Rail transit and neighborhood crime: the case of Atlanta, Georgia.

• Author: Ihlanfeldt, Keith R.

1. Introduction

   In a growing number of metropolitan areas, the construction or extension of a rail transit system has been advocated by policy makers as a way to address the numerous externalities associated with a heavy reliance on the automobile for the journey to work. However, in many of these same areas, rail transit proposals have encountered significant opposition from neighborhood groups (Pendered 1997; Carlson 2000; Stingl 1996; Byrd 1989; Gayle 1989; Armacost 1994; Kane and Lee 1994; Tandon 1999). These groups fear that a rail station placed in or near their neighborhood will increase neighborhood crime, because this would provide criminals improved access to the neighborhood. (1) If these fears are justified and transit stations do increase neighborhood crime, this may counteract one of the principal objectives of the new transit proposals: to reduce urban sprawl by attracting population and employment to station areas. Crime has been found to strongly affect the intrametropolitan location decisions of firms and households (Mills 1992; Cullen and Levitt 1999; Bollinger and Ihlanfeldt 2000). Hence if station-induced crime is a reality, station openings may worsen, rather than mitigate, urban sprawl. In addition, to achieve transit-oriented development, significant subsidies may be required to offset the cost of crime.

   There are reasons to believe, however, that the opening of a rail station may actually cause neighborhood crime to fall, rather than rise. Although the station may increase access to the neighborhood by outside criminals, it may also induce criminals living within the neighborhood to commit their crimes elsewhere by lowering commuting costs between the home neighborhood and other neighborhoods. Moreover, the station may increase the job accessibility of neighborhood residents, especially that of crime-prone youth, thereby increasing the opportunity cost of crime. Hence the net effect of rail transit on neighborhood crime can be either positive or negative, depending on the relative magnitudes of the above three factors.

   This paper provides some rare evidence on the relationship between rail transit and neighborhood crime. First, the economic model of crime is extended by adding a spatial dimension: criminals are mobile across neighborhoods and commit their crimes where net returns are the highest. Based upon this spatial model of crime, estimable models of neighborhood crime are developed, which include the percentages of the neighborhood served by rail transit as an explanatory variable. To estimate these models, a unique panel database for the Atlanta region that includes four consecutive years of crime data for each of 206 census tracts is used. With these data, neighborhood crime models are estimated for total crime and separately by crime category (property crime versus violent crime) using both fixed effects (FE) and random effects (RE) models.

   The results show that rail stations have a statistically significant effect on neighborhood crime and that the effect varies with three characteristics of the neighborhood: median income level, density of poverty, and average distance to poor people living outside the neighborhood. The mix of characteristics found within central city neighborhoods has resulted in transit increasing crime there, whereas in the suburbs transit has reduced crime in white neighborhoods and has had no effect on crime in black neighborhoods. The results suggest that the fears expressed by suburban residents over station-induced neighborhood crime are unfounded and that extensions of rail transit into the suburbs will not cause further decentralization of population and employment. However, to maximize transit-oriented development around central city stations, greater police surveillance within transit neighborhoods may be necessary.

2. Literature Review

   Only four studies provide evidence on whether rail stations increase neighborhood crime, with findings evenly split against and in favor of the hypothesis that transit causes crime. In a study of the Baltimore Metro system, Piano (1993) looked at trends in crime rates for neighborhoods surrounding three rail stations using data from years just before and just after the stations opened, comparing these to trends in crime rates for Baltimore County as a whole. No significant relationship was found between the trends in crime rates and the opening of the rail stations. However, there were no controls for other factors that might have affected crime rates.

   In a study of the Metropolitan Atlanta Rapid Transit Authority (MARTA) system, Poister (1996) also looked at time series data of criminal activity near rail stations before and after station opening dates. The study considered only two contiguous stations on the same section of the rail line, using neighborhood and county monthly crime data over a four year period. The poststation period was limited to just 18 months. The graphical "event study" depiction of neighborhood crime rates before and after the station openings showed little or no overall impact on crime rate trends in the neighborhoods surrounding the stations. A simple regression model attempting to measure changes in the overall trend in crime rates near the stations both before and after the station opening dates also offered little evidence to support the hypothesis that rail stations cause an increase in local crime. However, like Plano (1993), Poister (1996) used only the opening dates of the stations to explain changes in crime rate trends, ignoring other explanatory factors that may have been correlated with these dates.

