Eforensics Analysis of the 2024 Presidential Election in Pennsylvania

🔗 https://websites.umich.edu/~wmebane/PA2024.pdf
Posted on Jan 1, 1

Summary

Professor Walter R. Mebane Jr. applies his eforensics model to precinct-level data from Pennsylvania’s 2024 presidential election. Using Bayesian statistical techniques and mixture modeling, Mebane estimates:

  • Between 210,000 and 225,000 votes may be eforensics-fraudulent.
  • The presidential margin between Trump and Harris was approximately 120,000 votes.
  • Fraudulent vote types are categorized as:
    • Incremental Fraud (e.g. strategic ballot inflation or suppression)
    • Extreme Fraud (e.g. ballot box stuffing or deletion)

Key findings:

  • Philadelphia and Huntingdon Counties stand out as having significant signs of malevolent vote distortions.
  • Many precincts showed perfect turnout (100%), prompting concern and imputation of missing data.
  • The analysis flags 1,804 precincts as having some form of eforensics-detected fraud, with a small number showing extreme values.
  • The model shows posterior multimodality, indicating multiple plausible fraud patterns.

Mebane warns, however, that some flagged votes may stem from strategic voter behavior (e.g., coordination or tactical voting) rather than outright fraud, as observed in clean German elections. Still, even under conservative assumptions, more than 25,000 votes may have stemmed from malevolent distortions—a nontrivial portion of the margin.

Citation

Mebane, W. R. Jr. (2025). Eforensics Analysis of the 2024 Presidential Election in Pennsylvania (Working Paper). University of Michigan. Retrieved from https://websites.umich.edu/~wmebane/PA2024.pdf