Continuity Trumps? The Impact of Interviewer Change on Item Nonresponse

Allocating the same interviewer to each respondent over multiple waves is typically recommended for panel surveys. While some studies have investigated the effect of this strategy on wave nonresponse, there is scarce empirical evidence on how interviewer (dis-)continuity affects item nonresponse. This is surprising, given that the amount and pattern of item nonresponse is a crucial aspect of data quality. Using the first seven waves of the German Family Panel pairfam, we investigate whether interviewer continuity indeed influences item nonresponse in a non-experimental setting. Our analysis differentiates between "I don't know" responses and the complete refusal to answer, both with respect to the specific question of household income as well as the entire question program. By applying cross-classified multilevel models and estimating within-respondent effects, we can base our results on an intra-individual comparison, controlling for time-constant unobserved characteristics and taking into account the complex structure of the data. Our analysis shows no detrimental effect of an interviewer change, per se, over the course of the panel;a new interviewer only collects more "I don't know" answers if the respondent belongs to the oldest age group (born 1971-73). Younger respondents, in contrast, have a lower likelihood to answer with "I don't know" to the income question if they experience a change in interviewer. Changes in social distance with respect to age and gender do not prove to be relevant mechanisms for this effect. Only female respondents of the youngest cohort exhibit a lower likelihood of "I don't know" responses on the income question when reassigned to a female interviewer. Older interviewers tend to get more "I don't know" answers, whereas reassignment to a more experienced interviewer (regardless of age) appears to encourage less "I don't know" answers