In response to allegations of discrimination by police officers, the Massachusetts legislature passed an act in 2000 mandating that the Registry of Motor Vehicles record data on the race, gender, and search status of drivers who were issued a written warning or a ticket. Several previous statistical studies examined this data using regression techniques. This previous body of work, which focuses almost exclusively on demographics, concludes that the disposition of citations—tickets versus warnings—reveal racial and gender bias. Unlike the earlier work, this paper analyzes the Registry data using Classification and Regression Trees (CART) to determine the extent, if any, of racial and gender profiling. In addition to demographic information, our models consider variables like the situation (e.g., time of day and location), background (e.g. make and model as well as registration of the car), and behavior (exceeding the speed limit and driving at a high rate of speed) as part of the CART analysis. We find that, if the seriousness of the speeding violation—high speed and significant speed over the limit—are considered, the effects of demographic factors such as race and gender disappear. Furthermore, since driving patterns are correlated with these same demographic factors, the demographic factors themselves behave as proxies for serious traffic offenses. This case is accessible to readers with an intermediate-level knowledge of statistics.
