Models Accounting for Country-Level Clustering

Consideration of country-level clustering reduced the observed associations between subnational gender inequality and immunization coverage outcomes. In the model additionally controlling for the average zero-dose prevalence or DTP3 coverage for the corresponding country-year, we find a significant association between gender inequality and both zero-dose DTP prevalence and DTP3 coverage. In the model limited to the most recent year of data available for each subnational region and clustering standard errors by country, we do not observe a significant association between gender inequality and immunization outcomes. In multilevel linear regression models accounting for nested random effects of subnational regions within country, we find significant associations between gender inequality and both zero-dose DTP prevalence and DTP3 coverage. Findings are similar when limited to the most recent year of data, utilizing a linear regression model with country random effects. To more directly compare findings between models, we present

predicted marginal effects of higher versus lower gender inequality, e.g., the predicted percentage point difference in coverage between subnational regions with higher gender inequality compared to those with lower gender inequality (see Table 4). We first present the adjusted model that does not account for country clustering, as well as the four models discussed above. Though the direction of association remains constant across models, the magnitude and strength of association is reduced for the models that take into account country-level clustering.

**Table 4.** Predicted marginal effects [percentage point difference] for zero-dose DTP prevalence and DTP3 immunization coverage by SGDI category (702 subnational regions across 57 countries, 2010–2019).


\* *p* < 0.01; \*\* *p* < 0.001.
