Relevance & Research Question: We study the (non)response behaviour in web surveys using paradata which offer a full insight into the questionnaire response process, allowing to observing the behaviour of respondents when going through the questionnaire. We use Bosnjak and Tuten’s (2001) classification of seven groups of web survey participants: complete responders, unit nonresponders, answering drop-outs, lurkers, lurking drop-outs, item nonresponderns and item non-responding drop-outs. This classification is important for understanding the web survey process and to develop standardized response metrics to measure web survey (non)response rates.
Methods & Data: We applied above classification to study the behaviour of respondents in a sample of 50 surveys selected out of more than 2,000 web surveys, hosted by a web survey service at our University. Participants were classified into seven categories based on the available paradata. These include time stamps, information on participants' equipment, data on how many pages of the questionnaire were viewed and how many responses given by each respondent before quitting the survey. We develop a methodology how to use paradata to classify units into the 7 categories. A two-level statistical analysis is then performed on collected data taking into account the level of individual respondents and the level of the survey.
Results: We show that classifying participants into several response categories and study their characteristics allow better insight into the survey response process. We also show which characteristics of participants (better equipment) and which characteristics of survey design (length, topic) influence certain response behavior.
Added Value: Studying several surveys with a meta-analytical approach and performing a two-level statistical analysis is relatively rare and new approach in this area. It provides precious insight into how the survey characteristics influence the behavior. Based on that, new knowledge is generated about tailoring the questionnaire design so to minimize various types of nonresponse behaviour. In addition, results enable to propose improvements of the seven group classification scheme.