x

NEW UPDATE: Completed trials may now upload and register supplementary documents (e.g. null results reports, populated pre-analysis plans, or post-trial results reports) in the Post Trial section under Reports, Papers, & Other Materials.
Identity and Polarization: Evidence from an Online Experiment
Last registered on July 19, 2020

Pre-Trial

Trial Information
General Information
Title
Identity and Polarization: Evidence from an Online Experiment
RCT ID
AEARCTR-0005046
Initial registration date
December 05, 2019
Last updated
July 19, 2020 5:00 AM EDT
Location(s)

This section is unavailable to the public. Use the button below to request access to this information.

Request Information
Primary Investigator
Affiliation
LMU Munich
Other Primary Investigator(s)
PI Affiliation
LMU Munich
PI Affiliation
LMU Munich
Additional Trial Information
Status
In development
Start date
2019-12-05
End date
2021-03-01
Secondary IDs
Abstract
We test a theoretical model by Gennaioli and Tabellini (2019) on identity and polarization: when a new political dimension becomes salient, individuals choose new identities along this dimension and, due to a postulated tendency to stereotype both in- and out-groups, individuals shift their beliefs and attitudes in the direction of new stereotypes. This mechanism yields a clear prediction for polarization: an increase in the salience of one political dimension should lead to more polarized beliefs and attitudes along this dimension, while at the same time leading to less identification and reduced polarization along non-collinear, now relatively less salient, political dimensions. To test this prediction, we conduct an incentivized online survey experiment in which we randomly split participants into three groups: 1) control (C), 2) redistribution (R), and 3) openness (O). To vary the salience of a particular policy dimension, we expose each group to a different simulated news feed based on recent real-world news items from online media outlets. As outcomes we measure stated as well as revealed attitudes and beliefs about redistribution and openness (immigration, globalization, and climate change) to capture ideological as well as affective polarization.
External Link(s)
Registration Citation
Citation
Esguerra, Emilio, Leonhard Vollmer and Johannes Wimmer. 2020. "Identity and Polarization: Evidence from an Online Experiment." AEA RCT Registry. July 19. https://doi.org/10.1257/rct.5046-1.1.
Experimental Details
Interventions
Intervention(s)
We run an incentivized online survey experiment in Germany. We recruit participants through a professional survey company to ensure that our sample is representative of the voting-age population with respect to gender, income, and age. Our intervention is designed to shift the relative salience of four political dimensions: redistribution, globalization, immigration, and climate change.
Intervention Start Date
2019-12-06
Intervention End Date
2019-12-16
Primary Outcomes
Primary Outcomes (end points)
Our set of primary outcomes consists of participants' views on the following four political dimensions: redistribution, globalization, immigration, and climate change. Based on participants' views on these four dimensions, we construct measures of ideological polarization for each dimension.
Primary Outcomes (explanation)
Views on our four political dimensions are measured as follows:

Redistribution: Participants’ attitudes and beliefs are captured by the answers to the following questions:

1. Are differences in income between the poor and the rich problematic?
2. Should workers refrain from demanding wage increases to secure economic competitiveness?
3. Should the rich pay higher taxes to promote equality?
4. What should be the income tax rates for the rich and the poor?
5. How should the federal budget be split between different policy domains?

Immigration: To measure participants’ stated attitudes and beliefs in the domain of immigration, we ask the following questions:

1. Is the number of immigrants problematic?
2. Does immigration enrich German culture?
3. Are German Muslims different than other German people?
4. Should the government care equally about immigrants and natives?
5. When should immigrants be eligible for benefits?
6. When should immigrants be able to get citizenship and the eligibility to vote?
7. When are immigrants considered to be "truly" part of the country?

Globalization: To measure participants’ stated attitudes and beliefs regarding globalization, we consider answers to the following questions:

1. Does international competition harm national/local businesses?
2. Should German interests enjoy priority over other countries' interests?
3. Do international organizations curtail German sovereignty?
4. Does Germany benefit from the European Union?
5. Do participants view themselves as cosmopolitan?
6. Does globalization have a negative impact on Germany?
7. Should imports of certain goods be restricted?

