Research & Working Papers

Empirical work at the intersection of climate science, political economy, development, and public health.

Current Projects

All papers combine large-scale panel datasets, causal inference methods, and policy-relevant empirical designs.

2025
Ongoing

Climate Extremes and Economic Stressors as Predictors of Conflict in Central America (2012–2024)

20-country monthly panel integrating ERA5 climate reanalysis, SPEI drought indices, MLEED, ACLED, and FAO food-price series. Applies ElasticNet regularization and a mediation framework to decompose climate-to-conflict pathways. With Erik Wibbels (UPenn).

ERA5 SPEI MLEED ACLED FAO Food Prices EM-DAT
Working Paper
2025–
2026

Machine Learning for Peace (MLP) — Event Data Infrastructure

Contributed to a project scraping and processing over 120 million news articles from local outlets across 60+ developing countries to generate high-frequency event data on political and environmental dynamics. Contribution focused on web scraping and count verification to ensure data integrity. With Erik Wibbels (UPenn).

MLP MLEED
MLP Website ↗
Project
2025

Climate Change, Agriculture, and Migration in Central America

Examines how El Niño and La Niña cycles affect agricultural productivity and food prices across the Dry Corridor, and how resulting climate shocks and environmentally motivated violence shape migration flows toward the U.S. border. Contribution: descriptive figures on temporal and geographic variation in environmental events. With Erik Wibbels, Irina Marinov, Anna Cabre & Jeremy Springman (UPenn).

ENSO/El Niño ERA5 Climate Security Atlas
Project
2025

The Credibility Revolution in Political Science

Assisted with data collection and validation for a study on replication practices and open science norms, including coding replication file policies across 175 journals and helping validate the analysis code. With Guy Grossman (UPenn) and collaborators.

View on SocArXiv ↗
Project
2025
Ongoing

School- and Community-Based Intervention Against Sexual Violence in Liberia

Cluster-randomized evaluation across 150 school-community dyads (2,160 girls, 1,187 caregivers). Conducted high-frequency data quality checks (HFCs) for baseline and endline surveys; monitored safeguarding trigger variables ensuring 26 sensitive disclosures were escalated under the differentiated referral protocol; identified instrument assignment errors across 8 schools corrected prior to endline. With Guy Grossman (UPenn) & IPA/WHI.

HFC IPA
Field Project

Key Datasets Explained

A guide to the primary data infrastructure underlying my research projects.

MLP — Machine Learning for Peace

A UPenn DevLab project that scrapes and processes news from 120M+ articles across 60+ developing countries using NLP pipelines to generate high-frequency, structured event data on political violence, protest, and environmental dynamics — filling gaps left by Western-centric data sources.

MLEED — Machine Learning Event & Entity Dataset

The structured output of the MLP pipeline — a high-frequency political event dataset coded from local-language news, covering protest, violence, and governance events across the Global South at sub-national resolution.

ACLED — Armed Conflict Location & Event Data

A globally recognized real-time dataset tracking political violence and protest events worldwide, coded from media reports and local sources. Provides precise geo-coded event data at the incident level, used here to validate and complement MLEED for Central America.

ERA5 — ECMWF Climate Reanalysis

The European Centre for Medium-Range Weather Forecasts' global climate reanalysis product, providing hourly data at ~31km resolution on temperature, precipitation, wind, humidity, and 200+ variables back to 1940. Used to extract climate shocks at the administrative-unit level.

SPEI — Standardized Precipitation-Evapotranspiration Index

A multi-scale drought index that combines precipitation and temperature into a single drought severity measure, allowing comparison across regions and time scales (1–48 months). Captures both drought onset and recovery, making it ideal for linking climate stress to conflict and food security.

EM-DAT — International Disaster Database

Maintained by the Centre for Research on the Epidemiology of Disasters (CRED) at UCLouvain, EM-DAT records natural and technological disasters globally since 1900, including deaths, affected populations, and economic damages — used here to capture extreme weather events beyond standard climate indices.

Methods & Data

Econometric Methods
  • Panel Fixed Effects
  • Distributed-Lag Models
  • Mediation Analysis
  • Interaction & Threshold Frameworks
  • Causal Inference (Quasi-experimental)
  • ElasticNet Regularization
Geospatial & ML
  • Automated Geospatial Extraction
  • Satellite & Remote Sensing Data
  • NLP Entity Extraction (GPT-4, Gemini)
  • Machine Learning for Event Data
  • QGIS & Python Geospatial Workflows

Interested in Collaborating?

I welcome inquiries about joint research projects and data partnerships.