Interviewer effects on latent constructs in survey research
ENInterviewers may have an influence on the answers given by their respondents. Survey researchers usually estimate these interviewer effects on the univariate distributions of survey variables. However, it is also possible that associations between survey items are influenced by interviewer effects. In this paper, we analyze the covariance structure for a set of survey items that can be used for factor analysis. It is hypothesized that interviewers affect the covariances between items belonging to the same, as well as to different, latent constructs. As a consequence, factor loadings and correlations between latent constructs may be biased when interviewer effects are ignored. In order to assess the effects of interviewers on latent construct analysis, multilevel covariance analysis is performed on nine items belonging to three latent constructs derived from data of eight different countries in the European Social Survey, round 5. Results indicate that interviewer effects on these associations occur, but that their impact on measurement models is rather modest.
