4. Longitudinal Data Collection

A longitudinal design, either independently sponsored or in collaboration with one or more other federal agencies, offers analytic capabilities that are not possible with repeated cross-sectional designs. This is especially the case for those designs related to the decision to apply for benefits, including both individual factors that influence the decision and the impact of environmental and macro level changes (e.g., economic) on the decision to apply for benefits.

Selected reinterviews from large intermittent national surveys could provide needed information to assess change in status in different age, occupation, and gender groups. Much more can be learned from studying changes in individuals and their environments than from one-time cross-sectional measurement research. Such a design has high response rates, more ease in locating, and often better response reliability. Particularly where expensive screening was required for the initial sample, it need not be repeated and further subselection at different rates is possible. For example, one might follow all those currently on disability rolls, half of those with disabilities but not covered, a small fraction of those not reporting disabilities but with some health problems, and a still smaller fraction of the remainder of the population. What this means is that a combination of periodic large national survey with screening, and efficiently designed follow-up mostly by telephone, could continue the research on the disability policy questions, and the effectiveness of the process for determining eligibility for disability benefits.

The committee also suggests that SSA consider sample cohort rotation and integration with other federal surveys for the design of its disability monitoring system. Since samples with planned overlap over time perform more effectively in measuring change than independently drawn samples at each time point, some sort of cohort feature might be considered for the system. Several possibilities in decreasing order of statistical effectiveness are cohorts at the person, address, and cluster levels. A cohort in which the same persons are followed over time has the advantage of following those for whom disability is measured although the cost of follow-up can be extremely high to retain a high percentage over time. Drawing respondents from the same first-stage sample cluster (or primary sampling units) is the least costly of these options but also the least advantageous statistically since clusters often account for a relatively small part of the total variation in disability measures. A compromise to these two extremes is to return to the same sample of addresses for each round of a continuous sampling process. This approach is operationally effective since one returns to the same place each round (although address samples must be updated to accommodate new construction), but people

The National Academies of Sciences, Engineering, and Medicine
500 Fifth St. N.W. | Washington, D.C. 20001

Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement