the independent and dependent variables. In no case was the measure of percent illegal immigrants significantly correlated with any of the three outcome crime measures, and the noncitizen measure was significant in only one case. As a further test of potential influence, we created dummy variables, with the SMSAs in the highest quartile on each variable coded as a dummy variable in poisson models of arrest counts. This operationalization was an attempt to capture the effects of more extreme concentrations of effected groups, with, for example, the dummy variable for proportion of illegal immigrants indicating SMSAs with more than 1 percent of the population in this group. This coding yielded more statistically significant but nonetheless inconsistent results, and so we do not present them. There is no consistent or compelling evidence at the SMSA level that immigration causes crime.

We have suggested that an overreliance on prison statistics is a problematic basis for developing our understanding of immigration and crime. It is of further concern, in the political and economic context of cost reduction and shifting described at the outset of this chapter, that the prisons that collect these crime statistics may have an understandable interest in attracting atten-