Measuring discrimination is crucial to key questions in political economy, but studies of how group members discriminate against outsiders have focused on instances where membership is sharply defined. To study common and important instances of fuzzy group membership, I propose two approaches: one based on notions of distance, and the other based on measuring subjective perceptions of group membership. Taking regions in Mexico as an example, I use survey, field experiment and administrative data to test for regional bias in donations, in short-term migration, and in long-term migration. The bias appears only if region membership is conceptualized as fuzzy. Subjective perceptions provide the most consistent regionalization for capturing regional bias, and allow for testing several mechanisms. First, national appeals and a stronger attachment to the nation diminish regional bias. Second, the way individuals draw regions exacerbates regional inequality. The novel approaches can be applied to other identities with fuzzy demarcations.
Intervention
We placed donation petition cards at supermarket cashiers in nine states in Mexico in six supermarket chains. These donation petition cards are a standard way for our NGO partner to raise revenue in supermarkets.
Experimental design
The cards were in the cashiers between one and 31 days. The donation petitions varied on two dimensions. The first dimension varied the state where the beneficiaries lived. There were six beneficiary states. The second dimension varied whether the donation petition displayed the colors of the mexican flag on the frame of the donation petition card and on the name of the beneficiary state.
Design details
Randomization method
We randomized in a computer.
Randomization unit
We blocked on state and supermarket chain. Then we formed tuples based on the Mahalanobis ditance of the state's location, past sales and the supermerket size. The size of the tuple varied since different states had a different number of types of cards, which depended on the number of supermarkets in a state Tuple size ranged from 4 to 12.
Clusters
We planned to put cards in 581 supermarkets. We clustered at the supermarket level. There were 2 types of cards in each supermarket. Although we randomized at the half-supermarket level, we randomized at the supermarket level as a conservative assumption. We ended up placing cards in only 440 supermarkets due to logistical complications. Only 330 supermarkets provided us with usable data, since some chains did not participate.
Observations
We planned for 1162 half-supermarkets, and ended up observing 660 half-supermarkets.
Arms
The assignment of cards from the supermarkets we were able to obtain data from (330 supermarkets) is a follows:
76 half supermarkets had a State of Mexico beneficiary state and a prime.
77 had a State of Mexico beneficiary state and no prime.
50 had a Guerrero beneficiary state and a prime.
47 had a Guerrero beneficiary state and no pirme.
48 had a Puebla beneficiary state and a prime.
47 had a Puebla beneficiary state and a no prime.
56 had a Sinaloa beneficiary state and a prime.
55 had a Sinaloa beneficiary state and no prime.
52 had a Veracruz beneficiary state and a prime.
55 had a Veracruz beneficiary state and no prime.
46 had a Yucatán beneficiary state and a prime.
50 had a Yucatán beneficiary state and no prime.
We were interested in testing whether individuals had an in-region bias, so the treatment of interest is not the state on the card itself, but rather how far the beneficiary state was from the donor state. There were 94 half supermarkets where the beneficiary state was the donor state, 144 half supermarkets where the beneficiary state was one state away from the donor state, and 199 cards where the beneficiary state was two states away.
Primary outcomes
Revenue from card sales.
Partners
Name: Un Kilo de Ayuda
Partner_type: ngo
Url: https://www.unkilodeayuda.org.mx/
Cited by
No citation data available for this registration yet.