National Academies Press: OpenBook

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

Chapter: 1. Issues in the Measurement of Criminal Careers

« Previous: Front Matter
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 1
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 2
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 3
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 4
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 5
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 6
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 7
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 8
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 9
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 10
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 11
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 12
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 13
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 14
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 15
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 16
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 17
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 18
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 19
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 20
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 21
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 22
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 23
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 24
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 25
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 26
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 27
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 28
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 29
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 30
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 31
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 32
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 33
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 34
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 35
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 36
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 37
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 38
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 39
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 40
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 41
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 42
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 43
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 44
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 45
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 46
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 47
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 48
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 49
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 50
Suggested Citation:"1. Issues in the Measurement of Criminal Careers." National Research Council. 1986. Criminal Careers and "Career Criminals,": Volume II. Washington, DC: The National Academies Press. doi: 10.17226/928.
×
Page 51

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.

1 Issues in the Measurement of Criminal Careers Joseph G. Weis Between 1850 and 1900, each decen- nial census in the United States included a question on mortality that simply asked the respondent to report the number of household members who tract died dur- ing the past year. One would assume that a death in one's family would be of suffi- cient significance and salience that it would be reported accurately. However, a comparison of the census estimates of mortality with the number of deaths offi- cially registered showed an "uncleresti- mation of deaths amounting to over 50 percent" (Shryock, Siegel, and Associ- ates, 1975:3921. The survey estimate of mortality was consiclered so flawed and unreliable that the practice was discon- tinued at the turn of the century." About Joseph G. Weis is director of the Center for Law and Justice and associate professor of sociology, University of Washington. The author is grateful to Jeanne Kleyn, Margie Ramsdell, George Bridges, Elizabeth Loftus, Lynn Frandsen, Kris Jones, and James McCann for their suggestions, assistance, and support in preparing this paper. iImproved record-keeping and survey tech- niques have reduced the rate of underreporting, but 1 70 years later questions that asked whether a respondent hac] been the vic- tim of a variety of crimes were acIded to census surveys. The earliest pilot na- tional victimization survey reported that more than 50 percent of victimizations are not reported to the police, which sug- gests that the official records of crime underestimate the volume of crime (En- nis, 1967~. The juxtaposition of Me two underesti- mations highlights two major issues in the measurement of crime- the adequacy of official records as measures of phenom- ena and We apparent discrepancy in esti- mates of the amount and distribution of crime generated by different measure- ment me~ods. The resolution of these issues in Me measurement of crime has not been as easily accomplished as in the there remain substantial differences between sur- vey and official-record estimates of mortality. For example, a one-time 1962 comparison ofthe number of deaths reported in a quarterly census household survey and the number of officially registered deaths shows survey underreports of approximately 21 percent + 7 (Shryock, Siegel, and Associates, 1975:392).

2 measurement of mortality. For example, the substantially Tower estimates of mor- taTity produced by using survey tech- niques were iclentifiecI as the source of discrepancy and the survey was aban- donec3 because of confidence in the rela- tively complete sample of deaths con- tained in official mortality records. But the substantially higher estimates of crime generates! by victimization sur- veys, as well as by self-reports of criminal involvement, have not been attributed solely to the survey measures because of the selective sample of crimes and crim- inals represented (i.e., underrepresentec3) in official crime records. In fact, instead of one measure being considered clearly su- perior to another, the different measures of crime have come to be viewed by many as competing approaches, and each has its champions. Some see the potential for complementary measures of crime, each tapping somewhat different but overIap- ping domains of the phenomenon. To date, this battlegrounct of competing . . . ~. ~ CRIMINAL CAREERS AND CAREER CRIMINALS research, but they are complicatect and exacerbated by temporal change and the apparently unique characteristics of seri- ous, chronic offenders. Unfortunately, only one study has addressed the valiclity and reliability of measures in research on criminal careers (among a sample of adult prisoners) and that compared two meth- ods of measuring criminal behavior, of- ficial records and self-reports (Marquis and Ebener, 19811. There is not a rich literature available on alternative ap- proaches to measuring individual offend- ing patterns or on ways to resolve appar- ent discrepancies among measures in estimates of criminal career parameters or their correlates. The purposes of this paper are, first, to describe briefly the alternative ap- proaches to the measurement of criminal careers, particularly self-reports and off?~- cial records; second, to discuss the actual and potential sources of bias and (listor- tion in the measures of parameters and correlates and their effects on the conver- perspechves has been confined almost gence or discrepancy of estimates; and, exclusively to research on juvenile clelin- third, to propose some research strategies quency, which has been characterized by for improving the measurement of crimi- cross-section designs and local samples nal careers to increase the convergent within which serious offenders, minori- validity among the different measures of ties and the economically clisacivanta~ed individual offending patterns and to im- prove their capacity for reciprocal adjust- ment or calibration. Theory construction, policymaking, and program implementa- tion and evaluation can be better in- formecl by convergent than discrepant sources of information on the basic facts are underrepresented. With the recent interest in the extent to which criminal career research can inform crime control policy, the field of controversy has been broadenecI. The basic facts of criminal careers are either unknown or unclear. The parameters of criminal involvement over time prevalence, indiviclual of- fending rates, patterns of criminal behav- ior, and duration of involvement-and the correlates of those parameters have not been specified, due in part to the paucity of rigorous research in the area and the concomitant unresolved mea- surement issues. All the problems regard- ing the validity and reliability of mea- sures of crime apply to criminal career ALTERNATIVE MEASUREMENT APPROACHES Five basic alternative approaches to the measurement of crime produce data that may be useful, directly or indirectly, in measuring individual offending pat- terns: official crime records, self-reports of criminal behavior, reports of personal

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS victimization, direct observations, and in- formant reports. It is clear that the most viable and useful alternatives for n~easur- ing individual offending patterns are offi- cial records and self-reports. One group of criminal career researchers considers bow as necessary, and self-reports, in particular, as "essential" (Greenwood, 1979:43~4) for measuring the frequency of individual criminal involvement ant! the relationship between offenses com- mittec! ant] offenses that result in criminal justice system processing. Victimization, observation, and infor- mant measures of crime are more indirect in their generation of estimates than the two other measures and have limited ap- plications in criminal career research. In general, they tend not to measure the parameters and correlates of criminal be- havior as comprehensively or directly, and they suffer from having somewhat unique or narrow sample characteristics. Victimization surveys are victim based, rather than offender based, ant! conse- quently provide only indirect estimates of parameters and correlates at the aggre- gate level for a small subset of basically personal crimes and an even smaller sub- set of victimizer (or offencler) characteris- tics (Hinclelang, Gottfredson, and Garo- falo, 19781. Perhaps most important, the fact that the results of victimization sur- veys cannot be compared with other data sources at the inclividual-offender level is the most critical weakness of using vic- timization surveys in criminal career re- search. Direct observation studies, on the other hand, particularly if clone within a prospective longitudinal design and car- ried out for a number of years, are better able to measure indiviclual-leve! parame- ters and correlates than victimization sur- veys. But, unfortunately, they typically use small, selective samples of high-risk juvenile subjects whose behavior is not representative of serious crime happen- ing in natural settings or out "on the 3 street." As for informant reports, the most severe limitation is that informants can only report on others whom they know and observe ~lirectly or hear about indi- rectly. The sample, therefore, is very lim- ite~1 and selective. Official-recorc] data are adequate for es- timates of of Acid prevalence and indivic3- ual offending rates, pattems, and dura- tion. In fact, as a measure of crime mix, crime switching, and duration, official records are probably more useful than self-reports, primarily because time mark- ers are incorporates! in the records, which permits relatively accurate estimation of official onset and termination and of the sequence of changes in offenses of rec- ord. However, official data are not as ad- equate as measures of the correlates ant] determinants of parameters, simply be- cause many of the important pieces of information on the social, demographic, economic, and psychological characteris- tics of offenders are not routinely, system- atically, or accurately collected. This, however, is one of the strengths of self- report surveys, which can collect a great variety of rich etiological and descriptive information along with the reports of per- sonal involvement in criminal behavior. Self-reports are also more adequate mea- sures of prevalence and in(liviclual of- fencling rates than official data but are weakest as a basis for studying patterns of involvement and duration, particularly within the typical cross-section study de- sign. Respondents have rlifficulty in re- membering and dating onset, termination is (by definition almost immeasurable, and respondents have even more cliffi- culty recalling the sequences of a variety of offenses. Official records are necessary to those agencies responsible for crime control; they constitute the standard extemal val- idation criterion to which other measures are compared and are the only continu- ous, lifelong recorc! of criminal involve

4 meet. Self-reports typically cover much shorter time spans than official records, although they provide the potential for long-term measurement of a variety of samples. More importantly, they provide more detailed and richer information on an indiviclual offender's involvement in crime and on other characteristics of de- scriptive and etiological interest. In sum, official and self-report mea- sures best meet the needs of criminal career research. They measure more com- prehensively and systematically the pa- rameters and correlates of individual of- fending patterns over time for serious crimes (e.g., robbery, burglary, assault, drug dealing, auto theft, larceny), can be compared more easily and directly with each other on an individual-off6nder ba- sis, and, therefore, inform crime control policy regarding the serious, chronic of- fender. Taken together, they are at the heart of the controversy over the question of convergence and discrepancy in esti- mates of correlates of crime produced by different measurement approaches. The issue is whether official records and self- reports of criminal behavior generate sim- ilar distributions and correlates of crime. Are the two measures consistent or incon- sistent in their representations ofthe gen- eral characteristics of crime anct crimi- nals? The answers to the questions above are very important for two reasons. First, the implications for theory, policy, and pro- grams will reflect the convergence or dis- crepancy of crime measures~iscrepant correlates suggest different strategies, while convergent correlates support com- plementary strategies and confirm infor- mation on the facts of crime. Second, as the convergence in the estimates of cor- relates approaches unity, one measure can be subshtutec! for the other (Nettler, 1974:8~90~. Of course, this is unlikely, but if there is "apparent" and "sufficient" convergence, the correlates-or charac- teristics of offenders and offenses-could CRIMINAL CAREERS AND CAREER CRIMINALS be taken into account to estimate param- eters of criminal careers, particularly prevalence ant! incliviclual offending rates (Chaiken, Rolph, and Houchens, 1981: 141. That is, the use of one measure to calibrate the other is enhancer! with con vergence. CONVERGENCE OR DISCREPANCY IN ESTIMATES In general, there is surprising consis- tency in the descriptions of the most im- portant correlates of crime based on of- ficial crime records and self-reports of criminal involvement. Indeed, there are fewer discrepancies than one might ex- pect, given the severe criticisms that have been leveled against each method of measurement. In fact, among the major sociodemographic and etiological corre- lates of crime, only one is clearly discrep- ant in its correlation with the two mea- sures race-and one may or may not be discrepant social class. The only other apparently discrepant correlation is for a variable that has not been treated in the- ory or research as a major correlate pulse rate (Farrington, 1983).~ The re- mainder of the major correlates are con ~Farrington (1983:15) reports that convicted youths tend to have low pulse rates, especially if they are violent offenders, but that low pulse rate does not predict high rates of self-reported violent behavior, even though convictions and self-reports of violence are significantly related. A possible ex- planation of this discrepancy could be that con- victed juvenile criminals, particularly those con- victed for serious crimes of violence, are more cool and callous than other respondents, who may not be as hardened by experience with the criminal justice system. They may simply be more likely to deceive and underreport involvement in crimes of violence and, therefore, attenuate the correlation. The situa- tion may be somewhat analogous to the cool, hard- ened criminal "beating" a lie-detector test. And we know from empirical research that those who are most involved in the most serious crimes and who have official records are most likely to give invalid responses and underreport (Hindelang, Hirschi, and Weis, 1981:212-214).

ISSUES IN TTIE MEASUREMENT OF CRIMINAL CAREERS vergent or very similar in their relation to official ant] self-report measures (Hincle- lang, Hirschi, and Weis, 1981~. Only a limited number of studies allow the kinds of direct comparisons of official ant! self-report data on indivicluals in the same sample that are necessary to address adequately the issue of convergence or discrepancy in the correlates of crime. However, among those studies, the esti- mates procluced by official or self-report measures are largely consistent for sex (see Hindelang, Hirschi, and Weis, 1981: 137-155 for an extensive review), age (Farrington, 1983; Langan and Far- rington, 1983:530-531), school perform- ance ant] achievement, intellectual abil- ity and general competence, delinquency of friends or peers, poor family supervi- sion, and other major correlates that have been identified as etiologically important (West and Farrington, 1977; Hinclelang, Hirschi, and Weis, 1981:199-207; Far- rington, 1982~.3 The two major correlates about which there is some controversy, social class ant! race, provide clues about the differential validity and reliability that may be responsible for the discrepant anct, hence, divergent estimates of crimi- nal career parameters. 3For each of the major correlates for which there is no apparent discrepancy in the correlation be- tween itself and official and self-report measures of crime, there may be inconsistencies in correlations across but, typically, not within subgroups of re- spondents. For example, the relation of school per- formance, whether measured by official grade aver- age, achievement test, or self-report of performance, is consistently inverse with official and self-reported crime (see Glucck and Glucck, 1950; Jensen, 1976; Hindelang, Hirschi, and Weis, 1981:19~202). However, for some respondents, specifically black, male official delinquents, even though there is an inverse relation between self-reports of grades and official and self-reported delinquency, both correla- tions are lower than should be expected. They are not discrepant by method of measurement, but the correlations produced by each are lower than ex- pected because of the apparent diminished validity of self-reports, for both grades and delinquent be- havior, among members of this subgroup. S Social Class Most research on the relation between social class and crime does not assess the issue. Rather, it assumes that there is a strong inverse relation between class and "official" delinquency and then compares self-report estimates with this established "thoroughly clocumentec3" fact (see Gor- clon, 1976~. This relationship was es- tablishec! with ecological correlations, which, as is well known, tend to overes- timate inclivi(lual-level correlations sub- stantially the "ecological fallacy" (Rob- inson, 1950; Hannan, 1971~. In addition, most of the studies relied on one measure of crime, either official ciata or self- reports, which precludes a direct compar- ative assessment of the correlations gen- eratec3 by each method of measurement. The finding of no important relation- ship between incliviclual-level measures of class an(1 self-reportecl crime is also a well-establishec1 and thoroughly (locu- mentecl fact, one that has been replicated by many self-report studies (Short and Nye, 1958; Akers, 1964; Gold, 1970; Hinclelang, Hirschi, and Weis, 1981~. Given the (different facts about social class and crime that are generated by official and self-report measures, there are Togi- cally only two ways to bring them into convergence. The first approach implies that, since there is little doubt about the inverse relationship between official crime and class, self-reports are defective and, if improved technically, would] also show an inverse relationship win class (Elliott and Ageton, 1980), hence conver- gence. The second approach implies that, since there is little doubt about the find- ing of no important relationship between self-reporte<1 crime and class, official measures are suspect but, if the problem was corrected, they would also show no relationship with crime (Hinclelang, Hirschi, and Weis, 1979), hence conver- gence. The relevant empirical question, then, is: what is the inclividual-level rela

6 tionship between social class and official and self-reportecT crime? Two meta-analyses of the cIass-crime relationship come to opposite conclu- sions regarding the correlation between class and official crime measures. Tittle, Villemez, and Smith (1978) report no im- portant relationship, claiming it to be a myth in criminology. In a later review Braithwaite (1981) concludes that there is some evidence to support an official crime-cIass correlation. Examining those studies that use inclividual-leve! official data only (Havighurst, 1962; Hathaway and Monachesi, 1963; Polk, Frease, and Richmond, 1974) shows that the relation- ship between social class and official cle- linquency is weakly negative to nonexis- tent (Hindelang, Hirschi, and Weis, 1981: 187-188~. A more recent Tongitudi- nal study of a cohort of 7,719 boys in Stockholm (Janson, 1982) also reports no important indiviclual-level relationship between a "famiTy's social position" and a boy's having a police record. Many, many more self-report studies typically also report no significant rela- tionship between self-reported crime and class (e.g., DentIer and Monroe, 1961; Akers, 1964; Hindelang, 1973; Bachman, O'Malley, and Johnston, 19781. However, there is a major exception to the apparent consensus of self-report studies, ancT that is the recent work of Elliott and Ageton (1980) ant! Elliott and Huizinga (1983~. Baser] on analyses of the National Youth Survey4 probability sample of approxi 4In 1967 the National Institute of Mental Health sponsored the first of a relatively regular series of national youth surveys of a representative sample of juveniles for the purpose of estimating the extent and nature of delinquency and substance abuse in the United States. The survey was repeated in 1972, and in 1976 the National Institute for Juvenile Justice and Delinquency Prevention became a co- sponsor of an annual survey of the self-reported and official delinquency of a national probability panel of youths aged 11-17 in 1976. CRIMINAL CAREERS AND CAREER CRIMINALS mately 1,500 youths, they report that there is a moderate inverse relationship between class and serious "person crimes" and serious "property crimes," particularly for person crimes among black youths of Tow socioeconomic status (Elliott and Ageton, 19801. Later analyses of the national youth panel for 197~1980 show small-to-moclerate class differences in prevalence and incidence for serious crimes, and the relationship is stronger for incidence than prevalence scales (El- liott and Huizinga, 19831. Official data have also been collectecl on the national youth sample, but they have not yet been reported in the literature in a direct as- sessment of discrepancy or convergence. Of the five major studies reviewed by Hinclelang, Hirschi, and Weis (1981: 18~193) that have both official and self- report measures on an inclividual level for the same sample, all showed convergent estimates of the correlation between so- cial class and official self-reported crime (Reiss and Rhodes, 1961; Hirschi, 1969; Williams and Gold, 1972; Wolfgang, Figlio, and Sellin, 1972; Elliott and Voss, 1974~. Only one of the studies showed convergent estimates of a small inverse relationship (Reiss and Rhodes, 1961~. Another stucly, based on the Cambridge (Englancl) Study in Delinquent Develop- ment, found an inverse relationship be- tween social class at age 14 and having an official conviction record at age 17-20, but no relation to self-reportecl crime at age 18 (Farrington, 1979; 1982: 17~.5 This discrepancy is only one among a much broacler pattern of finding no important differences between estimates of a vari 5European, primarily British, studies ofthe class- crime relationship tend to find an inverse relation- ship no matter what the measure, in good part because of the more differentiated, structured, and rigid class structure that exists in Britain, so that this finding of "discrepancy'' is somewhat unusual within the British context (see Belson, 1968; McDonald, 1969; Bytheway and May, 1971).

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS ety of correlates baser] on self-reports and official records. This apparent cliscrep- ancy may be interpretable in light of an important reanalysis of the follow-up in- terview ciata from the Philadelphia cohort study (Wolfgang, Figlio, and Sellin, 19721. Thornberry ant! Farnworth (1982) found no important relationship between status, as measured primarily by ecluca- tional attainment, and criminal involve- ment, whether measured by self-report interviews or police arrest data, when the cohort members were juveniles. How- ever, there was a significant inverse rela- tionship between status and both official and self-reportecI crime when the cohort members were adults. That is, conver- gent estimates of the status-crime rela- tionship are produced for juveniles and adults, even though they are in different directions. The findings for adults are particularly important for criminal career research, because among serious, chronic offenders the relationships between pa- rameters and correlates are apparently different during different stages in career clevelopment. In other words, the opera- tion of social class is not the same at age 12 as it is at age 25. Another major study shows that there is a very weak to nonexistent relationship between social class ant! official and self- reported juvenile crime. More impor- tantly, the study attempts to identify and test the substantive and methodological sources of discrepancy, not only for social class, but for a number of other variables that have some empirical and theoretical importance (Hindelang, Hirschi, and Weis, 1975, 1978, 1979, 1981; Weis, 1983b). Hindelang, Hirschi, and Weis (1975, 1979) propose that the apparent discrepancy between self-report and offi- cial measures for social class, as well as for gender and race, becomes "illusory" when standard but critical methodologi- cal considerations are taken into account, particularly comparisons of results by 7 level of measurement and by the domain (variety), type, and seriousness of crimi- nal behavior. They content! that, if the data are properly analyzed by comparing inclividual-level measures of both official and self-reportect crime with indiviclual 1 ~ r 1 level measures or class, ant! the types and r . . 1 .1 .~ seriousness of crimes in both are the same, the alleged (liscrepancy for social class is resolved.6 Simply put, one should not compare apples with oranges. Elliott ancI Ageton (1980) have recommended basically the same methodological adjust- ments; in acIdition they suggest that the proper analysis of self-report data should include more attention to scoring and scaling of items. Specifically, they sug- gest that the reported frequency of illegal acts should not be restricted in the scor- ing of incliviclual incidence of involve- ment (which is preferred to prevalence) and that the unique properties anct con- tributions of indiviclual crimes that may be related to social class (or other vari- ables) should not be lost in global scales of (lelinquency. Hinclelang, Hirschi, and Weis (1981: 181-198) cTiscovere(1 in a comprehensive empirical test of these hypotheses that there is a weak, insignificant, or nonexis- tent relationship with socioeconomic sta- tus when one controls for level of mea- surement and compares in(lividual-level data on both variables. It seems that no matter how one measures, scales, or scores the clata, there are small, typically negative relations between social class and juvenile (delinquency. The correla- tions (~) range from -.01 to -.08 between occupation of principal wage earner and the self-report delinquency scales (prev- alence scales for total, serious, general, 6The same logic, of course, can be applied to other correlates and has been by Hindelang, Hirschi, and Weis (1981~; only race proved not to be amenable to these methodological discrepancy-res- olution strategies and procedures.

