Linking experimental and survey data for a UK representative sample: Structural estimation and external validity of risk and time preferences

Presenter: Matteo M. Galizzi, London School of Economics

Author: Matteo M. Galizzi

We provide the first ‘artefactual field experiment’ that directly integrates experimental measures for risk and time preferences for a representative sample of respondents within the Innovation Panel of Understanding Society, the world-largest multi-scope panel survey. We randomly select a representative subsample of 707 respondents and randomly allocate them to either a face-to-face (n=452) or a web-mode (n=255) interview. In both groups, respondents’ discounting rates and a-temporal risk preferences are elicited using incentive compatible methods. The experimental tasks allow the ‘structural’ joint estimation of risk preferences and a broad class of inter-temporal discounting models, including exponential, quasi-hyperbolic, fixed cost, Weibull, and various forms of hyperbolic discounting. The different structural models are jointly estimated using Maximum Likelihood methods, calculating individual-specific levels of daily ‘background consumption’ from linked survey data. Our representative sample of the UK population shows high heterogeneity in the individual responses to the experimental risk and time preferences questions, but no significant interview-mode effects. In general, responses are broadly consistent with risk-aversion; soundly reject exponential discounting; and give broad support to non-constant discounting models, in particular to the quasi-hyperbolic, the generalised hyperbolic, and the Weibull discounting models. In order to explore the ‘external validity’ of the experimental risk and time preferences measures, the structurally estimated behavioural parameters are systematically cross-validated with a comprehensive range of linked survey data, providing mixed evidence.