Applying survey over-coverage bias to health disparities research

Population health researchers have traditionally assessed racial disparities by utilizing the concept of race as an “umbrella term,” without analyzing within-group heterogeneity. These comparisons are useful for tracking inequitably distributed health status metrics along racial lines, but may conceal important information about the determinants of racial health disparities and the most effective interventions for addressing them.

In this study, we concretely identify the magnitude of bias from ignoring within-race differences using the survey research framework of over coverage bias. Over-coverage occurs when observations are included within the sample from units beyond the population’s theoretical boundary. To the extent that responses from the ineligible units differ from those of the true target population, sample statistics will exhibit bias.

We decompose BMI data according to Hispanic ethnicity and reveal the more nuanced story of the risk of obesity among Hispanic Americans. Whereas the obesity rate among Puerto Rican Americans is roughly 7% higher than that of non-Hispanic white Americans, the obesity rate among Cuban Americans is roughly 7% lower than that of non-Hispanic whites. We find the bias of applying the overall Hispanic obesity rate is roughly 5% of the true obesity rate among Cuban Americans.

We recommend that researchers consider using this framework to identify cases in which ethnic-based weighting may be a useful supplement as the proportion of foreign-born Americans grows and simplistic racial classifications such as ‘Hispanic’ grow increasingly unsustainable.