Progress or Study Design: Declining Disparities in Diagnosed Chronic Disease Between Black and White Men in the Health and Retirement Study

Reducing racial health disparities was announced as a goal by the Federal government in 1990 and the American Medical Association in 1995.  Subsequent research, usually within a single cross-section context, has continued to document disparities on many dimensions.  Using data from the 2006 wave of the Health and Retirement Study (HRS) and its face-to-face supplement Chatterji et al., (2012) examine disparities in awareness of diabetes and high bold pressure.  They report that estimates are sensitive to selection associated with respondent choice not to answer--a survey design element.

 This paper asks whether progress in reducing disparities has been made—an inherently longitudinal question.  1996 and 2006 estimates from the Rand version of the Health and Retirement Study are analyzed for evidence of disparity change.  Relative conditional likelihood (Relative Risk) of black relative to white (the reference group) male respondents reporting having been told they have diabetes, high blood pressure, or heart condition by a doctor is the disparity indicator.  Change in estimates toward one over the decade suggests progress.

Sensitivity of estimates to the HRS attrition and replacement study design elements is also examined.  Knowing whether findings about progress in reducing racial disparities among chronic conditions that are highly associated with mortality and age may be influenced non-random attrition in longitudinal surveys of elders due to the increasing force of mortality with age (Thatcher, A. R., 1999) is important.  To that end, separate analyses were conducted on respondents in the 1996 wave lost to attrition, respondents who answered questions in both 1996 and 2006 survey waves, and replacement respondents. HRS replacement sample was added in 2004.

The research strategy is a follows. First, determine whether disparities in 1996 HRS data exist; next, determine whether chronic condition disparities between black and white mature men declined between 1996 and 2006 in the data without adjusting for the attrition, panel, and replacement design features; finally, determine whether results from step 2) are sensitive to attrition, panel, and replacement design features.  Relative conditional likelihood values are estimated by maximum likelihood using the method suggested by Zou (2004) as implemented in STATA: Generalized Linear Model with Poisson family, log link function, and robust standard errors.  Disparity is indicated by exponentiated estimated coefficient values.

Findings illustration: In the 1996 HRS data the high blood pressure disparity estimate for black men of 1.5 indicates that the likelihood of an affirmative response to the question is 1.5 times higher for black than white male respondents.  Without adjusting for attrition and replacement, the corresponding 1.3 estimate from 2006 HRS data suggests that progress was made over the decade.   Sensitivity of the finding to HRS attrition and replacement design features is examined by estimating the same model on data from respondents in the 1996 wave lost to attrition, respondents who answered questions in both 1996 and 2006 waves, and the set of replacement respondents only. Results indicate that design features matter and matter differently for different chronic conditions.

Chatterji, Pinka,.. (2012). Beware of Being Unaware: Racial/Ethnic Disparities in Chronic Illness. Health Economics