Playing Around With Benford's Law In Excel. Yeah, Those Graphs Are Fishy.
(media.patriots.win)
Comments (18)
sorted by:
DEFAULT COMMUNITIES
•
All
General
AskWin
Funny
Technology
Animals
Sports
Gaming
DIY
Health
Positive
Privacy
I think Benford's law wasn't applied properly in this case. You cannot run the formula on total numbers. You have to run it on the batches of numbers that are reported and added together. If they are smudged you would see an unnatural distribution.
In case of this election you have to run it on the batches of votes that are reported. If the counts are manipulated they would look differently. When people make up numbers they are not completely random.
Okay, so how would I change it to be more accurate? Not split into R and D?
Oh I see what you mean. Hmmm, okay I may mess around with those.
Benford's law would apply here if there were no human intervention.
Actually it works across magnitudes just fine
but that's only one sample per state. That would only work is the entire count of a state was 100% made up. Don't forgot this law is only about first digits.
Yeah, I dunno. I just took the numbers as of right now to see how the graphs looked.
LOL REEEEEEEEEEEEEEEEEEEEE
I made a spreadsheet last night for something else, decided to adapt it to do a similar analysis comparing 2016 and 2020.
https://i.imgur.com/UaIwZDv.png
Very nicely done!
Those graphs are Biden's law
Small suggestion: might be useful to include a line for the expected value to compare it to.
The formula should be: https://external-content.duckduckgo.com/iu/?u=http%3A%2F%2Feventlogxp.com%2Fblog%2Fwp-content%2Fuploads%2F2016%2F03%2Fbenfordlaw-formula.jpg&f=1&nofb=1
Nice, thanks pede!
Do it for individual counties or matches of votes in updates in the time series.
Between states it's very poor. It'll most drown out anomalies and it's also a smaller sample set at the same time meaning it should deviate more from Benford's law.
Also do Benford's over time do see if it deviates then when.
Got the numbers from Decision Desk HQ's site as of 11/10/20. If any pede would like the file, feel free to ask!