Same concerns, same responses? A Bayesian quantile regression analysis of the determinants for supporting nuclear power generation in Japan

Using the Internet survey data from 6500 individuals, this study examines the determinants for supporting the restart of nuclear power plants operation in Japan. The variable of interest is the level of support that is measured as a categorical and ordered variable, for which ordered logit or probit is commonly estimated. This study departs from the literature using Bayesian ordinal quantile regression (Rahman 2015, Bayesian Anal. doi:10.1214/15-BA939) to address whether covariates have differential effects at various conditional quantiles of the latent response variable. This approach allows us to explore, for example, whether three otherwise identical individuals, the first with an average unobserved preference for the restart, the second with a low unobserved preference, and the third with a high unobserved preference, respond similarly or differently to a change in a covariate. The results show that for most of the covariates examined, including concerns about meltdowns and concerns about global warming, the effects differ across conditional quantiles of the latent response variable. In other words, the covariate effects depend crucially on individuals’ unobserved preferences for the restart (conditional on observables). The results also show that there are considerable gender differences in response to changes in covariates.