1 of 10

IDEAS 2026

Paper presentation

8–11 November 2026 · Florianópolis, Santa Catarina, Brazil

Women in Local Government Leadership during the COVID-19 Pandemic in Brazil –

The Female Leadership Paradox

​

Ivan Filipe Fernandes

UFABC, Brazil

Gustavo Andrey Fernandes, FGV/SP

2 of 10

IDEAS 2026

Problem and aim

8–11 November 2026 · Florianópolis, Santa Catarina, Brazil

Fernandes· Paper 47

The Female Leadership Paradox

Female-led municipalities recorded lower COVID-19 mortality in 2020.

�The same municipalities had lower social-distancing compliance in every period.

�They adopted no stricter NPI packages. For several coercive instruments, the point estimates run negative.

�

3 of 10

IDEAS 2026

Foundations

8–11 November 2026 · Florianópolis, Santa Catarina, Brazil

Decentralized by Design

  • SUS is highly decentralized

municipalities deliver basic health services (Massuda et al., 2018).

  • Subnational authority affirmed

Supreme Court upheld state and municipal competence to adopt preventive measures.

  • Ideology did not organize the response

partisan cleavage does not systematically explain local COVID-19 action (Fernandes et al., 2025).

  • Electoral calendar matters

November 2020 municipal elections concentrated incentives in the second semester.

  • Political exposure is asymmetric

mobility restrictions contested (Fernandes et al., 2023); testing, screening and disinfection not comparably so (Ajzenman et al., 2023).

�

�5,570

�Brazilian municipalities, each with constitutional authority over basic health delivery

�Nov 2020

�municipal elections, concentrating electoral pressure in the second half of the first pandemic year

�3

�

 

 

Fernandes· Paper 47

4 of 10

�5,568

�municipalities held

elections in 2016

�

IDEAS 2026

Methods

8–11 November 2026 · Florianópolis, Santa Catarina, Brazil

Close-election regression discontinuity

Design essentials

​

  • Timing   mayors elected in 2016 governed throughout 2020;
  • Running variable   margin of victory of the female candidate over the male candidate; threshold at zero.
  • Estimator   local linear polynomials with bias-corrected robust inference and a local quadratic correction (Calonico, Cattaneo and Titiunik, 2014; rdrobust, Calonico et al., 2017).
  • Specification   MSE-optimal bandwidth and state fixed effects on every column.
  • Data   Brasil.io (cases, deaths), InLoco (mobility), CNM and IBGE-MUNIC (NPIs and sanitary measures), FINBRA (health expenditure).

�

�601

�with margin of victory

below 3 percentage points

�

�145

�male vs female contests

83 female | 62 male mayors

�

 

 

Fernandes· Paper 47

5 of 10

5

Three facts that compose the paradox

01 Mortality advantage

−0.487

deaths / 1,000, 2nd semester 2020

Unconditional second-semester coefficient. First-semester estimate is statistically indistinguishable from no effect.

02 Compliance deficit

lower

social-isolation index in every period

Deficit largest in the first semester, persists throughout the year. Direction is the opposite of what a behavioral account predicts.

03 No stricter NPIs

null or negative

CNM and IRT stringency indices

Decree, sanctions, movement restrictions and gatherings all negative. Sanctions carry the largest MUNIC coefficient.

IDEAS 2026

Fernandes & Fernandes - Female Leadership Paradox - IDEAS 2026

6 of 10

6

A selective reconfiguration of the response

01

SANITARY & TERRITORIAL

neighborhood disinfection and distribution of cleaning kits are positive and precisely estimated. Informational pamphlets positive. Sanitary barriers positive, small coefficient.

02

STATE-LED DETECTION

applied clinical testing and health screening or triage positive and significant. These capacities do not require citizens to change behavior to work.

03

SOCIAL PROTECTION

food baskets or food credit, economic benefits, and social assistance through CRAS. Among the largest first-stage gender coefficients in the paper.

What does not fit

Mask distribution negative and significant.

Hygiene-kit distribution statistically indistinguishable from zero.

Mandatory masks no gender difference.

Health-budget share, 2020 positive only at the 10% level, suggestive rather than established.

IDEAS 2026

Fernandes & Fernandes - Female Leadership Paradox - IDEAS 2026

7 of 10

7

Isolation is a suppressor, not a mediator

UNCONDITIONAL

−0.487

deaths per 1,000 inhabitants, second-semester 2020

SUPPRESSION-ADJUSTED

−1.245

deaths per 1,000, controlling for second-semester social isolation (SE 0.131, p<0.001)

Including isolation raises the coefficient in absolute magnitude by ~2.5×

A covariate that enlarges the treatment coefficient is offsetting it, not transmitting it. Lower compliance in female-led municipalities worked against the mortality advantage and concealed part of its magnitude. Every subsequent mechanism specification holds isolation constant.

IDEAS 2026

Fernandes & Fernandes - Female Leadership Paradox - IDEAS 2026

8 of 10

8

Which policies absorb the mortality effect

Baseline: suppression-adjusted female-mayor coefficient = −1.245 / 1,000 (holding second-semester isolation constant)

Mediator added to baseline

Adjusted coefficient

Attenuation

Verdict

Health screening (triagem)

−0.786

~37%

strong candidate channel

Clinical testing (realizou_testes)

−0.787

~37%

strong candidate channel

Neighborhood disinfection

−0.996

20.0%

moderate attenuation

CRAS (social assistance)

​

10.8%

modest

Health-budget share, 2020

​

9.2%

modest, suggestive

Sanitary barriers

​

0%

no attenuation

Food baskets (1.133, SE 0.190)

​

0%

no attenuation despite large adoption gap

Mandatory masks / school food / donation / economic benefits

​

0%

no attenuation

Cumulative: disinfection + testing + screening absorb 64.7% of the suppression-adjusted baseline, with the coefficient at −0.439 (p<0.001).

IDEAS 2026

Fernandes & Fernandes - Female Leadership Paradox - IDEAS 2026

9 of 10

IDEAS 2026

Conclusion

8–11 November 2026 · Florianópolis, Santa Catarina, Brazil

07 / 08 · 1 min 30 sec

Composition, not stringency

  • Mechanism: effect operates through what the local government builds and delivers, not through policy stringency or compliance.

​

  • Behavioral channel: does not account for the advantage; its compliance component works against it.

​

  • Detection and prevention: state-led sanitary and diagnostic capacity emerges as the strongest candidate transmission channel.

​

  • Social protection: a substantial and strongly gendered feature of the response, analytically separate from mortality.

�

10 of 10

IDEAS 2026

Closing

8–11 November 2026 · Florianópolis, Santa Catarina, Brazil

08 / 08 · 30 seconds

Female mayors saved lives not by restricting citizen behavior, but by proactively expanding state diagnostic and sanitary capacit

Thank you

Ivan Filipe Fernandes· UFABC