(Huge) delegate vote anomaly in Alabama verified

[tremendoustie]

As far as I know the delegate races were the same in all precincts. The delegate races also aren't between delegates of different candidates, they're races between two or more potential Newt delegates, or two or more potential Paul delegates, etc.

But, these races were not for national delegates, right? They were local -- so they'd differ by precinct ... correct?

Where there any cases where delegates ran unopposed?

Every candidate had many, many cases where they got more votes for their first delegate than they themselves received. Until we understand why and how that occurs, we don't even understand the data on a very basic level.

More time understanding what the data means, and less time plotting it a thousand different ways, might be helpful.

 

[drummergirl]

You're spending tens of hours trying to make the facts fit the result you want to see, not tens of hours really trying to understand the reasons for the data.

So... you've been around awhile, but I haven't seen your avatar on any of the data threads before this week. I suggest that you might want to know what you are talking about before you spout your mouth off like that.

Your statement is insulting, offensive and degrading.

I don't know what your background is, but most of the people doing serious analytical work on this problem are highly educated engineers and scientists. Unfortunately, we don't have a professional statistician (at least not as far as I'm aware), but most of the math in politics is pretty straightforward.

If you think that the data plots of 2012 are what is to be expected, I would really love to hear your explanation of the reams of historical data and data from elections in Europe, Canada, etc. that are vastly different.

Please, share your background and your great wisdom. Show us your analytical prowess and your greatness with statistics. Otherwise, quit with the name calling.

 

[drummergirl]

But, these races were not for national delegates, right? They were local -- so they'd differ by precinct ... correct?

Where there any cases where delegates ran unopposed?

Every candidate had many, many cases where they got more votes for their first delegate than they themselves received. Until we understand why and how that occurs, we don't even understand the data on a very basic level.

More time understanding what the data means, and less time plotting it a thousand different ways, might be helpful.

The delegate races are for the convention in Tampa; Alabama chooses theirs this way. Look at the ballot; it's posted multiple times in this thread.

I don't know how you do your serious analytical work. But plotting data in various ways and looking for trends, then doing statistical operations to analyze those trends is precisely how we see what the data is showing us. I can assure you that for every chart Ron Rules, Liberty1789, affa, or The Man has posted (apologies to others whose names don't come to the top of my head), they probably have 20 in the trash can (dead ends, rough drafts, etc.).

I really do recommend reading the summary:

https://docs.google.com/document/d/1EokVx9tDsrjAJ-7H9XoPv3KmZYDvVjSFJ4cuxJTo1iE/edit

After page 10 is really for the mathophiles, but if you are good at math it's easy to understand.

 

[dsw]

But, these races were not for national delegates, right? They were local -- so they'd differ by precinct ... correct?

Where there any cases where delegates ran unopposed?

Every candidate had many, many cases where they got more votes for their first delegate than they themselves received. Until we understand why and how that occurs, we don't even understand the data on a very basic level.

More time understanding what the data means, and less time plotting it a thousand different ways, might be helpful.

No.

Unopposed delegates did not show up on the ballot.

Near the beginning Liberty posted some graphs showing the rates at which that "fatigue" occurred. You could start there. Understanding why and how that occurred is the whole point, but you haven't caught up yet with what has already been covered about it.

You don't even understand what the delegate races were about, and when the question was answered you asked the question again but the answer is the same.

Look, I'm all in favor of healthy skepticism. I'd welcome having someone else here applying some healthy skepticism. But start by figuring out how the ballot was supposed to work, etc., or else it's not skepticism it's just snark. A link to the ballot images has been posted. And Liberty has posted the data in a nice spreadsheet format, so you can dig into it yourself. For example, I looked at some variations of the "just wow" plot from yesterday, but plotted in simpler ways that I thought might have a better chance (if the pattern were still clear) of convincing someone who is not in the inner circle. Most people don't know what to expect from a graph of the total candidate votes minus the average of the corresponding delegate votes plotted cumulatively with precincts sorted by the total number of votes cast, so their reaction is more likely to be "huh?" than "just wow". Especially if they haven't seen the proof that no possible combination of demographics, campaign strategies, clustering of precincts geographically by size and other factors could possibly result in the kind of correlations that would make the shape of the wow graph unremarkable. So read up a bit, grab the data, and dig in!

