In this paper, I develop a model of confirmation bias, which arises from a cognitive mechanism conceptually linked to the Bayesian brain paradigm and the efficient coding hypothesis in neuroscience. Specifically, I hypothesize that cost of information processing, quantified as the statistical divergence between prior and posterior beliefs, influences the precision of the sampling process, thereby allowing the probability of misperception to emerge endogenously. Consequently, disconfirming evidence, which are costlier to process, attenuates the impact of new information during Bayesian updating, resulting in posterior mean beliefs that are disproportionately influenced by prior mean beliefs. I further test the assumptions of the model in an experiment and find empirical support for the hypothesis that noise in perceived probabilities increases proportionally with the distance between the DM's initial expectations and the sample.

In an experiment, we study willingness to trust when reciprocation probabilities are fundamentally uncertain. We propose a simple method to elicit trustors’ beliefs about reciprocation as second-order subjective distributions over possible reciprocation rates. We then elicit monetary trust equivalents for bets on reciprocation and use an Arrow-Pratt approximation within the smooth ambiguity model to assess how beliefs and ambiguity attitudes shape trust. Using hierarchical Bayesian econometrics, we estimate the distribution’s mean, capturing expected reciprocation, and variance, capturing perceived ambiguity. Accounting for heterogeneity in individual-level parameters, we show that both belief dimensions and ambiguity attitudes affect willingness to trust.

I experimentally investigate the interplay between confirmation bias and base-rate neglect at the individual level. I hypothesize that both how subjects utilize information and their reliance on base-rates vary depending on whether the information confirms or contradicts the base-rate during belief updating. I test this hypothesis in a setup which enables the elicitation of subjects' posterior beliefs for both confirming vs disconfirming signals at each trial. The evidence supports the hypothesis: while a majority of subjects exhibits confirmation bias, they also show less severe base-rate neglect when the signal confirms the base-rate. I further evaluate the predictive accuracy and the explanatory power of the model that accounts for the dependence of both information and base-rate utilization on the signal type, comparing it with several alternative specifications. I find that, at the individual level, the former outperforms all other specifications, including the most commonly used model in the literature on asymmetric belief updating, which only considers signal-type dependence in how information is utilized. These findings provide empirical evidence of the coexistence and interaction between these two widespread biases and underscore the importance of correctly accounting for the base-rate utilization in belief updating.