The "risk elicitation puzzle" summarizes the lack of understanding of why people's stated risk preferences often do not match their actual risk-taking behaviour. Among the several potential explanations, one possibility is that stated risk preferences may not accurately reflect true preferences because of cognitive biases such as overconfidence, lack of self-awareness, or social desirability; another possibility is that people's risk-taking behaviour is influenced by situational or contextual factors that are not sufficiently internalized when assessing risk preferences in the lab. But even before all these explanations come into play, it is unclear to which extent our measures are good, i.e. they display nice psychometric properties and are free from noise and measurement error.
This paper focuses on measurement error and the theoretical frame we use to deal with risk. In particular, we investigate whether the correlations among measures and between measures and field behaviour, two different facets of external validity, can be substantially increased by reducing measurement error by aggregating behaviours and stated preferences over time (and the latter also across methods).
To do so, we expose subjects to 4 different risk elicitation methods (Holt-Laury, BRET, Investment Game and a loss aversion multiple price list) and two widespread risk-taking questionnaires (DOSPERT and SOEP) in a lab session, and then follow subjects’ behaviour for 14 days using a daily reconstruction method (DRM), whereby subjects keep a personal journal of their activities and decisions involving risks over 14 days. In the DRM, subjects are asked to report every day if they faced risky situations, and if so which was their decision; they are further asked to assess the perceived riskiness of the situation and the perceived level of risk taken with their decision. During the same 14 days, subjects answered the tasks 7 times (on even days) and to questionnaires 7 times (on odd days).
The rich data we collect allows us to asses test-retest reliability, convergent and external validity for each subject over time; and to aggregate observations to check whether smoothing out measurement error, either with simple means or with ORIV techniques, can account for a part of the “risk elicitation puzzle”, and how big a part.
Join at: imt.lu/aula2