2.4. Analysis

Descriptive statistics, including means (M), standard deviations (S.D.), skewnesses, kurtoses, and frequencies, were computed for all dependent and independent variables. Unadjusted linear regressions were used to identify significant background variables associated with attitudes, subjective norms, and perceived behavioral control. When the categories of the independent variables were very small (<20 observations), we combined them to form larger categories to reduce the risk of over‐fitting the model. The results of unadjusted linear regressions are available in Appendix S2. Stata (version 17) ⁴⁰ was used to run descriptive statistics and unadjusted linear regressions.

Structural equation modeling was conducted in Mplus 8.8. ⁴¹ Model building was sequential based on the hypotheses from regression models to the full path model. Four models were estimated to test our hypotheses. Model 1 examined the direct relationship between caregivers' vaccination intentions and attitudes, subjective norms, and perceived behavioral control. Model 2 tested the indirect relationships between attitudes and intention, and subjective norms and intention, through the moderation of perceived behavioral control. Model 3 tested the full TPB model. Model 4 tested all the previously identified significant paths and incorporated relevant background variables. All models used full information maximum likelihood assuming data missing at random. Due to the inclusion of noncontinuous endogenous variables, the weighted least squares mean and variance adjusted (WLSMV) estimator was used. ⁴¹ Model fit was assessed using Root Mean Square Error of Approximation (RMSEA), Comparative Fit Index (CFI), Tucker‐Lewis Index (TLI), Standardized Root Mean Square Residual (SRMR), and chi‐square. ⁴² Statistical significance for all analyses was set at p < 0.05, using a two‐sided test. The data analyzed for this study are available upon request.