The case for the occurence of algorithmic vote flipping

[arsenius]

One more note on the precincts in Dane county. The average precinct size is 529 voters, but in Madison the average is 412 voters. Assuming that trend holds true in other counties, I hope this lays to rest the argument that this is a rural vs. urban issue.


Data here: http://www.countyofdane.com/clerk/elect2012b.html

 

[arsenius]

I emailed all of them! You can get a list of everyone here:

http://gab.wi.gov/clerks/directory

However, not a SINGLE person responded to the summary. They only responded to some questions about their testing procedures.

Dr. K, I'm going to be calling the Dane county clerk later today. I'd like to chat with you before, if you have time. I'm living overseas, so earlier would be better, the next few hours if possible. PM me your number if that works for you.

 

[RonRules]

To celebrate finally being able to upload a new avatar, here's the associated music for your enjoyment and to calm everybody down:
http://www.televisiontunes.com/Flipper.html

For those of you from North Korea who have no idea what I'm talking about:
http://en.wikipedia.org/wiki/Flipper_(1964_TV_series)
http://bussongs.com/songs/flipper.php

 

[RonRules]

One more note on the precincts in Dane county. The average precinct size is 529 voters, but in Madison the average is 412 voters. Assuming that trend holds true in other counties, I hope this lays to rest the argument that this is a rural vs. urban issue.


Data here: http://www.countyofdane.com/clerk/elect2012b.html

Dane county is clearly a flipper. I know this stuff:

 

[dsw]

Dane county is clearly a flipper. I know this stuff:

How much slope does there have to be to make it "clearly a flipper"?

Here's Ron Paul in Dane county 2008 from http://www.countyofdane.com/clerk/elect2008a.html . With 20% of the vote counted he has 10.1%, but by the end he has only 6.6%. (McCain benefitted from this, gaining 4% over the same range.) For that sample size (calculated as if it were a random sample) the margin of error would be 1.23%.

What's the verdict? Fraud, noise, or demographics? And if it's not fraud, is there an objective way to tell?

 

[RonRules]

How much slope does there have to be to make it "clearly a flipper"?

Here's what I confirmed yesterday by a university Quantitative Analyst (a Quant):

"With a sufficiently sized county, those lines MUST be horizontal".

There can be little wiggles, but the average slope, past 20% data or so (depends on the sample size), MUST be zero.

So, ANY slope on a county like Dane, with 488,073 people (167,000 voters) indicates a problem.

In the 2008 chart you posted, please remember and go back to my Ohio charts, McCain was clearly benefiting from flipping. Earlier in 2008, Romney was the flipper, but when he pulled out, McCain became the benefactor.

Please remember that our charts are cumulative. The last few points on the right include ALL the demographics. It is practically impossible that a constant, high correlation line would result due to demographics.

 

[RonRules]

Here's an example of a county that probably not a flipper, but exhibits some small bumps.
Those little bumps come out because the data size is smaller.

 

[dsw]

Here's what I confirmed yesterday by a university Quantitative Analyst (a Quant):

"With a sufficiently sized county, those lines MUST be horizontal".

There can be little wiggles, but the average slope, past 20% data or so (depends on the sample size), MUST be zero.

Wow! Nothing ambiguous about that. I would have said that it must be horizontal unless there's a correlation with the total number of votes at a precinct for some reason, but I guess that's why I'm not a Quantitative Analyst!

Check out the one below. It's from the same Dane County web site. If I'm getting the hang of reading these things, the vote flipping theory would say that flipping began where it suddenly starts taking a sharp dive down on the red line, shortly before 20k votes. But then a little way past 120,000 votes, it finally flattens out, so did they give up on flipping at that point when they realized they couldn't drive that green line up enough to win? Before leveling out that green line climbs from 35.6% (with 20% counted) to 43.8%. Clear fraud? I've got more examples too.

 

[arsenius]

Just got off the phone with a very stressed Dane County clerk. It was an interesting conversation. She was very upset because she gets a lot of these types of phone calls, but did not believe there could be anything going on. She said she is not a math person, and was not interested in hearing any numbers. When I said there is a 99% correlation between precinct size and Romney's count, she said she doesn't know how people vote, but that people are acting differently in this election because things are so polarized. She insisted that the vote be certified, because she was certain there was nothing wrong with it.

Apparently they are using very old machines, 20 or 25 years old. All the machines for the county are programmed by the deputy county clerk, in a room that only two (or three?) people have access to. She said that in Wisconsin, there are two other counties (she thought they were Rock and LaCrosse) that program their own machines. They are serviced by the company that made them once a year, but apparently this does not include software updates. It's more of wear and tear type maintenance on rollers and such.

Anyways, I have to say that I am convinced of her sincerity. She really did not believe anything could be going on, and if what she said is true, then it would be hard to pull something off there.

Here is LaCrosse County. Santorum looks relatively flat, with Paul going down. You know which way Romney goes.

