MEDIA: investigate these states for 2002 voter fraud: MN, TX, GA, NH ..

[TruthIsAll]

The following are the results of 8 hotly contested 2002 Senate races
Note the deviations between the final polls and the actual results.
Four races fell outside the final poll margin of error and swung sharply from the Democrat to the Republican.

The odds that 4 out of 8 elections would fall outside the MOE (all for the Repubs) is 1 out of 43,040 (see below). The strange results are in MN,GA,TX,NH
FINAL POLL......ELECTION
Dem Rep diff Dem Rep Diff Net change
NC 42 48 -6 45 54 -9 -3
MN 47 39 8 47 50 -3 -11
AK 51 46 5 54 46 8 3
GA 49 44 5 46 53 -7 -12
TX 48 49 -1 43 55 -12 -11
MO 46 46 0 49 50 -1 -1
NH 46 40 6 47 51 -4 -10
CO 45 47 -2 45 51 -6 -4
Average 46.75 44.875 1.875 47 51.25 -4.25 -6.125

The probability that a given poll will be correct within the MOE is 19 out of 20 =0.95, where the MOE = + or - 3 points.
The average switch was a 6.125% move to the Republican, beyond the margin of error…
Out of 8 races, the probability that ALL will fall WITHIN the margin of error =0.663. This means the chances that 1 or more will fall outside the MOE = .33, or 1 out of 3.
But 4 out of 8 races fell outside the MOE!
The Probability is calculated using the Excel Cum. Binomial Distribution Function:
=BINOMDIST(4,8,0.95,TRUE) = 0.000371751


Probability Matrix
N= Number of elections (trials)
3,4,5,6 =Number of elections outside MOE
N........3........4.......5.......6

8 0.579% 0.037% 0.002% 0.000%
10 1.150% 0.103% 0.006% 0.000%
13 2.451% 0.310% 0.029% 0.002%
16 4.294% 0.700% 0.086% 0.008%
20 7.548% 1.590% 0.257% 0.033%
25 12.711% 3.409% 0.716% 0.121%
30 18.782% 6.077% 1.564% 0.328%
34 24.065% 8.813% 2.592% 0.627%

The probability that all the races would fall for the Republican:
Prob (all N) = Repub
.........3......4.........5......6
12.50% 6.25% 3.13% 1.56%


Therefore, the Joint Prob (Outside MOE and all Repub)

N........3........4.......5.......6
8 0.072% 0.002% 0.000% 0.000%
10 0.144% 0.006% 0.000% 0.000%
13 0.306% 0.019% 0.001% 0.000%
16 0.537% 0.044% 0.003% 0.000%
20 0.944% 0.099% 0.008% 0.001%
25 1.589% 0.213% 0.022% 0.002%
30 2.348% 0.380% 0.049% 0.005%
34 3.008% 0.551% 0.081% 0.010%

Odds =1 out of X
N.........3........4.......5.............6
8 1,382 43,040 2,077,263 159,672,547
10 695 15,557 502,435 23,234,010
13 326 5,156 111,666 3,240,553
16 186 2,284 37,326 791,105
20 106 1,006 12,432 194,355
25 63 469 4,466 52,763
30 43 263 2,047 19,497
34 33 182 1,235 10,208


The odds that 4 out of 8 elections would fall outside the MOE (all for the Repubs) is 1 out of 43,040.
The odds that 4 out of 16 elections would fall outside the MOE (all for the Repubs) is 1 out of 2,284.

The odds that 4 out of 34 elections would fall outside the MOE(all for the Repubs) is 1 out of 182.



The binomial distribution function.

BINOMDIST(number_s,trials,probability_s,cumulative)

Number_s   is the number of successes in trials.
s= number races within MOE
Trials   is the number of independent trials.
Trials is the number of CRITICAL (competitive, close) races

Probability_s   is the probability of success on each trial.
Here Probability_s = .95 or 19/20 that the result will fall within the MOE of (3%).So the probability = 1 -19/20=1/20 that the result will fall OUTSIDE the MOE

Cumulative   is a logical value that determines the form of the function.
If Cumulative is TRUE, then BINOMDIST returns the cumulative distribution function, .
which is the probability that there are at most number_s successes;
if FALSE, it returns the probability mass function, the probability that of number_s successes.

