The Palm Beach County Vote

Reallocation and statistical counting methods

Supplemental data to original argument at: Palm Beach County Vote- Reallocation and statistical counting

How to allocate

another way to think of it..suppose we had a primary election where 10 candidates were running for office, and every ballot cast was marked for candidates 1 and 2 even though the ballot said "vote for one." And suppose we are trying to decide which two candidates should compete in a  runoff. Should we throw out all the ballots, or allocate the votes (by some method) between 1 and 2? Which better captures voter intent?

Because of the confusion that was clearly evident (affidavits and statistical charts), it is reasonable for us to remedy this unfortunate ballot design by doing a better job ascertaining the will of the voters.

so using the SAME argument as on the 3400 votes (above), then we might count 1/2 vote for each candidate on those double punched ballots since this is clearly BETTER than discarding! The voter made an intent, but we have 50/50 chance of guessing the RIGHT answer and that is better than 0% if we discard. 

But much better would be to allocate on the single vote ratio in that county (or do it on a precinct level). The result would be that we'd more likely and more accurately capture voter intent this way, than if we threw out the ballot. And accurately capturing voter intent is what it's all about. This is not partisan. 

In other words, if we had 100 double punched ballots for Gore and Buchanan, we'd count those as 99 Gore votes and 1 Buchanan vote since those are the most likely statistics from what we know. In other words, if you knew that the ratio of Buchanan votes to Gore votes in the county were 99 to 1, and you were given 100 ballots and you were told that some were Gore and some were Buchanan, the most logical way to divide them that best captures voter intent is not 50/50, but 99 Gore, 1 Buchanan.

So this plus the 3400 votes should alone do the trick. in fact, this alone will do the trick, but I'd rather have many arguments rather than hinge on just this one.

see http://pages.about.com/bgspence/index.html for analysis of the double punches and a list of links to other analyses. note at the bottom of this page it gives the actual totals for each multi punch ballot for those 1% that were hand counted

Using the 50/50 technique (which isn't really as good as we can get), so based on the numbers on the web page, Gore (punch 5) picks up over 50 votes (40+10) on the 4/5 and 5/6 alone (he'll pick up more than this for all combos listed on that web page) while Bush (punch 3) picks up about 5 votes. of course, you'd do the math for all double punch combos, but there is a net gain of 45 votes for gore for 1% which means a 4500 vote net gain for Gore relative to bush if all precincts were analyzed. This number is close to the minimum number calculated in http://www.ssc.wisc.edu/~bhansen/vote/vote.html which shows Gore should pick up at least 4,270 votes relative to Bush.

If we instead allocate based on a percentage basis, the allocation is even more favorable to Gore!

But to do this, we have to hand count these double punched ballots and hand count them FAST, before the deadline and have a judge rule that they are valid which means the case has to be heard Friday, I believe.

we'd also have to have them handcount these double punches in Palm Beach no matter whether you can file on friday or not. I doubt they are doing this. IF my argument is right, Someone NEEDS to get them to start counting these ballots ASAP, or get someone to program a machine to do this (this may be the simplest way).

Another approach that doesn't require recount: 
Another way is to divide up those 19,120 ballots up based on Bush/Gore/Buchanan, etc. ratio in the rest of the county assuming all classes of voters are equally likely to be confused. You don't need to examine each ballot. You just take the 19,120 and give 62% to Gore, and 35% to Bush, etc.

Recommendation

The judge must decide on a remedy. If he decides nothing, he is, de facto, choosing Remedy A. But which is the remedy that better captures voter intent? 

Scenario: You hand the judge 100 ballots. You tell the judge that these ballots are either Buchanan or Gore votes. You also tell the judge that in the county that these votes are from, that the ratio of A voters to B voters is 99 to 1 (confirmed by exit polls (we'd have to get that data if we want from VNS) and by the tally registered by other voters). 

Question: You tell the judge that in light of the facts, how would you allocate those votes between the candidates? Most people would say 99 to A, 1 to B (we should pose this question to people on the street to show that that's what any normal person would choose).

Now, which of the 2 remedies shown below better captures the intent expressed by those voters that we just did?

Remedy A 

Count 10 votes for each candidate team listed on the ballot even though the ballot was not punched for these candidates. 

Remedy B

Agree that within a single ballot we can't know which way the voter intended. It was a mistake that was corrected. But we can allocate in the aggregate based on voter makeup in the precinct or county.  The absolute best judgment we can make is to allocate 99 to A and 1 to B.

Remedy A is statistically equivalent to ignoring the ballots. It is ridiculous to divine voter intent for a candidate that the voter never punched and far from the statistics in the county.

Rationale for Remedy B: The intent was there: these ballots were NOT randomly punched. Allocating via statistics is by far the most reasonable: if I handed the judge 100 ballots from Palm Beach and said these ballots were votes for Gore or Buchanan, would he allocate them among all candidates equally (even those that were not punched) since that's what he's doing by throwing them out? or allocate 99 to Gore and 1 to Buchanan? which is more fair a remedy for capturing voter intent?

Options

Let the judge decide whether to count 80% of the double punches or 100%. We'd argue 100% since we should count every legal ballot. And these were legal ballots, except the voter intent was hard to decipher.

Next, we recommend allocating on the ratio of the two that are punched in the county based on single votes. We are perfectly happy with allocation based on the ratio in the precinct. We let the Bush camp pick which one they want between those two choices.

Lastly, we'd encourage a full count of the 19,120 multi-punched ballots. If the count can't be completed in time allotted, extrapolate from the count you have. This will encourage the Republican campaign not to send in protesters, throw bricks through windows, etc. since Bush will end up worse if we have to extrapolate. 

