Effect of Private Police on Crime: Evidence from a Geographic Regression Discontinuity Design

Past literature in criminology has shown that increases in policing can reduce crime. Generally, most rigorous studies have been done on temporary deployments of extra police, or “episodic” changes rather than permanent changes. To better understand effects of a sustained increase in police deployment, a group of researchers working in Philadelphia decided to look at the effect of the University of Pennsylvania Police Department (UPPD), which patrols within a specific boundary surrounding the university, on the amount of crime happening in the surrounding area.

The UPPD provide supplemental police services in University City, a district of Philadelphia including the university and surrounding neighborhoods. They are the largest privately funded publicly certified police force in Pennsylvania. Though the Philadelphia Police Department (PPD) serves all of Philadelphia, within the patrol zone boundary of UPPD, those forces are supplemented by university police officers. At any given time, there are between 6-8 PPD officers in the entire University City district, as well as some private security officers, and UPPD has 16 officers within the Penn patrol boundary at all times. As the number of officers within the patrol zone is approximately doubled, the increase in policing coming from the UPPD officers is certainly significant.

The researchers used a geographic regression discontinuity design to study the effects of the extra police officers  by taking advantage of the arbitrary boundary in which the UPPD patrols. Using common indicators that affect crime rates, such as population characteristics and land use, they found that the blocks just inside and just outside the boundary were essentially identical in terms of crime rate prediction. This means that a difference in crime rates just inside and outside the boundary (or what is known as a discontinuity) could have been caused by the treatment applied within the boundary — in this case, the increased police presence.

The study used data from UPPD of all crimes in University City from 2005-2010. The crimes were classified into street crimes, property crimes, violent crimes, and aggregated into total crimes. The unit of analysis used was a city block, as those were easily identifiable by police and individuals, and are natural settings for police interventions and daily routines. Data on parking tickets and traffic accidents at the block level were also used to serve as control variables, since the UPPD does not issue parking tickets, and should not affect traffic accidents within or outside of the patrol zone boundary.

In the results, there is a clear jump at the boundary in the number of crimes, with different statistical significance depending on the types of crimes and certain estimators used. For all crimes by block, there were between 32 and 61 more crimes in blocks just outside the boundary, which was a significant increase of between 45%-86% relative to the average number of crimes within the boundary. To check the results, the researchers tested other scenarios in which blocks were assigned random distances from the boundary and not their actual physical distances. This resulted in only 1.3% of cases with a larger discrepancy between crimes inside and outside the boundary.

For street crimes, which past research has shown to be more easily deterred by police patrols, the increase is a statistically significant 45%-115%. Property crimes show less of an effect, with statistical significance depending on the estimators used, and effects of 27%-60%. Violent crimes also were impacted, with an increase of 119%-153% outside the boundary, though the statistical significance was somewhat dependent on the specification of the estimator.

To check that the likelihood of risky behavior and rule-breaking is not dependent on the geographic location, whether inside or outside the boundary, the researchers used the data parking ticket and traffic accident data to compare the blocks. There was no systematic effect of the patrol zone on the number of these incidents, indicating no substantial differences in those behaviors, at least, on either side of the boundary. This supports the conclusion that increased police presence within a given area reduces crime in that area.

Using the increases found, the study created elasticity estimates for each type of crime. These estimates calculated the amount of change in crime based on the concentration of police per area. The estimates found were very similar to estimates created by previous studies that studied increases in police presence. Though they did not match all previous studies on all types of crime, property crime and violent crime were found to match at least several estimates found in other studies.

Some of the limitations of this study are that the study did not identify the specific mechanisms by which police were affecting crime at the boundary of the patrol zone. If the police presence was noticed by criminals, and the boundary was identified, they may have chosen to commit crimes just over the boundary that otherwise they would have committed inside, leading to a greater measured effect since some crime would be displaced from just within to just outside the boundary. However, this noticing of police presence could also have caused criminals to be deterred from committing crimes even just over the boundary, leading to a smaller measured effect, since both blocks just within and just without would have experienced a decrease in crime because of the increased police. An experiment randomly changing the dosage of police officers in an area could help determine whether it was the police themselves, or simply perceived police presence that changed crime levels. Another limitation also may be that there are invisible differences between blocks just inside and just outside the boundary that may or may not be caused by the increased police. The study also did not determine whether reporting behavior changes inside and outside the boundary, or to UPPD versus to PPD (since UPPD would likely have only crimes within the boundary reported to them) which could have led to discrepancies between measured and actual behavior. Further research on these limitations should be done to determine the effects of these on the conclusions made by this study.

______________________________________________________

Takeaway:

  • Extra police decrease crime in adjacent city blocks by 43-73%
  • Permanent extra police presence, such as University Police, can reduce crime outside of their “campus” boundaries
  • The results are consistent with other research on increased police presence, specifically for reductions in property and violent crimes

______________________________________________________

Reference:

Macdonald, J., Klick, J., & Grunwald, B. (2016). The effect of private police on crime: evidence from a geographic regression discontinuity design. Journal of the Royal Statistical Society: Series A (Statistics in Society), 179(3), 831–846. https://doi.org/10.1111/rssa.12142

 

 

Disparity Does not Mean Bias: Making Sense of Observed Racial Disparities in Fatal Officer-Involved Shootings with Multiple Benchmarks

There has recently been much discussion in the United States on racial disparities in police officer-involved shootings (OIS). Comparing the fraction of black citizens who were fatally shot to their fraction of the general US population, many claim that there is racial bias in police officers’ decisions.

