L’indagine panel con rotazione su ingressi e uscite: alcune simulazioni con i dati INPS

The sampling procedure presented in this paper was used to collect panel data and it is based on the following rotation design. The panel life of the sample elements corresponds to their natural life. At different points of time, the deaths and outmigrants leave the panel; new elements are added, selecting them from the births and immigrants. Therefore, the rotation depends on “random” natural events, such as births and deaths, immigration and outmigration. The resulting panel should incorporate the target population dynamics and increase its representativeness. The unmatched and matched portions of successive samples provide estimates of the births, y(bt), and the continuing population, y(mt). In the matched portion, the doublesampling regression estimate is generally used, y’(mt). The best combined estimate of Y(t) is found by weighting y’(bt) and y’(mt) inversely as their variances (Cochran, 1977; Patterson, 1950). If y(bt) is systematically lower than y’(mt), then the doublesampling regression estimate does not work well. Alternative weights are the proportions of the births and the total units of population at time t. The matched portion may be estimated by a simple autoreressive model. This sampling design was applied to the administrative data of the National Institute of Social Security (INPS: Istituto Nazionale della Previdenza Sociale) for the years from 1982 to 1988, relative to the Province of Modena. The estimates of the population mean, total, and unit seemed to be satisfactory, as did their variances.