The case for the occurence of algorithmic vote flipping

[drummergirl]

MODERATOR NOTE: This is a contentious subject that members of this forum have expressed strong disagreement on. A number of members do not think there is any evidence of vote flipping, but in the interest of avoiding flame wars and derailments to the work that is being done here this thread is to be kept free of attacks on this project. To view arguments making the case against the occurrence of vote flipping or to post arguments yourself go here or in other previous threads making the case against vote flipping such as here or here.


-------------------------------------------------------------------------------------------------------


Here is a summary, written with the non-math oriented in mind, of the analysis so far.

https://docs.google.com/document/d/1XYSqtgWZj91085OGxjTvXYZ2u5NL-3a4pko7vMLTBVE/edit?pli=1

Since the original threads are now over 4000 posts, there is a TON of more data that's been analyzed since super Tuesday, and considering the time critical nature of the information, I thought it would be a good idea to start a new thread.

If your state primary is yet to be conducted, please bring this to the attention of your election administrators so they have a chance to prevent fraud in their elections.

If your state primary is past, please gather up your data and head to your attorney general.

For anyone who likes to be thorough, the original thread is most easily read by the first post and then backwards from the most recent

http://www.ronpaulforums.com/showthread.php?363915

 

[economics102]

+rep! Keep pushing, this is vital work!

 

[drummergirl]

thanks; next project is a technical summary with lots of math for those who want data.

 

[RonRules]

Please keep this new thread as clean as possible, with productive posts (Pro or con). If your pet argument has been shot down, please don't self-bump. Keep praise to a minimum, but please criticize in private.

Make every post about new information that has not been provided before. Learn to make charts through Excel or the new Java program made by RPF member "Program4Liberty". This is the quickest way to analyze a new state/county/precinct.

You can download the current version here:
http://sourceforge.net/projects/voteanalyze/?_test=b

All you need to do is to arrange your data in this format:

==================Using VoteAnalyze================
To Use:

I have included pregenerated output information for everyone, but if you should wish to verify that these numbers and this program are correct, the program source code is included here, and you can re-generate the information yourself by double-clicking VotesAnaylze.bat and following the prompts. Use compile.bat to compile the program.

Here is an example of generating the 2012 outputs:

VotesAnalyze.bat
Note: This program requires that no candidate's last name contain another candidate's last name - e.g. Joe Adams and Bill Adamson.
Enter the relative path and filename to the candidates text file: votes2012/candidates.txt
Note: This program requires that all data be provided in files with comma seperated values (.csv)
Enter the relative path to the folder containing the .csv files: votes2012
Generating data by state...
Generaling data for all states combined...
Outputting correlation table...
Generating chart images...
Program Finished.

That's it! "votes2012/candidates.txt" and then "votes2012" are the only things you need to enter to get the program to generate data for 2012.
Use 2008 and 2000 in place for those years.

---------------------------------------------------

Please note: it's not a polished program, so if you type anything in incorrectly, the input tables aren't formatted correctly, or something I've not mentioned here happens, then you'll just get a Java error.

Try again, tweak, and use the forums.

 

[Constitutional Paulicy]

Not sure how accurate this is, but I heard someone say that the reason Romney's numbers surge at the close of the process is due to larger districts. The argument was that he does better than other candidates in larger population centers (big cities) where the vote counts take longer to tabulate so they are the last districts to submit their results.

 

[drummergirl]

Unfortunately it's a bit more than that. If that were the case his graphs would have a bump to a higher percentage, then flat line. But what we see, over and over in nearly every state, is once 20-50% of the votes are tabulated, the graph of Romney's overall percentage climbs linearly to the end of the tabulation. Statistically, it's impossible (well, the odds are better of winning the Texas state lottery 10 weeks in a row on a single quick pick a week).

