Modelling health state valuation data

An important task in deriving a summary measure of health is to incorporate stated preferences into health state classifications. This paper examines a range of alternative econometric models for dealing with a skewed, truncated, non-continuous and hierarchical (at the individual level) data set of standard gamble valuations of health states defined by the health state classification the SF-6D. Alternative specifications incorporating random effects for the individual level clustering, Tobit models to deal with skewness and truncation and ordered probit for the noncontinuity did not produce models better than the simple mean model and were often worse in terms of inconsistency, fit and heterogeneity.