonwhite, or black, offenders were more likely than white offenders to move to violent offense types and less likely either to desist entirely or to move to less serious offense types. Only one study (Rojek and Erikson, 1982) included a sufficient number of females for a comparison of offense switching by sex. As might be expected, after each official con- tact, female offenders were much more likely than male offenders to desist from offending or to move to the juvenile status offense of runaway. The limited analyses of differences in offense switching by age of the offender do not support any strong con- clusions about age effects. Specialization. Specialization is the tendency to repeat the same offense type on successive arrests. Specialization is indi- cated by the diagonal elements of a transi- tion matrix, Pii~k), representing the proba- bility of a next arrest of the same type. Several studies have noted some degree of specialization in offense types, as indicated by a tendency of these diagonal elements to be elevated in magnitude compared with the levels expected if switching was inde- pendent of prior offense type. Wolfgang, Figlio, and Sellin (1972), for example, compared the magnitude of the diagonal element, Pii, with the other entries in the same column (`Pmi, m 76 i). Noting that the probability of repeating the same of- fense was moderately higher than the

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS . . ~ chances ot moving to that offense from any other offense type (see Table 46), they con- cluded "that there is some tendency to re- peat the same offense type [especially for theft offenses]" (p. 206~. Specialization was also more evident among white than nonwhite offenders. Similar results were reported by Smith and Smith (1984) for incarcerated juveniles. The analyses of adult offenders assessed the extent of specialization by comparing the diagonal entries of the transition matrix with the overall prevalence of an offense type (Moitra, 1981; Blumstein, Cohen, and Das, 1985~. One approach would involve taking the ratio of the diagonal transition probability, pesky, to the corresponding col- umn marginal probability, p.j~k), with zj~k) = p2~(k)1p.j~k).28 A z-value of unity indicates that repeating the same offense type is just as likely as switching to that offense from any other offense type. Specialization would be reflected in z-values greater than 1, where repeating the same offense type is more likely than the level of switching to that offense generally. The z-values found for Washington, D.C., and Michigan adult offenders are presented in Table 47. The z-values exceed 1, some- times substantially, in virtually all cases. The z-values are lowest (less than 2) for "other" offenses in the Washington, D.C., sample and for burglary and larceny in all the Michigan samples. The z-values are highest for homicide and rape, followed by fraud, drugs, and auto theft. Burglary and larceny in Michigan and "other" offenses in Washington, D.C., have the greatest diago- nal transition probabilities of all offense types; their lower z-values suggest that these high diagonal values are largely a result of a higher probability of switching to these offense types generally, and not a reflection of greater specialization. The studies of juveniles in Philadelphia 28The column marginal probability is given by, R R C p~(k)= ~ nu~k)/2 ~ nick) i=! `=1 j=1 for R rows and C columns in a transition matrix. 397 and New Jersey, and of adults in Washing- ton, D.C., and Michigan all found some elevation in the magnitude of diagonal tran- sition probabilities. These assessments of specialization, however, were not very pre- cise. While the diagonal elements may be greater than other switching probabilities, there is no basis for judging when a differ- ence is substantively important as an indi- cator of specialization. Is a ratio of 2:1 suffi- cient to indicate specialization, or must the ratio be considerably larger? The ratios alone are not sufficient for assessing specialization. The main problem is that the range of possible ratio values varies considerably with the magnitude of the column marginal probability. For a prevalent offense, like larceny, which had a column marginal of.148 in Washington, D.C., the maximum ratio can be no more than 6.75.29 By contrast, the ratio can get very large for relatively rare offenses. For homicide, with a column marginal probabil- ity of.015, the maximum value ofthe ratio is 67 (1.0/.015~. This dependence of preva- lence means that ratios cannot be compared across offense types that vary substantially in prevalence. A ratio of 5 may be quite large for an offense type with a maximum ratio of only 6.75, but the same ratio value of 5 reflects less specialization when the max- imum value is 67. The variability in ratios observed in Ta- ble 47 corresponds to the variability in the column marginal probabilities. In particu- lar, the ratio is smallest for the most preva- lent offenses of burglary and larceny (with column marginals of.200 to .275 in Michi- gan), and largest for the rarest offenses of homicide and rape (with column marginals of.015 or less in all samples). What is needed is some common basis of compari- son for assessing the significance of ele- vated values for diagonal probabilities. One such statistic is the AS R. Separate ASRs are computed for every entry in a transition matrix, including the diagonals (see note 26~. The ASH compares 29In a case of complete specialization, the diago- nal transition probability is 1.0 and the maximum ratio is 1.0/.148 = 6.75.

392 o U] al .,, ,1 .,' Q Q o o ·,' ·,' ~Q Ed o .,' a o o ·-, P: ·. U] SO a) o - 0 ~ :^ Ed N G. . - U1 _ ~ ~ a) C) ~ a) 0 U] ~ o ~- al d. C.) 1 ~ m ~ W ~ m ~ a: Ed Al At ~ s ~ c) 0 ~ CQ .1 11 Y - . - o .,, ~r; U] y m ~n .,' s~ Q ~n ' - o ~1 ~ U] ~: a u' U] Yo m U] .,, sg ~1 U] 0 ~ .,' s ~ eq C) :~ a C) . a 0 a) E~ O a U] C: aJ ~ u' n p~ 0 3 QJ C) . a ~ 0 ~= ~ O ~Q ~ ,~ O U] '~ .,' h ~ '~ =~ 3 QJ tQ O E~ In OIr) N N C)C) d' ~4 00 O ~ O ~ U~ ~ O ~ .. ~. . · . . . . O NC~ ~r ·- _I ·~ t~ ~I I .-1 1 1 O a ~) O ~ O ~) O w) N Co o a ~) O _1 ~ ~ tr) ~ ~ ~ N . ~. . · . . ~ . ~ ~ u ~ ~ _. _I ~ ~) ~I I -I Q N ~1 1 1 td ~ a o ~ ~ U~ CO \0 U~ _' ~CO ~ ~ (D . - r- N O ... ....... _1 ~ N O1 ~) ~1 ~ ~ N u-) II ~1 ~1 ~11 U] -1 a,J In M0 ~ ~ u~ ~ ~ ~ ~ -~ Q ~CO a~ ~ ~ In ~ ~ ~ · · a · · · · · · ~·- rO -I kD ~ ~ N ~ ~1 ~ ~ ~ I I Q ~_ ~G l l ~ Q U] O U] ~ ~ a O Q t~ GS ~ Q a) O ~ CO ~ ~ 00 ~ ~ ~ ~ ~ ~ ~ 0 ... ....... .. C ~ ~ ~ ~ (D ~ ~0 ~ ~ s 0 ~ ~ .,, ~ ~ 0 eq L4 ~ a) ~ m o C Q~ a1 0 ~ 54 ~ un ~ ~Co a' ~ oo ~ ~ ~ ~ ao ~ Q ~ ~ 0 ~ ~ 0 ~ ~ ~r ~ ~ ~ k4 O o ~ O ~ ~ O O O O ~ O ~ ... .... ~ · . E~ ~ . - (1) Q · ~ ~5 a' · a) :5 Q _I ~ ~ t~S tTS E~ C) ~ ·. n ~ e~ ~ ~ ~ ~ kg CO ~ O - ~ ~0 ~ co ~ t- t- 0 a~ u) a~ 3 ~ .. . . . . . . . . . - u, Q, · ·~ ~ a a .,' Q ~ ~ ~ `= ~O a) o Gs U~ u~ ~l aJ 0 ~5 ·- a) Q ~ U] ~ ~ V O ~ ~ ~ ~ 0) a) U] 54 04 Q ~ ~ ~ ~ ~ ~ ~ ~ Q O ~ . - 05 0 ~: ~ ~ ~: ~ m ~ ~ 3 ~ P~

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS an observed frequency count of switches between offense types with the frequency that would be expected if offense switching was independent of previous offense type. This controls for differences in prevalence. Each residual difference between observed and expected frequencies is then adjusted to form a standardized deviate whose distri- bution is approximately normal. Relying on this common statistical distribution, ASRs can be compared with one another, and their individual statistical significance can be evaluated. Specialization, reflected in a higher prob- ability of repeating the same offense type, would violate the independence assump- tion. This special form of dependence on previous offense type is reflected in signif- icant, positive ASRs for the diagonal ele- ments of a transition matrix. Diagonal ASRs were used to assess offense specialization in Bursik (1980) and Rojek and Erikson (1982~. The results of this test of specialization in these and other offender samples are re- ported in Table 48. The strongest pattern to emerge in Table 48 is the difference in specialization be- tween adults and juveniles. While some specialization is evident for juvenile offend- ers in the upper half ofthe table, specializa- tion is found in all offense types for adult offenders. The much higher magnitude of the ASR values for adults also indicates that specialization by adult offenders is stronger than that found among juvenile offenders. This seems to suggest a somewhat more exploratory approach to crime by juvenile offenders and a stronger commitment to particular offense types by those who per- sist in crime as adults. This difference between juvenile and adult offenders may reflect a developmental process in which more specialized offend- ing patterns emerge gradually as offending continues over time. It might also reflect a selection process wherein juvenile samples include a mixture of casual offenders who desist from offending very quickly and a core of committed offenders who are more specialized in their offending. As the casual exploratory offenders leave offending in the juvenile years, adult samples would consist 393 more heavily of the committed, specialized offenders. Sorting these rival hypotheses requires samples with data on both the juvenile and adult periods for the same individuals. The changes-or stability- in specialization over time could then be examined for the subset of offenders who begin as juveniles and persist into adulthood. Unfortunately, none of the available samples permits such an analysis. While specialization is pervasive among adults, it is not uniformly strong in all of- fense types. The most specialized offense types in all adult samples were drugs and fraud. Auto theft was also highly specialized among black offenders in the Detroit SMSA and among Washington, D.C., offenders. Higher specialization in these offense types is consistent with the frequent role of these offenses as part of larger, organized illegal economic enterprises. The least specialized offenses among adult offenders were the violent, and often impulsive, offenses of murder and weapons violations in all sam- ples (except blacks in southern Michigan) and rape in the Washington, D.C., sample. These results are different from those sug- gested by the high diagonal z-ratios in Ta- ble 47, where murder and rape had the highest ratios, but that difference reflects the much larger range of variability that is possible in the diagonal ratios for these very rare offenses. The ratio values are higher for these offenses, but not especially high rela- tive to the maximum possible values of these ratios. Specialization is more sporadic among juvenile offenders. The Pima County sam- ple exhibited the least specialization; only property and runaway offenders showed a significantly greater tendency to repeat those offenses on successive arrests than would be found if switching was indepen- dent of prior offense types. This result was found for all three racial-ethnic groups. This lower level of specialization cannot be at- tributed to the larger representation of fe- males in the Pima County sample. Special- ization is in fact less frequent for female offenders; it occurs only for runaway of- fenses. But even among male offenders,

