Determinants of risk attitudes using sample surveys: The implications of a high rate of nonresponse

Our purpose is to investigate the determinants of risk attitudes, using willingness to bet in a hypothetical lottery as a proxy variable. A question regarding willingness to bet was submitted to a sample of Italian household heads during the Survey of Household Income and Wealth, conducted by the Bank of Italy in 2010. However, observations are available only for a restricted subset of the sample. Therefore, an analysis of the determinants of willingness to bet must be conducted as a sensitivity analysis in which the nature of the sample selection, random or non-random, is controlled for. Model estimations were performed within a semiparametric modelling framework, enabling flexible modelling of the effects of continuous covariates on outcomes. Our findings suggest that willingness to bet is observed in a restricted, non-randomly selected sample. Thus, biased parameter estimates may result from using statistical models that ignore this issue. Neglecting such bias can affect interpretations of results as well as predictions of willingness to bet (risk attitudes) among non-participants. We also observe that the assumption of a linear relationship between continuous covariates and outcome may obscure the patterns of the relationship investigated.