And no one has been able to come up with a hypothesis as to why Romney's total does this at the expense of Ron Paul, Rick Perry, and Michelle Bachman in Iowa, Paul and Huntsman in New Hampshire, Santorum in Louisiana (Ron Paul untouched), Paul, Santorum, and Gingrich in Oklahoma, etc.

 

[Titus]

Hopefully asking for clarification is ok.

Your article calls these basic statistics. However, it has been about 10 years since I took statistics, and I don't think I dealt much with Omega (I think that's the one that looks like an O). Is that the symbol for standard deviations? If so, are y'all using a t-test?

I'm not quite understanding the captions underneath each graph with the diagonal lines. Are the precincts sorted from smallest to largest and then added together? If so, wouldn't that suggest that the percentage of each precinct is what is flipped, rather than some from each precinct?

I think this needs a lot more explanation for people like me who have taken some statistics and want to know more about which equations were used and how they were generated.

Please don't take this as criticism of the model but I really want to understand it so I can contribute. If I can't understand it, I can be near certain the MSM viewers won't. Heck, the MSM hasn't even added up Maine's numbers.

That being said. +rep. Keep up the good work.

 

[RonRules]

Not sure how accurate this is, but I heard someone say that the reason Romney's numbers surge at the close of the process is due to larger districts. The argument was that he does better than other candidates in larger population centers (big cities) where the vote counts take longer to tabulate so they are the last districts to submit their results.

I have said that and possibly other people in this forum. Most of us have watched the results come in on each of the primaries's election night, being filled with enthusiastic early on, to only be disappointed as the evening progressed.

The time-based evidence is hard to come by from official sources. You'd have to go to CNN/CBS/ABC and ask them for the file where they received the data along with time stamps for each piece of data as it came in.

If you are really interested in researching this, you could also go back to each thread were individual primaries were discussed in this forum, collect the results that we posted with post time in the thread and make charts from that data. The X-Axis would be time instead of cumulative precinct size. I believe that the charts will look somewhat similar to our current charts, because of what you state: "where the vote counts take longer to tabulate so they are the last districts to submit their results."

It would be interesting to see one more point of argument to use. The general population would be more likely to accept such a time-based chart than "cumulative increasing precinct vote tally".

It's a neat project and I'd love to see someone tackle that.

One thing to be mindful though is that precincts, regardless of size, that use Direct Entry Machines (touch screen), will report just as quickly for small precincts as for large precincts. The above concept will work only for manually counted precincts or precincts that use optical scan.

 

[Titus]

It would be interesting to see one more point of argument to use. The general population would be more likely to accept such a time-based chart than "cumulative increasing precinct vote tally".

I would certainly find that easier to understand.

One thing to be mindful though is that precincts, regardless of size, that use Direct Entry Machines (touch screen), will report just as quickly for small precincts as for large precincts. The above concept will work only for manually counted precincts or precincts that use optical scan.

Question. Is the graph suggesting that time changes the way the votes are "flipped" or is it the size of the precinct?


Did someone do Maine?

 

[arsenius]

Titus, it is by precinct size. However, smaller precincts often finish totaling first, and so may be reported first. The argument is not that only later precincts are flipped.

 

[dsw]

Unfortunately it's a bit more than that. If that were the case his graphs would have a bump to a higher percentage, then flat line. But what we see, over and over in nearly every state, is once 20-50% of the votes are tabulated, the graph of Romney's overall percentage climbs linearly to the end of the tabulation. Statistically, it's impossible (well, the odds are better of winning the Texas state lottery 10 weeks in a row on a single quick pick a week).

If the larger percentages are in the larger precincts, i.e., the data points that pull up the average tend to occur toward the end of the sequence, then it won't flatten out. It's only going to flatten out if you get to a point where the prior average is (roughly) the overall average, and the remaining precincts average out to the overall average ... but that's not the case if there's a correlation between the precinct ordering and the vote percentage.

What the non-flattening graphs show, in a convoluted way, is that there IS a correlation between precinct size and vote percentage. If there is such a correlation, the graph won't be flat, and the stronger the correlation the less flat it will be.