Climate change: To elicit participants’ attitudes and beliefs in the domain of climate change, we collect answers to the following questions:

1. Are sufficient actions undertaken to protect the climate?
2. Does climate change constitute a serious problem?
3. Do participants feel responsible for protecting the climate?
4. What are participants' beliefs about the impact of climate change?
5. Do participants think that climate protection should enjoy priority over economic growth?

Based on participants' responses to these survey questions, we construct our main outcome variables. Our main hypotheses on ideological polarization pertain to the dispersion of attitudes along the different political dimensions. We therefore summarize the various survey items outlined above into summary indices representing each dimension. In particular, we will construct the following summary indices for our main analysis:

A: Four separate indices for the following dimensions, using all variables for each dimension as outlined above:
1. Redistribution index
2. Immigration index
3. Globalization index
4. Climate change index

B: A composite openness index based on the indices for immigration, globalization, and climate change. We further split up the openness dimension to account for the possibility that individuals do not view it as a homogenous dimension. We check the correlations of the immigration, globalization, and climate change indices with the redistribution index in the control group sample to investigate inhowfar subjects view these dimensions as separate. We construct alternative version(s) of the composite openness index, where we omit the dimension(s) from the set of immigration, globalization and climate change that is/are most strongly correlated with redistribution. We describe how we construct the summary indices in more detail in our pre-analysis plan.

We use several different strategies to measure changes in ideological polarization based on our indices for redistribution and openness (including the sub-indices): First, our main measure for ideological polarization for each index is the squared deviation from the control mean, which reflects to what extent an individual’s views on each of the political dimensions differs from the views of the average respondent in the control group. Second, we measure changes in ideological polarization by calculating the absolute deviation from the control mean. Third, we also compare variances between our treatment groups and our control group. We spell out how we construct our main outcomes in more detail in our pre-analysis plan.

We use these measures to test, among others, the following two main hypotheses:
1. An increase in the relative salience of either the redistribution or the openness dimension should increase polarization along this dimension (“direct effect”).
2. An increase in the relative salience of either the redistribution or the openness dimension should decrease polarization along the other dimension (“cross effect”).

To test our main hypotheses on the 'direct effect' and the 'cross effect', we compare outcomes between each treatment group and the control group. Of main interest are the comparisons of the redistribution and the openness groups with the control group, but we will also compare each sub-group of the openness group (i.e. immigration, globalization and climate change) with the control group separately.
Secondary Outcomes
Secondary Outcomes (end points)
Our secondary outcomes fall into five categories:
1. Revealed attitudes
2. First-stage relationship
3. Mechanisms
4. Affective polarization
5. Further outcomes
Secondary Outcomes (explanation)
Our secondary outcomes fall into the five aforementioned categories. In the following, we describe every category in more detail:

Revealed attitudes: To move beyond stated attitudes, we give subjects the opportunity to donate to charitable organizations, one for each political dimension (redistribution, globalization, immigration, and climate change). For the amounts donated we also calculate the above polarization measures and conduct analyses analogously to our analysis of the indices outlined above, i.e. test for the 'direct effect' and the 'cross effect'. See our pre-analysis plan for more details on the non-profit organizations and our analyses.

First stage: Our experiment builds on the following first-stage relationship: exposing individuals to a news feed highlighting one of our political dimensions of interest should increase the salience of these dimensions. To assess the existence of this first-stage relationship empirically, we ask participants which of the four political dimensions (redistribution, immigration, globalization, or climate change) they consider to be the most important problem Germany is currently facing. In our analysis, we then test whether individuals in either of our treatment groups are more likely to state the political dimension highlighted in this treatment group as an issue of primary political importance.

Mechanism: We also collect data on the underlying mechanism, i.e. that a shift in the relative salience of political dimensions induces a change in which social groups individuals identify with. We ask all participants whether they consider themselves to be a supporter, an opponent or neither of more redistribution, more immigration, more globalization, and more climate protection. As an example, we consider the case of the redistribution dimension; our approach for the other (sub-) categories is identical. We test whether individuals in the redistribution group are more likely to state that they are either a supporter or an opponent of more redistribution than those in the control group ('direct effect'). Furthermore, we test whether individuals in the redistribution group are less likely to state that they are a supporter or an opponent of more openness than those in the control group ('cross effect').

