The empirical relationship between immigration and criminal activity — specifically the finding that immigrants have substantially lower incarceration rates than native-born Americans despite demographic characteristics that predict high incarceration, and the theoretical question of what drives this gap.
Male immigrants aged 18–40 have institutionalization rates roughly one-fifth those of native-born men (Butcher and Piehl 2007). This is paradoxical because immigrants have characteristics — low education, low wages, young, male, disproportionately Hispanic — that among the native born are strongly associated with incarceration. Applying native-born institutionalization propensities to immigrants' observed characteristics predicts an immigrant rate double the native average; the actual immigrant rate is one-tenth that prediction.
The gap widened from 1980 to 2000, and more recently arrived immigrant cohorts have even lower relative incarceration rates than earlier cohorts.
Butcher and Piehl (2007) test three explanations:
Deportation: Criminal noncitizens are deported after serving sentences, mechanically removing them from the incarcerated population. Ruled out: naturalized citizens, who cannot be deported, show the same reduction in relative incarceration rates — so deportation cannot be driving the pattern.
Deterrence: The general increase in criminal penalties in the 1980s–90s, plus specific legislation targeting criminal noncitizens (1994 Violent Crime Act, 1996 Anti-Terrorism and Effective Death Penalty Act), deterred immigrant crime. Confirmed: naturalized citizens also improved, consistent with general deterrence. Critically, native-born movers — people living outside their state of birth — also reduced their incarceration rates relative to nonmovers. This suggests that migration (domestic or international) selects for incentive-responsive individuals who respond more to general penalty increases than the average native.
Self-selection: The type of person who immigrates shifted toward those with lower criminal propensity. Confirmed as gradual trend: more recently arrived cohorts in 2000 were less likely to be institutionalized than comparable cohorts in 1990, controlling for time-in-country. However, no discrete break at the 1996–2000 cohort points to gradual multidimensional selection rather than a single-legislation effect.
The standard Borjas-Roy model of immigrant self-selection focuses on a single dimension of skill (wages). But criminal-justice outcomes and wage outcomes need not be correlated in the population: an immigrant may have low wage-relevant skills but high responsiveness to deterrent incentives, or strong social networks and low criminal propensity. Butcher and Piehl interpret their results as evidence of multidimensional selection — migration selects on criminal propensity and incentive-responsiveness independently of wage skill. The key insight is that migration costs decline with the strength of social networks; individuals who immigrate through well-established networks may have particularly good social outcomes even if their wage-relevant skills are limited.
As a validity check, metropolitan areas (MAs) with the largest increases in immigrant share had the largest decreases in crime rates from 1990 to 2000. This rules out the alternative explanation that immigrants escape detection by fleeing across borders — which would predict that areas with more immigrants have higher reported crime.