-
The Welfare Consequences of Individual-Level Risk Preference Estimation
Models of Risk Preferences: Descriptive and Normative Challenges (2023), Vol. 22, 227-254.
DOI: 10.1108/S0193-230620230000022005 -
Subjective Beliefs and Economic Preferences During the COVID-19 Pandemic
(with G.W. Harrison, A. Hofmeyr, H. Kincaid, D. Ross, M. Schneider, & J.T. Swarthout)
Experimental Economics (2022), 25, 795-823.
DOI: 10.1007/s10683-021-09738-3 -
A Case Study of an Experiment During the COVID-19 Pandemic: Online Elicitation of Subjective Beliefs and Economic Preferences
(with G.W. Harrison, A. Hofmeyr, H. Kincaid, D. Ross, M. Schneider, & J.T. Swarthout)
Journal of the Economic Science Association (2021), 7(2), 194-209.
DOI: 10.1007/s40881-021-00115-7 -
Eliciting Beliefs about COVID-19 Prevalence and Mortality: Epidemiological Models Compared with The Street
(with G.W. Harrison, A. Hofmeyr, H. Kincaid, D. Ross, M. Schneider, & J.T. Swarthout)
Methods (2021), 195, 103-112.
DOI: 10.1016/j.ymeth.2021.04.003 -
The Trust Game Does Not (Only) Measure Trust: The Risk-Trust Confound Revisited
(with R. Chetty, A. Hofmeyr, & H. Kincaid)
Journal of Behavioral and Experimental Economics (2021), 90, 1-14.
DOI: 10.1016/j.socec.2020.101520 -
The Statistical Power of Individual-Level Risk Preference Estimation
Journal of the Economic Science Association (2020), 6, 168-188.
DOI: 10.1007/s40881-020-00098-x
Working Papers
-
Estimating Higher Order Risk Preferences with a Flexible Utility Function: The Bézier Curve
(with Andre Hofmeyr)
CEAR Working Paper 2025-05
Link -
Mixtures of Risk Preferences with a Bayesian Hierarchical Model
(with X. Sherry Gao & Glenn W. Harrison)
CEAR Working Paper 2025-03 — Status: Under Review
Link -
Salience, Beliefs, and Social History: The Trust Game
(with Andre Hofmeyr & Harold Kincaid)
CEAR Working Paper 2023-05 — Status: Under Review
Link -
Evaluating the Normative Coherence of Stochastic Models
CEAR Working Paper 2023-04
Link -
Recovering Subjective Probability Distributions: A Bayesian Approach
(with Glenn W. Harrison & Eric R. Ulm)
CEAR Working Paper 2022-03 — Status: Revise and Resubmit, Experimental Economics
Link
In Progress
-
Models of Risk Preferences and Higher-Order Risk Attitudes (with Glenn W. Harrison & Andre Hofmeyr)
Economists have gained a richer understanding of the structure of risk preferences by defining key concepts using intuitive and directly observable choice tasks. These choice tasks define “behavioral reference lottery pairs” that can be used to identify risk preferences reflecting risk aversion, prudence and temperance. We show how to apply these ideas to a wide range of models of risk preference beyond Expected Utility Theory. An immediate corollary is to be able to evaluate the importance of higher-order preferences in cases that do not meet the exact requirements of these behavioral reference lottery pairs. Those cases characterize virtually all of the important applications of risk preferences of all orders.
-
The Hunt for Loss Aversion: A Bayesian Detective Story (with James Bland & Glenn W. Harrison)
Cumulative Prospect Theory (CPT) is a flexible model of risk preferences that seeks to rationalize apparent deviations from Expected Utility Theory by proposing the concept of loss aversion. The intensity of loss aversion is directly parameterized in CPT which allows for direct econometric testing of its magnitude. We develop a Bayesian Hierarchical Model to estimate a structural model of Cumulative Prospect Theory at the individual subject level.
-
Bayesian Hierarchical Models of Risk Preferences (with James Bland & Glenn W. Harrison)
To better characterize the heterogeneity of risk preferences at the level of the individual agent the use of Bayesian Hierarchical Models is attractive. We demonstrate how these models can be applied to a wide range of popular models of risk preference and inferential questions. We develop methods for efficiently solving these models consistently, even with large numbers of agents. We show how demographic covariates can be used to systematically re-examine questions such as the nature of gender differences in risk preferences. We illustrate the use of these methods for “granular meta-analyses” of risk preferences, where the individual choice data from each study are formally modeled, rather than summarized by some statistic and its standard error. And we examine how hyper-posterior estimates from one study can be used as priors from other studies to “fill in” for incomplete experimental designs. Finally, we provide flexible software and templates to implement these methods across popular statistical platforms.
-
Accuracy and Confidence in Eyewitness Identification: A Reconsideration (with Quanita Adams, Glenn W. Harrison, J. Todd Swarthout & John Theilman)
Eyewitnesses are expected to make judgements that reflect their subjective beliefs over a range of alternatives. Those judgments are evaluated in terms of their accuracy and their confidence. The best responses are regarded as those that are accurate and confident. But in practical settings of importance to individuals and society, the tradeoff between accuracy and confidence has to be identified and measured rigorously, and is not in current practice. We examine the case of someone who had witnessed a crime being asked to identify a possible suspect in a line-up that varies in terms of gender and race. We redefine what accuracy and confidence should mean in this setting, design a controlled experiment to demonstrate and measure the two in a rigorous manner, contrast our approach with current practice, report results on the implications of our reconsideration, and offer normative recommendations for better managing the future risks of testimony from witnesses of crimes.
-
Structurally Estimating Joint Belief Distributions
An existing belief estimation method is extended to estimate joint belief distributions by bisecting a joint event space an arbitrary number of times, and eliciting beliefs over each bisection. The estimation of a joint belief distribution allows one to draw inferences about the dependence structure individuals ascribe to multivariate events. Beliefs are estimated jointly with risk preferences and with a full stochastic specification which rationalizes deviations from utility models and violations of the axioms of probability as choice or behavioral errors.
-
Limitations of Eliciting Beliefs with a Bandwidth Task
A bandwidth task is an incentivized belief elicitation task in which a subject is queried for the answer to a particular question and only rewarded if their answer falls within a specified band of the true answer. This task has several limitations. Most importantly, the response to these tasks does not map to any moment of the distribution, nor common summary statistics such as the median or mode. Additionally, the choice of bandwidth has direct implications on the response given by the subject for non-symmetric belief distributions, and this response under-identifies belief distributions defined by more than one parameter. Proofs are provided for these limitations. The consequences of using this approach over various domains of interest are also reviewed.
-
Massively Parallel Adaptive Markov Chain Monte Carlo
A method is proposed for adaptively selecting the proposal distribution for Markov Chain Monte Carlo simulation that is based on the current state of many simultaneous MCMC chains. This method lends itself to massively parallel implementations such that several thousands of chains can be run, and sampled from, in parallel. The gains in ergodic convergence speed, combined with the ability to sample from thousands of chains in parallel, make this approach an attractive alternative to other adaptive samplers. Gains in efficiency are demonstrated in comparison to other popular samplers by estimating a Bayesian Hierarchical Model of a non-parametric utility function with several thousand estimands. The posterior distributions are essentially equivalent between samplers, but the newly proposed sampler is several orders of magnitude faster in reaching the ergodic distribution.