Wednesday, November 29, 2006

Correction on Wells

So, Paul Wells has on his blog a little note I wrote last September/October and sent to him last night after we shared a pint. He basically suggested that everyone's predictions about the outcome of the Liberal leadership were just full of wild assumptions and countless unknowns. And he's quite right. So, I sent him the memo I drafted which I think demonstrates that if you were betting on the convention with no knowledge of the preceding campaign, but a knowledge of every other leadership convention that's occurred, then you have to bet on Ignatieff. History is quite clearly in his favour. And I think the memo demonstrates why, particularly that the ratio of his support to the second place candidate, Bob Rae, is quite high. However, one statistic I did include is incorrect.

I stated that "In the history of federal leadership conventions, the largest first ballot first place showing of an eventual loser was Claude Wagner (22.5). Michael Ignatieff is fully 7.5 points above that number." This is incorrect. In 1983, Joe Clark scored 36% on the first ballot, only to lose to Mulroney three ballots later. I missed this case, though it is in my dataset. So, I was wrong. Fortunately, I think it does just about nothing to take away from the core argument, for two reasons:

* 1983 was not really an open convention so much as the defense of a leadership and a referendum on Clark, which suggests delegates where much more polarized. This is a substantive difference, but an important one nonetheless.
* More importantly, Clark's ratio to Mulroney (1.24) was much smaller than Ignatieff's, and I think that remains the most important number. This means Mulroney needed much less time (i.e. ballots) and coordination (i.e. the undiscounted support of other candidates) to push over the top. None of Ignatieff's opponents have that in this case.

So, I apologize profusely for the mistake, and I shall be more slope-shouldered than usual this evening. But the facts remain.

18 comments:

Anonymous said...

Maybe. But one should recall that Clark in 1976 only got 11.7% on the first ballot, and Wagner's margin over him was 1.9.

Peter Loewen said...

Yes, fair enough. But the more relevant number, I think, is that Wagner was just 1.48 to Mulroney. In that case, the pissing contest between he and Mulroney allowed Clark time to assemble his coalition.

Anyways, I've no particular stake in this. I just call the numbers as I see them.

As a post-script, I should note how rare Clark's performance was. His is the lowest first ballot percentage of a winner of all time.

Antonio G said...

There's another factor not being considered here, which is that the delegate selection methood is quite different in this leadership race from any previous one. In all the earlier federal races where results were not decided on the first ballot, delegates were elected on a winner-takes-all basis at each rising or association. This contest is done on a proportional-representation basis at each association. The dynamics are very, very different. The only really relevant precedent is the Ontario convention where McGuinty was elected. And that turned out very differently indeed - I believe he was 4th on the first ballot, no?

So if Iggy loses, it may well be historic - but it will be because of a change in delegate-selection rules, not because he's a uniquely bad candidate.

Anonymous said...

No, Iggy is a "uniquely bad candidate."

CuriosityCat said...

If the Clark convention was "polarized", and this was a factor, what about the polarization of the Ignatieff campaign? Second choice of only 6% (lowest of the Big 4). In favour of the nation thingey (while more than 70% of Canadians oppose it). In favour of Iraq (while Liberal Party rejected that war). Dug a fissure into Liberal support for Israel with his Qana war crimes comment. In favour of preemptive war and coercive interrogation of suspected or actual terrorists, while most Canadians support the Charter of Rights.

Talk about polarization ....

Gotta be a factor in this convention.

Paul said...

Corrections are all well and good. But most statistical inferences are only valid within a sufficiently large universe: exactly how large is your dataset? How many comparable conventions did you study? (For the sake of your credibility, you might wish to list them.)

You might as well posit the hypothesis that the leader in the polls will win x% (say, 90%) of the time. Trivial and useless, but at least correct.

Anonymous said...

Fascinating, we'll know in a few days how your theory holds up.

Peter Loewen said...

These are all good comments to come back to. Let me address them in turn.

First, I think the allotment of delegates proportionally certainly changes strategy pre-delegate election. But after the selection of delegates I just don't think it is a meaningful difference. The only difference now, at this stage, is that delegates are distributed more evenly across ridings for any given candidate. But I think this has little effect on convention floor dynamics.

On polarization, let me make two points. First, there were no unknowns with Clark. His leadership abilities were known, and I am inclined to think that opinion on him was quite crystallized. I just honestly don't think it's that bad with Iggy yet. And, as for poll numbers which suggest many would never choose him, I think these are fine, but I think they're also biased by question format quite substantially. From the private polling to which I am privy, Rae appears nearly as often as Ignatieff when respondents are asked to list all the candidates for whom they would never vote.

