Department of Economics
Yale University
In a recent issue of the Atlantic Monthly (June/July 2008), Hanna Rosin describes a correlation which one of its authors describes as a “hard thing to say or write” about. The question addressed by Rosin concerned the perceived rise in crime in many American cities. The answer according to her article, “implicates one of the most celebrated antipoverty programs of recent decades.” (American Murder Mystery, The
"Nobody in the antipoverty community and nobody in city leadership was going to welcome the news that the noble experiment that they’d been engaged in for the past decade had been bringing the city down, in ways they’d never expected."
The evidence is essentially this map from the article:
Even before we come to the usual question of how correlation relates to causality, the measures used in this graph raise a number of question. First, rather than use some notion of crime rates, the authors use crime incidents, and of course densely populated regions will probably have more crime just because there are more people, so at the very least, the map should be based on crime per capita. If not, then one entirely reasonable way to read the map in the article is that Section 8 residents simply relocate where there are more people. Where there are more people there are more lots of things: grocery stores, pets, streetlights, and crime. Surely the relationship of interest is if an influx of Section 8 residents led to an increase in crime rates. While we do not yet have the merged crime incident data and population densities by the census tracts for Memphis, here is the map of population density in 2000 Memphis (the darker spots are the more densely populated areas):
Visually, it is clear there is a strong relationship between the number of crime incident in the Atlantic map and the population density across tracts. It is clear from this that simply re-doing the Atlantic map using crime rates would substantially weaken the relationship, perhaps eliminating it entirely.
Another issue concerning the measure of crime used, even apart from whether we look at crime rates or the raw number of crime incidents per tract, is the type of crimes involved. The Atlantic map uses incidents of “violent crime”, which is not itself a single category of crimes in any Federal reporting system such as those used by the FBI and branches of the Department of Justice. If we take the conclusion of the Atlantic article at face value, we would have expected a much more “exportable” type of crime to be property crimes such as burglary and robbery, and not so much violent crimes such as assault and homicide. Presumably the particular nature of the old housing projects that were replaced with the Section 8 voucher and certificate programs lent themselves much more to the extremely violent crimes, whereas the increase in opportunities (not to mention the value) of property crimes in the relatively more “suburban” areas allowed by the Section 8 program would allow for the greatest “exportation” of this type of crime – property crime - from the old housing projects. Yet the Atlantic map did not show the patterns associated with this non-violent class of crimes, even though this would lead to a more compelling “cause and effect” interpretation of their 2006 map.
This brings us to the most fundamental problem with the research cited in the article, which is simply that correlation is not causation. In particular, there is no analysis of crime patterns prior to 1997. Along the lines of our explanation for the patterns seen with raw crime incidents given above, it is also entirely likely that Section 8 residents chose to locate in areas where crime and poverty already existed – areas where rents would be cheaper and landlords would be more willing to rent under the confines of the Section 8 program. We have not immediately obtained maps of crime rates by detailed census tracts, but we did easily locate maps of poverty rates by census tract for Memphis in 1990 and 2000 (again, darker spots are poorer) to shed some light on whether this is all just a “selection” story, whereby Section 8 residents did not cause crime or poverty, but instead simply moved into areas that were already saddled with poverty at the time the Section 8 voucher programs were being expanded. Compare the following two map for poverty rates in
Poverty Rate in 2000
Compare these maps with the map showing Section 8 rentals – Section 8 rentals seem to have sprung up in areas that were already the poorest in 1990. While we acknowledge we would ideally have these two maps with crime rates, nonetheless, this similar analysis with poverty rates indicates that Section 8 rentals are not causing poverty. We should also note that even if this “before and after” type of analysis had shown a relationship with increased poverty rates in areas where Section 8 renters moved in, we still should not infer that Section 8 renters caused that increase since there are many reasons to think that the two trends are linked without one causing the other – both could be a pure by-product of increasing urban sprawl, with the increased dispersion of Section 8 renters riding a stronger wave of migration out of city centers to the relatively more suburban areas that has nothing to do with the operation of government programs and everything to do with free markets in the housing sector.
(http://bjsdata.ojp.usdoj.gov/dataonline/Search/Homicide/Local/TrendsInOneVar.cfm)