394 ~q . - ., U] ::) U] a) ~ 0 CQ · - a) ,' ~ . - (1) Q 10 ~ Q O O ~, O a) . - ~ .,1 ·~1 U] · - E~ O · - ~ O N ~ · - .,' ~ a ~ 1`,1 ~ ~, V 0 a U] ~Q P: U) ~¢ - a U] a o U] 0 ~ . - U] U] a E~ a CD N 1 m ~ ~ ~5 m ~ ~: E~ u E Ll ~n ~C) ~Lt O U) u. 4, O s ~ :' r ~: q4 a s ~ a a, 4 a, ~q ~ ~ ~0 Ll ~ ~ ~ ~ C: a' a, ~ U] U) ~ a, ~ ~ n ~ ~n ~ ~ ~, 0 ~, ~ ,4 4, P4= ~ ~ ~ ~ ~ ~ ~ ~ ~ s 4 ~: ~ o:: <: ~ m~ ~ :~ ~P~ 0 ~n 4) i4 ~ ~a 4 . - .,, r1 o JJ O4 Q ~ 4, O .,' ~n . - _ 0 C ~o ~ a, \., ~ o a, ns c o G, 81 q~ ~1 54 O q~ a) ~n x ~: ~q a) ~ a, `4 ~ ~ s4 a' ~ ~q s~ ~ o a s q~ JJ O ~ O ,~ .,1 a) JJ 4~ n5 O C a' s u, h .,4 3 ~4 ~ a, 4J s eq a O eq ~ eq G1 4, ~ ,-1 ''t C n5 3 a ·. C cn O C ~n 4}~ s 3 C. - c 0 a ~ 4J u, 0~ ~ ~ ~ 0u' r' O ~ .q4 ~a a'c sn5 C C . n5 U ~O O ~ eq ns~ JJ ns C C~ ~ 'O O ns~, ~, ns c)~ .,~ ~ ~ C ~ O u' _ .-,C q4 :~: s Cn5 u, oo ~ ·~ ns ns .,~ ~n~ ~ ~ 3: n5 0~ 0 ~ O ·- 4J C _ 4, - ~ C: ns 4, Ll ns a, - 0 s ~ q~ tn n5 ·- - E~ n5 .. ~ o n5 oo q' ~a' ~ ·- _ ns _ ~ ~ r~ ~n ns y ns =_ ~ P u, q~ 0 :' n5 c m E~ n u, .. ~ E' u, 0 ~ o, q~ ns nsc n5 0 0 O u, Y ~ a.' q4 C a, ~ ~ ~ C .,. ~ n5 ~r, 4) Y 4, a\ c O ~ O 3 c' C Ul Y ~ O · - c,) s C :> O . - ns Q. ns ~ ~ ns n5 s ~ ~ p. n5 ·- ~ _ co q4 n5 ~ O ., s ~ ~ P1 ~n O O n5 C ~ n5 n5 . ns ~ 0 s~ 0 4, .. c,) ~ u, ~ 4, D3 ~ :. P: n5 ~ ~ ~ 4, O .. cn ~ ~ aJ a' _ 3 · ~ oe 0 a' a JJ a ~a .,' n5 n5 1 1 ~a) C C O O n5 n5 C C ~rl . - 0 0 ~:' aJ ~u~ 0 0 · · - 4J 4) C JJ ns ns o5 t) C q~ C ns ~ ns C) C O .,' ~ ~ ~ ~ ~4 ~ 0 ~ C. C E~ n5 .- O " ° cm n ,' 0 :3 u, - 0 n5 - 1 1 c~ C O SO n5 C) ~ . C o' ~ a, 4d 0 e n5 ~ E~ m a' c a, ·rl - ns ns ~ ~ .~ q~ O O :E: 0 C .,~ ·- 0 ns ~ C n5 ns - 1 .,. e n5 O ~ ~' n5 q~ n5 ~ n5 .,. q~ 4, 0 0 a, 'l ~, 0 4, 3 ~ ns 0 C 4~ n5 ~ e 0 n5 C ns .,' S O .,'

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS Erikson, 1982:22~. In the Philadelphia samples, white of- fenders were more specialized than . .. ~. '. . . 39S there is evidence of specialization only for fenders first arrested for robbery and injury property and runaway offenses (Rojok and offenses. The tendency to repeat these seri- ous offenses, however, may have been over- estimated because the arrest leading to the sampled incarceration was not excluded from the analysis. Compared with arrests generally (Smith and Smith, 1984:Table 1), nonwhite ottenders. Specialization was sig- nificant in all offense types among whites, but it was only significant for non-index, theft, and combination offenses for non- whites. Interestingly, despite the greater tendency of nonwhites to switch to violent offense types from other offense types (see discussion of subgroup differences), non- whites did not exhibit specialization in ei- ther injury or damage offenses. Thus, ar- rests for violent offenses were more perva- sive generally for the nonwhite offenders. However, they were no more likely to fol- low previous arrests for violent offenses than to follow arrests for any other offense types. The Cook County sample included seri- ous juvenile offenders. All had been adjudi- cated delinquent in juvenile court, and all had at least five arrests as juveniles. They were also more like the adult offenders and exhibited the greatest offense specialization by juveniles. Significant specialization was found in all offenses for both races, except personal injury for white offenders. The pervasiveness of specialization in this sam- ple of persistent juvenile offenders is illus- trated in Table 48; it suggests that the juve- nile-adult differences in specialization found in the other juvenile samples may be due to sample selection, particularly the presence of large numbers of less special- ized offenders among juveniles who desist early from offending and thus are less prev- alent in adult samples. Smith and Smith (1984) used a different approach to analyze specialization by juve- nile offenders; they examined other crime- mix attributes in addition to diagonal tran- sition probabilities between adjacent arrests. Conditioning on the offense type of the first arrest, they found that the mean number of arrests and the probability of any arrests were both highest for repeating the offense type of the first arrest. Within a sample of incarcerated juveniles, such spe- cialization was especially evident for of robbery and injury arrests were over- represented among offense types leading to incarceration for offenders first arrested for these offense types (Smith and Smith, 1984:Table 4~. Summarizing, there are differences in the levels of specialization by juvenile and adult offenders. Specialization is evident and strong in all offense types among adult offenders, but it is more sporadic and some- what weaker among juvenile offenders. Among adults, specialization is strongest for drugs, fraud, and auto theft all offenses that play a role in organized illicit markets. It is weakest, although still significant, for the more impulsive, violent crimes of mur- der, rape, and weapons violations. Offense Clusters. Closely related to spe- cialization, in which the same offense type is repeated on successive arrests, is the concept of specialization within "offense clusters." Clusters represent natural parti- tions of offense types such that offenders display an increased tendency to switch among offenses within a cluster and a cor- responding decreased tendency to switch to offenses outside a cluster. For adult offend- ers and incarcerated juveniles, violent and property offenses form distinct clusters; for other juvenile offenders, there is a sharp partition between common law crimes and juvenile status offenses. Clusters can be identified by examining patterns among ASRs for switching be- tween individual offense types. Signifi- cantly positive ASRs indicate greater switching between offense types than would be expected if switching was incle- pendent of prior offense type; significantly negative ASRs indicate less switching be- tween offense types than expected. To avoid swamping interoffense tendencies when there is strong intraoffense specializa- tion, a model of quasi independence is

396 usecI, which excludes diagonal elements (Haberrnan, 1973, 1979~. The analysis thus looks for systematic patterns of offense switching only among transitions that are not specialized. In the analysis of offense switching among juveniles, individual offense types were aggregated into larger offense catego- ries based on a priori conceptual similarities among the offenses. A richer array of offense types is available in the analyses of adult samples from Washington, D.C., and Mich- igan. These data can be examined to iden- tify offense clusters empirically based on observed offender switching patterns. The ASR values obtained in the Detroit SMSA (Table 49) exemplify the results found in other adult samples when special- ized switching found in the diagonals is excluded from the analysis of offense switching. Only ASRs reaching a minimum significance level of.10 or better (two-tailed test) are reported in the table. In addition to ASRs for switches between pairs of offense types, the x2 value for all switches com- bined is reported in the table. This x2 value for the "incomplete" table of frequency counts after the diagonals are excluded is highly significant, ant] the hypothesis that switching among different offense types is independent of prior offense type is easily rejected in all adult samples examined. Two distinct clusters of violent and prop- erty offenses are evident for all adult of- fender samples examined. Regardless of ju- risdiction or race, adult offenders exhibited definite tendencies toward increased switching within these offense clusters (positive ASRs) and decreased switching between the two clusters (negative ASRs). This tendency for offense types to cluster varied somewhat by race- the partition be- tween violent and property offenses was stronger among black offenders. In all Michigan jurisdictions, but not in Washing- ton, D.C.,30 there was also a limited ten 30Because some nonstationarity in offense switching was observed across successive arrests in Washington, D.C., the analysis of offense clustering was performed separately for each of the first three CRIMINAL CAREERS AND CAREER CRIMINALS dency for increased switching between the clusters of violent offenses and robbery, and decreased switching between the clusters of property offenses and drugs. The clear tendency toward distinct clus- ters for violent offenses and property of- fenses provides empirical support for the use of these aggregate categories in other studies of offense switching. Not only are these offense types conceptually similar, they are also behaviorally related, that is, offenders are more likely to switch among offenses within a cluster on successive ar- rests. The same approach just used to identify offense clusters among reasonably disag- gregated offense types for adults can also be used to examine the patterns of switching among the more aggregate offense catego- ries used in studies of juvenile offenders. After excluding specialization in the same offense category, one would like to know whether switching among the remaining offense categories is systematic or whether it is independent of prior offense. On the basis of a visual inspection of similarities in the rows of transition matrices, Wolfgang, Figlio, and Sellin (1972:188-189) con- cludec] that other than a limited tendency to repeat the same offense type, offense switching appeared to be generally incle- pendent of prior offense type. Examination of ASRs in a model of quasi independence (i.e., omitting diagonal switches) permits a more rigorous test of independence of prior offense type. Bursik (1980) is the only study reviewed here to employ such a test. Suffi- cient data are available in the published studies, however, to perform this analysis for all the juvenile samples. As was found in the adult samples, the x2 for switches among different offense types (diagonals excluded) are highly significant for juvenile samples (except nonwhites in Cook County), and the hypothesis that switching is independent of prior offense transition matrices. While there were some varia- tions in the results for individual offense types across these transitions, the same general clustering pattern was observed for separate transitions.

