A prediction approach to representative sampling

After a discussion on the historic evolvement of the concept of representative sampling in official statistics, we propose a prediction approach to it. 1 The birth of representative method The word birth should not be taken literally, as there are many fathers and few (if any) mothers involved in the process. As a matter of fact, social and demographic statistics emerged as a result of partial investigations, rather than census-like investigations. In Stephan (1947) is given a number of early statistical analyses based on incomplete investigations. In Kruskal and Mosteller (1979a,b,c) they investigate the origin of “representative investigations”, and give several examples, some of which go as far back as “times of Athens”. Finally, in Seneta (1985) is elucidated the origins and development of the representative method in the Russian Empire/Soviet Union. The authors do not agree on all issues, but there seem to be a general agreement that the Norwegian statistician A. N. Kiær may be credited with the real “fatherhood” of the use of the representative method in statistical practice. The Russian statistician concludes in his book (Kaufman, 1913): “The honourif not of the discoverybut in any case of the systematic treatment of the method, is essentially due to the Norwegian statistician Kiær who in the nineties of the last century made this method the basis for a number of inquiries and acted as a zealous propagandist of the same ...” Kiær introduced the method for an international audience at the ISI meeting in Bern, August 1895. He presented results from a survey based on the 1890 Census. The design used was what to day would be called a three-stage design. In the first stage, 128 counties and 23 towns were selected throughout the country. In the second stage, a sample of males with ages 17, 22, 27, 32 etc. was selected. Finally, in the third stage, males who’s name started with the letters A, B, C, L, M , N were selected. The 1890 Census was used as a frame. To evaluate the representativity of the sample, he produced a number of tables of marginals from the survey, and compared the results with the census tables. In his conclusions from the meeting, he stated: “I find that 1. the sampling methods should be of great importance for the development of statistics chiefly in the special representative investigations are arranged so that they can be controlled with regard to the chief points with Statistics Norway, Kongensgate 6, PB 8131 Dep, N-0033 Oslo, Norway. E-mail: [email protected]