A widely viewed video found ‘suspect’ votes in Georgia. We found a lesson in probability.



For the 2020 presidential election, The New York Times uploaded millions of votes from around the nation to give readers the latest tally on the race.

A third-party vendor culled the votes electronically from states’ official returns, often several times an hour, and sent them to the Times in batches of thousands. A batch might include votes from every corner of a state, or specific regions, all according to when individual counties updated returns.

A video viewed more than 133,000 times seized on one such batch to allege that Joe Biden collected a share of votes in Georgia that was likely evidence of fraud.

The author claimed Biden won 98% of a 23,000-vote batch uploaded to the Times’ site Nov. 4, the day after the election. Such a lopsided vote share was “impossible,” the video claimed.

Impossible? Not really.

Using the same methodology as the author, The Atlanta Journal-Constitution found six batches of ballots reported from around the country where one presidential candidate won at least 95% of a batch. Three of those batches went for Biden, the other three went for Donald Trump. In fact, Trump won the most polarized batch of all — taking 99.9% of a 130,000-vote batch in Pennsylvania.

Here are the details:

Credit: Nick Thieme

Credit: Nick Thieme

None of these batches suggest a possible conspiracy in favor of either Biden or Trump. They merely show that when you repeat a process over and over, sometimes you get results that look funny — but aren’t.

Get 10 heads in a row from a coin flip, you might doubt the fairness of the flipper, or the coin. Statistically, however, if you flip a coin 1,000 times, the probability of getting at least 10 heads in a row is about 21%. So, roughly every 5 times you flip 1,000 coins, you can expect to get at least 10 heads in a row.

Now consider that nearly 160 million Americans cast votes that were uploaded in batches to the Times’ and other news sites during the election. When data is reported in tens of thousands of snapshots like this, you end up with numerical results that look weird but aren’t, purely by chance.

For the data in the video to be evidence that Biden’s 98% batch in Georgia was impossible or a suggestive of a conspiracy, Trump’s 99% batch in Pennsylvania would also have to be evidence of a conspiracy. What’s harder to imagine is that such a conspiracy happened all over the country, both in favor of Trump and in favor of Biden, when it can arise randomly, instead.

Nick Thieme is a data specialist for The Atlanta Journal-Constitution.