Claimed evidence of algorithmic vote flipping in Republican primaries

[Kevin_Lowe]

Warning: big pdf file.

Black Box Voting, a bunch of folks who have been on the trail of problems with the USA's electronic voting system for some time now, have published this big pdf file with graphs and such purportedly showing something very odd with the vote counting in certain Republican primaries.

To explain the purported oddity as simply as possible, imagine you get all the votes in an election, mix them up, and start counting them in a totally random order. The percentage of the vote that each candidate has can be expected to fluctuate wildly as you count the earliest votes, but by the time you have counted ~40% or so of the votes each candidate's percentage will have settled down into something very closely resembling the real, final percentage of the vote they have.

Unless two candidates are neck and neck, it's statistically very unlikely that whoever is winning at 40% of the count will not be winning when all the votes are counted.

What's weird is that in certain electorates, but not others, Mitt Romney's percentage of the vote increases smoothly as the votes were counted. Which is really weird. Assuming that everything was legitimate, it means that Romney votes, in some electorates but not others, knew to hide at the end of the queue.

That's not the end of the weirdness though. In those same electorates, it's claimed, some but not all of the other candidates' votes were all rushing to the front of the queue. So instead of the percentage of the vote each candidate had stabilising when around 40% of the votes were counted, Romney's vote percentage sailed grandly north while one and only one of the other candidates' percentage went south.

Particularly striking were the Richland, SC results where Gingrich and Santorum's vote percentage stabilised exactly as you would expect them to do, while Romney zoomed up the charts at the expense of Paul. The normal laws of statistics worked just fine in that electorate for two of the four candidates, and went totally wonky for two others.

The BBV hypothesis is that someone monkeyed with the vote counting machines to flip a certain small percentage of votes from other candidates to Romney. Whether that's right or not, I have to confess that I'm utterly baffled as to what non-suspect effect can cause a positive correlation consistently between number of votes counted and percentage of the vote going to Romney. It seems to be Romney benefiting in every instance, which is again a very difficult thing to explain by appeal to some hypothetical random factor.

There's a standard, self-sealing argument against belief in any kind of US vote tampering that goes like this: Nobody would tamper with the vote in the USA in any way that would leave evidence. We have evidence. Therefore no vote tampering happened. Hopefully we won't be seeing that argument in this thread.

That argument aside, does anyone see a reason I've missed as to what (other than some form of electoral fraud) could cause these numbers?

For the record, this thread is about evidence for or against electoral tampering and discussion of relevant statistical issues. Posts on other matters, including calls for the thread to be moved to the conspiracy theories section, are off-topic. You can expect such posts to be reported and I will be requesting that they be sent to AAH or split into a new thread.

Also for the record I have no particular opinion on the Republican primaries - all four candidates, to the limited extent that I have any knowledge of them at all, are very much alien to my own views, and I don't know enough to say whether any of them are more or less likely to win a subsequent Presidential election.

[geni]

The effect is consistent with the first few precincts (ie those that start with a) being very out of wack with the general voter trend. In that case what we are seeing is effectively a regression toward the mean.

Statisticaly this is entirely possible it simply requires that your early precincts be downright odd while all the others are pretty normal.

So if Aabach and Aardvark precient vote 90% Paul 5% Santorum 5% Gingrich and 0% Romney and everyone else votes 30% Paul 10% Santorum 10% Gingrich and 50% Romney then those graphs are pretty much what you will get.

I suspect the issue is that paul is very popular in some areas while Romney is consistently lukewarm popular everywhere.

[Puppycow]

Quote:

When tabulating the results of a ballot, at County or State level, the precincts are usually tallied by alphabetical order.

I thought that the precincts were tallied at the precinct level and then reported to a central authority as they were completed.

One explanation that occurs to me is that different candidates had different levels of support in different precincts, and in some cases those precincts supporting one candidate reported their results later than the others.

I'd like to see this "peer reviewed" by other independent experts before getting too concerned about it.

[Puppycow]

Why do they end the paper with a Bible quote?

Quote:

"Then you will know the truth and the truth shall set you free."

John 8:32

Doesn't exactly scream professionalism or academic rigor to me.
No footnotes, endnotes or any bibliographical support. I don't think a quality academic journal would publish this as is.

[Kevin_Lowe]

Originally Posted by geni

The effect is consistent with the first few precincts (ie those that start with a) being very out of wack with the general voter trend. In that case what we are seeing is effectively a regression toward the mean.

Statisticaly this is entirely possible it simply requires that your early precincts be downright odd while all the others are pretty normal.

