Skip to content

Replication

This page describes how to replicate the results presented in the paper.

Replication Package

The full replication package includes all code, processed datasets, and manuscript source files needed to reproduce every table and figure in the paper.

Repository Structure

The replication materials are organized in version-specific directories. The v4 directory contains the current primary analysis in R, while v2 and v3 contain legacy Stata implementations.

Software Requirements

Primary Analysis (v4 -- R)

Software Version Purpose
R 4.5+ Statistical computing
fixest latest High-dimensional fixed effects estimation
data.table latest Fast data manipulation
modelsummary latest Regression tables
ggplot2 latest Publication-quality figures
arrow latest Reading Parquet files

Additional R packages: kableExtra, sandwich, lmtest, broom, scales, sf, viridis.

Legacy Analysis (v2/v3 -- Stata)

Software Version Purpose
Stata/SE 17+ Statistical computing
reghdfe latest High-dimensional fixed effects
ftools latest Fast Stata tools

Manuscript

Software Version Purpose
LaTeX TeX Live 2024+ Document typesetting
elsarticle latest Journal document class
chicago latest Bibliography style

Data Sources

Primary Dataset

BEC-G65-WORK1.parquet

  • Source: Bolsa Eletronica de Compras (BEC), Sao Paulo state electronic procurement platform
  • Format: Apache Parquet (~125 MB)
  • Observations: 479,330 bids
  • Variables: 180 columns
  • Coverage: All bids for BEC Group 65 (medical, dental, and hospital supplies), January 2009 -- December 2019
  • Unit: Bid level (firm x item x procurement event)

Data Access

The BEC procurement data is publicly available through the Sao Paulo state transparency portal. The processed dataset used in the analysis is included in the replication package.

Key Variables

Variable Description
purchase_type 0 = Ordinary, 1 = Administrative, 2 = Litigated
bid_price_ref Reference price (maximum the government will pay), in BRL
bid_price Negotiated (final) bid price, in BRL
bid_qty Quantity demanded in tender notice
n_firms_bids Number of distinct firms submitting bids
po_item_winner Tender success indicator (1 = successful purchase)
item_id Product identifier (used for item FE)
pbu_code Public buyer unit code (used for PBU FE and clustering)
year_n Year (used for time FE)

A full data dictionary is available in the replication package (DATA_DICTIONARY.md).

Running the Analysis

Quick Start (v4 -- R)

# 1. Clone the repository and navigate to the project
cd paper1-bitter-pills

# 2. Run the full v4 pipeline (~5 minutes on 16 cores)
Rscript v4/run_all.R

This single command executes all analysis scripts in sequence:

  1. 00_prepare_data.R -- Load and prepare the dataset (cached at /tmp/v4_prepared.rds)
  2. 01_desc_stats.R -- Descriptive statistics
  3. 02_balance_table.R -- Balance table for urgent subsample
  4. 03_main_regressions.R -- Main regression tables (4 FE specs x 3 clustering variants)
  5. 04_heterogeneity.R -- Heterogeneous effects analysis
  6. 05_fiscal_costs.R -- Fiscal cost estimates
  7. 06_robustness.R -- Robustness checks (120+ regressions)
  8. 07_graphs.R -- Generate all figures
  9. 08_pub_tables.R -- Publication-ready LaTeX tables
  10. 09_pub_figures.R -- Publication-ready PDF figures

Running Individual Scripts

# Must run data preparation first
Rscript v4/analysis/00_prepare_data.R

# Then any individual analysis script
Rscript v4/analysis/03_main_regressions.R
Rscript v4/analysis/07_graphs.R

Legacy Analysis (Stata)

# Install required Stata packages
stata-se -b -q do v2/analysis/install_packages.do

# Run the v3 pipeline
bash v3/analysis/run_all.sh

Output Files

Tables

Directory Format Count Description
v4/pub/tables/ LaTeX (.tex) 17 Publication-ready tables (booktabs + threeparttable)
v4/manuscript/ LaTeX (.tex) 59 Full regression tables (tabularray format)
v4/results/ HTML (.html) 59 Browser-viewable tables

Figures

Directory Format Count Description
v4/pub/figures/ PDF 8 Publication-ready figures (grayscale, 6.5 x 4 in)
v4/graphs/ PDF 8 Color figures for presentations

Manuscript

# Compile the manuscript
cd v5/manuscript/paper
pdflatex -interaction=nonstopmode main.tex
bibtex main
pdflatex main.tex
pdflatex main.tex

Computational Environment

The analysis was developed and tested on the following system:

Component Specification
OS Ubuntu 24.04 (WSL2 on Windows)
CPU 16 cores
RAM 15 GB
R 4.5
fixest Uses OpenMP for parallel estimation (16 threads)

Runtime

The full v4 pipeline (run_all.R) takes approximately 5 minutes on the reference system. The most time-intensive step is the robustness analysis (06_robustness.R), which estimates 120+ regressions across three winsorization levels.