8 drug, famiTy-schoo! offense inclexes) and the total official offenses index (Hincle- lang, Hirschi, anct Weis, 1981:1961. These finclings also hold when the additional suggestions of Elliott and Ageton (1980) are incorporated in the analysis: that is, one still finds consistently weak-to-non- existent relations that at times are in the wrong direction when one (1) uses their self-report scales scored for incidence or the total unrestricted frequency within the past year; (2) examines each of the three official delinquency inclexes (police contacts ever, police contacts nest vear. juvenile court referral ever); (3) does an item-by-item analysis of their 69 self- report items; and (4) examines each of their (1,2,3) relationships separately with a variety of indicators of socioeconomic status, including father's education, fa- ther's occupation, mother's education, mother's occupation, father employed, mother employed, father's socioeconomic index, and mother's socioeconomic index (Weis, 1981, 19861. These results are perplexing because some rigorous studies report a moderate inverse relation between social class and self-reportec3 crime (e.g., Elliott and Ageton, 1980; Elliott and Huizinga, 1983), while other rigorous studies show an absent-to-weak inverse relation for both self-report and official measures (e.g., Hindelang, Hirschi, anc1Weis, 1981; Thornberry and Farnworth, 19821. Given that there are no published or otherwise available results on a convergent inverse relationship, except for Reiss and Rhodes (1961), the conclusion for now is that there is much more evidence, across a variety of some ofthe most important data sets compiled on juvenile crime (e.g., National Youth Survey: GoIc3, 1970, and Williams and GoIc3, 1972; Richmond youth project: Hirschi, 1969; Philadel- phia birth cohort study: Thornberry and Famworth, 1982; Seattle youth stiffly: Hindelang, Hirschi, and Weis, 1981) that convergence is more likely to show that CRIMINAL CAREERS AND CAREER CRIMINALS there is a weak negative-to-nonexistent relation between social class ant! both official and self-report measures of crime. However, what is disturbing is that the agrees] stanciarcI methoclological adjust- ments that shouIcT be applier! to these data comparisons (same level of measure- ment, same domain of criminal acts, more sensitive scoring and scaling proceclures) do not go far enough in creating some kind of apparent consensus on the issue of convergent vaTiclity. Race Race is the only major correlate (Wolfgang and Ferracuti, 1967; Wolfgang, Figlio, and Sellin, 1972; Curtis, 1974; Hinclelang, 1978; Blumstein, 1982) for which the metho(lological adjustments proposer! by Hindelang, Hirschi, and Weis (1979), particularly the controls for type ant! seriousness of offense (espe- cially violence), JO not make the official and self-report relations consistent in the way they JO for sex, social class, and other important correlates of crime. Official data historically have implied that blacks are "more criminal" than whites simnIv by virtue of their overrepresentation in the criminal justice system and official records. Self-reports, on the other hand, suggest that there is less difference in both prevalence and incidence of crime (e.g., GoIcT, 1970; Chaiken and Chaiken, 19821. The discrepancy is so large that "the range of ratios in self-report samples restricted to a single sex ... does not overlap with the range of ratios fount! in official data.... The very strong relation between race and delinquency in official data is not present in self-report clata. In fact, the self-report relation wouIcl have to be characterized as weak or very weak, with a mean black-to-white ratio of less than 1.1:1 expected on the basis of previ- ous research" (Hindelang, Hirschi, and Weis, 1981:159; emphasis in original). Contrary to expectations that compar - 1- ~

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS ing the more serious and violent crimes would show race differences of a magni- tude similar to those represented! in of- ficial records, the race differences for both self-report serious-clelinquency scales and individual items are not very large or significant. Hindelang, Hirschi, and Weis (1981) report white-to-black ratios that approximate 1 for the total prevalence scale; paracloxically, for the frequency or incidence scales the ratios are less than 1 on five of their six scales, which implies more white than black involvement in crime. These ratios approach unity as one moves from the less to more serious of- fenses but do not, as the official-data ra- tios would suggest, exceed 1. The overall 2:1 black-to-white ratio in the Seattle study's offlcial-record data is approxi- matec] by prevalence ratios for self- reports of some types of face-to-face vio- lence, for example, carrying a weapon, hitting a teacher, using a club, gun, or other weapon to get something, beating someone seriously enough to send them to the hospital, robbery, and jumping and beating someone, as well as for official reaction, probation, and school suspen- sion (Hinclelang, Hirschi, and Weis, 1981:16~1691. One might interpret this to mean that there may be violent-offense specialization by race, but the frequency estimates do not support this, for either scales or individual items; it seems that either the criminal justice system clis- criminates substantially against blacks, particularly males, or that blacks may be underreporting the extent of their in- volvement in crime. Other research on juveniles and, more surprisingly, on adult prisoners supports these finclings regarding apparent uncler- reporting by blacks, particularly mates with official criminal records. Shannon (1982, 1983) reports that in the Racine, Wisconsin, cohort study of the relation- ship between juvenile and aclult criminal careers, self-reports of delinquency hac! little relationship to race, but there was a 9 disproportionate representation of blacks with official records of police contact. Black males hacI the most contacts and most extensive, serious involvement with the police 75 percent of the black males had a police contact while a juvenile, compared with 50 percent of the white males. It appears that juvenile black mates may be underreporting their crim- inal involvement because "while 80 per- cent of the whites in each cohort either reported their number of police contacts consistently with police records or over- estimatec3 them, only half the blacks re- ported the number of contacts consis- tently with the records and the other half reported fewer contacts in comparison with official recorcls" (Shannon, 1982:1, 101. Other studies on juvenile crime also suggest underreporting and potential va- li(lity problems among black respon- lents, particularly mates with official rec- or(ls (Chambliss and Nagasawa, 1969; GouIcl, 1969; Hirschi, 1969~.7 Intriguing corroborative finclings are also reported on mate adult prisoners, including serious, chronic offenders, in the first (Peterson, Braiker, and Polich, 1981) and second (Marquis and Ebener, 1981; Chaiken and Chaiken, 1982) Rand surveys of incarcerated felons. Peterson, Braiker, and Polich (1981:xxii-xxiii, 6(k 65) report that the results of an anony- mous survey of Califomia prisoners gen- eratecl "complex" results on race. The self-reports showed that black inmates were "less active and dangerous crimi- nals" than white inmates they were in- volvec3 in the smallest variety of crimes and at the lowest reported rates compared with whites and Chicanos. They also re- ported less involvement in crimes of vio- lence. Even when corrections are made fin a personal communication, Delbert Elliott indicates that similar underreporting and potential validity problems among this subgroup were also appearing in ongoing analyses ofthe National Youth Survey data.

10 for different apparent probabilities of ar- rest and estimated offense rates by race (the mocle] of street offenders developed by Chaiken), differences remain, with the exception that blacks are shown to be more involves! in crime than Chicanos. The interpretation ofthese differences by race in self-reports and official records, particularly the apparent underreporting by black prisoners, is considered "some- what ambiguous" ant! two possibilities are offered either blacks do, incleecI, commit fewer crimes and less frequently or the criminal justice system is more likely to imprison blacks, particularly Tow-rate, occasional criminals. A third source of the differences is only alluded to in a footnote-the clifferential reliabil- ity and validity of self-report responses by race. Unfortunately, the first Ranc3 inmate survey was anonymous, so the self- reports and official records could not be validated reciprocally at the incliviclual level (Peterson, Braiker, and Polich, 1981:2~26, 901. However, Chaiken and Chaiken (1982:226, 244, 245, 247) and Chaiken, Chaiken, and Peterson preaches to measuring;indiviclual offencT- (1982:17) suggest that there may have been underreporting among some sub- groups of prisoners in the second Rand inmate survey, particularly among blacks and the less educated.8 For example, there is the "puzzling finding" that blacks report much lower rates of assault than whites; but it is also reported that the responses of blacks are less reliable, i.e., have "substantially worse internal qual- ity," and that "respondents whose survey responses have poor internal quality tend estimates of parameters and correlates anal, ultimately' the convergence of the CRIMINAL CAREERS AND CAREER CRIMINALS to report Tow crime commission rates for the crimes they report committing." There seems to be consistency in both inmate surveys regarding the under- reporting of some subgroups, particularly black inmates, reacting to lower-than- expected individual offending rates when self-reports are compared with official records. In short, a number of studies of juve- nile and aclult offenders report a discrep- ancy between race and self-reportect and official crime data, and the discrepancy is likely explainecl by underreporting on the self-report instrument rather than by race discrimination within the criminal justice system. Therefore, attention neecls to be turned to the connection between self-reports and official records and possi- ble racial and other differences in the link between the two. In this regard the next section addresses two questions: What is the nature ofthe reliability and validity of the two measures? How c30 they affect 8It is interpreted here as a "suggestion" because the data on individual offending rates by race could not be found in Chaiken and Chaiken (1982~. Rela- tions between race and other variables, including "intemal quality" and "external reliability" of mea- sures, are reported extensively, but the distribution of self-reported individual offending by offense and race could not be found. , , v estimates generated by the two ap ing patterns? EFFECTS OF RELIABILITY AND VALIDITY ON ESTIMATES Given the apparent underreporting of criminal involvement in self-report sur- veys that was iclentifie(1 in a number of studies ant] its possible contribution to iscrepant correlations, the focus here is on the effects of the differential reliability and validity of measurement approaches on estimates. Of particular interest are "response effects" (Suciman an(1 Brad- burn, 1974) in self-report surveys as pos- sible sources of bias.9 A brief discussion 9Given the focus of most of the research in this area and space limitations, other sources of error and bias, such as processing, sampling, and design, are

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS of the current state of the reliability and validity of self-reports of crime precedes a more detailed treatment of some of the more important sources of apparent and potential distortion and response bias in self-report measures. Reliability of Self-Reports of Crime Although Reiss (1975) is correct that few self-report researchers have assessed the reliability of instruments, the evi- dence shows that the self-report method is very reliable, ant] that if there is a problem with this type of measure it is more likely in the area of validity (Hinde- lang, Hirschi, and Weis, 1981:841. In terms of intemal consistency, both self- reports of juvenile crime (Hinclelang, Hirschi, and Weis, 1981:81) and of crime among adult prisoners (Marquis anct Ebener, 1981; Chaiken and Chaiken, 1982) show "impressive reliability," al- though there is some evidence of system- atic variation in intemal consistency by respondent subgroups among aclult pris- oners. For example, inmates convicted of drug dealing "left significantly more questions blank," and inmates who saw themselves as a "thief, player, or aTcohol- ic/cirunk" clid significantly worse on the intemal consistency of responses. In ac3- clition, even though there clicI not seem to be much difference in extemal reliability (validity) among whites, blacks, or His- panics (although it was still lower, on the average, than intemal consistency), the "summary measure ('percent of bad inter- nal quality indicators') was worse than average for black repondents" (Chaiken and Chaiken, 1982:226, 244~. Test-retest reliability of self-report data not addressed directly but only as they relate to response effects. The limitations of official-record measures are discussed later; besides, the issue for official measures is not so much a matter of reliabil- ity and validity as it is of utility. 11 is also quite good, both for reports of criminal behavior anct for independent variables. A small number of studies on juvenile clelinquency have reported high test-retest reliability coefficients; the time between administration of the self-report instrument ranged from a couple of hours (Hinclelang, Hirschi, and Weis, 1981: 81-82) to a number of years (Farrington, 1973; Bachman, O'Malley, and Johnston, 19781. These and other studies (DentTer and Monroe, 1961; Clark and Tifft, 1966; Belson, 1968; Kulik, Stein, and Sarbin, 1968) show that there is substantial con- sistency in self-reports of the same crim- inal acts, but reliability diminishes with time. In general, the test-retest reliability coefficients are higher for criminal behav- ior items than for other items (Elliott and Voss, 1974; Bachman, O'Malley, and Johnston, 19781. For example, with a few hours between tests, the coefficients on four independent items (mother's educa- tion, grade-point average, parental super- vision, and peer delinquency) ranged from 0.4 to 0.9, but the delinquency items varied around 0.9 (Hinclelang, Hirschi, and Weis, 1981:821. The reliability of alternate forms, using split-half coefficients, is also impressive. Elmhorn (1965) reports a coefficient of.86 for 21 delinquency items, and Kulik, Stein, and Sarbin (1968) report coeffi- cients ranging from .70 to .92 for five subscaTes clerivec3 from cluster analysis of 52 delinquency items. Overall, self-reports are very consistent in the measurement of crime, in terms of internal consistency, test-retest stability, and split-half reliability. There is some eviclence, however, that there may be potentially important variation in reliabil- ity, particularly internal consistency, by respondent characteristics, with appar- ently more reliability problems among both black juveniles and adult offenders, the less educated, and those with drug and alcohol abuse in their records. These

12 respondent effects on reliability may be related to underreporting of individual criminal involvement, the potential clis- tortion of parameter estimates, and the attenuation of correlations with these re- spondent characteristics. Threats to vaTicI- ity seem more likely sources of serious measurement problems, particularly a- mong some subgroups of respondents. Validity of Self-Reports of Crime There are many types of validity; for example, Costner (1975) has identified eight, ranging from content to face valid- ity.~° The focus here is on those types of validity that allow empirical assessment of the accuracy of responses. Therefore, the discussion emphasizes different kinds of concurrent validity or the extent to which self-report results are consistent with other concurrent measures of crimi- nal behavior. "Convergent" validity-the extent to which two or more measures intended to measure the same concept (e.g., crime) produce convergent results (Campbell and Fiske, 1959)-is depen- dent on concurrent validity. Two types of concurrent validity have been examined in the self-report litera- ture: known group and correlational. In known-group validity, groups known to differ on some dimension, for example, official delinquency, are compared using another measure that is supposed to ctis- criminate between the groups, for exam- ple, self-reports of delinquency. The lat- ter is validated to the extent that it shows that known official clelinquents differ on self-reports of crime. For example, the original Short and Nye (195X) self-report study includecl known-group compari- sons that showed that their 23-item, self- report checklist couIct distinguish rather 1',The eight types of validity are convergent, discriminant, content, face, predictive, concurrent, construct, and factorial (Costner, 1975:2). CRIMINAL CAREERS AND CAREER CRIMINALS well between boys in state training schools and in high school. These known- group differences in self-reports of crime (Erickson and Empey, 1963; Hindelang, Hirschi and Weis 1981:9~96) are evi- dence of validity, but they do not tell us anything about the magnitude of the rela- tionship between self-report ant! official measures in more general samples. The degree of validity is better assessed through correlational validation. The question becomes: To what extent are the results of one measurement approach consistent with or correlated with those produced by another? The stronger the correlation, the greater the concurrent and convergent validity. There are a number of measures with which self-reports of criminal behavior can be compared (Hindelang, Hirschi, and Weis, 1981: Figure 5.1, 98~: (1) unof- ficial sources, such as direct observation or the informant reports of peers, teach- ers, family, and victims; (2) self-reports of official action, such as police contacts, arrests, probation, or incarceration (e.g., Christie, Andenaes, and Skirbekk, 1965; Elmhorn, 1965; Hindelang, Hirschi, and Weis, 1981; Marquis ant] Ebener, 1981; Chaiken and Chaiken, 1982~; (3) official criminal justice records, such as offenses known to the police, court referrals, con- victions, and incarceration (Gould, 1969; Erickson, 1972; Elliott and Voss, 1974; Hin(lelang, Hirschi, and Weis, 1981; Chaiken and Chaiken, 1982~; (4) official records from other agencies, such as school, mental health, social service, and drug abuse (Weis, fanvier, and Hawkins, 1982~. Researchers have relied on official measures as the "stan(larcI" external vali- d~ation criteria, because there is substan- tial evidence and consensus regarding the identification of delinquents and criminals through the official sampling processes of the criminal justice system. Admissions of prior known records are

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS particularly important for assessing the degree of reporting accuracy, typically underreporting, and are necessary for at least inferring some degree of accuracy for self-reports of criminal behavior for which there is no record. That is, the apparent vaTiclity of those respondents with records who admit to them is extrap- olatec] to their self-reports of criminal be- havior, as well as to the self-reports of criminal behavior of those respondents who do not have criminal records. It is typically assumed that the validity of self- reports of criminal recorc] can be extrapo- lated to self-reports of criminal behavior, but it is not clear that the degree of vaTiclity can be transferred from one type oftask characteristic (discussed below) or respondent to another. For those respon- dents who c30 not have official records of crime, other validation criteria become more critical. This critical assumption about the va- lidity of self-reports is relater! to other assumptions that are often macle by re- searchers. One assumption is that the "magnitude" of underreporting of official record can be used as a correction factor, typically involving ~lownward acljust- ments to self-reports of criminal behavior. A second assumption is that the fre- quency of self-reported crime and the validity of admissions are related the higher the self-report estimates, the higher the vaTiclity estimates. In other words, the more crime a respondent ad- mits, the more accurate the responses. ThircI, it is assumed that the validity of measures of crime hoists for measures of other variables, particularly those that are more etiological, attitudinal, or descrip- tive of other respondent characteristics. Fourth, it is assumed that vaTiclity is more likely universal in application than differ- entiated by response effects, whether task, respondent, or interviewer charac- teristics. There is evidence from correla- tional vaTiclation and reverse-record 13 check research that these assumptions are questionable and, more likely, not true. Of course, one of the implications is that extrapolation whether of validity, cor- rection factors, or estimates from one measure to another is more problematic than perhaps anticipated, at least for now. The relationships of "self-reports of criminal behavior" and "self-reports of official record" each to "official record" are the most important vaTiciation correla- tions to assess. The relationship between self-reports of criminal behavior and hav- ing an official record is perplexing. In general, the research evidence suggests that there is a relatively high degree of correlational validity when comparing these two measures, with correlations ranging from .3 to .6 (Hirsch), 1969; Erickson, 1972; Farrington, 1973; Elliott and Voss, 19741. However, one ofthe few studies to report validity coefficients by subgroup suggests that the validity of self-reports of criminal behavior varies by race, with black (.09) en cl Asian (.07) ju- veniles having Tower validity coefficients than white juveniles (.34) (Gould, 1969~. Hindelang, Hirschi, and Weis ~ 1981: 100) also discovered that validity coefficients "vary systematically by race," with black males having much lower valiclity coeffi- cients (.33 to .42) than white males (.65 to .66) when comparing scores on the total prevalence scale and serious-crimes prev- alence scale with three official clelin- quency measures. This diminished valicl- ity translates directly into lower estimates of delinquent involvement among these respondents and the attenuation of the correlation between self-reportec3 clelin- quency and race. The relation between self-reports of official records and actual official records is probably the most critical comparison, because it is here that the accuracy of response is put to its most direct and severest test. Even if one disagrees on what an official record "actually" repre