So if Aabach and Aardvark precient vote 90% Paul 5% Santorum 5% Gingrich and 0% Romney and everyone else votes 30% Paul 10% Santorum 10% Gingrich and 50% Romney then those graphs are pretty much what you will get.

If that were the case, we would expect to see as many electorates where Romney starts poorly and rallies, as electorates where Romney starts out strong and then weakens. Spikes of Paul supporters should be just as likely at the end of the voting process as the beginning.

It's possible that there are indeed just as many such graphs, and BBV have cherry-picked the data to conceal those. However in the data they've presented, every single time Romney's line slopes upwards from left to right. So based on that data the odds of the graphs looking like that given the mechanism you propose would be one on 2¹¹.

Also in Richmond City, VA and Hillsborough, NH (2012) those lines are nearly straight. That looks a lot more like a linear correlation with number of votes counted, than it does with non-linear regression to the real mean which should have sorted itself out fairly quickly leaving a mostly-level line by the time the last votes are being counted.

I like your idea though, and it could turn out that BBV have indeed cherry-picked their data and with equal effort you could find eleven electorates where Romney has similarly smooth declines from left to right.

Quote:

I suspect the issue is that paul is very popular in some areas while Romney is consistently lukewarm popular everywhere.

There's the start of a potential explanation here, but unless I've missed something it relies on the assumption BBV have totally screwed the statistical pooch, and it doesn't currently explain the shape of some of the graphs very well.

ETA:

Originally Posted by Puppycow

Why do they end the paper with a Bible quote?

Doesn't exactly scream professionalism or academic rigor to me.
No footnotes, endnotes or any bibliographical support. I don't think a quality academic journal would publish this as is.

Indeed. These people aren't pros, more like enthusiastic amateurs, so it would be foolish to take their pronouncements as gospel. It's entirely possible they've made some rookie mistake and there's nothing amiss. Then again, the fact that they aren't up to academic standards isn't conclusive proof that they are wrong, either.

To my knowledge there's no body of professional statisticians or electoral scrutineers doing this kind of analysis of US voting patterns in a more rigorous fashion currently. So this might, for its sins, be the best information available.

[Puppycow]

Here is a longer version of the "layman's executive summary" PDF in the OP.

Right in the first paragraph I see a red flag. Can you tell what it is?

Quote:

This document is a summary of posts from members of The Ron Paul Forum investigating possible voter fraud in the 2012 GOP Primary. The threads were a work in progress and all postings were kept in order. The data shows so far:

[Kevin_Lowe]

Originally Posted by Puppycow

Here is a longer version of the "layman's executive summary" PDF in the OP.

Right in the first paragraph I see a red flag. Can you tell what it is?

Please bear in mind that the thread topic is evidence for or against electoral tampering, and discussion of relevant statistical issues.

If you have something to say which is on topic please just say it rather than playing at being Socrates.

[Puppycow]

Originally Posted by Kevin_Lowe

Please bear in mind that the thread topic is evidence for or against electoral tampering, and discussion of relevant statistical issues.

If you have something to say which is on topic please just say it rather than playing at being Socrates.

Well the fact that this is coming from Ron Paul supporters suggests a strong possibility that they already had a conclusion in mind and then went combing through the data looking for anomalies which might support their conclusion.

[ddt]

I don't buy the premise of ordering precincts by size. Isn't there a correlation of size with demographics? On the county level there certainly is, with the smaller ones being rural and the bigger ones being urban.

If you'd plot, say, the Dutch elections of 1977 ordered by municipality in that way, you'd start with a whole slew of agricultural communities with the christian-democrats at 80+% and Labour next to nothing, and end off with the big cities like Rotterdam and Amsterdam where the christian-democrats had nothing and Labour around 50% - and they'd both end up around the 30% mark.

[BStrong]

I took a quick look, but didn't see a mention of absentee ballot counting.

Absentee ballots are of the old paper type ballot, and the general consensus is that the absentee ballot voters tend to be more republican than democrat.

How this would effect vote tabulation, I don't know.

[WildCat]

Everyone knows Ron Paul has the support of 99.9% of all Americans. That he isn't President proves that there is vote fraud on a massive scale!

[Kevin_Lowe]

Originally Posted by Puppycow

Well the fact that this is coming from Ron Paul supporters suggests a strong possibility that they already had a conclusion in mind and then went combing through the data looking for anomalies which might support their conclusion.

Agreed. The problem is that Ron Paul supporters are also the logical people to notice if someone was flipping votes away from Ron Paul. So the fact that Ron Paul supporters are suggesting that Ron Paul is on the wrong end of some kind of election rigging is a fact that should not incline us to greatly change our prior belief in the likelihood of this kind of election rigging. Whatever that belief might be.

