CMStatistics 2019: Start Registration
View Submission - CMStatistics
B0952
Title: Estimating the population partly conditional mean using longitudinal cohort data with non-ignorable drop-out Authors:  Maria Josefsson - Umea School of Business, Economics and Statistics (Sweden) [presenting]
Abstract: Understanding how cognition changes during normal aging is important, since these more subtle changes still may affect day-to-day function, as well as for differentiating between normal and pathological states. Studies of cognitive aging using longitudinal data often result in highly selected samples due to selective study enrollment and attrition. An additional methodological challenge is practice effects, resulting in improved or maintained test scores despite a cognitive decline. These challenges may bias study finding and severely distort the ability to generalize to the target population even in well-designed studies. We propose an approach for estimating the finite population average of a longitudinal continuous cognitive outcome conditioning on being alive at a specific age, i.e. the population partly conditional mean. Specifically, we develop a flexible Bayesian semi-parametric predictive estimator, when longitudinal auxiliary information is known for all units in the target population. By specifying priors for the sensitivity parameters our approach allows uncertainty about untestable assumptions. The proposed approach is motivated by 15-year longitudinal data from the Betula longitudinal cohort study. We apply our approach to study normal cognitive aging with the aim to generalize findings to the target population.