One recurring question in election-integrity research is whether irregularities, when they occur, are randomly distributed or concentrated in particular kinds of contests. Researchers studying this question have generally looked for a specific pattern: are discrepancies between independent verification data (like exit polling) and official results more common in competitive races than in lopsided ones?
The logic behind this line of inquiry is straightforward. If a discrepancy pattern is essentially random noise — the product of ordinary polling error, sampling variation, or minor administrative inconsistency — it should show up roughly evenly across competitive and noncompetitive contests alike. If, on the other hand, a pattern shows up disproportionately in close races and largely disappears in races decided by wide margins, that asymmetry itself becomes a data point worth examining further, since it doesn’t fit what would be expected from simple statistical noise.
Testing this kind of hypothesis requires careful comparative methodology: consistent definitions of what counts as “competitive,” consistent measurement of the discrepancy being studied, and enough contests in each category to draw a statistically meaningful comparison rather than anecdotal conclusions from a handful of races. Researchers in this space typically publish their full methodology and underlying data specifically so that other analysts can attempt to replicate or challenge the findings — a core practice of any legitimate empirical research.
It’s worth being clear about what this kind of analysis can and cannot establish on its own. A statistical pattern showing that discrepancies cluster in competitive races is a finding that invites further investigation — it is not, by itself, direct proof of any specific mechanism causing that pattern. Responsible researchers are careful to describe such findings as raising a hypothesis rather than confirming one, and to call for the kind of direct, ballot-level auditing that could actually confirm or rule out an explanation.
This is ultimately the throughline connecting statistical research like this to the more hands-on tools of election verification: paper ballots, chain-of-custody procedures, and risk-limiting audits. Statistical analysis can identify where to look more closely. Direct, physical verification of the ballots is what actually resolves the question.
Controlling for Confounding Variables
The central methodological challenge in any research comparing competitive and noncompetitive races is ruling out confounding variables — underlying differences between competitive and noncompetitive districts that have nothing to do with vote counting, but that could still produce a correlation resembling the one being studied. Competitive districts often differ from safe districts in turnout levels, campaign spending, media attention, and demographic composition, any of which could independently affect polling accuracy. Rigorous research in this space has to statistically control for these factors before treating any remaining correlation as meaningful.
Peer Review and Replication
As with any empirical research making a claim of this significance, findings in this area benefit enormously from peer review and independent replication using the same publicly available data. Research that survives scrutiny from statisticians and political scientists without a stake in the outcome carries more weight than research that hasn’t been tested this way. Publishing full methodology and raw data specifically invites this kind of independent check, which is a feature of credible research, not a formality.
What Would Resolve the Question
Ultimately, a statistical correlation between contest competitiveness and reporting discrepancies — however well-controlled — is a research finding, not a legal or administrative conclusion. What actually resolves the underlying question in any specific race is a direct, ballot-level audit: hand-counting the paper ballots from the specific competitive contests flagged by the analysis, and comparing that count to the machine-reported total. Statistical research can make the case for prioritizing which races get that closer look. Only the physical audit can settle what the ballots actually show.