Both Trump & Clinton supporters display less positive attitudes towards the opposing supporters; significantly more wealth is destroyed if the opponent is an opposing voter, effect mainly driven by Clinton voters

The cost of a divided America: an experimental study into destructive behavior. Wladislaw Mill & John Morgan. Experimental Economics, Nov 19 2021. https://link.springer.com/article/10.1007/s10683-021-09737-4

Abstract: Does political polarization lead to dysfunctional behavior? To study this question, we investigate the attitudes of supporters of Donald Trump and of Hillary Clinton towards each other and how these attitudes affect spiteful behavior. We find that both Trump and Clinton supporters display less positive attitudes towards the opposing supporters compared to coinciding supporters. More importantly, we show that significantly more wealth is destroyed if the opponent is an opposing voter. This effect is mainly driven by Clinton voters. This provides the first experimental evidence that political polarization leads to destructive behavior.

---

Own spite measure

Our main measure was aimed to mimic basic market interactions where spite has been observed. More specifically, our measure reflects a simplified and condensed version of a second-price auction where one player can reduce the payoff of the opponent by increasing the own bid (see Kimbrough and Reiss 2012, for such a situation). This measure consists of three distribution-decisions upon money. These distributions are shown in Table 1. We call this our own spite measure. We asked the participants to decide three times among nine possible allocations, similar to the SVO-Slider measure by Murphy et al. (2011). The participants were told that either their decision or their opponent’s decision would be implemented, depending on a computerized random draw.

In all sets, the allocation with the highest payoff for the other player also maximizes the own payoff. However, any deviation from this allocation reduces the payoff of the other player and never increases the own payoff. In contrast to a standard dictator game—where there is a trade-off between the own payoff and the payoff of the opponent—in this game, the participants who do not choose the Pareto-efficient outcome do this in order to harm the other player. Therefore, any deviation from the Pareto-efficient outcome resembles spiteful behavior in a market setting and can be interpreted as spite or joy-of-destruction.Footnote25

Discussion

This paper investigates whether partisanship—understood as the self-identified party affiliation—leads to dysfunctional behavior. In particular, we study which attitudes and, more importantly, which behavior voters show towards voters casting the same or the opposite vote. To the best of our knowledge, we are the first to study spite in a political context. For this purpose, we collected decisions of self-reported Clinton and Trump voters five times over a period of 4 years.

Most importantly, we were able to show that dysfunctional behavior—understood as the destruction of wealth—is significantly more likely if an opposing voter (outgroup) is impacted compared to a coinciding voter (ingroup). This effect is mainly driven by Clinton voters, who are significantly more likely to behave spitefully towards Trump voters compared towards fellow Clinton voters. This effect is not found for Trump voters—while Trump voters generally exhibit more spiteful behavior they do not differentiate between Trump voters and Clinton voters in their behavior.

These effects are supported by the attitudes of the voters: attitudes towards opposing voters are substantially and significantly more negative than attitudes towards coinciding voters. This effect is significantly stronger for Clinton voters. Further, the timing of the experiment does not substantially change attitudes, and it has no significant effect on dysfunctional behavior.

Several aspects of the results are worth elaborating on.

First, it is worth pointing out that we are not the first to find substantial polarization in the US (see Pew Research Center 2017a), but we are the first to show that this polarization leads to significantly increased destructive behavior (at least for Clinton voters). This is the main point of the paper, and this presents a significant and important contribution. We are able to show that even in a low key situation, like an online experiment, people are more likely to behave spitefully if matched with opposing voters. Hence, it seems plausible that in more salient situations where partisanship is even easier to detect and of more importance (e.g., collaborative work), the effect would be even stronger. More importantly, we know now that an increasing polarization leads to increased social and economic costs.

Second, it is interesting that the timing of the experiment hardly influenced attitudes and behavior. During this period Donald Trump won unexpectedly in 2016 and lost in 2020. However, these events seem not to spill over substantially into attitudes and behavior. This indicates that the destructive consequences of polarization are persistent and might be hard to eradicate.

Third, the differences between Clinton voters and Trump voters are worth elaborating on. On the one hand, it seems not too surprising that Trump voters did not differentiate between ingroup-members and outgroup-members because this would be perfectly in line with most papers on outgroup-bias who show that outgroup-bias lead rarely to purely hostile behavior (Brewer 1999, 2017; Balliet et al. 2014). It is also not too surprising that attitudes towards opposing voters are negative because this has also been shown in other papers (Tajfel 1970; Fowler and Kam 2007; Weisel and Böhm 2015).

