Systematic self-report bias in health data: Impact on estimating cross-sectional and treatment effects

Abstract

This paper examines the effect of systematic self-report bias, the non-random deviation between the self-reported and true values of the same measure. This bias may be constant or variable, and can mislead empirical analyses based on descriptive statistics, program evaluation and instrumental variables estimation. I illustrate these issues with data on self-reported and measured overweight/obesity status, and BMI, height and weight z-scores of public school students in California from 2004 to 2006. I find that the prevalence of overweight/obesity is 2.4-7.6% points lower in self-reported data relative to measured data in the cross-section. A school nutrition policy changed the bias differentially in the treatment and control groups so that program evaluations could find spurious positive or null impacts of the intervention. Potential channels for this effect include improved information and stigma. © 2011 Springer Science+Business Media, LLC.