If there was rigging we'd expect them to notice, and if there was not rigging we'd expect some of them to see it where it didn't exist. It's a wash.

That's why I'm trying to keep the thread focused on the statistical evidence. That can at least potentially tell us something.

[marplots]

People who vote for Romney have jobs and people who vote for Paul do not. Skews the vote toward after work hours.

[Kevin_Lowe]

Originally Posted by marplots

People who vote for Romney have jobs and people who vote for Paul do not. Skews the vote toward after work hours.

The fact that in some electorates the lines converge on the true mean rapidly as you would expect, and in others the lines slope smoothly up or down as more votes are counted, is difficult to explain with an effect that does not have a logical reason to stop at an electoral boundary.

So (under your theory) in some electorates the Ron Paul supporters are all jobless layabouts who vote early, but in other electorates all the Ron Paul supporters are just as likely to have jobs as the non-Ron-Paul supporters. That seems weird. Why would the demographics of Ron Paul supporters change so much between electorates?

Tampering with the machines within an electoral boundary, whatever other problems it might have, is an explanation which does not have that problem.

(Also, and this should go without saying, if we think this explanation is worth taking seriously the first thing we should do is see if there is any hard evidence of higher unemployment rates amongst Ron Paul supporters).

[AvalonXQ]

Everybody else has already nailed it -- a flaw of this kind is indicative of some kind of non-random sorting in the vote-counting procedure.

So, if precincts are added to the count in the order they are received, what sort of correlation might there be between candidate preference and likelihood of swift reporting?

By the way, I found the OP deceptive (although I'm sure it was unintentional) when saying this:

Originally Posted by Kevin_Lowe

To explain the purported oddity as simply as possible, imagine you get all the votes in an election, mix them up, and start counting them in a totally random order.

For the entirety of the OP, I understood that this is what the OP was claiming had happened. In subsequent posts, it's clear that there's no randomizing procedure for vote order used before tabulating the votes.

[Puppycow]

Is "algorithmic vote flipping" something that can be done with paper ballots?

According to this site, Iowa and New Hampshire use paper ballots.

[Kevin_Lowe]

Originally Posted by AvalonXQ

Everybody else has already nailed it -- a flaw of this kind is indicative of some kind of non-random sorting in the vote-counting procedure.

So, if precincts are added to the count in the order they are received, what sort of correlation might there be between candidate preference and likelihood of swift reporting?

According to the file I linked, precincts are usually counted in alphabetical order. Now there may be some weaselling hiding in that "usually", because it doesn't specify how it works in unusual cases or how frequent those unusual cases are. However if precincts are tallied in alphabetical order that eliminates explanations like "small precincts were tallied first and this explains it" or "people who voted early were tallied first and this explains it".

Quote:

By the way, I found the OP deceptive (although I'm sure it was unintentional) when saying this:

For the entirety of the OP, I understood that this is what the OP was claiming had happened. In subsequent posts, it's clear that there's no randomizing procedure for vote order used before tabulating the votes.

I was trying to explain the issue without a huge wall of text, but I apologise if you feel that simplification was deceptive. However as stated above, the source claimed that precincts were tallied in alphabetical order, which is not a randomisation procedure as such, but which seems likely to have the same effect unless it's a thing in the USA to name precincts in strict alphabetical order from the RonPauliest precincts early in the alphabet through to the MittRomneyest precincts at the end of the alphabet.

How they could do that while keeping the Newty and Santorumy precincts distributed all nice and even isn't clear to me though.

[Puppycow]

Originally Posted by Kevin_Lowe

(Also, and this should go without saying, if we think this explanation is worth taking seriously the first thing we should do is see if there is any hard evidence of higher unemployment rates amongst Ron Paul supporters).

According to exit polls, the Ron Paul voters tend to be younger, and have lower incomes than the Romney voters. They are also more likely to be male. Many may be college students (hence concentrated in certain college towns). The unemployment rate is higher among young people.

ETA: However, I would expect the over-65 crowd to also be early voters, and those are Romney voters.

[Kevin_Lowe]

Originally Posted by Puppycow

Is "algorithmic vote flipping" something that can be done with paper ballots?

According to this site, Iowa and New Hampshire use paper ballots.

Can it be done with paper ballots? Well, that depends. If we are talking about hand-counted paper ballots, no, of course not. If we are talking about paper ballots which are counted by machines, I see no technical reason why algorithmic vote flipping would be impossible.

