STORY UPDATED: check for updates below.

Rounding error

Does an analysis of election data from Michigan and Georgia show suspicious vote shifts after the polls closed that took the leads away from President Trump and suspiciously awarded a disproportionate number of votes to Democratic candidate Joe Biden? No, it does not: First of all, The Gateway Pundit claims to use New York Times data for its analysis, not official election tallies. Secondly, the analysis does not take into account the fact that its tally uses rounded figures for the ratios for Trump and Biden votes and those ratios can be jittery with each subsequent vote dump as the numbers round up or down.

The claim appeared in an article published by the Gateway Pundit on November 17, 2020 titled "WE CAUGHT THEM! Part 6: In Michigan and Georgia, Like in PA and VA, Caught in SAME PATTERN! -- Once Biden Gained Lead with MASSIVE Vote Dumps, The Remainder of Votes All Possessed Same Biden to Trump Vote Ratio - THIS IS IMPOSSIBLE!" (archived here), which opened:

We've reported on multiple incidents of 'glitches' that people were observing related to the reporting of the 2020 Presidential Election. These incidents seem odd and so we sought out a data set from the 2020 Presidential election and we found it. A group of IT patriots had already analyzed a data set of election vote...

Users on social media only saw this title, description and thumbnail:

WE CAUGHT THEM! Part 6: In Michigan and Georgia, Like in PA and VA, Caught in SAME PATTERN! -- Once Biden Gained Lead with MASSIVE Vote Dumps, The Remainder of Votes All Possessed Same Biden to Trump Vote Ratio - THIS IS IMPOSSIBLE!

We've reported on multiple incidents of 'glitches' that people were observing related to the reporting of the 2020 Presidential Election. These incidents seem odd and so we sought out a data set from the 2020 Presidential election and we found it. A group of IT patriots had already analyzed a data set of election vote...

The Gateway Pundit article appears to use the same data source that a user of the messaging board thedonald.win utilized in a November 12, 2020 posting. The data in both postings have a time time stamp, the total number of votes and the ratio of those votes for Trump and for Biden. It does not contain exact vote counts for the candidates and also did not include third party candidates. It is likely this data was used to create a live display of unofficial election results in real time.

Lead Stories analyzed the initial data on November 13, 2020 and found it lacking because of the rounding used in the ratio of votes for each candidate made it imposible to use the figures to calculate exact vote counts for the candidates.

The Lead Stories analysis of that data said:

Rounded numbers are by nature not exact, so any analysis relying on them to make statements about exact numbers is virtually guaranteed to be wrong.

Looking at the data, Lead Stories determined the files are incomplete, further throwing off their validity. Lead Stories noted:

Right away it is clear there are no exact vote counts for either candidate in the files and there also appears to be no data for third-party candidates. So, it appears these files are not meant to be complete representations of the election results.

More specifically, Lead Stories said:

The important factor Lead Stories found is that the graphics-friendly ratios in the scraped data only appear to have at most three decimals so the biggest change they can register is 0.001 (or 0.1%). If standard rounding is applied, this means the real ratio could be anything between 0.0005 higher or lower (or 0.05%). That may not seem like much but when you are dealing with hundreds of thousands (or even millions) of votes, it adds up.

... even with relatively small numbers of votes, the calculated numbers of votes are immediately off by a few compared to the real numbers. Only if the ratio by chance happens not to need rounding does the calculated number of votes line up with reality ...

You can see the detailed Lead Stories analysis of the stats used in the donald.win posting here.

You can see the entire sheet with Lead Stories' calculations on the stats used in thedonald.win posting here.

In addition to these claims the Gateway Pundit also makes a big deal out of the vote shares for new batches becoming more and more constant near the end of the count. However the vote share reported in the data is always calculated off the total number of votes at that moment, not off the current batch of added votes. To put it simply, if five million votes have already been counted and they give a hypothetical 31.1% to 58.5% split (ignoring the third party candidates like the data does) adding a thousand more votes will likely not change those percentages (even if they all go to one candidate). And when these votes do change the percentages (because they made the result go over the margin for rounding, i.e. 31.14999999...% to 31.15000000...% makes the rounded number jump from 31.1% to 31.2%) they may massively flip any calculation based on these rounded ratios.

And again, you cannot use these rounded numbers to calculate exact vote counts anyway, so based on this data making any statement about the number of votes gained by a candidate in a particular batch (or the ratio between votes for the candidates for a particular batch) is completely impossible.

Lead Stories reached out to the Gateway Pundit for comment, and this story will be updated as appropriate if a response is received.

NewsGuard, a company that uses trained journalist to rank the reliability of websites, describes thegatewaypundit.com as:

A partisan conservative website that regularly publishes hoaxes, conspiracy theories, and unsubstantiated claims, including those related to the COVID-19 pandemic.

According to NewsGuard the site does not maintain basic standards of accuracy and accountability. Read their full assessment here.