Modeling Interviewer Effects in the National Health Interview Survey

This chapter describes the use of multilevel models to estimate interviewer effects in the US National Health Interview Survey. The chapter uses multilevel models, adjusting for characteristics of the respondents and areas that make up the interviewers’ assignments, to estimate the amount of clustering in survey responses by interviewer. A second stage of analysis is conducted to examine how these interviewer effects vary by question characteristics (e.g., sensitive versus non-sensitive questions), and across groups of interviewers defined by interviewer characteristics (e.g., interviewer cooperation rates). A faster interview pace may lead to greater deviations from the interview protocol (e.g., reading questions exactly as worded) and, therefore, may be associated with larger interviewer effects on estimates. The interviewer effect, or the increase in variance of a sample statistic that is attributable to interviewers, is one way to measure this influence on survey responses. Interviewer characteristics are also associated with interviewer effects.