Research

My research asks how voters respond when climate shocks and energy infrastructure become politically visible, and what that means for accountability. I work with original geospatial data and causal inference designs, mostly in R.

Working papers

Does Democracy Die in Darkness? Storm-Driven Blackouts, Turnout, and Vote Choice in São Paulo

Working paper

  • Question. Disasters reach voters less through wind and water than through the services that fail afterwards. When a windstorm cut power to millions of São Paulo residents between the two rounds of the 2024 mayoral election, did the blackout change how they voted?
  • Data. An original outage dataset for the city’s 104 electricity distribution areas, linked to geocoded election returns from 2008 to 2024.
  • Design. Difference-in-differences with a continuous treatment and spatially robust inference; a two-way fixed-effects panel of mayoral runoffs; wind speed as an instrument for outage hours; rainfall and outages outside the storm window as placebos; randomization inference and entropy balancing.
  • Finding. The incumbent’s gain between rounds was about two points smaller in the hardest-hit quarter of the city. Turnout fell citywide, but local outages did not predict where.

Presented at ISA, MPSA, and the ISA–AMEI joint conference.

Watt Drives Voters: The Politics of AI Data Center Energy Demand in Georgia’s Public Service Commission Election

Working paper

  • Question. Georgia is one of ten U.S. states that elect their electricity regulators. As AI data centers drive up electricity demand, do voters hold those regulators accountable?
  • Data. A county-level panel of data-center openings linked to election returns and survey data.
  • Design. Synthetic control, synthetic difference-in-differences, and spatial spillover models.

Decoupling Without Democracy: Does Economic Openness Drive Green Transitions in Authoritarian States?

Working paper

  • Question. Can trade openness push autocracies to decouple growth from emissions, absent electoral pressure?
  • Data. A panel of 165 countries, 2001–2023.
  • Design. Staggered difference-in-differences, synthetic control, and machine-learning screening of mechanisms.

Work in progress

Disasters, blackouts, and voting in Japan. Extending the São Paulo design to prolonged typhoon and earthquake blackouts in Japan, to ask whether voters blame utilities, local governments, or the governing parties.

Methods

Difference-in-differences and event studies · synthetic control and synthetic DiD · instrumental variables · randomization inference · spatial econometrics and areal interpolation · statistical machine learning · expert surveys · R (fixest, sf, spdep, tidyverse)