Look at it this way. Assume there's a correlation between precinct size and vote percentage and for this example, assume it's increasing. If we look at some midpoint in the sorted sequence, then the correlation means that the points to the right will have a higher average than the points to the left. But the cumulative average is obviously the average of the points to the left, so points to the right will cause the cumulative average to continue rising as we continue accumulating them. So the correlation means that the curve is not flat. Fraud or no fraud, it's just the effect of sorting by a non-independent variable.

The statistical argument assumes that the precinct size and vote percentage are independent variables. If there's a correlation between precinct size and vote percentage, then the statistical argument is incorrect.

 

[drummergirl]

Hopefully asking for clarification is ok.

Your article calls these basic statistics. However, it has been about 10 years since I took statistics, and I don't think I dealt much with Omega (I think that's the one that looks like an O). Is that the symbol for standard deviations? If so, are y'all using a t-test?

I'm not quite understanding the captions underneath each graph with the diagonal lines. Are the precincts sorted from smallest to largest and then added together? If so, wouldn't that suggest that the percentage of each precinct is what is flipped, rather than some from each precinct?

I think this needs a lot more explanation for people like me who have taken some statistics and want to know more about which equations were used and how they were generated.

Please don't take this as criticism of the model but I really want to understand it so I can contribute. If I can't understand it, I can be near certain the MSM viewers won't. Heck, the MSM hasn't even added up Maine's numbers.

That being said. +rep. Keep up the good work.

Thank you for your questions. This summary was written to help explain the math for people who don't know math. next project is a technical summary. And yes, if you flip through the original thread and look for the t-test values, R squared, and f numbers your jaw will drop.

If you are wanting to look for yourself at the basics, page one of the original thread and then read backwards is my best advice for now. I apologize for not having more detail for you atm, but I have to get up early (ewwww!).

 

[drummergirl]

It is not time. It is precinct size, which is actually shorthand for "the number of ballots cast in a precinct". So "small" just means there were not a lot of votes cast there. The difference between small and large is not necessarily obvious. I know we have one state where the break point was a difference of 20 votes per precinct. 20 votes is not that big a change.

And I believe Maine statewide was done.

 

[RonRules]

Question. Is the graph suggesting that time changes the way the votes are "flipped" or is it the size of the precinct?
Did someone do Maine?

If a given state/county is re-plotted with time on the X-Axis, it will show some slope in the direction of vote flipping, much like the cumulative precinct vote tally charts. I expect the chart to be more bumpy, depending on the type of equipment used at the various precincts.

Maine was done at the state level as shown below, but I don't remember seeing charts for individual counties in Maine. You'll have to get the data from the friendly local GOP, but it is also an important project to pursue. Maine is one of those states were will will likely have a plurality of delegates, but lose (officially) the popular vote. This could happen in 6-7 states and I expect much commotion in the coming weeks as these delegates get announced.

It is important that we can make a strong case for vote flipping in these contentious states. From the looks of the chart below, I think Ron had the popular vote in Maine, but got flipped to second place. We need a solid analysis, county by county in Maine. Anybody want to do this?

 

[drummergirl]

The statistical argument assumes that the precinct size and vote percentage are independent variables. If there's a correlation between precinct size and vote percentage, then the statistical argument is incorrect.

I think what you mean is that if there should be a correlation between precinct size and vote percentage then the statistical argument is incorrect. That is absolutely true.

It is the central issue of what we see, as a matter of fact. The math we are talking about is pretty basic and old. If the math is wrong, the whole field of statistics will never be the same.

 

[dsw]

I think what you mean is that if there should be a correlation between precinct size and vote percentage then the statistical argument is incorrect. That is absolutely true.

It is the central issue of what we see, as a matter of fact. The math we are talking about is pretty basic and old. If the math is wrong, the whole field of statistics will never be the same.

