Objective: Characteristics of interviewers and respondents have been shown to influence the quality of data from survey research in various domains. There is little evidence for such effects in alcohol research, however. The purpose of the study reported here was to examine effects of gender and age of interviewers and respondents simultaneously This was done using hierarchical linear modeling, the advantage of which is that it can account for the clustering effects of respondents being nested within interviewers.
Method: Data were obtained from the first wave of an ongoing randomized longitudinal study on changes in alcohol consumption in Switzerland. The response rate was 77.9%. Analyses were based on 2,746 (1,749 male) subjects with an average of at least monthly consumption in the 6 months before the telephone interview. Consumption was assessed by means of a graduated frequency measure. Five different hierarchical linear models of increasing complexity were used to test several hypotheses of interviewer and respondent effects. Findings from hierarchical linear modeling were compared with those from "classical" analysis of variance.
Results: A theoretical design effect of 1.89 attributable to interviewers was found. Both analyses of variance and hierarchical linear modeling provide support for a structure with a main effect for gender of respondents, as well as a main effect for age of respondents and an interaction effect between interviewers' and respondents' ages.
Conclusions: Interviewer effects affect the estimation of statistics in survey research and must be adjusted for either by means of multilevel analysis or by the use of specialized sample survey software.