14 sents, respondents should have direct ex- perience and knowledge of their contact with the criminal justice system and sim- ply need to remember and report itch In general, the validity coefficients are high, in the .8 range all "juvenile studies show that most of those with official rec- ords can be identified through self- reports" of records (Hindelang, Elirschi, and Weis, 1981:1041. In fact, the correla- tions (gammas) between the three official crime indexes and three self-report indi- cators of official action range between .6 and .8; the highest average validity coef- ficients are for self-reports of court refer- ral (.8) and the lowest for police contact it In a personal communication, Delbert Elliott has indicated that some respondents in the National Youth Survey who denied they committed crimes of record claimed that the police were simply harass- ing them and the records represented these "ilium raps." There are rare examples of arrested or con- victed "innocents," but it is clear that those with records have very likely committed a crime, in some cases many, many more than result in official action, whether juvenile (Erickson and Empey, 1963) or adult.(Peterson, Braiker, and Polich, 1981:55-65) offenders. And the great majority of official offenders do not deny their guilt but readily confess to their crime. Wolfgang, Figlio, and Sellin (1972) report that 90 percent of the juveniles who had police contacts confessed to the police on apprehension. It is possible that once in a while the criminal justice system errs and "gets the wrong person" (Nloss, 1963), that an offender may report harassment by local law enforcement officials that leads to discrim- inatory and empty arrests, and that criminal justice system records are incomplete and, therefore, claimed to be inadequate as a standard external validation criterion. However, the possibilities-or facts, if you wish-do not negate the fact that crim- inal justice system record data are as good as any standard external validation criterion in the social sciences, probably better than most. Compared with the other alternatives, it is clearly the best, and one needs to remember that the validation of two mea- sures is reciprocal by definition, so if concurrent, convergent validity is demonstrated for self-reports and arrests, both are validated and by the "relation- ship" between the two. And, finally, there is no such thing as a "perfect" external validation criterion to which the relative validity of another measure can be established. CRIMINAL CAREERS AND CAREER CRIMINALS (.6) (Hindelang, Hirschi, and Weis, 1981:10~1051. The Tower validity for the police measure is also evident in research among adult offenders. Marquis and Ebener (1981:viii, 1~14) report that the accuracy of reports of arrest is "uncertain" in comparison with self-reports of convic- tions among a prisoner sample, and they suggest that this is attributable to appar- ent face-validity problems because arrest can mean several things on the rap sheets. This same kind of face-validity effect could be responsible for the Tower vaTid- ity coefficient for the item in the Seattle youth study that asked respondents whether they had been "picked up by the 1- '' pollee. Overall, the correlational evidence supports relatively high validity for both self-reports and indicators of official con- tact. However, the sources of imperfect correlation remain unclear. Reverse Record Checks The use of record checks facilitates a more rigorous investigation and assess- ment of the sources of diminished vaTid- ity, as well as of the magnitude of the problem. Having prior knowledge of a respondent's official record allows the re- searcher to assess directly the accuracy of responses regarding the existence of the record, as well as to compare self-re- ported crimes with official offenses of record. Reverse record checks, whereby respondents in a study are sampled using 12`'Reverse'' record check is used here by con- vention to indicate only that information in records is checked with information reported by a respon- dent. It does not necessarily mean that a sample was selected from a population of potential subjects who have records. Marquis (1980) and Marquis and Ebener (1981) have pointed out that this kind of sampling design, called a "partial" design, can only account for underreports, simply because there is no respondent without a record who could report hav- ing one.

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS the existence of record as a sampling criterion, have been used in health sur- veys (e.g., Cannell ant! Fowler, 1963; Marquis, 1978, 1980), victimization sur- veys (Law Enforcement Assistance Ac3- ministration, 1972), crime surveys (Erick- son and Empey, 1963; Blackmore, 1974; Hardt ant] Peterson-Harcit, 1977; Bridges, 1981; Hindelang, Hirschi, and Weis, 1981), studies of drinking and alcoholics (Sobell and Sobell, 1978), and research on drug addicts (Ball, 1967~. In general, the majority of respondents with official rec- ords will report having them; in one in- stance 100 percent of a group of juveniles who hac! court records acimittec3 the fact (Erickson and Empey, 1963:459~. There is not a random distribution of the underreporters but, rather, apparent systematic bias in the accuracy of re- sponses to queries about one's official record. In the delinquency literature, the rate of underreporting seems to vary be- tween 5 and 60 percent of the samples (Gibson, Morrison, and West, 1970; Harcit ant! Peterson-Harcit, 1977; Hinclelang, Hirschi, and Weis, 1981), depending on the official criterion and seriousness of the crime. What is perhaps the only data set on crime that has a record check built into the study clesign that allows sex ant! race comparisons shows no sex clifference but an apparent race difference in the underreporting of official contact, which averages about 10 percent for white males and 30 percent for black males (Hindelang, Hirschi, ancI Weis, 1981:122-124~. For the serious crime in- clex (burglary, robbery, vehicle theft, per- son crime, weapon offense), 20 percent of the white males but 57 percent of the black mates underreported official con- tacts. The difference is even greater for some of the individual serious offenses, for example, 67 percent nonreporting for burglary and 62 percent for weapon of- fenses among black males, compared with 18 ant! 22 percent, respectively, for 15 white males (Hindelang, Hirschi, and Weis, 1981:172~.~3 It is clear that black male respondents are much less likely to report crimes that fit the offense category of their crimeks) of record. Among general aclult samples, as well as different kinds of offender subgroups, the nature of the response bias is more ambiguous. Bridges (1983:571), for exam- ple, reports that among the follow-up in- terview sample of the Philadelphia co- hort stucly, the "respondents uniformly overstated the number of offenses result- ing in arrest, ages at the time of offenses, the seriousness of offense, and the num- ber of others arrested." But Sobell and Sobell (1978) find that alcoholics and ex- aTcoholics tend to underreport their fel- ony arrests. Finally, separate analyses of the Rand inmate surveys (Marquis and Ebener, 1981; Chaiken ant! Chaiken, 1982) indicate that there is either no re- sponse bias or, in some cases, overreport- ing of arrests and convictions. One of the limitations that these stud- ies of aclults share with some of the stud- ies of juvenile response bias (e.g., Erickson and Empey, 1963) is that the i3Contrary to what Elliott (1982) has suggested, the differences among the four sex-race groups do not show that white males are atypical and that black males, white females, and black females are in line with other estimates of nonreporting. If one looks at the total number of offenses and the per- serious-crime index category, one discovers that among black females there are only 22 serious offenses, that in four of the five offense categories there are only two offenses each, and that the three 100 percent nonreports fall within those categories, all of which contribute to an artificially high and undependable estimate of the nonreporting rate for black females. For white females the situation is the same-with a total of only 16 offenses, one category with no offenses and another with one offense, the percentages are misleading as a consequence. In short, for the serious crimes in particular, the ones of most concern in the Seattle youth study and here, the data for female offenders do not provide a sufficient basis for concluding much at all about the effects of their nonreporting on validity.

16 respondent knows that the researcher has had access to his official records and, therefore, can verify or clisconfirm re- sponses. This undoubtedly contributes to greater agreement and, perhaps, over- reporting. Another problem is the very Tong reference period for responses, e.g., adults may be asked to report events that occurred during their juvenile years, per- haps more than 10 years earlier. In the only "double blind" study of self-reports of official record, i.e., the questionnaire administrators did not know if the respon- clent had a record and the respondent did not know that official criminal status was used as a sampling criterion anct was, therefore, known to project management, there was variable and systematic under- reporting (Hindelang, Hirschi, and Weis, 1981). In short, there is evidence from re- search on both adults and juveniles that there is important variation by subgroups in the accuracy of responses regarding official records. There seems to be differ- ential validity by respondent characteris- tics, specifically and perhaps most impor- tantly by race. In the next section the types of apparent or potential respondent effects are examined in more cletail, par- ticularly as they and other task and inter- viewer sources of response effects may compromise validity ant] reliability and, therefore, affect estimates of prevalence, incidence, and correlates. SOURCES OF RESPONSE EFFECTS ON VALIDITY AND RELIABILITY The somewhat ambiguous and some- times conflicting findings regarding re- sponse patterns to questions about having official contact with the criminal justice system, as noted, suggest some kind of systematic response bias. If this is so, there is the possibility that diminishes! validity may also extend to respondents' self-reports of criminal behavior. There CRIMINAL CAREERS AND CAREER CRIMINALS may be generalized systematic error in the reports of certain subgroups of re- spondents for the whole range of crime, particularly for the most serious offenses. Obviously, this could distort self-report scores for the very offenses that might resolve discrepancy between self-report and official measures the more serious crimes, which are the concern ofthe crim- inal justice system. There are many real ant! potential distorting elects on responses to sur- vey questions- "response effects" that could be related to the validity and reli- ability of self-reports of crime and, there- fore, to estimates of criminal career pa- rameters. However, the evidence from general survey research indicates that some kinds of response effects may be more important than others, particularly . it.. . . ,, `` . . . .. when threatening, anxiety arousing, or"dangerous" questions are asked. Ad- ditionally, some response effects may be more important than others to validity and reliability in research on serious crime ant! chronic offenders, at least ac- corcling to the scant direct research evi- clence on these types of behaviors and respondents. Consequently, a selective subset of response effects and their appar- ent or potential impact on self-reports of individual offending patterns are the fo- cus of discussion in this section. There are three major sources of re- sponse effects on the accuracy and consis- tency of answers to survey questions- interviewer, task, and respondent charac- teristics (Suclman and Bradbum, 1974; Bra(lbum and Su~man, 1979; Dijkstra and van der Zouwen, 1982; Bradbum and Danis, 1984~. Interviewer characteristics are either"role indepen~lent" attributes, such as gender, race, age, and personal biography, or"role restricted," such as interviewing style, competence, experi- ence, and interviewer's expectations of respondents or responses (see Dijkstra and van der Zouwen, 1982:1(}111. Task

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS characteristics describe the survey situa- tion and can be divided roughly into "structural" characteristics, such as method of administration ant! perceives! anonymity, and "question" characteris- tics, such as worcling, length, content, specificity, response categories, threat, salience, and social desirability. Respon- dent characteristics are the sociodemo- graphic or personal attributes of respon- dents that affect validity or reliability directly or indirectly through interactions with task and interviewer characteris tics. i4 The very scarce research on response effects in the measurement of crime ac3- dresses some ofthese sources of response bias and shows substantial consistency with findings of survey-methodology re- search in other substantive areas. There are also some apparent differences, espe- cially regarding the effect of respondent characteristics. Interviewer Characteristics It is evident that interviewer character- istics play a minor role in distorting an- swers to survey questions. Not only cloes the general literature reach this conclu- sion (e.g., SuUman and Bradburn, 1974; . 14There are also four major types of error that are related to response effects (Sudman and Bradburn, 1982:19), particularly task and respondent charac- teristics and their interactions. The first is memory error, the unintentional forgetting of what happened and when it happened, the consequences being underreporting and telescoping of events, typically forward into the more recent past. The second is motivated error, the intentional underreporting, overreporting, or distortion of responses for positive (e.g., social approval) or negative (e.g., fear of con- sequences) reasons or motives. The third is com- munication error, i.e., a respondent does not un- derstand what is being asked and responds idiosyncratically. The fourth is knowledge error, i.e., a respondent may not know the correct or appropriate answer to a question but responds any- way; typically, there is no way for the researcher to measure the lack of knowledge. 17 Brenner, 1982; Hagenaars and Heinen, 1982; Bradburn ant! Danis, 1984), but also the one study in criminology that can assess interviewer effects within a quasi- experimental design (Hinclelang, Hir- schi, and Weis, 1981) comes to the same conclusion. The design of that study faciT- itates a rigorous assessment of interview- er-respondent interaction effects because white mate, white female, black mate, and black female interviewers were ran- domly assigned to test conclitions and respondents. Therefore, the "role-incle- pendent" characteristics of the race and gender of the interviewers and the same sociodemographic characteristics of the respondents couIcI be examined for their interactive effects on the accuracy of re- sponses to questionnaires and in inter- views, particularly face-to-face inter- views, for which one would hypothesize the largest interviewer effect. Preliminary data analyses suggest that role-indepen- dent interviewer effects cannot account for the apparent differential validity asso- ciated with black, mate, official delin- quents because there are no differences in the magnitude or validity of the crime estimates of respondents by interviewer characteristics (Weis, 1983b). Task Characteristics Researchers have examined a number of task characteristics that can generate response effects that may have potential implications for research on criminal ca- reers. Structural task characteristics and the characteristics of survey questions are examined below. Structural Characteristics Two structural task characteristics, method of administration and anonymity of responclent, have been assessed cli- rectly for their effects on the validity and reliability of self-reports of juvenile clelin

18 quency. Because item threat is so closely relater! to anonymity, even though it is a question characteristic, it is treater! here as well. It has Tong been assumed that inter- views are better than questionnaires in producing accurate responses an(l, there- fore, are in part responsible for a reported difference in the relation between social class and self-report interview responses (inverse) and questionnaire responses (nonexistent) (Gold, 19661. It has also been assumed Hat since questions on one's criminal behavior are threatening, anonymity enhances the vaTiclity of re- sponses (see Goocle and Hatt, 1952; Can- nell and Kahn, 1968; Colombotus, 1969; Lin, 1976~. However, the Seattle youth stucly design allows a rigorous examina- tion of the joint effects of method of ad- ministration, anonymity of respondent, ant! threat of item. The results of such an examination show that the effects on both reliability and vaTiclity of responses are negligible, which corroborates results from the more general survey-methodol- ogy research (e.g., Bradburn and Su(lman, 1979:1-13; Fox and Inazu, 1980~. Examining the effects on reliability first, it is clear that method of administra- tion floes not affect test-retest consistency in estimates or the internal consistency of responses; nor is there any important variation in any of the subgroups in the sample (Hindelang, Hirschi, and Weis, 1981:11~1201. Regarding vaTiclity, not one of the six methods of administration that were compared had a significant clis- torting effect on correlational validity co- efficients. There were some minor but unimportant differences by sex and race. For example, questionnaires hac3 higher validity coefficients than interviews for mates, and the responses of females un- der anonymous conclitions may have been slightly more valid than those under nonanonymous circumstances. The non- anonymous interview was the least valid CRIMINAL CAREERS AND CAREER CRIMINALS of the four conditions among white mates, and black males did worst on the anony- mous interview and nonanonymous ques- tionnaire (Hindelang, Hirschi, and Weis, 1981:1201. Again, these slight and con- flicting (differences do not portent] any systematic or consistent effects of method of administration on correlational validity among any of the respondent subgroups. The possible effects of method of act- ministration on the results of reverse recorc! checks, particularly on under- reporting of criminal record, are a more rigorous test of methods effects.~5 The overall rates of underreporting discussed earlier were basically replicated within each method of administration. Among females the rate of underreporting (22 percent) was identical for each of the two conclitions to which they were assigned, the nonanonymous interview and the anonymous questionnaire. Among mates the results for the four conditions to which they were assigned are more inter- esting. White mates did more under- reporting of official offenses when they were being interviewed (12 percent) or filling out questionnaires (9 percent) un- cler nonanonymous circumstances. Nota- bly, black rnaTes, who had an uncler- reporting rate that was about three times the rate for white males, did the least underreporting (22 percent) when they were interviewed under nonanonymous circumstances; their underreporting in the other three methods of administration ranger! from 34 to 39 percent (Hindelang, Hirschi, and Weis, 1981:122-1231. It i5The reader is reminded that both the official offenses of record and the self-reported crimes were grouped because some specific types of crimes were rare, particularly in the official records, as well as being vaguely described and represented in the records of the police department and juvenile court. Therefore, crime-by-offense comparisons could not be made with enough rigor and confidence to justify that type of analysis (see Hindelang, Hirschi, and Weis, 1981~.

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS seems that the standard "face-to-face in- terview" might be more effective for these types of juvenile offenders. Criminologists have also used and ex- amined the effects of other, less standard methods of administration, including the randomized response method and the lie detector model. The randomized re- sponse method has been used on adult offenders (Tracy and Fox, 1981) and juve- niles (Hindelang, Hirschi, and Weis, 19811. In the Seattle youth study, a ver- sion of the randomized response method (Warner, 1965) was used with each re- spondent after he or she had completed the survey under the main method of administration. The results show that the randomized response method is "no more efficient than other traditional techniques of measurement" in estimating the prev- alence of juvenile crime as measurer] by a subset of 8 offenses from the full set of 69 self-reported delinquency items (Weis and van Alstyne, 19791. That is, the prev- alence estimates did not vary by method of collecting data, even for one of the most anonymous survey methods that has been developed. In fact, when the ran- domized response method followed the anonymous interviews, there was "dou- ble anonymity" (i.e., of both the respon- dent and responses), and even this extra degree of anonymity made no difference in prevalence estimates. Apparently, the "threat" of questions about criminal activ- ity did not.matter that much to the juve- nile respondents in this sample or it mat- terect to the same extent across a variety of anonymous and nonanonymous methods of administration. A final method of administration, a vari- ation ofthe "lie detector" (Clark and Tifft, resee Tracy arid Fox (1981) for a randomized response procedure for estimating incidence in a sample. With refinement, this version of the tech- nique could prove to be more useful in etiological and criminal career research when individual-level measures are necessary. 19 1966) or"bogus pipeline" models (Jones and Sigall, 1971), was also partially tested by Hindelang, Hirschi, and Weis (1981:31, 133). After the nonanonymous interview only, respondents were inter- viewed again about a list of neutral and criminal acts, but the answers were tape- recorded and respondents were informed that a psychological stress evaluator (PSE) would enable the researchers to analyze the tapes and produce graphs that would be read for truthful or dishonest answers. Respondents were shown exam- ples of the graphs and reminded ofthe lie detector potential of the PSE. Another use, in addition to motivating respon- dents to be truthful, was direct validation of oral responses with the PSE evalua- tions of the recorded voice patterns. Preliminary analyses indicate that on reinterview with the PSE, changes are often made in initial reports, typically increases in reported crime. Approxi- mately 10 percent ofthe responses among males were changed, and about 90 per- cent ofthose changes were from denial to admission of an offense. These percent- ages were very similar for black and white mates (Hindelang, Hirschi, and Weis, 1981: 133). Other analyses of these data inform more directly the issue of the validity of this particular method of ad- ministration. If deception is related to diminished validity, one can hypothesize that respondents who apparently give less valid responses in general will be more likely to change more of their an- swers when reinterviewed using the PSE method. However, there was little varia- tion in change scores. Those respondents who may have been generating less valid answers about their criminal involvement uncler the main test conditions did not have higher change scores. Either the apparently less accurate or less truthful respondents remained consistent in their inaccuracy or deception- from the main interview to the PSE interview, regard

20 less of the apparent ability of the re- searcher to ascertain the truth, or similar change scores reflect similar levels of ac- curacy and truthfulness (or inaccuracy and dishonesty) across respondent cate- gories (Weis, 1983b). The stanclard, direct tests of validity ant! reliability indicate that no method of administration,--whether anonymous or nonanonymous, generates more valid re- sponses than others to questions about criminal behavior. However, more incli- rect information suggests that there are factors that may be affected by a particular me~od of administration and that, there- fore, inform the overall assessment of va- lidity and reliability. Of obvious interest is the difference in self-report estimates of prevalence and incidence of criminal behavior by method of administration. The research on juvenile delinquency shows the same consistent pattern of no difference in estimates by method of acI- ministration, which one might expect if methods clo not have any apparent effects on the reliability and validity of re- sponses (Hindelang, Hirschi, and Weis, 1981:12~1251. However, there are more omissions and nonresponses on self administered] questionnaires than under other methods of administration, particu- larly for longer recall periods and rela- tively nonsaTient events (Suciman and Brac3bum, 1982:2751. For self-reports of juvenile crime, the method of administra- tion is the "best predictor of the number of missing self-report responses, with both questionnaire conditions (anony- mous and nonanonymous) contributing about equally" (Hinclelang, Hirschi, and Weis, 1981:125), but the nonresponses and missing data are not randomly dis- tributed by methoc! of administration or anonymity. In the anonymous question- naire condition, 86 percent of the mates answered each of the 69 crime items, as clid 97 percent in the anonymous inter- view; in the anonymous questionnaire CRIMINAL CAREERS AND CAREER CRIMINALS condition, 90 percent of the females an- swerec! all self-report delinquency items, and 96 percent clid so in the nonanony- mous interview. The nonresponse rates on past-year fre- quency were substantially higher, con- trary to suggestions that an affirmative answer to an item sets the respondent free to report frequency of involvement with the threat of the item diluted (SucI- man ancl Braclburn, 1982~. Seventy-four percent of the mates and 79 percent ofthe females provided past-year frequency re- ports on all the offenses they admitted to ever committing. In the anonymous inter- view 92 percent of the mates gave all the required frequency estimates, but only 57 percent click so on the anonymous ques- tionnaire. In the nonanonymous inter- view 94 percent of the females had full response on frequency but only 64 percent (lic1 so on the anonymous ques- tionnaire. Both the mate and female non- response patterns suggest higher non- response rates on questionnaires than interviews, particularly the anonymous questionnaire, ironically the favored method of administration in self-report research for the past 30 years. Anct as reported for general reliability anti, espe- cially, vaTiclity, it seems that "uncler all relevant method conditions, court delin- quents, blacks, ancl mates had higher rates of nonresponse on the last year fre- quency items" (Hindelang, Hirschi, and Weis, 1981: 1261. Of course, clepending on how the missing data are managed in analysis, there need not be distortions in relationships, because the answers of these types of respondents are more likely missing from the analysis and, thereby, likely attenuating correlations between crime ancl other variables. Finally, method of administration may affect the vaTiclity ancl reliability of re- sponses to variables other than the clependent criterion ancl, therefore, the relationship between inclependent ancl

