Most information structures pose a signal extraction problem for belief udpating: a person wants to learn X, but information about X is confounded by Y.
Inattention to alternative causes Y could create misattribution to X.
A series of laboratory and online experiments show that most people do not take into account alternative causes (feature neglect), leading to excessively sensitive and overprecise beliefs.
There is pronounced heterogeneity, with up to 95% of beliefs corresponding to three updating rules: feature neglect, Bayesian updating and non-updates.
Exploring the underlying mechanism, I find that unawareness about the necessity to factor in alternative causes drives feature neglect,
whereas non-updating follows cost-benefit considerations.
Based on combined evidence from more than thirty treatments, I propose a conceptual framework in which selective attention causes an automatic, subjective representation of the updating problem.
While this mental model determines how information is processed, computational complexity and the size of incentives affect whether it is incorporated or ignored.
Learning is limited by unawareness, but cues and hints debias by changing people’s mental models directly.
Sources and Heterogeneity of Reference Dependence - An Empirical Examination with Lorenz Goette, Alex Kellogg and Charles Sprenger
This project examines the role of heterogeneity in loss aversion for identifying models of expectations-based reference dependence (Kőszegi and Rabin, 2006, 2007) (KR). Different levels of loss aversion lead to different signs for comparative statics previously used to test the KR model. In an experiment with 607 subjects, we show heterogeneous treatment effects over loss aversion types. Recognizing heterogeneity in loss aversion allows us to reliably recover the KR model’s central element of expectations-based reference points. Additional effects are discussed related to the subjective perception of exchange experiences.
Breaking Trust: On the Persistent Effect of Economic Crisis Experience with Tom Zimmermann
Recent empirical evidence documents substantial geographical and temporal heterogeneity in trust levels that has far-reaching implications for societies’ economic outcomes. However, little is known about the sources of this variation. This paper examines the breach of trust hypothesis on the impact of adverse experience in economic interactions. Using a variety of identification strategies in a large cross-country sample, we estimate a persistent and robust negative long-term effect of economic crisis experience on trust in other people. In line with the hypothesis, the effect is specific to living through crises in trust-intensive domains, most of all banking crises. The effect is not driven by distrust in financial institutions but is accommodated by a lack of confidence in the political class, and operates via beliefs rather than changes in preferences. Our baseline cross-country evidence is corroborated by a separate within-country replication of the findings that exploits regional variation in the experience of U.S. bank failures.
Intertemporal Altruism with Armin Falk and Philipp Eisenhauer
For most economically relevant decisions, costs and benefits occur at different points in time and affect both the decision maker and others. This paper provides a first systematic analysis of altruistic behavior in intertemporal contexts. We study self-other tradeoffs in a high-stakes laboratory experiment using a novel donation paradigm. We theoretically motivate and then estimate i) univariate discount functions both for individual payments and for charity donations, ii) subjective exchange rates between charity-euros and self-euros for different delays, and ii) multivariate discount functions based on choices involving both categories.
We find that the subjective exchange rate falls over time, i.e., the relative value of one charity-euro per contemporaneous self-euro increases with common delay. Surprisingly, this tendency of "being more altruistic in the future than today" is not driven by differences in univariate discount functions. Our measures of multivariate discounting instead suggest that discounting of future payoffs depends on the reference category in the choice set. We formalize the implications of our findings for discounted utility theory in multi-attribute settings. Our data from static, risky choices by the same subjects further allow us to characterize the atemporal, multi- attribute utility function both non-parametrically and under structural assumptions.
On the Delayed Negative Effect of Prosocial Behavior on Happiness with Armin Falk
Loss Aversion and Effort: Evidence from a Field Experiment with Ernst Fehr and Lorenz Goette
This paper contributes new evidence to a recent controversy in labor economics: Is labor supply affected by sensations of gains and losses relative to a narrowly-defined earnings target? We report evidence from a field experiment that exogenously and temporarily manipulates the progress of workers towards a possible earnings target, by randomly delaying subjects in their progress. We find that individuals who were delayed put in more effort. We also find that this effect is specific to the treatments in which we paid the subjects on a piece rate. In a control treatment with fixed wages, we find no effect that the subjects effort choices were affected by the productivity manipulations. This is inconsistent with a wide class of neoclassical utility functions. We argue that this is evidence that the individuals in the piece rate treatment were trying to catch up with the loss in output due to low productivity. There are two models to explain this behavior: Reference-dependent preferences, or concave utility over narrowly defined outcomes. To further probe into the pattern, we also measure risk preferences of the individuals in an experiment involving risky payoffs. We find that the pattern of catching up after a delay is only present for individuals who behave in a risk-averse fashion in simple choice experiments. Furthermore, we find that the effect is concentrated among strongly risk-averse individuals, and not present, for moderately risk-averse individuals. This is evidence against concave utility, thus favoring reference-dependent preferences to explain our result.