Ethnic/Race Differences in the Attrition of Older American Survey Respondents

The patterns of attrition among elderly participants in health research studies vary significantly by race and ethnicity. As population diversity increases, researchers should consider these differences when assembling survey panels in order to retain minority participants and avoid biased estimates.

Although there is extensive literature on minority inclusion in health research, this is the first study to compare “patterns of attrition” between elderly minority groups participating in health studies. More specifically, using data from the “most current, comprehensive, longitudinal study of aging”—the Health and Retirement Study (HRS) 1992-2008—researchers examined whether foreign birth, health insurance and health status are associated with attrition of elderly participants, and whether these patterns vary across racial/ethnic groups.

Key Findings:

  • Attrition patterns associated with immigration status, insurance and health status differed enough among racial and ethnic groups that they could lead to biased estimates.
  • For all ethnic groups, those elderly born outside of the U.S. appear to be more likely to attrite from the study than to die, the opposite of the probabilities displayed for non-Latino Whites considered alone.
  • Previous research examining the racial and ethnic differences in mortality rates and in the onset of health decline and disability may underestimate those differences, because of the bias introduced by the high relative attrition of Hispanics.
  • Both the “Hispanic paradox,” in which Hispanics appear in studies to have lower mortality compared to similar Whites and African-Americans, and the apparent lack of a mortality differential between Hispanics with and without health insurance, may be explained by ethnic differences in attrition.

This study was limited by the relatively small HRS sampling of Latinos but suggests researchers should pay attention to differences in attrition across aging minority populations when assembling study panels.