A large-scale field experiment to disentangle sources of statistical discrimination in a social setting

We are conducting a correspondence study to investigate the extent of discrimination of immigrants in the context amateur football clubs in 15 Latin American countries.
We will randomly vary the content of the applications to test for different types of statistical discrimination.

2024-03-01

2024-03-31

Share of positive responses

There are two dimensions of treatment in our experiment. The first dimension is relative to the content of the email. We will randomize the content of the application message that clubs receive. The reference treatment (T0) is a core neutral message. For the other treatment groups, we include an additional message to the core of the email.

For the Individual treatment (T1), the following sentence will be added: ``When I was young, I played in the academy of Cremonese in Italy.'' For the amateur clubs in our sample, the treatment would probably signal a higher level than their average player, so we expect it to have an effect on individual statistical discrimination (Hypothesis 4).

For the Collective treatment (T2), the added sentence is ``I did part of my high-school curriculum in Spanish [Portuguese in Brazil] so I speak Spanish [Portuguese] well.'' The Collective treatment is designed to signal the player's ability to communicate with others in the team, i.e., to have an effect on collective statistical discrimination (Hypothesis 5).

Lastly, we include a treatment that combines the individual and collective treatments (T1+T2). The order of the treatments is randomized in the text of the email to avoid order effects.

The second dimension of treatment is relative to the origin of the applicants. In each country, census data are used to determine the countries used. We use the four largest countries in each of the following three regions: Latin America, Europe, and Asia. We used census data to get the two most common first and last names for children born in 2000.

Each club will be randomly placed in one of the 16 (4x4) treatment cells. All the analyses are between clubs. Between-clubs analysis means that the clubs do not receive similar-sounding applications from two different players, and we are restrained from identifying individual measures of discrimination of non-consenting participants.

Randomization done by computer

Club

Yes

15 countries

15000 clubs

Approximately 950 per treatment cell (content x origin)

0.05 standard deviation, corresponding to approximately 2 percentage points, using data from a pilot experiment in Argentina

Discrimination_Football_Latin_America_PAP