According to this search on that exact site lots of places in NH are using Diebold optical scan machines.

This google search turned up a lot of sources, most of them flaky, all claiming that 90% of votes in the NH would be counted by such machines. I haven't seen a reputable source repeating that figure but I'd be surprised if the claimed 90% was off by 90%.

What I would like to see from the BBV people is confirmation that the results they think are suspicious are all from machine-counted areas. If any of them come from hand-counted areas, I'd call that convincing evidence that whatever the cause of the oddity is, it's not anything to do with the voting machines. If all the odd results come from machine-counted areas, the electronic fraud hypothesis would still be in play.

[marplots]

I admit I didn't read the paper. However, if you want to see why the numbers run differently by precinct, you'd have to drill down to the specifics of the precincts. So, for example, if you see the pattern in Detroit but not in a suburb, it is rational to look at why one subset of voters might go to the polls in a different order in the large city as opposed to a suburb.

One explanation might very well be employment or access to the polls. And, this reasoning might be different in different precincts, so that there was no single overarching explanation for the anomalous event. I think this is what makes the voter fraud explanation both appealing and simultaneously suspect. It seems too easy to lump anomalies into the same basket with a single explanation, when there are so many other possible and perhaps even unique results.

Besides demographic issues, reporting about how the vote is going so far could affect the results. If only part of the voting population is paying close attention they may either react by going to vote when they otherwise wouldn't, or not voting when they otherwise would. That gives you the "news aware" tranche and the "not paying attention" tranche. I'm sure there are many others and any one could apply to any particular precinct.

My objection is that it might be misleading to think there is a single explanation or to seek that while excluding others. Voter fraud is one explanation, but it would require more evidence than the statistics alone.

It would be possible to detect it by contacting voters and polling them to see how they voted, then comparing what you got to the numbers reported. Not too hard to do in a smaller precinct, since you'd only need a statistically relevant sample. This would also present a much stronger form of evidence.

It is also misleading to think the patterns are attached to precincts instead of geographical or other factors. Precincts that have larger populations could wash out an effect that might appear in a smaller precinct and patterns extracted from the overall vote by precinct are bound to only show precinct variations. This separation by precinct forces any anomaly to be discrete at the precinct level, it will always and only yield a precinct-based mechanism, such as voter fraud.

[Kevin_Lowe]

Originally Posted by marplots

I admit I didn't read the paper. However, if you want to see why the numbers run differently by precinct, you'd have to drill down to the specifics of the precincts. So, for example, if you see the pattern in Detroit but not in a suburb, it is rational to look at why one subset of voters might go to the polls in a different order in the large city as opposed to a suburb.

One explanation might very well be employment or access to the polls. And, this reasoning might be different in different precincts, so that there was no single overarching explanation for the anomalous event. I think this is what makes the voter fraud explanation both appealing and simultaneously suspect. It seems too easy to lump anomalies into the same basket with a single explanation, when there are so many other possible and perhaps even unique results.

Besides demographic issues, reporting about how the vote is going so far could affect the results. If only part of the voting population is paying close attention they may either react by going to vote when they otherwise wouldn't, or not voting when they otherwise would. That gives you the "news aware" tranche and the "not paying attention" tranche. I'm sure there are many others and any one could apply to any particular precinct.

My objection is that it might be misleading to think there is a single explanation or to seek that while excluding others.

I agree with you that simple, dramatic explanations are tempting and that boring, complicated explanations are not. That's almost certainly a large part of the reason why conspiracy theories exist. However "lots of things did it and it's complicated" is a general-purpose explanation that suits any oddity in any data set. Unless you can point to the things in question and explain how they manage to make people in the precincts with names late in the alphabet more likely to vote for Romney it's not an explanation in any meaningful sense.

That doesn't mean we should do a cursory search for innocent explanations, and then if we can't think of any fix the conclusion that it must have been fraud. However we also shouldn't accept the first thing that looks like an explanation without checking to see if it actually explains the shapes we are seeing in the data either.

Quote:

Voter fraud is one explanation, but it would require more evidence than the statistics alone.

It would be possible to detect it by contacting voters and polling them to see how they voted, then comparing what you got to the numbers reported. Not too hard to do in a smaller precinct, since you'd only need a statistically relevant sample. This would also present a much stronger form of evidence.

Better still would be a hand recount of the ballots in the suspect precincts. Hopefully without a repeat of the incident, in Ohio I think, where the people who were to do a recount in a suspect precinct arrived in the morning to find that the electoral workers had just pulled an all-nighter doing their own recount "to make sure it all added up right".

