The notion that “there’s no evidence of election fraud” that took hold among Democrats and the mainstream media the night of November 3 is precisely the sort of assertion that lends itself to clear follow-up questions that almost no one seems interested in asking:
What would evidence of election fraud look like? Suppose that some country, at some point, held a fraudulent election. How could anyone prove it? Furthermore, is the only acceptable evidence so clear on its face that it would be evident the moment the polls closed? Couldn’t at least some evidence be hidden where only after-the-fact investigations could find it?
These questions are critical. Short of perhaps widespread videos of soldiers threatening voters, what sort of evidence might the president have presented on election night? What sort of evidence must he now present in order to prevail? Surely, if material election fraud did occur, it must have left some evidence. What is it?
While the specifics of the Trump campaign’s allegations have not yet been formalized (at least to the public), their contours are clear: A handful of jurisdictions under Democratic control destroyed, altered and/or fabricated enough ballots to shift the outcomes in selected states. How would anyone go about proving such an allegation?
The answer lies in the concept of “evidence marshalling,” a methodology used to bridge the gap between legal reasoning and Bayesian statistics. Bayesian analysis quantifies the likelihood of competing hypotheses using whatever evidence is available. The certainty and precision of the answer increases as more evidence arrives to support a specific hypothesis, but in a Bayesian framework it’s always possible to assess which answer appears most likely given what is already known.
Without getting into technical details, a Bayesian analysis begins with a “prior probability,” or a judgment derived from experience prior to looking at any specific data. The law reflects this approach (if informally) with “rebuttable presumptions.” A legal presumption is a statement that something is assumed to be true. A presumption is rebuttable if a litigant is allowed to submit evidence establishing that it’s likely false.
Beginning with that prior, Bayesian analysis considers each quantum of evidence individually. Each quantum pushes the probability “posterior” to its consideration either higher or lower. When all evidence has been considered, the posterior probability indicates whether the election was more likely to have been legitimate or riven with fraud.
The general legal presumption that American ballots and tallies are valid means that prior to the consideration of any evidence, the law has placed the probability that an election’s reported results are legitimate somewhere between 0.5 (even money) and 1.0 (certain).
The party attempting to prove fraud must bring evidence indicative of fraud. Again, each new showing increases the probability of fraud until the election appears to be more likely than not to have been fraudulent. Then the law shifts the “burden of proof” to the Biden campaign to present evidence indicating that the election results are legitimate. That burden continues to alternate until neither side has anything new to present.
How could the Trump campaign meet its initial burden? The best place to start is with uncontested evidence. Because roughly 90 million of 160 million ballots were circulated by mail, more than half the ballots were placed in circumstances in which it’s impossible to know who could access them.
Does that prove fraud? No. But it does suggest that fraud is more likely than in an election whose custodians protected all ballots at all times. It thus moves the (probability) needle a bit lower.
The same is true with the observation that Democrats clamored to reduce the security of these ballots while Republicans called to protect them. Proof? Still no. But again, a slight reduction in the probability that the results are legitimate.
Then come the various sworn statements, reports and rumors. Some states introduced unusual pauses in ballot counting? Court orders were violated? Observers were blocked? Tally patterns show statistical anomalies? Votes in a particular category or batch differ wildly from those in similar categories or batches? Computer glitches? Oddities pointing overwhelmingly in the same direction?
Every such report that the president can validate reduces the posterior probability that the reported election results are legitimate. How many would he have to validate to convince a court that at least some state elections were so riven with fraud that they need to be nullified?
Sadly, no precise answer is possible; legal reasoning lacks the mathematical formalism of Bayesian analysis. For those truly interested in understanding what it means for there to be evidence of election fraud, however, Bayesian methodology provides a clear answer.
The Trump campaign has already uncovered evidence of precisely the strategic fraud it alleges. The only question remaining is whether or not it can present enough such evidence to convince the courts.