Affective Polarization and Social Identity: Preference-Based or Belief-Based?

In our encouragement design, we encourage people to increase their political participation, in order to artificially increase their political identity salience.

2022-11-01

2022-11-08

People's preferences towards co-partisans and counter-partisans
People's beliefs about co-partisans' and counter-partisans' personal traits

To measure preferences, we will ask participants to do an Implicit Association Test (IAT). This task asks them to match attributes with categories. The click time they spend measures their preference. The longer time it takes them to match, the less preferred the target attribute is.

To measure beliefs, we focus on two dimensions, warmth and competence. Previous studies demonstrate that these are the two dimensions people focus on when they judge others. More specifically, we want to elicit participants' beliefs about co-partisans' and counter-partisans' trustworthiness and intelligence, which correspond to these two dimensions.

We will ask participants to guess others' performances in a trust game and a cognitive task. In the trust game, how much a person chooses to return as a receiver in the trust game measure his trustworthiness. Therefore, a participant's guess on how much a co-partisan/counter-partisan choose to return measures her belief about the person's trustworthiness. The cognitive task consists of 7 computational questions. The number of questions a person can answer correctly within 5 minutes measures his intelligence. Therefore, a participant's guess on how many questions a co-partisan/counter-partisan answers correctly measures her belief about the person's intelligence.

People's political identity salience
People's perceived status of own party
The extent to which people display affective polarization

As a manipulation check, we will measure the strength of participants' political identity. We will use a word-completion task, where participants are given 5 incomplete words with only the first a few letters. Participants will be asked to complete each word with the first word that comes to their mind. The number of political words they fill in of measures the strength of their political identity.

To rule out the effect of perceived status, we will also ask about participants' perceived status of the political party they identify with. We will ask them to allocate 100 status points to the Democratic and the Republican Party. The more they allocate to one party, the higher status they believe that party has.

To measure the extent to which people display affective polarization, we will use two types of measures. The first one, a feeling thermometer, is a direct measure. We will ask people directly about how cold or warm they feel towards the two parties in general. They will be asked to mark their feelings on a 0–100 scale, where 0 means very cold, and 100 means very warm. The difference in their feelings towards own party and the opposing party measures the extent of affective polarization.

The second measure is an implicit measure, where we will ask the participants a set of questions which are designed based on the downstream consequences of affective polarization. The questions cover their attitudes towards relationships with counter-partisans, estimation of president's performance, and their opinions on social topics that Americans care the most. If a participant's answers to the questions are consistent with the direction of the downstream consequences, then the participants are regarded to show affective polarization.

We use an encouragement design to artificially increase people's political identity salience.

We use an encouragement design to artificially increase people's political identity salience. We assume that the increase in identity salience leads to the increase in the strength of political identity, thereby changing the extent to which participants display affective polarization. Our experimental design addresses two objectives: to measure how the extent of affective polarization changes due to the inauguration, to identify how participants’ preferences and beliefs shift due to the inauguration.

Our experiment will be conducted on Connect from Nov 2 to Nov 8, 2022. We will recruit 1,600 U.S. citizens who live in the US, identify with either the Democratic party or the Republican party.

Participants will be randomly assigned to one of two conditions: control and treatment. All participants will be asked to complete two surveys in this experiment. The first survey will be sent out to all participants on Nov 1. The second survey will be sent out on Nov 7, one day before the election day.

We will use an encouragement design. A group of participants will be incentivized to send nudging messages to 5 other people to encourage them to vote in the midterm election. The idea is to increase both quantity and quality of participants' political participation. Therefore, they will pay more attention to political information and their political identity will become more salient. To reduce the impact of selective attrition, we also incentivize the control group participants to send reminding messages about the winter time switch on Nov 6 to their friends. We will compare behaviors or participants who send political and nonpolitical messages. We assume that the change in salience affects the extent to which partisans display affective polarization. By investigating how people’s preferences or beliefs towards co-partisans and counter-partisans switch because of the salience change, we will be able to identify the underlying mechanism of affective polarization. This design allows us to estimate two causal effects by comparing the treatment to the control group. The first causal effect is the intent-to-treat effect. The second one is the local average treatment effect (LATE) obtained from an IV approach, where the treatment is sending the political messages, and the instrument is the monetary incentive of sending the messages.

Randomization done by Qualtrics. We use a block randomization method which is achieved by the randomizer function embedded in Qualtrics. Among the 1,600 participants recruited, 800 will be randomly assigned to the control group, and the other 800 will be randomly assigned to the treatment group.

Participants are randomly assigned to two different groups. The order of questions in the second survey will also be randomized.

No

No cluster

1,800 US citizens who are over 18 years old

800 participants in the control, 800 participants in the treatment group.

According to our previous experiment data (the experiment was pre-registered on AEA RCT registry, doi: https://doi.org/10.1257/rct.7070), to observe a change of 20% standard deviation (sd = 26), power of 0.8, and alpha of 0.05 (two-sided), we need a minimum detectable effect size of around 400. Assuming a 50% compliance rate of encouragement in our two-stage study, we need the minimum detectable effect size to be 800 per group.