The Effects of Segregation on Crime Rates. David Bjerk McMaster University and RAND Corp. I - Introduction.
McMaster University and RAND Corp.
Numerous authors in a variety of disciplines have documented the high rates of crime and victimization in poor minority neighborhoods (e.g. Kling, Ludwig, and Katz, 2005; Krivo and Peterson, 1996; Kotlowitz 1991; Patterson, 1991; Messner and Tardiff, 1996)
While these studies highlight the correlation between segregated neighborhoods and crime, they do not necessarily identify whether greater segregation in a city (be it racial or economic) has a direct impact on the overall amount criminal activity or simply concentrates criminal activity to certain neighborhoods.
In this paper, I use differences across cities with respect to how housing assistance to the poor is allocated, as well as variables related to the local public finance of each city, to identify the direct effect of segregation on crime using instrumental variables methods.
There are a variety of quite straightforward reasons for why poorer individuals may be more prone to criminal activity than richer individuals.
However, greater segregation may also directly influence criminal activity.
Hence, greater segregation may directly increase criminal activity, especially with respect to violent or interpersonal crimes.
A variety of other potential mechanisms through which greater segregation can directly effect criminal activity have also been proposed, such as:
Alternatively, there are also ways in which crime might directly affect segregation. For illustrative purposes, assume for the moment that segregation has no direct effect on crime.
(1) Consider a state of the world where cities start off relatively integrated, with well-off white families living in neighborhoods with poorer black families.
(2) Alternatively, consider a state of the world in which cities start off relatively racially segregated, where the relatively small number of poorer white families live in neighborhoods with richer white families, but richer black families live in neighborhoods with the relatively large number of poor black families.
To look at the relationship between segregation and crime empirically, I use data from a variety of sources by MSA:
Other MSA level control variables:
As discussed previously, interpretation of these results is complicated by the potential simultaneity bias between segregation and crime, meaning these estimates may vastly overstate or vastly understate the true direct effect of segregation on citywide criminal activity.
To obtain more plausible estimates of any causal impact of segregation on crime, we need to find and exploit differences across cities that may impact segregation, but are exogenous to current crime conditions.
Three instruments will be used for this analysis:
The other two instruments are taken from Culter and Glaeser (1997):
(2) The number of municipalities within each MSA in 1962.
(3) The fraction of MSA government revenue in 1962 coming from the Federal government.
There is no reason to think that either of these variables could be directly related to crime conditions around the year 2000.
Recall that many of the explanations for why racial segregation may directly affect criminality actually hinged on the relationship between race and poverty.
We can look at this hypothesis to some degree by creating other indices of economic segregation besides the racial segregation indices.
2SLS results using the indices of poverty segregation are generally very similar to those using the indices of racial segregation.
In general, the results of this analysis suggest that greater racial and economic segregation have:
In terms of policy implications, these results suggest that efforts to reduce racial and economic segregation in cities (e.g. through razing high-rise public housing) may lead to drastic improvements in quality of life for the poor, and arguably for city residents overall, through reducing the total amount of violence (but might come at the cost of more car thefts).
Given we have more instruments than potentially endogenous variables, the model is overidentified and we directly test our exclusion restrictions regarding our instruments.