Bayesian or biased? Analytic thinking and political belief updating

Abstract

A surprising finding from U.S. opinion surveys is that political disagreements tend to be greatest among the most cognitively sophisticated opposing partisans. Recent experiments suggest a hypothesis that could explain this pattern: cognitive sophistication magnifies politically biased processing of new information. However, the designs of these experiments tend to contain several limitations that complicate their support for this hypothesis. In particular, they tend to (i) focus on people’s worldviews and political identities, at the expense of their other, more specific prior beliefs, (ii) lack direct comparison with a politically unbiased benchmark, and (iii) focus on people’s judgments of new information, rather than on their posterior beliefs following exposure to the information. We report two studies designed to address these limitations. In our design, U.S. subjects received noisy but informative signals about the truth or falsity of partisan political questions, and we measured their prior and posterior beliefs, and cognitive sophistication, operationalized as analytic thinking inferred via performance on the Cognitive Reflection Test. We compared subjects’ posterior beliefs to an unbiased Bayesian benchmark. We found little evidence that analytic thinking magnified politically biased deviations from the benchmark. In contrast, we found consistent evidence that greater analytic thinking was associated with posterior beliefs closer to the benchmark. Together, these results are inconsistent with the hypothesis that cognitive sophistication magnifies politically biased processing. We discuss differences between our design and prior work that can inform future tests of this hypothesis.

Reasoning about climate change

“…This allows us to shed new light on competing accounts of the role of deliberation in argument evaluation surrounding climate change: Does deliberation magnify partisan bias, consistent with the identity protective cognition 11 ? Or does it facilitate accurate assessments or normatively rational evidence evaluation, consistent with a more classical perspective on reasoning 28,29,35 ? Furthermore, we specify the classical reasoning account more precisely than in prior work.…”

“…Instead, increased cognitive reflection was associated with placing greater emphasis on prior factual beliefs when evaluating new information. While evaluating new evidence in light of prior beliefs is sometimes called "confirmation bias", in fact such evaluation can be entirely rational and unbiased from a Bayesian perspective when there is uncertainty about the reliability of sources [32][33][34][35] : When considering evidence that is inconsistent with your prior factual beliefs, it can be rational to conclude that it is more likely that (i) the information source is unreliable than that (ii) the accumulation of all your prior knowledge is wrong. For example, if a stranger tells you that he was abducted by aliens, is it not irrational to conclude that that information is probably unreliable.…”

The Psychology of Fake News

“…Put differently, there is a robust (albeit somewhat small) main effect of the news' ideological concordance on belief, but typically not an interaction with the news' veracity. Furthermore, this main effect of concordance does not necessarily indicate politically motivated reasoning: This difference could also arise from unbiased rational inferences (e.g., via Bayesian updating) that emerge from prior exposure to different information streamsand thus different prior factual beliefs [34][35][36][37][38][39][40].…”

Thinking clearly about causal inferences of politically motivated reasoning: why paradigmatic study designs often undermine causal inference

“…Though we lack a precise estimate of the number of counterpart experiments that use belief updating as the outcome, our impression is that it is less common. Second, and perhaps explaining the previous point, the evaluations outcome variable is argued to provide clearer evidence of politically motivated reasoning than the belief updating outcome variable (4,8,11) because of the difficulty in establishing a clear benchmark for what politically unmotivated belief updating should look like (4,12).…”