An Analysis of Interviewer Effects on Screening Questions in a Computer Assisted Personal Mental Health Interview Herbert Matschinger, Sebastian Bernert and Matthias C. Angermeyer Abstract: This article is concerned with the explanation of unexpectedly low prevalence rates in the course of an epidemiological mental health survey. It is assumed that the probability of the endorsement of important screening questions is at least partially responsible for this, since it decreases in the course of the work of the interviewer. First the decrease in the probability of the screening variable is demonstrated by means of a conditional fixed within model, and then the probability of the screening variable is analysed by a mixture logistic regression within latent classes. Two models with 4 and 5 classes are estimated and compared with respect to the difference in interviewer behaviour. It is shown that only a small segment of the set of interviewers is responsible for the effect of the sequence of interviews on the probability of the screening variable and that the experience with the CAPI system is moderately associated with the latent classes. By means of this model, those interviewers could be identified who are responsible for the artefact under study. Keywords: Epidemiological survey, random coefficient model, latent class model, mixture model, logistic regression, interviewer effect
