Reducing Bias and Increasing Precision in the Measurement of Health Disparities

Measurement is the first step in addressing health disparities across groups: In order to effect change, we must know when discrepancies exist, understand their magnitude, and develop a correct etiological model. Measuring group differences is complicated by the fact that groups may differ not only with respect to their health outcomes, but also with respect to measurement of these outcomes (Burgard & Chen, 2014; Stewart & Napoles-Springer, 2003). This is called measurement noninvariances, and it can lead to inaccurate inferences regarding the nature and magnitude of health disparities if not modeled properly. Even worse, Millsap (1997) showed that measurement noninvariance propagates bias throughout a prediction model, causing inaccurate inferences regarding the causes and consequences of health outcomes and disparities.

Traditional approaches for attending to measurement noninvariance within social science research have called for limiting an item pool to include only those items which operate the same for all groups. This approach changes the definition of the outcomes of interest, and also reduces the internal consistency/reliability of the measure.

This presentation draws from Bauer and Hussong's (2009) moderated nonlinear factor analysis (MNLFA) model, which was developed for the purposes of integrative data analysis across studies, to incorporate differential item functioning (the cause of measurement noninvariance) into the measurement model, thereby increasing precision and eliminating bias in descriptive measurement and predictive modeling without limiting the item set. This approach can be used with traditional categorical groupings (like gender and race/ethnicity), as well as with continuous measures (like income or age). 

The empirical demonstration will include examples related to: gender differences in anxiety and depression; racial/ethnic differences in deviant behavior; and socio-economic differences in smoking.

References

Bauer, D.J. & Hussong, A.M. (2009). Psychometric approaches for developing commensurate measures across independent studies: traditional and new models. Psychological Methods, 14, 101-125.

Burgard, S.A. & Chen, P.V. (2014). Challenges of health measurement in studies of health disparities. Soc Sci Med, 106, 143-150.

Millsap, R.E. (1997). Invariance in measurement and prediction: Their relationship in the single-factor case. Psychological Methods, 2, 248-260.

Stewart, A.L. & Napoles-Springer, A.M. (2003). Advancing health disparities research: Can we afford to ignore measurement issues? Med Care, 41, 1207-1220.