Monday, May 2, 2011

Final Projection: Conservative Minority

Canadian Election Watch's final projection was posted on the left sidebar around 6:45am EDT, and it calls for a strong Conservative minority government, with the NDP as the Official Opposition. I encourage you to visualize the projection with these maps (Update: now posted). The expected national seat count and vote share for each main party are as follows:

152 seats, 37.3% for the Conservative Party of Canada
94 seats, 30.5% for the
New Democratic Party of Canada
46 seats, 20.0% for the Liberal Party of Canada
15 seats, 6.3% for the Bloc Québécois
0 seats, 4.9% for the Green Party of Canada
1 seat held by an Independent

Because the Independent, André Arthur, normally supports the Conservatives, this projection implies that the Tories are just 2 seats short of a working majority. Here are some unscientific estimated probabilities of various events based on the record of similar models in past elections (not that the projection model is totally rigorous, but the probabilities are even more ad hoc):

Conservative working majority: 45%
New Democratic win (most seats): 2%
Liberal Official Opposition: 1%

Of course, if the Tories actually finish only two seats short, they could still form a majority if they manage to convince MPs from the Liberal party to cross the floor.

As a reminder, the projection model is a uniform swing model (dividing Canada into six regions) adjusted for riding polls and by-elections (such as Elizabeth May running in Central Nova last time). Additionally, the following tweaks are made:

- Different swings are applied to the Toronto Census Metropolitan Area and other areas in Ontario, based on EKOS data.
- The Conservative polling averages are slightly increased, and the Bloc's, Greens' and NDP's slightly decreased, as is consistent with past history and commitment measures. (The magnitude of these adjustments, around 0.5-1.5% in each case, is admittedly ad hoc.)

Both of these changes favour the Conservatives by a few seats, and account for most of the difference in the Conservative seat count between this projection and that of some other websites using a similar methodology.

At this point, it is important to remember that the idea behind a uniform swing model is that mistakes on one side and the other tend to cancel out. Thus, this model is not designed to correctly predict the winner is every riding. Rather, it is supposed to give a good idea of the expected total count for each party. As a result, the riding-specific projections should be considered unreliable (Atlantic, Québec, Ontario, West). But, hey, they're fun, and Lawrence Cannon, Peter MacKay and Gilles Duceppe are all projected to be swept away by the orange wave (though all these races are very tight, so none of the three may come to pass).

This final projection is a little different from the ones made throughout the campaign because it has an extra ingredient. Rather than just projecting as the winner whoever the model points to and adding up the ridings, I also took a look at how many close races each party is projected to win or lose. I then manually adjusted the outcome in certain ridings in order to come up with an aggregate projection with roughly equal upside and downside risks.

For example, the Bloc Québécois would have won only 5 seats on a uniform swing model. Because my model also includes riding polls and the regional effects they suggest, it had the Bloc winning 13 seats. However, the Bloc was still losing more close races than it was winning, so I shifted two additional seats to the Bloc from the NDP. The other beneficiaries of this adjustment are the Liberals, who get one seat from the NDP in Québec and one seat from the Conservatives in Ontario.

Furthermore, I took another look at Portneuf--Jacques-Cartier, which is handled outside the model due to the presence of an Independent incumbent. I decided to move that seat back into André Arthur's fold, though of course it is a tossup.

Had the above changes not been made, the final projection would have been 153 C, 98 N, 44 L, 13 B. This is what appears on the campaign trends graph (Update: now posted) in order to maintain consistency.

Later today, I will post a more detailed analysis of the projection, which could also serve as a quick (and incomplete) guide to election night.

4 comments:

Kelly Wilson said...

As to who crosses the floor can sometimes be a surprise, but do we have any idea of the sort of person to be particularly suspicious about?

Election Watcher said...

No clue, though I think there will be fewer than there might normally be under such a circumstance. Harper doesn't seem like a very pleasant boss. I wonder if Emerson would have crossed the floor if Harper had the reputation that he has now...

Still, two is definitely attainable. Perhaps somebody from the 905 or Atlantic Canada.

JeffS said...

Thanks for all your work on this. I have enjoyed reading the updates. What baffles me most about public polling here was the lack or riding-specific polls. We had a decent number from Quebec, but pretty much nothing from the GTA and Lower Mainland, where it seems the election will be determined. In the US we get an orgy of "swing state" polls, but no "swing/bellwether" riding polls in Canada. The parties must have them, therefore know a lot more than we do.

Election Watcher said...

Thanks Jeff. My guess is that riding polls are not cost effective for the media and the pollsters. You can poll 1000-2000 people and have the results reported all over the country, or you can poll 500 people and have the results reported in a city only.

The situation is a little different in Québec, where a riding poll can interest people across the province. This may be why we are seeing more of them there.