Venue: The Fuqua School of Business, Duke University, 1 Towerview Drive, Durham, NC 27708-0120
Presentation
State Variation in Growth of Medicare Spending on Physician Services: Stylized Facts and Spatial Econometrics
Does state-level growth in Medicare spending on physician services ('physician spending') amplify Wennberg-type inefficiencies? Following Fisher and others (2003), policymakers have focused on one inefficiency in U.S. health care: across geographic areas, care-quality does not rise with physician spending. Related research (Wennberg and others (2002, 2004)) shows that the level of spending varies greatly geographically. For example, physician spending per beneficiary in California was 68% higher than in Oregon. However, to control physician spending, Medicare's mechanisms (e.g., SGR) have addressed its growth rate, not its level. Accordingly, using state-level data for the 1980-2004 period and sub-periods, we examine Medicare spending growth from several angles.
First, the paper develops stylized facts about the variation in growth rates of state-level physician spending. Specifically, we examine: (a) Convergence Did these state-level growth rates converge (relative to the median)? We classify states as convergent, divergent, or constant. For each grouping, we identify common characteristics (e.g., high degree of managed care penetration) and note temporal characteristics of the growth-rate paths (rate of convergence; monotonic; different pre- and post-RBRVS; etc.) (b) Regional patterns Were growth rates more similar within regions than between regions? (c) Neighboring-state patterns Was state i's growth rate more highly correlated with growth rates of its neighbors than with those of its non-neighbors?
Second, the paper examines the propagation of changes in the growth rate of one state's physician spending to other states' growth rates - in effect, whether state growth rates play a game of leapfrog. Specifically, we present econometric results from two models:
In the first model, the same regressors affect all states' growth rates in physician spending but the error terms admit spatial autocorrelation. In this seemingly-unrelated regression, a shock to the disturbance term in state i's equation can influence the disturbance term in state j's equation. Alternative prior restrictions are imposed on the spatial autocorrelation of state-level disturbances. Parameter estimates are compared between models with alternative restrictions.
In the second model, state i's spending growth-rate depends not only on its own lagged values and on other variables (e.g., income) characterizing state i but on lagged values of spending growth in other states. One example of this class of models is the cointegrated vector error-correction model. This approach tests for cointegration of the growth rates in adjacent states; extracts unit roots from the state spending growth-rate series; allows the lagged growth rate in state j to affect the contemporaneous growth rate in state i; and distinguishes short-run dynamics from long-run equilibrium relationships.
In both models, Bayesian restrictions are imposed - greater weight is given to information from adjacent states than from more-distant states.
Third, the paper presents Medicare policy simulations. For example, we depress the growth rate for selected states with high per-beneficiary physician spending and simulate the shock's propagation. The simulations suggest whether policies aimed at high-spending states could reduce trend growth in aggregate Medicare physician spending.
The paper concludes with implications for an efficient, sustainable Medicare; the study's limitations; and directions for research.