An extended hierarchical ordered probit model robust to heteroskedastic vignette perceptions with an application to functional limitation assessment.

Abstract

To improve interpersonal comparability of self-reported measures, anchoring vignettes are increasingly collected in surveys and modeled as the hierarchical ordered probit (HOPIT) model. This paper-based on the idea of psychological distance-relaxes the assumption of vignette equivalence in the HOPIT by allowing for heteroscedasticity in respondents' perceptions of vignettes. Particularly, we assume that respondents who are more similar to a vignette are more familiar with the condition described and therefore are capable of forming a more precise perception of the vignette. We show evidence in favor of this extended HOPIT through Monte Carlo simulations and an application concerning self-reported vision difficulty from the WHO Study on Global Aging and Adult Health (SAGE).