An experiment about discrimination

We create fictitious accounts on Twitter that resemble humans and that claim to be PhD students with interests in topics related to economics. The accounts differ on three observable characteristics: gender (male or female), race (black or white) and university affiliation (highly ranked or not). The accounts follow approximately 100 subjects who are part of the "EconTwitter" community. After being active for twelve days, we measure how many follow backs each bot account obtained.

2022-05-26

2022-12-20

See pre-analysis plan

See pre-analysis plan

We create fictitious accounts on Twitter that resemble humans and that claim to be PhD students with interests in topics related to economics. The accounts differ on three observable characteristics: gender (male or female), race (black or white) and university affiliation (highly ranked or not). The accounts follow approximately 100 subjects who are part of the "EconTwitter" community. After being active for twelve days, we measure how many follow backs each bot account obtained.

Our subject pool is composed of all Twitter accounts that either tweeted or re-tweeted a status containing the term "#EconTwitter" between January 1st and February 28th, 2022 (approximately 15,000 accounts). Then, we will create 240 fictitious accounts claiming to be PhD students and who differ in their gender, race and university affiliation (overall, we have 30 accounts of each group). Each account will be active for a period of twelve days and follow approximately 100 subjects from the ‘EconTwitter’ sample. Our main outcome is whether or not
each subject followed back the bot accounts. Given that the accounts are identical in all respects but in the three dimensions we varied, differences in follow-back associated with the groups will be due to discrimination.

The experiment will be conducted in 30 waves of twelve days. In each wave, 8 accounts will be active. They will follow their respective subjects on the first day of activation, and we will measure follow-backs twice a day.

The subjects followed by each bot account will be defined randomly. Specifically, we perform block randomization as a way to improve balance. For the block randomization, we will use the following variables: gender (male, female, missing); profession (professor; graduate student; other; missing); the number of followers (above or below median). This gives us 24 strata. We will sample randomly from within each stratum, assigning the same proportion of users in each stratum to each bot account. Specifically, each bot account will be assigned to approximately 100 accounts to follow.

Randomization done in office by a computer

Individual

Yes

240 accounts

24000 subjects (individuals)

3000 subjects

PAP_DiscriminationTwitter_Aug11.pdf

This paper documents discrimination in the formation of professional networks among academic economists. We created 80 bot accounts that claim to be PhD students differing in three characteristics: gender (male or female), race (Black or White), and university affiliation (top- or lower-ranked). The bots randomly followed 6,920 users in the #EconTwitter community. Follow-back rates were 12 percent higher for White students compared to Black students, 21 percent higher for students from top-ranked universities compared to those from lower-ranked institutions, and 25 percent higher for female compared to male students. Notably, the racial gap persists even among students from top-ranked institutions.

Ajzenman, Nicolás, Bruno Ferman, and Pedro C. Sant'Anna. 2025. "Discrimination in the Formation of Academic Networks: A Field Experiment on #EconTwitter." American Economic Review: Insights 7 (3): 357–75.

https://doi.org/10.1257/aeri.20240298