However a poll such as the one you propose could also provide strong evidence of an issue if indeed one exists.

[Skeptic Ginger]

OK, help me out here, am I reading this correctly?

In the first graph the vote count is alphabetical by county and in the graph on page 3 showing the anomaly the count is by numbers of votes in each the county, is that right?

Wouldn't the variable, county size, need to be ruled out in order to rule in voting machine tampering? You can't rule that variable out simply showing other candidates don't have that effect. You might look at polls before the voting to see if county size is a consistently correlating variable.

The data is worth investigating. I don't care it was brought up by Ron Paul supporters. There were other events noted in the news that would give them cause to question possible interference in the primary outcome, and there was evidence of potential tampering with the voting machines in Ohio in 2004, whether you think that evidence was weak or strong.

But unless other variables are ruled out, you have cherry picking the order of vote tabulating.

[WildCat]

Originally Posted by Kevin_Lowe

Agreed. The problem is that Ron Paul supporters are also the logical people to notice if someone was flipping votes away from Ron Paul.

THey also think Jews international bankers run the Fed with the aim to enslave us all, the US government did 9/11 and blamed it on Arab cavemen to steal Iraq's oil, and Oswald did not kill JFK. Just ask them!

Ron Paul supporters allege conspiracy, news at 10. Right after the "dog bites man" story.

[Skeptic Ginger]

There is a lot of variation in candidate support between counties: How NH Counties Voted (Based On Economic Drivers)

Quote:

To see statewide totals, move your cursor outside of New Hampshire. To get more details on how WNYC/Patchwork Nation determined the main economic driver of a county, just click on the colored icon in the map key. You’ll see a nifty pop-up including census data, poll numbers, and other info.

It looks like there were distinct differences in the populations by county including greater support for RP in more rural (i.e. smaller population) counties.

Quote:

Merrimack, Coos, and Sullivan County make up this economic category, and account for 17 percent of the primary electorate. This particular group of counties is characterized more by small and mid-sized towns which depend on service industries like hospitality, restaurants, and retail.

I'll try to look a little closer later if someone doesn't beat me to it.

[casebro]

And at worst we have statistical clusters, which are absolutely normal.

Hey, flip a coin 100 times, you'll probably have 50/50. But start to data mine, and you will easily spot some 10-in-a-rows. Now arrange a bunch of sets of 100, by where the ten-in-a-rows happen- beginning, middle ,or end or the 100s. That seems to what Paul's guys have data mined for.

Nothing to see here, unless we want to discuss the motivation behind the data mining.

[Aepervius]

Originally Posted by Puppycow

Well the fact that this is coming from Ron Paul supporters suggests a strong possibility that they already had a conclusion in mind and then went combing through the data looking for anomalies which might support their conclusion.

Isn't that identical to the argument from authority in reverse ?
In other word, the source is bad (or simply considered biased) therefore the data is suspect and their argument wrong ?
FYI I don't know either way, I am just very warry of qualifying an argument as red flag because of the source.

Heck my first idea was "ron paul supporter rally early and vote early, whereas traditional old people rally for romney and later in some circumscription".

[Puppycow]

Originally Posted by Aepervius

Isn't that identical to the argument from authority in reverse ?
In other word, the source is bad (or simply considered biased) therefore the data is suspect and their argument wrong ?
FYI I don't know either way, I am just very warry of qualifying an argument as red flag because of the source.

In a court of law, when an expert witness comes in to testify about something which may be outside the comprehension of the laypeople on the jury, his qualifications and possible conflicts of interest or ideological motivations are fair game to be brought up by the opposing attorney. I'm not saying the fact that they are Ron Paul supporters automatically makes what they are saying wrong, but it is one piece of evidence to be considered.

[Puppycow]

I suspect that the answer lies somewhere in the demographic peculiarities which show up quite clearly in the exit polls I mentioned earlier. There are huge demographic differences between Ron Paul supporters and Mitt Romney supporters.

Age:

In which age group are you?
Total Paul Gingrich Romney Perry Santorum Huntsman
18-29
12% 46% 3% 26% - 9% 13%
30-44
19% 32% 9% 37% 1% 6% 14%
45-64
48% 19% 9% 42% - 10% 18%
65 or over
21% 12% 14% 42% 1% 10% 20%

Marital status:

Are you currently married?

