Estimating interviewer effects for binary responses

In surveys where interviewers need a high degree of specialist knowledge and training, one is often forced to make do with a small number of highly trained people, each having a high case load. It is well known that this can lead to interviewer variability having a relatively large impact on the total error, particularly for estimates of simple quantities such as means and proportions. In a previous paper (Davis and Scott, 1995) the impact for continuous responses was looked at using a linear components of variance model. However, most responses in health questionnaires are binary and it is known that this approach results in underestimating the intra-cluster and intra-interviewer correlations for binary responses. In this paper,a multi-level binary model is used to explore the impact of interviewer variability on estimated proportions.