Predicting risk selection following major changes in medicare

Abstract

The Medicare Modernization Act of 2003 created several new types of private insurance plans within Medicare, starting in 2006. Some of these plan types previously did not exist in the commercial market and there was great uncertainty about their prospects. In this paper, we show that statistical models and historical data from the Medicare Current Beneficiary Survey can be used to predict the experience of new plan types with reasonable accuracy. This lays the foundation for the analysis of program modifications currently under consideration. We predict market share, risk selection, and stability for the most prominent new plan type, the stand-alone Medicare prescription drug plan (PDP). First, we estimate a model of consumer choice across Medicare insurance plans available in the data. Next, we modify the data to include PDPs and use the model to predict the probability of enrollment for each beneficiary in each plan type. Finally, we calculate mean-adjusted actual spending by plan type. We predict that adverse selection into PDPs will be substantial, but that enrollment and premiums will be stable. Our predictions correspond well to actual experience in 2006.

Time is money: Outpatient waiting times and health insurance choices of elderly veterans in the United States

“…This property, known as the independence of irrelevant alternatives (IIA) can be tested (Greene, 2003) and conditional logit models fail these tests when applied with similar data to the choice of Medicare plans (Atherly et al, 2004;Pizer et al, 2008Pizer et al, , 2009. We follow this literature by selecting a nested logit model to address the rejection of IIA assumptions in the simple conditional logit.…”

“…The IIA assumption still applies within nests, but it is more likely to hold if nests have been chosen that group similar choices together. Pizer et al (2008Pizer et al ( , 2009 show that a three-nest structure (HMOs, Medigap plans, and traditional fee-forservice) passes within-nest IIA tests with MCBS data. This nesting structure is illustrated in Fig.…”

Nothing for something? Estimating cost and value for beneficiaries from recent medicare spending increases on HMO payments and drug benefits

“…Lagged total drug spending and lagged number of prescriptions are obtained from the prior year's MCBS dataset. We adjusted lagged total drug spending for moral hazard and possible under-reporting as described in Pizer et al (2008). The remainder of this appendix describes the computation of the average copayment for a plan.…”

“…This adverse selection problem would force plans to raise premiums repeatedly until most enrollees were driven away. Simulation analyses of the subsidy and coverage structure established by MMA indicated that PDPs will experience adverse selection, but the subsidy is large enough to ensure stability of this plan type (Pizer et al 2008).…”

“…"(HHS 2006). 8 SeePizer et al (2008) for market share and selection predictions for these plan types and for a high-benefit PDP design as well.…”