Analysis

Opening remarks of the Symposium 2002: Modelling Survey Data for Social and Economic Research, presented by David Binder.

We review the current status of various aspects of the design and analysis of studies where the same units are investigated at several points in time. These studies include longitudinal surveys, and longitudinal analyses of retrospective studies and of administrative or census data. The major focus is the special problems posed by the longitudinal nature of the study. We discuss four of the major components of longitudinal studies in general; namely, Design, Implementation, Evaluation and Analysis. Each of these components requires special considerations when planning a longitudinal study. Some issues relating to the longitudinal nature of the studies are: concepts and definitions, frames, sampling, data collection, nonresponse treatment, imputation, estimation, data validation, data analysis and dissemination. Assuming familiarity with the basic requirements for conducting a cross-sectional survey, we highlight the issues and problems that become apparent for many longitudinal studies.

A commonly used model for the analysis of time series models is the seasonal ARIMA model. However, the survey errors of the input data are usually ignored in the analysis. We show, through the use of state-space models with partially improper initial conditions, how to estimate the unknown parameters of this model using maximum likelihood methods. As well, the survey estimates can be smoothed using an empirical Bayes framework and model validation can be performed. We apply these techniques to an unemployment series from the Labour Force Survey.

Estimation of the means of a characteristic for a population at different points in time, based on a series of repeated surveys, is briefly reviewed. By imposing a stochastic parametric model on these means, it is possible to estimate the parameters of the model and to obtain alternative estimators of the means themselves. We describe the case where the population means follow an autoregressive-moving average (ARMA) process and the survey errors can also be formulated as an ARMA process. An example using data from the Canadian Travel Survey is presented.

Goodness of fit tests, tests for independence in a two-way contingency table, log-linear models and logistic regression models are investigated in the context of samples which are obtained from complex survey designs. Suggested approximations to the null distributions are reviewed and some examples from the Canada Health Survey and Canadian Labour Force Survey are given. Software implementation for using these methods is briefly discussed.

Opening remarks of the Symposium 2002: Modelling Survey Data for Social and Economic Research, presented by David Binder.

We review the current status of various aspects of the design and analysis of studies where the same units are investigated at several points in time. These studies include longitudinal surveys, and longitudinal analyses of retrospective studies and of administrative or census data. The major focus is the special problems posed by the longitudinal nature of the study. We discuss four of the major components of longitudinal studies in general; namely, Design, Implementation, Evaluation and Analysis. Each of these components requires special considerations when planning a longitudinal study. Some issues relating to the longitudinal nature of the studies are: concepts and definitions, frames, sampling, data collection, nonresponse treatment, imputation, estimation, data validation, data analysis and dissemination. Assuming familiarity with the basic requirements for conducting a cross-sectional survey, we highlight the issues and problems that become apparent for many longitudinal studies.

A commonly used model for the analysis of time series models is the seasonal ARIMA model. However, the survey errors of the input data are usually ignored in the analysis. We show, through the use of state-space models with partially improper initial conditions, how to estimate the unknown parameters of this model using maximum likelihood methods. As well, the survey estimates can be smoothed using an empirical Bayes framework and model validation can be performed. We apply these techniques to an unemployment series from the Labour Force Survey.

Estimation of the means of a characteristic for a population at different points in time, based on a series of repeated surveys, is briefly reviewed. By imposing a stochastic parametric model on these means, it is possible to estimate the parameters of the model and to obtain alternative estimators of the means themselves. We describe the case where the population means follow an autoregressive-moving average (ARMA) process and the survey errors can also be formulated as an ARMA process. An example using data from the Canadian Travel Survey is presented.

Goodness of fit tests, tests for independence in a two-way contingency table, log-linear models and logistic regression models are investigated in the context of samples which are obtained from complex survey designs. Suggested approximations to the null distributions are reviewed and some examples from the Canada Health Survey and Canadian Labour Force Survey are given. Software implementation for using these methods is briefly discussed.