Longitudinal estimation in dual frame surveys

Traditionally, large surveys use a single sampling frame from which the sample is selected. As the population and methods used to collect survey data change, single frame surveys may miss parts of the population. In order to obtain better coverage of the population of interest and cost less, a number of surveys employ dual frame surveys, in which independent samples are taken from two overlapping sampling frames. Some current surveys follow the same households at regular time intervals so that longitudinal quantities such as transitions in employment status can be studied. Such surveys require new methods for analyzing the longitudinal aspects of the data. In this research, pseudo-maximum likelihood method and a two-stage procedure are used to develop the estimators in longitudinal dual frame complex surveys. These methods are necessary to estimate changes in poverty status or employment status of individuals over time. Properties of the proposed estimators are studied assuming that sample comes from a superpopulation. The estimators are shown to be consistent for estimating the population quantities, and to have an asymptotically normal distribution. In addition, corrected Chi-Squared type goodness of fit tests are developed and their asymptotic properties are derived. These tests allow us to use data from both frames as well as the sample design information to assess the model fit. A practical jackknife method is applied to estimate the variance of the estimators. The methods developed in this research are applied to the Current Population Survey and the Survey of Income and Program Participation, and result in more precision for estimating transitions in employment status.