Protest and the Politics of Blame: The Russian Response to Unpaid Wages

Appendix A. How the Survey Was Conducted

Appendix A How the Survey Was Conducted This study is based on ‹ndings from a nationwide survey of 2,026 adult (age eighteen and over) Russians. Face-to-face interviews were conducted between September 27 and October 12, 1998, by a Moscow-based research ‹rm, the Institute for Comparative Social Research (CESSI) for the U.S. Information Agency (USIA). The survey design involved a multistage strati‹ed cluster sample. For the ‹rst stage of sampling, Russia was divided into six geographically based strata: Center and North, Volga and Volgo-Vyatsky, North Caucasus, Urals, West Siberia, and East Siberia and the Far East, except for the two largest cities, Moscow and St. Petersburg , which were treated as self-representing strata. The non-self-representing strata were each divided into substrata of high-protest urban areas, low-protest urban areas, and rural areas, based on of‹cial data from the Russian State Statistical Committee for the number of people and enterprises involved in strikes in the past four years. Cities from oblasts in which 3.5 percent of the population or more engaged in strikes from 1995 through June 1998 were classi‹ed as high protest (Kemerovo, Komi, Rostov, Kurgan, Buryatia, Kostroma, Amur, Irkutsk, Archangelsk, and Bryansk). Cities from the remaining oblasts were classi‹ed as low protest.1 In the resulting design, the Volga and VolgoVyatsky stratum does not have a high-protest substratum, but two strata, Center and North and East Siberia and Far East, have two high-protest substrata each to maximize the chances for a wide dispersion of sampling points within these strata. The total number of strata, including Moscow and St. Petersburg, is twenty-one. To ensure that comparisons of politically active and less active Rus243 1. The data from 1995–98 for the number of strikers in each oblast were added and then divided by the population in that oblast. This is not a precise method of determining the true percentage of the population who have participated in strikes because some workers could have participated in many strikes and therefore been counted several times. We used these numbers only as rough approximations for the purpose of classifying oblasts and their cities into high and low protest strata. sians could be made with reasonably high levels of con‹dence, the sampling scheme was designed to achieve a greater number of interviews from politically active Russians than might otherwise be achieved through random sampling in proportion to the size of each stratum. The high-protest strata were systematically oversampled, and the low protest and rural strata were systematically undersampled. Fifty percent of the interviews were allocated to the high protest strata, and 50 percent were allocated to the low protest and rural strata. The data were later reweighted to re›ect more closely the true probability of selection based on population size and adjusted to re›ect educational distributions within strata. The number of interviews assigned was based on an assumed response rate of approximately 70 percent for cities and 85 percent for villages for a targeted achieved sample of at least 1,800 respondents. Twenty-eight interviews were assigned for every twenty achieved interviews in urban primary sampling units (PSUs) (and thus fourteen for every ten achieved interviews in urban sampling points), and twenty-four interviews were assigned for every twenty achieved interviews in rural PSUs (and thus twelve for every ten achieved interviews in rural sampling points). PSUs in the urban areas were cities, and sampling points were electoral zones. PSUs in the rural areas were regions (raiony), and sampling points were villages. In the self-representing strata, PSUs and sampling points were electoral zones. In all cases, PSUs and sampling points were selected by the method of probability proportionate to size. Households were selected randomly, and individuals were selected randomly using Kish grids. The ‹nal sample contained 92 PSUs and 184 sampling points with approximately ten interviews per sampling point. The overall response rate was 81 percent of the 2,516 targeted individuals , or 2,026 respondents. After weighting, the ‹nal sample size is 2,021. The margin of error for a simple random sample of this size would be ± 2.2 percent at the 95 percent con‹dence level. Although strati‹cation tends to reduce sampling error, clustering—by sampling among PSUs and sampling points—tends to increase it. We thus multiply the sampling error for a simple random sample by the square root of two, as a reasonable approximation of the design effect...