Fig. 4: Effect of pooled treatment on knowledge and attitudes among adults enumerated during census.

Intent-to-treat estimates of community treatment assignment for each outcome listed on the y axis. Treatment effects were estimated using OLS and included randomization block fixed effects and heteroscedasticity-robust standard errors clustered at the village level. Each dot is labelled with the exact coefficient (to three decimal places) and significance at the ***1,**5 and *10 per cent critical level. Bars represent 95% CIs of treatment estimates. The analysis includes 45 villages and 817 households surveyed at endline for which we observed complete randomization blocks. Associated regression results are provided in Extended Data Tables 6 and 7, including corresponding sample sizes. Reported estimates do not correct for multiple hypothesis testing. The Extended Data tables report the associated FDR-adjusted q values. The survey measures for the ‘Believes COVID-19 is real’ comes from the survey question: “Do you believe that COVID-19 exists in the world?” (yes or no). ‘Knows about the COVID-19 vaccine’ comes from the survey question: “Do you know about the COVID-19 vaccine/marklate?” (yes or no). ‘Vaccines are effective’ is 1 if respondents completely agree with the statement: “Vaccines are effective.” ‘Vaccines are safe’ is 1 if respondents completely agree with the statement “How much do you agree with this statement: vaccines are safe.” Trust in sources of information are from a multiple-select question: “Who do you most trust getting information about COVID-19?” (community health clinic (CHC), MoHS, media (news, TV), social media (Facebook, among others), family and friends, among others).