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS type can be rejected. The differences across rows of a transition matrix are sufficient that knowledge of the prior offense type can be useful in predicting the next offense type. The particular nature of dependencies on prior offense type is indicated by significant ASRs, as illustrated in Table 50 for juvenile offenders in Pima County, Arizona. A strong partition between violent and property of- fenses-reflected in significant negative ARES tor switches Between tnese offense categories was found among adults and incarcerated juveniles, but the partition was generally not as sharp among juveniles. Aside from incarcerated juveniles in New jersey, only nonwhite offenders in Phila- delphia and white offenders in Cook County (data not presented) displayed a tendency not to switch between the injury and theft offense categories. In addition, there is evidence of a tendency by some juveniles to switch between violent and property offenses. Offenders in Pima County, for example, exhibited higher-than- expectec] switching between offenses against persons and property, as indicated in Table 50, by significant positive ASRs for switches between these offense catego- ries. The Pima County study provides the only data that separately identify juvenile status offenses (runaway and "other status". The results reported in Table 50 indicate a sharp partition between status offenses and tradi- tional crimes.3i First, a tendency not to switch from status offenses to the traditional crime categories is evident. Also evident is a tendency not to switch from the traditional crime categories to Resistance; Resistance is more likely after a juvenile status offense. Interestingly, running away is distinct from other status offenses; the negative ASRs indicate a tendency not to switch between the two juvenile status categories. To summarize, offense switching does not appear to be independent of the prior offense type. Instead, knowledge of the Manly the results for all offenders in Pima County are reported here. Similar results were also found in each racial-ethnic group. 397 prior offense type provides information use- ful in predicting the next offense type for both repeated and nonrepeated offenses. The analysis of offense switching by adults and incarcerated juveniles provides behav- ioral support for the frequently used parti- tion between violent and property offenses. Adults and incarcerated juveniles exhibited definite tendencies to switch among of- fenses within clusters of violent or property offenses and a tendency not to switch be- tween-those two clusters. This partition between violence and theft, however, is not as sharp among juvenile offenders gener- ally. Moreover, based on one study of juve- niles, there appears to be a sharp partition between traditional crime categories and juvenile status offenses: switching from a status offense to a traditional crime on the next arrest is less likely than would be expected if switching was independent of prior offense type. This partition may reflect a difference in the offending patterns of males ant] females represented in the data, i.e., females may have been more heavily represented in status offenses and males in traditional crimes. Escalation. Escalation, a tendency for offenders to move to more serious offenses as offending continues, is probably the most widely held, commonsense view of crimi- nal offending. Data on offense transitions provide an opportunity to investigate of- fense escalation empirically. Various ap- proaches are used to study escalation in offense seriousness. Rojek and Erikson (1982) searched for long-term trends based on the results ofthe test for stationarity across transitions. On the basis of their finding of stationarity in offense-switching patterns across transi- tions for juveniles in Pima County, they concluded that there was no support for a process of escalation to more serious of- fenses (1982:17~. The conclusion of no escalation for Pima County juveniles was not confirmed in the reanalysis conducted for this report. On the basis of the published data, nonstationarity was found for switching from runaway and

398 a) JJ U] ~5 .,_, a: v .,' · - I' U) 1 U) U] ,1 o a s U] ~ ~ 0 a: a) a: . U] a) ~ s s a) 3 U] 3 U] o , in . - ~ a U] s 3 ~ U] U] ~ a: ~ _ o U] O ::' I' U] u' a) a, En O' N 1 m m a: ~ En En U] a) SO SO U] o a) En OR o lo. o P. L' o .~ v :- ~ s En v SO SO . m U] a Id Q Q o OR o :~ 1 _I ] tQ - - ~c u~ . 11 ~x E~ a' [Q Y ~s~ o o ~ l ~ o - 1 ao ~o · c ~ 1 1 1 ·e u] o s :~ tD ~ cr 0 . · 1 l O1 1 ao ~ 1 ~ID ·~ 1 ~ o Ln . . 1 c~ _1 1 1 o 1 1 ~ o 1 . . 1, _I o L~ . 1 In1 ~I. O d' ~0o o ~c ~ t- CD 1 ~1 o [__i 1~i d' 1 1 ~ ~o tg ~o 1 -- ] ~ ~ s (v ra ~ ~ ~ ~ ~ ~ ~ v o ~ Q. ~ ~ ~ {Q a `e ~ ~Q ~ ~ n5 O ~ ~ ~: 3 Q ~O m . - 0 ~ ~ a ~

399 oD ~ · ~ 1 1 to to co · · ~ 1 1 1 ~- ~ ~d. 1 ~ A us ~ ~ a' . · · 1 -~ 1 1 1 a, At · 1 1 · 1 1 to to ~ 1 . . 1 ~ ~ 1 1 co a, ~ (D · 1. . . 1 ~ _I 1 1 1 ) d. 0~ a, · · 1 ~1d. ~1 1 1 _ 1 ~ - ~c up . up 11 x ·e 0 o m 0 ~5 0 Up · ~ .,' tQ 0 a, .,, 3 - s~ a) Q o ~ - 1 · 1 d' 1 Cal CO o ~ . . 1 1 ED 1 1 o o ~ 1 · . 1 ~ ~ 1 1 a' cat ~ 0 · . 1 1 a, . Up . kD ~AD · ~ 1 1 AD ~9 . 1 ao o ~ · c~ ~ 1 1 a eQ ~ >, a)~ eQ 0 ~ ~ a) ~ a, u, ~'~ ~ V 0 3 Q. ~ ~ ~O u' :, ~ ~ ~Q ~ cs O £ o: ~ 3 Q ~O al ~1 .,, o ~, ~ a) s~ a a) o Q · O s a v .,, . - ,0 ~ ~, 0 o U] U) o o ,, ·. ~L, W E~ O Z ~ 0 ,, o o . v ,' .,, .,, cn 0 ,, x ~C

400 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-50 Test of Independence of Switching Among Offense Types for Pima County Juveniles: Observed and Expected Transition Probabilities and Adjusted Standardized Residuals (ASRs) When Diagonal Elements Are Omitted (all offenders) Offense Type Offense Type of k + 1st Official Contact - of kth Other ~ Contact Persons Property Crimes Runaway Persons Observed Expected (ASR) Property Observed Expected (ASR) Other Crimes Observed .034 Expected .031 (ASR) Runaway Observed Expected (ASR) Other Status Other Status Desist] __ __ .375 .232 (3.9)*** .117 .158 .125 .094 .146 .168 .163 .084 .131 .151 (4.2)*** (3.7)*** (1.7) .036 --.110 .157 .028 (2.1)* .287 .235__ (2.6)** .015 .022 (-1.7) .150 .285 (-3.4)*** .182 .255 (-7.9)*** -- .140 .148 .167 .169 .048 -- .107 .068 -- .122 (-3.0)** (-1.8) __ .185 .227 .170 .288 (-2.8)** .251 .207 (4.3)*** Observed .028 .205 .084 .119 -- .431 Expected .034 .256 .103 .161 -- .314 (ASR) (-4.0)*** (-2.0)* (-3.7)*** (8.1)*** x2 = 136.4*** (15 d.f.) Offenders were characterized as desisting when no further official juvenile contacts were observed. Since the data were limited to juvenile official contacts, however, offenders who appeared to desist as juveniles may have had subsequent arrests as adults. *Significant at the .05 level. **Significant at the .01 level. ***Significant at the .001 level. SOURCE: Derived from data in Rojek and Erikson (1982:Table 2). ASRS for an incomplete table (excluding diagonal values) are calculated using the program provided in Haberman (1979:Appendix). Only ASRs significant at the .10 level or better (two-tailed test using a standard normal distribution) are reported. from other status offenses. On the basis of the analysis of ASRs for switches on succes- sive transitions, switches from runaway and status offenses to person offenses and to the category of"other crimes" were less likely on early transitions and more likely on later transitions. Conversely, switches from juve- nile status offenses to Resistance were more likely on early transitions and less likely on later transitions. This suggests possible es- calation in seriousness for juvenile status offenders who remain criminally active. These results, however, must be regarded cautiously because of possible imprecisions in the reanalysis.32 An alternative approach to detecting es

