Washington State Recount, Part Deux

[Fmr. Gov. NickG]

If you are so suspicious then what is your theory on where the ballots came from?  Do you think they were cast after the election?  Or are they actually completely fake, totally unconnected to an actual voter?

[Fmr. Gov. NickG]

Again Carl, if the ballots that are found in King County are not valid ballots, then what exactly do you they they are?  Do you think they are fraudelent votes, or do you just think that voters should just be out of luck if an election worker screws up and their votes aren't counted the first time?

[Fmr. Gov. NickG]

Again Carl, if the ballots that are found in King County are not valid ballots, then what exactly do you they they are?  Do you think they are fraudelent votes, or do you just think that voters should just be out of luck if an election worker screws up and their votes aren't counted the first time?

For the moment, let's make a 'leap of faith.'

Lets just suppose, for the sake of argument, that the ballots are totally legitimate.

Why did it take Logan over a month to find them?

The Keystone Kops weren't this incompetent!

Also, would you convict someone on the basis of evidence that wasn't available at an original trial, or the first retrial, but which the prosecution magically 'found' on the third retrial?

Chain of custody problems do arise.

You can't convict someone based on that subsequent evidence of guilt if double jeopardy had attached, but it can be used to exonerate someone who had already been convicted.  That's because their is a presumption of innocence, and the burden of proof is on the side of the prosecutor; we are willing to tolerate more false negatives in exchange for fewer false positives.

In the case of ballots, their should be a presumption that everyone's vote should count.  The burden of proof must be on the person who wants to disqualify a person's vote.  Absent some specific evidence that a person's vote is invalid, it should be counted, no matter how much the election workers screwed up.

There are probably errors of this sort in most elections; it's just that we don't hear about them because the elections aren't close enough to necessitate this level of scrutiny.   The whole point of having a recount is correct errors in the first count in the case where such errors might matter.  If Republicans (or anyone else) want to challenge specific ballots for specific reasons, that's fine, but they should be counted in the meantime.

[Fmr. Gov. NickG]

For those of you familiar with statistical mathematics, I have a question for you.

If you take the original total for all votes counted for Govenor in the state of Washington this year by county, and then take the square root of that number, which one county has already exceeded adding on more than the square root of initial votes?

Hint, its the one county which has not yet posted its recount total.

Is this a joke?
Why is the square root of the number of votes relevant?  Of course the answer is whatever county is largest, but it has nothing to do with anything.  You are assuming that large counties should find drastically fewer mistakes in proportion to their population, and I don't see why that would be true. 

[Fmr. Gov. NickG]

For those of you familiar with statistical mathematics, I have a question for you.

If you take the original total for all votes counted for Govenor in the state of Washington this year by county, and then take the square root of that number, which one county has already exceeded adding on more than the square root of initial votes?

Hint, its the one county which has not yet posted its recount total.

Is this a joke?
Why is the square root of the number of votes relevant?  Of course the answer is whatever county is largest, but it has nothing to do with anything.  You are assuming that large counties should find drastically fewer mistakes in proportion to their population, and I don't see why that would be true. 

I would suspect that Carl is looking for statistical anomolies.  While I don't think he'll find any, it would be interesting to check.

NickG really doesn't understand statistics.

The square root principle is a well established parameter for error.

In the first vote count, King county counted 874,928 votes cast for Govenor. 

The square root of this is 935 (rounded to the nearest whole number).

One the first (machine) recount, King counted added 971 votes, which was slightly in excess of the easily explained margin of error (935).

No other county, in neither the first (machine) recount nor the second (hand) recount has had a change from the original vote which equalled, much less exceeded, the square root of the original vote count.

I understand statistics just fine....

Your "square root" principle only works if their is an equal chance that a given county will overcount its vote and undercount its vote.  This assumes that "mean" result after the count will be the same as the mean result before the count.  But this is clearly not true...most of the changes are from counties adding votes that weren't counted previously, not from subtracting votes that were mistakenly counted. 

