Getting out the (swing) vote: evidence from Argentina

We test the Pivotal Voter Model in a real-world setting: a mid-term election in Argentina. To manipulate the perception of being more or less pivotal, we send letters to citizens that did not vote in the mandatory primary elections, making more/less salient how relevant their vote could be to change the result of the election.

We designed a randomized controlled trial with three arms, each of them consisting of 6,000 citizens. The treatment groups will receive letters by postal mails, while the control will not receive any letters.

The content of the treatment 1 letter will emphasize in a relatively neutral way the importance of voting. The content of the treatment 2 letter will explicitly say that the vote of the individual that received the letter is important to change the result of the election. It will also show a graph indicating that if less than half of the citizens that did not vote in the primary, showed and vote for the opposition candidate, they will win the election.

All the letters will be signed by the candidate of the opposition party. Our main outcome is turnout (at the individual level). The elections are for local counselors in the city of Berazategui (Province of Buenos Aires).

2021-11-15

2022-03-31

- Turnout (individual level, count): the citizen voted or not in the mid-term election (November, 2015) in the city of Berazategui.

We plan to analyze heterogeneous results by variables such as location and (presumed) political preference.

Turnout: takes a one if the individual voted and zero otherwise (count).

- Vote for Frente de Todos (incumbent):
- Vote for Juntos por el Cambio (opposition, polling booth level, count)

- Vote for Frente de Todos (incumbent): number of votes (count) for the incumbent party (Juntos por el Cambio) at the polling booth level

- Vote for Juntos por el Cambio (opposition): number of votes (count) for the opposition party at the polling booth level

There are 720 polling booths. We divided them into three groups:

Control: 240 booths, 25 individuals per cluster
Treatment 1: 240 booths, 25 individuals per cluster
Treatment 2: 240 booths, 25 individuals per cluster
Total: 720 booths, 18,000 observations in total

The control group does not receive letters. The treatment arms receive letters. In the two arms, the text aims at motivating individuals to vote. Although treatment 2 makes salient the fact that the individual may be pivotal.

Randomization done in office by a computer

There are two randomization units.
First, we randomized 720 polling booths into three groups:
- Control
- Treatment 1
- Treatment 2

Each polling booth has approximately 70 individuals that did not vote in the primary elections. We randomly selected 25 individuals per booth (out of 70).

Yes

720 clusters (polling booths)

18,000 in total (6,000 per arm)

Control: 240 clusters, 25 individuals per cluster
Treatment 1: 240 clusters, 25 individuals per cluster
Treatment 2: 240 clusters, 25 individuals per cluster
Total: 720 clusters, 18,000 observations in total

Power 80%, Confidence 95%, ICC 0.03 and assuming a prevalence of 10% in turnout (that is, 10% of the targeted individuals would vote independently of our intervention), we would be able to detect an effect of 2% of the individuals (0.5 individuals per polling booth).