Semi-parametric identification of determinants of health expenditures – evidence from inpatients in China

Purpose – The purpose of this paper is to examine the determinants of average health expenditures for inpatients in China with national data for period 2002-2010 and regional data during 2005-2010. Design/methodology/approach – The semi-parametric framework is established to identify the determinants of health expenditures with local-constant least squares (LCLS) and local-linear least squares (LLLS) techniques. The LCLS technique aims to identify correlative determinants among all considered variables, and LLLS technique aims to further distinguish linear decisive and nonlinear control variables among all correlative determinants. Findings – First, root mean square error tends to decrease with irrelative variables smoothed out in regression model, validating the modelling reasonability of the semi-parametric approach. Second, the determinants of average health expenditures for inpatients exhibit considerable variation among regions despite the fact that governmental health expenditure, GDP per-capita, and urbanization do impact average health expenditures for inpatients to a certain extent. Third, both linear decisive and nonlinear control variables vary greatly with national, provincial, and regional data. Practical implications – First, the illiteracy rate should be further reduced nationally. Second, urbanization development and the average treatment number of inpatients for each physician per day should be strictly controlled in region A and C, respectively, in order to control average health expenditure for inpatients. Originality/value – First, the semi-parametric framework with LCLS and LLLS techniques allows for data structure-oriented model in regions rather than a uniform and definite model for underlying structure. Second, the research undertakes for the first time a comprehensive data analysis of the determinants of average health expenditures for inpatients with national and regional data in China.