636 – Real-World Approaches to the Knotty Problems of Outliers, Faulty Values, and Covariates in Complex Sampling Designs
Reliability and Stability of the Six-Question Disability Measure in the Survey of Income and Program Participation
Matthew Brault
U.S. Census Bureau
Researchers have long acknowledged that disability is a dynamic characteristic (Adler 1992, Verbrugge, Reoma and Gruber-Baldini 1994, Wolf and Gill 2008). Nonetheless, the concept is often treated as static over short periods in longitudinal studies. The disability status of a respondent is asked during one interview and assumed to remain constant over several interviews or for the life of the panel. I explore this assumption using reoccurring data on disability status from the Survey of Income and Program Participation (SIPP). In the 2008 panel, the six-question set of disability questions from the ACS were added to a reoccurring topical module. I employ structural models from Heise (1969) and Wiley and Wiley (1970) to separate reporting error from real change under two assumptions about the measures' reliability. Both methods assume that disability status follows a first-order Markov process. With these methods, I find that the disability measures in the SIPP had relatively moderate to low reliability with coefficients between 0.414 and 0.638. Conversely, an individual's true disability status is strongly correlated with the person's status one year later (r=0.937). Thus, the supposition that disability remains relatively consistent over short periods has some validity.