 

[RonRules]

most of the math in politics is pretty straightforward.

Maybe not! I've got to re-learn the Chi-Squared test, because I think that what we need to prove this thing. Unfortunately the statisticians are not quickly rushing to the rescue here.

The more advanced math may be what's needed, but when most people can't read the most basic chart, I cringe in advance at what the response will be.

If statisticians were on the nightly news like the sports heroes, this vote fraud would be solved long ago.

 

[drummergirl]

The more advanced math may be what's needed, but when most people can't read the most basic chart, I cringe in advance at what the response will be.

If statisticians were on the nightly news like the sports heroes, this vote fraud would be solved long ago.

No kidding!

I had a similar reaction when I saw the scatter plots a few pages back. I thought, that looks like exponential distribution; I don't have time to relearn that today, and I definitely don't have the time to relearn it and teach it through forum posts.

Because... RON PAUL IS COMING TO TEXAS TODAY!!! And I leave for College Station in a few hours, WOOT! <happy dance>

 

[dsw]

https://docs.google.com/document/d/1EokVx9tDsrjAJ-7H9XoPv3KmZYDvVjSFJ4cuxJTo1iE/edit

After page 10 is really for the mathophiles, but if you are good at math it's easy to understand.

But you have to be willing to overlook the fact that what it says on page 10:

Another way to look at the data is to chart it and consider, what are the odds of getting to the endpoint from somewhere in the middle? This is essentially what the news agencies do on election night. They look at the votes that have come in, calculate the margin of error, figure the odds of that candidate ending up with more than 50% of the votes, and project a winner as soon as possible.

is not essentially what exit polls do, or even close to what exit polls do. What the exit polls do, among other things, is very carefully analyze their sample (which they've already tried to randomize by selecting locations with different demographics, and trying to randomize the selection of people to poll at those locations), compare the demographics of that sample to the demographics that their model predicts for the overall election turnout, and weight the sample they have in ways that will hopefully make it closer to the unattainable ideal of a truly random sample. They augment this with polling from just before the election (trying, for example, to get data on early and absentee voters who tend to be skewed toward older voters among other things), and sometimes telephone polling on the day of the election. If you understand why exit polls have to do all of this in order to get results that are (usually) fairly accurate, then seeing what is going on in the argument being presented in that document will be a lot easier.

 

[drummergirl]

But you have to be willing to overlook the fact that what it says on page 10:

is not essentially what exit polls do, or even close to what exit polls do. What the exit polls do, among other things, is very carefully analyze their sample (which they've already tried to randomize by selecting locations with different demographics, and trying to randomize the selection of people to poll at those locations), compare the demographics of that sample to the demographics that their model predicts for the overall election turnout, and weight the sample they have in ways that will hopefully make it closer to the unattainable ideal of a truly random sample. They augment this with polling from just before the election (trying, for example, to get data on early and absentee voters who tend to be skewed toward older voters among other things), and sometimes telephone polling on the day of the election. If you understand why exit polls have to do all of this in order to get results that are (usually) fairly accurate, then seeing what is going on in the argument being presented in that document will be a lot easier.

Yes, I've really simplified what the pollsters do. And some of what they do is art as much as science. But news agencies don't like headlines like "Dewey defeats Truman" making them the laughing stock of the media, so they do use a lot of math.

 

[dsw]

Yes, I've really simplified what the pollsters do. And some of what they do is art as much as science. But news agencies don't like headlines like "Dewey defeats Truman" making them the laughing stock of the media, so they do use a lot of math.

Yes, but you've simplified it to the point of it being false. Just like on the first page where you write "Basic statistics told the pollsters that if they polled 1200 likely voters, they would have a margin of error of +-3% and they bet that would be close enough to make a good prediction." In fact, basic statistics said no such thing. It said that if they polled 1200 *randomly selected* likely voters they could draw some conclusions. But without that qualifier (which is what they're trying to approximate when they go through so much analysis and adjustment of the data in an exit poll) you can't draw any such conclusion. And that's also what's wrong with the way you're using the marble counting analogy, comparing something that is defined to be randomly selected with something else that is not. And so on throughout the document.

