Estimating the stability and crystallization of public opinion

In the past, survey researchers assumed that the respondents held opinions which could be obtained by asking the proper questions. This idea the “file drawing model” for survey response was criticized strongly by Converse in 1964, but also strongly defended by Achen (1975), Erikson (1978, 1979), and Judd et al. (1981). Zaller (1992) introduced the idea that people create their response on the spot when they are asked a question on the basis of the most salient considerations. The saliency of the different considerations can vary with the context in which the question is asked and the formulation of the question itself. This approach has gained a lot of attention in the last 20 years (Tourangeau et al., 2000; Tourangeau and Rasinski, 1988; Wilson and Hodges, 1991). In this study a model has been developed that unifies these ideas into one model (van der Veld and Saris, 2004). For this purpose a theory is needed that describes how a question leads to a response. This theory should account for the fact that the response could be different from the opinion, due to the (bad) quality of the measurement procedure. Furthermore, the model should allow that an opinion is made up of preexisting or stable considerations and temporary or unique considerations. A structural equations model that follows from the mathematical description of the survey response for an individual has been developed. With this structural equations model it is possible to decompose the variance of responses into three different components that also can be linked to commonly used concepts in survey research. The variance of the responses will be decomposed in variance due to pre-existing considerations, temporary considerations, and measurement error. These three variance components can be linked to opinion stability, opinion crystallization, and measurement quality respectively. 1 Data analyzed in this paper comes from the Russian Socio-Economic Transition Panel (1992-1999) that was funded by the Dutch Organization for Scientific Research (NWO). We acknowledge support from the Netherlands Institute of Advanced Studies (NIAS) where we resided during the academic year of 2003-2004, and worked on this paper. In addition we acknowledge the support of the Amsterdam School of Communication Research (ASCoR). 2 William van der Veld is assistant professor in Methodology of the Social Sciences at department of Communication Science of the University of Amsterdam. 3 Willem E. Saris is Full Professor in Methodology of the Social Sciences at the University of Amsterdam and professor at the ESADE business school of Ramon Llull University in Barcelona.