Analysis of Detroit, Michigan's 2020 Absentee Ballot Images

In 2020, Detroit processed a record number of absentee ballots due to the pandemic. The central counting location was the TCF Center, where ballots were organized into 134 Absent Voter Counting Boards (AVCBs). Each AVCB handled ballots with a specific "ballot style" (different versions based on precincts).

Through Freedom of Information Act (FOIA) requests, we obtained:

• Digital images (PDF files) of the actual paper ballots.
• Hand written logs on how many ballots each worker processed per AVCB (called "FOIA Ballots Count" in the table).

I compared these to the official "Statement of Votes Cast" — the county's reported totals from election night.

Key Findings from the Table

Total official ballots reported (Statement of Votes): 174,384
Total from county scanner logs (FOIA Ballots Count): 172,584 → short by about 1,800
Total ballot images received and counted from PDFs: 171,510 → short by about 2,874 compared to official

To measure overall discrepancies, we look at the absolute variance (ignoring whether over or under, just the total size of differences across all boards):

• Between official totals and FOIA scanner logs: 7,886 ballots
• Between official totals and actual PDF images received: 3,926 ballots

Common Explanations for the Differences

Most variances appear to stem from normal challenges in a massive, high-speed scanning operation rather than anything intentional. Here are the main patterns I observed:

• Double-scanning: In several AVCBs, the same ballots were accidentally scanned more than once. This inflated the scanner logs (making FOIA counts higher than official). Because ballots weren't shuffled between scans, duplicates were easy to spot and remove when counting images. This accounts for more than half of the larger positive variances — likely just human error in a rushed process handling hundreds of thousands of ballots.
• Clerical/reporting errors: Some boards reported fewer ballots scanned than expected, but the PDF images showed the correct (or very close) number. For example, AVCB 6 reported 1,681 scanned, but we received images for 2,538 — nearly matching the official total. These seem to be simple mistakes in recording the numbers.
• Missing or delayed PDF files: In cases where far fewer images arrived (e.g., AVCB 35: reported 1,789 but only 758 images), it usually meant one or more PDF files hadn't been sent yet. I notified the county about roughly 20 such missing files, and they provided most of them shortly after. This explains many of the shortages.
• Small differences: Many boards were exact matches or off by just a handful — typical in any large manual process.

Overall, the data shows a huge effort to process absentee ballots accurately under difficult conditions. The remaining variance (especially after receiving additional PDFs) is small relative to the total volume and consistent with logistical or clerical issues, not systemic problems.

This review supports the importance of public access to election records via FOIA — it allows independent verification and helps build confidence in the results.

Download the official Wayne County, Statement of Votes Report from this link.