The determinants of health expenditures: evidence from US state-level data

Abstract

Most macro studies of what determines health expenditures have used the same panel of OECD country-level data. Based on a more homogeneous panel data set of US states we constructed, this note applies the model selection procedure to identify the determinants of health expenditures at the state level. We find that the four key factors are gross state products, the proportion of the population over the age of 65, the degree of urbanization and the number of hospital beds. The cross-section income elasticity of health care is around 0.7, implying that health care is a necessity rather than a luxury good at the state level. The (relative) price of health care varies significantly across states but does not appear to have real effects on the amount of resources (measured in real dollars) a state devotes to health care.

Notes

¹ Due to data availability, neither Di Matteo and Di Matteo (1998), Giannoni and Hitiris (2002) nor Freeman (2003) considered price variations across regions (e.g. states and provinces). Instead, Di Matteo and Di Matteo (1998) used national price indexes to deflate state/provincial income and health expenditures. Freeman (2003) did not deflate the two series. Di Matteo (2005) deflated the 51 US state data (including Washington DC) using eight regional CPI indexes published by the Bureau of Labor Statistics (BLS).

² Since participation is voluntary and the survey is not randomized, we cannot exclude the possibility that factors affecting an area's decision to participate in surveys may also affect its health care expenditures. If this is the case, then the omitted variable problem arises. However, it can be difficult to identify which explanatory variables we use are correlated with the decision to participate. Therefore, the direction of bias, if any, is not predictable.

³ Note that we assume health expenditure outcome of each state is a realization of the same stochastic process. The state economy or population size itself is of no relevance. States would be assigned the same weight if they had the same participation rate in the surveys.

⁴ The endogeneity problem may arise in Model (1) since HCE and GSP may cause each other contemporaneously. To address this issue, we used one-year lagged GSP as the instrument variable for the current GSP. The basic results reported in the paper still hold, which suggests that the endogeneity bias is likely to be small.

⁵ The results based on AIC are similar to those in Table 3 with one significant difference: both price and the number of active physicians are selected in the model for year 2000. The coefficients are 0.30 and 0.16 and the corresponding t-statistics are 1.51 and 2.03, respectively.