Total Paul Gingrich Romney Perry Santorum Huntsman

Yes

69% 19% 9% 42% - 11% 17%

No

31% 34% 6% 35% - 5% 16%

Income:

2011 total family income:
Total Paul Gingrich Romney Perry Santorum Huntsman
Under $30,000
11% 35% 11% 31% - 9% 11%
$30,000 - $49,999
15% 28% 11% 31% 1% 11% 16%
$50,000 - $99,999
37% 22% 10% 35% 1% 11% 19%
$100,000 - $199,999
27% 20% 8% 47% 1% 7% 17%
$200,000 or more
10% 12% 10% 52% - 7% 18%

History of voting in Republican primaries:

Before today, had you ever voted in a Republican presidential primary?

Total Paul Gingrich Romney Perry Santorum Huntsman

Yes

86% 20% 10% 43% - 10% 15%

No

12% 38% 3% 24% - 8% 21%

Somewhere in there, I'll bet, lies the explanation. It could be that young people are concentrated in particular voting precincts. And/or that people with high incomes are concentrated in other particular voting precincts. And/or that single people and married people are concentrated in different particular voting precincts.

[Kevin_Lowe]

Originally Posted by Skeptic Ginger

OK, help me out here, am I reading this correctly?

In the first graph the vote count is alphabetical by county and in the graph on page 3 showing the anomaly the count is by numbers of votes in each the county, is that right?

I'm not quite sure what "numbers of votes in each the county" means to you, but the x axis is showing the percentage of the total vote counted, with the votes being counted precinct-by-precinct in alphabetical order. Or so it is claimed, anyway.

Quote:

Wouldn't the variable, county size, need to be ruled out in order to rule in voting machine tampering?

I don't think county size is relevant, unless county size goes in alphabetical order.

Originally Posted by casebro

And at worst we have statistical clusters, which are absolutely normal.

Hey, flip a coin 100 times, you'll probably have 50/50. But start to data mine, and you will easily spot some 10-in-a-rows. Now arrange a bunch of sets of 100, by where the ten-in-a-rows happen- beginning, middle ,or end or the 100s. That seems to what Paul's guys have data mined for.

Nothing to see here, unless we want to discuss the motivation behind the data mining.

This is a general argument which can be directed at any data set. As such it means nothing.

Do you have anything to say which specifically pertains to the thread topic? This story of yours doesn't explain how you can get a straight, tilted line out of counting a bunch of votes in alphabetical order by precinct.

However you slice it, the graphs for Polk and New Hampshire are weird.

[geni]

Originally Posted by Kevin_Lowe

If that were the case, we would expect to see as many electorates where Romney starts poorly and rallies, as electorates where Romney starts out strong and then weakens. Spikes of Paul supporters should be just as likely at the end of the voting process as the beginning.

Yes but if they happen at the end what you would see would be a pretty much straight line with an almost too small to see blip at the end.

Quote:

To my knowledge there's no body of professional statisticians or electoral scrutineers doing this kind of analysis of US voting patterns in a more rigorous fashion currently. So this might, for its sins, be the best information available.

There are quite a number however they don't publish their results since they work for private polling organisations, political parties and campain groups.

[Skeptic Ginger]

Originally Posted by Kevin_Lowe

I'm not quite sure what "numbers of votes in each the county" means to you, but the x axis is showing the percentage of the total vote counted, with the votes being counted precinct-by-precinct in alphabetical order. Or so it is claimed, anyway

I don't think county size is relevant, unless county size goes in alphabetical order.

So what am I missing then? The first graph looks like a normal regression toward the mean and the graph on page 3 is different. Are they not showing the same vote count just recorded in a different order?

[Roger Ramjets]

Originally Posted by geni

The effect is consistent with the first few precincts (ie those that start with a) being very out of wack with the general voter trend. In that case what we are seeing is effectively a regression toward the mean.

In some cases this 'regression to the mean' is quite quite striking. For example in Richmond, Vermont, the votes start out 100% Paul, 0% Romney, linearly regressing to ~50% each by the end.

There are two reasons I can think for this apparent anomaly:-

1. Massive election fraud! Not just a small percentage of votes being flipped, but whole ballot boxes being stuffed - with votes for Ron Paul. How else can we explain the incredible 80%+ lead that he had at the start?

2. Ron Paul voters have weird voting habits. We already know that Ron Paul voters are weird, so this is the most likely explanation.

Or it could all just be normal statistical noise. Without corroborating evidence there is no way to prove that it was election fraud rather than some other innocuous effect. So what about exit polls? Did they indicate that Ron Paul should have won by a large margin?

[Rolfe]

I find this sort of statistical analysis fascinating. (Geek confession: As a teenager I once figured out how many episodes of Classic Star Trek had been made by graphing the overlap between episodes shown by the BBC - apparently in random order - and episodes included in the James Blish novelisations which were being published at the time, also with an apparently random episode order. It was obvious where the graph was converging at a pretty early stage, and it was right.)

