Dropouts on the Web: Influence of changes in respondents' interest and perceived burden during the Web survey

Abstract

Dropouts in Web surveys may pose a serious threat to data quality, yet their causes are often unknown. This study aims to shed light on how respondent's motivation changes throughout the survey and how that affects dropping out. While most existing studies examine survey response at a single point in time, we try to grasp the dynamics of respondents' behavior by registering their perceptions of survey situation throughout the survey. The questionnaire consisted of an introductory part followed by 20 Web pages with several questions apiece. The pages differed in content and format of the questions, as well as in number of questions and average completion time. The order of the pages was randomized. At the end of each page, respondents were asked how interesting the questions on that page were and how much effort they invested in answering them. Changes in interest and perceived burden were analyzed for groups of respondents who dropped out at each page. The results show that interest usually decreases and perceived burden increases just before the respondent drops out. These changes in interest and burden are related to characteristics of the questions, respondents, and survey design. Respondents' tendency for dropout can be observed even before the dropout actually happens, through lower levels of interest, higher levels of effort, and higher item nonresponse in question blocks that precede the point of the dropout. Global evaluations of survey experience are not affected by overall duration of the questionnaire, but mostly by certain key moments during the questionnaire, most notably the maximum and ending levels of interest and effort. The results could help in practical work and open new paths to theoretical explanations of survey behavior.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 118 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.