Job Market Paper
aggregate random consideration sets: theory and evidence
with Victor Aguiar, Nail Kashaev, and Jeongbin Kim
We study the problem of identifying preferences and discriminating across competing choice models for a population of individuals that may or may not pay attention to all available alternatives. We design and implement a novel online experiment that exogenously varies choice sets and consideration cost in a large population of individuals. We extend exiting models of consideration-mediated choice to allow for preference heterogeneity, and provide a testing procedure similar to Kitamura and Stoye (2016). We show that, if there is a positive mass of fully attentive individuals, we can uniquely recover the stochastic consideration set rule and identification of preferences is as good as in random utility model (RUM). Our findings suggest that RUM does not provide a good description of the population’s behavior without conditioning on the consideration cost. In contrast, a model of random consideration with logit attention (Brady and Rehbeck (2016)) cannot be rejected at the 95 percent confidence level when allowing for preference heterogeneity. When facing a high cost of consideration, behavior can be described by the independent consideration model of Manzini and Mariotti (2014), which is nested in the logit attention model and RUM.
satisficing and stochastic choice
with Victor Aguiar and Mark Dean - JET(2016)
Satisficing is a hugely influential model of boundedly rational choice, yet it cannot be easily tested using standard choice data. We develop necessary and sufficient conditions for stochastic choice data to be consistent with satisficing, assuming that preferences are fixed, but search order may change randomly. The model predicts that stochastic choice can only occur amongst elements that are always chosen, while all other choices must be consistent with standard utility maximization. Adding the assumption that the probability distribution over search orders is the same for all choice sets makes the satisficing model a subset of the class of random utility models.
POWER OF REVEALED PREFERENCE TESTS AND PREDICTIVE (UN)CERTAINTY
Revealed preference theory defines the behavioral conditions that are equivalent to many models of decision providing a nonparametric test for it. For utility maximization these conditions are given by the Generalized Axiom of Revealed Preference (GARP). For some data sets these conditions may be so unrestrictive that is (nearly) impossible to detect violations regardless of the data generating process, hindering the interpretation of the explanatory performance of the model. This paper establishes a natural trade off between fit and power in terms of their contribution to narrow down predictions if behavior is utility maximizer with error. The proposed measure exploits the connection between power and the precision to which preferences can be identified from behavior. The less precise the identification of preferences, the bigger deviations need to be in order to be detected. This measure allows to differentiate among behavior and designs that have similar power and fit under standard measures in the literature, as shown in the empirical applications for the data sets studied by Beatty and Crawford (2011) and Choi et al (2007).
WHAT CAN BE LEARNED FROM BEHAVIOR? PREDICTIVE ABILITY IN DISCRETE CHOICE ENVIRONMENTS
Revealed preference restrictions provide (nonparametric) testable implications for many theories of consumption. The analyst does not know the choice model but seeks to learn about it from behavior. However, learning is hindered due to identifiability issues because of incomplete data sets and/or inconsistent choices, both common in empirical studies. This paper provides a joint treatment of these in terms of the performance of the model to predict behavior given data. Incomplete data may translate into multiplicity of preferences that rationalize behavior which drives uncertainty when predicting choices. Likewise, inconsistent choices also induce noisier predictions. These two concerns relate to the power and fit problems respectively. Therefore, the predictive approach makes explicit the trade off between fit and power for revealed preference theories in terms of the predictive precision of the model given data.
the impact of confidence on risk taking behavior and performance
with Nicholas Coleman