Development of an Unbiased Longitudinal Model for Transitions in Functional Limitations
Xian Liu, Uniformed Services University of the Health Sciences (USUHS)
Charles Engel, Jr., Uniformed Services University of the Health Sciences (USUHS)
Han Kang, Department of Veterans Affairs
Kristie Gore, Walter Reed Army Medical Center
Direct application of linear mixed models would lead to inconsistent estimates of the effects on transitions in the number of functional limitations. In this research, we develop two generalized mixed models to address this selection bias problem, one parametric and one non-parametric. The parametric model is a two-stage perspective, developed as an extension of the traditional Heckman’s two-step linear regression model and involving three estimating steps. The non-parametric mixed model describes the process of functional status transitions as a joint distribution of two sequential events, developed to obtain unbiased estimates of transitions in the number of functional limitations when the assumptions for the parametric model do not hold. Our empirical example demonstrates that without considering the selection biases in the process of health transitions, estimation of the effects on transitions in the number of older persons’ functional limitations can be severely biased.
Poster Session 7