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS depenclent variables. This means that there should be different correlations by method, but the only research on this issue in criminological-methocts research concludes that correlations between juve- nile crimes and a number of often-used etiological variables (such as parental su- pervision, respect for the police, and amo- rality) vary by method, but that no method is consistently better or worse than the others in producing accurate re- sponses to attitucle-type items (see Hindelang, Hirschi, ant! Weis, 1981: 127-128). There has been more research on method of administration and anonymity than on other task characteristics, but some of the "question-related" sources of response bias apparently contribute more to unreliability and invaliclity than the structural task characteristics. Question Characteristics In a systematic review of the post-1970 survey-research literaturei7 on the re- sponse effects of question characteristics, DeLamater (1982) concludes that four as- pects of survey questions are potentially more likely to affect respondents' behav- ior than others: (1) the wording of the question, (2) the salience of the topic addressed by the question, (3) the social desirability of responding one way or an- other to a question, ant! (4) the degree to which a question arouses anxiety or is threatening to a respondent. Each of these topics is addressed sequentially in this section, as well as two task character- istics that have been identified as impor- tant question-related sources of response bias-response categories and recall pe- riod (Bradburn and Danis, 19841. Recog- nizing that there are important interac- tions among these sources of error, for i7Studies appearing before 1970 are reviewed in Sudman and Bradburn (1974). 21 example, less salient events have shorter effective recall periods (Cash ant] Moss, 1972), interactions and other sources of error are also addressed in the discussion. Question Wording. A major source of response bias can be attributed to com- munication errors, particularly when questions JO not communicate effectively the meaning of an item. Questions may slider in a number of ways that affect response, but three seem particularly im- portant when the questions are about "threatening" behaviors-content, speci- ficity, and length. The fact that even the slightest changes in the content of a question can substan- tially alter responses has been thoroughly documented in survey me~oclology (Schuman and Presser, 1981; Braciburn and Danis, 19841. Changing words in a question can lead to concomitant changes in the meaning attributed by a respon- clent and, therefore, to ctifferent re- sponses. For example, in the measure- ment of criminal behavior a critical element is "intent," and its inclusion or exclusion can change- an item from a crime to a legal act. The question "Have you ever broken a car window?" cloes not necessarily reflect a crime, ant! would yield higher estimates than an item that required that it be "on purpose" with the intent to damage it or to enter the car for illegal purposes. There is evidence from adult-offender samples that question wording is an im- portant source of response effects. Chaiken, Chaiken, and Rolph (1983) have compared the first and second Ranc! in- mate surveys, primarily on differences in question wor(ling and response catego- ries that might be responsible for the "remarkable differences" in prevalence estimates and individual offending rates between the two surveys. They conclucle that the "differences between the first and second survey's results for robbery,

22 assault and franc! can be attributed pri- mariTy to differences in wording on the survey instrument" and that for some of the crimes for which the worcling is "al- most iclentical," differences in wording could "account for substantial disparities in answers for some respondents." For example, on the second survey burglary included breaking into houses, busi- nesses, ant! cars, but on Me first survey it excluded cars and, therefore, affirmative responses on burglary for those who only break into cars (Chaiken, Chaiken, and Rolph, 1983:19, 31. The specificity of a question may also affect responses, surprisingly leading to typically higher estimates because the question is a better sect to recall (Brad- bum ant! Danis, 1984: 1~18~. This is shown in a comparison of a general and a specific vandalism item on the same sur- vey for juveniles. The prevalence esti- mates were more than 10 percent higher on the more specific item, "broken the windows of an empty house or over un- occupied building," than on the more general item, "banged up something that did not belong to you on purpose" (Hinclelang, Hirschi, and Weis, 1981:401. It is likely that the higher estimates on the more specific items are an example of "aided recall" there is more information in the specific question that "sects" or "prompts" or "cues" recall, and it pro- duces a higher rate of response. For prev- aTence estimates this is an advantage, but for frequency estimates with bounded or specifier] recall periods, it can cause more telescoping and even higher, although more inaccurate, incidence estimates (Bradbum and Danis, 1984:181. A final element of question wording that seems to make a cli~erence is the length of the question. According to Tourangeau (1984), the best aid to recog- nition ant! recall of a past event is the item itself. Research shows that longer ques- tions reduce the number of omissions and improve recall in a variety of survey set CRIMINAL CAREERS AND CAREER CRIMINALS sings (Cannel!, Marquis, and Laurent, 1977; Cannell, Oksenberg, and Converse, 1977~. The longer question is apparently more efficient because of the memory cues that the item provides and the greater length of time it allows the re- spondent to remember an event; in inter- view situations, question length is even related to length of reply (SuUman and Bradburn, 1982:50~. Unfortunately, there has been no criminological research on the effects of question length, but it seems likely that, in combination with question content and specificity, question length has important ejects on responses. In fact, these kinds of question-wording effects are considered! sufficiently sub- stantial that Hinclelang, Hirschi, and Weis (1981:40) conclucle that the "level of self-reportecl delinquency within a sam- ple appears to fluctuate broadly as a func- tion of apparently minor changes in item content" and that the "magnitude and even the direction of race and sex (liffer- ences in delinquency are often contin- gent on item content or worcling." These effects of question worcling are primarily attributable to communication errors, which can be corrected in item and instrument construction and testing. Memory errors are less amenable to ma- nipulation ant! are significant contribu- tors to other sources of response effects, particularly (differences in response cate- gories, recall periods, and salience of queried events. Response Categories. Self-report re- search on crime began with "closed- ended" questions that used "normative" response categories. The response cate- gories in Short and Nye's original 23 de- linquency items, for example, were "very often, several times, once or twice, no" (Nye, 1958:13-141. Some self-report stud- ies have continued to use similar ordinal- level response categories, losing impor- tant information in the process and, more importantly, maintaining information that

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS includes considerable error. Normative- response categories make comparisons between respondents clifEcult ant! impre- cise because one respondent can check "several times" anc3 mean nine times, whereas another may check the same re- sponse and mean three times. Each re- sponds according to personal experiences anc3 norms, which are related to personal behavior, as well as to that of peers and reference groups. Unless one "norms" each responclent's answers and thereby standardizes comparisons across re- sponse categories (see Braciburn and Sucl- man, 1979:15~162), accurate com~ari- sons are obviated. The most direct anc3 best solution is to ask the respondent to report the frequency of commission of each crime-more information is pre- served and comparisons among respon- clents are more straightforward and pre- cise, but the question ofthe validity ofthe frequency estimates then becomes a crit- ical one. It has been shown that "open-ended" questions, particularly those that might be regarclec3 as threatening (which might react to underreporting), are better than closed-enclec3 questions in producing prevalence anc3, especially, incidence es- timates. For example, Brac3burn anc3 Sud- man (1979) indicate that longer, open- endec! questions increased the reported estimates of beer consumption by 60 per- cent over shorter, closed-ended ques- tions. They also consider the responses to open items more valid (Su~man and Bradburn, 1974), which is the more criti- cal issue. Contrary to what they suggest, higher estimates clo not necessarily mean more valid estimates, anc3 if the estimates are not more valid, error is introduced at the high end of the distribution by over- reporting. In criminological research, es- pecially research on serious, chronic of- fenclers and on correlations that often depenc! on the tail end of a distribution, this kind of response bias could be partic- ularly problematic. High-frequency but 23 less valid estimates of criminal involve- ment could distort correlations, likely am- plifying the correlations between crime and other variables. A good example of the impact of re- sponge-category format is the difference in estimates of individual offending within anc3 between the first anc3 second Rand inmate surveys. The first survey used open-ended frequency questions anc3 a modified closec3-open-endecl for- mat that included four frequency catego- ries (0, 1-2, 3 5, ~10) anc3 a fifth category that asked for a frequency if the respon- clent had engaged in the crime 11 or more times. The differences for burglary are substantial; the mean rate from the open- ende(l, frequency-response format was 16 times the mollifier! categorical rate (Peterson and Braiker, 1980:261. This is also higher than the burglary rate esti- mates from the second survey, in which two other modified frequency-response formats were used (see Chaiken and Chaiken, 1982:Appenclix E, 16, 41~. It is also reported that for high-frequency crimes, for example, drug (leafing, the frequency reports based on unit of"time on the street" produced even higher esti- mates than simply asking for a total fre- quency count over a stan(larc3 unitary re- call period, such as the past year. The implications for vaTiclity are not straight- forwar(1 the higher estimates couIcl sim- ply reflect even greater inaccuracy of responses. As always, the problem is val- iclating reports of criminal behavior for which no external validation criteria are available. Short of this, what is known about the cognitive psychology of fre- quency estimates of past events may be helpful in interpreting findings ancl, ulti- mately, in developing instruments. Recall Period and Salience. Since open-endecl questions that ask for respon- clent estimates of "frequency" of criminal behavior are relatively stanclar<1 in self- report research and are essential for mea

24 suring incTiviclual offending patterns, the accuracy of reported frequencies be- comes a paramount concern. Recalling the frequency of past events is also re- lated to two other question-task character- istics "recall period" ant! "salience" of the topic. Hence the discussion turns to the interactions among the three, as well as the contributions of other related sources of memory error, for example, the "similarity" of events, "retroactive inter- ference," ancI"primacy-recency" effects. In general, respondents provide more accurate estimates of low-frequency events anct much more variable responses for higher frequency events (Loftus, 19801. When the event is criminal behav- ior and the respondent is asked to locate it within a specified recall period, say the past year, the task becomes much more complex and the sources and magnitude of potential error increase. Two basic sources of memory-performance error are referred to in the criminological litera- turc "forgetting" what happened (mem- ory clecay) and "distortion" of memory by misplacing an event temporally (telescop- ing), which usually involves remember- ing it as more recent than it was in actu- ality (Garofalo and Hindelang, 19771. A third source of memory error, "construc- tion," is usually not considered but is a source of substantial error according to recent research in cognitive psychology (TalIand, 1968; Klatzky, 1975; Loftus, 1980, 19821. Beginning with telescoping and mem- ory decay, there is little doubt that the former occurs in the reporting of victim- ization and criminal behavior. For exam- ple, Garofalo and Hindelang (1977) re- port that about 20 percent of the victimizations that respondents placed within the recall period actually hap- pene(1 before it, according to police rec- orcis. Telescoping is one of the conse- quences of"primacy-recency" effects in memory we tend to remember best CRIMINaL CAREERS AND CAREER CRIMINALS those events that happened recently and worst those that occurred in the distant past, with the qualification that those events at the beginning and ens] of a series will both be remembered better than those in the miciclle (a "serial position" effect) (Lotus, 1980:2~251. This means that the longer the recall period, the more criminal acts the respondent will forget, but report- ing of the earliest and most recent criminal experiences will probably be most accu- rate. For example, Bridges (1983) reports that in the Philadelphia cohort follow-up sample, self-reports of the respondent's most recent offense were the most vaTic! and reliable. The emphasis in criminological re- search on "memory clecay" as the most important source of forgetting and under- reporting in both victim and self-report surveys seems misplaced both theoreti- cally and empirically. Forgetting behav- ioral events such as criminal acts involves the loss of ability to remember informa- tion that has been encocled and stored primarily in episodic, long-term memory (Natzky, 1975:124-127, 17~179~. There are two kinds of forgetting: memory de- cay, which is a passive form of forgetting as a simple function oftime passing, and a more active form, interference, which en- tails one experience interfering with the remembrance of another (Murclock, 1974~. "Retroactive" interference and "similarity" interference exert the strong- est effects on remembering past events. The former refers to the interference cre- atecl by subsequent events in remember- ing prior events, and the latter refers to repeated similar events interfering with the ability to distinguish unique charac- teristics of any one in a series of similar events (Hunter, 1957:258-268; Murclock, 1974: 11-12; Klatzky, 1975: 12~127; Lof- tus, 1980:66~7~. In retroactive interfer- ence, if there are many interpolated events and they are also similar in their characteristics, the forgetting can be sub

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS stantial, particularly the ability to clistin- guish one from another and, therefore, to remember how many discrete events oc- currecI. The primary role of interference and the secondary role of memory decay in forgetting has been highlighted by Murdock's (1974: 12) conclusion that per- haps "interference accounts for 8~98 percent of the variance and clecaY mav account for the remainder." In short, fre- quent, similar retroactive interference is a very potent force in forgetting, even more so when there has been faulty or incom- plete encoding and storage of an event to begin with. The implications for selecting appro- priate recall periods for items in a survey are different from an interference versus a memory-decay perspective on forgetting. The recall periods that are typically used are basest on the principle that forgetting increases with time, without taking into account other influences on memory, such as "salience" of the question topic, which has demonstrable effects on recall and the perception of time (Bradburn and Danis, 1984:30~. The arbitrarily favored recall period is 1 year, on the grounds that longer time periods increase forgetting, frequency estimates, and error, and that shorter time periods might not produce enough reports of low-base-rate behav- iors for analytic purposes. Whether 1 year or any recall period is too short or too Tong is an empirical question, one that is com- plicated by one of the "most influential of the factors" that affect responses, the sa- lience of the topic (DeLamater, 1982:451. Research shows that salient events, those that are important, relevant, and conse- quential, are remembered better than those that are not. Cash and Moss (1972) have shown in self-reports of car acci- clents that the memory of events with high salience was adequate up to about 1 year; the period for intermediate salience was 1~ months, and for low salience, from 2 weeks to 1 month. 25 The salience ofthe event is also related to perceptions of time. Salient events are perceived as more recent than they really are, which means that forward telescop- ing is more likely for behaviors with high salience (Braciburn and Danis, 1984:3001. This, in turn, inflates frequency estimates of the more salient events. Recent events are also perceived as more salient (DeLamater, 1982:22), which has similar effects on respondent's reports of fre- quency. One of the other characteristics of salient events is that they are usually not relatively frequent within an individ- uaT's experience. Behaviors that are more frequent and also similar to each other will also be less salient; in addition they increase retroactive interference, which in turn, decreases the ability to cTifferen- tiate discrete events and distorts fre- quency estimates. In self-reports of criminal behavior, these effects may be problematic for re- ports of the most serious offenses and for the most serious offenders. For most peo- ple, even most offenders, committing a serious crime would be a salient event, relatively easy to remember. In general, one would expect more accurate esti- mates for serious, less frequent offenses than for less serious, more frequent of- fenses. However, we know that the ma- jority of serious crimes are committed by a small subgroup of serious, chronic of- fenclers who commit a variety of serious crimes with high frequency for exam- ple, in one study the 10 percent most active incarcerated adult burglars acimit- tecl to an average of 232 burglaries per year (Chaiken and Chaiken,1982:48~. For those offenders, burglary is a routine, high-frequency behavior, one much like another. Based on what cognitive psy- chology and survey methoclology suggest about the interactive effects of frequency, salience, recency, similarity, and retroac- tive interference, one could hypothesize that their estimates of the number of bur

26 glaries they commit contain substantial error. The problem is determining the direction of the response bias the com- bination of similar, Tow-saTience, high- frequency behaviors could very likely leac] to underestimations, although this is hard to believe given the very high fre- quency estimates that were reportect. There is no empirical reason to rule out the possibility that a respondent might overreport for other reasons that have little to do with the cognitive aspects of memory, for example, social approval, cle- ceit, or other motivated sources of error that are more related to respondent char- acteristics than to question characteris- tics. Social Desirability. A question may cause the respondent to consider the so- cial desirability of the response, rather than its accuracy. A respondent may give an answer out of a desire or need for "social approval" or because the Question has a "trait clesirability" that elicits an approving response (Edwards, 1957~. Sucl- man and Bradburn (1974) report a consis- tent social-desirability response effect, but one that varies by the nature of the clepenclent variable and the respondent's sociodemographic characteristics. How- ever, the effect of social desirability on the validity of self-reports of crime among juveniles is not relatecl to apparent cliffer- ences in the validity of their responses. Since this type of response effect may be more important for respondents who are actively involved in crime and may want to deny it, Hindelang, Hirschi, and Weis (1981) inclucled two clesirability variables in their analysis: a five-item acquiescence (yea-saying/nay-saying) scale and a five- item social desirability scale. Within all methods of survey administration and sex, race, and class categories, these so- cial-desirability effects accounted for about 2 percent of the variance in invalid CRIMINAL CAREERS AND CAREER CRIMINALS ity, as measured by a deceit score that equalecl the sums of nonreports across the 13 offense categories identified in a re- verse record check (Weis, 1983b). In short, social desirability does not seem to be an important source of response bias in survey questions about crime, at least among juvenile respondents. Respondent Characteristics Hindelang, Hirschi, and Weis (1981: 219) have concluded from their analysis of the validity ant! reliability of self-reports and official records of juvenile crime that Delinquency exists most clearly in the minds ofthose least likely to engage in it." To the extent that this is true, the impli- cations for research are increasingly im- portant because, as samples become more representative of serious crime and crim- inals, "measurement problems are inten- sified." These problems seem to be more related to characteristics of subgroups of respondents and to motivated error than is reported in the literature on general survey methodology. Of the three major categories of re- sponse effects (interviewer, task, and re- spondent characteristics), the impact of respondent characteristics on answers is probably the most ambiguous. The sur- vey-methodology research on the effects of such sociodemographic attributes as gentler, race, and education tencTs to show no "general effects" (Suciman and Braclburn, 1974; DeLamater, 19821. How- ever, clepending on the task, there are observed differences for example, so- cial-(lesirability response effects may be greater among women (Clancy and Gove, 1980; Gove, 19821. A variety of factors may be related to respondent effects, but here the discussion focuses on only those that seem particularly pertinent to the potential response bias to questions about crime created by the characteristics of

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS serious, chronic offenders. First the char- acteristics of these offenders are cie- scribed briefly, and then the apparent ant] potential response effects ofthese respon- clent characteristics are examined. Characteristics of Serious, Chronic Offenders . . . . . From research on juvenile and adult criminals, there Is a relatively clear pic- ture of the small subgroup (between 5 and 10 percent) of offenders who are responsible for committing the majority of serious crimes (Wolfgang, Figlio, and Sellin, 1972; West and Farrington, 1977; Hamparian et al., 1978; Strasburg, 1978; Hinclelang, Hirschi, and Weis, 1981; Pe- terson, Braiker, and Polich, 1981; Chaiken and Chaiken, 1982; Shannon, 19831. In general, both juvenile and aclult serious, chronic offenders commit a greater variety of crimes more frequently than other offenders, particularly the more serious crimes ant] those involving violence (e.g., burglary, robbery, assault, drug dealing, auto theft, grand larceny). They begin their careers at an early age, both committing crimes and being o~- cially sanctioner] in their early teenage years. The onset of their criminal behav- ior typically involves substance abuse, which often becomes a frequent and heavy habit. The early onset and long duration of frequent and serious criminal- ity indicates commitment to a criminal life-style among many of these offenders. The "violent predators," those inmates who have been identified as the most serious, chronic offenders in the Rand sample, are very young compared with other offenders, are more involved in vi- olence, and, typically, are more serious substance abusers (Chaiken and Chaiken, 1982;6~65, 86 871. In aciclition to cleal- ing clrugs, violent predators started using hard drugs frequently as juveniles and 27 have continuer! their substance abuse, including heroin addiction, into adult- hood. In fact, it is reporter! that 83 percent of the violent predators were using hard drugs (heroin, barbiturates, amphet- amines) in large-quantities on an almost daily basis during the measurement pe- riod (Chaiken and Chaiken, 1982:65, 155~. They are also more likely than other offenders to have heroin habits, although "extreme" polydrug use is more charac- teristic, especially heroin with amphet- amines, barbiturates with alcohol, ant] barbiturates with amphetamines. As the cost of their abuse increases, so does their criminal activity. Drug use is identified as "one of the major factors in serious high- rate crime" and is associated with "virtu- ally every type of crime" (Chaiken and Chaiken, 1982:1551.~8 With the exception of heroin-only addiction, the other com- binations of drug use are related to vio- lence. This symbiotic relationship between substance abuse and crime among many serious, chronic offenders is interwoven into a broader deviant life-styTe that en- compasses cultural norms and values that maintain the relationship ant! exert addi- tional, independent cultural influences on He offencler's behavior (Liebow, 1967; Foster, 1974; Valentine, 19781. Chaiken en cl Chaiken (1982:176) estimate that"a relatively large fraction of the inmate population, up to 65 percent, may take part in those life-styles." i8This is also apparently true among samples of the general youth population; lIuizinga and Elliott (1981:80) report that for the majority of young peo- ple the use of drugs is not related to criminal behavior, but, that among those who use drugs and do crime, the levels of juvenile criminal involve- ment are highest among those who use the greatest variety of drugs (alcohol, marijuana, and other drugs) most frequently. In short, more drug use is related to more criminal behavior among juveniles.

