Improving Data Quality for Web Surveys in Real Time through Predictive Modeling Using Paradata

Paradata are a rich source of data that are collected through little additional effort by researchers. However, paradata are often underutilized. This study suggests a novel approach to use paradata to alter the survey itself in real time in order to improve data quality.

Through a predictive model, paradata about the responses will be utilized to alter the presentation of the survey questions themselves. First, if respondents straight-line through a grid section of the survey, following grids could be changed to single item questions in order to discourage straight-lining. Second, if respondents display multiple indicators of poor data quality, key questions could be moved forward in the survey to present earlier. This second option reduces survey length, lowers cognitive burden for respondents that are taking short cuts, and prevents drop-offs. Both of these techniques could help to improve data quality.

Though programming a survey to adapt in real time may involve a large effort in the beginning, once employed it could be used across projects for little additional cost. Improving data quality should be a goal of everyone in the survey research community. As web surveys continue to increase in frequency of implementation, the focus on data quality of this mode should be a priority.