Interviewer Effects on Wave Non-Response in a Longitudinal Survey: A Multilevel Analysis
Rebecca Vassallo, Gabriele B. Durrant, Peter W.F. Smith
Southampton Statistical Sciences Research Institute, University of Southampton, Southampton, Hampshire, United Kingdom

The presence of significant interviewer effects on non-response in both cross-sectional and longitudinal surveys has been confirmed across several studies (Campanelli et al., 1999; Durrant et al., 2010; Pickery et al. 2001). However, further research is required to identify interviewer characteristics which explain some of this variation as well as interaction effects between sample members and interviewers. For longitudinal studies research is needed to identify the relative importance of interviewers across several waves. The aim of this paper is to analyse interviewer effects on wave non-response in a longitudinal study. The analysis is restricted to response conditional on contact. The main research questions are: is the probability of a respondent refusal at a particular wave dependent on the interviewer allocated? What is the relative importance of interviewers from different waves? Which interviewer-level variables explain part of this variation? Are there interaction effects between interviewer and respondent characteristics? Do any individual-level or area-level predictors of non-response vary randomly across interviewers?

The data for this study comes from the longitudinal Family and Children Study, which gathers information on the health and socio-economic status of households with children in the United Kingdom (Lyon et al., 2007). This dataset benefits from linkage to administrative data and detailed information on interviewers from a survey of both wave 7 and wave 8 interviewers, which in addition to survey data from wave 7, some participation history variables, and interviewer call record data provide a rich selection of potential explanatory variables for wave 8 non-response. The final analysis sample contains 7092 cases within a cross-classification of 329 interviewers at wave 7, 335 interviewers at wave 8, and 150 areas (primary sampling units). A cross-classified multilevel model is fitted (Durrant et al., 2010) modelling non-response at wave 8, with random effects for wave 7 and wave 8 interviewers and primary sampling units. The results will be used to inform best practices regarding interviewer allocation and training with the aim of reducing non-response in longitudinal surveys.


Campanelli, P. & O'Muircheartaigh, C. (1999). Interviewers, interviewer continuity, and panel survey nonresponse. Quality & Quantity, 33, 59-76.

Durrant, G. B., Groves, R. M., Staetsky, L., & Steele, F. (2010). Effects of Interviewer Attitudes and Behaviors on Refusal in Household Surveys. Public Opinion Quarterly, 74, 1-36.

Lyon, N., Mangla J., Tait, C., & Scholes, S. (2007). Families and Children Study (FACS) 2005, Wave 7 Technical Report (National Centre for Social Research, London).

Pickery, J., Loosveldt, G., & Carton, A. (2001). The effects of interviewer and respondent characteristics on response behavior in panel surveys - A multilevel approach. Sociological Methods & Research, 29, 509-523.

Keywords: interviewer effects; non-response; multilevel modelling; paradata

Biography: Rebecca Vassallo is currently reading for an MPhil/PhD in Social Statistics at the University of Southampton, under the supervision of Dr Gabriele B. Durrant and Prof. Peter W.F. Smith. Her general research topic is the analysis of paradata in social surveys with specific focus on the implications of the results on survey practice. She graduated with a B.A. (Hons) in Social Policy from the University of Malta and obtained an M.Sc. degree in Social Statistics from the University of Southampton.