How Does Confirmation Bias Affect the Persuasiveness of Policy Predictions?

Love Christensen

An important way for political actors to shape public opinion is by influencing voters’ beliefs about policy outcomes. For example, in debates on trade reforms, U.S. politicians have frequently made predictive appeals about the effects of free trade on, for instance, manufacturing employment.

But political actors face a trade-off when they try to influence the electorate’s beliefs. They want to sway voters maximally, yet voters may discount predictions that are inconsistent with what they already hold to be true. Should political actors moderate or exaggerate their predictions to maximize persuasion?

In the paper, I address both theoretically and empirically whether voter confirmation bias – the propensity for individuals to assimilate information which confirms their beliefs – affects how they learn from information about the effects of political reforms. Theoretically, I provide a formal model of confirmation bias. The intuition of the model is that voters use the distance between a prediction made by a politician about the effect of a reform and their prior belief about this effect as a shortcut for inferring the credibility of a prediction. The greater this distance is, the less credible the prediction. I show that it is only under strong confirmation bias that extreme predictions, i.e., predictions far from the voters’ priors, undermine their own credibility to such an extent that more moderate predictions would be more effective in shaping the beliefs of voters. Thus, it is only if voters exhibit strong confirmation bias that political actors are incentivized to moderate their predictions.

Empirically, I test the theoretical model with a survey experiment. The gist of the experiment consists of presenting respondents with predictions about the effect of joining the TPP on U.S. manufacturing employment. I then examine how respondents (1) update their beliefs about the effect of the reform and (2) perceive the credibility of the prediction, both conditional on the distance between their prior belief and the prediction. A summary of the results is shown in the figure below.

The x-axes show the distance between the prediction that respondents are exposed to and their prior belief. The scale is 0.1 million jobs, so the maximum difference between the respondent’s prior and the prediction is a predicted loss of 3 million jobs. The predictions are always pessimistic relative to the respondents’ beliefs. The left panel shows the effect of the difference between the respondents’ priors and the predictions on how much the respondents update their beliefs. It is clear that respondents initially assimilate the information, but as the prediction grows more extreme, they increasingly rely on their prior beliefs, consistent with strong confirmation bias. The center panel shows that this may be explained precisely by respondents discounting the credibility of the prediction as it grows more extreme. And in the rightmost panel, we see that it is only predictions which are neither too close nor to distant from respondents’ priors that are able to shift support for the TPP.

For a politician who wants to sway public opinion, the message is clear: predictions have to be spaced just right to effectively change public opinion. Predictions close to voters’ priors are perceived as credible but do not necessarily shift beliefs, or if they do, do not shift them enough to shift support for the reform. Predictions far from voter’s priors may actually shift beliefs, but are not perceived as credible enough to shift support.

Besides the importance for understanding how voters think and reason, I believe that this work is especially important for two reasons. First, it shows that the beliefs of voters may in fact constrain the behavior of political actors, which has important implications for the strategic behavior of politicians. If politicians must craft their messages with the electorate’s beliefs in mind, shaping voter beliefs may induce competing politicians to change their campaign messaging for strategic reasons. Second, it acts as a corrective to the very bleak picture of confirmation bias that is often presented in mass media and the political psychology literature. Confirmation bias does not necessarily imply that voters are irrational or that they cannot learn inconvenient truths. As the results show, voters do in fact learn when the predictions are spaced just right. Rather than being a sign of ignorance, confirmation bias may equally well reflect a form of sophisticated reasoning, where voters use the content of the message to infer the credibility of the message and the sender.

Go to full article