There is a bit of buzz going around the Internet recently about an article published in the last couple of days by Nathan Newman which alleges that Google’s Adwords program is allowing advertisers to racially profile Google users.
In his piece, entitled ‘Racial and Economic Profiling in Google Ads: A Preliminary Experiment’, Newman conducts an “experiment” where he observed the ad results returned when certain name-subject combinations were entered in Gmail. He was attempting to find a link to African-American/Latino sounding names and negative advertising similar to some of the proven studies showing bias towards minority groups in hiring based upon names on resumes.
Newman’s experiment was simple: He picked nine names that were closely associated with a specific racial category. Two male and one female name were chosen for African-American, White, and Latino groups. He then included the names along with other simple search terms in the subject line of a Gmail account and recorded a screen capture of the ads that were returned.
When the name “Connor Erickson” was linked with “Arrested; Need a Lawyer”, ads for bankruptcy and criminal fraud attorneys showed up in the sidebar. But when the name “DeShawn Washington” was used instead, two different DUI attorney advertisements were returned. Similarly, less favorable ads were returned for the African-American and Latino names when ‘Education’, ‘Buying a Car’, and ‘Need a Job’ were used in the subject line. You can check out a full listing of Newman’s results here.
While some of the more innocuous of these results can probably be attributed to improvements needed in Google’s own massive search algorithm — there was one instance where a Latino sounding name with the ‘Need a Job’ subject was linked with a company called Salsa Labs (which is not Latino) — most of the credit/blame should rest with the advertisers themselves.
Through the Adwords program, advertisers choose which ads to display based on certain keywords that they designate. So in Newman’s experiment, with everything else in the subject line being equal except the names, the name itself then becomes the determining factor in which ad gets displayed.
One distinction to note in Newman’s experiment was that the some of the ad results could also be shown to have a location-based bias, independent of the name used. When the search was done in more affluent neighborhoods, results were noticeably different from when it was performed in poorer areas, suggesting that some geo-location profiling was happening as well.