- The number of overvotes should be linearly related to the population of the county.
- The number of undervotes should also be linearly related.
- If there is an equal tendency for undervotes and overvotes, the NET ERROR will be related by some function of the square root of population.  But if there are only overvoters, the net will still be linearly related. 
- In the real life situation, where there are a few undervotes and far more overvotes, the net result will be somewhere in between, but much closer to a linear function than an exponential one.

Moreover, even if there is an equal chance of overcounting and undercounting, their is no reason why the square root of population will be the "easily explained margin of error".  This would more accurately be the square root of population times some constant, but setting this constant at 1 is completely arbitrary on your part.

[Fmr. Gov. NickG]

As evidence of my assertion that, because the recount is more likely to add votes than it is to subtract them, the net change in vote will be more closely linearly correlated to population than the square root of population, I ran the correlation for all the Washington counties excluding King county:

(Note that I am using the total number of votes counted in a county instead of population)

Linear correlation b/w (change in votes during manual recount) and (voting population) = .8261
Linear correlation b/w (change in votes) and (square root of voting population) = .7347

[Fmr. Gov. NickG]

I'm not sure I understand how turnout rates figure into your "square root" idea.

Yes, turnout rates vary slightly from precinct to precint, but not by that much; there is an obvious and overwhelming linear relationship between population and votes.  I ran the correlation between 2004 turnout and 2000 census population data for Washington counties, and the correlation was .995 (where 1.00 would mean absolutely no variation in turnout).   So any formula or correlation that applies to population will also apply to turnout.  And even if turnout were more varied, what does this have to do with your "square root" idea?

You said:

You are very wrong.  The method I employed estimates the likelihood of an error (undercounting or undercounting).  When a recount exceeds the paraments of this methodology, it is highly suspect.


I assume by this you mean "undercounting or overcounting".  This is not a good assumption.  Your  "square root" formulation would be correct (with the caveat that I mentioned before that the constant isn't necessarily 1) if you were talking about two recounts with the same methodologies, in which there was no particular reason why the second count would over- or under-count more votes than the first. 

But in this case, the second recount (the hand recount) is designed to be more inclusive than the first.  Almost all of the changes will be adding new ballots rather than subtracting.   In this situation, the relationship between population and change in votes will be highly linear rather than exponential, which I think is supported by the statistical evidence I posted above.

[Fmr. Gov. NickG]

I'm not sure I understand how turnout rates figure into your "square root" idea.

Yes, turnout rates vary slightly from precinct to precint, but not by that much; there is an obvious and overwhelming linear relationship between population and votes.  I ran the correlation between 2004 turnout and 2000 census population data for Washington counties, and the correlation was .995 (where 1.00 would mean absolutely no variation in turnout).   So any formula or correlation that applies to population will also apply to turnout.  And even if turnout were more varied, what does this have to do with your "square root" idea?

You said:

You are very wrong.  The method I employed estimates the likelihood of an error (undercounting or undercounting).  When a recount exceeds the paraments of this methodology, it is highly suspect.


I assume by this you mean "undercounting or overcounting".  This is not a good assumption.  Your  "square root" formulation would be correct (with the caveat that I mentioned before that the constant isn't necessarily 1) if you were talking about two recounts with the same methodologies, in which there was no particular reason why the second count would over- or under-count more votes than the first. 

But in this case, the second recount (the hand recount) is designed to be more inclusive than the first.  Almost all of the changes will be adding new ballots rather than subtracting.   In this situation, the relationship between population and change in votes will be highly linear rather than exponential, which I think is supported by the statistical evidence I posted above.

Let me deal with two fundamental misassumptions you make.

First, you seem to believe that the number of votes cast is little more than a reflection of the population.  I tried gently toi tell you this is not so, and gave you both the concept and specific examples which disprove your assertion previously.  Apparently I was not clear enough as you still do not understand, so I will specifically specll it out with a couple of examples:

County          Population          GE Vote for Govenor          Ratio
                      7/1/03

San Juan        14,762                  9,853                               1.4982

Whitman        40,702                 17,549                               2.3193

Second, you seem to have a rather fundamental misunderstanding of tha law regarding recounts.  Suggest you read the decision of the Washington Supreme Court and the Superior Court judge I previously cited with a link.

