Assessing (and Addressing) Reporting Heterogeneity in Visual Analogue Scales (VAS) with an Application to Gender Difference in Quality of Life

Working paper number
2019-26
Publication Year
2019
Authors
Paper Abstract
In this study, we propose several new methods to account for reporting heterogeneity in self-reported data coming from Visual Analogue Scales (VAS) using corresponding VAS-based anchoring vignettes. Compared to usual Likert scale measures, VAS have the advantage that they lead to more nuanced assessments. Yet, like responses to Likert scale, VAS may suffer from individual-specific reporting heterogeneity. To the best of our knowledge, such reporting heterogeneity and potential solutions to solve this problem in the context of VAS measures have not yet been addressed in the literature. Using VAS-based anchoring vignettes and standard vignettes assumptions (vignette equivalence and response consistency), we show how standard fixed-effect approaches and double-index models can be used to address individual-specific reporting heterogeneity in VAS. We also show that several other methods such as Generalized Ordered Response models and Hierarchical Ordered Probit (HOPIT) models can be used to meaningfully adjust for potential reporting heterogeneity under the weaker assumption that VAS responses should be interpreted as ordered rather than cardinal data. We then apply our methods to real data assessing gender differences in Quality of Life (QoL) among students in Switzerland. While female students report higher levels of QoL than male students -as commonly found in the literature- we also show that female students tend to rate the QoL of corresponding comparable anchoring vignettes higher than male students. Accounting for these gender differences in response behaviors, we show that female students actually appear to be worse off in terms of QoL than male students. This finding suggests that reporting heterogeneity may be important in assessing gender differences in QoL and that the commonly found female advantage in QoL assessments may at least be partially due to differences in reporting behavior.