28 Effects of Serious, Chronic Offender Characteristics on Responses Of interest to most researchers are the effects of offender characteristics on each other, for example, drug use on crime. However, another way to examine the effects of respondent characteristics is on the measurement of criminal behavior, specifically on the validity ant! reliability of the answers of respondents who pos- sess those attributes. Unfortunately, this has not been the focus of any major re- search effort, but it is apparent that fre- quent involvement in serious crime over a long period of time, substance abuse, and engaging in violent acts may have cleleterious response effects on self- reports of criminal behavior. As reported earlier, Chaiken anct Chaiken (1982: 22~226, 244) concluded that most incti- vidual characteristics are not related to the validity ("external reliability") and reliability ("internal quality") of re- sponses of prison inmates, but they ctid find a few exceptions. For each of the exceptions to respondent effects, what is important is that validity and reliability problems lead to suppression of self- reported indiviclual offending rates. Among an adult-prisoner sample, the ev- idence suggests blacks, the less educated, drug dealers and users, alcoholics, and life-styTe criminals probably have less ac- curate, Tower estimates of self-reported criminal involvement. It is hypothesized that the unique characteristics of more serious, chronic offenders contribute to both increased memory error and moti- vated error in their responses to questions about their official criminal records and criminal behavior. There is little doubt that drugs, ranging from alcohol to heroin, impair a person's ability to encode and store information about an event during which the person was under the influence of drugs. This diminished capacity to learn and pre CRIMINAL CAREERS AND CAREER CRIMINALS serve the characteristics of a past event due to intoxication can occur even when moderate amounts have been consumed. Parker and Noble (1977) report that, although the decrements are greatest among heavy drinkers, they are also evi- dent among light drinkers. Encoding and storage are impaired by alcohol, but re- trieval is not that is, if an event occurs while the respondent is in a sober condi- tion, it can be retrieved equally well whether the respondent is sober or intox- icated, unless the respondent is very in- ebriated. However, if an event occurs cluring an intoxicated state, the informa- tion is less available for retrieval whether the respondent is sober or intoxicated, although there is some research evidence of"state-dependent" recall (Bower, 1970; Birnbuam and Parker, 1977:106; Wein- gartner ant! Parker, 1984~. This type of research has typically been carrier] out with "normal" or "social" drinkers. Among alcohol abusers and al- coholics the effects are even more power- fu! and dramatic; there may be chronic cleficits in storage (Cermak and Butters, 1973) due to heavy episodic consumption or continual intoxication and the influ- ence of the "alcoholic haze" or "black- outs" (Loftus, 1980:93-95~. It is possible to forget everything about an event, even a salient one; other events may be re- memberec3 only in part and only when remincled of what happened. The same kinds of effects have been documented for other ([rugs, including marijuana, PCP, and heroin. There is ev- idence that PCP drastically impairs learn- ing and memory functions (Perry, 1976~. The research on marijuana shows that, like alcohol, there are substantial effects on new learning and storage but not on retrieval (DarIey et al., 1973; Loftus, 1980; Wetzel, Janowsky, and CIapton, 1982), and there is evidence that free recall is hampered more than recognition memory (Miller and Cornett, 1978~. The

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS research on heroin and polydrug abuse effects is even more interesting because there is some evidence of a racial differ- ence. Penk et al. (l9Sla,b) report greater decrements in short-term memory among black and Hispanic heroin and polydrug abusers, which suggests that heavy drug abuse may have a more cleleterious effect on memory among those groups than among white abusers. Given that in a typical adult prisoner population about 40 percent were intoxi- cated daily during the year prior to last arrest, 40 percent of the crimes for which they are incarcerated involved alcohol intoxication, 65 percent of the inmates used other drugs cluring the year prior to last arrest, and 40 percent of the crimes for which they are incarcerates] involved other drugs (Sykes, Roccolo, and Thomp- son, 1980:60), and, given the sordic! drug history of most serious, chronic offenders, it is a short logical leap to the hypothesis that among both juvenile and adult seri- ous, chronic offenders there will be sig- nificant memory error in responses to questions about their past criminal be- haviors. A substantial proportion of them wflT be uncler the influence of drugs at the time of the crime, and this will diminish their ability to encode and store informa- tion pertinent to the cletafls ancI even the essentials of the event. Depending on the amount ant! kinds of drugs used, and the duration of the drug-using pattern, the completeness and accuracy of recall wflT vary, but in general the greater the intox- ication, the more the memory errors and the Tower the estimates of involvement in crime. The effects of drugs may interact with another factor in forgetting the nature of the behavior or event. There is evidence that alcoholics "selectively forget" un- pleasant events that occur while they are drunk (Jones and [ones, 19771. This ten- dency to have a poorer memory for bad experiences is characteristic of nonaTco 29 holics as well (Erdelyi and Goldberg, 1979~. Events that arouse stress, anxiety, fear, guilt, shame, worry, and other nega- tive emotions ant] feelings hinder accu- rate perception and encoding of the cle- taiTs of events, as well as the ability to retrieve and recall what happened (Sie- gel and Loftus, 1978; Loftus,1980:77~21. In short, we tent! to remember the good ancI forget the bacT. A related factor is the trauma, clanger, or violence that is inher- ent in an event. Loftus ant] Burns (1982) have demonstrated that the recall of wit- nessing a violent event a film of a boy being shot in the face-was substantially inferior to the recall of a nonviolent con- cTition. Given that serious, chronic offenders engage in serious property and violent crimes, such as robbery and assault, one can hypothesize that the memory of seri- ous criminal acts, and particularly those that involve interpersonal face-to-face confrontations and the threat or actuality of violence, will be impaired signifi- cantly. If one also takes into account the probability of drug-intoxication effects, the impairment would be even greater. This memory error shacles into motivated error when a respondent consciously wants to forget things that involve pain, acute anxiety, blows to self-esteem, threats to values and norms, feelings of guilt and shame, and so on (TalIancI, 1968:89-911. An interpersonal crime is likely to include most if not all these elements and, therefore, is a goof] cancli- clate for being forgotten; at a minimum it is not the kink! of behavior that one wants to remember. In some people this desire to forget can make memories more or less unretrievable through "regressive for~et- ting" (Hunter, ~. ~ ,, ~ 1957:245) and outright Rena tnat tne event or behavior ever occurred (TalIancl, 1968: 106~. It is also likely that other characteristics of the respondent contribute to the inten- tional, motivated distortion of memory.

30 Dishonesty, deceit, and lying may moti- vate inaccurate responses for a variety of personal and cultural reasons. Any time a past event includes ourselves, there is a tendency to remember He good and for- get the bad characteristics of not only the experience but also of ourselves. There is a "self-serving bias" (Nisbett and Ross, 1980) in operation that preserves or en- hances our self-esteem, particularly when our role in the event for example, knif- ing someone is not considered a laud- able act according to usual cultural stan- dards of propriety. The operation of this bias may lead to lying about what hap- pened. As Loftus (1980:142) has pointed out, in constructive memory "people tend to rewrite history more in line with what they think they ought to have done than what they actually did." If a respondent does not values a behavior and sees it as discrepant with self-image, the response may range from denial to reporting less than actual involvement. On the other hand, if the behavior is valued and en- hances self-concept, the response may be inflated. The former condition is consid- ered to typify involvement in and report- ing of criminal behavior, but for some criminal respondents the latter condition may be more typical. What is self-serving, socially desirable, or socially approved will reflect ethno- centric as well as egocentric criteria. For the estimated 65 percent life-style crimi- nals (Chaiken and Chaiken, 1982:176), what is valued will more likely reflect criminal cultural values, making the self- serving bias much more unpredictable in its effects on response accuracy. Another cultural source of response bias may be the black, mate, street-corner culture, wherein not cooperating with "the man" has been refined over many years of op- pression into a highly functional form of defensive and, more subtly, offensive ad- aptation. Here anyone, even a black mate researcher, is a figure of authority not to CRIMINAL CAREERS AND CAREER CRIMINALS be trusted but rather to be hustled and cajoled into believing in the sincerity and veracity of the respondent. These same kinds of attitudes and values exist in white working-cIass, street-comer cul- tures (Miller, 1958) and may have similar kinds of effects for those types of of- fenders. It is also likely that intellectual compe- tence contributes to unintentional mem- ory error among serious, chronic offend- ers, more so than among other offenders or nonoffenders. We know that intellec- tual competence has an important rela- tion to crime in general (Reiss and Rhodes, 1961; Wolfgang, Figlio, and Sel- lin, 1972; West, 1973; . . ~. ~ __\ , Hirschi and Hindelang, 1977) and an effect on the validity and reliability of self-report re- sponses about criminal behavior among juveniles (Hindelang, Hirschi, and Weis, 1981:202-204) and adult offenders (Chaiken and Chaiken, 19821. The effect is apparently so substantial that it ac- counts for some of the racial difference in validity discovered in the Seattle youth study. The relation between a 20-item general-knowledge scale and official and self-reported delinquency is negative, as one would expect, but there is apparent underreporting by black males with Tow knowledge scores. The difference be- tween white and black mates in self- reported delinquency is, in part, a conse- quence of differences in "knowledge." Similar differences are apparent for self- reports of school grades, i.e., relationships between knowledge scores and official and self-reported delinquency indicate differential validity by race, with black males giving the apparently least accurate answers (Hindelang, Hirschi, and Weis, 1981:19~201, 202-2041. Apparently, general cognitive process- ing, symbol manipulation, problem solv- ing, information processing, and even the capacity for storage are related to mea- sured intelligence (Klatzky, 1975:31

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS 3211. Since many of these processes are activated directly in a survey situation, one should expect more memory errors among respondents with less capacity to perform those processes because of lower intelligence, competence, or knowledge. The evidence among juvenile and adult offenders suggests that this is the case, and one can hypothesize that the effect would be even more acute among a group of serious, chronic offenders. Another complication is that knowI- ecige errors increase as the ability to count past events diminishes as a func- tion of time and frequency. The respon- dent is forced to guess more at the high- frequency end of the distribution, claiming knowledge that he floes not have, with the result that most respon- clents will underestimate, but some will overestimate, which contributes to the greater variation and measurement error in high-frequency estimates (Hasher and Chromiak, 19771. This clifficulty in recall is evident in the "large disparity" be- tween aclult offenders' self-reports of ju- venile crime and their official juvenile records (Chaiken and Chaiken, 19821. Many crimes were acimittec3, but the rec- ords showed no criminal involvement. This coup! be a consequence of forgetful- ness ant! memory distortion among a group of respondents with characteristics that are likely to produce these kinds of response errors. Of course, another possi- bility is that they clic! not, indeed, have official contact while juveniles. After all, the odds of being processed by the juve- nile justice system are quite low, even among active offenders (Erickson and Empey, 1963; Farrington, 1973~. Obviously, there is another possible explanation. Chaiken and Chaiken (1982: 224) point to the inadequacies of the rec- orcis available to criminal justice agencies as the source of the disparity. That is, the limitations of official records also contrib- ute to validity and reliability problems in 31 the measurement of crime, including the measurement of individual offending pat- terns in criminal career research. LIMITATIONS OF OFFICIAL RECORDS Most of the major limitations of mea- sures of the amount and distribution of crime based on official records hac3 been identified by the end of the nineteenth century, including the primary problem of the "ciark figure" of unknown crime (Sellin and Wolfgang, 19641. Over the years these limitations have been docu- mented repeatedly. For example, Hinde- lang (1974) discusses 14 and Savitz (1982) 27 limitations of, or problems with, of- ficial crime statistics, particularly data from police records and the Uniform Crime Reports (UCR). These limitations range from cli~erences in counting rules for property and person crimes to compa- rability difficulties across jurisclictions. However, they are not clebilitating limi- tations in fact, the vaTiclity and reliabfl- ity of official measures, as evaTuatecl by the standard social-scientific criteria of comparing results over time and with other measures, are quite good. For ex- ample, the correlates of offenses known to the police are stable over the 50-year period covered by the UCR, and the cle- scription of the characteristics, distribu- tion, and correlates of crime is similar to that produced by both victimization ant! self-report survey measures (Belson, 1968; Hinclelang, 1974, 1978; Maltz, 1975; Gordon, 1976; Nettler, 1978; Hinclelang, Hirschi, and Weis, 1981; Thornberry and Farnworth, 19821. A number of the limitations of official- record data are cliscussecT briefly below; those problems that have more bearing on criminal career research are discussed in more cletaiT. Among the variety of official data on crime, "offenses known to the police" are

32 the most widely consulted and utilized, because they are considered closest to the crime itself and in a more constant rela- tionship with crimes committed. The UCR, publishecl annually by the FBI, provide statistics on the amount and dis- tribution of crimes known to the police, arrests, and other information about the criminal justice system. A variety of prob- lems with these police ciata can affect validity and reliability and, therefore, the synthesis of estimates of criminal career parameters proclucec3 by offical data and self-reports. In general, o~cial-record clata, whether police, court, or prison data, uncleresti- mate the amount and distribution of crime, at both the aggregate and indivic3- ual level, although here the latter is of more concern and interest. Individual offenders are clifferentially vulnerable to acquiring a record, for a variety of reasons, all of which compro- mise the validity anct utility of official- record data for research purposes. First, police practices-for example, arrest pol- icy, number of personnel, allocation of resources, organizational ideology-vary over time and place, which creates poten- tially different arrest risks and enforce- ment. Second, laws vary by time and place, and this affects the probabilities of offense and official reaction. Thirst, changes in local data needs may refocus priorities and resources ant] leac! to dif- ferent recording practices across ant] within departments. Fourth, because of agency and jurisdictional differences in policy and practice, the accumulation and comparison of information on an indivic3- ual who has records from more than one jurisdiction are problematic. Fifth, typi- cally only the most serious offense in a criminal event is recorded; other offenses are not necessarily repr~-sent~--in the official record. Sixth, the number of of- fenses counted varies by type of crime- for "person" crimes the number of vic CRIMINAL CAREERS AND CAREER CRIMINALS tims in any event equals the number of offenses, while for"property" crimes the event equals the crime. Seventh, most crimes of violence those that produce bodily injury are not included in the UCR index of crimes and, therefore, are less likely recorded as rigorously by local law enforcement personnel. There are many other methoclological ant] crime- cIassification problems with official rec- orcts, as well as other problems that im- pact the calculation of offense rates, but they will not be acIdressed here (see Hindelang, 1974; Savitz, 19821. Overall, these general limitations compromise the validity, reliability, an(1 utility of official- record data, especially police statistics, in criminal career research. Other limita- tions that affect the measurement of crim- inal careers more directly are the focus of the remainder of this section. Four limitations of official records- insufficient information, discontinuity, time cocling, and samplingi9 point to the fact that both self-report ant! official mea- sures have limitations and that official- recorcT clata contribute to the apparent vaTiclity an(1 reliability problems. In short, the limitations of official-recorcT data make the identification of the serious, chronic offender more difficult, the caTi- bration of official to self-report estimates less precise, and the reciprocal validation comparisons of the two measures more difficult. Official records are not maintained pri- marily for research purposes, an(1 conse- quently, whoever the researcher or what- ever the research task, there will be i9There are many other problems, but they are not discussed because of space limitations or be- cause they are not considered problematic for crim- inal career research for example, the coding and scoring problems that arise when~harge~-colmts, and arrests in police records have to be distin- guished by a coder (Marquis and Ebener, 19811. The reader is referred to Savitz (1982) for a treat- ment of this and other kinds of problems.

