The abstract states "[p]eer transportation companies such as Uber and Lyft present the opportunity to rectify long-standing discrimination or worsen it", but the study appears (from my quick glance) not to have directly compared Uber, Lyft, and taxis with the same methodology. The (scant) evidence regarding likely racial discrimination by taxis seems to indicate that Lyft and Uber are much less discriminatory.
That matches my intuitive opinion about it. And it only gets worse with the sharing economy. As soon as you choose your peer, it introduced human judgement.<p>One interesting thing would be to measure which race is the most racist, so we can sensibilize the right population.<p>Another interesting thing would be to have objective metrics to judge whether a transaction went well (whether the passenger was at the predefined location, on time, and whether the car was used with care).<p>I'll say it: Sometimes racism is based on a correct evaluation of the risk. So what can we do to diminish the risk?
Consider a world where you are assigned a random phrase (i.e. "Correct Horse Battery Staple") when you request a ride. The driver is never shown a name. When they are shown the set of people they could pick up, they are not shown a photo or passphrase until they accept the ride. As soon as they accept your ride, they are shown your passphrase and your photo (the former to ensure that they are picking up the right passenger, and the latter to ensure that a previously vetted passenger isn't cheating and providing their account to someone else / buying a ride for someone else).<p>Wouldn't this go a pretty long way towards removing obvious sources of discrimination?
Interesting, but the standard deviations of the metrics are as large or larger than the mean in every case. Also, there is a lack of trend between independent replicates.<p>So, very noisy data to say the least. I think their taxi figure in the appendix highlights the problem of discrimination much more clearly than their analysis of this dataset.