"Facts are meaningless. You could use facts to prove anything that's even remotely true!"
-Homer J. Simpson

Tuesday, October 03, 2006

Ranking the Blogs' Liberal Leadership Prediction Abilities

I have all my predictions and analysis of the Liberal Leadership Race here for you to review

As I said in one of my earlier posts I believe in the wisdom of crowds. So my technique for predicting the Liberal Leadership Delegate Numbers was to take all of the analysis of the others that I could find and then average them together. This would act as a "stock market" or averaging of all of the sources of Data.

For interest's sake, as both an analysis of this method of predicting political outcomes, and as a bit of a hat tip to the people who gave me the data, I present:

The Prognosticator Award - Liberal Leadership Delegate Selection Division

So I will list each of the people/organizations who's work I used. I will list their estimate, followed by the actual results, then the average deviation from the results and then my analysis of their prognostication abilities :-)

Entry 1: Cerberus


Actual Prediction Diff




Ignatieff 29.80% 24.00% 5.80%
Kennedy 16.80% 19.00% -2.20%
Rae 19.80% 19.00% 0.80%
Dion 16.60% 14.00% 2.60%
Dryden 4.60% 8.00% -3.40%
Brison 3.90% 7.00% -3.10%
Volpe 4.60% 6.00% -1.40%
Hall-Findlay 1.00% 3.00% -2.00%

Average deviation: 2.7%
Worst Estimate: 5.8% underestimate for Ignatieff

Cerberus had pretty good numbers but (like most) he underestimated Iggy (Which is odd, given that he is an Ignatieff supporter). Nothing to be ashamed of, but not quite our winner


Entry 2: Jason Cherniak (note that he gave ranges and I took the average of the range)


Actual Prediction Diff




Ignatieff 29.80% 27.50% 2.30%
Kennedy 16.80% 22.50% -5.70%
Rae 19.80% 15.00% 4.80%
Dion 16.60% 13.50% 3.10%
Dryden 4.60% 8.00% -3.40%
Brison 3.90% 9.00% -5.10%
Volpe 4.60% 6.00% -1.40%
Hall-Findlay 1.00% 3.00% -2.00%

Average deviation: 3.5%
Worst Estimate: 5.7% overestimate for Kennedy (even if you allow Jason his "range" Kennedy was still outside it)

Given that Jason was using ranges it is a bit unfair to him to use his average deviation against him, but......oh well :-) Still, as you will see, he did better than some

Entry 3: democraticSPACE

Update: (I was using earlier numbers from this site and somehow, due to my own stupidity, didn't use the most up-to-date numbers, Greg was quick to point out, and rightly so, that his final numbers were far more accurate than I was giving him credit for - I leave my initial numbers up becasue that is what is used for my calculations at the bottom, however, Greg's numbers that he included in the comments section are the ones you should use for him - he had the most accurate numbers and should have been awarded the winner - all I can say is...Whoops!
No slight was intended for Greg and I was foolish to think that my numerical ability was anything close to him , or many of the others making predictions )



Actual Prediction Diff




Ignatieff 29.80% 24.20% 5.60%
Kennedy 16.80% 16.90% -0.10%
Rae 19.80% 14.30% 5.50%
Dion 16.60% 13.50% 3.10%
Dryden 4.60% 9.20% -4.60%
Brison 3.90% 9.90% -6.00%
Volpe 4.60% 6.60% -2.00%
Hall-Findlay 1.00% 3.60% -2.60%

Average deviation: 3.6%
Worst Estimate: 6.0% oversupport for Brison

[section deleted due to my own stupidity as explained above]

OOPS!

Greg pointed out that he had more up-to-date numbers than the ones that I was using. I inlcude those here for you (from his comments):

IGGY 29.8 / 26.8 / 3.0
RAE 19.8 / 18.2 / 1.6
KENNEDY 16.8 / 15.1 / 1.7
DION / 16.6 / 16.4 / 0.2
DRYDEN / 4.6 / 9.1 / +4.5
BRISON / 3.9 / 5.8 / +1.9
VOLPE / 4.6 / 4.7 / +0.1
HALL FINDLAY / 1.0 / 2.5 / +1.5
Average deviation: 1.8
Worst Deviation: 4.5% over-support for Dryden

Much better showing for him then many of the others. In fact, if I had been paying attention then I would have had even better numbers in my avergaing anaysis.

mea culpa!


