Twitter Users of Both Sex More Likely to ‘Follow’ White Women

Between June and October of 2009, digital agency Box UK conducted two sequential social experiments to test how Twitter users reacted to being followed by strictly controlled test accounts. The results strongly suggest that given a choice of following black and white people of either sex, Twitter users are more likely to ‘follow’ white women, and least likely to follow black women.

The first experiment was conducted between 3 June and 20 Aug 2009. Four almost-identical Twitter accounts were created with the same name, biography, location and profile colors. All four accounts posted the same tweets, simultaneously.

Tweets were manually posted, at the rate of one or two per day, on a range of topics that couldn’t be easily used to identify their sex or race; including design, culture and art.

The only difference between each account was the profile photograph: a realistic looking stock photograph of a black woman, black man, white woman and white man; each of similar age and ‘attractiveness’.

Over the period of the experiment, the accounts followed ‘random’ Twitter users, at a rate of a few hundred per day. Users were selected by randomly choosing from the millions of followers of the Top 1000 Twitter accounts.

The accounts also followed-back anyone that followed them, and periodically removed any friends who didn’t reciprocate the follow, in order to maintain a realistic following-to-follower ratio.

At the end of the experiment, the Black Female had amassed 812 followers, the White Female 935, the Black Male 824 and the White Male 825.

This hinted at a preference for the White Female account, who attracted a statistically significant additional number of followers to the other accounts.

To confirm and expand on these results, a second experiment was conducted with an additional set of controlled parameters, between 6 August and 1 October 2009.

In this experiment, five accounts were created. As before, each had a similar profile and posted the same tweets. Rather than using photographs, which could have skewed the first results based on subtle variations in attractiveness, in this experiment illustrated ‘avatars’ were created for each account. The female avatars were exactly the same except for skin colour; similarly for the male avatars. A fifth ‘ambiguous’ avatar was created using an abstract blue shape that could not be used to identify sex or race.

Random Twitter users were again selected from followers of the Top 1000, which each account randomly followed, at the rate of a couple of hundred users per day. This time, potential users were first filtered to remove any obvious ‘spam’ accounts: those that had unrealistic following/follower ratios, those that had strange tweet timing patterns, etc.

The pool of random users was also compared against the US Census list of male and female first names, so that they could be assigned a sex (only users whose real first names could be assigned a sex were used). The five test accounts subsequently followed an equal number of male and female users.

At the end of the experiment, the Black Female had amassed 513 followers, the White Female 755, the Black Male 631, the White Male 562 and the Ambiguous account 722.

We further filtered these results, as a random element of ‘re-tweeting’ could have affected the numbers: if the followers of one test account ‘re-tweeted’ them, then they will have been given a greater exposure than the other users, potentially skewing their follower numbers upward.

To remove this skew, we re-calculated the follower numbers using only those users that each account had originally followed from our random pool, i.e. not including any subsequent users that had followed our test accounts ad-hoc.

Using this metric, the Black Female received 311 reciprocal followers, the White Female 522, the Black Male 333, the White Male 375 and the Ambiguous account 504.

This distribution also holds when the data is sub-divided into male followers and female followers for each account, showing that both sexes are most likely to follow White Female or Ambiguous accounts, and least likely to follow Black Females. We can also deduce that on average, female twitter users are 30% less likely (than male users) to follow a request from a stranger.

Conclusion

Twitter is an increasingly important platform for conducting social experiments, with its ability to tap-into and measure human communication and behavior on a massive scale. As the platform grows, we expect to see businesses and academics harnessing this capability to ‘invisibly’ survey the real behavior and reactions of people, enabling a new wave of social research and customer intelligence.