I couldn't figure out how to post an image in a reply so I put in this new post.
It was based on an exchange regarding state vote fraud summaries and me arguing that Benford's law is not valid for determining voter fraud because the underlying dataset is very structured. It is structured because election commissions plan the different voting areas.
They plan voting districts/wards/precincts based on things like population density, distance from other voting locations, staffing, facilities, machines, logistics, whether voters will walk or drive, etc. This planning process results in vote district sizes that tend to cluster in certain ranges.
Benford's law works best for data spanning many orders of magnitude and needs an underlying dataset that is unstructured. A particular voting dataset may or may not follow Benford's law and may or may not have had fraud.
Someone replied "put up or shut up" so here I have downloaded the data for Milwaukee in 2020, totaled the votes in each ward (what they call them there) and performed a histogram to show the data structure.
Now even if the vote districts didn't show this structure, but instead number of votes in districts are correlated with population density, and votes for a candidate are also correlated with population density. (less dense areas trend DJT and higher density populations trend Biden) Well that would also strongly structure the dataset and make Benford's law invalid.
When I say "invalid" I really means a high enough percentage of false positives and false negatives that it isn't a reliable indicator of whether fraud did or didn't occur.
I couldn't figure out how to post an image in a reply so I put in this new post.
It was based on an exchange regarding state vote fraud summaries and me arguing that Benford's law is not valid for determining voter fraud because the underlying dataset is very structured. It is structured because election commissions plan the different voting areas.
They plan voting districts/wards/precincts based on things like population density, distance from other voting locations, staffing, facilities, machines, logistics, whether voters will walk or drive, etc. This planning process results in vote district sizes that tend to cluster in certain ranges.
Benford's law works best for data spanning many orders of magnitude and needs an underlying dataset that is unstructured. A particular voting dataset may or may not follow Benford's law and may or may not have had fraud.
Someone replied "put up or shut up" so here I have downloaded the data for Milwaukee in 2020, totaled the votes in each ward (what they call them there) and performed a histogram to show the data structure.
Now even if the vote districts didn't show this structure, but instead number of votes in districts are correlated with population density, and votes for a candidate are also correlated with population density. (less dense areas trend DJT and higher density populations trend Biden) Well that would also strongly structure the dataset and make Benford's law invalid.
When I say "invalid" I really means a high enough percentage of false positives and false negatives that it isn't a reliable indicator of whether fraud did or didn't occur.