But the probability argument (that the odds of Mitt's graph doing such and such is some huge number) only makes sense if there's no correlation between the precinct size and the vote percentage. The slope of the graph proves that there IS a correlation, and therefore the calculation of the odds of it happening is no longer valid. So I'm confused about what the probability argument is supposed to show. It assumes something proven to be false.

It's obviously not impossible for there to be a correlation between precinct size and vote percentage, and therefore a curve that doesn't flatten out, without needing any fraud to explain it. The Virginia Beach data show one way that could happen. There's a large contiguous area with two properties, (a) the precincts in that area are among the largest in the city (they account for 60% of the top 23 precincts), and (b) Romney was very popular in those areas. With no fraud at all having those two conditions coincide is sufficient to have a graph that starts out looking good for Ron Paul but doesn't flatten out and ends with Mitt winning.

I'm not saying that this is the whole story in VBC -- obviously not since the bias for Romney over Paul shows up well before that point in the ordered list of precincts (and well before the alleged flipping is supposed to start according to the algorithm). But it illustrates a kind of situation that could cause this sort of correlation, without fraud.

A key point here is that, as I noted in the other thread, there's a strong pattern in the precinct ordering that tends to cluster precincts of similar size together. I don't know if that's true elsewhere, but it's clear in VBC. So when an area spanning multiple precincts is strongly pro-Mitt, and it happens to be one of the areas with large precincts, a correlation with precinct size is the natural result.

 

[affa]

It's obviously not impossible for there to be a correlation between precinct size and vote percentage, and therefore a curve that doesn't flatten out, without needing any fraud to explain it.

This is why we've spent tens of hours charting historical elections, showing that Romney is, by far, Mr. Precinct Size and that, barring 2008, it has only really been seen in nature to any significant degree Louisiana, where the guy in charge of elections ended up in jail for election fraud.

 

[kathy88]

Get this off the front page.Anyone interested already knows you are in hot topics.It's just a bright light flashed in the eyes that detracts from current real important business.

Wouldn't that be a moderator decision? Wead just wrote a blog about the fraud going on, that is in grassroots. The only "detraction" from other business occurs if you click on the topic. So don't click.

 

[jbauer]

First I can say I stoped reading the 4k+ thread weeks ago.

Second, whether you disagree or agree with the methods is irrelevent here. That horse has been beaten to death several times over. So don't "tinkle" in their cheerios.

You guys have put a ton of work into this, work that you wouldn't have done if you didn't really beleive what you were working on would make a difference. I'm not sure if its valid or not. In my mind for validation it would have to be taken to an outside source, re-run the work re-checked the numbers by someone without a horse in the race. This would ultimatley be a huge factor in proveing or disproving the results and your conclutions.

Have you looked into the costs of having an outside source pour over your data and give their conclutions? I'm not sure what that would cost or what ammount of time it would take. What I do know is if Dr. Paul can raise $1M in a couple days. Surley with properly asking and showing your need you could come up with the several $1k's to take this to the next logical step and if kept transparent you'd get what you're asking for.

If the conclutions are right it would fundementally change the voting world. If they're wrong it would only cost what people were willing to chip in.

What is the next step?

 

[arsenius]

I agree with jbauer, what is the next step?

At this point I don't see this coming to anything, without some sort of breakthrough. The phenomenon has been duplicated all over. It's established. There needs to be a next step to prove whether it's fraud or not, or get it into someone in the media's hands (a tall order, I know!). Drummergirl, your updated summary is great, although I would personally change the beginning to be less emotional and jump right into the report. Thank you for starting work on a technical summary as well!

I am currently emailing drummergirl's summary to the county clerks in Wisconsin, because it is voting next week. I am trying to make it clear in the email that I am contacting them as a concerned citizen who wants to ensure that the voting process is on the up and up, whether the claims are true or not. I am including some questions about how voting machine testing is done, based on some concerns that came out in the thread.

I would definitely donate to a chipin of some sort to get this looked at by an independent party also.