Modeling Final Outcome and Length of Call Sequence to Improve Efficiency in Interviewer Call Scheduling

Abstract

Survey practitioners are increasingly interested in how best to use paradata to improve data collection processes. One particular question is if it is possible to identify early on during fieldwork sample cases that may require a long time, and therefore a lot of financial and staff resources, until interviewing is completed. More specifically, we aim to identify cases with long unsuccessful call sequences. This paper models call record data predicting final call outcome and length of a call sequence. Separate binary and joint multinomial logistic models for the two outcomes are presented, accounting for the clustering of households within interviewers. Of particular interest is to identify explanatory variables that predict final outcome and length of a call sequence. The study uses data from Understanding Society, a large-scale UK longitudinal survey. The work has implications for responsive and adaptive survey designs. The results indicate that modelling outcome and length of a call sequence jointly improves the fit of the model. Outcomes of previous calls, in particular from the most recent call, are highly predictive. The timing of calls and interviewer observation variables, although significant in the models, only slightly improve the predictive power.