Questionnaires are foundational tools in social, behavioral and cognitive sciences. While the past century witnessed significant advances in the design and deployment of questionnaires, comparatively little attention was paid on how to structure the resulting data. The ubiquitous current practice consists in adopting a tabular data format, where each row in a table represents a respondent and each column a specific question, response or other attribute. This format has clear advantages when applied to small datasets. However, when dealing with more complex datasets, this format quickly becomes impractical, error prone and difficult to reuse. Furthermore, it prevents questionnaire research from taking advantage of modern digital technologies, multi-modal assessment and advanced data analytic methods. As an alternative, we recommend instead the tidy data format and show how it can handle most of those limitations. This change of format reflects a deeper change in perspectives: focusing on the person-by-question interaction as being the key observational unit rather than the person-by-questionnaire(s). The tidy format also makes it more apparent that the same data model could in principle be used to structure data not only from a wide range of questionnaires but also from computerized cognitive tests where the tidy format is already commonly used (i.e., each row in such tables representing a “trial”, referring to a person x stimulus interaction). A shared data model for behavioral and questionnaire data–as proposed by the Behaverse data model–simplifies data documentation and reuse, the sharing of data analysis methods across both types of instruments (e.g., analysis of response times in questionnaires), but also affords the development of a new generation of software tools, data analytic methods, and research questions, which may ultimately improve scientific research quality and yield novel insights about the human mind.
Cardoso-Leite et al. (Thu,) studied this question.