[anamandujano]

1. A few questions

First--I'm on board for an email campaign should you decide to go that route.

But--

1. Will they be able to understand all the stats? The information is impressive. Can it be simplified and still be impressive?

2. Can you get a name statistician (if that's even the right field) to hit them with this? (Maybe you are a name statistician.)

I also think that they cheated like crazy (again) for the 2002 elections. This stuff burns me up. They need to be exposed for trifling with our democracy like this.

[NJCher]

2. info needs to be translated into a news piece

Who did the numbers?


Cher

[papau]

3. Anyone can do the numbers - first year Stat course is all that is needed

But Media will want resume to include PHD and University Name.

[spotbird]

7. Try this link

http://www.lightscion.com/voxnyc/archives/00000051.htm

It is a little more clear.

[CafeToad]

4. Didn't the same thing happen in 1994 mid-term elections?

Specifically, all the close races broke the repub's way at the last minute?

I doubt that much can be made statistical anomolies from a legal point of view, but if some publicity is forthcoming to let them know that we're onto them, then at least the chance of re-occurrences in the future may be diminished (is that your point?).


on edit:

I suppose you realize that 95+% of DU'ers aren't going to buy your arguments, which seem to hinge on the fact that POLLS ARE ACCURATE (at least is seems like I've seen an awful comments about how inaccurate polls are, if that's what people really believe, then why should election results match polling results?). Note that personally I believe that well-worded polls are generally accurate, but I'm just wondering if this is the best forum for you to be promoting your idea???

[Jackpine Radical]

5. There are polls and polls.

In fact, election-prediction polls are generally rather accurate because you have an external validator criterion, namely, the election itself. Thus over a relatively long history of comparing results against (reasonably) fair elections, it was possible to develop an accurate polling technology.

The point here is that when the poll stops predicting the validator criterion, then something has apparently gone wrong with the validator values. Like fraud for instance.

On the other hand, there is no external validator for things like popularity polls, etc., so they can never be trusted in the same way as polls that cdan be subjeted to validation. You just have to take their word that, say 90% of the population thinks Bush is doing a good job. The only thing you really have in these cases is reliability indicators--do other polls sing different questions & different samples produce similar results? Reliability is only a measure of whether the measurement is cdonsistent or repeatable, not whether it is in some greater sense "accurate."

[TruthIsAll]

18. I am not promoting an idea, just using basic probability analysis

and applying it to the polling vs. actual resuts, to get a ball-park estimate of the probabilties.

The +3/-3% standard MOE takes into account polling error. The odds are 19 out of 20 that the final result will fall within the MOE.

People sometimes confuse the MOE with correctly picking the winner. The projected winner can lose, but the forecast is still correct within the MOE.

For instance assume the final poll has A defeating B by 51-49. If B defeats A by 52-48, then the projection was correct - it fell within the MOE. But if B wins by 53-47, then the projection was wrong. This happens approximately 1 out of 20 times.

[spotbird]

6. How about letters to the editors about 18,181?

http://www.cleveland.com/news/plaindealer/index.ssf?/base/news/1059125554154350.xml

http://www.sunspot.net/bal-te.md.vote25jul25,0,5882973.story?coll=bal-home-headlines

The black box story is starting to get legs but there are no questions from places where it has been used. In fact these stories report that everything was fine there. The 18,181 story is certainly the hardest to explain and everyone can understand it. Letters to the editor at least get that part of the story in print.

[ShaneGR]

8. Question about polls

Are you talking about exit polling or pre-election polling? If it's pre-election polling then the data is, unfortunately, irrelevent. Why? Because we suffered from horrible turnout in 2002. While someone called a few days before may have voiced their support of a Dem candidate, it doesn't mean they actually showed up to vote.

[Eloriel]

9. I don't understand

a thing -- or, more accurately, my eyes glaze over when someone pulls strings and strings of numbers out.

However, I think this deserves a :kick:

Eloriel

[TruthIsAll]

10. In 2000, fraud in broad daylight; in 2002, fraud in Cyberspace...

I have posted this analysis at least 4 times on DU, and am doing so again in light of Bev Harris' investigation, which has officially proved that the Diebold voting machines could have been easily hacked. Now, who here at DU believes for one moment that they were not?

