By: Patrick Zimmerman

In shocking news, Donald Trump isn’t drawing on evidence when making his sweeping statements about the danger posed by immigrants (including but not limited to illegal ones).  Most commonly using this argument to tout the need to build a wall on the US-Mexico border or lambast those pesky “sanctuary” cities in California, he has repeatedly and unequivocally associated all immigration with spikes in crime rate, in particular violent crime rate.¹  To Trumplandia, people from “shithole countries” contaminate the places they emigrate to, both moral and physical contagions that must be quarantined in order to Make America Great Again through the glory of scapegoating.

Funny, it seems as if the FBI has a whole lot of data on crime rates,² and the US Census appears to pay attention to immigration (and has a handy-dandy API).³  So let’s add our two cents to the number of studies that show no causal relationship between the two.

The question

Is an immigrant population associated with an increase in violent crime rates?  Specifically, is the immigrant population that most vexes the #MAGA crowd, Latinos “pouring” in over the border, corresponding to a spike in the violent crime rate?

The short-short version

Nnnnnope.  Big fat zero.  Not at all.  Nada.  Zilch.  Math says, “uh uh,” at least when looking at 2014 (the most recent year in the full UCR database).

When controlling for known correlations (namely, median income and black population), the non-citizen %, non-native born %, and % of population using languages other than English of a given Census-designated place all showed no correlation with violent crime rate at a high level of significance (p<0.0001).  Latino % showed no correlation at a moderate level of significance (p<0.1).

The residents of sanctuary cities appear to be at no more risk from violent crime due to an immigrant influx than any other place.  So much for the immigrant crime wave…

The Wall, in addition to frequent questions about its potential effectiveness and expense, isn’t going to solve the problem of violent crime in the US.  Moreover, it’s certain that a fair number of people in the DHS, DoJ, and even within the Trump Administration are well aware of this.  It is pretty hard to think of the Southern Border Wall, then, as anything other than an overt attempt at social engineering; making America “Great” is clearly synonymous with making it more White and less Latino.

The results

After controlling for confounding variables, the conclusion is clear: Immigration has nothing to do with violent crime.  Thank you for playing, Mr. President; better luck with your next racist argument.

Citizenship status actually has the weakest correlation to violent crime of any of our tested variables.

Mouseover for details.

Non-citizen population percentage has a correlation with violent crime of -0.095, with a p-value of <0.0001.  It is, for all intents and purposes, completely random and non-predictive of violent crime in any way.  It’s almost as if the behavior of non-citizen immigrants was indistinguishable from the rest of the population.  Hmmmmm……

Ok, let’s take into account naturalized citizens who managed to make it here (either legally at the time or ex-post-facto legalized).

Mouseover for details.

Still a big fat doesn’t matter.  Correlation of -0.145, p<0.0001. 

“Fine,” says the Trump Advocate-brand rhetorical device.  “What about people who haven’t embraced Amurika enough to speak English at home?  Everyone knows that assimilated folks are more likely to get along with people here.”

Mouseover for details.

Sorry red hat.  Nothing there, either.  Correlation between speaking non-English languages at home and crime rate was -0.183, p<0.0001.

Time to stop beating around the bush.  Are Latino populations correlated with crime?

Mouseover for details.

At the risk of sounding repetitive: No.  While the significance was milder (p=0.083, so satisfying the p<0.1 rather than the <0.05 level), the data suggest that there isn’t any correlation there, either.  -0.033.

It’s pretty clear that Trump’s border wall, even one gives the plans the most generous interpretation of their potential effectiveness, is supposed to protect people from….a whole lot of nothing. 

The methodology

We pulled all data from the UCR database for 2014 for police departments reporting on violent crime rate (n=3422), then cross-indexed all of those that had a matching Census Designated Place in the US Census FIPS database (n=2828). 

We then wrote a bash script to loop over every one of the resulting CDPs and run an API call on the Census’ American Community Survey 5-year database for 2011-16 for the following variables: Total population, non-native born population, non-citizen population, spoke languages other than English at home, and Latino population.⁴  We also pulled info on potential confounding variables: median income (correlation of -0.370 with violent crime rate, correcting for black population %) and black population (correlation of 0.477 with violent crime rate, correcting for median income).

We ran multiple regression analyses (and a multiple regression analysis) in SPSS on the dataset.  The latter identified that only median income and black population affected the violent crime rate model, so we then went and ran correlations on the other variables correcting for those two factors. 

What’s next?

First, we’re planning to look at non-violent crime rates, to see if larceny, burglary, and the like have any relationship to immigration.  Next….um….we should probably look at why violent crime spikes in areas with high black communities, even when taking income into account.  Finally, these conclusions would be even stronger as a historical model, incorporating data back to the UCR’s 1985 start point.  That’s a longer-term project, but quite doable. 

Notes:
¹ Violent crime is defined by the FBI as “those offenses that involve force or threat of force” and broken down into “murder and nonnegligent manslaughter, rape, robbery, and aggravated assault.”^
² The FBI Universal Crime Reporting searchable database goes from 1985-2014.  2015 & 2016 are available, but without the custom searching functionality. We used the 2014 data.^
³ The 5 year surveys have a much more complete database for the optional reporting variables like race, occupation, and income than do the 1 or 3 year surveys. The Census requires one to get a free API key, then the call on the 2016 ACS 5-year survey is made on the following endpoint (using the following example variables: Washington – FIPS code 50000, DC – FIPS code 11, and non-native born – census variable B05002_013E) –   https://api.census.gov/data/2016/acs/acs5?get=NAME,B05002_013E&for=place:50000&in=state:11&key=whateveryourkeyishere.^
⁴ Important note: the US Census considers Latino an ethnicity, not a race.  It further breaks down the Latino category along racial lines (white, black, indigenous, etc).  So even in a hypothetical place where no non-Latino people of color existed and the only categories were “white” and “latino”, it would still be the case that Latino ≠ nonwhite. For example, immigrants from Spain or Portugal.^