Modelling the process leading to cooperation or refusal using interviewer call record data Seminar

Abstract
In recent years, survey agencies have started to routinely collect call record data in interviewer administered surveys, which includes both telephone and face-to-face surveys. Such call record data collect as a minimum, information about the day and time of the call, the outcome of the call, the household or sample member id and the id of the interviewer. It may also include further information, for example, characteristics of the person the interviewer talked to, interviewer observations about the household and information about the interaction between the interviewer and the household. Such call record data is a type of paradata (Couper, 1998) since it represents information about the survey data collection process. Survey agencies hope that analysis of such call record data may inform best interviewer calling practices. This paper aims to analyze the process leading to cooperation or refusal in several face-to-face household surveys using call record data. The analysis benefits from an unusually rich dataset, the UK Census Link Study, which combines paradata from six UK face-to-face household surveys, including detailed call record data, interviewer observations about the household and information about the interviewer-household interaction. These data were linked to information about the household from the UK Census. A key advantage is that all of this information is available for both responding and nonresponding households. The data have a multilevel structure with households nested within interviewers. A multilevel multinomial logistic regression approach is used which jointly models the different types of outcomes at each call to predict the likelihood of interview or refusal at the current or future call. The effects of both time variant and time invariant correlates are considered in the model.