Averages can hide the truth in a dataset. My next step was to isolate specific deviations from Benford expected at the digit and vote count levels. To do this, I flagged all election years and all candidate votes whose digit percent error was more than 2 standard deviations (prob ~ 2%) from the mean.
The anomalous years and candidate votes are:
.
Even while the means of all digits seem low, these years and digit errors are fairly substantial relative to all the rest.
Digit 4 and 5 on Democrat Vote 2020 really jump out at me.
The actual percent error for the Democrat vote in 2020:
Consider the weight of this Benford error rate analysis (2000-2020):
- This century, there have been only 5 vote counts where the error rate exceeds two standard deviations from the mean.
- Of the two major parties, the Democrat vote count is the only party whose votes have exceeded two standard deviations from the mean. That happened in 2012 and 2020.
- The election year 2020 had the most anomalies (3 of the 5).
- The anomalous votes for Democrats in 2020 were in counties where the digits shifted to 4 or from 5 in 2016.
Next step, closer inspection of those counties whose votes shifted to 4 or from 5 in 2020.
I'll call this group the Benford Anomalous Group (BAG): BAG