Problems in russian opinion polls

Many researchers, beginning with Campbell and Fiske (1959), Andrews (1984), and Saris and Andrews (1991), have suggested to use the multitrait multimethod approach to estimate the quality of measurement instruments. This approach has also been used in Russia, building on the experience of Andrews (1984) in the USA, Kbltringer (1995) in Austria, and Scherpenzeel and Saris (1997) in The Netherlands. Analyses of these data sets revealed no insurmountable problems, whereas analysis of the Russian data by this approach has raised serious problems with many of the experiments. In particu¬ lar, the multitrait multimethod experiments failed with topics where the correlation between repeated observations of the same trait within the interview is extremely high compared with the correlation between repeated observations of the same trait across the interviews. This indicates a contradiction, i.e. the high correlation within the inter¬ views indicates a stable opinion, whereas the low correlation across the interviews in¬ dicates a very unstable opinion. Assuming that the cross-interview correlation gives a more realistic picture, we argue that the high correlation within the interviews are contaminated by memory effects and/or response set effects, such as acquiescence. Al¬ though this is the first time that we have found these effects in multitrait multimethod experiments they may be more general than so far realised. Correspondence to: Drs. W. M. van der Veld ing. (vdveld@worldonline.nl) University of Amsterdam Department of Communication Sciences (ASCoR) Oude Hoogstraat 24, 1012 CE Amsterdam. Note: the above e-mail address can also be used to ask for the survey questions and the correla¬ tion matrices of the topics dealt with in this paper.