I do, however, recall a screen shot from BBC election night coverage of a UK general election. At an early stage in the proceedings the SNP had more seats than the Conservatives. ALL THREE SNP seats had declared very early for some reason I don't know, and only two of the hundreds of Conservative seats had declared by then.

Rolfe.

[The Central Scrutinizer]

Originally Posted by Kevin_Lowe

Warning: big pdf file.

Black Box Voting...

I stopped reading here. I would suggest others do the same. Folks, this is a bunch of loonies who just can't get over the fact that W won the election in 2004, and who don't understand that opinion/exit polls and election results are not the same thing.

Mods, please move this to the Conspiracy section ================>

[Kevin_Lowe]

At this stage none of the offered explanations explain the shape of the problematic graphs satisfactorily.

If the graphs approached the final values asymptotically, then the explanation that these are just electorates where a precinct starting with "a" went overwhelmingly to Paul and the rest went overwhelmingly to Romney would fit. However some of the graphs just don't look like that at all. The line is straight, which is really hard to explain.

To explain this neatly, there needs to be some effect which is proportional to the alphabetical position of the precinct. The only possibility I can think of is that someone decided to name the precincts in alphabetical order east/west or north/south or similar, so that geographical position correlated tightly with alphabetical position, and that geographical position also correlates tightly with the proportions of Paul and Romney supporters.

However that still leaves the total failure of the anomaly to affect Gingrich and Santorum. Even if we accept that Paul support is going to be highly patchy for whatever reason, we need some explanation for why 100% of the people not voting for Paul because of this effect went to vote for Romney instead.

Is there some clear political reason why this might be the case? In Australia, for example, it would be unsurprising if people defecting from the Greens went to the Labor party, since it's further to the left than the other major party. Is Romney clearly the most Paul-like of the other candidates, so that it would be incredibly unlikely that a Paul defector would ever vote for Gingrich or Santorum? I don't know enough about the candidates' policies and appeal to answer this one myself.

If it turns out that precincts are indeed named in an order such that alphabetical order goes with geographical order, and Romney is the only logical candidate a Paul-defector would support, that might just explain things. However if either of those guesses is wrong I don't think any of the proposed explanations fit the data.

ETA: This nifty interactive map makes it looks like there's no correlation between location and precinct name in New Hampshire, at least. Scrubbing over it at random the names appear to be all over the alphabet with no particular correlation between position and name.

[marplots]

I'm confused. You mention alphabetical order, but the paper mostly talks about precinct size. For example, from pg 5:

Quote:

"Keep in mind that many of those voters are the very same people. It is difficult to imagine demographics that could explain the sudden appearance of a massive positive correlation between Romney's results and precinct size in only four years. Ditto for Paul's..."

But that's exactly what we've been trying to give you -- odd reasons why it might matter.

My question is why the alphabetical order anomaly, which I think is much more interesting, is dropped in favor of "tally counts" and "precinct size" later on. This actually argues against an algorithm based on alphabetical order, doesn't it?

Still, to make the case, I suggest they poll voters and see if the numbers match those given. Seems like that would be the gold standard.

[Kevin_Lowe]

Aha! Progress, although I feel a bit stupid. Then again that's the usual result of investigating any anomaly - most anomalies turn out to be boring.

The vital bit in the original document is when they say "Now let's order the precincts by number of votes cast and let's start counting from the smallest all the way to the
largest". So they aren't ordering by alphabet at all - that was a misreading on my part.

Since they've ordered by precinct size that makes things potentially a lot less mysterious. Alphabetical order shouldn't correlate with anything at all. Precinct size could logically correlate with all sorts of things like being rural, being poor and so on. So while there didn't seem any prospect at all of explaining things if we were counting in alphabetical order, there's at least a possibility of explaining things if we're counting in ascending order of precinct size.

However this doesn't magically solve the problems. The graphs from previous elections, ordered the same way, show that there can be wild fluctuations in the early count which seem highly likely to be due to small-precinct voters being different to big-precinct voters. See the Alachua, FL graphs, for example. However these fluctuations sort themselves out completely before even 20% of the vote is counted, and the lines are mostly flat from there. So these results are still weird.

Also, although they don't show graphs to go with the claim, the article claims that the weirdness is not exclusively a Romney vs Paul weirdness:

Quoting from the linked article:

"The anomaly adjusts to Romney’s political agenda for a given state election: it takes from Paul in Iowa and New Hampshire, Gingrich in Florida and South Carolina, Santorum in Arizona and Ohio. Romney’s surge leaves only Santorum totally unaffected in South Carolina, whereas Paul is the only one untouched in Alabama."

