Semiparametric Bayesian estimation in an ordinal probit model with application to life satisfaction across countries, age and gender

Abstract

We employ a novel semiparametric Bayesian ordinal probit model to re-examine the relationships between age and life satisfaction (“happiness”) across countries and gender. Within our ordinal choice framework we introduce a new scheme for cutpoint simulation and develop a locally adaptive method for smoothing an otherwise erratic collection of age dummy variable coefficients. We find strong evidence that employment status is deeply intertwined with the reported happiness of men, that employment status is relatively more important for men than it is for women, and that the common use of quadratic models to report ages when happiness is minimized can produce inference that is inaccurate and misleadingly precise. We also find evidence that refines the “U”-shaped pattern between age and happiness that has been reported and/or debated in this literature: For men in most of the European countries we consider, happiness is found to rise sharply before traditional retirement age and then into the beginning of retirement, followed by a flattening or decline in the right-tail of the age distribution. Often-used models that are quadratic in age fail to reproduce this pattern and are frequently at odds with the data.