APPENDIX B: RESEARCH ON CRIMINAL CAREERS calation makes use of a seriousness scale to weight arrests differentially by offense type. The average seriousness on succes- sive arrests can then be compared for up- ward or downward trends. The average se- riousness (Sellin-Wolfgang scale) of police contacts for Philadelphia juveniles is shown in Figure 3.33 A distinctive upward trend is evident; average seriousness increased by about 4.3 on each police contact, from an average level of 100 on the first contact up to an average of 165 on the fifteenth. Seri- ousness scores in this scale can vary over a very large range-from values under 100 to over 2,600. An increase in the average from 100 to 165 might arise, for example, if all first contacts had scores of 100 (associated with such offenses as forcible entry and thefts of under $10) and later contacts had a mixture of scores with 89 percent at 100 and 11 percent at 700 (associated with offenses involving injuries that require hospitaliza- tion of the victim or thefts of over $80,000~. In race-specific analyses, a similar upward trend in seriousness was found only for nonwhite offenders; average seriousness for white offenders was far more variable over successive arrests. The analysis of escalation by Wolfgang, Figlio, and Sellin (1972:16~168) focused on changes in seriousness within individual offense types. Focusing on theft offenses, for example, revealed very little change in the seriousness of theft offenses on succes- sive arrests. Thefts on the tenth police con- tact were of about the same average serious- ness as thefts occurring on the first police contact. Only injury offenses exhibited a 32The replication results may be distorted by errors in the frequency counts used in the reanalysis arising from (1) imprecision in estimating frequency counts from published transition probabilities and (2) possible typographical errors in the published data, which are suggested by inconsistencies in frequency counts calculated from the various pub- lished tables. 33The Sellin-Wolfgang seriousness scale (Sellin and Wolfgang, 1964) was used to weight the serious- ness of every police contact. The scale was extended here to include scores for non-index offenses, and all scores were multiplied by 100 to avoid decimal values. 40 1 80 1 70 - - 1 a' o C, in an a) c o ._ U) ~ 120 to ~ 1 10 1 60 1 50 140 130 1 00 , ~S(k)= 100.730+4.315k R2 =.69 (t = 5.389) · / . 90, ;~ I i 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1r 6 8 10 12 14 16 2 4 A crest N u m ber-k FIGURE B-3 Trends in average seriousness on successive police contacts for offenders in Phila- delphia bird cohort (all offenders). Data from Wolfgang, Figlio, and Sellin (1972:Table 10.4 and Figure 10.2). significant positive trend on successive ar- rests. That trend, however, can be attrib- uted completely to very high seriousness scores on the last two observations. No trend was observed when these later police contacts, which were based on very small numbers of offenders, were excluded.34 On the basis of the general absence of strong upward trends in seriousness for individual offense types, the authors concluded that "the notion that offense severity is posi- tively related to the number of offenses that one commits is not strongly substantiated by these data" (p. 1671. This is a narrow view of escalation. While changes in seriousness within an offense type are certainly one component of escala 34Using average seriousness-of-injury offenses, S(k), on the first 15 police contacts (k = 1,2, . . ., 15), the regression results are, S(k) = 313.20 + 17.81k, R2 = .28. (t = 2.256) Excluding the last two observations, S(k) = 395.48 + 2.05k, R2 = .01. (t = .354)

402 lion, increases in the seriousness of offend ing may also result from changes in the mix of offense types on successive arrests. The . . Increase In average seriousness on succes sive police contacts evident in Figure 3 reflects a combination of shifts to more se rious offense types on later police contacts and any increases in seriousness within individual offense types. Overall, offense-switching patterns in the Philadelphia sample were generally sta tionary over successive transitions. A more detailed look at switching for individual offense types using ASRs, however, re vealed some isolated shifts toward more serious offenses on later transitions. The tendency for non-index offenses to be fol lowed by another non-index offense on the next police contact decreased on later tran sitions, and the tendency to switch from non-index to injury offenses increased. Likewise, the tendency to switch from "combination" to injury offenses increased on later transitions, and switches from com bination to damage onenses aecreasec~. all cases, there were increases in switches to more serious offenses and decreases in switches to less serious offenses on later reflect differences among offenders rather transitions. Just as was found in the thana trend towardless serious offenses as reanalysis for Pima County juveniles, there individual offending continues. To control ' ~- - for this potential selection effect, offenders were partitioned by the number of arrests in their arrest records, and the analysis of trends was restricted to the common se quence of arrests for a subset of offenders. Thus, in comparing average seriousness on each arrest up to the sixth, for example, only offenders who had at least six arrests were included. As illustrated in Figure 4 and Table 52, when record length was controlled, average seriousness was generally stable over suc cessive arrests within all offender sub groups, except white offenders in Detroit. Moreover, in a simple regression, average seriousness was always lower for offenders who had longer records, and there was a significant negative effect of record length CRIMINAL CAREERS AND CAREER CRIMINALS seriousness generally declined except for the last observation, which was anoma- lously high. When this last observation is excluded from the regression, the negative trend coefficient becomes significant [S(k) = 3.11 - .0329k, with t = 3.2951. To ensure that the results on seriousness were not sensitive to the particular seriousness scale used, a number of alternative scales were compared in the analysis of Washington, D.C., adults (Moitra, 1981), and significant declines in seriousness were observed for all scales used. The analyses of adults also reveal a po- tentially important confounding selection effect in the observed trends of seriousness. Because offenders had different numbers of arrests, the same offenders were not ob- served over the full sequence of arrests. For early arrests, average seriousness reflects Me contributions of a mixture of offenders, some win only a few arrests and some win long records of arrests. As the arrest number increases, however, average seriousness is increasingly restricted to offenders with large numbers of arrests. Thus, the decline in seriousness observed in Table 51 could It 1 __ ~1 A__ appears to be some escalation In ser~ous- ness for Philadelphia juveniles, especially for nonwhite offenders in that sample. All Me adult samples reviewed here were also examined for changes in average seri- ousness on successive arrests. In contrast to Me increases in seriousness found among juveniles, average seriousness declined on successive arrests in all adult samples. When the same seriousness scale was used on all samples, small but statistically signif- icant downward trends were found over successive arrests for all adult samples, ex- cept for black offenders in Detroit (Table 51~.35 Even in this lamer subgroup, average Lathe seriousness scores are directly comparable across the various samples. Public order and "other" offenses, which were excluded from most of the analyses of Michigan samples, were included in computing average seriousness. The scale scores ranged from O to 12. Homicide was assigned a score of 12, rape 9, robbery 7, aggravated assault 5.5, drugs 4.4, burglary 3.4, larceny 2.6, auto theft 2. 1, weapons 1.5, fraud 1.1, public order 0.7, and "other" 0.3.

APPENDIX B: RESEARCH ON CRIMINAL CAREERS TABLE B-51 Changes in Average Seriousness, S(k), on Successive Arrests (k)for Adult Offenders: Regression Results of S(k) = a + bk Value Detroit SMSA Southern Michigan SMSA Washington, of k All Races Blacks Whites All Races Blacks Whites D.C. 403 1 2.72 3.10 2.35 2.41 2.72 2.17 2 2.74 3.10 2.38 2.39 2.79 2.07 3 2.70 3.14 2.28 2.34 2.74 2.01 4 2.62 2.96 2.53 2.27 2.61 1.97 5 2.53 2.87 2.27 2.40 2.86 1.99 6 2.45 2.84 2.17 2.34 2.66 2.06 7 2.38 2.73 2.14 2.36 2.77 1.99 8 2.43 2.83 2.18 2.21 2.70 1.81 9 2.41 2.57 2.30 2.47 2.88 2.12 10 2.45 2.84 2.22 2.28 2.56 2.05 11 2.40 3.02 2.06 2.08 2.38 1.82 12 2.41 2.76 2.21 2.06 2.42 1.73 13 2.39 2.87 2.13 2.35 2.91 1.79 14 2.08 2.44 1.87 2.01 2.21 1.84 15 2.34 3.18 1.82 2.27 2.39 2.15 _____________________________________________________________ 3.37 3.27 3.26 3.09 3.16 3.13 3.07 2.95 2.98 2.91 __ a 2.74 3.03 2.42 2.43 2.85 2.09 3.38 b -0.0335 -0.0188 -0.0299 -0.0179 -0.0258 -0.0142 -0.047 (-6.42) (-1.58) (-5.20) (-2.59) (-2.40) (-1.83) (-9.20) NOTE: The t-statistics of estimates are reported in parentheses. SOURCE: Blumstein, Cohen, and Das (1985:Table 19). in all samples except white offenders in southern Michigan. These results illustrate the importance of controlling for potential selection effects. Without adequate con- trols, systematic differences in the mix of offenders who are compared can be incor- rectly interpreted as reflecting develop- mental changes during individual criminal careers. A similar negative relationship between record length and offense seriousness was also noted in Frum (1958) and in Soothill and Gibbons (1978~. The lower average seriousness for adult offenders who have long records may be due to differences in incarceration experiences for offenders. In- carceration is routinely imposed on offend- ers convicted of serious offenses, and a record of serious offenses further increases the likelihood of long prison terms at sen tencing (Blumstein et al., 1983~. The ex- pected reductions in time free in the com- munity that accompany serious offenses would make it more difficult for offenders to accumulate large numbers of arrests for se- rious offenses. Just as the apparent decline in serious- ness for adult offenders results from differ- ences in seriousness across offenders, dif- ferences among offenders may also account for the increases in seriousness on succes- sive police contacts for juveniles. As the number of police contacts increases, aver- age seriousness measures depend increas- ingly on offenders with large numbers of contacts. If more serious offenses are more common in the records of these more active juvenile offenders, perhaps because of a more limited use of incarceration for juve- nile offenders, the changing mix of offend

404 - y - cn 3.5 _ a cat `,, 3.0 v, a, c v' :' o ._ ~ 2.5 Number of Prior Arrests > p all= 6 p = 8 I,, lap= 10 0 5 10 Arrest Sequence (k) FIGURE BY Average seriousness on successive arrests, k, for arrestees win at least p arrests, Wash- ington, D.C., adults. Source: Troika (1981:Figure 2.4). ers alone could produce the observed . . . increases in average seriousness on succes sive contacts. TABLE B- 5 2 CRIMINAL CAREERS AND CAREER CRIMINALS The analyses in Bursik (1980) ant] Smith and Smith (1984) explicitly controlled for record length among juvenile offenders. The analysis in Bursik (1980) was restricted to the first four switches for juveniles with five or more contacts. Bursik (1980:856) re- ported that the aggregate switching process was stationary over those four transitions. Unfortunately, the slate necessary to repli- cate the analysis and to examine switching for individual offense types were not pub- lished. Smith anti Smith (1984:Table 6) ex- amined offense patterns for offenders who had at least six arrests as juveniles. Their data indicate a trend toward increased prob- abilities of robbery and injury offenses on later arrests for these recidivistic offenders. Since all offenders have at least six arrests, only the sixth arrest is potentially biased toward more serious offense types by in- cluding the last arrest that led to the sam- pled incarceration in the analysis. While these results Swim appropriate controls for record length are consistent with escala- tion in seriousness on successive arrests for Changes in Average Seriousness, S(k), on Successive Arrests (_) for Adult Offenders with m Prior Arrests when Sampled: Regression Results of S(k) = a + bk + cm Southern Michigan Parameter Detroit SMSA Values Blacks Whites Blacks Whites Washington, D.C. a 3.142.552.842.133.7 A - b -.0058.0341.0188-.0079 (-.52)(3.94)(1.45)(-1.03) c -.0300-.0581-.0279-.0087-.0640 (-2.31)(-6.7)(-2.15)(-1.14)(-6.44) Degrees of freedom117117 ~2 117 117 .072 .28 .04 .04 63 .50 NOTE: The t-statistics of estimates are reported in parentheses. SOURCE: Blumstein, Cohen, and Das (1985:Table 21).

