of Nursing Homes Specification Tests of the Markov Model

In Norton (1990), I presented evidence that changing the reimbursement sys-tem to nursing homes to account for performance had a positive effect both on quality and on controlling costs. The analysis used a simple Markov model to estimate transition probabilities between states of health in the nursing home. A comparison of the probabilities for the control group (no incentives) and the experimental group (positive incentives) found them to be different. People in the experimental group stayed for a shorter time and had better outcomes. The simple Markov model maintains several strong assumptions. For ex-ample, it assumes that the transition probabilities are constant over time, in-dependent of past states, and the same for all people. If any of these assump-tions are false, the conclusions of the previous paper may be ill founded. This paper extends the analysis to more general models and in doing so subjects the simple Markov model to a series of rigorous specification tests. Most of the specification tests are done on data from the control group nursing homes only so as not to mix effects of the experiment with those of the assumptions