Replication Data for: Procurement Certainty, Coordination Amplification, and Acreage Lock-in: Graduated Intensity Identification and Panel Quantile Evidence from Maharashtra, India

Description

This replication package contains the data and R code required to reproduce all tables and results reported in the manuscript "Procurement Certainty, Coordination Amplification, and Acreage Lock-in: Graduated Intensity Identification and Panel Quantile Evidence from Maharashtra, India" (submitted to Journal of Quantitative Economics). The package contains five files: 1. RODE_Analysis_Panel_238obs.csv Analysis dataset. Balanced panel of 238 district-year observations covering 34 agricultural districts of Maharashtra across seven Kharif seasons (2016-17 to 2022-23). RODE_01_Codebook.R. 2. RODE_01_Codebook.R Documents all variables including identifiers, acreage share dependent variables, the continuous cooperative mill density variable (log_mills), rainfall SPI-3 instrument, structural controls, PMFBY state-level control, and AgriStack DiD indicators. 3. RODE_02_Build_Dataset.R Data construction script. Builds RODE_Analysis_Panel_238obs.csv from six primary Indian government administrative data sources: (i) Directorate of Economics and Statistics, MoAFW (data.desagri.gov.in); (ii) India Meteorological Department district daily rainfall series; (iii) Maharashtra Cooperative Sugar Factories Federation official factory registry (mahasugarfed.org); (iv) PMFBY Administrative Dashboard (pmfby.gov.in); (v) Ministry of Agriculture and Farmers Welfare Digital Agriculture Mission rollout records (agristack.gov.in); (vi) Companion district panel constructed from Maharashtra Season and Crop Reports and ICRISAT district-level database. 4. RODE_03_Estimation.R Estimation script. Reproduces all regression results reported in Tables 3 through 6 of the manuscript: panel fixed effects with continuous mill density (dose-response), coordination amplification interaction specification, panel quantile regression using the Canay (2011) two-step estimator, two-way fixed effects DiD with pre-trend tests, Generalized Synthetic Control estimation (Xu, 2017) via the gsynth package, and robustness specifications. Requires R packages: plm, lmtest, sandwich, quantreg, gsynth, tidyverse. 5. RODE_04_README.txt Documentation. Describes data sources, sample construction, variable definitions, missing value handling, AgriStack cohort assignments, estimation steps, and known limitations. Required reading before running any script. All underlying crop acreage and production data are publicly accessible through Indian government portals as documented in RODE_02_Build_Dataset.R and RODE_04_README.txt. Software: R version 4.4.3 (aarch64-apple-darwin20). Key package versions: plm 2.6-3, quantreg 5.97, gsynth 1.2.1. Prakash Gurumurti Vhankade, Associate Professor, Gokhale Institute of Politics and Economics, Pune, prakash.vhankade@gipe.ac.in

Steps to reproduce

SYSTEM REQUIREMENTS ------------------- Software: R version 4.4.3 or later Operating system: macOS, Windows, or Linux Memory: minimum 4 GB RAM Storage: 50 MB for all files Required R packages (install before running): install.packages(c("tidyverse", "plm", "lmtest", "sandwich", "quantreg", "gsynth", "readxl", "remotes")) The gsynth package may require installation from GitHub if the CRAN version is unavailable for your R version: remotes::install_github("xuyiqing/gsynth") STEP 1: VERIFY THE DATASET --------------------------- Open R and run: df <- read.csv("RODE_Analysis_Panel_238obs.csv") cat("Rows:", nrow(df), "| Cols:", ncol(df), "\n") cat("Districts:", length(unique(df$District)), "\n") cat("Years:", sort(unique(df$year)), "\n") Expected output: Rows: 238 | Cols: 39 Districts: 34 Years: 2016 2017 2018 2019 2020 2021 2022 For full variable documentation, open RODE_01_Codebook.R and read the comments. No code needs to be run in this file. STEP 2: REPRODUCE ALL TABLES (Tables 3-6 in manuscript) source("RODE_03_Estimation.R") Table 3 — Panel fixed effects: Models A, B, C, D Table 4 — Panel quantile regression at Q25, Q50, Q75, Q90 Table 5 — AgriStack DiD results: TWFE ATT overall and cohort-specific (Cohort 1 and Cohort 2) GSC ATT via gsynth (Xu 2017, r=1 selected by cross-validation) Table 6 — Robustness checks: Baseline, excluding Konkan districts, pre-COVID period Outputs from Table 3 (Model B): log_mills:s_sg_lag1 = 0.223 (SE = 0.047, p < 0.001) s_sg_lag1= 0.344 (SE = 0.101, p = 0.001) Expected key output from Table 5 (TWFE): AgriStack ATT = -0.037 (SE = 0.022, p = 0.088) Output from GSC (Xu 2017): ATT (average) = -0.005, optimal factors r = 1 STEP 3: REBUILD DATASET FROM PRIMARY SOURCES To rebuild RODE_Analysis_Panel_238obs.csv from the original government data sources, open RODE_02_Build_Dataset.R and update the six file paths at the top of the script to point to your local copies of the source files. Primary source files required: (i) DES crop data: download from data.desagri.gov.in (Season: Kharif, All districts, Years: 2009-10 to 2022-23) (ii) IMD rainfall: district daily rainfall series (2009-2022) (iii) Cooperative mill registry: mahasugarfed.org (iv) PMFBY data: pmfby.gov.in (2018-2022) (v) AgriStack dates: agristack.gov.in rollout records (2020-2021) (vi) Companion district panel: constructed from Maharashtra Then run: source("RODE_02_Build_Dataset.R") The script saves the rebuilt panel to the path specified in PATH_OUT at the top of the script. RUNTIME ESTIMATES Step 1 (verification): < 1 minute Step 2 (all tables): 3-5 minutes Step 3 (rebuild from sources): 15-20 minutes (data sourcing not included; varies by internet speed and portal availability) CONTACT prakash.vhankade@gipe.ac.in

Institutions

• Gokhale Institute of Politics and Economics

  Maharashtra, Pune

Categories

Licence