Important update 11 November 8:45 PM The missing votes appear to have been a data input error. The totals at the end of the early data I was given were correct. It was only the one time point that was in error. The time step where Trump goes down was due to missing data at that time: the old file I was given at the time stamp has data missing for the totals. The totals before and after are the same in the corrected data. Meaning the first curiosity has been solved: simple data error entry, which only affected the one time point. Here is the PA web site for data. Here is the corrected version of the data I was given (Excel file).
A brief note. I’ve been asked to examine the Pennsylvania votes. That work is ongoing. Update See below for a serious critique of Benford’s law.
I’m showing here (with permission) the one analysis I found most curious.
This is official county-level timed voting data that started at 2020-11-04 11:00:00, a day after the election, to 2020-11-07 11:29:00 which is Saturday night. That is, these are all late vote counts. They start, county by county, where the vote left off on election night.
This is a picture of the running totals by the time the votes were added, summed across all counties, during those time periods. They do not start at 0, but at the totals given after election night.
Important update This is the old figure with uncorrected data.
The early gains for Biden are from, mainly, Philadelphia, Allegheny, Montgomery, Chester and Berks counties. A simple plot (click to see: it’s large) shows the size of vote additions for both candidates, when new vote totals (greater than 0) were added by county (and not all counties added votes after election day).
All goes well for Trump until 2020-11-04 21:15:00 when he loses just under 10,000 votes, but curiously from three different counties simultaneously: -1,063 Allegheny; -2,972 Bucks; -7,135 Chester. Biden never lost any votes (at least, in this late voting).
Important Update This has been solved. Missing data at the one time point. Since the totals at all other points match, this curiosity does not appear to have been deliberate, and is just a common result of looking at early data, which can be messy.
Here is the figure with the corrected data:
Understand that this does not mean the decreases happened at this time, but that they were recorded in the official data as happening at that time. And the same is true for our next observation.
Biden’s next curiosity was the big increase of 27,396 votes at 2020-11-06 08:53:00 over one consecutive reporting period. This bump is just like the blue-red F-memes you have seen: this only seems more spread out because of the finer time scale used.
These two curiosities account for a 37,263 vote swing for Biden. Biden’s total, as of the end of this data, was 3,344,528, and Trump’s 3,310,326. Biden therefore “won”, in this dataset anyway, by 34,202 votes.
Important Update The late Biden addition is still there and real.
There is more to come. Stick around.
Update Benford’s law is only useful in uncovering multiple and on-going instances of cheating. As in somebody consistently cooking financial books. As I showed above, assuming the curiosities are cheats, it only took two instances to tip the balance. Benford’s law will never pick this up: never.
I’m skeptical of what I’m seeing in other analyses, because if somebody turns something up with Benford, it implies that many, many vote totals were tampered with, which increases the possibilities of getting caught. And you don’t need to tamper with many. Only a few.
To support this site and its wholly independent host using credit card or PayPal (in any amount) click here