Adjusting for nonresponse bias due to attrition and intermittent wave nonresponse in longitudinal surveys of older adults

Understanding the dynamics of age-related health decline is increasingly a high priority for society because of the realities of population aging. In longitudinal studies, participants may be present for some waves of data collection and missing for others, that is, wave nonresponse. The first research topic in this dissertation is to explore and compares the determinants of multiple sources of nonresponse. The results indicate that there are potentially varying missingness mechanism underlying nonresponse due to refusals, noncontacts, proxy interviews, and death. The second topic is to examine the impact of incorporating multiple sources of attrition in missing data adjustment using sensitivity analysis with latent pattern-mixture models. The substantive model estimated is memory decline trajectory among the older adults in the United States. The results show little impact of incorporating sources of attrition, but incorporating it helps estimate a parsimonious model. The third topic examines the sequential regression multiple-imputation method as an approach to handle multivariate missing data. The results show that this approach produces comparable results with the maximum likelihood estimation.