If Bush can lie about Nigergate and tell us that Saddam would not allow inspectors in Iraq, then we can assume he is capable of fixing the elections to gain control of the Senate.

As a programmer analyst, math graduate (3 degrees) and student of politics going back to Eisenhower, I am absolutely convinced the Sebnate elections were stolen. They did it in broad daylight in 2000; they did it in Cyberspace in 2002.

Look at the vast discrepancies: an average switch in the eight hot races of 6%+; and a whopping 11% in the four races which switched to the GOP way beyond the MOE. I computed probabilities for deviations greater than the MOE; but we have 4 states which were so far beyond the 6% cutoff that the analysis is conservative, in my view.

The conservative odds, based on the analysis, is 43,040 to 1 that the elections were fixed; the true odds are much greater.

I don't buy the argument of low dem voter turnout; it's a canard. The votes were swithced to make it appear so. Why should the turnout in these elections be any different from others? There is no valid reason other than F-R-A-U-D. Bush needed the Senate. If he could lie to take us into a war in which thousands died, he is capable of rigging the votin machines to win the Senate.

To Zogby and all you other pollsters: Are you listening? It's past time. Zogby has already said that he never saw anything like this turnaround fiasco. The results are truly unprecedented.

So here is your chance, dera media and dear pollsters, to confirm or refute the analysis. Bev Harris has proved that hacking the Diebold machines is possible.

The laws of probability and statistics as they relate to polling provides circumstantial evidence of fraud, solid enough so that they could be cited in a court of law.

Let's now hear it from the media. Bev has done more work than all you so-called investigative reporters out there. Now its your turn.

[TruthIsAll]

11. I love the smell of the Cumulative Distribution Function in the evening

...

[ISUGRADIA]

12. This seems to be questionable statistical analysis

1) How were undecided % allocated among the candidates? It does not seem to have been. It's not unheard of for the undecided to break towards one party during a year when there is a partisan trend, examples: 1980, 1994 GOP, 1974, 1986 DEMS.

2) Where is South Dakota? It was obviously among the top 8 Senate races, more than Texas at least.

3) While on TX, where did these polls come from? Were these all the last polls of the race? Are they all Zogby? All polls I saw for Texas in the last week were showing Cornyn winning fairly easily. The last Zogby poll put Cornyn ahead 50-46 11/3-11/4. The one you may be citing with the race at 49-48 was from 10/30-11/2. It would seem to compromise your results this using of outdated poll numbers if that is the case.

4) How are other polls (different polling firms) accounted for in this analysis? Or were the ones showing the Dem in the best position used? The Minnesota one was at least a week old and the poll released by the Star Tribune on 11-3 had the race at 46% Mondale, 41% Coleman, a dead heat given the 3.2% MOE. Zogby put it at Mondale 51%, Coleman 45% on 11/3-11/4.

5) Why is a 3% MOE assumed for all these polls? Why not the actual MOE?

In closing, polls are just a snapshot of the electorate at the moment and I do not think given the number of variables that are not controlled in this analysis that you can state that statistics alone prove there fraud.

The Republicans had a good 2002 election for various reasons in the Senate just as the Dems did in 2000 (See FL, DE, MI, and WA that year, all close, all DEM in the end).

Anyone who cares can see some of the last polling numbers for some of these races at: http://www.realclearpolitics.com/Congressional/Senate_02_Polls.html

[althecat]

13. Zogby did say Georgia was off the scale... he said it here...

Pollsters defend their surveys in wake of upsets for more coverage of this issuehttp://www.coxnews.com/cox/news//cns/overview.html/stories/2002/1107-POLL.html

Which is sadly now a dead link... it would be good to pull this back up into the collective consciousness...

*****

The full details of the Scoop analysis follow below. In summary Scoop found:


- 14 races showed a post opinion poll swing towards the Republican Party (by between 3 and 16 points);
- 2 races showed a post opinion poll swing towards the Democratic Party (by 2 and 4 points);
- In three races the pollsters were close to correct;
- The largest post opinion poll vote swings occurred in Minnesota and Georgia where pollsters got the final result wrong (see… Pollsters defend their surveys in wake of upsets for more coverage of this issue);

and for the benefit of the skeptic above...