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS ~nsufEcient, inadequate, or biased infor- mation in the records. This is particularly true for ciata on incTiviclual offenders, sim- ply because official data systems are cle- signec! for bookkeeping and the aggrega- tion of data. In trying to identify violent predators from the array of official-recorcI data available, Chaiken and Chaiken (1982:12), for example, concluded that the information that is routinely collected within the criminal justice system cannot be usect to make meaningful distinctions between violent predators and other of- fenclers. They consider the validity of the records to be "as suspect as the respon- clent's veracity: Records are often missing or incomplete" (Chaiken, Chaiken, and Peterson, 1982:91. And those data that are essential for criminal career research conviction and disposition records are among the least utilized in the criminal justice system and in research. Apparently, the information that most accurately predicts high-rate offenders is not available because it is not systemati- cally and routinely collected by the po- lice, courts, or corrections system. For example, in the Rand inmate surveys, only 3 of the 10 variables that predict high-rate robbers are from official rec- ords; data for the others were collected in the self-report survey (Chaiken and Chaiken, 1982:86 871. Those offender characteristics found to best distinguish serious, chronic offenders extensive drug abuse, employment instability, and a record of early juvenile violence" are not easily accessible within the criminal justice system. And the offense-history attributes of these offenders exacerbate the problem there is greater likelihood of error in their records simply as a func- tion of their more frequent contact with the criminal justice system, more time being coverer] by the records, more juris- dictions and agencies being involved, more information recorcled, more varied and complex cases, more mobility, more 33 discontinuity in record over time and place, and so on. In other words, a long- term, complicated offense history is prob- ably less valid and less reliable than a recent, simple one. One can hypothesize that crime is reco']ec3 least accurately in the records of those more involved in crime. The likelihood oftnore error in the records of serious, chronic offenders will make vaTiciation and synthesis with self: reports even more clifEcult than for less serious offenclers. The problem of incomplete, inade- quate information is pervasive, ranging from aclult arrest records (Marquis and Ebener, 1981:86) through juvenile rec- orcis. Paradoxically, automated record sys- tems do not necessarily mean that more complete and acleauate information will be provided. In fact, there is a tendency to streamline and economize, to choose a select subset of items that have been collected for entering into the system. This information can be retrieved more easily, but the detailed information that one might need for research purposes is either lost for good because it was not entered or is only available when re- trievecl manually, just like in the oIcl system. juvenile records may present even more problems than aclult records, in two ways. First, juvenile police records tend to be less complete and cletailed than adult police records, primarily because the rehabilitative i(leology of juvenile jus- tice diminishes the importance of and need for good police data. This is partic- ularly so if the juvenile court and police department have a close working rela- tionship, for example, a police juvenile division that works directly with the court-intake system. On the other hand, juvenile court records typically are richer in details about the social, economic, and psychological characteristics of the of- fender than are aclult records, again due primarily to the nee(1 for this kink! of

34 information in making disposition deci- sions that may be more therapeutic than punitive. In some juvenile court jurisclic- tions, this separation in functions may be institutionalized in separate record-keep- ing systems, a "legal" file for offense- relatec3 information and a "social" file for offencler-relatec3 information.20 In short, juvenile records, police records in partic- ular, are incomplete and c30 not represent the seriousness of actual criminal in- volvement of either those juveniles who clo have records or, obviously, the many serious juvenile offenders who do not (Tracy, 19811. The seconc] way in which juvenile rec- orcis are problematic reflects the cli~culty of accessing information, however incom- plete or inadequate, about juvenile crim- inal activity ant! other characteristics by adult criminal justice system officials (Chaiken ant! Chaiken, 1982:2244. Given that aclult offenders who are most in- volvect in crime began their careers, ac- corcTing to self-reports, before they were 13 years old (Peterson, Braiker, and Polich, 1981:xxix), the unavailability of juvenile record information, particularly on such factors as onset, drug abuse, and violence, is a substantial problem. This is primarily a problem of continu- ity between juvenile and adult records, but it also affects the ability to measure change from juvenile to adult status with standard units of comparison measured and recorclec3 in similar ways. Other im- portant official sources of information, and even of correlational validation (see Hindelang, Hirschi, and Weis, 1981: Fig- ure 5.1, 989), are not typically coordi- natec3 or integrated with either juvenile or adult criminal justice record systems. Se 20This discussion is based on the author's re- search experience with a number of juvenile court and police search systems, particularly the one in King County and Seattle, Washington. CRIMINAL CAREERS AND CAREER CRIMINALS rious, chronic offenders often exhibit early troublesome behavior in or are ser- vicecl by the school system (West and Farrington, 1977; Farrington, 1982), men- tal health system (Cocozza and Steadman, 1974), drug abuse agencies, and other social service institutions. Although at the national level there has been some orga- Sized effort to coordinate fecleral statistics on children, youth, and families, coorcli- nation among local agencies, particularly in sharing record information, is rare ant] certainly not systematic Will, Peterson, anct Moore, 19841. This kind of informa- tion on social, economic, and psycholog- ical characteristics would be useful in confirming anti supplementing data on current or incipient serious, chronic of- fenders. Even though official-record informa- tion on temporal sequencing;-onset, crime switching, termination is more trustworthy than self-reports of time, the fact that official records are the only half- way reliable source oftime-cocling makes error more critical. The ciates recorded in criminal history records have been iclen- tified as particularly problematic because termination markers, such as "release ciates" from institutions, may not be sys- tematically coded and recor(lecl (Green- wooc3, 1979:411. This means that caTculat- ing "time at risk," "street time," or "free time," which is crucial to estimating indi- vidual offending rates, is hampered by this source of error. If measuring tempo- ral changes in criminal behavior were not so central to criminal career research, and if respondents' memories for ciates and sequential orders were not so unreliable, these time-co(ling problems would not be consi(lered as critical as they are here. And the problem is not specific to crimi- nal history records; it is a pervasive prob- lem from the point of police contact through discharge from parole. Finally, there is the issue of sampling, ~v

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS both of offenders and offenses by the juvenile justice and criminal justice sys- tems. Official measures underestimate the actual number of offenders, offenses in the population, and offenses of known offenders. This underestimation is not necessarily a problem if, as Quetelet (1833) suggested, there is a constant ratio between the actual amount of crime and officially recorded crime-the latter then "represents" the former and is an accept- able measure of crime (Weis, 1983a:3791. But if the difference in the ratio between unknown and officially known crimes is not due to random error but to systematic bias that cannot be identified, specified, and corrected, the offlcial-record data can- not produce accurate estimates of the ac- tual amount and distribution of crime. Unfortunately, the ratio of unknown to known crime is clifflcult to specify with precision, and it certainly is not invari- able. The criminal justice system may behave in relation to certain offenses and offenders in ways Mat create differential probabilities of reaction for example, ar- rest that concomitantly produce incon- stant ratios between crimes committed and arrests. These may be variations by offense type and seriousness, as well as by offender attribute. The biggest prob- lem with official clata is the unclerreport- ing by victims to the police. An unknown number of crimes are never discovered, and, of those that are, the majority are not reported to authorities by the victim or other citizens (Skogan, 1977~. Unfortu- nately, this underreporting or nonreport- ing to the police varies by offense type, 35 occurred, whether reported on the survey or not.2i Of those crimes that are reported to authorities, some may not be recorded, but more likely they will be codecl and recorder! in such a way that using them for many research purposes is clifflcult. Among the variety of coding and recorcI- ing problems (see Savitz, 1982), one is particularly problematic for comparisons of official and self-report data, whatever the purpose. The legal categories that are user] to process and code offenses are broad and do not necessarily correspond to the criminal behavior that led to official processing.22 There may be substantial disparities between the actual criminal behavioral event and the legal category within which it is processed and re- corded, as well as substantial variation within the legal category. For example, as a consequence of plea bargaining, the beating of one person by another could be processed, coded, and recorded as lisor~lerly conduct. If there is disparity, perhaps substan- tial, between the criminal behavior that an incliviclual commits (and admits on a self-report questionnaire) and the legal designation of the offense of recorcl, the research and measurement problems are significant. Comparisons of self-reports and offenses of record on an individual- offencler basis will be problematic. Even for relatively serious offenders, there may be few offenses in the record, which may force aggregation into even broader of- fense categories to make comparisons victim and offender characteristics per- ~ ~ It Is also possible that there are differential ceptions of police efficiency, anti a num- validity effects by respondent characteristic similar her of other variables (Hinclelang, to those observed for self-reports. If so, it makes the Hirschi, and Weis, 1978~. This variation is victimization data less useful in extrapolating esti not random or precLictable in any rigor- mates of total crime rates from official records. Ous, precise way. As with self-reports, there is no way to know or to verify the number of victimizations that "actually" ~ t course, these broad legal categories and what is included in them will also vary over time and across jurisdictions as well, while the behavior . . . can remain Invariant.

36 possible with self-reports (see Hindelang, Hirschi, and Weis, 1981:121~. Crime-by- offense-of-record comparisons are cliffi- cult unless the sample is very large or very criminal. These disparities would also diminish validity, particularly cor- relational validity and reverse record checks. The amount of error introclucec! wouIc! likely attenuate correlation coeffi- cients between having an official record and both self-reports of criminal behavior and self-reports of official record. The usual reverse-record-check analysis, wherein the percentage of offenses in police records that are not self-reportec] is interpreter] as an indicator of relative va- lidity, also is affected detrimentally by the error introduced by the noncorrespon- dence between self-reportect criminal be- havior ant] offense of record. One has to make coding decisions about the place- ment of reported crimes in the available legal offense categories, which is likely to lead to more misclassifications than nec- essary. Unfortunately, the sampling process is less amenable to direct modification than some of the other problems with official- record data. As Hindelang, Hirschi, and Weis (1981:21) have concluded, "there is no practical way to directly modify or refine official statistics to compensate for uncliscoverecI, unreported, and unsolved crimes." However, one can study the sampling process more carefully to ascer- tain its relationship to criminal behavior; for example, clifferential arrest probabili- ties can be estimated from self-reports of crime and official records of individual arrest for a subset of those crimes, and changes can be made in administrative and technical aspects of official-recorc3 data collection, as well as in the self- report method. These kinds of changes and research are what may be necessary to enhance the validity and reliability of both official and self-report measures, to improve convergence on parameter CRIMINAL CAREERS AND CAREER CRIMINALS estimates and their correlates, and, ulti- mately, to create the best possible combi- nation of measures of individual offencl- ing patterns. RESEARCH STRATEGIES TO IMPROVE MEASUREMENT Even though official records and self- reports have relatively good general va- lidity and reliability as measures of crime, it is clear that both approaches to mea- surement are imperfect ant! have unique problems and limitations. Together, they may provide the most adequate, comple- mentary description of individual offencl- ing patterns in criminal career research, particularly if they are both improved and refined to maximize reciprocal validity ancI reliability and, hence, the prospect for even more convergence than already exists on the representations of the basic characteristics of crime and criminals. Given what we know, what can be done to use official and self-report measures in ways that buiTcI on their complementary characteristics to take advantage of their unique strengths and compensate for their weaknesses? What kind of research might improve our knowledge of validity and reliability problems and their effects on estimates of parameters and correlates anal, ultimately, the convergence or dis- crepancy of official ancl self-report mea- sures of crime? Many things that can or should be done to improve measurement are simply good research practice. However, there are also many issues, questions, and prob- lems in both official and self-report mea- surement that require the illumination of empirical research. Four research issues seem particularly important for improv- ing the measurement of individual of- fencling patterns. First is time coding, both in official records (various transac- tion dates) and in self-reports (recall-pe- riod bounclaries). Second, the isomorph

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS ism of the domains of criminal behavior contained in official records and repre- sented in self-report measures is critical in a variety of ways, ranging from item selection and reverse record checks to attempts to calibrate unknown self-report- ed crimes to officially known offenses. Third, the method of measurement is probably more critical in research on se- rious, chronic offenders than on less pred- atory samples of offenders, ant] the deter- mination of the most effective way to collect valid and reliable data from these kinds of respondents is critical. Fourth, the potential respondent effects that may be attributed to drug and alcohol abuse, violent criminal behavior, official status as a serious offender, intellectual compe- tence, and race (anc] their interactions) need to be investigated to determine their actual contribution to response bias. Given the nascent status of research on the measurement of criminal careers, the development of Tong-term research strat- egies is a necessary and logical first step toward achieving greater convergence in the estimates of career parameters produced by different measurement approaches. A strategy, as a "plan" to achieve some end, establishes the bound- aries within which more concrete means to that end will be implemented. In the context of this paper, the goal is to achieve synthesis in estimating the pa- rameters of criminal careers. To this end, six "research strategies," of the many pos- sible, are discussed, each of which ad- dresses one or more of the above-ident- ified research issues, as well as other problems, and each of which includes suggestions for the kinds of research and studies that might be incorporated into the Tonger-range research agenda implied by the strategies.23 23This coverage, admittedly, is not comprehen- sive. Nor does it claim the selection of the four "most important" research issues, or that the six 37 The research strategies focus on issues in empirical research on the measure- ment of criminal career parameters and their correlates. It is assumed that the research on measurement issues will lead eventually to the kinds of improvements in criminal career research that will en- hance the convergence of estimates and make it more useful for theory, policy, and programs. 1. Research on the measurement of criminal careers should include a com- prehensive, quantitative meta-anaZysis of the relationship between research meth- ods andfindings on parameters and cor- reZates. Since research on criminal careers is a relatively new area of concentrated re- search effort and work on the measure- ment of criminal careers is rudimentary, organizing and synthesizing the relevant research literature is an essential step in building a better understanding of the effects of research methods on estimates of prevalence, incidence, and correlates of serious, chronic criminal behavior. This type of analysis, whereby published estimates of crime are systematically coded and analyzed statistically, facili- tates the investigation of the effects of research methods (independent vari- ables) on, for example, estimates of indi- vidual offending rates (dependent vari- able). By comparing estimates produced by different methods, one can assess con- vergence and discrepancy (Campbell and Fiske, 1959; Magnusson, 1966) and iden- tify potential sources of error that can be attributed to differences in measurement approaches (Weis and Bridges, 1983: 35 391. Treating important methodological characteristics of studies as data (see Crain and Mahard, 1983; Bridges, Weis, . research strategies are the only or preferred ways to address these measurement topics.

38 and Crutchfield, 1984), one can examine statistical relationships between methocI- ological stucly characteristics anc3 re- portec] findings on criminal career charac- teristics, such as involvement in the serious crimes of robbery, assault, and burglary. For example, by including mea- surement approaches used in a study as a ciata point (self-report, official records, ob- servation, informant), one can assess the association between type of measure anc3 estimates of prevalence, incidence, and correlates across the sample of studies to determine if there are effects by measure- ment method. This type of of meta-anal- ysis has been used extensively in ecluca- tional research, particularly in reviewing and synthesizing experimental research (see Hunter, Schmidt, ant] Jackson, 19821. These types of meta-analyses are amena- ble to relatively sophisticates! statistical analysis, including regression (e.g., TittIe, Villemez, and Smith, 1978), Tog linear, anc3 structural equation techniques. Thus, the potential exists for rigorous synthetic analysis of the relevant litera- ture on measurement in criminal career research. There have been a few recent applications of the approach in criminol- ogy. Both TittIe, Villemez, anc3 Smith (1978) and Braithwaite (1981) performed quantitative meta-analyses of the litera- ture on the relationship between social class and crime, treating the published associations as clependent variables and characteristics of the studies as inclepen- dent variables. A current effort (Weis anc3 Bridges, 1983:35~1, 49~51; Bridges, Weis, and Crutchfielc3, 1984) that focuses on improving the measurement of violent behavior has direct relevance to measure ment issues in criminal career research; it includes as one of four major aims a com- prehensive, systematic meta-analysis of Me violence literature. The findings anc3 methods of major studies of violence are being analyzed to identify methoclologi- cal sources of convergence and cliscrep CRIMINAL CAREERS AND CAREER CRIMINALS ancy in estimates of prevalence, inci- clence, and correlates of violence. 2. Research on the measurement of criminal careers should attempt to spec- ify anct estimate the direction and mag- nitude of response effects in seZf-report measures. No research has focused on response effects on self-reports of criminal behav- for, especially among samples that in- clude serious, chronic offenders, whether juvenile or adult. There are isolated ex- amples of research on one or two sources of response bias in the criminological literature (Hindelang, Hirschi, and Weis, 1981; Chaiken and Chaiken, 1982), but only as secondary considerations within broader research efforts. The research on survey methodology includes many stud- ies on response effects and many lists of practical suggestions for reducing them in survey practice (DeLamater, 1982:38~5; Sudman and Bradburn, 1982:3~51, 71~3, 207-208, 22~231; van der Zou- wen and Dijkstra, 1982:220~. For exam- ple, regarding asking threatening ques- tions about behavior, such as criminal activity, the recommendations include, among many more: using open-ended questions; assuring the respondent of an- onymity; using competent interviewers; placing the threatening questions toward the middle ofthe questionnaire; conduct- ing pilot studies when the target popula- tion is unique in identifiable ways; using memory aids, such as cued recall; con- structing multiple-item measures; mea- suring social desirability, sensitivity, and other attitudes that may affect response; formulating specific questions; using ap- propriate recall periods; using longer rather than terse or incomplete questions; phrasing questions in clear, simple lan- guage; deliberately loading questions; embedding the most threatening ques- tions first in a list to maximize validity on later questions of interest; making reli

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS ability checks; providing full information in introductions; and constructing a Togi- cal order of questions. Many of these suggestions, if imple- mented in survey-research practice, will probably reduce response effects. But the problem is that the recommendations come out of research on behaviors and samples that are very different from what one finds in criminology, and there re- mains the need to identify the sources of response bias to estimate direction and magnitude ancl, ultimately, correct for them in analysis. Fortunately, a model clevelopect by van der Zouwen and DiJkstra (1982:22~ 223) for estimating and correcting response effects could facilitate this kind of re- search. The mode] specifies the relation- ship between responses and the charac- teristics of the task (e.g., anonymity), interviewer (e.g., race), and respondent, as well as the conditions for the occur- rence of response effects (e.g., social de- sirability). Collecting the necessary data to test this mode} on a sample that in- clucles both juvenile and aclult serious, chronic offenders could go far in improv- ing the unclerstanding and correction of response effects in self-reports of criminal behavior and, therefore, the utility of the slate for theory, policy, and program pur poses. 3. Research on the measurement of criminal careers should investigate the effects on response bias of respondent characteristics, specificaZly official crimi- nal status, drug and alcohol abuse, vio- [ence, intellectual competence, and race. Contrary to findings in survey-metho- c30Iogy research that respondent effects are not substantial in typical research set- tings with normal samples, evidence in the criminology literature suggests that in self-report surveys of criminal behavior among samples inclucling juvenile and aclult offenders, some respondent charac 39 teristics are sources and others are poten- tial sources of response bias. In the Seat- tle youth study, the combination of being a mate, black, official clelinquent, ancT having low general knowledge was re- lated to apparent invalidity of responses to self-report items and other etiological questions (Hindelang, Hirschi, and Weis, 19811. In the RancI inmate surveys, the answers to questionnaires among blacks, the less educated, and drug-alcohol abus- ers were much more internally inconsis- tent and unreliable, which apparently suppresses estimates of individual crimi- nal involvement. And the research in cog- nitive psychology suggests that respon- dents who engage in violent acts or commit crimes while intoxicated on drugs will have impaired memory of the behavior and event (Birnbuam and Parker, 1977; Loftus, 19801. The role of drug abuse as a source of response bias seems particularly significant among seri- ous, chronic offenders, the majority of whom have long-duration, heavy drug- use histories that are related to their in- volvement in other kinds of crime. Studying the effects of these respon- dent characteristics on measurement, specifically on the validity and reliability of responses about criminal behavior, is not the way they are usually examined. Howeverj it seems that the potential for substantial response bias that is attribut- able to these respondent characteristics warrants serious research consideration. What are their effects on answers to self- report items? How are the effects real- ized? What are the origins and processes that lead to differential response bias on the basis of these personal attributes? Answers to these questions may be difficult to get, but a few ways promise at least some illumination of the questions. First, in data sets many of these respon- dent characteristics are represented in the sample (e.g., Seattle youth study, Rand inmate surveys, Philadelphia co

40 hort study, National Youth Survey, Cambridge Study in Delinquent Devel- opment), which would allow reanalyses of vaTiclity and reliability taking them into account as covariates.24 For example, if one hypothesizes Tower estimates of indi- vidual offending among heavy drug users because of memory impairment, compar- ing their self-report estimates and validity and reliability scores with those of other respondents should inform the question. Obviously, more complicated multivari- ate analyses can be performed on as-many of these respondent and other character- istics as are available in the respective data set. Second, the experimental re- search that addresses those personal at- tributes that affect memory could be rep- licated with expanded samples that include serious, chronic offenders. This would be a good test and potential confir- mation of the hypothesized effects and, therefore, a source of data on the extent of memory decrement that can be attributed to drug histories, violent past, intellectual competence, and so on. Third, given the central role that drug and alcohol abuse plays at the onset and in the development of serious criminal careers and the prob- able powerful effect it has on memory, research focusing on drugs and crime seems warranted. This is not to suggest that the "relationship" between the two needs to be specified better. Rather, the role of drugs in and during the commis- sion of crimes and the effect on perceiv- ing, encoding, storing, and retrieving the details, and even the essentials, of the event need to be investigated. For exam- ples: What are the effects on recall and response accuracy of respondent intoxica 24These and similar kinds of analyses are cur- rently under way at the University of Washington as part of the reanalysis of data sets on violence in the project "Improving the Measurement of Violent Behavior," sponsored by the National Institute of Mental Finale, Center for the Study of Antisocial and Violent Behavior (Weis and Bridges, 1983). CRIMINAL CAREERS AND CAREER CRIMINALS lion, nodding, or withdrawal is there uniform impairment or enhanced state- dependent retrieval of information? What kinds of drugs are used during the com- mission of what kinds of crimes? How much is used per criminal event? Why are drugs used? What is the correspon- dence between self-reports of drug-crime involvement and official accounts? Fours, a study could be launched with a simple factorial design that included the necessary sample size and variation on respondent characteristics-official crim- inal-noncriminal, drug user-nonuser, competent-less competent, black-white, violent-less violent that would allow a simple assessment of independent and interactive respondent effects. Compari- sons on self-reports of individual offend- ing, drug use, official records, and validity and reliability tests could show the hy- pothesized effects. 4. Research on the measurement of criminal careers should include experi- mentaZ research on the valictity anc] reli- ability of measurement methods. The evidence that exists on the differ- ential validity and reliability of official and self-report methods of measuring crime comes from research on juvenile, general-population samples (e.g., Hinde- lang, Hirschi, and Weis, 1981~. There is a clear need for research on the compara- tive efficacy of measurement methods for a sample (or samples) that includes adult and serious, chronic offenders. If, indeed, these types of offenders may be more likely to give inaccurate responses, we need to verify that, but, more importantly, we need to identify under what condi- tions response bias is most likely to occur and in what direction and magnitude. The best way to address this research need is probably with an experimental study that compares two or more ap- proaches to measuring criminal behavior on the same sample. The Seattle youth

