Beyond Facebook: Other Online Networks
Lastly, we set up six Facebook profiles to check the impact of sexual-preference: a highly-sensitive personal attribute. Two profiles (male control) are for males interested in females, two (female control) for females interested in males, and one test profile of a male interested in males and one of a female interested in females. The age and
location were set to 25 and Washington D.C. respectively. Figure 6 plots the similarity scores for 1 week of data. While there is more noise in general, unlike in 4(c) there is
a measurable difference between the control and test pairs; we further manually verified based on ad content that this difference is qualitative in nature (e.g. ads for gay bars were never shown for the control profiles, but shown often for the test profiles). The median similarity score for gay women was 0.15 higher than for gay men, indicating that advertisers target more strongly to the latter demographic.
Alarmingly, we found ads where the ad text was completely neutral to sexual-preference (e.g. for a nursing degree in a medical college in Florida) that was targeted exclusively to gay men. The danger with such ads, unlike the gay bar ad where the target demographic is blatantly obvious, is that the user reading the ad text would have no idea that by clicking it he would reveal to the advertiser both his sexual-preference and a unique identifier (cookie, IP address, or email address if he signs up on the advertiser’s site). Furthermore, such deceptive ads are not uncommon; indeed exactly half of the 66 ads shown exclusively to gay men (more than 50 times) during our experiment did not
mention “gay” anywhere in the ad text.
Overall we find that while location affects Google ads, behavioral targeting does not today appear to significantly affect either search or website ads on Google. Location, user demographics and interests, and sexual preference all affect Facebook ads.”