Social Sciences Brown Bag Seminar

Joint with Erik Snowberg and Leeat Yariv.

Measurement error is ubiquitous in experimental work due to the inherent imprecision of lab-based elicitation techniques of behavioral proxies. This measurement error biases inference on causal effects of such behavioral proxies towards zero and attenuates correlations between different proxies.  Measurement error also reduces the effectiveness of such elicited behavioral proxies as controls.  In this paper, we develop a simple methodology for handling measurement errors in experiments by performing multiple elicitations of behavioral proxies.  We illustrate the potential usefulness of our methodology with data from the Caltech Cohort Study, a longitudinal study tracking the entire student body at the California Institute of Technology.  Replicating several classical experiments, this application allows us to characterize the impact of measurement error on classic results and to illustrate how results change after accounting for measurement error.