presdifffromparty = presperformance - partyprecinctperf // y = n - x
precinctinfo['precinctrep'] = partyprecinctperf // x
...
jsondict['x'] = precinctinfo['precinctrep'] // x
jsondict['y'] = presdifffromparty // y
Looks like the same problem.
Instead try...
jsondict['y'] = presperformance
I could probably give some indentation / whitespace and naming advice but not really critical.
Just stuff like...
[ precinct.get('precinct')]
Should be...
[precinct.get('precinct')]
And a lot of inconsistency. I usually set my IDE to show white space and tabs. It looks as though at least two people worked on this or parts were copies in or multiple editors were used. I see double spaces, then quad spaces, then tabs.
Consider using _ to separate words in variable names. For example bunchofwordssquashedtogether is not particularly easy on the eye.
I'll look at it closer tomorrow to understand exactly what these colums are and how to properly put them to use (full define them so I know exactly how they relate to what's on ballot, maybe dig around for a ballot image too). Saying that it looks like individual names so it's a bit different. I guess the concept of straight R is in the code. Names don't have party but I can google it. I'll probably just plot all the candidates performance first.
Note that this data is limited you may not be able to detect fraud with it even if these is fraud.
Ha! Hopefully you're on the right side of the Balmer curve.
I had my editor crash in the middle and switched so that's probably the spacing issue. I was also copying from the VA work. That was even dirtier because I got it wrong so many times before it worked and I'm doing this on the side of my job.
Also, take a look at the pure numbers in a spreadsheet from the state. For the counties this method has said there was shenanigans and you'll see the trend by looking at Trump vs. the worst of the Senate/House candidate in each precinct. That's how I originally figured out.
The real problem is that a lot of this analysis isn't necessarily going to show anything. You might get lucky and it might turn out that one country cheated or was flipped and it blurts out as a big anomaly.
The level of fraud you need to perform to flip an election for a state where it's naturally close isn't necessarily something that's going to show up on most statistical analysis.
To make a simple point, shuffle all the point on your graph randomly 1%. Does the new graph tell you anything different? Would you even know anything was different if you hadn't see the original graph? When you think of it like that then you can understand the problem we're dealing with.
We still need to look at it and this is a start but it's also magnification X1.5 and we probably need to go to X10000 to fully uncover the full scope of fraud and corruption.
If there's cheating and a state swings over to another candidate by a few tens of a percent to a percent then that might indicate how small your fraud indicator has to be.
Numerically and statistically speaking if you have illegals voting in the election in a significant proportion (merely a tenth of a percent can be significant) there's no reason that would register in an obvious way in numerical analysis.
The lawyers will likely have much better data to work with, for example there's no reason why they should not be furnished with a database of all registered voters and the individual votes as well as whatever it is the poll workers would have access to including the physical ballots.
A single snapshot of 2020 and the results on its own is only of limited use. It might however gleam something but if not you get a lay of the land.
Comparing the relative performance of candidates is useful any interesting though it might not reveal anything you can be sure is fraud. People have already found in some data sets it appears Biden got lots of votes from people who apparently only filled that part of the ballot which if true on its own is already very suspect.
It's more useful to compare this data with 2016 and other datasets like populations for each region, registered voters, etc.
Most of the cheating I suspect though will be in cities or densely populated locations. In those it's a natural haystack you can bury your needle into.
Realistically they want to take a few thousand ballots at random for the state and try to track down the voter.
If I were making algorithms to cheat in the election they I would probably do something like take the top 50 closest states where Trump wins then flip them. Basically spread out the risk so it's hard to detect.
I believe a lot of the cheating however is not really grandoise. It's just things like optimising for mail-in ballots, fiddling the rules, etc.
If I was going to cheat in this election then I'd use multiple approaches including BLM. It's a criminal syndicate. Basically criminals, it's an anti-police movement so go figure. The criminal system has an underground network that can reach tens of millions or Americans or more. Compared to peddling crack cocaine a bit of ballot harvesting is a breeze and for someone who wants to buy the election it's far more cost effective than directly donating to the campaign to instead donate to organised crime. If organised crime can organise all these riots then having loads of ballots ready to drop late at night all for Biden is an absolute breeze.
I'd also look for USPS workers disgruntled about reforms to make it profitable and who want to get at Trump and who likely already to some degree have a network of employees or union or whatever.
There are two massive vulnerabilities they can very easily and subtly exploit. First is that they were allowed to collect after deadline knowing how much to make up for. Secondly is that they know mail-in goes more to Biden. The entire design of the election system and the rules changes for COVID-19 actually broke it. If they know after 12 mail-in ballots likely favour Biden then it's really easy to just optimise for that and accept as many as possible. They can also store them somewhere secret as a buffer, sorted buffer, sort them by region and proportion of democrats in said region then slowly release then in batches as needed.
This is actually just a standard CS practice. Buffering and priority queue.
What I could probably do is write a small simulation of an election system that can do things like this just to see what kinds of patterns I get.
I agree that this is one piece of a much larger puzzle. Let me know if you find any other additional concepts for how fraud might have surfaced at the precinct level data. The one that showed how absentee algorithm may have been based on Trump election day seemed promising
New post is here: https://thedonald.win/p/11Q8cBUnxd/ne-2nd-congressional-algorithmic/c/
I included a download to the python code. This is same method I used for Virginia but cleaned up quite a bit. Can you review?
