Thread by @EduEngineer

As a consistent third party voter living in a state that wasn't going to swing a close election, I don't vote R or D.

But this is the time to throw weight behind an investigation into the possibility of a stolen election. We must investigate the statistical evidence.

So, let us investigate the statistical evidence broadly and as a community. Carefully and honestly. Benford's law is a clever technique, but there is an underlying reason behind it that should be understood to best apply the fundamental test.

The reason behind it is that population pools grow exponentially, so they move through orders of magnitude at an invariant rate. If we take the logarithm, the results are then linear. Discarding the integer parts, the fraction parts should form a uniform distribution.

The object to this is that range constraints throw off the uniformity of that distribution. We can correct for this! Constrained distributions tend to either be monotonic or convex with a local maxima/minima.

When we choose a smaller logarithmic base, our 9 or 10 intervals (that's arbitrary for the lead digit count, so just think k intervals) become modular and stack the different parts of the distribution in a way that in practice induces uniformity to a high degree.

To convince yourself, try it with a data set. Instead of counting the lead digits, take the log base 1.2 of the vote counts (or try different bases). Now discard the integer part. Pick k intervals from 0 to 1 and count the number of the remaining fraction parts in each interval.

Will we see that election results have high smoothness over those k intervals? I have seen this for myself to a high degree so far, but what I want for you to do is test it on several elections and see for yourself.

Now, run the same filter on Biden data for this election or other elections.

Your jaw may drop to the floor. One of these candidates does not look like all of the other ones, historically speaking. If you're a half-decent spreadsheet jock, see for yourself.