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS juveniles, the authors interpret the results as reflecting increased diversification for offenders first arrested for nonindex or prop- erty offenses and increased specialization for those first arrested for robbery or injury offenses. Markov Property. Markov processes are distinguished by the fact that transitions at any point depend, at most, on the current state of the process. In a Markov process, knowledge about the sequence of previous states leading to the current state provides no useful information in predicting the next state. Characterizing offense-type switching as a Markov process involves assuming that, with the exception ofthe offense type ofthe current arrest, the offense type of the next arrest is independent of all other previous arrests. Markov processes are particularly useful analytic devices for predicting various fu- ture attributes of a switching process. If offense switching on successive arrests can be adequately represented as a Markov chain, it is a relatively easy matter to esti- mate the chance that some future arrest will be for a specific offense type. The chance of a robbery charge on the third arrest, for example, can be obtained directly from the estimated transition matrix and the initial distribution of offense types. The expected distribution over different offense types can also be estimated for any anticipated num- ber offuture arrests. With respect to special- ization, one can estimate the expected num- ber of arrests in other offense types between returns to an offense specialty, or the ex- pected number of successive arrests within an offense specialty. If the transition proc- ess includes transitions to Resistance asso- ciated with real termination of offending or with truncation of observation periods, it is possible to estimate the expected number of arrests before desistance will be observed. Similar predictions of future offending are usually possible if offense switching is not Markovian, but they are typically more dif- ficult to estimate. While Markov processes are very useful from an analytic perspective, they are not particularly satisfying substantively or oper- ationally. Markov processes are "memory 405 less" in that dependence on prior history is limited only to the current offense type. If offense switching were to be adequately represented as a Markov process, then prior history of offense types would not usefully distinguish future offending. An offender with a long sequence of arrests for some offense type, say robbery, would be no more likely to repeat with robbery on the next arrest than would an offender with no other prior arrests for robbery other than the current arrest. All the studies of juvenile offenders re- viewed here invoked the Markov property in characterizing offense switching. The limited manner in which this property was applied, however, suggests a failure to rec- ognize that the Markov property is not es- sential for estimation and analysis of transi- tion matrices. While the Markov property was invoked, it was tested in only a limited fashion or not at all, and, with the exception of Smith and Smith (1984), none ofthe useful predictions that derive from the Markov prop- erty was explored in the analyses. The one study of juveniles reporting a limited test of the Markov property is Wolfgang, Figlio, and Sellin (1972~. Here a second-order transition matrix was consid- ered in which the offense type on the next arrest depended on the offense types of both the current and the immediately pre- ceding arrest. A x2 test was performed on the estimated transition matrix for the first three police contacts to assess the adequacy of this dependency structure against a null hypothesis that switching is dependent only on the current offense type. The result- ing x2 value was significant at the .07 level (X2 = 100.03, 80 d.f.), and thus there is a reasonable basis for rejecting the null hy- pothesis of first-order independence. How- ever, using a .05 threshold of significance, the authors accepted the null hypothesis of independence, but they advised caution with respect to this analysis because of the small number of cases available to generate some of the transition estimates (1972:18 1871. Aside from the problem of small sample sizes, the aggregate x2 test is not sufficiently sensitive to violations of independence for

406 individual offense types. A more sensitive test of the Markov property is available. This test relies on the well-known property of Markov chains: the j-step transition ma- trix, Pj, for offense-type switches between arrests separated by j - 1 intervening ar- rests should be given by the product of the j intervening one-step matrices, P(k), with j Pj= ~ P(k)e k=1 In comparing offense types on the first and fourth arrests, for example, offense switching would be represented by a three- step matrix, P3. The independence of Markov processes means that this three- step matrix can be obtained directly by successively applying the transition matrix from the first to second arrest, P(1), and then the transition matrix from the second to third arrest, P(2), and finally the transition matrix from the third to fourth arrest, P(3~. In this example, the three-step matrix is given by P3 = P(1) x P(2) X P(3~. For a stationary Markov process, where P(kJ = P for all k, P3 = P3. Moitra (1981) was the only study re- viewed that explicitly applied this test of the Markov property to offense-type switch- ing. Using data on offense switching be- tween arrests for adults arrested in Wash- ington, D.C., Moitra (1981:Table 2.11) compared the two-step transition matrix to the product of the two intervening one-step matrices. As is commonly observed in anal- yses of other social processes (e.g., residen- tial migration, social mobility), greater spe- cialization (indicated by larger diagonal values) was found in the observed two-step matrix, P2, than was expected from the prod- uct ofthe one-step matrix, p2. This tendency toward greater specialization over succes- sive arrests was a basis for rejecting the simple Markov characterization of no de- pendence on prior history in offense switch- ing. Ordinarily, failure of the Markov prop- erty introduces unacceptable errors into long-term predictions, but short-term pre CRIMINAL CAREERS AND CAREER CRIMINALS dictions based on the Markov property are reasonably accurate. The failure of the Markov property in the short term, repre- sented by a two-step matrix, however, sub- stantially limits the usefulness of any pre- dictions of offense switching based on a simple Markov model. Rather than offense switching being inde- pendent of prior history, it appears that offenders may in some way be distinguish- able in terms oftheir accumulated record of prior offense types. In particular, offense switching appears to be more specialized than expected under a simple first-order Markov assumption. This failure of the Markov property has both analytic and sub- stantive consequences. The Markov model is most useful as an analytic tool for making projections about future offending patterns based on available estimates of one-step switching patterns. These projections-e.g., the expected distribution of offense types for future arrests are less reliable when the process violates the basic Markov as- sumption. While often useful for the in- sights they provide, such projections are not necessary, and empirically based transition matrices can still be estimates! from ob- served offense switching and their structure analyzer] without satisfying the Markov property. Any substantive insights into of- fense-switching patterns that emerge from such analyses, however, are limited by their failure to go beyond the immediately pre- ceding offense type when characterizing the influence of prior criminal history on future offending. Offenders whose histories include many instances ofthe same offense type, for example, may exhibit greater spe- cialization in future arrests than is found when the analysis is restricted to only the current offense type. When the Markov property fails, analyses that do not rely on transition matrices may be more appropri- ate. Methodological Issues in Analyzing Offense Seriousness Studies of offense seriousness during criminal careers attempt to varying de- grees to characterize both the mix ant!

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS sequence of offense types found during ca- reers. Two general approaches are avail- able, one that attempts to summarize entire careers, and another that focuses on succes- sive arrests. Included among the summary approaches are various statistical measures, such as a simple percentage distribution of the offense types found in a career, the geometric scaling proposed by Shannon (1968) to capture the mix of different offense types during a career, and an actuarial tech- nique of risk assessment proposed by Rob- ins and Taibleson (1972) to assess the effect of one offense type on the probability that another offense type subsequently occurs. The approach focusing on changes over successive arrests includes, primarily, anal- yses of transition matrices of switching be- tween offense types first introduced in Blumstein et al. (1967) and Blumstein and Larson (1969~. A number of methodological concerns have emerged in this review of empirical research on offense seriousness. Since most recent analyses of offense seriousness have relied on transition matrices, the various problems are discussed below primarily in relation to this technique. The same prob- lems, however, are sometimes of equal con- cern for other, simpler descriptive tech- n~ques. Requirement for Large Sample Size The data requirements for analyses of offense switching are greater than those for analyses of other aspects of individual of- fending. While participation, frequency, and duration can be partitioned by single offense categories, analysis of offense switching is based on pairs of successive offense types. Thus, the collection of all robbery events that together form the datum for estimates of participation, frequency, and duration is partitioned into smaller sub- sets depending on the offense type of the next event when estimating offense-switch- ing patterns. Even a reasonably large number of ar- rests often reduces to unreliably small num- bers of observations for offense pairs when offense switching is analyzed by individual 407 transitions and for a reasonable number of distinct offense types. Compared with a simple linear growth in data needs for other analyses, data needs grow multiplicatively in analyses of offense-type switching. Lack of sufficient data often limits the usefulness of analyses of offense-switching patterns. Data limitations are especially a problem when there is extensive variability in individual offense-switching patterns as a result of population heterogeneity (i.e., dif- ferences across individuals) or nonstationar- ity (i.e., variability in the switching process on different transitions). By using transition matrices, transition probabilities can be es- timated from the proportions observed for various offense pairs in samples of offend- ers. The adequacy with which these proba- bility estimates characterize offense-switch- ing patterns depends on the extent to which the collection of offense pairs used results from a reasonably homogeneous switching process. As variations in the switching proc- ess increase, the estimates of transition probabilities are a less accurate reflection of offense-switching pattems. When there is high variability in the switching process, it is desirable to partition the data, by popula- tion subgroup or by transition number, to form subsets that are more homogeneous with respect to offense switching. Such par- titions, however, greatly increase data re- quirements. While the results varied across data sam- ples, the studies reviewed here indicate some important differences in switching patterns for different population subgroups (especially race and sex) and some non- stationarity on successive transitions (espe- cially for adults). Analyses on separate sub- groups and for different transitions are thus important in any study of offense-switching patterns. Ensuring the capacity to partition the data adequately puts increased de- mands on the sample size required for anal- ysis. When the sample size is inadequate to the task, transition matrices should not be used; other, simpler techniques that de- scribe the changing mix of offense serious- ness are more appropriate. These might include differences in the percentage re- peating an offense type based on the num

