The Political Economy of Forest Loss

Our study investigates how information about local deforestation and employment impacts voter preferences in Brazilian municipal elections. We aim to understand how voters evaluate tradeoffs between environmental conservation and economic development when choosing candidates.

The intervention involves providing voters with an information treatment embedded in a survey, which highlights local forestation rates and local employment growth statistics. For both treatment dimensions we will provide information on whether levels increased, stayed the same, or decreased. The pre-election survey, conducted before the election, captures voting intentions, with the treatment group receiving the information and the control group receiving none. A follow-up survey after the election will assess whether this information influenced voting behavior. Through comparison, we aim to determine if exposure to environmental and economic data shifts voter preferences.

2024-09-28

2024-10-05

- We expect voting intentions to change for the incumbent if they receive information about forestation/employment. Pre-election intentions will be measured at the extensive margin (who they intend to vote) and the intensive margin (how likely they will vote for their preferred candidate). In the post-election survey we will ask who they voted for.

- For forested area the effect can go either way depending on preferences for/against deforestation. Someone in favor of (against) deforestation will be more (less) likely to vote for incumbent if forested area decreased. Someone in favor of (against) deforestation will be less (more) likely to vote for incumbent if forested area increased.
- For the information provided on employment growth in the local area
- We expect effects to be greater if information received differs from previously stated perceptions
- We expect effects to be greater for those with stronger preferences in favor of/against deforestation as indicated by WTP/WTA elicited before treatment

1) Most important issues in the election
2) Open text of why they want to (pre-election survey) and why they did (post-election survey ) vote for a candidate
3) We will conduct a lottery where the choices in terms of donations to three charities of one respondent will be implemented by the research team. Out of the three possible charities, one is in favor of deforestation, one fights for forest preservation, and one for education of children in favelas.

1) For the most important issues in the election we provide 5 options, two of which are clearly related to the treatments. We hypothesise that providing information might make these issues more salient.
2) We will use embeddings from a pre-trained transformer model (Universal Sentence Encoder or more modern model, such as Gecko) to study the open text written by respondents. The hypothesis is that providing information might make embeddings related to the treatment information more salient. We will also reduce the high dimensionality of these embeddings (e.g. 512) to 2-3 dimensions using TSN-E to study whether the distributions of responses differ for treated and untreated, again in relation to the cross-derivates with preferences for/against deforestation and perceptions. We will also do a simple keyword analysis using TF-IDF for descriptive purposes. This is meant find insights about mechanisms through exploration.
3) The hypothesis is that receiving information about forest growth might shift donations to pro/con forestation groups depending on pro/con pre-treatment preferences. This is not a key measure, but meant to study salience and whether preferences shifted, or only accountability of politicians changed decisions.

Public description:
Our study uses a randomized controlled trial (RCT) design to evaluate how information about local deforestation and employment affects voter behavior in municipal elections. Participants will be randomly assigned to either receive additional information related to local environmental and employment conditions or not. Surveys will be conducted before and after the election to measure changes in voting intentions and behavior. This design allows us to explore the influence of targeted information on electoral choices.

Hidden description:
Our study employs a two-wave randomized controlled trial (RCT) design. In the first wave, a pre-election survey is conducted with a representative sample of voters in local municipalities, capturing their baseline voting intentions and preferences. Participants are randomly assigned to two groups: a treatment group, which receives an information intervention highlighting local deforestation rates and employment statistics, and a control group, which does not receive this information.

The information treatment aims to provide voters with detailed data on how environmental conservation efforts (forest preservation) may trade off with local employment opportunities, particularly in regions where deforestation contributes to economic activities such as agriculture or logging.

The second wave, a post-election survey, is conducted after the municipal elections to assess whether the provided information led to shifts in voter preferences or electoral behavior. This follow-up survey will be brief, asking participants whether they changed their vote or maintained their initial voting intentions based on the new information they received.

We will analyze the differences in voting behavior between the treatment and control groups to determine the causal effect of the information treatment on voter decision-making. This analysis will help us understand how voters weigh environmental and economic factors when selecting candidates in local elections.

Randomization occurs within the Qualtrics survey software

individual

No

5000

5000 Pureprofile participants

1000 receive information about forest growth
1000 receive information about employment growth
1000 receive information about both
2000 receive no information (control group)