1970 HUNGARIAN CENSUS OF POPULATION Information on the data collection and processing (1977)

I. THE COMPREHENSIVE REWIEV OF THE DATA COLLECTION AND PROCESSING - 6. Sampling techniques applied in the census

the point in question is a grouped sample, the size of the sampling error is different according to the various categories considered. Consequently, the convenient sample size was defined by the categories which proved the least favourable from the point of view of the dispersion (occupational and cultural categories of the population). According to the preliminary results - for the sake of the convenient processing and treatment - the sample size was defined as the fourth part of the population. In order to prepare the planned representative sample survey connected with the census, examinations were made in October 1968 as to the confidence intervals of the sample taken in a census district of large representativity rate. At that time, only the numerical results of the one per cent sample used in the experimental registration of 1960 were at disposal, on the basis of which the dispersion values were examined and published in the case of seven categories applied (three demographic and four occupational categories). The mathematical computations performed on this basis ensured the comparison of the confidence intervals of the district sample, of the sample of the flats as well as of the individual sample - for the case of a simple and a stratified sample, according to different types of communities. Finally, the order of magnitude of the data lying within the confidence intervals have been fixed, using different sampling methods. Experimental evidence can be summed up in the following: - In the case if samples of the same size are compared, the confidence intervals of the district sample are considerably larger than in the case of selected flats or individuals (persons). With demographic criteria, the difference is negligible, while in the case of occupational criteria (being more homogenous within the district) the sampling error is a multiple value of the error estimated in the case of flats or individuals. - The correcting effect of the stratification is, however, the most conspicuous in the case of the district sample, mainly in the case of occupational problems where the standard error of the stratified district sample is but 0, 63 - 0, 68 per cent of the non.-stratified one. The correcting effect due to the stratification of the communities realized in the samples of the flats and persons has been scarcely felt. - In the case of the seven criteria examined, the confidence intervals of the stratified district sample are, in the 95 per cent significance level, i.e. calculating with 1,96-tuple standard error, rather large. As far as the demographic and several occupational criteria are concerned, they remain under 1 per cent while the maximum value amounts to 1, 6 per cent. It has to be considered, however, that the selected criteria are rather populous. Accordingly, the survey has pointed to the fields where a district sample based on high representativity rate (for example, 20-25 per fent) is the most suitable. Such is for instance the migration movement examined on the basis of the place of birth and the domicile at the former date, the survey of the fertility and so on. A possibly more accurate picture will be obtained as to the application of the 25 per cent district sample, if it is not compared with samples of flats or individuals of the same representativity (the idea of which never emerged), but with samples consisting of less numerous elements, as for example with samples being of 1 or 2 per cent representativity. In order to have the preliminary data of census processed, a 1 per cent sample of flats has proven reliable enough and ensured a rather disaggregated data publication. The existence of a sample of such unusually high representativity (25 per cent) was motivated by the larger confidence intervals of the selection according to census districts. If for the sake of the experiment, the standard errors of the district sample of 25 per cent representativity are compared with a 1 per cent individual sample or a 2 per cent sample of flats, the usefulness of the large district sample will be conspicuous. Its accuracy proves higher than that of the 1 per cent sample applied with the former data processing. By means of the micro-census of 1973 or the "panel" underlying it, the performing of more detailed computations and the opportunity of a more elaborate analysis will be ensured. Namely, the electronic data processing was able to clear up the scattering of the data originating from several census districts about the county mean, from the point of view of twenty important criteria. 44