Affective polarization: To measure affective polarization, we implement a ‘feeling thermometer’ to capture the extent of affective polarization as proposed by Iyengar et al. (2019). We ask participants to rate both supporters and opponents of each of the four dimensions (redistribution, globalization, immigration, and climate change) on a 101-point scale from -50 (cold) to +50 degrees (warm). The measure of affective polarization for individual i is then computed as the difference between the score an individual gives to the group she identifies with herself (in-group) and the score she gives to the out-group. Again, we test for the existence of the 'direct effect' and the 'cross effect'.

Further outcomes:
1. We may also consider polarization measures based on Esteban and Ray (1994).
2. We also elicit individuals’ partisan political preferences as well as past and future intended voting behavior. We use participants' answers to test whether highlighting the salience of one political dimension induces individuals to vote for a party catering to the highlighted dimension. Again, we consider both the 'direct effect' as well as the 'cross effect'.
3. Finally, we also ask respondents to classify themselves with respect to the traditional political terminology on a political spectrum ranging from very left-wing to very right-wing.

We provide more details on how we construct and use our secondary outcomes in our pre-analysis plan.
Experimental Design
Experimental Design
In the online survey, we elicit participants’ views on four political dimensions: redistribution, globalization, immigration, and climate change. In addition, we collect detailed information on participants' demographics, such as age, gender, income, or educational and employment status as well as political outcomes such as voting behavior.

As part of the survey, we experimentally vary the relative salience of the aforementioned political dimensions. More specifically, we employ the following five-arm design: first, we randomly allocate participants to either the 1) control group (C), 2) the redistribution group (R), or the 3) openness group (O). Each group is then exposed to a different stimulus varying the salience of a particular political dimension. More specifically, group R receives an upward shift in the relative salience of the redistributive conflict dimension; group O an upward shift in the relative salience of the openness dimension; while the control group C receives a placebo. We further split the openness group into three subgroups which receive a stimulus either increasing the salience of immigration (O1), globalization (O2), or climate change (O3). This leaves us with a total of five different groups.

Our treatment shifts the relative salience of either of the two political dimensions, i.e. redistribution or openness. We implement these shifts in the relative salience of political dimensions by exposing participants to simulated news feeds based on recent real-world news items from German online media outlets.

For further details on the experimental design, please refer to our pre-analysis plan.
Experimental Design Details
Not available
Randomization Method
Randomization done in office by a computer.
Randomization Unit
Randomization takes place at the individual level.
Was the treatment clustered?
No
Experiment Characteristics
Sample size: planned number of clusters
As randomization is conducted at the individual level, the number of clusters equals the number of individuals.
Sample size: planned number of observations
We expect ca. 5,150 individuals to complete the survey.
Sample size (or number of clusters) by treatment arms
We randomly assign participants with the following probabilities to five different groups:
1) control (C): 1/3
2) redistribution (R): 1/3
3) openness (O): 1/3
3.1) immigration (O1): 1/9
3.2) globalization (O2): 1/9
3.3) climate change (O3): 1/9

For clarification, individuals have a probability of one third of being assigned to the openness group. Within the openness group, each participants is randomly assigned either to the globalization, immigration or climate change condition. Hence, participants effectively have a chance of 1/9 of being assigned to the globalization, immigration or climate change condition, respectively.
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
Supporting Documents and Materials

There are documents in this trial unavailable to the public. Use the button below to request access to this information.

Request Information
IRB
INSTITUTIONAL REVIEW BOARDS (IRBs)
IRB Name
Ethics Commission, Department of Economics, University of Munich
IRB Approval Date
2019-12-04
IRB Approval Number
2019-16
Analysis Plan

There are documents in this trial unavailable to the public. Use the button below to request access to this information.

Request Information