Opting Out of Political Discussions

Abstract

Methodological limitations have hindered our ability to understand the conditions that make individuals seek or avoid political discussions. We introduce a methodological approach to assess communication preferences in contexts where these choices are difficult to measure. We conduct three experiments to examine how the characteristics of the people in a discussion, as well as its topic, influence an individual’s “price” to participate. Participants indicated how much they would need to be compensated to participate in a short discussion about a randomly assigned topic (political or nonpolitical) under different group compositions (co-partisans, out-partisans, or a mixed group). We find that individuals demanded significantly more compensation to engage in a discussion with out-partisans than with co-partisans, for both political and non-political topics.

Keywords:

• political discussion
• polarization
• experiments
• discussion networks
• avoidance

Disclosure statement

No potential conflict of interest was reported by the author.

Supplemental material

Supplemental material for the article can be accessed on the publisher’s website at https://doi.org/10.1080/10584609.2018.1561563.

Notes

1. See Frimer and colleagues (2017) for an application in which individuals give up the chance to win money to avoid something they do not like: unpleasant cognitive dissonance from hearing an argument from someone who disagrees on a political issue.

2. In other work (Carlson & Settle, working paper), we explore more of the motivations, such as accuracy motivations, that have been shown to influence people’s willingness to seek out discussion with those with whom they disagree, or seek social information from a more ideologically diverse set of views (Pietryka, 2016).

3. In all studies our dependent variables come from participants typing in their demand amounts. We chose to allow participants to freely type their compensation demands because we did not want to impose an upper limit with a drop-down menu or other form of a closed-ended question. However, this means that simple typos are observationally equivalent to extreme outliers. For example, if a participant typed 100 as his or her compensation demand, we have no way of knowing whether he or she intended to demand $100 in compensation for participation in a five-minute discussion, or if he or she forgot to include the decimal point on his or her intended $1.00 demand. Given this potential complication, we are sensitive to our treatment of outliers in each of the four studies. Detailed information about outlier thresholds and results including and excluding outliers can be found in the online supplemental Appendix.

4. We note that this was a very plausible “new study” that could be conducted in the lab. In fact, approximately half of the students who participated in the study did actually come into the lab where they had political conversations with other participants lasting approximately five minutes. At the end of the omnibus survey, participants are always given the option to give their contact information to indicate their interest in participating in future studies for pay.

5. The topics were abortion, gun control, immigration, mandatory vaccinations, criminal justice reform, charter school policy, the 2016 Republican presidential candidates, and the 2016 Democratic presidential candidates.

6. We acknowledge that “easy issues” in the Carmines and Stimson framework might indeed be those most difficult to discuss with others given their controversial nature and the smaller range of policy options making it harder to moderate or censor one’s view in a discussion.

7. The group compositions used in this study were: a group of Hillary Clinton supporters, a group of Bernie Sanders supporters, a group evenly split between Clinton and Sanders supporters, a group of Donald Trump supporters, a group of Ted Cruz supporters, a group of Marco Rubio supporters, a group evenly split among Trump, Cruz, and Rubio supporters, and a group evenly split among supporters of all the candidates listed.

8. For example, the 2016 Emmy nominations could be construed as political after concerns were raised about the underrepresentation of minorities at the 2016 Academy Awards. Similarly, the 2016 Olympics could bring up discussions about infrastructure and development in Brazil, given the media attention dedicated to Rio’s struggle to prepare to host the event. State-sponsored doping programs were also discussed leading up to the start of the 2016 Olympics, which could include some political elements. Pokémon Go is a smartphone game that allows players to track and capture Pokémon at various public places in their communities. The game was very popular at the time our data were collected, but it also proved to be disruptive to places of worship and schools, which were often flagged as places to catch Pokémon. There might also be political discussions about the benefits to local businesses who experienced increased foot traffic due to the game. Finally, Uber and Lyft could yield some political discussion, as some states and localities discuss whether to allow these transportation options to operate. Moreover, the political component of these issues, as well as their overall salience, could vary cross-nationally.

9. We used a modified version of the scale Theiss-Morse uses to examine national identity (Theiss-Morse, 2009). Our version is simply generalized to capture one’s propensity to identify with groups in general as opposed to specific groups.

10. Those in the apolitical conditions were asked, “How interested are you in pop culture?” Those in the political conditions were asked, “How interested are you in politics and public affairs?”

11. In Study 3, we collected enough basic demographic information that we were able to compare differences between potential outliers and the rest of the sample. As shown in a series of tables in the online supplemental Appendix, outliers were not substantially, nor statistically, significantly different from the rest of the sample on most characteristics that we could measure. Two exceptions are that the outliers were slightly younger and more educated than the rest of the sample.

12. We excluded 232 pure Independents from the analysis. We had 121 Independents who leaned toward the Democrats and 71 Independents who leaned toward the Republicans, who we coded as leaning partisans and included in the analysis. In making our coding decision, we build on Klar and Krupnikov (2016) and Klar, Krupnikov, and Ryan (2018), who argue that many Americans have negative affect toward partisans in general, as opposed to just negative affect toward members of the out-party exclusively.

13. We present these results including outliers in the online supplemental Appendix. The coefficients on the Identity (T2) treatment are all in the same direction as shown in Table 2, but the results are statistically significant across all four models. The interaction term in Model 2 is also statistically significant at the .10 level.

Additional information

Funding

This work was supported by the National Science Foundation Division of Social and Economic Sciences [SES 1423788].