Fig. 2.

Impact of female leadership on COVID-19 deaths and cases — Dynamic local DID estimates.

Notes: This figure displays four graphs reporting the effects of electing a female mayor in the 2016 election across the year-quarters between 2017.1–2020.4 and their 90% confidence interval. Treatment effects are estimated using quarterly data from 2017.1 to 2020.4. The local DID regression model has the form $y_{mst}={\sum }_{k\ne 2020.1}{\beta }_k\cdot FemaleMayo{r}_{ms}+f_t\left(FemaleVoteMargi{n}_{ms}\right)+{\theta }_{ms}+{\lambda }_{st}+{ϵ}_{mst}$ where ${\theta }_{ms}$ captures municipality fixed-effects and ${\lambda }_{st}$ captures state-year-quarter fixed effects. $FemaleMayo{r}_{ms}=\mathbb{1}\left(FemaleVoteMargi{n}_{ms}\right)$ is an indicator variable equal to one when the municipality $m$ in state $s$ selected a woman as a mayor in 2016. To mirror our baseline RD specification in a dynamic setting, we control for $f_t\left(FemaleVoteMargi{n}_{ms}\right)=f_t\left(FemaleVoteMargi{n}_{ms}\right)\cdot \mathbb{1}(FemaleVoteMargi{n}_{ms}>0)+f_t\left(FemaleVoteMargi{n}_{ms}\right)\cdot \mathbb{1}(FemaleVoteMargi{n}_{ms}<0)$, a year-quarter specific polynomial in the vote-share of female candidates with parameters that vary flexibly for municipalities that elected a man and a woman as a mayor in 2016. Treatment effects are normalized concerning the first quarter of 2020, the last quarter before the COVID-19 outbreak. Each local DID specification uses the sample of the RD baseline estimates with the same outcome and RD polynomial that is reported in Table 1. We cluster standard errors at the municipality level.