COVID-19 vaccine hesitancy cannot fully explain disparities in vaccination coverage across the contiguous United States

Abstract

Vaccine hesitancy has been identified as a major obstacle preventing comprehensive coverage against the COVID-19 pandemic. However, few studies have analyzed the association between ex-ante vaccine hesitancy and ex-post vaccination coverage. This study leveraged one-year county-level data across the contiguous United States to examine whether the prospective vaccine hesitancy eventually translates into differential vaccination rates, and whether vaccine hesitancy can explain socioeconomic, racial, and partisan disparities in vaccine uptake. A set of structural equation modeling was fitted with vaccine hesitancy and vaccination rate as endogenous variables, controlling for various potential confounders. The results demonstrated a significant negative link between vaccine hesitancy and vaccination rate, with the difference between the two continuously widening over time. Counties with higher socioeconomic statuses, more Asian and Hispanic populations, more elderly residents, greater health insurance coverage, and more Democrats presented lower vaccine hesitancy and higher vaccination rates. However, underlying determinants of vaccination coverage and vaccine hesitancy were divergent regarding their different associations with exogenous variables. Mediation analysis further demonstrated that indirect effects from exogenous variables to vaccination coverage via vaccine hesitancy only partially explained corresponding total effects, challenging the popular narrative that portrays vaccine hesitancy as a root cause of disparities in vaccination. Our study highlights the need of well-funded, targeted, and ongoing initiatives to reduce persisting vaccination inequities.

Keywords: COVID-19; Disparity; Social vulnerability; Structural equation modeling; Vaccination coverage; Vaccine hesitancy.

Fig. 1

Conceptual Diagrams for mediation analysis.${\mathit{c}}_{\mathit{E}\mathit{M}}$ means the variance of the error term of mediators; ${\mathit{c}}_{\mathit{E}\mathit{O}}$ means the variance of the error term of outcomes; ${\mathit{c}}_{\mathit{r}\mathit{g}},{\mathit{c}}_{\mathit{i}\mathit{t}},$${\mathit{c}}_{\mathit{s}\mathit{e}},$${\mathit{c}}_{\mathit{h}},$${\mathit{c}}_{\mathit{h}\mathit{t}},$${\mathit{c}}_{\mathit{d}},$${\mathit{c}}_{\mathit{p}},{\mathit{c}}_{\mathit{s}\mathit{f}}$ mean the variances of exogenous variables; ${\mathit{c}}_{\mathit{E}\mathit{M}\mathit{E}\mathit{O}}$ refers to the covariance matrix between error terms of mediator and outcome.

Fig. 2

Temporal evolution of coefficients of univariable regressions (a) and scatter plot of vaccine hesitancy versus vaccination rate. Panel (a) covers data from 2021-01-01 to 2022-01-27 on a weekly average basis. The error bar depicts the robust 95 % CI. Panel (b) was plotted based on data during January 21–27, 2022. Each spot represents one county in the contiguous US.

Fig. A1

Temporal evolution of fully vaccinated rate (%) across counties stratified by vaccine hesitancy. Sample comprises 2,657 contiguous US counties. Texas, Colorado, and Virginia are excluded because C.D.C. vaccination data (county level) were missing before November 2021.

Fig. A2

Spatial distribution of county-level (a) vaccine hesitancy and (b) vaccination rate. Vaccination rate was plotted based on C.D.C. vaccination data during January 21–27, 2022.

Fig. A3

Temporal evolution of difference in mean of vaccination rate between vaccine hesitancy quintiles (Q1 as reference). The error bar depicts the robust 95 % CI of the mean difference. Gray error bars denote those reporting P-values greater than 0.05 in T-tests (i.e. not significant).