While I am not a mathmetician. To me this looks crooked.. It was crooked then and it is crooked now.

All the workings for the above analysis are here...

http://www.scoop.co.nz/mason/stories/HL0211/S00078.htm

[TruthIsAll]

14. Diebold + Simple Mathematics = ?

THE PURPOSE OF THE ANALYSIS
A statistical analysis of the Senate 2002 elections has employed basic probability theory to determine the odds that four out of eight hotly contested Senate races would dramatically turn from the Democrat to the Republican, based on the latest polling numbers taken just prior to the election.

These results should be taken in the context of the paper "Analysis of an Electronic Voting System", written by four computer scientists. They assert that their analysis of the Diebold electronic voting system shows that "it is far below even the most minimal security standards applicable in other contexts".

The statistical analysis does NOT constitute proof of fraud, but nevertheless are highly incriminating as circumstantial evidence.

THE POLLING MARGIN OF ERROR
The probability that a given state poll will be correct within the +/- 3% margin of error (MOE) is 19 out of 20 or 95%.

The odds that 4 out of 8 elections would fall outside the MOE (and ALL go for the Repubs) is 1 out of 43,040.

THE STATES IN QUESTION
The four states which experienced these remarkable turnarounds were:

1-Minnesota: The Democrats were leading 47-39% in the final polls; the Republicans won by 50-47%, an 11 point switch.

2-Georgia: The Democrats were leading 49-44% in the final polls; the Republicans won by 53-46%, a 12 point switch.

3-Texas: The Democrats were trailing by 48-49% in the final polls; the Republicans won by 55-43%, an 11 point switch.

4-New Hampshire: The Democrats were leading by 46-40% in the final polls; the Republicans won by 51-47%, a 10 point switch.

Each of these races turned around with deviations significantly beyond the 6% margin of error range.

Out of 8 races, the probability that ALL would fall WITHIN the margin of error =66.3%, or 2 out of 3. Stated another way, the chances that 1 or more states would fall OUTSIDE the MOE =33.7%, or 1 out of 3.

THE CUMULATIVE DISTRIBUTION PROBABILITY FUNCTION
The statistical analysis utilized the Cumulative Binomial Distribution function. This function computes the probability that there would be at least (N) successes in a series of (T) independent trials, where (P) is the probability of success in any trial. For any values of N,T,and P the Probability is calculated using the Excel Function: =BINOMDIST(N,T,P,TRUE)

For the case of 4 out of 8, with 95% probability
= BINOMDIST(4,8,.95,TRUE)= 0.000371751

In addition, the probability is 1/2 (50%) that any given election falling outside the MOE would go for the Democrat (or Republican). Therefore, the probability that ALL four would fall for the Republican is the product 1/2*1/2*1/2*1/2= 1/16

SUMMARY OF RESULTS
The odds that 4 out of 8 hotly contested Senate elections would fall outside the MOE (all for the Repubs) is the Joint Probability: 1/16*.000371751= .0000232, or 1 out of 43,040.

Assuming that 16 elections were hotly contested, The odds that 4 out of 16 elections would fall outside the MOE (all for the Repubs) is
1 out of 2,284.

Assuming that ALL 34 elections were hotly contested, the odds that 4 out of 34 elections would fall outside the MOE(all for the Repubs) is
1 out of 182.

[ISUGRADIA]

16. I'll just point out one inaccuracy using a state, NH, as an example

It was stated that in New Hampshire "The Democrats were leading by 46-40% in the final polls; the Republicans won by 51-47%, a 10 point switch"

Looking at site I mentioned in my earlier post, here are the final polls listed for NH:

Polling Firm Date GOP DEM UNDECIDED SPREAD
ARG 11/2-11/3 48% 44% 8% GOP +4
Univ of NH 10/30-11/2 47% 46% 7% GOP +1
Concord Monitor 10/29-10/31 46% 47% 7% Dem +1
ARG 10/28-10/30 48% 46% 6% GOP +2

