*Wissenschaftliches Programm*   *Liste der Vortragenden*

Sektion 8
Montag, 18.09.2000, 15.30–15.50 Uhr, POT 106

Use of Register Data in a Sample Survey

Tiberiu Postelnicu, Centre for Mathematical Statistics,Bucharest

Auxiliary data obtained from administrative and other register are frequently used in National Statistical Agencies to improve the estimation in a sample survey. There are alternative techniques toincorporate register data in an estimation procedure. In this paper,an overview on the use of register data for statistical purpose,especially for social statistics,is given. Polytomous survey variables are commonly encountered in survey practice. A new estimation procedure was introduced by Lehtonen and Veijanen for the estimation of class totals of such variables. Logistic generalized regression estimation (LGREG) is based on a multinomial logistic model, which might be more realistic for class indicators than the linear model norrmally used in generalized regression estimation (GREG). The aim of this paper is to compare LGREG and GREG estimators, which yield identical results for complete poststratification, but differ for other models such as ranking. LGREG is closely related to the standard GREG. In standard GREG estimation, the available register data are incorporated in the estimation procedure by a linear model fit. In LGREG estimation, the joint distribution of the class indicators is described by a multinomial logistic model.
REFERENCES:

[1]
Lehtonen R. and Veijanen A. (1998): Logistic generalized regression estimators. Survey Methodology, 24, 51-55.
[2]
Hold D. and Smith T.M.F. (1979): Post stratification. Journal of the Royal Statistical Society A, 142, 33-46.
[3]
Nordberg L. (1989): Generalized linear modeling of sample survey data. Journal of Official Statistics, 5, 223-239.