We may be able to count the 19,120 ballots by machine if suitably programmed. We'd want to program a machine asap to be able to do this!! The programming can be verified using controlled runs (republicans pick the deck and verify the tally) and by random sampling of the ballots counted and extrapolation.

 

 


old stuff

1. this web page gives a new argument that the palm beach ballot was illegal because it says vote for group, instead of vote for ONE. Here is the page: http://pages.about.com/bgspence/index.html. I had my corporate attorney look into it, and based on his analysis, he thought the ballot was legal.

2. we should ask court to reallocate the 3000+ miscast Buchanan ballots based on Bush/Gore vote ratio already tabulated. See below for case citations on this. For bar and scatter plots of the vote in palm beach, see: The Palm Beach balloting

3. reallocate double punches using 1/2 vote for each candidate (see section below). this alone would do it. Rationale for doing this is in http://pages.about.com/bgspence/index.html which mentions the "vote for group" problem, instead of "vote for one". By reallocating, we capture voter intent and it is better than doing nothing (see argument on reallocating Buchanan below)

4. statistical recount in miami dade (see section below)

Can't go into court until she certifies.

Statistical tie (not pursuing)

Since within margin of error of the machines, it is a statistical tie. Argue to split the electors between the two candidates.

Legislature determines how select electors. could do it by number of counties. Legislature may have final say. courts may divide up electors because law unclear. 

Reallocating Buchanan (not pursuing)

Reasons for not pursuing this is that this is harder argument to make than the double punches. 

It is compelling in the number of papers written by academics. Despite arguments to the contrary, the fact that the absentee ballots from this county indicated that the true ratio is less than what was counted.

This is really the key argument here since the Buchanan vote can be argued that the .79% Buchanan vote is not out of the ordinary in a general sense since there were counties in Florida that had a lot more (on a percentage basis) than this. So we stick to the multi-punch argument for now because a judge without a math background would have a tough time seeing through a smokescreen.

Miami-Dade idea (not pursuing)

here's an idea for Miami Dade to meet the deadline that they haven't considered. approach #2 is new.

approach #1 
machine recount the ballots and handcount just the 10,000 ballots that are in question. I don't see why they aren't doing this...it seems obvious.

approach #2
 if you realize that machine recounts have a margin of error, then realize the same is true of hand counts. In view of the time set by the supreme court, you do a hand count that is STATISTICAL so that you will have a margin of error, just like with machine counts. Process is:

step A: pick 1% of the ballots AT RANDOM from the pile (RANDOM pick is critical) and count them

put those ballots back and repeat the process in A 10 times

Result: you have 10 separate counts. You've done 10 statistical samples of the data.

You look at the mean and std deviation of those samples.

That gives you confidence about the mean and std deviation of the extrapolation (something you don't get on a "single" count of ALL ballots).

So you've counted 10% of the ballots, but you have excellent statistical handle on the result.

In fact, in an error prone process such as hand counting, THIS IS MORE ACCURATE THAN A 100% HAND COUNT!!!!! Let me explain why this is, but if I can contact a real statistics expert, he could argue it better than I can.

Here's why this process is probably more accurate:

suppose each of our 10 "counts" was done by a different set of people. and suppose one group was a bunch of people of the same party pretending to be the other party. well, we'd immediately see this from the stats because the counts in their group would be outside of the mean and std dev established by the other groups. So we could throw that out. in a process where all groups contribute to a SINGLE count, we'd never have spotted these cheaters since we'd have no statistical basis for identifying them since there is only one count at the end.

Secondly, suppose our "cheater" group in the "normal hand count" process handled more than their "fair share" of the ballots, i.e., they counted faster than other groups and were given more work. Again, they'd skew the results much more in a "single count" process than in a process where they were given exactly the same number of ballots as other groups.

In simple terms, this is like judging for olympic gymnastics....you look at the 10 scores from the judges and you can see if one judge is biased. you might throw out the high and low scores, average the result and get a pretty good score like they do in the olympics.

In reality, you do this 1% sampling as many times as is required to get the error rate you are looking for. Since voting cards tabulated by machine have a .1% error rate (even for ideal operation and perfect handling per the references I gave to the team down in florida), you need only repeat your 1% samples sufficient times to establish a similar std deviation.

Again, a statistician could tell you more and could probably tell you exactly how many samples you'll need to make and the size of each sample. it might be a lot smaller than we think.

as far as error rates for the FLA machines, best source is the "independent testing authority" for florida (i don't know who this is), or talk to Mike McCarthy at Election Systems and Software in Omaha, NB or Derry Hobson at Sequoia Pacific (HQ in st. louis, but he's located in Exeter, CA).

MIT contact:

name: Barnett, Arnold I

email: abarnett@MIT.EDU

phone: (617) 253-2670

 

my contact number is 650 279 1008

-----------

PARTIAL LIST of Palm beach attorneys who have filed 

 W. Dexter Douglass*

 Douglass Law Firm 

850/224-6191

douglass@nettally.com

 

 Patrick W. Lawlor**

Young & Lawlor

 954/426-8226

no email address

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

 Gary M. Farmer**

Gillrspie, Goldman, Kronen & Farmer

 954/771-0908

no email address

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

Henry B. Handler*

Weiss & Handler

561/997/9995

whpa218@aol.com

hbh@weissandhandlerpa.com
harryatt@emi.net

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

Mark A. Cullen*

 The Szymoniak Firm

561/989-9669

SARPA@FRONTIERNET.NET

search for attorneys: http://199.44.15.3/Members.nsf/ME+Search?OpenForm


Return back to basic arguement: :Palm Beach County Vote- Reallocation and statistical counting

Steve Kirsch Political Home Page