Introduction

Many researches already exist on disparities in US policing and this research extension, inspired by this article, modifies the focus to assess a potential bias due to the driver’s gender. Indeed, this factor is susceptible to influence the sensibility of the officer, which may act differently with men or women. Search decisions and stop outcome could therefore differ. This is evaluated in this study through a data analysis on several US states or cities provided here. The aim is to determine if there are disparities in drivers' treatment and whether it is comparable across the USA. This is done by first balancing the datasets, and then comparing stop, search and hit rate statistically. Average outcome differences are then computed for datasets of states that present the required features, to quantify 1. the difference in searches effectively done (for same reason to search) and 2. the difference in stop outcomes for same stop reasons. This study ends with a study case on Florida, that permits to assess a potential effect of officer's gender.

Ratio of male and female drivers

Stop Rate

The stop rate is the proportion of drivers stopped compaired to the population of the city or state. For all cities the stop rate is significantly higher for men except for Maryland state, which means that generally, more men are stopped than women. At this point of the analysis, the dataset is balanced only to have the same racial ditribution in men and women subsets.

Search Rate

The search rate is the proportion of stopped drivers that have been searched. It is represented for the search of the car as well as the search of the person. The person search rate appears to be higher for women but the vehicle seach rate is higher for men. At this point of the analysis, the dataset is balanced only to have the same racial distribution in men and women subsets. It means that the distribution of infractions types, of drivers age and of policer suspicion degree that the drivers is carrying contraband are not the same in men and women subsets. This issue is adressed further in the analysis.

Hit Rate

The hit rate is the proportion of searched drivers for which contraband (drugs, guns) has between found. We performed a statistical Fisher test to assess if the proportion of men carrying contraband among all searched men is significantly different than the one of women. The darker the color of the dot, the higher the significance. With the exception of Greensboro, for which women have a significant higher hit rate, we observe higher hit rates for men. However, only one is significantly different than women's hit rate (Raleigh) for a confidance level of 5%. It seems generally not justified to search men more than women.

Search and arrestation/citation decision

Equal treatment for male and female drivers having committed the same infraction ?

To assess if men and women are treated equally, they are paired and their outcome decision taken by the police officer is compared. The matching allows to pair one male driver with one female driver presenting a similar subjective signal to the police officer : same race (to avoid a potential racial bias), same age category (more or less than 45 years old), same reason for search (to compare drivers offering to the officer a similar suspicion level to find contraband) and same search issue. The age is used as a proxy for the attractiveness of the drivers.

The results are displayed on the maps bellow.

SEARCH DECISION

The map below presents the average outcome difference (AOD) in the personal search decision between men and women. If the driver is searched, the outcome is set to 1 and kept to 0 otherwise. The difference is computed as the outcome for men minus the outcome for women. A positive value indicates that men are more searched. For all the states analyzed, the difference is significantly positive (95% confidence intervals can be displayed by clicking on the map). Men appear to be more searched than women for similar suspicion degree to be carrying contraband.

STOPS AND ARRESTATION/CITATION DECISION

Now, when analyzing the stop outcome decision, the matched pairs have also the same reason for stop (instead of reason for search) and the same search issue (contraband found or not). This allow again to compare only drivers having committed the same infractions. The map below presents the average outcome difference (AOD) of a stop in the arrestation/citation decision. If the driver has been arrested or a citation has been issued the outcome is set to 1 (severe decision). If the driver receives a warning only, the outcome is set to 0. Except for police departments of North Carolina and Tennessee, the AOD is significantly negative (95% confidence intervals can be displayed by clicking on the map). For these cases, police officers tend to let women go more easily than men for the same infraction.

When the driver is searched but not his/her car...

When a police officer is suspecting the driver to be carrying contraband, he can search him/her and his/her car. While looking for drugs or guns it would be surprising to search the person but not the car. These cases could be instances of unwarranted person searches and sexual assault.

The figure bellow compares the AOD in the search decision of stops where the car has been searched to stops where it wasn’t. We observe AODs of 0 for some police departments. These cases are specific to the dataset, as appearently, if there is a reason for stop, some police department necessarily have either the vehicle or the person searched (other state have also frisk that can be performed, or simply indicate no search even if there was a reason for that is written). The search reason feature refers also to the person search meaning that by matching pairs on the search reason, we have only drivers that have been searched when their car wasn’t. For the datasets of Charlotte and Greensboro, we observe lower AOD for cases where the car was searched than where it wasn’t. The difference between men and women is then smaller.

For which infraction are women let free more than men ?

This question can be answered by looking at the AODs presented in the figure bellow. All different reasons for stop are represented. Some of them are not infractions and some other are not very precise. When looking at precise infraction types (seat belt, speeding, ..), we observe that men are in general less easily let free than women.

Case study on Florida state : influence of the officers' gender on the stop outcome decision

Let's dig into even more details concerning stop decisions. What happens if we split female from male officers, and look for the average outcome difference ? The aim of this analysis is to find out if the officer's gender has an impact on a potential bias in decision making. For this purpose, a similar methodology is applied as before to balance our dataset, and outcome difference for each type of potential violation will be compared. Very few datasets contain all information needed for this analysis, hence it is performed only on the big dataset of Florida state (3.2 Mio stop records for men officers, 0.25 Mio for women officers before filtering).

The AOD for the woman officers is near 0, and the confidence interval ranges from negative to positive values:

at this point, it is not possible to determine whether woman officers take biased decisions. However, it seems different for men : their average outcome when deciding on the outcome of a stop seems to indicate that they tend to search men more than woman drivers, even if the age range, race, subjective indicator (reason for stop) are identical for them.


From the results when splitting men from women officers, and splitting the group by type of reason, some interesting findings appear :

- Women officers are more gentle with male drivers that drive at too high speed, reprehending them less than female drivers, whereas men officers do not make a significant difference between them.

- For seatbelt violation, it is the exact opposite : women do not make any difference in their judgement, whereas men officers are more gentle with female drivers.

- Both men and women officers seem significantly more gentle with female drivers that have faulty equipment.

- Men officers seem particularly biased regarding registration and TAG violations, as well as traffic control devices violations : they decide themselves for harder outcomes when facing a male driver. This is however not the case for female officers, acting equally on male and female drivers for these violation types.

- Female officers seem particularly biased with female for CMV inspections : they tend to let them get away less easily than men drivers.

Conclusion

This analysis permits to show very interesting, statistically verified, findings : altough there is similar hit rates for men and women, US male and female drivers get treated differently by officers when stopped, independently from their age range, gender, race and propensity to carry contraband. These differences may vary depending on the type of violation, and particularly whether the officer is a man or a woman. Tendancially, men officers show most bias, being more gentle to woman drivers than to men. It was shown that policers behave differently depending on if they searched the car of the driver : if not, the difference in outcome is very often quite inexistant, showing an equivalent treatment for male and female drivers. However, a significant difference appears if the car's searched : men are more often searched than women. Across different states, these differences and conclusions may vary, and not precise pattern can be made explicit. Weaknesses are of course present in this study, particularly on the officers' sexual orientation (which is partly assessed in the case study). Further improvements could be achieved if the reasons for stop and search would be more precise and better documented. Police officers who are conducting gender profiling might also cover themselves by overstating reason for stop and search. This would bias the study. This issue should be adressed in further work.