1990 POPULATION CENSUS Detailed data based on a 2 per cent representative sample (1992)

III. THE METHOD OF SAMPLING

III. THE METHOD OF SAMPL1NG The 1 and 2 per cent samples used for preliminary processing of the 1970 and 1980 censuses, respectively, were selected by stratified systematic sampling: in each area unit (i.e. stratum) every lOOth dwelling was selected in the first case and every 50th dwelling in the second; all persons living in the dwellings selected were included in the sample. The programme of the 1990 census was not uniform in that two kinds of questionnaires were used, namely, basic and so-called representative questionnaires assigned to 80 and 20 per cent of the ED's (enumeration districts), respectively. The representative questionnaire covered numerous details and issues which the basic questionnaire did not deal with and contained all questions of the latter: this related both to the housing and the personal questionnaire. It was obvious that the 2 per cent sample of the 1990 census had to be restricted to the ED's with representative questionnaires in order that reliable data based on those questionnaires may be published in this volume; the systematic method used earlier would have produced a sampling fraction of 0.4 per cent for those ED's. This consideration led to a sample consisting of every tenth dwelling of each ED to which representative questionnaires were assigned. The two-stage sampling outlined above — ED's being selected in the first stage and dwellings in the second — ensures that characteristics corresponding to issues or questions of the representative and the basic questionnaire have the same level of precision as both they have a common 2 per cent sample basis. This level, however, is in the case of basic characteristics lower than what would have been obtained if we had had a systematic sample consisting of each 50th dwelling, since both the variability of ED's and that of dwellings within ED's contribute to sampling errors of the estimates. Selection of the 20 per cent sanple The 20 per cent sample of the census was selected from a universe of stratified clusters of different size; after the selection of 20 per cent of the ED's by computer somé corrections by hand took place to ensure the exact sampling fraction of 20 per cent alsó for dwellings as well as for persons. In the course of selection by computer the use of the 1980 enumeration districts as sampling units seemed straightforward. Deviation from this principle was envisaged only in the case of scarcely populated ED's in the outskirts where sampling units were created by collapsing those ED's. The actual selection process, however, definitely indicated that the 1980 ED's were too large sampling units in many cases, which would have implied strong heterogeneity in the size of sampling units and thereby difficulties in keeping the sampling fraction of dwellings at 20 per cent on the one hand, and less reliable data, on the other. Replacing the 1980 ED's by their subdistricts defined earlier enabled us to overcome these difficulties. In the case of institutional households, the sampling units were not the institutions but the institutional districts. In the case of smaller institutions, however, the creaxion of mixed districts consisting of both institutional and priváté households was accepted. The selection of the 20 per cent sample was carried out by counties and within counties by strata defined in terms of towns and villages (úrban and rural areas). In this stratification each big town was regarded as an individual stratum where the attribute "big" was defined as "at least of the size of the least county seat in 1989" which, due to preliminary data, amounted to 42,000 people. Two further strata 237