close× Call Us
close×
Title How to Improve Elite Perceptions of Public Opinion? Experimental Evidence
Post date 11/20/2018
C1 Background and Explanation of Rationale

To be responsive, elected officials often have to rely on beliefs about public will. Previous research suggests that representatives often have distorted images of the electorate. However, it remains unclear why political elites misperceive voter preferences, and how these constraints can be overcome. I argue that misperceptions result from a combination of unequal exposure and cognitive biases.

If elected officials rely on availability heuristics to gauge public preferences, inequalities in political voice can generate distorted images of the electorate. In turn, a systematic tendency to project their own preferences on others – known as social projection – may also hinder the development of accurate beliefs.

These arguments are tested in a survey experiment with local elected officials in Switzerland. Subjects will be asked to predict the outcome of two upcoming referendums in their constituencies after being assigned to different informational nudges designed to avoid availability heuristics and social projection. Referendum results provide local-level measures of voter preferences and allow me to test the effect of the nudges in reducing misperceptions. The study also offers an opportunity to understand what explains the capacity of elected officials to gauge public preferences.

C2 What are the hypotheses to be tested?

Reducing Misperceptions:
H1: Information about the composition of the electorate increases the accuracy of elite perceptions of public preferences
H2: Informing public officials about the tendency to project our own preferences on others increases the accuracy of elite perceptions of public preferences

Heterogeneous Treatment Effects:
H1b: The effects of providing information about the composition of the electorate [H1] should be higher among local officials whose prior image of the electorate was less accurate.
H2b: The effects of informing officials about social projection [H2] should be higher among local officials whose individual position on the federal initiatives differs from the majority opinion in their municipality.

Correlates of Misperceptions:
H3: When officials’ preferences differ from the majority of voters, perceptions of voter preferences should be less accurate among local officials who performed well in the previous election.
H4: Perceptions of voter preferences are less accurate for less salient issues.
H5: Perceptions of voter preferences are more accurate among full-time/professional politicians
H6: Perceptions of voter preferences are more accurate among politicians with higher levels of progressive ambition
H7: Perceptions of voter preferences are more accurate among delegates than trustees.

C3 How will these hypotheses be tested? *

Respondents and data collection
Public official survey: Follow-up online survey of 1000-2000 members of local governments in Switzerland recruited as part of the 2017 National Survey of Local Executive Members (Nationale Befragung der Gemeindeexekutivmitglieder).

Voter preferences: Federal referendum results disaggregated at the local level. Source: Swiss Federal Statistical Office.

Dates of administration
Public Official Survey: November 7 – November 24, 2018
Federal Referendums: November 25, 2018

Eligibility and exclusion criteria for participants
As part of the 2017 National Survey of Members of the Local Executive Members (N=7,825), respondents were invited to provide their email contacts if interested in taking part in a follow-up study about federal referendums. A total of 5,393 officials provided their contact information. An email invitation with a link to the Qualtrics survey was sent to this group on November 7, 2018.

Experimental design
A 3x1 experimental design tests the key expectations derived from the theory (H1. H1b, H2, H2b). A control group receives no additional information prior to the referendum predictions task (described below). A second group receives information about the composition of the electorate in their constituency (electorate composition treatment). A third group receives this same informational treatment plus a recommendation to avoid projecting their own preferences on the electorate (electorate composition + active debias treatment). The specific wording of each treatment is described below.

[Electorate composition treatment]
When thinking about the preferences of voters, if often helps considering the views of different groups in the constituency. To help you in this process, think about the residents of [CITY NAME] and give us your best guess to the following questions:

[For Horned Cow Initiative]
What share of the workers in [CITY NAME] works in the primary sector (including agriculture, forestry, or fishing)? (responses recorded from 0 to 100)
In the 2015 federal elections, what was the combined share of votes received by the SP and the GPS in [CITY NAME]. (responses recorded from 0 to 100)

[For Self-Determination Initiative]
In the 2015 federal elections, what share of voters in [CITY NAME] voted for the SVP? (responses recorded from 0 to 100)
What share of the population in [CITY NAME] was born outside of Switzerland? (responses recorded from 0 to 100)

NEXT PAGE

To give you some reference points, here are your answers along with the most recent data from the Federal Statistical Office in your municipality and in Switzerland as a whole.

[Table with respondent’s own perceptions of electorate composition alongside data from the Federal Statistical Office]

This information is intended to help you consider the preferences of the average voter in your municipality.

[Active debias treatment]
[After the electorate composition vignette] Finally, one last tip. Decades of research show that people tend to project their own preferences on others. Without noticing, we often overestimate approval for issues we support, while underestimating approval for issues we oppose. Try to take this into account when making your prediction.

Randomization
Survey participants will be randomly assigned to one of three groups: control group, electorate composition group, and electorate composition + active debias group. To avoid contagion effects, randomization was made at the official level and not at the official-initiative level. Randomization is performed with the R package blockTools. Multivariate continuous blocking is used to maximize balance between conditions on (1) respondents’ self-reported partisanship, (2) language (German, French, or Italian), (3) municipality size (population), local support for (4) the SVP and (5) the SP in the previous general election, and (5) canton. Individual-level information required for the randomization comes from the original survey from which respondents were recruited.