 

[RonRules]

Wow! Nothing ambiguous about that. I would have said that it must be horizontal unless there's a correlation with the total number of votes at a precinct for some reason, but I guess that's why I'm not a Quantitative Analyst!

No, I'm sorry, but you still don't understand our charts. The accumulation of votes as you go from left to right does not show a SINGLE precinct anywhere except the very FIRST point on the left. The single point on the left is the smallest precinct, a trivial contribution to the chart. EVERY other point on the chart includes more than one precinct and the last point on the chart includes ALL precincts.

Please try real hard to understand that. I want to make you a believer because you can then take over and answer posts from other people like you.

 

[RonRules]

Just got off the phone with a very stressed Dane County clerk.
Apparently they are using very old machines, 20 or 25 years old. All the machines for the county are programmed by the deputy county clerk, in a room that only two (or three?) people have access to. She said that in Wisconsin, there are two other counties (she thought they were Rock and LaCrosse) that program their own machines. They are serviced by the company that made them once a year, but apparently this does not include software updates. It's more of wear and tear type maintenance on rollers and such.

A couple of things. The individual voting machines are not "programmed" by the staff. They just enter the names of the candidates in the previously written "program"

But that's all irrelevant. It's not about the precinct machines. I am convinced those are fine.

It's ALL about the central tabulator. You could vote with stone chisels at the precinct, it won't matter. It's the central tabulator that flips the votes. It does so when the precinct size is about 250 and does so linearly as a function of cumulative precinct size.

 

[dsw]

No, I'm sorry, but you still don't understand our charts. The accumulation of votes as you go from left to right does not show a SINGLE precinct anywhere except the very FIRST point on the left. The single point on the left is the smallest precinct, a trivial contribution to the chart. EVERY other point on the chart includes more than one precinct and the last point on the chart includes ALL precincts.

Please try real hard to understand that. I want to make you a believer because you can then take over and answer posts from other people like you.

Well of course the last point includes all precincts. I don't see how that explains why there can't be a naturally-occurring correlation between %vote and precinct size, i.e., the precincts that show up near the right side of one of these graphs would not have the same %vote that the candidate got overall. But I'm not a Quantitative Analyst and there's more math that I don't understand than math I do understand! So maybe if I keep looking at these kind of graphs it will sink in.

Anyway, more from Dane. What's the story on one like this? They waited until almost 60% of the vote has been counted and THEN start flipping? Why would they do that? Does that make any sense at all?? I mean, by the time you've got 60% counted what are the odds of your opponent dropping from 27% to 23% over the remaining precincts? Seriously, how would you calculate the odds of something like that?

 

[RonRules]

Let me explain Dane county. I predict that Dane will be the center of all the attention, for several reasons.

To answer your question:
First: Check your chart, it doesn't look right. For Wisconsin, forget Ron Paul. The action is between Romney and Santorum. Please chart those and for consistency in the future, use the following colors: Romney (ugly green); Santorum (Dark Blue); Ron Paul (Red); Gingrich (Indigo or Turquoise, if you care to chart him)

Second:
a) There are 193 counties in Dane, and the average size in 109 and the median size is 88 votes. No flipping will occur with less than ~250 votes per precinct. (I need to do stats to find out exactly where the flipping starts)
b) There are only 14 precincts with votes > 250.

That's why the flipping does not always start at the same cumulative count on the X-Axis. If all the precincts were <250 votes (approximately) there would be no flipping at all. You have to look at the statistical distribution of the precinct counts.

Dane has some very odd irregularities that I am investigating. I'm looking at Madison, where Ron Paul did great. In one precinct 6 times better than Romney. This of course shoots down the theory that "Romney does better in big cities".

Next about Dane county, it is one of those counties, which uses a central tabulator by ES&S. This information will come handy later, I promise.

 

[drummergirl]

Anyway, more from Dane. What's the story on one like this? They waited until almost 60% of the vote has been counted and THEN start flipping? Why would they do that? Does that make any sense at all?? I mean, by the time you've got 60% counted what are the odds of your opponent dropping from 27% to 23% over the remaining precincts? Seriously, how would you calculate the odds of something like that?

Good question. The simplest way is to chart the data by deciles and use the standard deviation and z statistic (tables or with software like excel). This one is from Iowa.

So, here if we look at Romney going from 18.2% with 30% of votes counted to 23.7% at the end and we are generous and use the 0.5% standard deviation, that 's a change of 11 standard deviations (z=11.0)

As to why start when 60% is already counted? They are trying to hide.

Think of a couple of kids splitting up the candy they took in on halloween:

One for you, one for me...then (after little sister starts falling asleep), one for you, two for me, one for you, three for me...

*And if my sister is reading this thread, the example is purely hypothetical and never actually happened.

 

[dsw]

First: Check your chart, it doesn't look right.

