i'm a fan of your work--

what do you think of Nate Silver?

he seems to poohpooh exit polls. has anyone written a critique of him?

Welcome to PI.

If I may ask, when and where did you become familiar with my posts (DU?). Please PM.

Before I get started, I just put this new Excel model online.
http://www.geocities.com/electionmodel/TrueVoteCalculator.htm
This 200k model calculates the 2004/2008 true vote for all states.

You may already be familiar with the old Election Calculator. It calculates the True vote for all elections since 1988 as well as for 2004 NY, OH, FL etc.

The new True Vote Calculator is only one sheet, but is quite powerful in that it calculates the True Vote instantly. Just input 2004 or 2008 and that's it.

I have always questioned Nate's pre-election (Monte Carlo simulation)and post-election analysis (exit poll). His knowledge of modeling (and exit polls) appears to be limited. Here are some reasons I have come to that conclusion:

1. Pre-election polls. Nate misses the #1 rule: KISS (KEEP IT SIMPLE STUPID). He tried to incorporate pollster rankings into his model. That was not only overkill; it was wrong. He gives a high rank to pollsters who were close to the bogus 2004 vote. In 2000, Rasmussen had the worst performance. Zogby and Harris were nearly perfect. In any case, it's a moot point because the rankings add unnecessary complexity and bias to the model

2. His model did not allocate undecided voters to Obama, who was the de-facto challenger. Most pollsters gave 60-90% to Obama. Gallup gave 90% of undecideds to Kerry in 2004, bringing him from 47-49 to 49-49.

3. He tried to project the election 3 months out. He was going to correct due to the inevitable "tightening" in favor of the GOP. But the tightening is mostly due to replacing RV polls with LVs (RCP listed the final 15 LV polls, not one RV). It makes more sense to assuming the election is held today.

4. His simulated EV probability was not compatible with the projected vote share. With a 52% share, you have close to a 95% EV win probability. For instance, my final Election Model projection gave Obama a 100% EV win probability (so did others) when he won all 5000 election trials. Nate low-balled Obama's win probability in the 70-80% range, and only reached 90% a eek or so betore the election-too conservative. More importantly, the low prob. indicated that his simulation methodology was skewed. He was informed to that effect by others and some folks at Columbia U. adjusted the model.

5. Exit polls. Nate is no different than all the media pundits who have it exactly backwards. They dismiss the accuracy of the early, unadjusted exit polls, yet quote the bogus Final Exit poll- despite the fact that the Final is always forced to match the vote count using impossible weights and implausible vote shares. I have shown that this was the case in the 2004, 2006, 2008 elections. All had a 4% GOP fraud factor.

6. Nate never discusses election fraud. That's a good way to get on Olbermann and Maddow.

7. Nate does not use sensitivity analysis to stress-test his model. But who does? I do it all the time. He's an accountant, not a mathematician. Experts such as Freeman, Baiman, Palast, Dopp and myself all have advanced math/systems degrees. Independent and diverse analysis confirms that the exit polls are close to the truth and verifies the math.

8. Kerry won by 8-10 million; the Dems won more than 30 seats in 2006; Obama won by over 20 million votes (see the Election Calculator model). Nate has not done his homework on the 2004,2006,2008 elections.

9. My 2004 Election Model matched the unadjusted exit polls (51.3% to 51.2%).

10. My 2006 Generic poll trend model exactly matched the unadjusted exit polls (56.4 to 56.4%)

11. My 2008 Election Model EV (365.3 expected) exactly matched the result (365) and the 53.1% projected vote share was within 0.2% of the recorded 52.9%...BUT IT WAS WRONG.

12. OBAMA WON THE TRUE VOTE BY OVER 20 MILLION WITH A 56-58% SHARE. THE BIASED GOP RASMUSSEN, FOX AND BATTLEGROUND TRACKING POLLS BROUGHT DOWN OBAMA'S AVERAGE SHARE. I ASSUMED A N=BASE CASE 60% UNDECIDED ALLOCATION TO OBAMA. IT WAS TOO CONSERVATIVE. WITH 75% UVA, OBAMA WOULD HAVE HAD A 380 EXPECTED EV.

13. The final state polls were mostly LV (likely voter) subsets of the full RV(registered voter) polls RV polls WOULD HAVE ADDED 3-4% TO OBAMA'S SHARE - AND OVER 50 ELECTORAL VOTES....

14. THE EXIT POLLSTERS HAVE NOT YET RELEASED THEIR 2008 ELECTION REPORT WHICH WILL ONCE AGAIN SHOW THAT THE DEMOCRATS DID MUCH BETTER THAN THE RECORDED VOTE. BUT THE PUNDITS WILL ONCE AGAIN REPEAT THE MANTRA THAT THE EXIT POLLS ARE "CRAP". AND EVERYONE WILL CONTINUE TO BE MISLED BY PEOPLE LIKE NATE SILVER.

Oh, one other thing. I'm writing a book for the layman (and the pundits) on election analysis. I will also talk about my interesting experience as a spreadsheet-wielding conspiracy "math guy" on the Net.

Thanks for this response!

Will pm shortly.

2. 90% !
that's amazing.

3. yeah i noticed he does that a lot. for example, he posted a comparison of polls regarding cannabis legalization, trending the lines, and proclaiming that cannabis will be legal in x years. totally ignoring how fast political views can change in either direction due to events.

4. not surprised. what or who was the help from columbia? iirc, nate is a u of chicago product.

5. yes, the end justifies the means.

6. bingo. cue the liberal swooning.

7. i did not know palast had that degree.

8. he's never analyzed 2008?!

13. they haven't?! incredible.


looking forward to your book!