BACKTESTING THE 2000 ELECTION: PROOF THE ELECTION MODEL WORKS!

[TruthIsAll]

I ran an experiment to check the accuracy and methodology of the model. I backtested the model by plugging in the final 2000 state percentages for Gore and Bush, as if they were current POLL NUMBERS.

I wanted to see how close the model would come to the actual 2000 selection, in the national voting percentages and final electoral votes. I was in for a pleasant shock. Here are the results:

In the 2000 selection, Gore had 48.38% vs. Bush 47.87%.
The rest went to Nader (2.74%) et al.

RESULTS OF THE TEST:
In the backtest simulation, Gore received 48.43% vs. Bush 47.83%.
The model was correct within .04%!

Gore also won the simulated election (5000 trials) with an average of 272 EV with a 55% win probability.

METHODOLOGY:
The model utilizes the average 2-party vote spread for each state (Dem and Repub only), based on the total of the last three (3) presidential elections. This way, we can determine the appropriate state voting weights to use in order to calculate the current national vote percentages for Kerry and Bush.

Right now, based on the latest state polls, the Election Model has Kerry leading Bush nationally by 47.80% - 46.65%, with a 99% win probability.

Just thought you would like to know this.

http://www.geocities.com/electionmodel/

[Merlin]

1. Congrats, TIA. You may have yourself a new career!

Sure hope your methodology holds up thru 11/2.

[TruthIsAll]

6. Kerry might even carry Lancaster, PA.

heheheheheh

[Merlin]

16. Tell you what. If he does, I'll buy you a brand new Mercedes.

Not that I can afford one. But I am not a bit worried.

[Lex]

2. Thanks TIA.

This is really GREAT news about the model you are running!

I look forward to your posts!

Thank you.

[Wapsie B]

3. TIA I always look forward to your posts.

Very informative and uplifting.

:yourock:

[autorank]

4. Nice Discriminant Function Truth. UDAMAN!

At least I hope it's a discriminant function, sounds like it but its been years...you rock!

[ciaobox]

5. I know I am biased but...

We really are just so much smarter than Conservatives.

Great job! :kick:

[dolstein]

7. Sorry, but predicting the results of past elections is easy

Predicting the results of future elections is somewhat more difficult.

[TruthIsAll]

8. As usual, Dolstein, you missed the point.

I am only saying that the METHODOLOGY is correct in the application of state weighting, based on the last three elections. That was the purpose of the backtest.

I am not predicting anything, other than to say that right now the odds favor Kerry - by 98%. The odds change daily.

I'll wait until the weekend before the election to make a prediction.

[5thGenDemocrat]

9. Fine work, TIA

I've gotta say -- I think your numbers are at least as accurate as any other methodology I've seen this year and better than most. Thank you for the hard work and know that it's appreciated.
John
Who lives in a very safe Dem city in a pretty safe county in a fairly safe state (Saginaw, Michigan -- now, as always, The Center of the Universe).

[fujiyama]

10. Wow

That's pretty cool. That's very close to the real results.

BTW, which date's polls was the test using? Were these polls taken a week before the election or just a day or two before? Which firms did you use in the test? It seems as though there are more polling firms and agencies this year. What about the state polls? I thought you were using the data from electoral-vote.com this time?

Either way, let's hope that Bush is toast.

[TruthIsAll]

15. I plugged in the actual results, pretending they were polls.

The purpose of the test was to compare the simulation model to the actual results of the 2000 election. I used the same proportional state voting-weights (from the last 3 elections), combined with the Gore/Bush actual percentage split in each state. I compared the output of the model to the actual Gore/Bush national voting split.

The model came within .04% of the actuals and calculated an average 272 EV for Gore. That's for 5000 simulated election trials.

THIS WAS NOT A TEST OF THE MODEL TO SEE HOW CLOSE IT WOULD COME TO THE ACTUAL ELECTION BASED ON POLL NUMBERS.

THIS WAS A TEST TO SEE IF THE MODEL METHODOLOGY IS VALID BASED ON THE ACTUAL GORE/BUSH STATE SPLITS.

i WANTED TO SEE IF THE CALCULATED STATE-WEIGHTED NATIONAL VOTE, BASED ON THE LATEST STATE POLLS IS ACCURATE. THE TEST PROVES IT IS.

IN ADDITION, I WANTED TO SEE IF THE SIMULATED GORE ELECTORAL VOTE MATCHED UP WITH HIS ACTUAL. IT DID.

AGAIN, THIS WAS NOT A TEST TO SEE HOW ACCURATE THE MODEL IS.

BUT WE CAN SAY THIS: IF KERRY GETS 60% OF THE UNDECIDED VOTE, BASED ON CURRENT UNBIASED POLLS, HE HAS A 98% PROBABILITY OF WINNING THE ELECTION WITH 323 EV.

[oasis]

11. Thanks. Keep up the great work.

:thumbsup:

[Awsi Dooger]

12. Here's a link with the state polls from 2000

http://www.duke.edu/~rml5%20/RA/Aldrich/wh2000gen_state4.htm

Might be helpful if you want to backfit using old poll numbers and see what they would have forecast at the time, using your election model. Also a good way to evaluate various pollsters, and in which direction polls tend to err from state to state.

[JCMach1]

13. Give '96 or '92 a try as well...

[TruthIsAll]

14. Kick from work. Will update model around 7 PM

tia