The Effect of Gender Quotas in Local Bodies on Women's Representation in Rural India

Using a novel dataset of over 67,000 rural local governance bodies (Gram Panchayats) spanning 20 years across Rajasthan and Uttar Pradesh, we find that randomly implemented gender quotas for women do not substantially increase the chances of women winning elections once the seat is unreserved. Even fifteen years of exposure to local women leaders has limited effects.

Authors: Varun, K. R. and Gaurav Sood

Quick Start

# 1. Clone and setup
git clone https://github.com/soodoku/quota_raj.git
cd quota_raj

# 2. Install R dependencies
R -e "renv::restore()"

# 3. Obtain external data (see Data Dependencies below)

# 4. Run full pipeline
Rscript scripts/99_run_all.R

# 5. Compile manuscript
cd ms && latexmk -xelatex main.tex

Pipeline and Crosswalk Architecture

Because administrative boundaries change frequently and spelling is highly irregular, the pipeline uses the Local Government Directory (LGD) as a "universal translator" to link election records to Census covariates at the lowest possible level: the Gram Panchayat (GP).

1. Election to LGD Block: Election samiti or block names are mapped to official LGD Block codes via manual crosswalks (data/crosswalks/active/).

2. Election GP to LGD GP: Within each matched block, GP names are matched in two passes:

   • Exact match: Normalized GP names are joined directly (handles ~40-50% of GPs)
   • Fuzzy match: Remaining unmatched GPs are matched using Jaro-Winkler distance < 0.20 (optimized for precision over recall)

   Deduplication resolves edge cases:

   • One-to-many (election GP → multiple LGD GPs): Keep lowest distance
   • Many-to-one (multiple election GPs → same LGD GP): Keep lowest distance within district/block

3. LGD to SHRUG: The LGD GP code is linked to the SHRUG Village ID (shrid2) using a pre-built crosswalk (data/shrug_gp_xwalk/).

4. SHRUG to Census: Covariates merged via shrid2 key.

The match_distance column is preserved in output files for robustness checks at stricter thresholds.

Directory Organization

data/
├── crosswalks/
│   ├── active/             # Active election-to-LGD mappings (Essential)
│   └── audit/              # Tie-resolution, unmatched, and diagnostic reports
├── lgd/
│   ├── raw/                # Original .xls and .csv from LGD portal
│   ├── processed/          # Cleaned LGD hierarchies
│   └── manual/             # SHRUG manual matches
└── shrug/                  # Raw Development Data Lab datasets

Data Dependencies

Included Data

• Rajasthan: GP election data 2005-2020 in data/raj/source/

External Data (must be obtained separately)

1. SHRUG (Development Data Lab)

Source: https://www.devdatalab.org/shrug_download/

Download and place in data/shrug/:

• shrug-pca01-csv - Population Census Abstracts 2001
• shrug-pca11-csv - Population Census Abstracts 2011
• shrug-vd01-csv - Village Directory 2001
• shrug-vd11-csv - Village Directory 2011
• shrug-ec05-csv - Economic Census 2005
• shrug-ec13-csv - Economic Census 2013
• shrug-secc-mord-rural-csv - SECC Rural
• shrug-pc-keys-csv - PC Keys for linking

2. SHRUG-LGD Crosswalk

Source: https://www.devdatalab.org/shrug_download/ (User Contributed Modules section)

Download "SHRUG GP to shrids" by Pratik Mahajan, extract to data/shrug_gp_xwalk/

3. LGD (Local Government Directory)

Source: https://lgdirectory.gov.in/

Download for Rajasthan and UP, place in data/lgd/raw/:

• Village-GP mapping files
• Block/Panchayat Samiti files

4. UP Election Data

Source: in-rolls/local_elections_up

Code Organization

Scripts are numbered sequentially by processing stage to guarantee the exact execution sequence.

scripts/
├── 00_*.R          # Stage 0: Configuration and utilities
├── 01_*.R          # Stage 1: Data extraction, standardization, crosswalks
├── 02_*.R          # Stage 2: Panel creation and recoding
├── 03_*.R          # Stage 3: SHRUG integration and audits
├── 04_*.R          # Stage 4: Descriptives, Balance & Validation
├── 05_*.R          # Stage 5: Main Analysis - Short Term
├── 06_*.R          # Stage 6: Main Analysis - Long Term
├── 07_*.R          # Stage 7: Extensions (Candidates, Replications)
├── 08_*.R          # Stage 8: Phone Surveys
└── 99_run_all.R    # Stage 99: Master pipeline orchestration

Notes on Analysis

Random Rotation Subsample (06c)

The random rotation analysis restricts to districts where chi-square tests fail to reject independence of quota assignment across transitions (p > 0.05).

10-year analysis (2005→2015): Requires independence for 05→10 AND 10→15

• Rajasthan: 9/32 districts (592 open seats)
• UP: 10/44 districts (1,772 open seats)
• Results reported in tabs/long_term_random_rotation.tex

15-year analysis (2005→2020/21): Requires independence for ALL THREE transitions

• Rajasthan: 3/32 districts (208 open seats in Churu, Jaisalmer, Sikar)
• UP: 0/44 districts pass

We do not report 15-year random rotation results because the sample is too small. With only 3 districts and 208 observations, the 7-coefficient interaction model cannot be reliably estimated, and any findings would likely reflect district-specific patterns rather than generalizable effects.

Outputs

Tables (tabs/): Balance tests, treatment effects, candidate characteristics (~25 LaTeX tables)

Figures (figs/): Coefficient plots, treatment persistence, maps

Manuscript (ms/main.pdf)

Requirements

• R 4.5+
• XeLaTeX (for manuscript compilation)

Related Repositories

• in-rolls/local_elections_up - UP Local Election Data
• in-rolls/quota - Effects of Reservations on Allocation and Development Outcomes
• in-rolls/local_elections_kerala - Kerala Local Government Data