ables representing location within 100 miles of the U.S.-Mexican border or in Texas (if not on the border) or in another border state (Arizona, California, or New Mexico). The results of these regression analyses revealed that the border effects were all negative for violent crime and largely so for property crime. The conclusion from this analysis is that "if the data suggest anything about the border's impact on crime, it is that crime is lower on average in border areas than in other U.S. cities when the characteristics of the urban population are held constant" (1994:20). Nevertheless, the Commission concedes that a more direct test of the effects of immigration, especially illegal immigration, requires more specific measures.

We undertook a more direct test of immigration effects at the level of standard metropolitan statistical areas (SMSAs) by joining measures of legal and illegal immigration, developed by Bean et al. (1988) and based on a methodology used previously for the nation and states by Warren and Passel (1987), with Uniform Crime Report data collected by the Federal Bureau of Investigation. The estimate of the legally resident noncitizen population was generated using alien registration data for 1980 from the Immigration and Naturalization Service and data on legally admitted aliens for January-March 1980 for SMSAs. This number was then subtracted from the figure for aliens counted in the 1980 census (corrected for nonreporting of country of birth, misreporting of citizenship, and misreporting of nativity) to obtain an estimate for undocumented aliens. Application of these procedures also yielded an estimate of legal noncitizens. The corrections and adjustments used in developing these estimates are described in greater detail by Warren and Passel (1987) and Passel and Woodrow (1984). Although these estimates of illegal immigrants are dependent on the numbers of undocumented immigrants included in the 1980 census, not the number actually present, Bean et al. (1988:39) indicate that "it is likely that the distribution … of the undocumented population not included in the 1980 Census might be similar to the distribution of those included."

We began our analysis with the 47 SMSAs considered by Bean et al. (1988) that are located in the five southwestern states of Arizona, California, Colorado, New Mexico, and Texas, with four border SMSAs in Texas deleted because they include substantial areas and populations in Mexico. Thirteen of these SMSAs were not represented in the Uniform Crime Report data, so 34 SMSAs were ultimately available for our analysis. The three outcome measures of crime included violent crime rates, property crime rates, and total crime rates. Results are presented in Table 9-3.

We initially regressed logged arrest rates on the proportions of the population age 15 and over who were estimated in the 34 SMSAs to be illegal immigrants or noncitizens, black, and living in poverty. The illegal immigrant and noncitizen measures were introduced separately in equations to avoid collinearity problems. Box-Cox transformations were applied in this analysis and we selected the variant that produced the highest R2, which involved taking the logs of