Indicators for SIT and SI assignment were included in all models; other covariates were retained if they were significant at the .01 level. Variables that were not included in the stepwise modeling then were tested one at a time for inclusion in the model; these additional covariates were retained if they were significant at the .10 level or if they had a strong impact on the coefficients of other variables in the model. Because of the potential importance of targeting interventions to specific groups, interactions between race or ethnicity and other covariates, as well as those between age and other covariates, were tested. A significant (p < .05) interaction was found between being White and length of time homeless; consequently, White persons were divided into newly (less than 1 year in their lifetime) and chronically homeless subgroups (1 year or greater). Multicollinearity was assessed, and model fit was examined using the Hosmer–Lemeshow test. Finally, recruitment site was used as a random effect in a mixed-model logistic regression analysis for vaccine completion. Statistical analyses were performed with SAS/STAT (SAS Institute, Cary, NC) and Stata (Stata Corporation, College Station, TX).