   Bowes and Ihlanfeldt (2001) also used Atlanta data to estimate a simple neighborhood crime model that served as an auxiliary equation to their hedonic price analysis of the impact of MARTA rail stations on residential property values. The crime model was used to estimate the indirect effects that stations may have on the values of nearby properties by attracting criminal activity to station areas. In their basic model, Bowes and Ihlanfeldt (2001) found that the density of neighborhood crime is higher in those census tracts whose centroids are within a quarter mile of a station. However, in a model containing interactions, the latter effect was found to vary with neighborhood income, distance from downtown, and whether the station had a parking lot. (2)

   Finally, Block and Block (2000) mapped reported street robberies (actual and attempted) on Chicago's Northeast Side and within the Bronx borough of New York City. For both of these places they found that there was a strong relationship between street robbery and propinquity to a rapid transit station. The number of robberies tended to peak a few blocks away from the stations.

3. A Spatial Economic Model of Crime

   The economic model of crime (Becker 1968) is aspatial in the sense that all crimes are assumed to be committed by residents of the home community. This assumption is untenable if the objective is to explain neighborhood crime when criminals are mobile across neighborhoods. In this section a simple model that adds a spatial dimension to the standard crime model is presented. The objectives are twofold. The first objective is to identify those factors that account for differences in the amount of property crime across neighborhoods. The model is also relevant to that subset of violent crimes that are economically motivated. The second objective is to relate these factors to the percentage of the neighborhood that is served by rail transit.

   Consider the average resident (R) of neighborhood H. Assume that within a given time period, he must decide whether he will commit a crime and whether this crime will be committed within the home neighborhood. The joint probability that he is a criminal and commits his crime in H equals the marginal probability that he is a criminal times the conditional probability that he commits the crime in the home neighborhood given that he is a criminal:

   (1) [P.sup.R.sub.H] = [P.sup.R](C) > [P.sup.R](H|C).

   The factors that determine the marginal probability are identified by the standard crime model. In this model, the expected net return ([pi]) from committing a crime is defined as the expected payoff (w) minus the direct cost incurred in committing the crime (c) minus the product of the probability of apprehension and conviction (p) and the prospective penalty if convicted (f):

   (2) [pi] = w - c - pf.

   The expected net return from committing a crime can be defined separately for a crime committed by the resident within his home neighborhood ([[pi].sub.H]) and outside his home neighborhood ([[pi].sub.O]):

   (3) [[pi].sub.H] = [w.sub.H] - [p.sub.H]f ,

   (4) [[pi].sub.O] = [w.sub.o] - [p.sub.o]f - [t.sub.o],

   where c is limited to the costs of commuting to the crime ([t.sub.o]), which are assumed to be negligible if the crime is committed within the home neighborhood. The resident will be a criminal if the maximum expected net return from crime exceeds the benefit of being law-abiding, which equals foregone expected earnings in legitimate activity (e) net of journey-to-work costs (j) plus the monetary equivalent of the psychic return from good citizenship (g):

   (5) max([[pi].sub.H],[[pi].sub.o]) > e - j + g.

   Given that Equation 5 is satisfied, the resident who is a criminal will commit his crime in H only if the expected net return from crime is higher there than elsewhere:

   (6) [[pi].sub.H] > [[pi].sub.o].

   Equation 5 defines the variables that affect the marginal probability, whereas Equation 6 defines the variables that affect the conditional probability. Together these are the variables that affect the joint probability: (3)

   [P.sup.R.sub.H] = f([w.sub.H], [w.sub.O], [p.sub.H], [p.sub.o], [t.sub.o], e,j, g).

   Although the above model adds a spatial dimension to the economic model of crime, Equation 7 provides limited guidance on how this dimension might be incorporated into an empirical model...