Fig 3. Treatment effects of the most effective reminders by the health status of the recipient.

Notes: The effects are percentage changes relative to the probability of staying home in the control group, which receives no reminder (i.e., the regression coefficients in Table 3 are divided by the probability of staying home in the control group). Among respondents who are in bad health conditions (N = 603), the share of those who declare they will stay home more than doubles after receiving a reminder that emphasises risks for family (p = 0.036), and the share of those who actually stay home increases by 80% (p = 0.034). Similar impacts of the “you” treatment are not statistically significant. Among respondents who are in good health (N = 4,704), the impacts of the reminder are much smaller and not statistically significant. It is important to note that while impacts of over 30% may appear sizeable, they a relative to a low compliance level in the control group (where less than 15% of respondents stay home). Respondents classify their health status on a 5-point scale. The lowest two values are considered bad health conditions for the purpose of this analysis (health information is missing for 3 observations used in the main regressions). Stars reported at the top of the bars express the level of significance of the coefficient (*** p<0.01, ** p<0.05, * p<0.1).