Using estimations of the relationships between several sets of good data to infer causation is an old endeavor, of course. The difficulties to which I pointed above are not new. For a little over one hundred years, the social sciences have emerged largely with the mission to solve or circumvent such difficulties. Their efforts have been broadly productive; inferences of causation based on quantitative estimations that respect state-of-the-arts social science rules are more trustworthy than practically everything else. In brief, the more complex the issues under study, the more back-of-the-envelop calculations suffer in comparison with modern social science methodology. Issue of comparative immigrant vs native born criminality are pretty much at a level of complexity for which those methods were developed.

The rules of social science include an obligation to publish in scholarly journals where the findings will be subjected both to pre-publication and to post-publication critical assessment. (See my didactic essay on scholarly submission: “What’s Peer Review and What it Matters”) Publication in scholarly journals also facilitates eventual attempts at replication with its potential to root out major research-based fallacies. I realize that duplication is a rare event, but the threat of it keeps researchers on their toes. Note: This doesn’t mean that the degree of confidence one should award to serious social science products should be high, in absolute terms. There is a difference in practice between, “bad,” and “very bad.” Also, in spite of formidable recent successful hoaxes against pseudo-journals, some disciplines hold the line, including left-leaning Sociology. (Read Gabriel Rossman’s “Sokal to the Nth Degree” in the November 8th-11th issue of the Weekly Standard.)

As it happens, there is a recent study that addresses the topic of immigration and criminality that fulfills good social science criteria. It’s Michael T. Light and Ty Miller, “Does Undocumented Immigration Increase Violent Crime?” published in Criminology, March 3, 2018. The study relies on data from all 50 states plus the District of Columbia, collected from 1990 to 2014, inclusive. Its design is reasonable; it allows for the observation of change over time in the relationships of interest. The 24-year period of observation is a convenience sample of any longer period one would prefer, but it’s not known to what extent it is representative of a longer period. This is a common limitation on interpretation. Twenty-four years of observation is a lot better than one year for the purpose of generalization though. The estimation methods used in their study to express the relationship between numbers of illegal immigrants, on the one hand, and several well accepted measures of serious criminality, on the other, are up-to date. The same methods allow for the elimination of alternative formulations – that is, they allow for “controlling” for variables other than the main variable of interest, the number of illegal immigrants. The article contains a useful and thorough review of the literature. It’s written with remarkable clarity, given the inherent complexity of the endeavor it describes.

The study gives a straightforward answer to the straightforward question it poses:

The increase in the number of illegal immigrants (in the US) is associated with a decrease in serious crime.

The authors dispose fairly well of an interpretation of these counter-intuitive findings based on the idea that more illegals results in less crime reporting in the relevant populations, rather than in actual decrease in crime. This explanation would make their startling main finding practically spurious.

It is not equally clear to me that the authors have disposed completely of the hypothesis that an influx of illegal immigrants is causally linked to stepped-up law enforcement, and only thence to crime reduction. This formulation has important policy implications. It says: Illegal immigration does not increase crime, provided you do what needs to be done about it. And, you may be lucky and overshoot your mark. Ideally, I would  have liked to see a measure of cost per some unit of crime reduction included in the estimation models.

It’s unfortunate for my purpose that this study focuses on illegal immigration specifically, since my own primary interest is in legal immigration. It matters little in the end because a secondary analysis within this study indicates that increase in legal immigration (considered separately from illegal immigration) is also associated with a decrease in serious crime.

As quantitative social scientists are unfortunately inclined to do, Light and Millet give us a literal expression of their main finding, like this (I think this practice, of making findings shout instead of whispering should be heavily taxed.):

A one-unit increase in the proportion of the population that is undocumented corresponds with a 12 percent decrease in violent crime.

We don’t have to take literally this metric wording and the causality it suggests. For example, there is no need to believe that if enough additional illegal immigrants enter the US, at some point, serious crime will disappear completely. It’s enough to acknowledge that a very good study on the relationship between immigration and serious crime leaves little room for the possibility that the more of the one, the more of the other.

Note, however, that the Light and Miller findings do not exclude this formulation completely. It’s possible that in some states an increase in illegal immigration is associated with a surfeit of serious crime. It’s possible even that for all states, but for a brief period, an increase in illegal immigration is quickly followed by a rise in serious crime. Because of these possibilities, the public perception and the startling results of this study may well be compatible. These would probably not be casually detected. Few regular observers, be they politicians, journalists, or public servants are likely to have a clear view of events in 51 separate entities sustained for twenty-four years. Intelligent, rigorous minded observers may be right about what they know and drastically wrong about what they have not studied through hard facts.

Light’s and Miller’s is a classically good article. It’s thorough without sacrificing detail; it offers a good quality and a useful review of the relevant sociological literature; it’s tightly reasoned. The estimations it reports on seem impeccable. I think Light and Miller is the standard against which all reports on the relationship between immigration and crime should now be assessed. Practically, this article should contribute to switching the burden of proof: Although they are less educated, poorer, and younger on average than the native-born population, immigrants appear to commit less serious crime than the latter. More surprisingly to some, illegal immigrants, who begin their American career by demonstrating their willingness to violate American law, do not appear to be prone to criminal violence.  But, as I have mentioned before, an illegal status is a strong incentive to keep one’s nose clean.

I am not proposing here that Light and Miller’s article should forever block the progress of more conventional ideas  to the effect that more immigration is associated with more crime. All it would take would be a single study of similar quality to overturn this remarkable study’s findings. In the meantime, it would be reasonable to shift the burden of proof  away from where it has implicitly stood: immigrants tend to be criminal.

[Editor’s note: in case you missed it, here is Part 6]