Essays on Information and Non-Bayesian Beliefs

Abstract

In this dissertation, I present a comprehensive discussion of a class of biases within the realm of probabilistic reasoning, namely confirmation bias (encompassing or closely related to commonly seen terms in the literature such as motivated reasoning and wishful thinking). The dissertation consists of three main chapters. In Chapter 1, I propose a new and improved belief updating model that can accommodate both motivated and unmotivated confirmation bias. The model improves upon existing models in its ability to explain data better, and its applicability to settings beyond binary-state spaces. I characterize the model with three intuitive axioms. In two extended applications, I show that the model establishes a link between confirmation bias and several well-known phenomena, such as the significance of first impressions, the polarization of beliefs, and the perseverance of inaccurate beliefs. In Chapter 2, I turn to the experimental elicitation of motivated and unmotivated confirmation bias. Previous experiments have provided evidence for motivated and unmotivated confirmation bias individually, but never discussed the possibility that the two can occur together in depth. This chapter presents one of the first experiments that examines both forms of confirmation bias together. Subjects were asked to update their beliefs regarding both politically contextualized questions and neutral questions. Subjects exhibited both motivated and unmotivated confirmation bias, but there was also significant heterogeneity among them. Notably, motivated confirmation bias is significantly stronger in later rounds of the experimental tasks, which may be correlated with the shorter response times in the later rounds. In Chapter 3, which is joint work with Emel Filiz-Ozbay, we discuss wishful thinking (motivated confirmation bias) within a major application. In a rational inattention setting where consumers acquire information on the good’s quality before making purchasing decisions, we examine the implications of the presence of consumers with wishful thinking. These biased consumers are unaware of their bias, and weigh any good news about the product quality more heavily than a Bayesian consumer. The firm, which aims to increase the volume of sales, can strategically constrain the accuracy of the information that consumers can acquire. We show that in the presence of biased consumers, a firm would find it profitable to constrain information acquisition unless the prior belief on the quality of the product is too low. We characterize the conditions under which the entry of a competitor firm can effectively alleviate this type of exploitation. Our findings shed light on the incentives of review platforms for bombarding wishful consumers with low quality product reviews and limit consumers’ ability to identify to reviews with informative contents.