This repository replicates the empirical analysis from Alix-Garcia, McIntosh, Sims, and Welch (2013) examining the environmental effects of Mexico’s Oportunidades conditional cash transfer program.
The original study investigates whether poverty alleviation programs influence deforestation outcomes in rural areas. Using quasi-experimental variation in program eligibility, the authors test whether transfers to poor households affect land use behavior and forest cover.
This project reproduces the main empirical results using the publicly available replication datasets and implements the analysis in R.
The goal of the project is to evaluate whether the empirical findings can be reproduced using modern statistical software and to better understand the econometric mechanisms underlying the results.
Does the Oportunidades poverty alleviation program influence deforestation outcomes in rural Mexico?
More broadly, the study investigates whether income transfers to poor households alter incentives for forest clearing and land use decisions.
The analysis evaluates whether municipalities that became eligible for the program experienced different deforestation outcomes compared with otherwise similar municipalities that did not receive transfers.
The replication uses datasets originally constructed for the published study.
The repository includes three datasets used in the analysis:
Deforestation dataset
Contains municipal-level measures of forest cover and deforestation outcomes.
Household dataset
Contains household-level information on consumption, demographics, and program participation used to examine potential mechanisms through which transfers affect land use decisions.
Distance dataset
Contains geographic information used to measure distance to treatment thresholds and support the identification strategy.
All datasets are stored in Stata .dta format and are imported directly into R for analysis.
The empirical strategy follows the identification framework used in the original paper.
The analysis begins by importing the replication datasets and constructing the variables used in the empirical models.
This includes:
- treatment eligibility indicators
- geographic distance measures
- deforestation outcomes
- household consumption variables
Summary statistics and descriptive comparisons are constructed for treatment and control groups.
The core empirical specifications estimate how program eligibility influences deforestation outcomes.
The baseline model takes the form:
where:
- ProgramEligibility identifies municipalities eligible for Oportunidades
- X represents control variables included in the original specification
The analysis exploits geographic thresholds used in program targeting.
Distance to the eligibility cutoff is used to compare locations just inside and just outside the program eligibility boundary, providing quasi-experimental variation in treatment assignment.
Household-level regressions examine whether transfers affect economic behavior through channels such as:
- consumption patterns
- agricultural investment
- land use decisions
These results help explain potential mechanisms linking poverty alleviation to environmental outcomes.
The replication successfully reproduces the main findings of Alix-Garcia et al. (2013) with a high degree of accuracy.
Summary statistics show that localities eligible for the Oportunidades program experienced higher rates of deforestation than otherwise similar non-eligible localities near the marginality index cutoff. In the restricted sample of communities near the eligibility threshold, the share of polygons experiencing deforestation increases from approximately 4.9% in non-eligible localities to about 7.9% in eligible localities.
Initial regressions using eligibility as a proxy for treatment reveal a positive and statistically significant relationship between program eligibility and forest loss. Across Tobit specifications, the eligibility coefficient is consistently positive and significant at conventional levels, indicating a discontinuous increase in deforestation at the program eligibility threshold.
The causal estimates using the fuzzy regression discontinuity and instrumental variable framework confirm this result. Instrumental variable Tobit estimates show that program participation significantly increases the probability and intensity of deforestation. The marginal effects imply that treatment raises the probability that a locality experiences any deforestation by approximately 1.8 to 3.8 percentage points relative to the baseline probability of roughly 4.9%, implying nearly a doubling of deforestation risk for localities just above the eligibility threshold.
Among localities that do experience forest loss, the program also increases the share of land deforested by roughly 15–33% relative to baseline levels. These results are robust across specifications, including both Tobit and IV-OLS estimators, as well as across full and restricted samples around the marginality cutoff.
Household-level analysis using the randomized Progresa pilot data provides evidence on the mechanisms behind these effects. The program significantly increases consumption of land-intensive foods. Treated households consume beef approximately 0.11–0.12 more days per week relative to a baseline of 0.39 days, and drink milk about 0.33 additional days relative to a baseline of 1.44 days. These increases suggest that higher household income leads to greater demand for livestock products, which are associated with land-intensive production and can contribute to local forest clearing.
Taken together, the results support the central conclusion of the original study: poverty alleviation through conditional cash transfers can unintentionally increase deforestation when higher incomes raise demand for land-intensive goods in rural economies with limited market integration.
poverty-deforestation-rdd-replication
│
├── README.md
├── LICENSE
├── .gitignore
│
├── code
│ └── poverty_deforestation_replication.Rmd
│
└── data
└── raw
├── ecological_footprint_restat_deforestation_data.dta
├── ecological_footprint_household_analysis.dta
└── ecological_footprint_distance_data.dta
The entire analysis is implemented in an R Markdown workflow.
Required packages include:
- tidyverse
- haven
- fixest
- ggplot2
- modelsummary
- dplyr
Install packages in R if necessary, then run:
rmarkdown::render("code/poverty_deforestation_replication.Rmd")
All regression results, tables, and figures will be generated within the report.
Replication studies depend on the availability and completeness of the original datasets. While the empirical framework follows the published methodology, minor differences in software implementation may lead to small differences in estimated results.
The objective of this project is to reproduce the analytical framework rather than extend the original model.
This repository is intended for research and educational purposes only.