Determinants of healthcare spending: a state level analysis

Abstract

Within the high and rising level of healthcare spending for the US as a whole is substantial variation in spending across states. Yet relatively little attention has been given to the empirical analysis of interstate differences in aggregate healthcare expenditures, and therefore little empirical evidence exists at the state level to guide policymakers. Using data for all 50 states for the year 1998, we estimate an empirical model that includes structural and reduced-form healthcare spending equations and a health production function to assess the significance, size and relative importance of factors that prior research indicates, may play an important role in explaining interstate variation in medical care expenditures, and the main pathways through which they operate. Our results indicate higher levels of healthcare spending for state populations with higher income, less education, fewer uninsured residents, less healthy lifestyles, larger proportion of elderly residents, greater availability of medical care providers and less urbanization. Our findings suggest that the most effective cost containment measures may be those that increase education and promote healthy lifestyles. Not only do these actions lead to reductions in healthcare spending, they also improve the health status of the population, and may help to achieve other important social policy goals.

Notes

¹Aggregate healthcare spending includes spending on hospital services, physician services, other licensed professional services (e.g. chiropractors, optometrists, podiatrists, independent practice nurses), dental services, prescription drugs, nondurable medical products (nonprescription drugs and sundries), durable medical products (e.g. optical goods), home healthcare, nursing home care and other personal healthcare (e.g. child, school, veterans, military and native American health programs).

²Research indicates that individuals often travel across state borders to receive healthcare services, especially residents living in rural areas that are located near large metropolitan areas, and also travel long distances to receive specialized services at well recognized facilities. Failure to account for exporting and importing of healthcare services between states will result in over-stating (net exporter) or under-stating (net importer) state level healthcare spending.

³The variable HMO does not measure managed care penetration, which includes preferred provider organizations as well as HMOs.

⁴For example, refer (Auster et al ., 1969; Silver, 1972; Fuchs, 1974; Hadley, 1982; Rosen and Taubman, 1982; Blair et al ., 1995; Lee et al ., 1995; Elo and Preston, 1996; Backlund et al ., 1999; Calle et al ., 1999; Deaton, 2002; Meyers et al ., 2002; Thornton, 2002; Smith et al ., 2005).

⁵ For example, refer (Auster et al ., 1969; Silver, 1972; Fuchs, 1974; Grossman, 1975, 2000; Newhouse and Friedlander, 1980; Kenkel, 1991; Elo and Preston, 1996; Backlund et al ., 1999; Thornton, 2002).

⁶For the structural spending equation, an approximate F-statistic of 1.95 (p=0.12) does not reject the overidentifying restrictions at conventional significance levels. An F-statistic of 24.41 (p<0.001) for the joint significance of the identifying instruments in the first-state regression indicates strong instruments.

⁷ We choose as instruments for DOC, the number of medical schools in a state and population density. For the quasi reduced-form equation, an approximate F-test of the overidentifying instruments provides evidence of exogeneity (F=0.07; p=0.80), and an F-test of their joint significance in the first-stage regression (F=9.49; p<0.01) provides evidence of their relevance. A Hausman test of exogeneity of DOC cannot be rejected at any reasonable level of significance (t=1.18; p=0.24). For the structural spending equation, we also cannot reject the overidentifying restrictions (F=1.15; p=0.35). A general methods of moments test (Newey, 1985) of the hypothesis that DOC is exogenous assuming CDR is endogenous cannot be rejected at the 5% level of significance (t=1.88; p=0.07).

⁸As Paul Feldstein has observed, if physicians were able to pass on higher malpractice insurance costs to patients and third-party payers, then ‘it is unlikely they would spend their time in marches on their respective state capitals seeking legislative relief’ (1988, p. 188).

⁹We estimated our general models using several alternative measures of MAL. (1) Premium measure reported in the article not adjusted for state cost of living differences. (2) Premium measure using census division average premiums as a proxy for all 50 states, with and without cost of living adjustments. Our results were very similar for all of these alternative measures.