Attrition bias is a problem in virtually all panel surveys and a variety of statistical methods have been developed to correct for it (see, e.g. the special edition of the JHR, vol. 33, no.2 which reports on attrition bias models in the PSID, SIPP and NLSY). Unfortunately, good validating data are rarely available to test the efficacy of these statistical methods. With the addition of Medicare claims records for members of the Asset and Health Dynamics Among the Oldest Old study (AHEAD, now integrated with the expanded Health and Retirement Study) it is now possible to validate these statistical measures of correcting for attrition bias for health event models. While the evaluations can not be generalized to all panels for all classes of models, they can greatly inform the debate. One of the primary research issues motivating the AHEAD is the need to understand better the relationship of aging and health transitions. The proposed research aims to analyze Health and Retirement Study public-use data on the AHEAD cohort in conjunction with HRS sample management information and linked administrative data from the Health Care Financing Administration. Specifically we propose to use Medicare record data for the AHEAD cohort to investigate the relationship between health events and panel attrition; to conduct secondary analysis of the combined AHEAD/Medicare data to estimate attrition bias in health event-history analyses of the elderly and to validate statistical methods of detecting and correcting for any such bias; to develop a systematic coding procedure for survey management data such as the non-interview forms and to produce a supplemental data file containing relevant information from these survey management records; to develop improved survey data acquisition protocols and to implement, under alternative funding mechanisms, these on future waves of the HRS and AHEAD study; and, to develop a dataset for all users of the restricted AHEAD/Medicare files that will assist in classifying diagnoses from Medicare claims data for comparison with the self-reported health conditions data collected in the AHEAD.
