From Typical Areas to Random Sampling: Sampling Methods in Russia from 1875 to 1930

Abstract

ArgumentWhile we know that Russian mathematicians contributed to the application of probability in the theory of statistics, Western literature on the history of statistics generally overlooks the role played by statisticians working in zemstva statistics departments. This paper argues that the quantity and the diversity of statistical data needed by zemstva administrators stimulated methodological innovations in the field and influenced the rise and development of sampling surveys in Russia between 1875 and 1930. The first surveys on parts of the whole were done by statisticians who were seeking solutions to practical administrative problems, and their sampling techniques evolved as Russian administrators put these statistical surveys to use. Finally, the paper argues that Kovalevskiy’s mathematical treatment of the theory of stratified sampling, published in 1924, ten years before Jerzy Neyman’s, is in fact a synthesis of the zemstva statisticians’ sampling practice and the Russian university statisticians’ theoretical works before 1917.

Representativeness in Statistics, Politics, and Machine Learning

“…Although stratification appeared in some earlier sampling work [28,109], Neyman helped establish it for a wider audience [18]. Stratification bridged the purposive-random tension, providing a way to increase the precision of estimates by incorporating knowledge of group differences into random sampling [47,148].…”

“…However, the representative method itself also emerged from changing economic and political demands on statistical "descriptions of the world" and related changes in thinking about parts and wholes [47, p. 211]. While earlier monography work intended to illustrate "holistic" truths by selecting the average type, sampling emerged from a need for information on variation among individuals [47,109]. As local charity solutions to poverty became inadequate in the context of national economic crises, surveys needed to provide not holistic knowledge of the typical family but nationally applicable knowledge of the condition of individuals so that governments could allocate and distribute aid.…”

Untitled

“…For example, Tschuprow (1923) and Kowalsky in his 1924 text both derived the optimum allocation formula for stratified sampling a decade before Neyman did so in his famous 1934 paper (after learning of Tschuprow's paper, Neyman (1952) recognized Tschuprow's priority for the results). Mespoulet (2002), Zarkovic (1956), Zarkovic (1962), andSeneta (1985) provide further details about early survey research and research on survey sampling in Russia.…”

Creating a Socialist Society and Quantification in the USSR

“…The first tests of surveys on a "part of the whole" probably took place in the 1870s. The period from 1885 to 1917 was marked by numerous efforts to perfect sampling methods, and by serious thinking by the Russian statisticians in this respect (Mespoulet, 2002). From 1895, they had the same discussions between themselves as their colleagues in other European countries about the shift from an exhaustive count to a partial survey, then about the shift from a sample constructed in a reasoned manner, based on the use of type, to a random sample based on calculated probabilities (on the history of sampling methods and sampling surveys, see Desrosières, 2002;Gigerenzer et al, 1989;Stigler, 1986).…”

“…The exuberant production of statistics in the USSR in the early 1920s created particularly favourable conditions for the rapid spread of sample surveys in the country (Mespoulet, 2002). Intensive use of such surveys from 1919 onwards stimulated thinking about sampling methods.…”