408 her of prior arrests for that same type found in a career, or using some weighted statistic for that offense type that gives greater weight to more recent occurrences of the offense type. Reliance on Official Records All of the studies ot ottense switching reviewed here relied on official-record data on sequences of police contacts or arrests for samples of offenders. This dependence on official-recorc! data arises from the re- quirement for data that document the exact sequences of offense types over time" in- formation that is readily available in official records. The picture of offense switching that emerges from analyses of official-record data, however, confounds patterns of of- fense switching by offenders with patterns of law enforcement, especially by the po- lice. As noted earlier, the offense types observed on successive police contacts or successive arrests will vary with the levels of police effectiveness in apprehending of- fenders for different offense types. Offense types with higher detection and apprehen- sion rates will be overrepresented among official contacts compared with their repre- sentation in successive crimes committed. If enforcement rates vary substantially for different offense types, the patterns of switching observed in official-record data will provide a distorted view of offense switching between actual crimes commit- ted. This confounding effect is recognized in virtually all studies of offense switching, and the studies are careful to note that the reported results apply to successive official contacts for offenders. The variability in arrest risk for different offense types is illustrated in Table 53. Based on data for the United States, the ratio of arrests to reported crimes (in column 3) varies from a low of .12 for auto theft to a high of.42 for aggravated assault. The ratio of arrests to reported crimes alone, how- ever, is an inadequate estimate of the chance of arrest for a crime for an indi- vidual. Crimes committed but not reported to the police are not included, and arrests CRIMINAL CAREERS AND CAREER CRIMINALS sometimes include arrests of more than one individual for the same crime incident. The number of reported crimes can be adjusted for nonreporting by using the re- porting rates for various offense types avail- able from national surveys of criminal vic- timization. A further adjustment for multiple offenders per crime incident is also available in these national surveys.36 The adjusted estimates of the probability that any individual offender is arrested for a crime committed, whether reported to the police or not, are shown in the last column in Table 53. The risk of arrest is highest for offenses involving direct contact between offenders and victims (robbery and aggra- vated assault) and lowest for property of- fenses without contact. The final arrest risk for the various offense types is generally low, averaging only 1 arrest for every 20 crimes committed. De- snite the reduction in arrest risk after the adjustments, there is still a threefold differ- ence between the highest risk (aggravated assault) and the lowest risk (larceny and auto theft). Variability in arrest risk for dip ferent offense types is thus a very real con- cern in analyses of offense switching that rely on official contacts only. Two strategies are available for dealing with distortions in offense-switching pat- terns that arise from use of official-record data. The first is to expand the scope of self-reports of offenses committed to in- clude data on the actual sequence of dif- ferent offense types. To date, such data on sequences of crimes actually committed have been unavailable. It is only recently Mat self-report studies have begun to col- lect data on frequency of offending during a reporting period. Collecting data on the sequence of offense types will require 36The adjusted arrest risk for a crime of offense type k, qk, is given by Arrests of Type k rk qk = X- Reported Offenses of Type k °k where rk is the rate of reporting offenses to the police by crime victims and °k is the average number of offenders per crime incident.

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS 409 TABLE B-53 Probability of Arrest for a Crime, Adjusted for Nonreporting to the Police and Multiple Offenders per Crime Proportion Offense Type Reported Offenses in U.S., 1980a Arrests in UPS., 1980 Ratio of of Total Arrests to Offenses Probability Number of of Arrest for Offenders a Crime, A, Reported Reported per Crimg for Individual Offenses to Police- Incident- Offenders Robbery548,809146,270 Aggravated assault654,957277,470 Burglary3,759,193513,300 Larceny7,112,6571,191,900 Auto theft1,114,651138,300 .27 .42 .14 .17 .12 .57 .54 .51 .27 .69 2.3 2.6 1.6 1.6 1.8 .07 .09 .04 .03 .03 federal Bureau of Investigation (1981:Table 1). bFederal Bureau of Investigation (1981:Table 24). bureau of Justice Statistics (1982a:Table 89). dReiss (1980b:Table 2). panel designs that include data collection from the same sample of offenders at fre- quent intervals. Depending on the antici- pated rates of individual offending, monthly or perhaps even weekly reports may be required. Given that the focus of the research is offense seriousness, the strategy of repeated and frequent self-reports is best limited to samples of known ollenders. Those offend- ers might be identified from self-reported offenses in a more widely used screening instrument, or through arrest or police con- tact associated with an offense. As noted earlier, self-report studies involving re- peated and frequent reporting will be costly, will require a reasonably long-term commitment-of at least several years- to data collection, and will involve difficult logistics in order to maintain contact throughout the study with samples whose members are likely to be uncooperative and mobile. While the self-report approach is certainly possible, the various implementa- tion problems in addition to the large sam- ple sizes required to estimate switching patterns make pursuit of this research strategy all the more difficult. The data re- quirements are somewhat less demanding if repeated self-reports are used to estimate changes in offense mix during successive reporting periods. Such analyses would not require data on the exact sequence of of- fense types and smaller samples of offend- ers would be adequate. An alternative strategy for analyzing the actual sequence of crimes committed builds on the current reliance on official-record data, extending it to address offense switch- ing between crimes actually committed. As we learn more about the links between individual offending and the criminal jus- tice selection process, we will be better able to model the selection process. By incorporating models of the selection proc- ess with readily available official-record data, we can begin to draw inferences about the switching process for undetected crimes that intervene between official-record events. This inferential strategy has begun to be employed with some success in stud- ies of individual crime rates based on offi- cial-record data. Biases Associated with Sample Selection The most obvious biases arising from the sampling process are distortions introduced by the sampling event itself. These are most

410 likely to arise when sampling is based on some threshold of seriousness in offense types. The analyses of criminal histories for a sample of prison inmates by Frum (1958) and incarcerated juveniles by Smith and Smith (1984) are excellent illustrations of this problem. The samples were drawn from among inmates in state correctional facilities. Since all sample members were incarcerated for the last arrest in their rec- ords, that last event was likely to be for a serious offense type or to follow a record of repeated convictions for serious offense types. The sampling strategy of using incar- cerated offenders, and the failure to exclude the last offense type from the analysis, were no doubt major factors contributing to the findings of escalation toward more serious offense types over the course of criminal careers and of the tendency for some of- fenders to specialize in serious offense types. When sample selection is based on a seriousness threshold, it is essential that the sampling event be excluded from analysis of offense-switching patterns. Failure to ex- clude the necessarily more serious sam- pling event will bias estimates of offense .. . . switching patterns toward these more serious events. It was precisely to avoid such biases that the more serious sampling event was excluded from analyses of adult arrestees by Moitra (1981) and Blumstein, Cohen, and Das (1985~. Sample selection in Rojek and Erikson (1982) and Bursik (1980) was based on ei- ther processing by the juvenile court or an adjudication as a delinquent in the juvenile court. Given that the discretion to resolve juvenile cases informally is available to both the police and to intake officers at juvenile court, the formal involvement of the juvenile court likely increases the seri- ousness of the sampling event in both sam- ples. The sampling event was not excluded from either analysis. The sampling event, however, was not restricted to the last event in the record; it could appear anywhere in the record, depending on the age of the offender during the sampling period. This distribution of the sampling event over dif- ferent points in a record limits the biasing CRIMINAL CAREERS AND CAREER CRIMINALS effect toward more serious events at the end of the record. On the other hand, it may be responsible for findings of stationarity over transitions as intermittent escalations in se- riousness associated with the sampling event are randomly distributed over indi- vidual arrest histories, obscuring any pat- terns over time that may otherwise exist. Two strategies are available to control for distortions arising from the sampled event. First, the sampled event can be excluded from the analysis entirely. This strategy is especially appropriate when the sampled event falls at the end of arrest history data. The alternative is to include the sampled event in the analysis, but to limit the anal- ysis to similar events. Thus, in the two juvenile court samples, analysis of offense switching would be limited to contacts processed by juvenile court (in the case of Rojok and Erikson, 1982) or to offenses that were adjudicated delinquent (in Bursik, 1980~. In this way, the sampling event is indistinguishable from other events in the analysis. This strategy of only analyzing events similar to the sampling event was employed in the study of offense switching by juveniles in the Philadelphia cohort (Wolfgang, Figlio, and Sellin, 1972~. Like the adult analyses, this study of juveniles is free of biases associated with the sampling event. Aside from biases introduced into the switching process by the sampling event, the sampling process itself selectively lim- its the population of offenders who are stud- ied. All analyses of offense switching re- quire at least one official-record event (police contact, arrest, juvenile court proc- essing, juvenile court adjudication, convic- tion, or incarceration). Those that exclude Resistance require at least two contacts for each offender. The switching patterns ob- served thus apply most accurately to sub- sets of offenders with official records. Even if the criminal justice selection process was completely random, offenders with official contacts would be a random sample of all offenders only if all offenders are homoge- neous in offending. Any variability in of- fending (e.g., higher frequency rates for some offenders compared with others, or