A lot of the "art" in exit polling is coming up with an accurate model for the demographics of the overall voter turnout. That's not something that can be known exactly, and if you get it wrong then the adjustments they work so hard to make to their samples would be incorrect. But the "science" (math) part of it is more straightforward. If they didn't do those things their calculation of the probability (margin of error) would be invalid.

The way you calculate probabilities is invalid for exactly the same reason. The part you've left out about exit poll methodology in your over-simplified explanation is the part you leave out in your own calculations, but the exit pollsters actually *do* make those corrections that you don't mention. Without that the calculation is bogus.

 

[affa]

The "flipping" b.s. is absurd. Sorry, it is. If you asked me to draw what I'd guess the plots would look like, sorting precincts by size, I'd draw exactly that -- guaranteed, depending on the region, certain candidates will do better or worse in districts near or in certain municipalities, than they will in the surrounding countryside.

Which is why we actually look at historical election data, for comparison. You know, rather than just 'guessing' what we think they should look like. We don't 'guess' what the charts should look like, we've charted elections worldwide and back to the 80's. Perhaps if you actually did any research, you might understand why the charts we see look so odd. But instead, you do the absolute definition of jumping to conclusions.

 

[drummergirl]

The way you calculate probabilities is invalid for exactly the same reason. The part you've left out about exit poll methodology in your over-simplified explanation is the part you leave out in your own calculations, but the exit pollsters actually *do* make those corrections that you don't mention. Without that the calculation is bogus.

Well, if we were doing random sampling of 1000 people and expecting to get results like CNN, that would be true. But we are not polling. We are looking at full sets of actual election data so the sample size = 100%, or close to it (still about 15% or so uncertified in Alabama). The laws of large numbers really ought to apply. And when we see historical data, like Alabama 2008, the math works fine. So the question still remains, what is wrong in 2012?

 

[RonRules]

depending on the region, certain candidates will do better or worse in districts near or in certain municipalities, than they will in the surrounding countryside.

Look up the word "Cumulative"
http://lmgtfy.com/?q=Cumulative

The key to accept our arguments is to understand our relatively simple charts:
1) They are actual election votes, not polls.
2) I personally use ALL the vote data that is available. (If I don't include a point in the chart, I explain why, usually because it screws up scaling)
3) And most importantly, the rightmost part of the charts (and the rightmost point specifically) includes ALL VOTES, therefore ALL DEMOGRAPHICS.

The demographics argument fails once you understand our basic charts.

 

[dsw]

Well, if we were doing random sampling of 1000 people and expecting to get results like CNN, that would be true. But we are not polling. We are looking at full sets of actual election data so the sample size = 100%, or close to it (still about 15% or so uncertified in Alabama). The laws of large numbers really ought to apply. And when we see historical data, like Alabama 2008, the math works fine. So the question still remains, what is wrong in 2012?

The law of large numbers ought to apply IF the way you sample the data satisfies a precondition of independence, a precondition you go on to show is not satisfied. But then you go on to apply the law of large numbers as if that precondition were satisfied.

Who is your target audience? Do you expect them not to understand that there's a necessary precondition that you are omitting? And not to realize that the precondition clearly doesn't hold in the cases where you would need it to hold for your argument to be valid?

 

[tremendoustie]

Which is why we actually look at historical election data, for comparison. You know, rather than just 'guessing' what we think they should look like. We don't 'guess' what the charts should look like, we've charted elections worldwide and back to the 80's. Perhaps if you actually did any research, you might understand why the charts we see look so odd. But instead, you do the absolute definition of jumping to conclusions.

Yep, and you get the same thing in many earlier elections -- e.g. huck did better in many rural areas too. It's completely expected, to anyone actually interested in using common sense.

But, I don't expect anyone here to change their mind -- it's practically a religion at this point.

 

[tremendoustie]

Look up the word "Cumulative"
http://lmgtfy.com/?q=Cumulative

The key to accept our arguments is to understand our relatively simple charts:
1) They are actual election votes, not polls.
2) I personally use ALL the vote data that is available. (If I don't include a point in the chart, I explain why, usually because it screws up scaling)
3) And most importantly, the rightmost part of the charts (and the rightmost point specifically) includes ALL VOTES, therefore ALL DEMOGRAPHICS.

The demographics argument fails once you understand our basic charts.

You apparently don't understand your own charts. You might as well sort the results by age, and then get shocked when romney picks up more old people.