Primary and secondary outcome measures
[After being presented with the official description of the Self-Determination Initiative] It is always hard to anticipate the results of popular votes, but we would like to ask you to predict the share of voters in [CITY NAME] that will SUPPORT the Initiative for Self-Determination. Please give us your best guess from 0% to 100%:

[After being presented with the official description of the Horned Cow Initiative] It is always hard to anticipate the results of popular votes, but we would like to ask you to predict the share of voters in [CITY NAME] that will SUPPORT the Horned Cow Initiative. Please give us your best guess from 0% to 100%:

Referendum results disaggregated at the municipal level will be used to measure voter preferences in each municipality. The key outcome variable is the following measure of misperceptions: |Perceived % voter support – % voters supporting the referendum|.

Manipulation check
After the prediction tasks, respondents are asked the following question:
In the prediction tasks above, which tips did you receive? [Response options: No tips; Information about the electorate; Information about the electorate and a suggestion to avoid projecting my own preferences; The results from an old referendum.]

Analyses
H1 and H2 will be tested with two-sample t-tests. H1 implies testing the difference in misperceptions between the electorate composition group and the control group. H2 will be tested by comparing average misperceptions in the electorate composition + active debias group and the electorate composition group. These tests will be done for each federal initiative separately and by pooling both initiatives. The pooled analysis will be tested using OLS with clustered standard errors by local official and fixed effects for initiative. Finally, to test the sensitivity of the results to distance between the day when respondents took the survey and day of the referendum, I will estimate OLS models accounting for the number of days between submitting the survey and Election Day.

H1b and H2b will be tested using OLS. H1b requires interacting the treatment conditions with a measure of the average absolute distance between the two questions on electorate composition and their true figures. Respondents in the control group are asked the same set of questions at the end of the survey, after the prediction task, to allow for this comparison. H2b is tested by interacting the treatment conditions with a binary measure that takes the value of 1 if the respondent’s position on the initiative is aligned with the revealed preferences of the majority of voters in the same municipality, and 0 otherwise.

The hypotheses on correlates of misperceptions (H3-H7) will be tested using OLS with models estimated for each initiative and a model where both initiatives are pooled. Controls: fixed effects by state, municipality size (population), distance between survey submission and referendum day, years in office, political ideology, age, sex, and education. The key predictors required to test each hypothesis are described below:

[Past electoral performance – H3]
Electoral performance will be measured as a 5-point scale based on the following question:
Given your original expectations about the election, how did you feel about the result obtained? The election results were:
1. Very much below my expectations
2. Slightly below my expectations
3. Just as I expected
4. Slightly above my expectations
5. Very much above my expectations

[Issue saliency – H4]
This hypothesis is tested by comparing average absolute misperceptions for the Self-Determination Initiative (high salience) and the Horned Cow Initiative (low salience). This coding decision is based on a series of contacts with Swiss experts on direct democracy.

[Professionalization – H5]
The effect of professionalization will be tested with two variables: (1) a binary measure of whether the official receives a salary from the municipality, and (2) a 3-point scale where 0 is volunteer, 1 is part-time employee, and 2 is full time employee. The following questions are used to build these variables:
Are you volunteering or employed by the community?
1. No salary
2. Hired
Do you have another professional activity? (Yes/No)

[Progressive ambition– H6]
Progressive ambition will be measured as a 4-point scale based on the following question:
How would you describe your position regarding a candidacy for a different political office outside of your municipality in the future?
1. Definitely interested
2. I might be interested if there is a good chance
3. I would not rule it out completely, but I have no interest right now
4. I would absolutely exclude it

[Representative role – H7]
Officials’ self-reported representative role will be measured in two ways based on the importance given to the opinion of the population and the inner convictions of the respondent. A binary version of the variable takes the value of 1 if the respondent states that public opinion matters more than inner convictions, and 0 otherwise. A continuous version of the variable will simply be the difference between the two scales. The specific wording of the question is the following:
How important are the following factors when making important decisions? [Response options: Not important, Rather unimportant, Rather important, Important]
1. Opinion of the population

6. Inner convictions and rules of actions

Notes:
- Appropriate auxiliary quantities from our models will be computed and reported, including treatment effects by subgroup and differences in marginal effects between subgroups.
- I will compute all marginal effects appropriate to test the hypotheses of interest from any interaction models described above. In some cases, I may present treatment effects estimated on different subsets of the data for expositional clarity. If so, I will verify that I can reject the null of no difference in treatment effects in a more complex interactive model reported in an appendix when possible.
- Don’t know responses will be considered missing data for the factual belief outcome measures.
- Summary statistics for the sample will also be reported. I may also collect and report response timing data to capture respondent attention.
- The order of hypotheses and analyses in the final manuscript may be altered for expositional clarity.
- Where applicable, regression results for binary dependent variables will be verified for robustness using logit. Regression results for individual ordered dependent variables will be verified for robustness using ordered probit.

C4 Country Switzerland
C5 Scale (# of Units) 1000-2000 local officials
C6 Was a power analysis conducted prior to data collection? No
C7 Has this research received Insitutional Review Board (IRB) or ethics committee approval? Yes
C8 IRB Number Washington University in St. Louis IRB (#201810084); University of Geneva ethics commission (CER-SDS-12-2018)
C9 Date of IRB Approval Washington University in St. Louis: 11/05/18; University of Geneva: 10/18/2018
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? No
C13 JEL Classification(s) D72