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS N longer criminal careers) would increase the representation of more active offenders in the sample. Because their greater criminal activity increases their exposure to risk, of- fenders with higher frequencies and longer criminal careers are more likely to experi- ence an official contact and thus are more likely to be found in samples.37 Most anal- yses of offense-switching patterns, there- fore, reflect offense switching for the more active offenders who are found in the sam- ple and may not apply to all offenders. Role of Frequency Rates, Career Length, and Incapacitation in Switching Patterns Frequency rates, career length, and time spent incarcerated vary for different offense types. As reported in the review of fre- quency rates above, individual crime rates are higher for property crimes and lower for violent crimes. Analyses of the length of criminal careers (Blumstein and Cohen, 1982) report an opposite relationship: shorter average careers in property crimes and longer average careers in violent crimes. Time spent incarcerated is also likely to be longer for violent crimes than for property crimes. These differences can affect the switching patterns observed, es- pecially when observation periods are lim- ited in length. In particular, offense types that occur at high rates, and thus involve short intervals between events and short periods of incarceration, are likely to be more prevalent as switching destinations. Conversely, switching to offense types that involve longer average intervals between events and longer periods of incarceration is likely to be underrepresented, especially when observation periods are short. The possible distorting effect of the dis 37This tendency to oversample more active of- fenders holds under a variety of conditions. The only exception is those instances in which selection risk per offense committed is strongly inversely related to individual offending patterns win high- rate or long-career offenders having a much lower risk of official contact per offense than low-rate or short-career offenders. 4~] tribution in the various offense types was evident in the analysis of specialization for adults. Without controlling for differences in the distribution of offense types, burglary and larceny have the largest diagonal switching probabilities of all offense types for adults, which suggests greater special- ization in these offense types by adult of- fenders. Examination of the column marginals for these offense types, however, reveals that switching to these offense types is higher generally. Thus, the tendency to specialize in burglary or larceny is not es- pecially great relative to the generally higher frequency of switching to burglary and larceny as the next offense type. Con- versely, even apparently small diagonal val- ues may reflect significant specialization when switching to an offense type is gener- ally quite rare. The distribution in different offense types is explicitly controlled in all analyses in which the observed frequency of switches is tested against a model of complete independence in switching. Dif- ferences in the distribution of the offense types are reflected directly in the frequency of switches expected in an independent process. Variations in the number of events in a criminal history are one indication of dif- ferent levels of offending and differences in incapacitation experiences. Most directly, differences in the number of events will reflect variations in individual frequency rates and in career length. High-rate of- tenders are more likely to accumulate large numbers of events, as are offenders who remain criminally active for long periods of time. Extended periods of incarceration during careers. bv contrast. will limit the c , , , number of events in a career. Because lev- els of offending and incapacitation experi- ences may also be associated with the of- fense types found in a record, differences in the number of events can affect analyses of offense-switching patterns. This potential source of bias in analyses of switching was illustrated most dramati- cally in the earlier examination of escalation effects. Without controls for differences in the number of arrests for different individ- uals, average seriousness appeared to de

412 cline with each additional arrest for adults. The analysis, however, was not based on the same sample of individuals at each ar- rest. Offenders with only a few arrests con- tributed to the average seriousness of early arrests, but seriousness on successive ar- rests was increasingly based on offenders with larger numbers of arrests. Thus, the observed decline in seriousness could re- flect differences among offenders, and not a change as individual offending progresses. The key role of population heterogeneity was confirmed when controls for this sam- ple-selection effect were introduced. Con- trolling for the number of arrests in a his- tory, average seriousness was generally stable on successive arrests for adults. Av- erage seriousness, however, was lower for adult offenders who had larger numbers of arrests. Variations in record length among offend- ers is a similar concern in estimating switch- ing probabilities more generally. More ac- tive offenders, with their larger numbers of arrests, will contribute disproportionately to estimates of a single, summary transition matrix that combines all offense switches together. To the extent that switching pat- terns vary with record length among of- fenders, the combined matrix estimate will be biased to reflect the pattern of offenders who have long records. This potential bias is partially controlled by estimating sepa- rate transition matrices for each offense switch; variations in switching with record length will be evident in the variability (nonstationarity) across the separate matri- ces. These separate matrices, however, are subject to the same sample-selection biases affecting average seriousness. Successive matrices are based on an increasingly more selected sample of offenders those with larger numbers of arrests. Thus any trends in switching observed over successive ma- trices may reflect population heterogeneity and not a progression in switching during individual criminal careers. The potential role of selection effects in successive transition matrices was illus- trated in the reanalysis above of the data on juvenile offenders in Pima County. The reanalysis found nonstationarity in switch- ing probabilities from juvenile status of CRIMINAL CAREERS AND CAREER CRIMINALS tenses, with more Resistance on early tran- sitions compared with later transitions, and more switches to personal and "other" crimes on later transitions compared with early transitions. This pattern suggests an escalation in seriousness for status offend- ers. The successive transition matrices, how- ever, were not estimated using the same sample of offenders on each transition. Of- fenders with only a few police contacts were selected out of the analysis through early Resistance. Thus, the apparent trend to more serious offending for status offend- ers may reflect a selection effect in which status offenders with a small number of contacts were also less serious offenders. These less serious offenders, however, only entered the estimates of early transition matrices. Later transition estimates were based increasingly on status offenders who had larger numbers of police contacts. If these more active status offenders were also more serious offenders generally, the trend to more serious offending observed on suc- cessive transitions would reflect this popu- lation heterogeneity and not a tendency to move to more serious offenses for individ- ual status offenders. As in the analysis of trends in average seriousness, the effects of this form of pop- ulation heterogeneity can be explored by estimating successive transition matrices af- ter controlling for the number of arrests in a history. This, however, places increased de- mands on the sample size necessary for analysis. Population heterogeneity, especially with respect to record length, represents a strong competing hypothesis in accounting for differences in offense patterns observer! in the studies reviewed here. In comparing adult and juvenile offenders, for example, greater specialization was observed for adult offenders than for juvenile offenders. In the juvenile years, offender samples may consist of some casual offenders whose of- fending is exploratory and ends quickly and of other more committed offenders who are specialized in their offending. As explor- atory offenders leave offending in the juve- nile years, adult samples would consist more heavily of committed, specialized of

APPENDIX B: RESEARCH ON CRIMINAL CAREERS fenders. In this event, the difference in specialization for adults and juveniles might arise from differences across offend- ers and not from a developmental process toward greater specialization as offenders get older. Sorting out these rival hypotheses requires analyses of offense-switching pat- terns for a common sample of offenders who begin offending as juveniles and persist into adulthood. Aside from the potential distortions asso- ciated with variations in the number of events in criminal histories, variations in the length of observation periods may also affect the switching pattern observed. Of- fenders' frequency rates and career lengths, as well as their incarceration experiences, all affect the length of intervals between events. Inter-arrest intervals, as notecl, will be sho* when individual arrest rates are high. When individual arrest rates are low, by contrast, or when long periods of incar- ceration are likely to substantially reduce the time at risk for subsequent arrests, inter- arrest intervals are more likely to be long, and these intervals will only be observed in longer careers. To the extent that frequency rates, career length, and incarceration risk vary across different offense types, the asso- ciated differences in intervals between events for different offense types can affect the mix of offense types observed in switch- ing data. In particular, offense types charac- terizec] by short inter-event intervals are more likely to be observed when observa- tion periods are short. ~. Correspondingly longer observation periods are required if offense types characterized by longer inter- event intervals are to be adequately repre- sented. Variations in inter-arrest intervals for dif- ferent offense types may affect the estimates of transition matrices. All the analyses of offense switching reviewed here have sup- pressec] differences in the time intervals between events. Switching events were de- fined by the occurrence of a next arrest (or police contact), and switching events were aggregated regardless of the differences in the time interval to that event. The pattern of switching among offense types, however, may vary with the length of inter-arrest intervals. 4~3 Building on the differences in frequency rates and career lengths observed for dif- ferent offense types, for example, it might be expected that switches to property of- fenses-with their higher individual fre- quencies and short careers would be more likely when the intervals between events are short. Conversely, when intervals are long, greater switching to violent offenses- with their lower frequencies ant] longer careers- would be expected. The data on offense switching between arrests for adults in Washington, D.C., and Michigan are user] here to explore these potential differences. The estimated transi- tion probabilities for selected offense types for Washington, D.C., arrestees are pre- sented in Table 54. The significance of differences in switching was assessed using the ASRs of Goodman's (1962, 1968) contin- gency table approach. Taking one offense type at a time, a test was made of whether switching patterns from that offense were independent of the length of the interval to the next arrest. Although not presented here, results similar to those for Washing- ton, D.C., arrestees were also found for both whites and blacks in the Detroit and south- ern Michigan samples. Systematic variations in switching were found with differences in the length of in- tervals between arrests. Consistent with the lower frequency and longer careers in ag- gravated assault, the most persistent differ- ence was an increased tendency to switch to aggravated assault as the length ofthe inter- val between arrests increased (indicated by a shift from negative to positive ASRs). Switches to robbery, with its higher fre- quency rate and shorter careers, were more likely after short intervals (indicated by a shift from positive to negative ASRs). A decline in specialization as intervals in- creased was also observed for robbery and burglary (indicated by a shift from positive to negative ASRs). - Alternatives to Simple Markov ModeZs A simple Markov property was invoked in several analyses of offense-switching pat- terns. Under this Markov assumption, of- Sense switching Lepers, at most, on the

414 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-54 Variations in Offense-Type Switching with Length of Interval Between Arrests for Washington, D.C., Arrestees Prevalence Offense Length of Probability of of Offense Type on Interval Offense Type on k + 1st Arrest Type on kth kth Between Aggravated Arrest Arrest Arrests Assault Robbery Burglary (percent) Aggravated <1 year.300 .086.059 11.6 assault (NS)a (2.7)**(NS) (x2 = >1 year and.269 .070.065 16.9 51.03,* <2 years(NS) (NS)(NS) 33 d.f.) >2 years and.259 .026.043 18 <4 years(NS) (-2.1)*(NS) >4 years.385 .049.045 19.3 (3.0)** (-1.8)(NS) Robbery <1 year.109 .312.074 10.4 (-2.7)** (3.7)***(NS) (x2 = >1 year and.133 .301.062 10 57.03,** <2 years (NS) (NS) (NS) 33 d.f.) >2 years and .143 .214 .071 8 <4 years (NS) (NS) (NS) >4 years .195 .134 .101 10 (3.0)** (-4.4)*** (NS) Burglary <1 year .071 .083 .316 10 (-3.8)*** (NS) (4.8)*** (x2 = . . 68.96,*** >1 year and .113 .056 .185 11.3 33 d.f.) <2 years (NS) (NS) (-2.6)** >2 years and .067 .067 .225 14.2 <4 years (NS) (NS) (NS) >4 years .180 .087 .174 12.5 (4.7)*** (NS) (-3.5)*** pOnly ASRs significant at the .10 level or better (two-tailed test using standard normal distribution) are reported in parentheses. All other nonsignificant values are indicated by NS. *Significant at the .05 level. **Significant at the .01 level. ***Significant at the .001 level.

