Commentary
Study Finds Racial Bias In Paid-Search Ads
- by Laurie Sullivan , Staff Writer @lauriesullivan, February 5, 2013
It's not black-and-white, though Latanya Sweeney might make it seem that way. The Harvard professor published a report suggesting Google searches expose
racial bias through AdSense. Depending on the perceived race of the names searched for, the ads that serve up siggest "statistically significant discrimination."
The paper, based on 2,184 searches of racially associated personal names, shows names typically associated with black people are 25% more likely to bring up advertisements related to criminal activity. Google says it doesn't racial profile. The platform relies on brand marketers to decide on keywords when targeting specific audiences on publisher sites.
Names like DeShawn, Darnell and Jermaine seem to generate ads suggesting an arrest between 81% and 86% of name searches on one Web site and between 92% and 95% on another. Two Web sites were used to support the paper. Names assigned to white children at birth, such as Geoffrey, Jill and Emma, generated more neutral copy. The word "arrest" appeared in 23% to 29% of name searches on one site and 0% to 60% on the other.
A Google search for the name "Latanya Sweeney" and "Latanya Lockett" also show ads suggesting arrests. While no arrest records were found for Sweeney, she did find records for Lockett. The ads appeared on google.com and on a publisher's site where Google serves up ads. The ads linked to instantcheckmate.com.
Instances similar to Sweeney's Claire McCaskill example make me wonder the percentage of a marketer's budget that goes in the trash, and whether algorithms need a bit more tweaking. She points to an ad for McCaskill, U.S. Senator from Missouri, where the word "arrest" appears below an ad for her U.S. Senate campaign.
Admittedly, a few names don't follow the pattern, according to Sweeney. Dustin, a name typically given to white babies, generates an ad suggestive of an arrest 81% and 100% of the time.
Laurie Sullivan is a writer and editor for MediaPost. You can reach Laurie at lauriesullivan@gmail.com.
Wonder if Sweeney asked the marketers where they got their name lists. Since arrest records are generally available under FOIA and are public records one could crunch through the records at a city by city level and then geotarget the names. Certainly if I were a search marketing manager working on an account that provided services to those recently arrested, one strategy would indeed be to predict hit rate based on historic patterns.