Affective Polarization: The Role of Beliefs

Participants will be randomly assigned to one of four experimental groups. Two groups focus on beliefs about intelligence: a baseline group and a treatment group that receives accurate information about partisans’ intelligence. The other two groups focus on beliefs about trustworthiness: a baseline group and a treatment group that receives accurate information about partisans’ trustworthiness.

Intelligence is measured by how many questions a randomly selected Democrat and Republican answered correctly on a cognitive test in our previous study. Trustworthiness is measured by how much a randomly selected Democrat and Republican returned in a trust game.

The cognitive test is created based on the expanded version of the Cognitive Reflection Test (CRT) (Toplak et al., 2014). In the cognitive task, participants are given 7 computational questions. They need to answer as many questions correctly as possible within 5 minutes. To get correct answers to these questions, they need to override prepotent intuitive responses that are incorrect and engage in further reflection. The number of correct answers measures participants’ intelligence. This measure is a popular tool to measure cognitive ability (Cipriani et al., 2020; Sofianos, n.d.) and has been proven to be correlated with other established measures, such as the Raven’s Progressive Matrices test (Corgnet et al., 2015).

In the trust game, there are a sender and a receiver, and each of them has a starting capital of $0.25. The game has two steps. In the first step, the sender decides whether to send $0.25 to the receiver or to keep that money for herself. If the sender chooses “keep”, both players earn their starting capital. The game ends. If the sender chooses “send”, the money gets tripled, and the receiver receives $0.75. The game enters the second step. The receiver then has $1 in total and need to choose to return any increments of 10 cents. Participants were asked to make decisions when they played each of the two roles, but we are only interested in their choices as receivers. In this task, how much participants choose to return as receivers measures their trustworthiness.

2025-07-14

2025-07-31

Participants' level of affective polarization and their willingness to have political conversations with co- vs counter-partisans

Affective polarization is measured by feeling thermometer. This measure asks respondents to rate Democratic/Republican voters on a scale ranging from cold (0) to warm (100). Affective polarization is the difference between partisans' feelings towards co-partisans and counter-partisans.

We elicit participants' willingness to have political conversations with different partisans through the Becker-DeGroot-Marschak (BDM) method (Becker et al., 1964). Specifically, we ask participants the minimum amount between 0 and $10 they are willing to accept for a chat about politics with a randomly selected Democrat and a randomly selected Republican, respectively. The outcome we construct is the difference between their minimum amounts for counter- vs co-partisans. This difference captures how polarized they are in having social interactions.

Social distance items
Participants' willingness to have non-political conversations with co- vs counter-partisans

We include social distance items as a complementary measure of affective polarization. The questions are adapted from Broockman et al. (2023). Participants will be asked whether they feel upset having their children married to counter-partisans, and if they feel comfortable being friends or neighbors with counter-partisans.

Similar to political conversations, we use the BDM method to elicit participants' willingness to chat with different partisans on non-political topics. We focus on the difference between their minimum amounts for counter- vs co-partisans.

Recruitment and eligibility:
Our experiment will be conducted on Prolific. We will recruit U.S. citizens who live in the US. We plan to recruit a balanced sample of participants who identify with the Democratic party and the Republican party. The experiment will be coded using Qualtrics.

Procedures:
1. Tasks: Participants complete the cognitive task or the trust game, depending on which condition they are assigned to, to be familiar with the tasks.
2. Prior elicitation: Participants report beliefs about Democrats' and Republicans' intelligence (or trustworthiness)
3. Intervention: Treatment groups receive correct information about Democrats' and Republicans' intelligence (or trustworthiness)
4. Posterior elicitation: we elicit participants' beliefs again to examine how much they update to our information
5. Measure affective polarization
6. Investigate downstream consequences of affective polarization: We elicit participants' willingness to accept (WTA) to chat with co- vs counter-partisans on different topics.
6. Chat realization: Based on the BDM results, participants have chats with their assigned partner on an assigned topic.

Prior Elicitation
At the beginning of the experiment, we will elicit participants’ prior beliefs about Democrats’ and Republicans’ intelligence (or trustworthiness). Specifically, in the intelligence condition, they will be asked to estimate how many questions on average the Democratic and Republican participants answered correctly in the cognitive test in our previous study. In the trustworthiness condition, they will be asked to estimate how much on average the Democratic and Republican participants returned as the second-mover in a trust game. To provide a reference point, we will inform participants about the overall average performance across all partisans. This allows participants to focus on assessing whether Democrats and Republicans performed above or below the average, rather than making completely uninformed guesses.

Intervention
The treatment group in each condition will receive correct information about the average performances of Democrats and Republicans in the tasks. In contrast, participants in the control group will not receive any information.

Posterior Elicitation
We then elicit participants’ posterior beliefs to assess how much treated participants update their beliefs in response to the provided information, serving as a manipulation check. To minimize consistency bias and demand effects, we will use a different elicitation method from the prior belief, helping participants distinguish between the posterior and the prior. Specifically, we will frame the task as a prediction exercise, asking participants to predict the performance of Democrats and Republicans in the current study. Instead of providing an estimate of the average, participants will make guesses about a randomly selected Democrat/Republican.

Affective Polarization Measurement
We will ask participants to complete the feeling thermometer and social distance items.

Willingness to Accept (WTA) to Chat
To better understand the behavioral consequences of affective polarization, we also measure participants’ willingness to engage in a 10-minute anonymous online text chat with a randomly selected participant. Each participant is presented with three topic areas as potential conversation starters: (1) contentious political issues, (2) contentious pop culture issues, and (3) contentious personal finance strategies. These topics are chosen to vary in required levels of cognitive effort and mutual trust. Pop culture topics are expected to require lower levels of both, while political and finance topics are expected to require higher levels.

For each topic, participants indicate the minimum amount (between $0 and $10) they would accept to chat with a randomly selected Democrat and a randomly selected Republican. Responses are incentivized using the Becker-DeGroot-Marschak (BDM) method (Becker et al., 1964).

Before entering their responses, the computer pre-selects a random offer between $0 and $10. After participants submit their minimum acceptable amounts, one topic and one partisan identity (Democrat or Republican) are randomly selected. Participants are told that these two chat partners are the same as the ones they make predictions about in the posterior elicitation. If the computer’s offer exceeds the participant’s stated minimum for that combination, the chat takes place and the participant receives the payment. If not, no chat occurs and no payment is given, and the study ends.

Randomization will be done using embedded features of Qualtrics.

Individual

No

1,600 clusters (individuals)

1,600 observations in total

400 participants in each experimental group.

Based on the data of our pilot studies, to observe a 0.2 SD treatment effect in affective polarization, we need at least 187 participants per group in the intelligence condition, and at least 394 participants per group in the trustworthiness condition