Testing Intrinsic Preferences for Information
Abstract: We present experimental results from a broad investigation of intrinsic preferences for information. We examine whether people prefer negatively skewed or positively skewed information structures when they are equally informative, whether people prefer Blackwell more informative information structures, and how individual preferences over the skewness and the degree of information relate to one another. The wide scope of our investigation not only reveals new insights regarding intrinsic preferences for information, but as we show, also allows for testing of existing models in this domain. We nd that models based on the framework of Kreps and Porteus (1978) and Caplin and Leahy (2001), are the most consistent with the data we observe.
Overview: Economists have traditionally assumed that individuals only care about material payoffs. Recently, researchers have come to recognize that utility can also depend on beliefs about themselves and future outcomes. While people cannot directly choose their beliefs, they can choose the sources of information they are exposed to that shape those beliefs.Therefore, belief-based utility can have important implications for information gathering and search behavior, and individuals may take actions detrimental to their material payoffs in order to improve their emotional state. For example, a large proportion of individuals who are at risk for Huntington’s disease refuse to take a diagnostic test, even though the results of the test could improve their ability to plan for the future. Thus, identifying how utility depends on beliefs can have significant implications for welfare and potential practical nudges to aid with decreasing such endogenous information frictions. For example, understanding how beliefs affect utility can inform government policy in a variety of information disclosure settings, including health care, media markets, and investment decisions.
In this joint research program (with Yeşim Orhun at the University of Michigan, Ross Business School and Collin Raymond at the University of Oxford, Department of Economics) we study individuals’ preferences over “information structures,” i.e. the way in which information is provided to individuals, using both experimental and theoretical perspectives. Studying choices over information structures, we will be able to distinguish between different theoretical models that give rise to very different practical implications.
Recently, a variety of theoretical models have been developed to try to understand belief-based utility. These include models of dynamic reference dependence, such as Kőszegi and Rabin (2009), where individuals care about changes in their beliefs; models of anticipatory utility such as Brunnermeier and Parker (2005) and Caplin and Leahy (2001), where individuals care about having optimistic beliefs today; and models of preferences for the timing of the resolution of information such as Kreps and Porteus (1978) and Dillenberger (2010).
Although these theoretical models have been applied in a variety of important settings,including medical testing, media bias and asset pricing, very little has been done to directlytest their primary predictions — that individuals will exhibit preferences over information,even when it cannot affect the decision-maker’s actions.
The proposed research is a series of experiments that will allow us to distinguish between competing theoretical models of intrinsic preferences for information. We will directly elicit preferences for information in an environment where the information, by construction, cannot influence actions. Thus, any preferences must be due to belief-based utility. Moreover, our design allows us to cleanly distinguish between different classes of models by testingfor a new type of informational preferences — a preference for slanted information. Weask individuals whether they would prefer to eliminate more uncertainty about the good outcome or bad outcome, or in other words, whether they would prefer negatively slanted or positively slanted signals. Preferences for slant are often observed in the real world (e.g. preferences for biased media reports), but were not one of the original motivating factors in the development of many of models of belief based utility. However, difference classes of models generate very different preferences for preference over slant, and so can serve as ancillary predictions that allow us distinguish between competing models of belief based utility in our experiments.
Update: We have run the first part of experiments for this project. We are currently working on finalizing manuscript so that it is ready for submission. The paper is presented by my co-authors in different places.