Applying Critical Race Theory to Research: Assumption of Non-Independence
Critical Race Theory is not, in fact, a singular theory in the traditional sense of cause and effect. It does not lay out a full set of theoretical components that produce an expected outcome, as is typical in most major public health theories. Although some researchers have essentialized this tradition (e.g. the five tenets of CRT), such distillation does not account for the breadth and depth of CRT. Arguably, this has been more harmful than helpful. CRT has offered a critique of the failings of the liberal state, mounted a vigorous challenge against so-called objective principles in law, drawn attention to systems thinking in educational and public health theory, and facilitated robust dissension among its adherents.
CRT is more accurately understood as a school of thought spanning multiple disciplines including critical legal studies, philosophy, education, and research methodology. CRT theorists share a belief that race and racism are indispensable social constructions that exert major influence throughout political, social, and economic life and warrant constant attention and evaluation. When applied fully, CRT in public health shapes every aspect of program and research planning and implementation - from defining the problem, designing a program or study, and to interpreting the outcomes and results. CRT largely remains in its infancy in public health because few research studies have sought to build upon CRT meaningfully.
The purpose of this article is to advance CRT in public health research. It seeks to apply the CRT assumption of structural racism to analyzing data. CRT has a major assumption that can be tested and proven. If racial membership reflects non-random distribution whereby members of different races are not independent because of the shared racial oppression or privilege, then the implication is that they are nested within contextual units. Further, that nesting could mean that two observations in a sample are related to each other, which would constitute a violation of the assumption of independence. It is analogous to taking multiple samples in a neighborhood survey from the same household, a violation that any researcher would easily recognize. In the case of CRT, that “household” is race. Stark economic, health, and social racial differences have been well-established, which raises serious concerns whether the assumption of independence could ever be met in a society as racially stratified as the United States. Random sampling is used to avoid violating this assumption. However if we generalize the logic of CRT to all of society, then random sampling is still ineffectual because of the pervasiveness of racism. These have profound implications for research given that regression and correlation assume independence. Previously published research may have Type-1 errors.
Methods: We use the 2020 Behavioral Risk Factor Surveillance System (BRFSS) public dataset to conduct multilevel modeling, which is an appropriate statistical method when observations are organized at more than one level. General health (1=excellent, 2=very good, 3=good, 4=fair, 5=poor) is the outcome variable. Race is the level-2 variable defined as: 1=White only, non-Hispanic, 2=Black only, non-Hispanic, 3=American Indian or Alaskan Native only, Non-Hispanic, 4=Asian only, non-Hispanic,5=Native Hawaiian or other Pacific Islander only, Non-Hispanic, 6=Other race only, non-Hispanic, 7=Multiracial, non-Hispanic, 8=Hispanic. The level-1 variable is income 1=less than $15,000, 2=$15,000 to less than $25,000, 3=$25,000 to less than $35,000, 4=$35,000 to less than $50,000, 5=$50,000 or more. We used full information maximum likelihood (FIML) for our estimation method. Don’t know/ Not sure/ Refused were excluded from analysis.
Results: 104731 observations were included in our analysis. At level 1, income (b=-.130, s.e.=.003, p<.001) was negative and a significant predictor of general health (Table 1).
At level 2, race was a significant predictor (b=.027, s.e.=.104, p=0.025). Since Wald Z test is a two-tailed test, the p-value is calculated by splitting the p-value in half. The ICC is .97 (1.23/1.25). This result indicates that income and race were significant predictors of general health.
Discussion: This brief research study supported CRT assumptions about the role of race in research analysis. It also raises concerns with regression or correlation analyses about Type-1 errors using BRFSS data. Further research is needed to explore multilevel modeling using race.
Limitations: This study has several limitations. There may be bias arising from the observations removed based on our exclusion criteria. Second, we only included one Level-1 variable. Future research should explore whether our results hold up with additional covariates.
Table 1: Level-1 Analysis
Level-2 Analysis
