P(l)ay-As-You-Go: Large-Scale Longitudinal Risk-Elicitation in the Field

Abstract

We use the on-site pump displays of a retail fuel chain to create a unique panel of over 7,000 visitors to whom we repeatedly and frequently administer an incentivized (Bomb Risk Elicitation Task) and non-incentivized (SOEP risk question) risk elicitation task over the course of four years. This new approach to data collection overcomes the challenges of user time constraints and limited control that often hinder the use of incentive-compatible experimental methods in the field. We find that 82% of all decisions in the BRET-task are consistent with risk aversion, a larger share than typically reported in lab studies with student samples. The correlation with responses to the SOEP risk question is positive and significant, but small in magnitude. Within-subject correlations of choices in the BRET and the SOEP increase in early rounds and stabilize around 0.48 and 0.65 respectively in later rounds. This suggests that task repetition is needed for the stable measurement of risk attitudes. Both tasks reveal that female subjects are more risk averse. We find limited evidence that the elicited risk attitudes correlate with actual risk-related behaviors. Major events during the sample period, such as COVID-19 and spikes in the fuel price due to the invasion of Ukraine, do not measurably impact the risk preferences of the sampled population.