Understanding voter turnout among the youth: evidence from Chile

We sent three text-messages before the Chilean presidential election of November 19, 2017, and three more messages before the runoff of December 19th. The messages attempted to encourage youngsters to vote, by either providing them relevant information or by appealing to their desire of being represented in government. More specifically, we sent the following messages:

Text Message 1
Hi [NAME]! In the last election, just X% of the young people in your municipality voted. Find out where you vote here: [link]

Here, the idea was to see whether the information of turnout rates in the past election could affect the decision to vote. It is worth noting that in Chilean municipal election of 2016 (the "last election" alluded in the message), turnout rates among the youth were very low. In this sense, we thought that the less youngsters voted in the past, more compelled they would feel to vote in the next election.

Text Message 2
Hi [NAME]! If the young people don’t vote, others are going to choose instead! Participate in the elections this Sunday.

In this message, we wanted to appeal to their desire of being represent in government decision-making, by participating in the election.

Text Message 3
Hi [NAME]! Today is election day! At what time do you go vote?

In this third message, the idea was to prime voters to think about when they vote, and therefore, plan the day accordingly.

For the runoff of December 19, we sent the same messages, but to half of the treated units, in order to see whether there is an effect of receiving messages for both elections and to analyze if the messages of the first-round had a durable effect towards the runoff.

As an additional exercise, we implemented a survey-experiment as follows. In an ex-post survey (after the runoff) we randomized the way of asking voters wether they actually voted, along the lines of Della Vigna et al (2017). Essentially, we reduced costs of lying to one group, by telling them that the survey was going to be much shorter if they did not vote. We aim to compare their self-reported answers with the actual administrative data on turnout, which will provide us a measure of how much voters lie (if they do) about their electoral participation.

On October 25th, 2017 we sent a list of 1,500 subjects of the treatment group to our counterpart in Ekhos, a market research consultant company. The list included the name and the national id number for each person. Then, Ekhos obtained the cell phone number of 714 cell phone numbers through Equifax, a consumer credit reporting agency. Afterwards, Ekhos proceeded to validate these numbers, by calling each of them and implementing a five-question survey about activities during leisure time. At the end of the process, Ekhos was able to validate 180 cell phone numbers, meaning that in 180 cases, the cell phone corresponded to the person assigned to the treatment group.
On November 17th, we sent the first text message approximately at 4 pm. The next day, we sent the second text message at 11:00 am. On election day, we sent the message at 10:0 am. Only 5 people decided to unsubscribe from the list. For the runoff of December 17, we also sent three text messages, following the same pattern that in the first round: a message two days before the election, another the day before and the last in the morning of election day.

For the post-election survey, we needed to obtain the cell phone number of the subjects in the control group. Therefore, on November 29th, we sent on to our counterpart in Ekhos a list with 1507 subjects from the control group, and they followed the same procedure than for the treatment group to obtain and validate the phone numbers. Between December 21th and December 28th, the company was able to contact 81 subjects from the treatment group and 105 subjects from the control group. They sent us the data of this survey on January 10th.

2017-11-19

2017-12-17

1) Individual-level voter turnout: indicator variable equals to 1 if the subject voted in the 2017 Chilean presidential election and 0 otherwise.
2) Self-reported voter turnout: indicator variable equals to 1 if the subject reported having voted in the 2017 Chilean presidential election and 0 otherwise.

Self-reported vote choice: set of indicator variables equal to 1 if the subject reported having voted for a particular candidate and 0 otherwise.

We designed and implemented a field experiment to test the effectiveness of a set of low-cost behavioral tools in increasing youth turnout.
Our sample of 3,006 youngsters was composed by voters who live in the Metropolitan Region of Santiago, Chile. Eligible voters were registered to vote in new ballot boxes in the presidential election of 2017 in six municipalities, stratified by electoral booth. The six municipalities were located throughout the city, ensuring socioeconomic diversity. We randomized half of the individuals to a treatment group and half to a control group. All registrants had a mobile phone number, but whether those were valid or not would unravel only after trying to reach them out with our SMS-intervention.

Randomization done by a computer (with replicable code).

Individuals, stratified by municipality and electoral booth.

No

20 ballot boxes.

3006 voters.

1499 individuals the treatment group, 1507 in the control group.

We assumed a two-sided hypothesis test with a 5% significance level, a desired power of 80%, and that equally-sized groups to compute the sample size required to detect a minimum detertable effect size of five percentage points.