APPENDIX B: RESEARCH ON CRIMINAL CAREERS offense type of the current arrest. The lim- ited tests available for assessing the ade- quacy of the Markov assumption suggest that offense switching is not adequately modeled as a first-order Markov chain. De- pendence on prior offense types appears to extend beyond the current offense type and results in greater specialization than would be expected in this simple Markov model. The tendency for observations to bunch on the diagonals of transition matrices has been observed in a variety of social pro- cesses, most notably residential migration and status mobility. Failure of simple Markov models in these processes is often attributed to population heterogeneity, and a variety of alternative modeling strategies have been proposed (see, for example, Singer and Spilerman, 1978, for a discus- sion of the various approaches). In its most common form, the population is presumed to vary in its tendency to stay in the same state on successive transitions. In the case of offense switching, offenders would vary with respect to offense specialization. At one extreme, some offenders might be highly specialized and thereby have a high likelihood of repeatedly engaging in the same offense type. At the other extreme would be generalists, whose offending would vary randomly over many different offense types. Various alternative models have been proposed to address population heterogene- ity satisfactorily. Many ofthese models pre- serve the Markov property for switching within different population subgroups, but specify different transition matrices for each subgroup. The non-Markov aggregate tran- sition matrix reflects the combined effect of these separate Markov transition processes. The simplest, and one of the earliest, ap- proaches to population heterogeneity was the "mover-stayer" model first introduced by Blumen, Kogan, and McCarthy (1955~.38 If this model is applied to offense switch- ing, the population of offenders would be 38Various later extensions and tests of the mover- stayer model are available in the research literature; see, for example, Goodman (1961), White (1970), and Spilerman (1972b). 4~5 TABLE B-55 Distr ibution of Youths with Over Half of Their Police Contacts in a Single Offense Category White (percent) Nonwhite (percent) Personal injury Personal property Impersonal property Other ~ No n specialization" Total 1.5 1.5 34.3 28.4 34.3 100 (N = 134) 1.5 0.9 32.5 14.9 50.1 100 (N = 335) NOTE: Cook County Juveni le Court s able of youths with at least five police contacts. SOURCE: Bursik (1980:Table 5). divided into two groups-the "stayers," who would always repeat the same offense type, and the "movers," who would switch among offense types according to a common Markov transition process. Switching by both groups can easily be combined and various predictions of expected future switching patterns for the aggregate popu- lation are available. The finding of specialization in a variety of offender samples suggests that this parti- tion of offenders into specialists and gener- alists may be a fruitful approach to model- ing offense switching. While there is evidence of specialization in all samples, some offenders seem to be more likely to specialize than others. As indicated in Ta- ble 55, specialization within aggregate of- fense categories was widespread among Cook County juvenile offenders. For of- fenders with at least five police contacts as juveniles, one-half of the nonwhites and two-thirds of the whites in the sample had over 50 percent of all their contacts in a single offense category. The distribution of specialists in different offense categories reflects the relative distribution over these offense categories generally. On the basis of data on adult arrestees in Washington, D.C., the proportion of spe- cialists varies considerably for different of- fense types. As indicated in Table 56, spe

416 CRIMINAL CAREERS AND CAREER CRIMINALS TABLE B-56 Proportion of Specialists Found Among Adult Arrestees in Washington, D.C. Offense Type of Arrest Number of in 1973 Arrestees Percent with Prior Arrests for Index Offenses Percent Specialists Among Those with Prior Index Arrests Percent Specialists Among All Arrestees Murder277 65.0 - 21.1 13.7 Rape253 63.3 22.0 13.9 Robbery1,230 65.2 53.4 34.8 Aggravated assault1,930 59.6 57.0 34.0 Burglary902 55.5 44.5 24.7 Auto theft496 52.4 35.7 18.7 NOTE: "Specialists" are arrestees with prior arrests for the same charge as the sampled arrest in 1973. For arrestees with only one prior arrest for an index offense, that one prior index arrest is for the same charge as the sampled arrest. With two or more prior index arrests, the preponderance of the prior arrests are for the same charge as the 1973 arrest. Under the "predominance" criterion, about one-half of all index arrests in a record--including the 1973 sampled arrest-- must be for the current charge. For a record with a total of 3, 4, or 5 index arrests, including the sampled arrest, at least 2 must be for the current charge. For a total of 6 or 7 index arrests, at least 3 must be for the current charge. More generally, if n is equal to the number of prior index arrests of any type, and m is equal to the number of prior index arrests of the same type as the current arrest, a person satisfies the "specialist" criterion if for n > 3, m > (n - 1)/2 for n odd or m > (n - 2)/2 for n even, and for n = 2, m > n/2 = 1. SOURCE: Derived from data in Cohen (1982:Table 3-3). cialists within an offense type were most often found among offenders arrested for robbery and aggravated assault. One-third of all arrestees in these offense types had prior records ant! a predominance of arrests for the same offense type. Specialists in robbery and aggravated assault represented over one-half of those arrestees who had any prior arrests for index offenses. Special- ists were least prevalent in murder and rape, accounting for only 14 percent of all arrestees for these offense types. This lower prevalence of specialists was not due to a lower likelihood of any prior arrests. Two- thircls of the arrestees for murder and rape had prior arrests for index offenses, but less than one-quarter of those recidivistic ar- restees were specialists in those offense types. A similar mix of specialists and general- ists was evident among respondents to the second Ranc! inmate survey. As in~licatecl in Table 57, diversity in offending was very common; most inmates indicated that they committed several different offense types during the observation period. Never~e

APPENDIX B.: RESEARCH ON CRIMINAL CAREERS less, more than one-quarter of all respon- dents reported that they committed only one offense type. Only robbery was rarely committed as a sole offense type. Even among the category of"Iow-level robbers," 64 percent of the respondents (N = 153) also reporter] that they committed burglary and theft crimes during the 1- to 2-year observation period. The simple mover-stayer model and vari- ations ofthe model that permit a continuous distribution of differences among offenders (see, for example, Spilerman, 1972b) rest on an assumption of population heterogeneity. 4~7 The transition process varies across the population, but within any subgroup the transition process is invariant over time. An alternative explanation offered for the ten- dency of switching processes to bunch on the diagonals explicitly incorporates vari- ability in the process with time. This is most often Lone by allowing for duration clepen- dence in the switching process. In analyses of residential migration and status mobility, duration depenclence re- flects a phenomenon of cumulative inertia (McGinnis, 1968), whereby the probability of remaining in the same state increases as TABLE B-57 Combinations of Offense Types Committed by Respondents to the Second Rand Inmate survey Offenses Reported During Observation Period Combinations a b Drug Robbery Assault- Burglary Theft- Deals Number of Respon- Per dents cent Violent predators (robbery-assault drug deals) Robber-assaulters Robber-dealers Low-level robbers Mere assaulters Burglar-dealers Low-level burglars Property and drug offenders 0 Low-level property offenders Drug dealers Totals + + + + ? ? ? ? O o o o + o + o o + ?? + o o ? + 306 ? 0 160 ? + 188 240 105 199 171 + O O ?? O 128 0 0 0 + 0168 112 1,777 O O O + 15.0 7.8 9.2 11.8 5.1 9.8 8.4 6.3 8.2 5.5 87.1 NOTE: + Respondents commit this crime by definition. 0 Respondents do not commit this crime by definition. ? Respondents may or may not commit this crime; analysis shows that nearly all in this category do. ?? Respondents may or may not commit this crime; analysis shows that most in this category do not. Assault includes homicide arising out of assault or robbery. bTheft includes auto theft, fraud, forgery, and credit card crimes. CThe remaining 12.9 percent did not report committing any of the offense types surveyed. Respondents with missing data (150 out of 2,190) were excluded in calculation of the percentages. SOURCE: Chaiken and Chaiken (1982a:Table 2.5).

418 time already spent in that state increases. Switches to a different state are more likely the shorter the duration in any state. In offense switching, duration dependence is reflected in variations in switching patterns with increases in the length of the intervals between arrests. The preliminary analysis of the role of different intervals (Table 54) suggests that duration dependence may be a factor in offense switching. Contrary to the cumulative inertia observed in studies of mobility, however, repeating the same of- fense type seems to be more likely when intervals between arrests are short. The tendency to specialize appears to decrease for longer intervals. A variety of modeling strategies have been proposed for incorporating duration dependence. These include expanding the state space to include duration explicitly as a defining attribute (Cox and Miller, 1965; McGinnis, 1968), introducing inde- pendent variables into Markov chain mod- els (Coleman, 1964; Spilerman, 1972a), and semi-Markov processes (Ginsberg, 1971~. (See Hoem, 1972, for a review of various models that incorporate duration depen CRIMINAL CAREERS AND CAREER CRIMINALS dence.) These approaches may be fruitfully applied to analyses of offense switching as well. The analytic treatment of offense switch- ing is currently in the earliest stages of development. Only the simplest first-order Markov models have been explored, and then in very limited ways. Analysis in this area may gain substantially from the many developments in modeling already avail- able in other fields, especially the treatment of mobility processes in demography and sociology. Attempts to model offense switching may also benefit from expanding the process to include consideration of the role of intervening, but undetected, of- fenses in the observed switching process between arrests. Such models would char- acterize switching between arrests in terms of the basic switching process between of- fenses committed and the selection process that transforms some offenses into observed arrests. Alternatives to Markov fo,,~ula- tions, with their limited focus on successive events, might also be fruitfully explored to accommodate the role of prior history in future offense seriousness.

Next: Appendix C: Workshop on Research on Criminal Careers: Program and Participants »
Criminal Careers and "Career Criminals,": Volume I Get This Book
×
Buy Paperback | $114.00
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

By focusing attention on individuals rather than on aggregates, this book takes a novel approach to studying criminal behavior. It develops a framework for collecting information about individual criminal careers and their parameters, reviews existing knowledge about criminal career dimensions, presents models of offending patterns, and describes how criminal career information can be used to develop and refine criminal justice policies. In addition, an agenda for future research on criminal careers is presented.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!