Systematic Measurement Error in Self-Reported Health: Is anchoring vignettes the way out?

This paper studies the pattern of non-random measurement error in self-assessed health responses across population subgroups and examines whether anchoring of vignettes can be used to identify this bias. It uses unique data from the World Health Survey (WHS)-SAGE survey(wave 1) from India, that has self-reported assessments of health linked to anchoring vignettes as well as objective measures like measured anthropometrics and performance tests on a range of health domains. Both estimations using individual fixed effects and anchored-vignettes response reveal strong systematic reporting bias across subgroups. Controlling for a battery of objective health measures, we implicitly test and confirm the validity of the ‘response consistency’ assumption used in vignettes technique. Further analysis using individual fixed effects in a two-stage regression estimation reveals substantial individual reporting bias even after accounting for the usual covariates controlled in a regression. The analysis finds that non-random measurement error in SAH cannot be simply dealt with by controlling for socio-economic covariates in a typical regression framework. This exposes the problem of cross-comparability using self-reported health response in the context of a developing country setting and lends support to the use of vignettes for identifying this bias.