ISSUES IN THE MEASUREMENT OF CRIMINaL CAREERS study (Hindelang, Hirschi, and Weis, 1981) and an extencled replication study that focuses on very similar issues in measuring violence among a stratified random sample of criminals, mental pa- tients, and the general population (Weis and Bridges, 1983) are available models for the suggested research. In fact, the latter sample undoubtedly includes a number of serious, chronic offenders in the institutionalized juvenile- and adult- offender subsamples. What is proposed is that a sufficiently large sample of juvenile and aclult seri- ous, chronic offenders be drawn for studying within an experimental design the comparative measurement properties of two self-report methods (the face-to- face interview and the nonanonymous questionnaire) and offlcial-record clata. Randomization would be built into the design, each respondent would partici- pate individually (compared with the group administration in the Ranct inmate surveys, for example; Chaiken and Chaiken, 1982), and all the suggestions listed above to reduce response effects wouIct be appropriately taken into ac- count. The face-to-face interview would be the focus of the research, for a number of reasons. First, the information cannot be anonymous for research purposes, and besides anonymity does not seem to af- fect responses to questions about criminal behavior (Hindelang, Hirschi, ant] Weis, 19811. Second, there are more nonre- sponses and missing data on question- naires than in interviews (Chaiken and Chaiken, 1982; Suciman and Bradburn, 19821. Third, available research evidence suggests that the face-to-face interview may be the most effective method of acI- ministratior~ for these types of respon- dents (Bra(lburn and Suciman, 1979; Marquis anct Ebener, 1981:viii). For ex- ample, Hindelang, Hirschi, and Weis (1981) discovered that among those re 41 spondents with the highest rate of under- reporting- black, male official delin- quents-reporting was most accurate during the face-to-face interview. Fourth, an interview setting, with a competent, experienced interviewer, provides a greater clegree of control and flexibility in eliciting information from the respon- dent. For example, to cue recall, test va- liclity of answers, match self-reported crimes to offenses of record, and establish the perception of a "knowing" inter- viewer, it is recommended that the of- ficial record be used (physically, if appro- priate) as an integral part ofthe interview. It can set up "guilty knowledge" ques- tions, i.e., the interviewer refers to of- fenses of record, for example, three bur- glaries, and asks about pertinent vaTi- dating information: Of the three burglar- ies you were arrested for, how many were of houses or apartments? These kincis of detailed data are extremely difficult to obtain in other ways. This approach can also be user! in comparisons of recogni- tion and cuecI-recall memory of offenses of record. A "creep probe" would also be part of the interview for four purposes: (1) to collect cletailecl information on offenses to ascertain face validity and general accu- racy of answers; (2) to gather data about other important aspects of the criminal event (e.g., intent, motive, number of crime partners, relationship to victim) that could be comparer! with official dep- osition statements for crimes that become offenses of record; this would allow a crude assessment of the disparity be- tween a respondent's memory for details of criminal events and the official version; (3) probing might also be necessary for these kinds of respondents simply to get a relatively normal response pattern (Erick- son and Empey, 19631; and (4) as sug- gestec! by Hindelang, Hirschi, and Weis (1981), to adjust scores on delinquency items based on corrections made in initial

42 reports as a consequence of probing. The nonanonymous questionnaire ant] a fol- low-up ranclomized response questioning or a real or simulated "lie detector" inter- view woulc] be included in the design to test these approaches on this type of sam- ple and provide potential alternatives for comparisons; it is less expensive to give respondents a checklist, but the data may not be as complete or useful. Due to the nature of the sample, this kind of experimental design couIc3 also facilitate a more rigorous assessment of the effects of the apparently important respondent characteristics on responses since serious, chronic offenders personify those characteristics. In general, this type of experimental study is the preferred way to determine the best way to collect valid and reliable data from serious, chronic offenders and to validate recipro- cally data from official records. Other- wise, we are simply guessing, on the basis of scarce, ancillary evidence, about the most effective method of measure- ment. This is certainly not a firm founda- tion on which to build theory, make pol- icy, or c resign programs. 5. Research on the measurement of criminal careers should treat the im- provement of measurement as a research and development enterprise. For a variety of reasons, the preferred research clesign in criminal career re- search is the prospective, longitudinal study. Its primary purposes are to observe changes in behavior and correlates in a cohort. However, there is another pur- pose that should be considered, and that is incorporating into the study design a research and development component on measurement methods. This could be done within an existing longitudinal study or built into one in the future. There are a number of reasons for pro- posing that instrument <levelopment and vali~lity-reliability research be incluclecT CRIMINAL CAREERS AND CAREER CRIMINALS in longitudinal studies. First, we simply need to learn more about measuring crim- inal behavior over time. Knowledge is inadequate anti, until necessary improve- ments are made and we learn more, re- search is hanclicappe(l. Different measurement approaches are more or less effective at clifferent stages of career development, and some are more rudimentary than others. For example, official records are not very useful as in- cTicators of anything for young children, simply because children do not engage in the criminal behaviors that get them into trouble with the law. On the other hancl, direct observations and informant reports (e.g., teachers, parents, peers) are useful approaches in gaining some sense oftrou- blesome behavior among children that might have substantial predictive validity (Farrington, 1982~. The application of sur- vey methods to samples of young chil- dren is a good example of a standard method of measurement being at a rudi- mentary stage of development, but hav- ing the possibility of improvement if treated as a research and development task incorporated into a longitudinal study design. In fact, self-report surveys of"deviant" behavior and attitudes in a cohort of first graders were defined a priori as a research and development task within a larger prospective longitudinal project on delinquency prevention (Weis, Janvier, and Hawkins, 19821. Each ad- ministration of the survey was used to modify and refine the original instru- ment, increasing internal consistency, variance on items, and efficiency of ad- ministration (Weis and Worsley, 1983~. Validity and reliability testing should also be treated as a dynamic process that is built into the research design, rather than something that is attended to after all the data are in and analysis is under way. One should expect validity and reliability to change over time as do the respon- dents, measures, behaviors, and corre

ISSUES IN THE MEASUREMENT OF CRIMINaL CAREERS 43 lates. And, like the measures, improving record check in which the response inter the accuracy and quality of the data vat (the time elapsed between the offense shouIct be a "given" as a research objec- of record and the response to questions five and constantly monitored. For exam- about it) is sufficiently Tong that memory pie, what is the differential validity over may be faulty or at least not as accurate as time of responses based on recognition it could be. versus recall memory and how floes it In sum, treating the improvement of effect estimates of criminalcareerparam- measuring criminal behavior as a re eters? Or, does the magnitude of re- search and development task, incorpo sponse bias by race vary over time, and rated in a prospective longitudinal study of the development of criminal careers, wouIc3 improve substantially the validity and reliability of measures and, ulti mately, the ability to describe criminal career parameters and correlates. with what implications for measurement? If the improvement of measures is treated as a research and clevelopment task, it also means that one can "tinker" or try to test different techniques as the project unfolds. For example, respon- clents could be trainee! to fill out ledger- type diaries of their activities and, given the design, monitored in the process. One couIcT also use different types of response categories, recall periods, question spec- ificity, ant! question length in the survey instrument, all with the objective of im- proving response accuracy over time. Given a cohort sample, the potential ef- fects of"small changes," such as those in question worcling, couIcl be clirectly and accurately assessed. And recall periods could be "bounded" from wave to wave of data collection, making the analysis of temporal changes both easier and more accurate. The time between waves could be varied, covering shorter and longer recall periods, perhaps rotating two or more subsamples to control for testing effects. Finally, one would have Me unique opportunity to observe the development of official records as well. If the research component is designee! so that data are collected at relatively frequent intervals from both respondents and records, the capacity to assess reciprocal validity and subjects such as clifferential arrest proba- bilities is enhanced substantially. A rela- tively contemporaneous record check is bound to be more accurate and useful for "triangulation" of results than a reverse 6. Research on the measurement of criminal careers should assess the isomorphism of domains of seZf-reported crimes and official offenses. There is insufficient and inadequate information in official records regarding the characteristics of offenders, as well as the behavioral referents of the legal cate- gories of offense. Offenses of record are the organizational products of a process that transforms initial criminal behavioral events that were detected into sanctioned ancl, therefore, recorcled official offenses (Black, 19701. The problem for research on the measurement of criminal careers is the disparity between the criminal be- havior and the official offense. To the extent that there is not a "good fit" behav- iorally or conceptually between the two, the implications reverberate throughout research on valiclity and reliability of both self report and official measures. The disparity between what an of- fencler actually did and what is repre- sented in the record can affect all compar- ative validity tests, including reverse record checks and, ultimately, the conver- gence or discrepancy of estimates of pa- rameters and correlates. This is not a trivial or esoteric problem, but a very practical problem in comparing the two methods of measurement. If the units of

44 comparison are not the same, compari- sons are obviated. We have seen these difficulties in the reverse-record-check comparisons ofthe Seattle youth study, in which self-reported behaviors were grouped within offense categories to en- able comparisons, rather than on the basis of official offenses (lIindelang, Hirschi, and Weis, 1981~. To the extent that there is meaningful disparity, extrapolating from offenses known to total crime, either within a ju- risdiction or, more unreliably, across ju- risdictions, may be more difficult to do with optimal precision. Even the appar- ently simple identification of a self-report item that fits an offense category may not be that easy (Hindelang, Hirschi, and Weis, 1981:171-1801. In short, the dispar- ity between criminal behavior and legal classification is a critical measurement problem. To determine the degree of isomorph- ism between self-report and official do- mains, one should investigate the struc- ture ant] content of official records (Marquis and Ebener, 1981:viii). More research is needed on the characteristics of official-record data: What is the domain of criminal behaviors that official records represent? To what extent do the do- mains of official and self-report data over- lap? Is the apparent multidimensionaTity of criminal behavior in self-report data replicated in official-record data? An- swers to some of these questions can be gained through the usual dimensional- izing procedures, for example, factor and cluster analysis of official data. Others require more time-consuming and dif- ficult data collection and analysis. To get closer to the domain of criminal behaviors circumscribed by official records, one needs to delve deeper than the legal cat- egories recorded in the active records of the police department or court. One should go to the offender's case file for the much more detailed information given in depositions by the victim, witnesses, po CRIMINAL CAREERS AND CAREER CRIMINALS lice, and offender. This should enhance the agreement of the two data sources, provide more accurate information from which to select self-report items, increase correlational validity coefficients, make reverse record checks more precise, and so on. In short, more research on the measurement properties of official-record data is needed, particularly on the degree to which the conceptual and behavioral domains of self-report and official mea- sures are isomorphic. To the extent that they are not, the complementarily of the two measures is compromised. CONCLUSION Theory, policy, and programs on crim- inal careers depend on accurate informa- tion about the characteristics, distribu- tion, and correlates of criminal behavior over time. Fortunately, the two most via- ble approaches to measuring the parame- ters and correlates of criminal careers, self-reports and official records, generally describe the same essential features of crime and criminals. There is much more convergence than discrepancy in their representations of the phenomena. How- ever, research on measurement proper- ties, particularly validity and reliability, shows that both approaches to measuring crime are still plagued by a variety of problems and weaknesses. The improve- ment of self-report and official measures should be a fixed objective and an ongo- ing enterprise. The goal should be to maximize the precision and utility of measures of the parameters and corre- lates of individual offending patterns over time, particularly for serious and violent crimes. In so doing, the refined measure- ment approaches may generate more sim- ilar estimates of parameters but, more likely and more usefully, should produce even more convergent estimates of the correlates of offenses and offenders. With this knowledge about systematic varia- tion, adjustments and fine-tuning are

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS more within the realm of possibility, if not on the basis of current knowledge at least on the basis of future research. A number of research strategies have been described that could improve research on the measurement of criminal careers and, therefore, our ability to understand and address the problem of serious, chronic criminal behavior. REFERENCES Akers, Ronald L. 1964 Socio-economic status and delinquent be- havior. A retest. Journal of Research in Crime and Delinquency 1:38-46. Bachman, Jerald G., O'Malley, Patrick, and Johnston, Jerome 1978 Youth in Transition, Vol. Vl: Adolescence to Adulthood: Change and Stability in the lives of Young Men. Institute for Social Re- search. Ann Arbor: University of Michigan. Ball, John C. 1967 The reliability and validity of interview data obtained from 59 narcotics addicts. Ameri- canJournal of Sociology 72:65~654. Belson, William 1968 Development of a Procedure for Eliciting Information from Boys about the Nature and Extent of Their Stealing. Survey Re- search Centre. London: London School of Economics. Birnbuam, Isabel M., and Parker, Elizabeth S. 1977 Acute effects of alcohol on storage and re- trieval. Pp. 99-108 in Isabel M. Birnbuam and Elizabeth S. Parker, eds., Alcohol and Human Memory. New York: Ichn Wiley & Sons. Black, Donald J. 1970 Production of crime rates. American Socio- logical Review 35:73~748. Blackmore, Jane 1974 The relationship between self-reported de- linquency and official convictions amongst adolescent boys. British Journal of Cr~mi- nology 14:172~176. Blumstein, Alfred 1982 On the racial disproportionality of United States' prison populations. Journal of Crim- inal Law and Criminology 73:1259-1273. Bower, Gordon H. 1970 Organizational factors in memory. Cognitive Psychology 1:18-46. Bradburn, Norman M., and Danis, Catalina 1984 Potential contributions of cognitive research to survey questionnaire design. Pp. 101-129 45 in Thomas B. Jabine, Miron L. Straf, Judith M. Tanur, and Roger Tourangeau, eds., Cog- nitive Aspects of Survey Methodology. Re- port of the Advanced Research Seminar on Cognitive Aspects of Survey Methodology, Committee on National Statistics, National Research Council. Washington, D.C.: Na- tional Academy Press. Bradburn, Norman M., and Sudman, Seymour 1979 Improving Interview Method and Question- naire Design. San Francisco, Calif.: Jossey- Bass. Braithwaite, John 1981 The myth of social class and criminality re- considered. American Sociological Review 46:36-57. Brenner, Michael 1982 Response effects of"rc~le-restricted" charac- teristics of the interviewer. Pp. 131-165 in Wil Dijkstra and Johannes van der Zouwen, eds., Response Behavior in the Survey-later- view. New York: Academic Press. Bridges, George 1981 Estimating the effects of response errors in self-reports of crime. Pp. 59-76 in James A. Fox, ea., Methods in Quantitative Criminol- ogy. New York: Academic Press. 1983 An empirical study of error in reports of crime and delinquency. In Marvin E. Wolfgang, Robert M. Figlio, and Terence P. Thornberry, From Boy to Man,from Delin- quency to Crime. Chicago, Ill.: University of Chicago Press. Bridges, George, Weis, Joseph G., and Crutchfield, Robert 1984 The Measurement of Violent Behavior: Fac- tors Affecting Estimates of the Prevalence and Correlates of Violent Behavior. Paper presented at the American Sociological As- sociation Meeting, San Antonio, Texas, Au- gust. Bytheway, W. R., and May, D. R. 1971 On fitting the "facts" of social class and criminal behavior: a rejoinder to Box and Ford. Sociological Review 19:585-607. Campbell, Donald T., and Fiske, Donald W. 1959 Convergent and discriminant validation by the multitrait-multimethod matrix. Psycho- logical Bulletin 56:81-105. Cannell, Charles F., and Fowler, Floyd J. 1963 Comparison of a self-enumerative procedure and a personal interview: a validity study. Public Opinion Quarterly 27:25(}264. Cannell, Charles F., and Kahn, Robert L. 1968 Interviewing. Pp. 449-487 in Gardner Lind- zey and Elliott Aronson, eds., The Handbook of Social Psychology. Vol. I. Reading, Mass.: Addison-Wesley.

46 Cannell, Charles F., Marquis, Kent H., and Laurent, Andre 1977 A summary of studies of interviewing meth- odology. Vital and Health Statistics 69:1- 78. Cannell, Charles F., Oksenberg, Lois, and Con- verse, Jean M. 1977 Striving for response accuracy: experiments in new interviewing techniques. Journal of Marketing Research 14:306~315. Cash, William S., and Moss, Abigail J. 1972 Optimum recall period for reporting persons injured in motor vehicle accidents. Vital and Health Statistics 50:1~3. Cermak, L., and Butters, N. 1973 Information processing deficits of alcoholic Korsakoff patients. Quarterly Journal of Studies on Alcohol 34: 111~1132. Chaiken, Jan M., and Chaiken, Mareia R. 1982 Varieties of Criminal Behavior. Santa Moniea, Calif.: Rand Corporation. Chaiken, Jan M., Chaiken, Mareia R., and Peterson, Joyce E. 1982 Varieties of Criminal Behavior: Summary and Policy Implications. Santa Moniea, Calif.: Rand Corporation. Chaiken, Jan M., Chaiken, Mareia R., and Rolph, John E. 1983 Self Reported Commissions of Crimes: Comparison of Two Questionnaire Formats. Santa Moniea, Calif.: Rand Corporation. Chaiken, Jan M., Rolph, John E., and Houehens, R. L 1981 Methods for Estimating Crime Rates of In- dividuals: Executive Summary. Santa Moniea, Calif.: Rand Corporation. Charnbliss, William J., and Nagasawa, Richard H. 1969 On the validity of official statistics- a eom- parative study of white, black, and Japanese high school boys. Journal of Research in Crime and Delinquency 6:71-77. Christie, Nils, Andenaes, Johannes, and SkirbeLk, Sigurd 1965 A study of self-reported crime. Pp. 8~116 in Karl O. Christiansen, ea., Scandinavian Studies in Criminology. Vol. 1. London: Tavistoek. Claney, Kevin, and Gove, Walter 1980 Sex differences in mental illness: an analysis of response bias in self-reports. American Journal of Sociology 80:205-216. Clark, John P., and Tiffl, Larry L. 1966 Polygraph and interview validation of self reported delinquent behavior. American So- ciological Review 31:516-523. Coeozza, Joseph J., and Steadman, Henry J. 1974 Some refinements in the measurement and CRIMINAL CAREERS AND CAREER CRIMINALS prediction of dangerous behavior. American Journal of Psychiatry 131(9~: 1012-1014. Colombotus, John 1969 Personal vs. telephone interviewers' effect on responses. Public Health Report 84: 77 782. Costner, Herbert 1975 The Validity of Indicators. Paper presented at American Sociological Association Meet- ing, San Francisco, California, August. Crain, Robert L., and Mahard, Rita E. 1983 The e£ect of research methodology on de- segregation achievement studies: a meta- analysis. American Journal of Sociology 88 (51:83~854. Curtis, Lynn 1974 Criminal Violence National Patterns and Behaviors. Lexington, Mass.: D. C. [Ieath. Darley, C. F., Tinklenberg, J. R., Roth, W. T., Hollister, L. E., and Atkinson, R. C. 1973 Influence of marijuana on storage and re- trieval processes in memory. Memory and Cognition 1:196-200. DeLamater, John 1982 Response-effeets of question eonter~t. Pp. 1~48 in Wil Dijkstra and Johannes van der Zouwer~, eds., Response Behavior in the Sur- vey-laterview. New York: Academic Press. Dentler, Robert A., and Monroe, Lawrence J. 1961 Social correlates of early adolescent theft. American Sociological Review 26:73~743. Dijkstra, Wil, and van der Zouwen, Johannes 1982 Introduction. Pp. 1-12 in Wil Dijkstra and Johannes van der Zouwen, eds., Response Behavior in the Survey-Interview. New York: Academic Press. Edwards, Allen L. 1957 The Social Desirability Variable in Person- ality Assessment and Research. New York: Dryden. Elliott, Delbert S. 1982 Review essay on Measuring Delinquency. Criminology 20~3~):527-537. Elliott, Delbert S., and Ageton, Suzanne 1980 Reconciling race and class differences in self-reported and o~eial estimates of delin- queney. American Sociological Review 1:95-110. Elliott, Delbert S., and Huizinga, David 1983 Social class and delinquent behavior in a na- tional youth study. Criminology 21: 14~177. Elliott, Delbert S., and Voss, Harwin L. 1974 Delinquency and Dropout. Lexington, Mass.: D. C. Heath. Elmhorn, Kirsten 1965 Study in self report delinquency among school children in Stockholm. Pp. 117-146