FPC/WNDS-TV 10/27-10/30 40% 45% 15%Dem +5
Univ of NH 10/23-10/28 42% 46% 12% Dem +4

So let's take the ARG poll, the last one, factor in the undecided breaking the same way the poll did (48% GOP 44% DEM = 92%; 48% is 52.17% of 92%, 44% is 47.83% of 92%) thus GOP gets 4.2% of the undecided and Dem gets 3.8% of the Undecided. Add this to the poll numbers to get the following:

NH Senate
GOP 52.2%
DEM 47.8%
which is pretty close to the actual numbers if you take out 3rd party candidates and look at the two major party results only:

Sununnu GOP 227,229 52.27%
Shaheen DEM 207,478 47.73%

So NH did not fall outside the margain of error in an analysis of several competing polls of the end of election period.

Look at the last 4 polls, the actual result was reasonable with the margain of error in all four polls. How do you explain these polls away?

The point of the analysis that started this discussion was "... to determine the odds that four out of eight hotly contested Senate races would dramatically turn from the Democrat to the Republican, based on the latest polling numbers taken just prior to the election."

If you are only using one set of polling data before the election (that which is most favorable to Democrats) and not factoring any other polling data (Neutral or favorable towards Republicans) I don't see how you can say this is an honest analysis of the polls.

If your point is to say the polls showed Democrats should have won more of these races and that there was the possibility that elections were stolen by tabulator tampering, I don't think you can without examining all the polls relating to a race. Some of these races were just close looking at the assorted polling results and as in NH did not"dramatically turn from the Democrat to the Republican"

Dems sometimes lose because of bad campaigns, national trends and momentum, local events, which a straight stastical analysis of flawed numbers may not take into account.

[TruthIsAll]

17. A reply to your very legitimate comments...

Even using the poll results you justfiably cite for MN and TX, the change is still beyond the MOE (+/-3%)

Using your numbers,
...Dem Rep...Dem Rep
MN 51 45 6 47 50 -3 -9
TX 46 50 -4 43 55 -12 -8

The changes for the Republicans were 9% and 8%, beyond the 6% range.

As far as NH is concerned, you are correct. The SCOOP site does not show the polling numbers I used when I did this analysis last November, and I cannot explain why. They were there originally, but they are not there now.

As far as SD was concerned, you are right, it should have been included.

So let's assume that only 3 out of 8 execeeded the MOE, taking out NH.
From the Joint Probability table, we see that .00072 is the probability of this occurrence, or 1 out of 1,382. This is still quite remote.

N........3........4.......5.......6
8 0.579% 0.037% 0.002% 0.000%
10 1.150% 0.103% 0.006% 0.000%
13 2.451% 0.310% 0.029% 0.002%
16 4.294% 0.700% 0.086% 0.008%
20 7.548% 1.590% 0.257% 0.033%
25 12.711% 3.409% 0.716% 0.121%
30 18.782% 6.077% 1.564% 0.328%
34 24.065% 8.813% 2.592% 0.627%

The probability that all the races would fall for the Republican:
Prob (all N) = Repub
.........3......4.........5......6
12.50% 6.25% 3.13% 1.56%


Therefore, the Joint Prob (Outside MOE and all Repub)

N........3........4.......5.......6
8 0.072% 0.002% 0.000% 0.000%
10 0.144% 0.006% 0.000% 0.000%
13 0.306% 0.019% 0.001% 0.000%
16 0.537% 0.044% 0.003% 0.000%
20 0.944% 0.099% 0.008% 0.001%
25 1.589% 0.213% 0.022% 0.002%
30 2.348% 0.380% 0.049% 0.005%
34 3.008% 0.551% 0.081% 0.010%

Odds =1 out of X
N.........3........4.......5.............6
8 1,382 43,040 2,077,263 159,672,547
10 695 15,557 502,435 23,234,010
13 326 5,156 111,666 3,240,553
16 186 2,284 37,326 791,105
20 106 1,006 12,432 194,355
25 63 469 4,466 52,763
30 43 263 2,047 19,497
34 33 182 1,235 10,208

[PaDUer]

15. Utah

should also be checked closely..lot of strange musings there..