That does make life more complicated. Up until now people have put forward a lot of suggestions about how the demographics of Paul supporters could explain the anomaly (and pointed out that it was Paul supporters claiming that there was a problem). However we have to chuck all those suggestions out, if in some precincts Paul was totally unaffected by this effect while Gingrich and Santorum were hit my the mystery effect transferring votes to Romney in the bigger electorates.

So we need an effect which correlates very strongly with the size of an electoral precinct, but which is independent of whichever of Romney's opponents happens to suffer in any given race, because the opponent suffering the loss seems to vary while the effect remains similar.

[Aepervius]

Originally Posted by Roger Ramjets

In some cases this 'regression to the mean' is quite quite striking. For example in Richmond, Vermont, the votes start out 100% Paul, 0% Romney, linearly regressing to ~50% each by the end.

There are two reasons I can think for this apparent anomaly:-

1. Massive election fraud! Not just a small percentage of votes being flipped, but whole ballot boxes being stuffed - with votes for Ron Paul. How else can we explain the incredible 80%+ lead that he had at the start?

2. Ron Paul voters have weird voting habits. We already know that Ron Paul voters are weird, so this is the most likely explanation.

Or it could all just be normal statistical noise. Without corroborating evidence there is no way to prove that it was election fraud rather than some other innocuous effect. So what about exit polls? Did they indicate that Ron Paul should have won by a large margin?

Just out of curiosity I took the picture you quoted , approximated to make it easier for me to check.




Then I used an intial tally of 4000 votes, with only 2 candidates, and thought, OK how many voted for romney by that point or for Paul ? As you can see in the last column:

	Romn	Paul	Vote Number	Voted Romn	Voted Paul	Delta Romn	Delta Paul
0	10	90	0	X	X		
5	12,25	87,75	200	25	176		
10	14,5	85,5	400	58	342	34	167
15	16,75	83,25	600	101	500	43	158
20	19	81	800	152	648	52	149
25	21,25	78,75	1000	213	788	61	140
30	23,5	76,5	1200	282	918	70	131
35	25,75	74,25	1400	361	1040	79	122
40	28	72	1600	448	1152	88	113
45	30,25	69,75	1800	545	1256	97	104
50	32,5	67,5	2000	650	1350	106	95
55	34,75	65,25	2200	765	1436	115	86
60	37	63	2400	888	1512	124	77
65	39,25	60,75	2600	1021	1580	133	68
70	41,5	58,5	2800	1162	1638	142	59
75	43,75	56,25	3000	1313	1688	151	50
80	46	54	3200	1472	1728	160	41
85	48,25	51,75	3400	1641	1760	169	32
90	50,5	49,5	3600	1818	1782	178	23
95	52,75	47,25	3800	2005	1796	187	14
100	55	45	4000	2200	1800	196	5

The result unsurprisingly is this , for every 5% time slice of the tally , more and more people come to vote for Romney, while less and less people vote for Paul :


So where I am stuck is that I see two way to interpret this. The simplest and most valid interpretation is that Ron Paul demographic was "vote early, vote in mass" whereas Romney demographic simply came late massively. My gut feeling would be that this would be consistent to what I observed in voting bureau, with young people coming early, and old people coming late in the afternoon massively.

But i am speculating here. As this was anecdotal observation while being voted booth observant/counter , not really noting the vote of people per hour.

[Kevin_Lowe]

Originally Posted by Aepervius

So where I am stuck is that I see two way to interpret this. The simplest and most valid interpretation is that Ron Paul demographic was "vote early, vote in mass" whereas Romney demographic simply came late massively. My gut feeling would be that this would be consistent to what I observed in voting bureau, with young people coming early, and old people coming late in the afternoon massively.

Unless I've misunderstood what's going on, the order in which votes are counted in the graphs in the article has nothing whatsoever with the order in which votes were cast.

All the votes in the smallest precinct are counted, then all the votes in the next precinct smallest, and so on. As I understand it. So the Paul demographic in those places (and the Gingrich and Santorum demographics in other electorates) would have to be a "vote in small precincts" demographic, not a "vote early" demographic.

[Beerina]

I don't understand. Are they claiming actual vote progression should match as if you jumbled the votes?

That's completely wrong. Different people vote in the morning and during the day than do after work. Also, people can panic, get despaired, or want to jump on the winning side as the day progresses, which can also make the distribution distort as the day marches on.