Neural network imputation applied to the Norwegian 1990 population census data

Abstract

Abstract: When adequate administrative registers are available, population censuses can be carried out by supplementing the administrative data with data compiled from sample surveys. A problem is, however, to obtain reliable estimates from the survey data for small areas and groups. This article reports results from empirical experiments with neural network models for imputing individual values of survey attributes utilizing the available administrative data. The imputed values were used to prepare estimates of proportions for a population and for smaller subgroups of the population. The data used in the experiments were from the 1990 Population Census in Norway.