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS in Karl O. Christiansen, ea., Scandinavian Studies in Criminology. Vol. 1. London: Tavistock. Ennis, Philip H. 1967 Criminal Victimization in the United States: A Report of a National Survey. National Opinion Research Center. Chicago, Ill.: Uni- versity of Chicago. Erdelyi, Matthew H., and Goldberg, Benjamin 1979 Let's not sweep repression under the rug: toward a cognitive psychology of repression. Pp. 36~402 in John Kihlstrom and Frederick I. Evans, eds., Functional Disor- ders of Memory. Hillsdale, N.~.: Lawrence Erlbaum. Erickson, Maynard 1972 The Changing Relation Between Official and Self-reported Delinquency. Unpub- lished manuscript. University of Arizona, Tucson. Erickson, Maynard, and Empey, LaMar T. 1963 Court records, undetected delinquency and decision-making. Journal of Criminal Law, Criminology, and Police Science 54:456- 469. Farrington, David 1973 Self-reports of deviant behavior: predictive and stable? Journal of Criminal Law and Criminology 64:99-110. 1979 Environmental stress, delinquent behavior, and convictions. Pp. 9~107 in I. G. Sarason and C. D. Spielberger, eds., Stress andAnx- iety. Vol. 6. Washington, D.C.: Hemisphere. 1982 Stepping Stones to Adult Criminal Careers. Paper presented at Conference on the De- velopment of Antisocial and Prosocial Be- havior, Voss, Norway, April. 1983 Executive Summary of Further Analyses of a Longitudinal Survey of Crime and Delin- quency. Final Report to the National Insti- tute of Justice. Washington, D.C.: U.S. De- partment of Justice. Foster, Herbert L. 1974 Ribbin', Jivin', and Playin' the Dozens: The Unrecognized Dilemma of Inner City Schools. Cambridge, Mass.: Ballinger. Fox, Greer L., and Inazu, Judith K. 1980 Patterns and outcomes of mother-daughter communication about sexuality. Journal of Social Issues 36:7-29. Garofalo, James, and Hindlelang, Michael I. 1977 An Introduction to the National Crime Sur- vey. Washington, D.C.: U.S. Department of Justice. Gibson, H. B., Morrison, Sylvia, and West, D. J. 1970 The confession of known offenses in re- sponse to a self-reported delinquency sched 47 ule. British Journal of Criminology 10:277- 280. Glueck, Sheldon, and Glueck, Eleanor T. 1950 Unraveling Juvenile Delinquency. Cam- bridge, Mass.: Harvard University Press. Gold, Martin 1966 Undected delinquent behavior. Journal of Research in Crime and Delinquency 3:27- 46. 1970 Delinquent Behavior in an American City. Belmont, Calif.: Brooks/Cole. Goode, William J., and EIatt, Paul K. 1952 Methods in Social Research. New York: McGraw-Hill. Gordon, Robert A. 1976 Prevalence: the rare datum in delinquency measurement arid its implications for the theory of delinquency. Pp. 201-284 in Mal- colm W. Klein, ea., The Juvenile Justice Sys- tem. Beverly Hills, Calif.: Sage Publications. Gould, Leroy C. 1969 Who defines delinquency: a comparison of self-reported and officially reported indices of delinquency for three racial groups. Social Problems 16:325~336. Gove, Walter 1982 Systemic response bias and characteristics of the respondent. Pp. 167-187 in Wil DiJkstra and lohannes van der Zouwen, eds., Re- sponse Behavior in the Survey-laterview. New York: Academic Press. Greenwood, Peter 1979 Rand Research on Criminal Careers: Pro- gress to Date. Santa Monica, Calif.: Rand Corporation. Hagenaars, Jacques A., and Heinen, Ton G. 1982 Effects of role-independent interviewer characteristics on responses. Pp. 91-130 in Wil Dijkstra and Johannes van der Zouwen, eds., Response Behavior in the Survey-Inter- view. New York: Academic Press. Hamparian, Donna M., Schuster, Richard, Dinitz, Simon, and Conrad, John P. 1978 The Violent Few. Lexington, Mass.: Lexington. Hannan, Michael T. 1971 Aggregation and Disaggregation in Sociol- ogy. Lexington, Mass.: Lexington. Hardt, Robert H., and Peterson-Hardt, Sandra J. 1977 On determining the quality of the delin- quency self-report method. Journal of Re- search in Crime and Delinquency 14:247- 261. Hasher, L., and Chromiak, W. 1977 The processing of frequency information. Journal of Verbal Learning and Verbal Be- havior 16:17~184.

48 Hathaway, Starke R., and Monachesi, Elio D. 1963 Adolescent Personality and Behavior. MMPI Patterns of Normal, Delinquent, Dropout and Other Outcomes. Minneapolis: Univer- sity of Minnesota Press. Havighurst, Robert J. 1962 Growing Up in River City. New York: John Wiley & Sons. Hindelang, Michael I. 1973 Causes of delinquency: a partial replication and extension. Social Problems 20:471~87. 1974 Uniform Crime Reports revisited.Journal of Criminal Justice 2: 1-18. 1978 Race and involvement in common law per- sonal crimes. American Sociological Review 43:9~109. Hindelang, Michael I., Gottiredson, Michael R., and Garofalo, James 1978 Victims of Personal Crime: An Empirical Foundationfor a Theory of Personal Victim- ization. Cambridge, Mass.: Ballinger. Hindelang, Michael I., Hirschi, Travis, and Weis, Joseph G. 1975 Self-Reported Delinquency: Methods and Substance. Proposal submitted to the Na- tional Institute of Mental Health, Depart- ment of Health, Education, and Welfare, MH27778. Department of Sociology, Uni- versity of Washington. 1978 Social Class, Sex, Race and the Discrepancy between Self-Reported and Official Delin- quency. Grant Number MH2277~03. Cen- ter for Studies of Crime and Delinquency, National Institute of Mental Health, Depart- ment of Health, Education, and Welfare. Department of Sociology, University of Washington. 1979 Correlates of delinquency: the illusion of discrepancy between self-report and official measures. American Sociological Review 44:99~1014. 1981 Measuring Delinquency. Beverly Hills, Calif.: Sage Publications. Hirschi, Travis 1969 Causes of Delinquency. Berkeley: Univer- sity of California Press. Hirschi, Travis, and Hindelang, Michael J. 1977 Intelligence and delinquency: a revisionist review. American Sociological Review 42:571~87. Huizinga, David, and Elliott, Delbert S. 1981 A Longitudinal Study of Drug Use and De- linquency in a National Sample of Youth: An Assessment of Causal Order. Boulder, Colo.: Behavioral Research Institute. Hunter, Ian M. L. 1957 Memory. Middlesex, England: Penguin Books. CRIMINAL CAREERS AND CAREER CRIMINALS Hunter, I. E., Schmidt, F. L., and Jackson, G. B. 1982 Meta-Analysis: Cumulating Research Find- ings Across Studies. Beverly Hills, Calif.: Sage Publications. Janson, Carl-Gunnar 1982 Delinquency Among Metropolitan Boys. Department of Sociology. Stockholm: Uni- versity of Stockholm. Jensen, Gary F. 1976 Race, achievement and delinquency: a fur- ther look at delinquency in a birth cohort. American Journal of Sociology 82:379-387. Jones, Ben M., and Jones, Marilyn K. 1977 Alcohol and memory impairment in male and female social drinkers. Pp. 127-140 in Isabel M. Birnbuam, and Elizabeth S. Parker, eds., Alcohol and Human Memory. New York: John Wiley & Sons. Jones, Edward E., and Sigall, Harold 1971 The bogus pipeline: a new paradigm for measuring affect and attitude. Psychological Bulletin 76:349-364. Klatzky, Roberta L. 1975 Human Memory: Structures and Processes. San Francisco, Calif.: W. H. Freeman. Kulik, James, Stein, Kenneth B., and Sarbin, Theodore R. 1968 Disclosure of delinquent behavior under conditions of anonymity and nonanonymity. - Journal of Consulting and Clinical Psychol- ogy 32:506~09. Langan, Patrick A., and Farrington, David P. 1983 Two-track or one-track justice? Some evi- dence from an English longitudinal survey. Journal of Criminal Law and Criminology 74:519-546. Law Enforcement Assistance Administration 1972 San Jose Methods Test of Known Crime Victims. Statistics Technical Report No. 1. Washington, D. C.: U.S. Government Print- ing Office. Liebow, Elliot 1967 Tally's Corner: A Study of Negro Street Corner Men. Boston, Mass.: Little, Brown. Lin, Nan 1976 Foundations of Social Research. New York: McGraw-Hill. LoDcus, Elizabeth F. 1980 Memory. Reading, Mass.: Addison-Wesley. 1982 Memory and its distortions. Pp. 12~154 in A. G. Kraut, ea., The G. Stanley Hall Lecture Series. Washington, D.C.: American Psycho- logical Association. LoDcus, Elizabeth F., and Burns, T. E. 1982 Mental shock can produce retrograde amne- sia. Memory and Cognition 10:318-323. Magnusson, David 1966 Test Theory. Boston, Mass.: Addison-Wesley.

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS Maltz, Michael D. 1975 Crime statistics: a mathematical perspective. Journal of Criminal Justice 3: 177-193. Marquis, Kent H. 1978 Inferring Health Interview Responses from Imperfect Record Checks. Santa Monica, Calif.: Rand Corporation. 1980 Hospital Stay Response Error Estimates for the Health Insurance Study's Dayton Base- line Survey. Santa Monica, Calif.: Rand Cor- poration. Marquis, Kent H., and Ebener, Patricia A. 1981 Quality of Prisoner Self-Reports: Arrest and Conviction Response Errors. Santa Monica, Calif.: Rand Corporation. McDonald, Lynn 1969 Social Class and Delinquency. London: Faber & Faber. Miller, Loren L., and Cornett, Teresa L. 1978 Marijuana: dose elects on pulse rate, subjec- tive estimates of intoxication, free recall and recognition memory. Pharmacology, Bio- chemistry and Behavior 9:57~577. Miller, Walter B. 1958 Lower class culture as a generating milieu of gang delinquency. Journal of Social Issues 14:5-19. Murdock, Bennet B. 1974 Human Memory: Theory and Data. Poto- mac, Md.: Lawrence Erlbaum. Nettler, Gwynn 1974 Explaining Crime. New York: McGraw-Hill. 1978 Explaining Crime. 2d edition. New York: McGraw-Hill. Nisbett, Richard E., and Ross, Lee 1980 Human Interference: Strategies and Short- comings of Social Judgment. Englewood Cliffs, N.J.: Prentice-Hall Nye, F. Ivan 1958 Family Relationships and Delinquent Be- havior. New York: John Wiley & Sons. Parker, Elizabeth S., and Noble, Ernest P. 1977 Alcohol consumption and cognition filing in social drinkers. Journal of Studies on Alco- hol 38: 1229~1232. Penk, W. E., Brown, A. S., Roberts, W. R., Dolan, M. P., Atkins, H. G., and Robinowitz, R. 1981a Visual memory for black and white male heroin and nonheroin drug users. Journal of Abnormal Psychology 90:486~89. 1981bVisual memory of male Hispanic-American heroin addicts. Journal of Consulting and Clinical Psychology 49:771-772. Perry, D. C. 1976 PCP revisited. Clinical Toxicology 9:33 348. Peterson, Mark A., and Braiker, Harriet B. 1980 Doing Crime: A Survey of California Prison 49 Inmates. Santa Monica, Calif.: Rand Corpo- ration. Peterson, Mark A., and Braiker, Harriet B., with Polich, Suzanne M. 1981 Who Commits Crimes: A Survey of Prison Inmates. Cambridge, Mass.: Oelgeschlager, Gunn and Hain. Polk, Kenneth, Frease, Dean, and Richmond, F. Lynn 1974 Social class, school experience and delin- quency. Criminology 12:8~96. Quetelet, Adolphe 1833 Recherches Sur le Penchant au Crime aux Differens Ages. 2d edition. Brussels: Hayez. Reiss, Albert I. 1975 Inappropriate theories and inadequate meth- ods as policy plagues: self-reported delin- quency and the law. Pp. 211-222 in N. J. Demerath III, O. Larsen, and K. F. Schuessler, eds., Social Policy and Sociol- ogy. New York: Academic Press. Reiss, Albert J., and Rhodes, A. Lewis 1961 The distribution of juvenile delinquency in the social class structure. American Sociolog- ical Review 26:72~732. Robinson, W. S. 1950 Ecological correlations and the behavior of individuals. American Sociological Review 15:351 ;357. Savitz, Leonard 1982 Official statistics. Pp. ~15 in Leonard Savitz and Norman Johnston, eds., Contemporary Criminology. New York: John Wiley & Sons. Schuman, Howard, and Presser, Stanley 1981 Questions and Answers in Attitude Surveys. New York: Academic Press. Sellin, Thorsten, and Wolfgang, Marvin E. 1964 The Measurement of Delinquency. New York: John Wiley & Sons. Shannon, Lyle W. 1982 Predicting Adult Criminal Careers from Ju- venile Careers. Iowa Urban Community Re- search Center. Ames: University of Iowa. 1983 The Prediction Problem As It Applies to Delinquency and Crime Control. Iowa Ur- ban Community Research Center. Ames: University of Iowa. Short, James F., Jr., and Nye, F. Ivan 1958 Events of unrecorded juvenile delinquency: tentative conclusions. Journal of Criminal Law and Criminology 49:296~302. Shryock, Henry S., Siegel, Jacob S., and Associates 1975 Methods and Materials of Demography. Rev. edition, Vols. I and II. Washington, D.C.: U. S. Government Printing Office. Siegel, Judith M., and Loftus, Elizabeth F. 1978 Impact of anxiety and life stress upon eye

50 witness testimony. Bulletin of the Psycho- nomic Society 12:479~80. Skogan, Wesley G. 1977 Dimensions of the dark figure of unreported crime. Crime and Delinquency 23:41-50. Sobell, Mark B., and Sobell, Linda C. 1978 Behavioral Treatment of Alcohol Problems: Individualized Therapy and Controlled Drinking. New York: Plenum Press. Strasburg, Peter A. 1978 Violent Delinquents. A Report to the Ford Foundation. New York: Vera Institute of Jus- tice. Sudman, Seymour, and Bradburn, Norman M. 1974 Response Effects in Surveys: A Review and Synthesis. Chicago, Ill.: National Opinion Research Center. 1982 Asking Questions. San Francisco, Calif.: Jos- sey-Bass. Sykes, Thomas M., Roccolo, Thompson, Joann K. 1980 An Analysis of Program Needs of Prison Inmates in Washington State. Olympia, Wash.: Office of Research. Talland, George A. 1968 Disorders of Memory and Learning. Har- mondsworth, England: Penguin. Thornberry, Terence, and Farnworth, Margaret 1982 Social correlates of criminal involvement. American Sociological Review 47:50 518. Tittle, Charles R., Villemez, Wayne J., and Smith, Douglas A. 1978 The myth of social class and criminality: an empirical assessment of the empirical evi- dence. American Sociological Review 43: 643 656. Tourangeau, Roger 1984 Cognitive sciences and survey methods. Pp. 7~100 in Thomas B. Jabine, Miron L. Straf, Judith M. Tanur, and Roger Tourangeau, eds., Cognitive Aspects of Survey Methodol- ogy. Report of the Advanced Research Sern- inar on Cognitive Aspects of Survey Meth- odology, Committee on National Statistics, National Research Council. Washington, D.C.: National Academy Press. Tracy, Paul E. 1981 Analysis ofthe Incidence and Seriousness of Self-reported Delinquency and Crime. Pa- per presented at the American Society of Criminology Meeting, Washington, D.C., November. Tracy, Paul E., and Fox, James A. 1981 Validation of randomized response. Ameri- can Sociological Review 2:187-200. Robert L., and CRIMINAL CAREERS AND CAREER CRIMINALS Valentine, Bettylou 1978 Hustling and Other Hard Work: Life Styles in the Ghetto. New York: Free Press. van der Zouwen, Johannes, and Dijkstra, Wil 1982 Conclusions. Pp. 209-222 in Wil Dijkstra and Johannes van der Zouwen, eds., Re- sponse Behavior in the Survey-luterview. New York: Academic Press. Voss, Harwin L. 1963 Ethnic differentials in delinquency in Hono- lulu. Journal of Criminal Law, Criminology and Police Science 54:322 327. Warner, Stanley L. 1965 Randomized response: a survey technique for eliminating evasive answer bias. Journal of the American Statistical Association 60:63 69. Weingartner, Herbert, and Parker, Elizabeth S. 1984 Memory Consolidation: Psychobiology of Cognition. Hillsdale, N.J.: Lawrence Erl- baum. Weis, Joseph G. 1981 One More Time: Social Class and Delin- quency. Paper presented at the American Society of Criminology Meeting, Washing- ton, D.C., November. 1983a Crime statistics: reporting systems and meth- ods. Pp. 378~392 in Stanford H. Kadish, ea., Encyclopedia of Cr~me and Justice. New York: Free Press. 1983b Delinquency from the self-report perspec- tive. Pp. 73 88 in Raymond R. Corrado, Marc LeBlanc, and Jean Trepanier, eds., Current Issues in Juvenile Justice. Toronto: Butter- worths. 1986 The elusive correlation: social class and crime. in Michael R. GottEredson, ea., Posi- tive Cr7minology: Essays in Honor of Michael J. Hindelang. Beverly [Iills, Calif.: Sage. Weis, Joseph G., and Bridges, George S. 1983 Improving the Measurement of Violent Be- havior. Proposal to Center for the Study of Antisocial and Violent Behavior, National Institute of Mental Health. Department of Sociology, University of Washington. 1984 A Meta-analysis of Correlates of Violence. Paper presented at American Society of Criminology Meeting, Cincinnati, Ohio, No- vember. Weis, Joseph G., and van Alstyne, David 1979 The Measurement of Delinquency by the Randomized Response Method. Paper pre- sented at the American Society of Criminol- ogy Meeting, Philadelphia, Pennsylvania, November.

ISSUES IN THE MEASUREMENT OF CRIMINAL CAREERS Weis, Joseph G., Janvier, Richard L., and Hawkins, J. David 1982 Delinquent Behavior and Its Prevention: Overview. Washington, D.C.: Office of Juve- nile Justice and Delinquency Prevention. Weis, Joseph G., and Worsley, Katherine 1983 Measuring Deviant Behavior Among Chil- dren. Washington, D.C.: Office of Juvenile Justice and Delinquency Prevention. West, Donald J. 1973 Who Becomes Delinquent? London: Heine- mann. West, Donald J., and Farrington, David P. 1977 The Delinquent Way of Life. London: Heinemann. Wetzel, C. Douglas, Janowsky, David S., and Clapton, Paul 1982 Remote memory during marijuana in 51 toxication. Psychopharmacology 76:278- 281. Williams, Jay R., and Gold, Martin 1972 From delinquent behavior to official delin- quency. Social Problems 20:20~229. Wolfgang, Marvin E., and Ferracuti, Franco 1967 The Subculture of Violence: Toward an In- tegrated Theory in Criminology. London: Tavistock. Wolfgang, Marvin E., Figlio, Robert M., and Sellin, Thorsten 1972 Delinquency in a Birth Cohort. Chicago, Ill.: University of Chicago Press. Zill, Nicholas, Peterson, James L., and Moore, Kristin A. 1984 Improving National Statistics on Children, Youth, and Families. Washington, D.C.: Child Trends, Inc.

Next: 2. The Impact of Substance Abuse on Criminal Careers »
Criminal Careers and "Career Criminals,": Volume II Get This Book
×
Buy Paperback | $110.00
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

Volume II takes an in-depth look at the various aspects of criminal careers, including the relationship of alcohol and drug abuse to criminal careers, co-offending influences on criminal careers, issues in the measurement of criminal careers, accuracy of prediction models, and ethical issues in the use of criminal career information in making decisions about offenders.

  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!