Refugees and Crime: What is the Evidence?

It is hardly a surprise to anyone that the idea of some sort of a link between refugee resettlement and crime is pervasive. It was a central topic of debate in the 2016 US election. Due to the center stage of this topic, within the first week of President Trump’s term, the US refugee resettlement program was suspended for 120 days. So far in 2018, the US is on track to accept 77% fewer refugees than in 2016. Additionally, citizens from a number of countries are now banned from entering the United States. Chief among the reasons for enacting these policies is the idea that recently resettled refugees may cause unwanted crime in local communities, and that the refugee resettlement system could be used as a pipeline for terrorists to enter the United States.

Now, to be clear, for a long time (see this post from 2015) I’ve pointed out the baseless nature of these claims. This is for a couple of factors: First, of the over 20 million refugees around the world right now, less than 1% of them will ever be resettled in a third country (i.e. not their own country or the country they fled to). Second, refugees are screened by the UNHCR, the US Department of Human Services, the US Department of Homeland Security, and travel on regular passenger airlines. So, while the possibility that criminals or terrorists could use the refugee resettlement system to “sneak” into a country theoretically exists, it is quite unlikely hysterical fear is at all justifiable.

Okay, that is just my opinion, but what does the data say? Is there any evidence that refugees raise crime levels and put settlement areas at risk for terrorist activities? Recent research by Catalina Amuedo-Dorantes, Cynthia Bansak, and Susan Pozo examines this exact question. The paper, entitled, “Refugee Admissions and Public Safety: Are Refugee Settlement Areas More Prone to Crime?” is currently a working paper at the Institute for Labor Economics (IZA). Here is the abstract to the paper:

According to United Nations High Commissioner for Refugees, the number of refugees worldwide rose to 21.3 million in 2015. Yet, resistance to the welcoming of refugees appears to have grown. The possibility that refugees may commit acts of terrorism or engage in criminal behavior has served as fuel for the Trump Administration’s position in 2017. Is there any basis for these fears? We exploit the variation in the geographic and temporal distribution of refugees across U.S. counties to ascertain if there is a link between refugee settlements and local crime rates or terrorist events in the United States. We fail to find any statistically significant evidence of such a connection.

The idea behind the data analysis in the paper is to investigate the relationship between the number of refugees in a given US county within a given year and the number of criminal or terrorist incidents in the same US county within the same year. The authors control for a host of observable factors such as education attainment, employment rates, per capita income, proxies for economic growth, and spending on police protection. The authors also include county and time fixed effects which help control for time- invariant unobservable factors within a given county or year. One of the threats to the ability of the researchers to identify the causal effect of refugee resettlement on crime is the fact that refugees are not exactly randomly resettled across the US.

If refugees were randomly resettled across the US, then we would not need to worry about some sort of omitted variable, reverse causality, or some other form of statistical endogeneity that could bias the results. In reality, however, some counties are more welcoming to refugees or have better access to refugee resettlement service providers and thus refugees are more likely to live in these places. If these places also happen to have higher crime rates then the impact of refugee resettlement on crime rates will be biased upwards.

The authors try to get around this potential problem by implementing a “shift share” instrumental variable that is defined as the share of all US immigrants from a particular country of origin who resided in a given county in 1980 multiplied by the number of refugees from a particular country of origin within a given year.

This instrumental variable is clearly relevant since refugees tend to be resettled (with a few exceptions) into local areas where individuals from their home country already live. The authors claim that this instrumental variable is excludable because both the share of US immigrants from a particular country of origin in 1980 and the number of refugees from a particular country of origin within a given year are both unlikely to be correlated with present day crime and terrorism, except through the potential mechanism of the present day share of refugees within a given county.

So, what are the results?

The instrumental variable analysis show no statistically significant effect of the number of refugees on any type of crime or acts of terrorism. The authors visualize these results by plotting these relationships. If anything there is a slight negative relationship between the number of refugees and acts of crime or terror.This paper provides some serious and relatively rigorous evidence to the debate about the relationship between refugees and acts of crime and terrorism. Despite this, the paper is not perfect. The authors briefly mention the secondary migration of refugees. This is a topic I’ve researched in the past (see here and here). I think this is an important aspect to explore in future drafts of this paper.

I also think it might be worth looking into the lagged effects of refugee resettlement on crime rates. Since the authors have multiple years of data, this sort of lagged effect analysis seems like it would be relatively easy to implement. I’d also encourage the authors to spill additional ink defending the excludability of their instrumental variable. The results align with my priors, so I’m probably less likely to spend too much energy searching for a violation of the exclusion restriction, but many obviously don’t share my perspective. Finally, on an aesthetic note, get rid of the default Stata graphs. It is petty, but I suggest the authors make the y-axis labels vertical and at least type “scheme(s1mono)” when generating their figures.

Nevertheless, these results suggest that the fear that President Trump has stoked over the last several years are statistically unjustifiable. At best these fears are based off of anecdotal evidence and hear-say that is not representative of reality. At worst these fears are the result of deep-seated animosity embedded in far too many US citizens toward people from different countries, cultures, and who practice a different religion.

For those interested in more research on this topic, here is a nice Twitter “moment” thread by John Holbein.