Doesn't look right in what sense? I very well might have made a mistake, but those weren't from the 2012 presidential election. I wrote a little program to parse all of the election data on the site (easy since they always used the same format). Then generated the .csv files for cumulative graphs but rather than graphing immediately, I had the program compare the 20% value to the 100% value and flag the ones that had the largest change, because of course without any fraud they absolutely MUST have zero slope according to the Quantitative Analyst.

The first one I posted was an RKBA referendum from 1998. (If they were flipping, it was an attack on the second amendment but they failed.) The other was 2002 race to elect a sheriff. The math should still work though, right? Is there some reason that the slope MUST be zero (past a certain point) in presidential elections, but not when people are voting on other things? Because if the math should still work I've got more to share from 1998 to 2004 ballots, for Dane county races that have a large enough number of votes cast.

 

[RonRules]

Something else to be on the lookout for.

I charted the city of Madison only (I used all the precincts with the name "Madison WD (Ward) nn).

Ron Paul did great in Madison and he got flipped really hard by Romney. Paul got flipped so hard that you barely see any flipping on Santorum in comparison.

Here's the new thing to be on the lookout when you want to see filipping evidence:

Notice the slow exponential downward curve at the beginning of the chart on Ron Paul's line. That's an additional flipper identifier.

Now, you all tell me why I said that.

All the lines should have an early little wiggle and straighten horizontally pretty fast, just like my favorite county in Wisconsin: Outagamie.

BTW, what's there not to like about Outagamie, Wisconsin. Check out their website: http://www.outagamie.org/

 

[drummergirl]

Just got off the phone with a very stressed Dane County clerk.

Anyways, I have to say that I am convinced of her sincerity. She really did not believe anything could be going on, and if what she said is true, then it would be hard to pull something off there.

That is one of the reasons all of this is so insidious. It's being done in a way that is specifically designed to avoid detection. (Well, most thieves want to avoid detection, but few are good enough at it to succeed.)

With the exception of a few bad apples (like that fella in Louisiana who copped a plea in the 90s), election administrators are as serious as serious can be about maintaining the integrity of their elections.

The trouble is, historically speaking, there have been 3 ways to rig an election. 1) stuff the ballot box. 2) bribe or strong arm voters. 3)pad the voter registration rolls (with dead people, those who've moved, etc) so your people can vote more than once.

There are numerous protections in place to prevent these from happening and to catch them quickly when they do.

The flippers have found a new way to steal votes; the vote totals remain the same (so they pass the canvass) and they don't mess with small vote total precincts where a change would be really noticeable. They take a few votes here and there from the larger precincts; in a close race a little change is a big difference.

 

[RonRules]

Doesn't look right in what sense? I very well might have made a mistake, but those weren't from the 2012 presidential election. I wrote a little program to parse all of the election data on the site (easy since they always used the same format). Then generated the .csv files for cumulative graphs but rather than graphing immediately, I had the program compare the 20% value to the 100% value and flag the ones that had the largest change, because of course without any fraud they absolutely MUST have zero slope according to the Quantitative Analyst.

The first one I posted was an RKBA referendum from 1998. (If they were flipping, it was an attack on the second amendment but they failed.) The other was 2002 race to elect a sheriff. The math should still work though, right? Is there some reason that the slope MUST be zero (past a certain point) in presidential elections, but not when people are voting on other things? Because if the math should still work I've got more to share from 1998 to 2004 ballots, for Dane county races that have a large enough number of votes cast.

I didn't recognize the shapes because they were not 2012. By now, I think you can just show me the chart and I'll tell you what county it's from (in Wisconsin, and most of VA).

" because of course without any fraud they absolutely MUST have zero slope according to the Quantitative Analyst. " Yes.

A race for Sheriff, county wide will probably have pretty small precinct vote totals. If they're over 250 each and you ignore the little bumps on the road, it should be a horizontal line.

Note that in Ohio in 2008, I show a chart of a Republican Attorney general race that flips a Democratic attorney's race. It's a GOP trick, not necessarily a Romney thing. McCain used it quite a bit in 2008, after Romney quit.

If your Sheriff is a strong Republican, he may flip up. Check Joe Arpaio's last election. That would be fun to look at.

 

[dsw]

A race for Sheriff, county wide will probably have pretty small precinct vote totals. If they're over 250 each and you ignore the little bumps on the road, it should be a horizontal line.

I'm going to go back and double-check everything, now that I have confirmation that it *should* be flat if there's no fraud (according to flipping theory, I mean, not according to math as I understand it). But for that sheriff race, what I have is 192 precincts with only about 35 or so under 250, and 144k total votes.
http://www.countyofdane.com/clerk/electres/elect2002d.html
Sheriff, Hamblin vs Benedetto, near the bottom of that page.

 

[dr.k.research]

Dr. K, I'm going to be calling the Dane county clerk later today. I'd like to chat with you before, if you have time. I'm living overseas, so earlier would be better, the next few hours if possible. PM me your number if that works for you.

OK. As early as you want. Send me a TM. And your number.