I really wouldn't mind the obviously flawed logic -- it's the arrogance that frosts me.

P.S. I graduated fourth grade; I know what the word cumulative means. Do you know what the word "patronizing" means?

 

[tremendoustie]

Whatever. I really wanted to look into why so many candidates received more votes for their delegate than they themselves did, which would seem impossible. You know, actually try to understand the meaning of the data.

Apparently I accidentally stumbled into a cool-aid festival instead, so never-mind. Instead of checking into the reason behind a behavior, we instantly declare "fraud!!!" then plot the same phenomenon 1000 different ways, and call it "mathematical proof" of a-priori assumptions. Have fun ...

 

[dsw]

Look up the word "Cumulative"
http://lmgtfy.com/?q=Cumulative

The key to accept our arguments is to understand our relatively simple charts:
1) They are actual election votes, not polls.
2) I personally use ALL the vote data that is available. (If I don't include a point in the chart, I explain why, usually because it screws up scaling)
3) And most importantly, the rightmost part of the charts (and the rightmost point specifically) includes ALL VOTES, therefore ALL DEMOGRAPHICS.

The demographics argument fails once you understand our basic charts.

Obviously the demographics of the right-most point are identical to the demographics of all voters. By definition.

You left out the part that shows why the demographics (and other factors that could be relevant) of the first N% of the precincts sorted by total votes cast can't be significantly different from the demographics of all voters.

The smallest 40 counties by vote totals are 21% of the overall vote, but only 8.5% of the vote from the 100 top precincts. The largest 8 counties are 52% of the overall vote, but 71% of the vote from the 100 top precincts. So as you get toward the right-hand side of the graph you're seeing a higher percentage of votes coming from those eight larger counties than you are when you're on the left-hand side of the graph. What is your argument that weighting the largest counties less heavily on the left and more heavily on the right, and the smaller counties more heavily on the left and less heavily on the right, isn't going to introduce any correlation with voter preference as you move from left to right in the graph?

 

[affa]

Yep, and you get the same thing in many earlier elections -- e.g. huck did better in many rural areas too. It's completely expected, to anyone actually interested in using common sense.

But, I don't expect anyone here to change their mind -- it's practically a religion at this point.

You're just trolling this thread anyway - telling people 'good job this looks suspicious' while intentionally insulting the very same people over and over in the same breath. You didn't even understand the ballot, and have been asking super basic questions anyone that spent even 30 seconds doing their own research would know the answer to... answers that are required to even understand what's being discussed in this thread, let alone have an intelligent opinion on it.

As for 2008, I love that you pick Huckabee, the one candidate we've identified in 2008 as having the same anomaly in a couple counties. Love it! You're paying just enough attention to be divisive.

 

[drummergirl]

The law of large numbers ought to apply IF the way you sample the data satisfies a precondition of independence, a precondition you go on to show is not satisfied. But then you go on to apply the law of large numbers as if that precondition were satisfied.

Who is your target audience? Do you expect them not to understand that there's a necessary precondition that you are omitting? And not to realize that the precondition clearly doesn't hold in the cases where you would need it to hold for your argument to be valid?

That's not exactly the case. What we see (over and over again) is a linear correlation after a certain point (also called hinge). Before the hinge point, behavior is as expected. Then there is a change. And then the statistics change too. We start looking at R-squared, t-test, f-stat (is this really a straight line and just how well does it fit the data?). Those numbers don't assume a random sample or independence at all.

As to the target audience; it's pretty darn broad. The first part of the summary is geared towards any literate adult. It uses as little in depth technical information as possible since most people despise math and don't even understand how to balance a checkbook. The technical addendum is for people with a math background; the idea being, we really need some help over here asap. I'm sure the analysis to date is not perfect, but I do know enough to to know we have a serious problem in our electoral system. I can't tell you who dun it, or precisely how (though we are getting closer on this all the time), but this data smells fishier than the dumpster behind a Red Lobster.

 

[drummergirl]

You apparently don't understand your own charts. You might as well sort the results by age, and then get shocked when romney picks up more old people.

Look, I'm truly sorry that you never made it past the fourth grade. But please understand that the math we are using here is taught in basic statistics classes everywhere. If you have questions, by all means ask them. But making careless remarks is counterproductive and does nothing to further your education.