Entry 4: delegate.count



Actual Prediction Diff






Ignatieff 29.80% 33.11% -3.31%

Kennedy 16.80% 17.95% -1.15%

Rae 19.80% 14.38% 5.42%

Dion 16.60% 18.37% -1.77%

Dryden 4.60% 7.45% -2.85%

Brison 3.90% 4.96% -1.06%

Volpe 4.60% 2.33% 2.27%

Hall-Findlay 1.00% 1.45% -0.45%

Average deviation: 2.3%
Worst Estimate: 5.4% underestimation of Rae

More of a tracking site than a prediction site, but he did do some "regional adjsutments" and other stuff to predict the outcome. Actually did quite good. The only site to come close to predicting Iggy's suport. Best of all the blogger that we have seen so far (one is better, but you will have to wait)

Entry 5: The Strategic Counsel/CTV/Globe&Mail poll


Actual Prediction Diff




Ignatieff 29.80% 19.00% 10.80%
Kennedy 16.80% 9.00% 7.80%
Rae 19.80% 17.00% 2.80%
Dion 16.60% 13.00% 3.60%
Dryden 4.60% 9.00% -4.40%
Brison 3.90% 3.00% 0.90%
Volpe 4.60% 3.00% 1.60%
Hall-Findlay 1.00% 3.00% -2.00%

Average deviation: 4.2%
Worst Estimate: 10% underestimation of Ignatieff

One of two entries from the MSM and, as usual, they got it totaly wrong. Every blogger in the race beat them both on average and in almost every specific. Moral = blogs rule, polls suck.

Entry 6: The Gandalf Group (i.e. David Herle, the Liberal's Pollster/Campaign Manager)


Actual Prediction Diff




Ignatieff 29.80% 13.00% 16.80%
Kennedy 16.80% 7.00% 9.80%
Rae 19.80% 19.00% 0.80%
Dion 16.60% 8.00% 8.60%
Dryden 4.60% 19.00% -14.40%
Brison 3.90% 4.00% -0.10%
Volpe 4.60% 1.00% 3.60%
Hall-Findlay 1.00% 1.00% 0.00%

Average deviation: 6.73%
Worst Estimate: 16% underestimation of Ignatieff

This was just sad. Very, very sad. The Liberals should hire any of the blogger above (or our winner below) to be thier pollster for the next election.

Entry 7: Our Winner - Calgary Grit


Actual Prediction Diff




Ignatieff 29.80% 26.70% 3.10%
Kennedy 16.80% 17.70% -0.90%
Rae 19.80% 17.90% 1.90%
Dion 16.60% 14.60% 2.00%
Dryden 4.60% 7.20% -2.60%
Brison 3.90% 6.30% -2.40%
Volpe 4.60% 7.80% -3.20%
Hall-Findlay 1.00% 1.90% -0.90%

Average deviation: 2.2%
Worst Estimate: 3% underestimation of Ignatieff

Calgary Grit has the lowest deviation and a pretty low worst-guess. He was closer on Iggy than most as well (good, given that he is a Kennedy supporter) . He absolutly kicked the ass of the MSM abuot six ways from Sunday and edged out the rest of the bloggers.

Results of Averaging

So I took the average for each delegate percentage (see my work in the link at the top of the post) and the results confirm my methodology:


Actual Prediction Diff




Ignatieff 29.80% 25% 4.80%
Kennedy 16.80% 17% -0.20%
Rae 19.80% 19% 0.80%
Dion 16.60% 15% 1.60%
Dryden 4.60% 9% -4.40%
Brison 3.90% 6% -2.10%
Volpe 4.60% 5% -0.40%
Hall-Findlay 1.00% 2% -1.00%

Average deviation: 1.9%
Worst Estimate: 5% underestimation of Ignatieff

The wisdom of crowds resulted in the lowest deviation from the actual results and the only points that were out worse than any others were the Dryden/Iggy numbers.

Yes, this was the most accurate, but it was only due to the hard work and dilligance of the other bloggers. Hat tip to all those who put work into the process and gave us the numbers to play with that they did.