Wednesday, April 4, 2012

Results from Wisconsin and Maryland.
Down goes Silver! Down goes Silver! UPDATED WITH GRAPHS!

To describe last night's prediction contest, I paraphrase the most famous boxing call in history, when Howard Cosell at ringside yelled "Down goes Frazier! Down goes Frazier!"  Absolutely NO disrespect to the great champion (here's a link to my remembrance of him), but what my system did last night to Nate Silver's system is what George Foreman did to Frazier all those years ago.

I could say it was close or I was lucky, but neither of those statements are true.  My system and Nate Silver's system are both good in their own ways, but in Wisconsin, a bad late poll was released and both of our systems took it into account.  My system barely noticed its ill effects, but it knocked Nate's system on its ass.  Here are the final numbers.

Romney: 42.5% (I guess too high, Nate too low. I gain 0.6%)
Santorum: 37.6% (Both too low. I gain 2.9%)
Paul: 11.7% (I'm too low, Nate's too high. I gain 1.2%)
Gingrich: 6.1% (Both too high. I gain 3.8%)
None of the Above: 2.1% (Both too low. I gain 1.6%)
Final score: Hubbard 93.2%, Silver 83.4%

Nate actually lost ground in both his Gingrich pick and his NOTA pick, which is why his column has no gray and no black.  This is also the reason his column is much, much shorter than mine.  This contest was the biggest butt-kicking so far in our sixteen skirmishes so far.

Romney: 49.1%(both high, I gain 2.8%)
Santorum: 28.9%(Both low, I gain 0.5%)
Gingrich: 10.9%(I'm high, Nate's low, Nate gains 0.1%)
 Paul: 9.5% (Both low, Silver gains 0.3%)
None of the above: 1.6% (Both of us low, I gain 1.5%)
Final score: Hubbard 97.0%, Silver 92.6%

With this graph you see we both got over the 90% mark, always a good sign, but I did much better on predicting Romney (marked in green for money, of course) and that makes up most of the difference in our scores.

The long and short of it.  In overall contests, I now lead 10-6, not massively dominating but not just luck, either. (One standard deviation ahead for my mathy readers.)  Last night, my system got the biggest margin of victory yet, a 9.8% win in Wisconsin, and the most precise poll yet, 97% accurate in Maryland.

I could correctly say I was lucky, since any time you try to predict the future you are at the mercy of a random world. But I'm going eschew false modesty today and say that my luck last night was the result of a better prepared system created by a better trained mind.

The things that make my system better are pretty simple.  I factor in the median as well as the average to keep outlier polls from knocking everything out of whack. I'm looking at averages after a reasonable None of the Above vote is factored in, while Nate is trying to extrapolate trends over time. My way is simpler and more mathematically sound, especially with just a few data points like we had in Maryland, a total of two polls.

So I have a lead right now of 10-6 and it's three weeks until several East Coast states have their primaries on April 24. It would be silly to think my system is fated to have an equally good evening later this month, but since I opened this up paraphrasing Howard Cosell, let me once again show my age by paraphrasing Mel Brooks, a line from the crazy playwright in The Producers.

I'm a mathematician. Nate's a statistician. I outrank him.



Karlacita! said...

Booyeah! You're the winner!

Math is awesome. #mathematiciansoutrankyou

BobManDo said...

...Congrats on the win... and here is something you might want to dig flipping fraud... clearly explained in this document:
Titled: "Evidence of Algorithm Vote Flipping in GOP Primary Elections Layman's Executive Summary."
Matty Boy, I'd love your analysis of this... Thanks! -BobManDo

Matty Boy said...

Hi, Bob. I read the paper and a lot of what he says makes no sense. It sounds like a Ronulan cherry picking data to fit with the conspiracies that sound so good in his head.

Does marijuana make you paranoid? I've smoked so much I forget.

(/sarcasm I have smoked very little in my lifetime.)