Title How do Brazilian voters trade school quality and other government outputs when evaluating municipal politicians?
Post date 09/14/2018
C1 Background and Explanation of Rationale

In a working paper (“Evaluating students and politicians: Test scores and electoral accountability in Brazil”), we find that voters who receive information about the performance of municipal schools are less likely to vote for the incumbent when such signals are positive than when they are negative. This goes against established theories of retrospective voting and electoral accountability.

Our findings stem from two research designs and two datasets from Brazil. First, in a regression discontinuity design (RDD) exploiting data for the municipal elections of 2008, 2012 and 2016 across Brazil, we study the effect of municipalities meeting a highly visible school quality target (IDEB) on electoral outcomes. Second, in a randomized controlled trial (RCT) in a sample of municipalities in Pernambuco, we study the effect of informing voters before the election about their municipality’s rank in the performance in an early student assessment (ANA) on individual vote choice.

Contrary to expectations, we find in both the RDD and the RCT that signals about school quality decrease rather than increase electoral support for the incumbent. In the RDD, we find that municipalities that meet the IDEB target have a lower vote share for the mayor than those that miss their target. In the RCT, we find that voters are less likely to vote for the mayor the better the performance of the municipal schools.

C2 What are the hypotheses to be tested?

The main hypothesis stemming from the RCT and the RDD is that voters are less likely to vote for the incumbent when they receive positive signals about the quality of public schools (H1). The main goal of the survey is to explain this counter-intuitive finding. We have developed a number of hypotheses that could explain it:

Trade-off hypotheses:
• H2: Education quality vs public jobs trade-off. Voters punish good performers because they perceive a trade-off between the quality of schools and the number of people employed by the municipality (in education or other sectors) or, more generally, a trade-off between government performance and jobs.
• H3: Education quality vs healthcare quality trade-off. Voters punish good performers because they perceive a trade-off between improving the quality of municipal schools and the quality of municipal clinics, which they value more.
• H4: Education quality vs poverty alleviation trade-off. Voters punish good performers because they perceive a trade-off between the quality of municipal schools and the municipal government’s efforts at reducing poverty, which they value more.
• H5: Education quality vs government responsiveness trade-off. Voters punish good performers because they see politicians who improve school quality as less responsive to their needs.

Mediated trade-off hypotheses:
• H6: People who give less importance to education, relative to other policy areas, are more likely to punish the incumbent when informed about good school performance because they perceive such information as a signal of the incumbent not prioritizing their interests.
• H7: Parents (H7a), parents of primary- and middle-school aged children (H7b), or parents of primary- and middle-school aged children enrolled in municipal schools (H7c) have a direct stake in education quality and thus respond positively to good school quality signals (i.e., in the opposite direction than hypothesized in H1-H5). We expect parents with a more direct stake in school quality to respond more positively to good school quality signals (i.e. our prior belief about the likelihood of these sub-hypotheses is higher for H7c than for H7b, and higher for H7b than for H7a).

C3 How will these hypotheses be tested? *

To test these hypotheses, we will conduct a short online survey with Brazilian respondents recruited via Facebook. The core of the survey is an information experiment, which will be preceded by some questions that help us collect pre-treatment covariates and simulate the conditions in which voters make decisions during electoral campaigns.

Before the experiment we will expose respondents to a battery of questions about how important they think different policy areas are for their personal interests. In particular, respondents will be asked to rank improvements in education, healthcare, employment in the municipal sector, economic activity in general, security, and fight against poverty. This will allow us to (i) collect data about saliency of different policy areas for voters; and (ii) prime respondents to think about the importance of different policy areas, and thus simulate the conditions in which voters decide their vote during electoral campaigns.

Immediately before the experiment, we will inform voters about IDEB (what it is and what it means for a municipality to meet its IDEB target), and ask them whether they had already heard about it.

Rather than use a vignette experiment with information about a hypothetical mayor, we provide real-world information about the performance of the mayor (or former mayor) of the respondent's self-reported municipality. Respondents will be randomly assigned to one of two groups: control (where they are not given information about whether their municipality met its target) and treatment (where they are informed about whether their municipality met its school quality target, as per the data published in 2016). Those in the treatment group read the following question: “In the year 2016, at the end of the term of Mayor [Name], [Municipality] [did not achieve / achieved] its IDEB target. Have you heard about that result?”

The treatment information is accompanied by two images designed to reinforce the message: a photograph of the mayor (obtained from the official repository of the Supreme Electoral Court) and a red cross or a green checkmark, depending on the municipality's IDEB performance. Those assigned to the control group do not read any version of this question or a substitute.

Next, all respondents, regardless of random assignment, read the following text: "Thinking about [Mayor], do you agree or disagree with the following statements about [his/her] term as Mayor of [municipality] during the period 2012–2016?" Response options are on a 4-point Likert scale ("totally agree", "partially agree", "partially disagree", "totally disagree"). To collect outcome data, we will ask about their agreement with the following statements, with the order of items randomized within respondents:
• As mayor, he/she hired a lot of public servants.
• As mayor, he/she improved municipal health clinics.
• As mayor, he/she helped people like you solve problems.
• As mayor, he/she reduced poverty in the municipality.
• As mayor, he/she improved municipal schools.
• As mayor, he/she invested a lot of money in education.

After these main outcomes, we will ask respondents whether they would vote for that mayor if they were to run for another term, again with response options being in a 4-point Likert scale from "great chance" to "no chance".

C4 Country Brazil
C5 Scale (# of Units) 3,000
C6 Was a power analysis conducted prior to data collection? Yes
C7 Has this research received Insitutional Review Board (IRB) or ethics committee approval? Yes
C8 IRB Number not provided by authors
C9 Date of IRB Approval not provided by authors
C10 Will the intervention be implemented by the researcher or a third party? Researchers
C11 Did any of the research team receive remuneration from the implementing agency for taking part in this research? No
C12 If relevant, is there an advance agreement with the implementation group that all results can be published? not provided by authors
C13 JEL Classification(s) not provided by authors