Unobserved Factors Linking Functional Decline and Depression Among the Oldest Americans

摘要

This study considers the dependence between functional decline and depression in a nationally representative sample of older Americans from the Survey on Assets and Health Dynamics among the Oldest Old (AHEAD) covering the years from 1993 to 2002. Previous research has shown that depression is a significant predictor of functional decline and, conversely, functional limitations lead to more depressive symptoms. While this cross-dependence is an established fact in the literature, relatively few prior papers formally modeled the association between functional decline and depression. In this paper, functional decline is defined as 2 or more limitations in 5 Activities of Daily Living (ADL) and 2 or more limitations in 5 Instrumental Activities of Daily Living (IADL) from the baseline to the last available follow-up interview. Depression is defined as 3 or more points on the 8-item Center for Epidemiological Studies Depression (CES-D) Scale over the same range of time. In the analytic sample of 5,470 oldest Americans, each of the three measures is initially estimated in the univariate probit model controlling for a rich set of available risk factors identified in the previous studies. Then it is argued that univariate probit models are incapable of capturing individual differences (for example, predisposition to both physical and emotional ill health) that may link functional decline and depression in the oldest Americans. Thus, a more advanced multivariate probit model is employed, and three discrete equations are estimated jointly. In this way, unmeasured factors specific to the individual will become part of the error terms, and statistically significant correlations in the variance matrix will indicate dependence between functional decline and depression. Estimation of multivariate probit model reveals substantial unobserved heterogeneity in the dynamics of ADLs, IADLs, and CES-D score over time. Thus, previous results based on univariate methods should be interpreted with caution.