I'm not misunderstand this at all...it is not an "assumption".
I ran an analysis of the data and found that votes correlated to population in this election with a coefficient of .995...a near perfect correlation.  Giving me one example of an exception doesn't disprove the correlation for the entire data set.

Are you suggesting that the Washington state election law  mandates that the two recounts have an equal chance of gaining and losing votes?  No matter what the law is, this clearly isn't the case. 

In the hand recount, excluding King County, the candidates together gained a total of 1211 votes, and lost a total of 46.

So a candidate is overwhelmingly more likely to gain votes in the recount than lose them. 

When this happens, the change in votes will be much more closely correlated to population than the square root of population.  Even if you don't believe the theory, run the numbers yourself!  As I posted above, this is empirically true....the coefficient is .83 for the first relationship, and .73 for the second.

Finally, even if all of your assumptions are correct, you haven't given any reason at all why the "easily explained margin of error" should be the square root of population rather than the square root of population times some constant related to the standard deviation.

[Fmr. Gov. NickG]

You said:

I'm not misunderstand this at all...it is not an "assumption".

I reply:

Please cite where I said what I am emphatically denying!

You said:

I ran an analysis of the data and found that votes correlated to population in this election with a coefficient of .995...a near perfect correlation. 

I reply:

Your "correlation" is faulty. People used to "correlate" women's hem lines with stock market results, etc.  I gave you a specific example of how turnout rates vary markedly and you reply with with a dam the specific examples.

You said:

Giving me one example of an exception doesn't disprove the correlation for the entire data set.

I reply:

How many examples WILL it take to disprove your fallacious correlation?

You said:

Are you suggesting that the Washington state election law  mandates that the two recounts have an equal chance of gaining and losing votes?  No matter what the law is, this clearly isn't the case.

I reply:

I specifically did NOT say what you are imputing to me.  Your technique of assuming things I did not say is reminescent of the debating style of the late Richard Nixon.

You said: 

In the hand recount, excluding King County, the candidates together gained a total of 1211 votes, and lost a total of 46.

So a candidate is overwhelmingly more likely to gain votes in the recount than lose them. 

I reply:

So, based on this one example, all recounts will result in an increase in votes counted?

You said:

When this happens, the change in votes will be much more closely correlated to population than the square root of population. 

I reply:

I don't know whether you are unable to understand what I have posted or are simply deliberately mistating my posts.  I NEVER based my analysis on population.  YOU based YOUR analysis on population, which is faulty for the reasons I have repeatedly pointed out (giving specific examples).

You said:

Finally, even if all of your assumptions are correct, you haven't given any reason at all why the "easily explained margin of error" should be the square root of population rather than the square root of population times some constant related to the standard deviation.

I reply:

Talk to a professor of social sciences familiar with statistical analysis.  He/She may be able to explain it to you.  It is Generally Accepted as a form of analysis.  Your 'linear' method is rejected!

You must be logged in to read this quote.

It occurs to me that this whole argument about turnout vs. population doesn't matter.  The initial regressions I ran were for total votes vs. change in votes, and those are the coefficients I am quoting (.83 vs. .73)....although the coefficients are the same if you substitute population for total votes.   Although I misunderstood you initial assertion, I mentioned what I was running the regression on when I initially posted the numbers, so I'm not sure how we got into this argument.

I never said all recounts will result in an increased vote total.  Indeed, if the two recount methods were identical, the absolute value of the error between them would have would correlate to the square root of turnout (again, times some constant).

But in this case you have two complete different methodologies.  The second by its nature will tend to find many more votes that it subtracts.  This is supported by the actual results.  Because the mean is positive and not 0, this will tend toward a linear correlation.  And once again, this assertion is supported by an analysis of the actual results.  I have yet to see any statistical evidence that your theory has any basis in reality.