However, it is puzzling that Clinton voters have significantly less positive attitudes towards their outgroup-members compared to Trump voters. More importantly, Clinton voters behave relatively more spitefully towards outgroup-members compared to ingroup-members—which cannot be found for Trump voters. This result is particularly interesting in the sense that it does not only show a mere group identity effect (as our aggregate results are driven by Clinton voters only) but an effect of political identity. The asymmetry in destructive behavior between Clinton and Trump voters suggests that political identity functions differently than just plain group identity.

One possible explanation for the asymmetry in behavior is that Trump voters are considered morally wrong in supporting Donald Trump. In that case, Mummendey and Wenzel (1999) argue theoretically that “inferior” groups are more likely to experience discrimination and hostility. Similarly, Brewer (1999) argues that negative discrimination might be present if participants are fighting for political power. Further support is provided by Parker and Janoff-Bulman (2013), who show that morality based groups lead to less positive emotions. More importantly, Weisel and Böhm (2015) demonstrate a significant increase in help avoidance if the group difference is morality-based: “When given the chance to benefit a strong-enmity outgroup, and even more so a morality-based outgroup, many group members decline to do so” (Weisel and Böhm 2015, p. 118). In Online Appendix we discuss differences in ascribed morality between Clinton and Trump voters in our experiment. We find that Clinton voters consider Trump voters substantially less moral than vice versa. In parallel, polls also reveal that a majority of Democrats express to feel angry going into the midterm elections of 2018 while only 30 percent of Republicans say the same.Footnote37 This indicates that the group difference might be morality-based and consequently drive hostile behavior.

Another possible explanation for the heterogeneous effects between Clinton and Trump voters is the expectation of Clinton voters that Trump voters will generally behave more spitefully and therefore retaliate in expectation. Trump voters on the other hand just generally are more prone to spiteful behavior independent of the opponent. Thus, Clinton voters just increase their spiteful behavior to match the spiteful behavior of Trump voters, which results in a heterogeneous effect.

However, these explanations are only conjectures and it might be valuable for future research to take a closer look at the justifications and motivations of Clinton and Trump supporters to engage in hostile behavior.

While we believe the results to be robust, some possible limitations should be noted. First, our experiment might be prone to experimenter demand effects as the opponent’s political orientation is made salient. While this saliency is essential for the treatment to work, it might reveal the experiment’s purpose and, thus, lead participants to shift their behavior. To obtain a bound on a possible demand effect we conducted a demand-effect treatment as suggested in de Quidt et al. (2018) and reported in detail in Online Appendix. Inducing demand does not change the behavior in our setting.Footnote38 Thus, while a demand effect cannot be excluded, we find that inducing a demand effect does not alter the behavior of participants substantially. Another limitation of our experiment is the non-representativeness of our sample. While our sample is much more representative of the US population than typical student samples, it is still not representative of the US population, as discussed in detail in supplementary material. Thus, such a selection might bias our results and reduce the generalizability of our findings. In Online Appendix, we try to deal with this issue by adjusting the weights of our estimations to make our sample artificially representative. While our results remain robust, we cannot exclude the possibility that our findings would differ using a representative sample. A similar concern is that sampling Trump and Clinton voters via MTurk might be problematic as, for example, Trump voters on MTurk are different from Trump voters in the general population. Reassuringly Huff and Tingley (2015) show that Mturkers behave similarly to the general population with regard to their voting behavior and suggest that Mturk is a great source to study voters. Further, we find striking similarities in demographics and voting patterns between our sample and nationally representative samples as discussed in supplementary material. Nevertheless, we should caution the reader that we cannot exclude the possibility that, for example, Clinton voters on Mturk are particularly spiteful compared to Clinton voters in general. It is also worth pointing out that we made our treatment rather salient by providing the political identity of their opponents to participants. While this was essential for the experiment it might reduce generalizablity. Partisan affiliation would most likely be less salient in the vast majority of human interaction. Thus, our experiment might not speak to everyday situations but primarily to environments where political identity is salient (such as around elections, rallies, etc.). Another concern we have to think about is possible spillover effects between the auction experiment (which is reported in Mill and Morgan (2020)) and the main task of this experiment (the spite task). It is possible that conducting an auction prior to the spite task might have increased spiteful behavior as participants might have been put into a competitive frame. Behavioral spillovers are discussed in detail in Dolan and Galizzi (2015). In particular, Cason and Gangadhara (2012), and Savikhin and Sheremeta (2012) find spillover effects between competitive games and cooperative games. However, we have two “competitive” games, which most likely will reduce the spillover effect. Further, we can see that the behavior in the SVO-task reported in Online Appendix is very similar to behavior reported in other studies (see also Footnote 19) which indicates that behavior has not been influenced substantially. More importantly, while such a spillover effect might shift overall behavior towards more spite, this spillover effect is identical between the treatments and also influences Clinton and Trump voters arguably to the same extend.Footnote39 Thus, while the absolute level of spite might have been affected by the auction, the auction is unlikely to account for our heterogeneous results.