Finally, I agree in principle on the point about statistics and samples. My sample is 20 federal and provincial leadership conventions since 1960. It's a fine sample, though I'd be happy to see any contrary data from cases I've missed. I remain fairly confident in saying that a loss by Ignatieff would be quite unlikely, though nothing is out of the realm of possibility. Mostly, I am just trying to get people to think about these numbers in the most clear-eyed way possible, and this explains my search for relevant metric/statistics.

Anonymous said...

The real issue is the vote count of delegate votes on the “first” ballot does not include the 850 possible ex-officio delegates. So Iggy did not win 30% of the TOTAL delegates but 30% of the vote cast therefore if you look at the chart below he only won 24.8% of possible delegates.

How can you compare the total that does not include ex-officio votes - to totals at other conventions that included all the ex-officio delegates - and make any meaningful analysis?

You are comparing apples and oranges.
.






IG
1377
29.3
24.8

Rae
943
20.1
17.0

Ken
820
17.5
14.8

Dion
753
16.0
13.6

Dy
238
5.1
4.3

Vl
226
4.8
4.1

Br
181
3.9
3.3

Fin
46
1.0
0.8

Un
112
2.4
2.0

4696
4696



Exofficio
850
18.1
15.3

5546
5546

Peter Loewen said...

Yes, there are certainly a large number of unknown delegates right now given the ex-officios. I am just trying to make an analysis given what we know now.

If you'd prefer, wait until the first ballot is announced. If Ignatieff's lead ration persists with the ex-officio support, then I think he is the favourite, based on history.

Incidentally, your second percentages are more or less meaningless. They are the apples to oranges. You can't just add into the denominator by including ex-officios and not include them in the numerator and then expect to make a meaningful analysis given past results.

Anonymous said...

Are you using the final results from the delegate selection meetings? For what it's worth, by my calculation the Ignatieff to Rae ratio is 1.46 (1377/943)

Peter Loewen said...

No, the memo which i sent Wells was written in October immediately after the meetings.

hosertohoosier said...

Walter Harris lost to John Wintermeyer in the 1958 Ontario Liberal Leadership convention, despite a first ballot performance of 39% (the number 2 was stronger than Rae though, with 34%) - outside of your sample admittedly.

Larry Grossman was at 45% on the first ballot and came within inches of losing in 1985.

Also, if you want to increase your N, here, take cases after the first ballot as well. It seems to me that there are many comparable cases to the present Liberal race that may have appeared on the second, third or eigth ballot. The dynamics, however, are the same.

Oh and the worst violation of your case is the 1995 NDP leadership race, where Robinson lost with 37% of delegate support.

Peter Loewen said...

The Svend case is a tough one, as it's unclear whether he dropped off because he would lose or because he knew that if he squeaked out a victory it would be for a bitterly divided party.

I do include the 1958 Liberal example in my memo. If you read the Wells post, it is included in the second point.

Agreed on the relevancy of later ballots. One can spend all day looking for disconfirming cases or for new interpretations of the data, and that's a fair enough one.

Anonymous said...

To elabourate on my question above - In your original memo, you said: "Michael Ignatieff has passed the crucial 'ratio' threshold. He has 1.6 delegates for each delegate which his next closest competitor has won. In the history of federal leadership conventions, no one has lost with this large a first ballot lead."

But as discussed above, the Ignatieff:Rae ratio is 1.46 (based on committed delegates). And in 1976, you note that the Wagner:Mulroney ratio was 1.48. Isn't this an example of a higher (well, effectively equivalent) ratio for a leading candidate who went on to lose? In other words are you still drawing the same conclusion with the revised numbers? Of course we'll have the real first ballot numbers soon...

Peter Loewen said...

Yes, you are quite right that with the revised, up to date numbers, Ignatieff is more vulnerable. that said, my sources on his turnout numbers suggest he should come back close to that ratio (1.6). So, tomorrow should be an immensely interesting day.

Please do keep in mind that I am not shilling for anybody here - i am just trying to point out historical realities. And you've just done the same, so thanks.

Anonymous said...

i think in your apology you meant to say slope shouldered, a sloop is a sail boat with dual rigging and a particular mast placement.

alex mackenzie

Peter Loewen said...

Right you are. I actually meant to say that I was Nonesuch-shouldered.