Prosecutor Recall Elections and Public Opinion

Overview

This repository contains the replication materials for our study on prosecutor recall elections and public opinion, published in Public Opinion Quarterly. The research examines voter attitudes toward prosecutor performance and policy positions in the context of recall elections.

Key Findings

• Voters distinguish between extensive margin (who enters the system) and intensive margin (severity of treatment) policy positions
• "Progressive" voters who supported the recall show distinct policy preferences
• Treatment framing significantly affects political support for prosecutor candidates

Technical Highlights

Component	Technologies
Statistical Analysis	R (estimatr, modelsummary)
Machine Learning	randomForest for voter classification
Visualization	ggplot2, custom theme development
Survey Methods	Experimental design, survey weighting

Project Structure

prosecutor-recall-study/
├── analysis/
│   ├── da_ana.Rmd                # Main analysis (R Markdown)
│   └── data/
│       └── sample/               # Sample data for replication
├── poq/
│   ├── poq_margins.Rmd           # Publication figures
│   ├── poq_margins_descriptives.Rmd
│   └── readme.txt
├── figures/
├── LICENSE
└── README.md

Experimental Design

The study uses a randomized survey experiment with three treatment conditions:

Condition	Description
Get Tough	Prosecutor emphasizes punitive policies
Reduce Extent	Prosecutor emphasizes diversion/declination
Reduce Intensity	Prosecutor emphasizes reduced sentences

Key Analyses

Treatment Effects by Recall Vote

# Linear regression with clustered standard errors
recall_models <- list(
  "Recall Proponents" = lm_robust(out1 ~ group1, data = da_recall, 
                                   clusters = response_id),
  "Recall Opponents" = lm_robust(out1 ~ group1, data = da_no, 
                                  clusters = response_id)
)

Random Forest Classification

# Predict recall vote using policy preferences
rf <- randomForest(
  out_vote ~ out2 + sal + puni_t + prog + sym + rr,
  ntree = 2000,
  mtry = 1,
  data = match_da
)

Installation

# Install required packages
install.packages(c(
  "here", "tidyverse", "estimatr", "modelsummary",
  "randomForest", "ggplot2", "haven"
))

Replication

# Set working directory to project root
library(here)
here::i_am("analysis/da_ana.Rmd")

# Render the analysis
rmarkdown::render("analysis/da_ana.Rmd")

Data

The analysis uses survey data from three sources:

1. San Francisco voters (n=866): Local residents during the Boudin recall
2. IGS California Poll: Statewide representative sample
3. National Lucid sample: Cross-validation of findings

Sample data provided for code testing. Full data available upon request.

Citation

@article{yogev2025recall,
  title={Prosecutor Recall Elections and Public Opinion},
  author={Yogev, Dvir},
  journal={Public Opinion Quarterly},
  year={2025},
  publisher={Oxford University Press}
}

Author

Dvir Yogev
Post-Doctoral Researcher
Criminal Law & Justice Center, UC Berkeley Law

License

MIT License - see LICENSE for details.