SOEPpapers on Multidisciplinary Panel Data Research 846
We propose a new non-linear regression model for rating dependent variables. The rating scale model accounts for the upper and lower bounds of ratings. Parametric and semi-parametric estimation is discussed. An application investigates the relationship between stated health satisfaction and physical and mental health scores derived from self-reports of various health impairments, using data from the German Socio-Economic Panel. We compare our new approach to modeling ratings with ordinary least squares (OLS). In one specification, OLS average effects exceed that from our rating scale model by up to 50 percent. Also, OLS in-sample mean predictions violate the upper bound of the dependent variable in a number of cases.
quasi maximum likelihood bounded dependent variable German Socio-Economic Panel