- Project 5 ? Retention-Early Warning Project This project builds on two lines of inquiry that are rarely integrated. First, we draw on prior research that has investigated the impact of attrition in longitudinal studies of aging. Second, we incorporate research that has probed the impact of ?early warning? signs for later life health decline. Putting these together, we submit that those lost to attrition over time are likely to have worse cross-time health profiles compared to continuing participants (CP). The reasons for predicting greater decline among attriters are that: (a) they had more extensive profiles of early warning markers at baseline ? something we demonstrate with preliminary studies, and (b) these starting vulnerabilities are expected to increase their risk for subsequent stress exposures as well as worsening psychosocial profiles across time, the combination of which we hypothesize will account for their greater mental and physical health decline over time compared to continuing participants (CP) in MIDUS. To examine these issues, our specific aims are to: (1) locate and reinstate attriters from MIDUS 2 and 3 using proven survey methods; (2) characterize longitudinal change in physical (including biomarkers) and mental health between attriters and CP; and (3) test hypotheses about early warning vulnerabilities, subsequent stress exposures, and worsening psychosocial profiles to account for differences in cross-time health between attriters and CP.
These aims will be accomplished by using high incentives and state of the art methods for tracking and contacting attriters who will be given a condensed battery of psychosocial, cognitive, and health assessments via personal home visits. We have excellent current contact information on 85% of MIDUS attriters and expect to collect new data from about 20% (500?600 individuals) of the full attrition sample. Our work stands in contrast to other longitudinal aging studies, which have concluded (using multiple imputation methods) that attrition is not biasing obtained findings. We submit that these assertions may be problematic when those lost to attrition are not missing at random, the case for which we seek to test. To sharpen understanding of these differing accounts, we will examine findings on cross-time health from reinstated attriters to CP to alternative findings derived from multiple imputation methods that adjust for attrition using observed baseline data. We predict that multiple imputation will under-represent the magnitude of health decline observed among reinstated attriters. Further, we will explore an alternative imputation strategy that uses newly collected data on reinstated attriters to multiply impute values for attriters not reinstated. All such comparisons will be preceded by analyses to assess the degree to which the 20% reinstated attriters are representative of all attriters, and if necessary, correct for bias between the 20% and the 80%. To facilitate this widened range of analyses, the project now includes two biostatisticians with expertise in missing data methodologies in longitudinal research.
Showing the most recent 10 out of 278 publications
