The Quest for Parsimony in Behavioral Economics: New Methods and Evidence on Three Fronts
104 Pages Posted: 17 Jan 2017 Last revised: 26 Jan 2017
Date Written: January 2017
Behavioral economics identifies myriad deviations from classical economic assumptions about consumer decision-making, but lacks evidence on how its diverse phenomena fit together and whether they are amenable to modeling as low-dimensional constructs. We pursue such parsimony on three fronts, with success on two and instructive failure on the third. Elicitation parsimony reduces impediments to data collection by streamlining standard methods for directly measuring a person’s behavioral tendencies. We do so for 17 potentially behavioral factors per individual in a large, nationally representative sample, and several sets of results indicate that our streamlined elicitations yield low-cost, high-quality data. Behavioral sufficient statistic parsimony aggregates information across behavioral factors, within-person, to create two new lower-dimensional, consumer-level measures of behavioral tendencies. These statistics usefully capture cross-sectional variation in behavioral tendencies, strongly and negatively correlating with a rich index of financial condition even after (over-)controlling for demographics, classical risk attitudes and patience, cognitive skills including financial literacy, and survey effort. Our quest for common factor parsimony largely fails: within-consumer correlations between behavioral factors tend to be low, and the common factor contributing to all 17 behavioral factors within-individual is weakly identified and does not help explain outcomes conditional on the other covariates. Altogether our results provide many new insights into behavioral factors: their distributions, inter-relationships, distinctions from classical factors, and links to outcomes. Our findings also support the two leading approaches to modeling behavioral factors—considering them in relative isolation, and summarizing them with reduced-form sufficient statistics—and provide data and methods for honing both approaches.
Suggested Citation: Suggested Citation