However, this is not necessarily a correct conclusion to draw, since the population at risk of being fatally shot may not be the same as the general US population. If most of the population does not encounter the police in any given year, there is little to no chance of them being fatally shot by police, except in cases of ricochet or non-purposeful shooting, which are very rare.

In order to actually calculate racial disparity, the comparison benchmark should be a group of people who have a chance of being involved in a fatal OIS in the first place. In terms of social science research, this group of people is similar to the at-risk population for a disease or symptom, against whom the actual rate of people affected are compared. Determining a benchmark population to properly estimate the population that is ever at risk of being involved in a fatal OIS is a difficult problem, but is invaluable to properly assessing the racial disparities in OIS situations.

In 2017, a group of researchers used seven different benchmarks for the at-risk of fatal OIS population, and found varying levels of racial disparities in fatal OIS across the US, using data from 2015-2017. Their study shows the importance of selecting the appropriate benchmark for conclusions to be drawn from. It also notes the assumptions that are made with the different benchmarks, and why those are necessary to be aware of when calculating disparity and making inferences about its meaning.

The formula used to find the odds ratio of a black citizen being fatally shot as opposed to a white citizen being fatally shot was simply the black fatality rate over the white fatality rate, or:

[Black Fatally Shot ÷ Black Benchmark]

____________________________

[White Fatally Shot ÷ White Benchmark]

 

Using publicly available data from the U.S. Census for the first benchmark, data from the Police-Public Contact Survey for benchmarks 2-4, and data from the Uniform Crime Report for benchmarks 5-7, the researchers compared data from the Washington Post’s fatal OIS data to calculate the impact a person’s race (black or white) had on the likelihood of being fatally shot by an officer compared to the at-risk population.

They found that the odds ratios varied greatly, with results across all three years (2015-2017) for each benchmark as noted here. Results greater than 1.0 indicate that blacks are more likely to be fatally shot while results less than 1.0 indicate that whites were more likely to be fatally shot.

  • In the general U.S. population: black citizens are about 3 times more likely to be fatally shot by an officer
  • In police-initiated contacts: about 3.25 times more likely
  • In traffic stops: about 3.5 times more likely
  • In street stops: about 2.5 times more likely
  • In total arrests: about 1.3 times more likely
  • In arrests for violent crime: about 0.8 times as likely
  • In arrests for weapons offenses: about 0.67 times as likely

The results show that in the first five benchmark cases, black people are more likely to be killed in a fatal OIS, but with very different likelihoods, whereas in the last two cases, white people are more likely to be killed in a fatal OIS. For interpretation purposes, results can be grouped into those greater than 1.0, indicating situations where black people are more likely to be fatally shot, and those less than 1.0, indicating situations where white people are more likely to be fatally shot.

The assumptions being used to craft these benchmarks include: OIS occur in response to perceived imminently dangerous citizen behaviors, criminal behavior is a reasonable proxy for imminently dangerous behavior, and arrests are a reasonable proxy for criminal behavior. By that logic, people should be comparing the fatal OIS rate to one of the other benchmarks which take into account the population among whom an OIS has a chance of happening in the first place. Essentially, if the opportunity of an OIS is not even an option, that case should not be used to tell whether or not the officer was more likely to fatally shoot someone based on their race.

The researchers acknowledged some limitations of their study, including the difficulty of getting accurate data from all law enforcement agencies, especially as the government does not track much of the OIS-related data, though hopefully they will begin doing so soon. Since the data being used is collected at the national level, these results cannot be used to infer anything about a more specific group, such as an individual case, a single department, or county law enforcement offices.

The study specifically focused on the racial disparity between black and white citizens who are fatally shot by officers. This data does not consider non-lethal interactions with the police, and so cannot be used to find anything about those cases. Since they only measure the racial disparity among fatal OIS, the results also cannot be used to determine anything about disparity or potential bias that may occur before the chance of an OIS, such as any disparity or bias by officers when stopping people for traffic stops or street stops. Finally, this data cannot be used to show that racial bias is the driving force behind any uncovered racial disparities.

This study also does not deal with cases when the at-risk group being used as the benchmark does not always encompass all the situations in which a fatal OIS may happen, such as the controversial cases of officers shooting and killing citizens who posed no imminent threat. The researchers mention that these scenarios complicate the use of certain benchmarks that do not take these situations into account.

Overall, the study emphasizes the importance of choosing a benchmark population against which to compare the rate of people who are fatally shot by officers. Because many people do not encounter the police and many police interactions occur with no force, let alone lethal force, ever being used, the benchmarks for finding racial disparities must be more carefully considered before being used to determine bias.

__________________________________________________________________

Takeaway:

  • The comparison benchmark for fatal OIS should include those who have a chance of being fatally shot by police, instead of the entire general population
  • Black people are more likely to be shot and killed in OIS in police-initiated contacts, traffic stops, street stops, and total arrests
  • White people are more likely to be shot and killed in OIS in arrests for violent crime and in arrests for weapons offenses
  • Benchmarks for determining racial disparity must be carefully considered before being used to determine bias

__________________________________________________________________

Reference:

Brandon Tregle, Justin Nix & Geoffrey P. Alpert (2018): Disparity does not mean bias: making sense of observed racial disparities in fatal officer-involved shootings with multiple benchmarks, Journal of Crime and Justice, DOI: 10.1080/0735648X.20181547269