I'm a bit drunk but...
...
Looks like the same problem.
Instead try...
I could probably give some indentation / whitespace and naming advice but not really critical.
Just stuff like...
Should be...
And a lot of inconsistency. I usually set my IDE to show white space and tabs. It looks as though at least two people worked on this or parts were copies in or multiple editors were used. I see double spaces, then quad spaces, then tabs.
Consider using _ to separate words in variable names. For example bunchofwordssquashedtogether is not particularly easy on the eye.
I'll look at it closer tomorrow to understand exactly what these colums are and how to properly put them to use (full define them so I know exactly how they relate to what's on ballot, maybe dig around for a ballot image too). Saying that it looks like individual names so it's a bit different. I guess the concept of straight R is in the code. Names don't have party but I can google it. I'll probably just plot all the candidates performance first.
Note that this data is limited you may not be able to detect fraud with it even if these is fraud.
Ha! Hopefully you're on the right side of the Balmer curve.
I had my editor crash in the middle and switched so that's probably the spacing issue. I was also copying from the VA work. That was even dirtier because I got it wrong so many times before it worked and I'm doing this on the side of my job.
FYI, i was able to get New Hampshire data and it doesnt show anything.
https://rumble.com/vb4c2b-new-hampshire-by-precinct.html
Also, take a look at the pure numbers in a spreadsheet from the state. For the counties this method has said there was shenanigans and you'll see the trend by looking at Trump vs. the worst of the Senate/House candidate in each precinct. That's how I originally figured out.
The real problem is that a lot of this analysis isn't necessarily going to show anything. You might get lucky and it might turn out that one country cheated or was flipped and it blurts out as a big anomaly.
The level of fraud you need to perform to flip an election for a state where it's naturally close isn't necessarily something that's going to show up on most statistical analysis.
To make a simple point, shuffle all the point on your graph randomly 1%. Does the new graph tell you anything different? Would you even know anything was different if you hadn't see the original graph? When you think of it like that then you can understand the problem we're dealing with.
We still need to look at it and this is a start but it's also magnification X1.5 and we probably need to go to X10000 to fully uncover the full scope of fraud and corruption.
If there's cheating and a state swings over to another candidate by a few tens of a percent to a percent then that might indicate how small your fraud indicator has to be.
Numerically and statistically speaking if you have illegals voting in the election in a significant proportion (merely a tenth of a percent can be significant) there's no reason that would register in an obvious way in numerical analysis.
The lawyers will likely have much better data to work with, for example there's no reason why they should not be furnished with a database of all registered voters and the individual votes as well as whatever it is the poll workers would have access to including the physical ballots.
A single snapshot of 2020 and the results on its own is only of limited use. It might however gleam something but if not you get a lay of the land.
Comparing the relative performance of candidates is useful any interesting though it might not reveal anything you can be sure is fraud. People have already found in some data sets it appears Biden got lots of votes from people who apparently only filled that part of the ballot which if true on its own is already very suspect.
It's more useful to compare this data with 2016 and other datasets like populations for each region, registered voters, etc.
Most of the cheating I suspect though will be in cities or densely populated locations. In those it's a natural haystack you can bury your needle into.
Realistically they want to take a few thousand ballots at random for the state and try to track down the voter.
If I were making algorithms to cheat in the election they I would probably do something like take the top 50 closest states where Trump wins then flip them. Basically spread out the risk so it's hard to detect.
I believe a lot of the cheating however is not really grandoise. It's just things like optimising for mail-in ballots, fiddling the rules, etc.
If I was going to cheat in this election then I'd use multiple approaches including BLM. It's a criminal syndicate. Basically criminals, it's an anti-police movement so go figure. The criminal system has an underground network that can reach tens of millions or Americans or more. Compared to peddling crack cocaine a bit of ballot harvesting is a breeze and for someone who wants to buy the election it's far more cost effective than directly donating to the campaign to instead donate to organised crime. If organised crime can organise all these riots then having loads of ballots ready to drop late at night all for Biden is an absolute breeze.
I'd also look for USPS workers disgruntled about reforms to make it profitable and who want to get at Trump and who likely already to some degree have a network of employees or union or whatever.
There are two massive vulnerabilities they can very easily and subtly exploit. First is that they were allowed to collect after deadline knowing how much to make up for. Secondly is that they know mail-in goes more to Biden. The entire design of the election system and the rules changes for COVID-19 actually broke it. If they know after 12 mail-in ballots likely favour Biden then it's really easy to just optimise for that and accept as many as possible. They can also store them somewhere secret as a buffer, sorted buffer, sort them by region and proportion of democrats in said region then slowly release then in batches as needed.
This is actually just a standard CS practice. Buffering and priority queue.
What I could probably do is write a small simulation of an election system that can do things like this just to see what kinds of patterns I get.
I agree that this is one piece of a much larger puzzle. Let me know if you find any other additional concepts for how fraud might have surfaced at the precinct level data. The one that showed how absentee algorithm may have been based on Trump election day seemed promising
You can find 2016 info on MIT at https://electionlab.mit.edu/data