move, so only a portion of the sample is retained from one time to the next.
Longitudinal designs require that additional decisions be made concerning the length of the panel (that is, the number of years individuals are followed) and the frequency of data collection. In addition to the questions outlined above describing a periodic rich data collection effort supplemented by monitoring of the population through an abbreviated set of measures, the development of an ongoing, continuous panel design would have to address: the size of the sample needed to achieve analytic capabilities for a single calendar year versus pooled estimation across contingent years; the life of a single panel, that is, the number of years individuals will be followed through time; the periodicity of the data collection; the acceptability of mixed modes for data collection and its effect on the measurement and nonresponse properties of the resulting estimates; the use of a panel design requiring consistent response from the same respondents versus a mix of self and proxy response over time; and the ramifications of the decision on the error properties of the estimators. Several panel designs among the federal data collection efforts are shown in Table 5-1.
However, pure person and cohort samples also have the disadvantage of higher respondent burden since respondents will be asked to participate in several rounds of data gathering. To control the added burden of fully retained cohort samples, some type of rotating cohort sample might be used in the design of a disability monitoring system. For example, SSA might consider something comparable to the 4–8–4 rotation scheme that has been used by the CPS, in which a sample household is sampled for four consecutive months, not interviewed for eight months, then interviewed again for four consecutive months (Census and BLS, 2000). A very different design—a continuous overlapping panel design—is used in the Medical Expenditure Panel Survey. In that survey, members of a panel are interviewed five times over 24 months; a new panel begins at the start of each calendar year so that panels can be pooled to produce calendar-year estimates.
Survey data can be made richer by linking with appropriate administrative files maintained by SSA for both the SSDI and the SSI programs. Administrative data usually have no information on persons who have not applied for benefits and little information on socioeconomic variables. Household population surveys, on the other hand, provide information on persons who have not applied and on a wide range of socioeconomic variables but contain little or no information on the person’s interactions