This session discusses the value of combining empathy with data driven analysis. What does empathy add? What value is there to bring empathy to the data and the data to empathy. How do we do it? Let's start with the data and analytics side. We'll review one research study, a very good one that was published not long ago. This paper uses a large, very large data set to answer the question is there bias and police stops? That's one major risk to the goal of arriving at your destination. You may be late or not make it at all if you're stopped along the way, if there's bias in those stops than certain groups are more exposed to this risk than others. We're not saying all police stops are wrong, but they shouldn't hold people to different standards based on race or ethnicity. The researchers start by filing public records requests for arrest records is also called the Freedom of Information Act or FOIA. And arrive at a data set of about 95 million police stops by 56 state and local police agencies. The records include the race, gender and age of the driver, the date and time of the stop, where the stop occurred and whether a search was conducted. The authors conduct three main analyses on this data to try to answer their question, first is the veil of darkness test, which asks are minorities stopped more often in daylight than in darkness? The idea is that racial profiling is easier during the day than it is at night because you can actually see the driver during the day. So the researchers identified twilight for a given traffic stop location and then compared police stops 30-90 minutes before sunset to police stops after dark. What they find is that blacks are pulled over at higher rates before twilight as opposed to after, when drivers appearances are more visible. If bias didn't exist in these stops, then you'd expect the stops to be the same before dark as they are after dark. The second test focuses on the percentage of traffic stops that lead to searches and, when searched, do the searches reveal contraband. This test gets even deeper into bias because there is a single standard for when police can search a vehicle that's probable cause a fair probability that a search will result in evidence of a crime being discovered. Evaluating this probability should be the same across all traffic stops and all groups, but it's not, in this test stopped black and Hispanic drivers are searched nearly twice as often as stopped white drivers. Even worse searches of black and Hispanic drivers detect contraband less frequently than searches of white drivers. So minorities are stopped more often and searched more often, even though a smaller percentage of these searches appear to have been justified. It seems like that fair probability is assessed as lower when the driver is black or Hispanic than when the driver is white. The third test examines whether these vehicle searches change after a key justification to search, vehicles is taken away. Do searches and bias in these searches decrease after cannabis is legalized at the time of the paper, several states in the US had legalized cannabis, and the paper focuses on the first two Colorado and Washington. The reality is that cannabis is a type of contraband that has a smell and not just a smell, but a smell that is detectable even by the human nose. It's a common source of probable cause for a search for this reason, the authors find that in Colorado and Washington search rates significantly decreased after cannabis legalization for all groups for black drivers. Hispanic drivers and white drivers, while there were no decreases in states that didn't legalize, unfortunately, the bias and traffic stops remained. What does this tell us? What can data tell us that maybe we didn't know before. As we know, we can say bias exists without doing this kind of analysis based on the vivid, impactful stories about people's experiences with discrimination. Nonetheless, it does provide some useful insight. For instance, this analysis sought to provide multiple quantifiable, objective measures of bias. It provides a clearer definition of the risk itself and how we can measure that risk. Black and Hispanic drivers were more likely to be stopped, more likely to be searched? Less likely for the search to actually find illegal things this allows us to state with confidence that this risk is higher for certain individuals. A more precise insight and that likelihood of a police stop is higher for blacks and Hispanics, it also provides some insight into the impact of this risk and the potential effectiveness of risk responses. We can say with confidence that the source's bias, it's not just a perception of bias that would require its own type of response. In this case, an effective response requires not only addressing the thoughts, feelings and actions that stem from biased police stops, but also decreasing the stops themselves. Now in the next session, we'll blend these data driven insights with empathy driven insights and maybe see if we can refine our problem statements.
