The Myth of the Rational Voter: Experimental Edition
By Bryan Caplan
I’ve long argued (here, here, and here for starters) that weak incentives are a major cause of political irrationality, but I’ve failed to convince Gerry Mackie, Jeff Friedman, Tyler Cowen, and many other critics. So I was delighted to discover some new evidence. Bullock, Gerber, Hill, and Huber’s new “Partisan Bias in Factual Beliefs About Politics” runs a series of experiments to test the effect of incentives on partisan bias. (gated and ungated versions) Their estimated effects are so large they surprise even me:
In both experiments, all subjects were asked factual questions, but some were given financial incentives to answer correctly. In both experiments, we find that the incentives reduce partisan divergence substantially–on average, by about 55% and 60% across all of the questions for which partisan gaps appear when subjects are not incentivized. But offering an incentive for accurate responses will not deter cheerleading among those who are unsure of the correct factual response, because such people stand to gain little by forgoing it. In our second experiment, we therefore implement a treatment in which subjects were offered incentives both for correct responses and for admitting that they did not know the correct response. We find that partisan gaps are even smaller in this condition–about 80% smaller than for unincentivized responses. This finding suggests that partisan divergence is driven by both expressive behavior and by respondents’ knowledge that they do not actually know the correct answers.
To be fair, the authors never use the words “irrational” or “irrationality” in their paper. They think they’re testing the sincerity of partisan bias:
This paper, however, considers a distinct alternative: partisan differences in survey responses may not solely indicate differences in true beliefs. Instead, they may also reflect the expressive value of offering survey responses that portray one’s party in a favorable light. A partisan pattern of survey response follows when this expressive value outweighs the utility that survey respondents would receive from stating their sincere beliefs. Partisan divergence may therefore reflect the joy of partisan “cheerleading” rather than sincere differences in beliefs about the truth. Despite the reality that survey respondents have limited incentives to respond accurately to survey questions, almost no research has attempted to determine the extent to which partisan divergence in responses to factual questions with partisan implications reflects sincere beliefs.
The authors’ results are admittedly consistent with insincerity. Yet they’re equally consistent with my rational irrationality story: partisans sincerely believe their answers, but exert more intellectual self-discipline when the price of error rises. And introspectively, my interpretation is more plausible. Are we really supposed to believe that in their heart of hearts, seemingly self-righteous partisans are being willfully deceptive? Isn’t it far more likely that they’re simply being intellectually lazy?
The deeper problem with Bullock et al., though, is that they bizarrely regard their results as good news for democracy:
Our results have important implications for both political science and understandings of contemporary public opinion. Regarding the former, persistent partisan gaps, if sincere, suggest important limitations to democratic accountability. If Democrats and Republicans perceive different realities, then the incentives for incumbent politicians to pursue policies that generate objectively good policies may be reduced. Our results imply that such concerns are overstated. Democrats and Republicans may diverge in their survey reports of facts, but such responses should not be taken at face value as sincere expressions of partisan worldviews.
Question for Bullock et al.: Are real-world democratic elections more like your treatment condition with incentives, or without? Given voters’ low probability of decisiveness, the answer is clear: Real-world democratic elections closely resemble the treatment without incentives. After all, if a voter votes under the influence of partisan bias, what happens to him? The same thing that would have happened if he voted responsibly. As I once objected to Arthur Lupia:
Yes, if voters were paid for correct answers, they would know more. But in the real world, they aren’t paid. Tests without incentives mimic real world conditions; tests with incentives don’t.
From the standpoint of democratic performance, it doesn’t really matter if partisan bias stems from insincerity or irrationality. If weak incentives are a major cause of partisan bias, we should expect partisan bias to infect democratic outcomes from now till doomsday.
P.S. Don’t blame me, I am only a messenger.