A Comparative Analysis on the Social Determinants of COVID-19 Vaccination Coverage in Fragile and Conflict Affected Settings and Non-fragile and Conflict Affected Settings

Background: The coronavirus disease 2019 (COVID-19) pandemic has coerced various resources of all the countries. While the high-income nations redirected financial and human resources to understand specific determinants of vaccination coverage, fragile and conflict-affected setting (FCS) nations were waiting for global bodies to cater to their ever-growing need for vaccines and other lifesaving drugs. This study aimed to determine various factors influencing vaccine coverage in the FCS context. Methods: World Bank’s classification of FCS states was the primary source for country classification. The study utilized data from various other open sources. The study models cross-country inequities in COVID-19 vaccine coverage and we have employed multi-variate log-linear regressions to understand the relationship between COVID-19 vaccine coverage and cross-country macro-level determinants. The analysis was conducted on two samples, non-FCS Countries and the FCS countries. Results: Socio-economic determinants such as gross domestic product (GDP) per capita, socioeconomic resilience; health system determinants such as density of human resources, government spending on health expenditure; and political determinants such as effective government, more power to regional governments, political stability and absence of violence play a pivotal role in vaccine coverage. We also found that FCS countries with a higher share of people strongly believing in the vaccine effectiveness have a positive association with COVID-19 vaccine coverage. Conclusion: The study confirmed that political factors, government effectiveness and political stability are also important determinants of vaccine coverage. The result further draws attention to few policy implications such as promoting future research to explore the linkages between the perceived equality before the law and individual liberty and its effect on vaccination coverage in the FCS.


VIII: Diagnostic tests for omitted variable bias, multi-collinearity, linearity and heteroscedasticity.
For each level of analysis, i.e., socio-economic determinants, political determinants, healthsystem determinants, we perform a series of diagnostic tests to check for endogeneity.

Model specification
We perform a link test to assess model specification. This test is based on the idea that if a regression is properly specified, one should not find additional independent variables that are significant (except by chance). This means that the dependent variable needs a transformation or "link" function to relate to the independent variables. Operationally, we add an independent variable to the equation that is likely to be significant if there is a specification error. For our aggregate analysis, we expect the link test to show misspecification as we only focus on social determinants of health. On one hand, our narrow focus allows us to clearly see specific relations between regressors and vaccine coverage, but on the other hand, leaves scope for further additions to our framework. Consequently, we find that our link test provides evidence of model misspecification. Specifically, when we regress our dependent variable on its predicted values and the square of predictions, we find the squared term to be statistically significant. This means that we can either change the link function (to a generalised model as by Zhu et al.,) or we change measurement of individual variables 11 .

Omitted Variable Bias
We perform Ramsey's (1969) regression specification error-test for omitted variables on each of our regressions under the null hypothesis that the model has no omitted variables. For the aggregate analysis, we conclude that there is scope for omitted variable bias which is what we expected based on the link test and Ramsey's test.

Collinearity
To detect the collinearity of the regressors with the constant, we compute variance inflation factors (VIF) for each of our regressors. Multi-collinearity is seen if, a) the largest VIF is greater than 10, and b) the mean of all VIFs is considerably larger than 1. For the aggregate analysis, we do not find evidence of multi-collinearity as we do not find any regressor to have variance inflation factor greater than 10 and the mean of all VIF's is around 3.02 which, albeit far from 1, can be accommodated in our analysis as insufficient evidence for multi-collinearity.

Heteroscedasticity
We employ White's test to check for heteroscedasticity under the null hypothesis that there is no heteroscedasticity with the test statistic having a chi-square distribution with K*(K+3)/2 degrees of freedom (K = number of independent parameters). For our aggregate analysis, we find that p-value for the chi-squared test-statistics to be 0.59 which is greater than 0.05. Consequently, we fail to reject the null hypothesis and can conclude that we have limited the extent of heteroscedasticity in our regression specification. To substantiate this, we also use Breusch-Pagan test for heteroscedasticity (by lifting the normality assumption of error terms) and find that p-value (p=0.046) for the test-statistic (Chi-squared = 19.95) is statistically insignificant at 1% level of significance. The null hypothesis in the Breusch-Pagan test is of constant variance among the independent variables. By failing to reject the null hypothesis, we again conclude that we account for heteroscedasticity in our regression model.