Mitigating biased measurement's effect on health disparities: An illustrative example of item response theory-structural equation-based models

Accurately understanding a diverse population's health and investigating health disparities across subgroups (e.g., majority and minority groups) requires equivalent cross-group measurement. Yet, little research addresses the possibility that systematic measurement error influences disparities research. Measurement bias refers to the possibility that individuals with identical health respond dissimilarly to questions about their health as a function of their race or ethnicity. Bias can obscure differences, decrease reliability and validity, and render comparisons impossible. Without establishing unbiased measurement, the field cannot draw strong conclusions about disparate outcomes, support evidence-based practice and policy, and address disparities.

Item response theory (IRT) and structural equation modeling (SEM) offer methods to test for measurement bias. They also provide model-based methods for mitigating bias. Unfortunately, bias studies have seen few public health applications, partly because too few receive training in their use or interpretation. In this presentation, I address this. Using data from the National Epidemiologic Survey on Alcohol and Related Conditions, I take an applied, non-mathematical approach and describe an SEM-IRT-based method to evaluate and correct for measurement bias across Whites (n = 16,480), African-Americans (n = 4,139), and Hispanics (n = 4,893). I show how bias as a function of minority status leads to erroneous conclusions about alcohol abuse disparities. I show how measurement bias subsequently systematically biases research examining the correlates of alcohol abuse and investigations of the effectiveness of alcohol abuse interventions across race and ethnicity. I conclude by showing how model-based estimates can mitigate this error, leading to more accurate conclusions.