A latent measure of cultural racism and its association with US mortality and life expectancy

Abstract

The US social context is characterized by systemic racism, a key driver of racial disparities. While structural and interpersonal racism are well-established components of systemic racism, cultural racism-a system of beliefs (ideology) and societal way of life (culture)-has been less robustly described and currently lacks a validated measure. We conducted confirmatory factor analysis on nine key indicators to define a cultural racism latent measure. In an analysis of US mortality data between 2018 and 2021 obtained from the Centers for Disease Control and Prevention's WONDER system, each unit increase in the cultural racism factor was associated with ~136 (95% confidence interval, 90 to 182) all-cause deaths per 100,000 and a one-year decline in life expectancy (~-1 (-2 to -1)); these associations were consistent for both Black and white adults. The cultural racism factor substantially advances the science of racism and health and provides an empirical basis for efforts to address US health inequities.

Fig. 1. Conceptual overview of mechanisms linking SR and health outcomes.

The dashed arrows indicate mutually reinforcing dynamics between component dimensions of SR. HPA, hypothalamic–pituitary–adrenal; CNS, central nervous system.

Fig. 2. Final one-factor CFA model.

The final model accounted for correlated errors between (1) online racial animus and lack of white support for Medicaid expansion and (2) the number of Confederate symbols per 100,000 and racial resentment. CFA loadings are indicated by solid arrows from the latent construct to each indicator; the standardized covariance of correlated errors is indicated with dashed arrows between indicators with correlated errors. The sample size is 51 (that is, all 50 US states and DC).

Fig. 3. Distribution of CRF scores for each US state and DC.

The top panel shows the geographical distribution of the CRF, while the bar chart in the lower panel shows the value for each state and DC.

Fig. 4. Distribution of state-level CRF values by history of Jim Crow laws and history of Confederacy.

a, The CRF and history of Jim Crow laws. b, The CRF and history of Confederacy. The CRF value for each state is indicated using state abbreviation labels. In each box plot, the centre line indicates the median, the centre solid dot indicates the mean, the box limits indicate the upper and lower quartiles, and the whiskers indicate values within 1.5 times the interquartile range. The sample size is 51.

Fig. 5. Interaction effects for the CRF and cancer mortality by region and for the CRF and diabetes mortality by race.

a, The CRF and cancer mortality by region. b, The CRF and diabetes mortality by race. The results are from random-effects linear regression models for cause-specific deaths per 100,000 including interaction terms and adjusted for demographic factors (total population size, study year and geographic region). The sample size is 204 (that is, four observations for each of 50 US states and DC). The double segment on the y axis in a indicates truncation of values between zero and the next value on the y axis, and the y-axis ranges differ between a and b. The lines in the plots represent estimated β coefficients, while the shaded error bands represent 95% CIs around the estimated β coefficients, indicating the range within which the true values are expected to fall with 95% certainty. The width of the bands reflects the uncertainty in the estimates, with narrower bands signifying greater precision. Fully reported statistical results for the interaction models on which a and b are based are presented in Supplementary Table 6.

Extended Data Fig. 1. Test-retest reliability for CRF.

CRF estimated using data for 2017 (y-axis) is strongly correlated with CRF using data for 2004 (x-axis). Fit statistics for confirmatory factor analysis model from which CRF for 2004 was extracted were: Chisq=1.16, df=1, p=0.28, CFI=0.99, TLI=0.97, RMSEA=0.06. This model contained only four indicators (showing factor loading and p-value; anti-immigrant sentiments (0.87, p0.001), implicit bias (0.46, p=2.5e-03), racial resentment (0.60, p0.001), and number of public symbols of the confederacy per 100,000 (0.49, p=2.5e03); correlation between racial resentment and number of public symbols of the confederacy was specified (covariance and p-value; -0.18, 0.37). Each solid dot indicates value for one of 50 states and DC and is labeled with state abbreviations. Statistical test for P: Pearson product moment correlation coefficient, df=49, two-tailed.

Extended Data Fig. 2. Correlation between CRF and anti-Black cultural racism measure from Price et al (2002).

CRF was positively and strongly associated with the only other metric of anti-Black cultural racism (r (90% CI; p) =0.66 (0.5, 0.77; p0.001)), providing evidence of convergent validity. Each solid dot indicates value for one of 50 states and DC and is labeled with state abbreviations. Statistical test for P: Pearson product moment correlation coefficient, df = 49, two-tailed.

Extended Data Fig. 3. Discriminant validity for CRF.

Showing Pearson product moment correlation coefficient between CRF and in panel A) Racialized economic segregation (ICERACEINC; -0.5 (-0.66, -0.31), df=49, p0.001; Krieger et al 2016), B) Standardized structural racism factor (SSRF; -0.47 (-0.66, -0.23), df=35, p= 3.1e-03; Brown and Homan 2024), and C) State racism index (SRI; 0.15 (-0.09, 0.37), df=48, p=0.29; Alvarez 2023). Each solid dot indicates value for one of 50 states and DC. Discriminant validity was assessed by comparing these correlations to the square root of the average variance effect (AVE=0.45, 0.38, 0.54) for the CRF obtained from the CFA model, which is 0.67. All three correlations are less than 0.67, which suggests that the CRF attains discriminant validity of concept in relation to structural racism. Very small p-values are reported in scientific notation to preserve precision. Statistical test for P: Pearson product moment correlation coefficient, two-tailed.

Extended Data Fig. 4. Relationship between state-level cultural racism factor and predicted age-adjusted deaths.

All-cause deaths (panel A), cancer-specific deaths (panel B), CVD-specific deaths (panel C), diabetes-specific deaths (panel D) per 100,000, and life expectancy at birth (panel E). Graphs are based on coefficient estimates from random effects linear regression model adjusted for demographic factors (total population size, study year, and geographic region). Double segments on y-axis indicate truncation between zero and the next value indicated on y-axis in the plot, and y-axis ranges differ across all five panels. Lines in the plots represent estimated beta coefficients, while shaded error bands represent 95% confidence intervals around the estimated beta coefficients, indicating the range within which the true values are expected to fall with 95% certainty. The width of the bands reflects the uncertainty in the estimates, with narrower bands signifying greater precision. No adjustments were made for multiple comparisons.