W Eight Adjustm Ents in Estim Ates for the 1999 M Edicare Health Outcom Es Survey

1.Introduction Large national sample surveys often use sample designs to cover the complexity of study objectives. Sample weights based on the features of the sample design are often incorporated into calculations to generate population estimates. The most common components of the sample weight are the base weight that reflects the different probabilities of selection and nonresponse adjustments. In general, the higher the probability of sample selection, the more likely a givenperson will be selected. Further, the greater the difference among subgroups in response rates and the analytic characteristics of interest, the greater the need to adjust survey weights for nonrespsonse. These weightadjustments are designed to reduce the bias in the survey estimates (Cohen 1998). M edicare Health Outcomes Survey (HOS) is a new quality measure for M edicare + Choice Plans, which is designed to measure two year change in physical and mental function among plan enrollees. The purpose of this study was to assesshow much difference in the estimates from a regression model the adjustment for nonresponse makes. Potential characteristics related to the HOS survey response status were investigated. The resultwas then used to establish the adjustment for nonresponse through the construction of homogeneous weighting class, in which the probability that a sampled individual will respond is the same for all individuals within a given class. Two weight components were considered: one related to the sample selection and the other related to the nonresponse. The effect of these weight adjustments was investigated through an analytic model.