Measuring racial ad delivery
A couple of points on this, from the paper:
[quote]The Facebook Marketing API allows advertisers to request a breakdown of ad delivery performance along a number of axes but it does not provide a breakdown based on race. However, for the purposes of this work, we are able to measure the ad delivery breakdown along racial lines by using location (Designated Market Area, or DMA*) as a proxy.
*Designated Market Areas  are groups of U.S. counties that Neilson denes as “market areas”; they were originally used to signify a region where users receive similar broadcast television and radio stations. Facebook reports ad delivery by location using DMA regions, so we use them here as well
DMA Region(s)  Aud. Records
Wilmington White 450,000
Raleigh–Durham White 450,002
Greenville–New Bern Black 446,047
Greensboro Black 446,050
Table 1: Overview of the four North Carolina custom audiences used to measure racial delivery. We divide the most populated DMAs in the state into two groups, and in each
group, create two audiences with ∼450,000 users of the same race. We then use the statistics Facebook reports about delivery by DMAs to infer delivery by race.
Not very accurate way to determine race and I am am quite sure race is more complex than black or white.