Keywords: Methodological challenges and improvements, including in the areas of sampling, measurement, survey design and survey response or non-response

In this paper, we examine two types of item nonresponse in a face-to-face survey: “don’t know” (DK) and “item refusal” (IR). Based on the cognitive model of survey response and the theory of survey satisficing, we derive hypotheses and explanatory variables on three levels: question, respondent, and interviewer characteristics. This analytic approach advances previous research on item nonresponse which only included one or two of those levels in their explanatory models. For example, research which used a mono-dimensional explanatory model showed that ambiguous wordings cause more measurement error than a well-formulated question which are easy to understand and to answer. However, those models on the question level do not allow the comparison to factors on the respondent and interviewer levels. In addition, relying on a mono-dimensional model discards the possibility to control for relevant variables on the omitted levels.
To test our explanatory model, we used data from the German Longitudinal Election Study 2013 (N = 2,003), a representative face-to-face survey of the German population. In order to derive measures on the question level, each question was coded with respect to the following question characteristics: number of words, question type, matrix question, sensitive question, hypothetical question, recall question, question on facts, and knowledge question. In addition to various respondents’ characteristics on the respondents’ level such as respondents’ education, age, and gender, the dataset was expanded by supplementary data about the interviewers.
The results of the multi-level models show that variables on all three levels significantly affect both types of item nonresponses. In detail, we show that some variables are associated with DK as well as IR (question difficulty, question format, respondents’ ability, and interviewer ability), whereas other variables relate to the two item nonresponse types differently (question type, question position in the questionnaire, respondents’ motivation, and interviewer ability). Finally, we derive recommendations in order to minimize both types of item nonresponse.