Sunday, September 26, 2010

Mind the Gap

As I've pointed out on several occasions, a full week has passed since the last days of polling. Tomorrow is voting day in New Brunswick - so how can we make an attempt at estimating voting behaviour when almost the last 1/4th of the campaign has not had any polling?

This is not the first time something like this has happened in New Brunswick. Corporate Research Associates has a policy of not polling in the last week of a campaign, and that policy still existed in 2006.

The last CRA poll during the 2006 campaign was taken on September 10th, with the vote taking place on September 18th. It's an identical situation to this electoral campaign: the last day of polling was September 19th, while the vote will be taking place eight days later on September 27th (tomorrow).

In that last CRA poll during the 2006 campaign, the Progressive Conservatives were at 42%. The Liberals were at 44% and the New Democrats were at 10%.

Eight days later, the parties actually got 48%, 47%, and 5%, respectively.

That isn't exactly on the mark. In the end, the PC vote increased by a factor of 1.143 and the Liberal vote by a factor of 1.068. The NDP vote eroded by a factor of 0.5.

Was that a polling error, or the permutations of the final stages of a campaign? The likely answer is a bit of both.

To determine how people will vote tomorrow, based on the last available CRA poll, we could apply the same adjustment that took place during the 2006 campaign. Of course, that campaign had its own issues and flashpoints, and using the 2006 campaign as our only point of data limits the predictive capability of this exercise. In truth, I'd consider that capability to be almost zero. But it is a fun little exercise nevertheless, and gives us a little preview of what could happen tomorrow.The last CRA poll in this campaign (the one I reported on a few days ago encompassing data from both the TJ and CBC polls) had the Progressive Conservatives at 43%, the Liberals at 38%, and the New Democrats at 11%.

Applying the same level of adjustment that took place in 2006, we end up with 49% for the Progressive Conservatives, 41% for the Liberals, and 6% for the New Democrats. Those are convenient numbers, as they leave four percentage points that we can distribute comfortably to the Greens and People's Alliance.

With those levels of support, I project that David Alward's party would win 36 seats and form a majority government. The Liberals would win 19 seats and the New Democrats would win none.

But, again, I ask you to take these numbers with a massive grain of salt. It is much more likely than not that the changes in support in the last week of the campaign will come about differently than they did four years ago.

1. Eric you took my projection. Thanks for all the good work on the NB election!

Regards and compliments,

Earl

2. Thanks Earl.

My projection came about independently, I assure you!

Really looking forward to tomorrow. This is the first time I get to test the model since the 2008 Quebec election. Because of the limitations in polling, I'm out on a longer limb than I normally will be, but it should be a good test and a way to see how I can improve the model.

Aside from the calculations, my gut also tells me I'll be right or close, so that's comforting.

3. In 2006, the NDP campaign in NB totally collapsed. They only had a partial slate of candidates, the leader flopped and apparently only 3 or 4 candidates even campaigned locally. So its not surprising that the vote fell to 5%. (but I think other polls in NB at the end of the campaign had them at 6 or 7%). Its a different story this time since the NDP has active campaigns in wayyyy more ridings. If you look back at the polling in previous NB elections, the NDP got about what the final polls projected. I think 2006 was very much an outlier.

4. I agree (though I did take into account the fact that not every NDP riding had a candidate, but with rounding to the nearest full number it didn't change anything), which is why I said people need to take this with a grain of salt.

5. Ed are you going to post your prediction?

6. I find it interesting to look at the projections but without much data it must be difficult to project the outcome. I wish I knew more about the subject, but I don't...what I do know is the NB voters tend to vote parties out. In the Caraquet riding, although it is liberal the anger towards that party is strong, and towards to PC, all because of lies (liberals) and the PC closed down the local very needed hospital. It could go any way, and might be more like in 1995 where the ndp & pc were neck in neck...and the liberals took it because of a popular politician. In 2003, the current liberal barely made it in by only 99 votes. I enjoy ..your work though Eric, thanks!

7. Here's my predictions:
PC: 48
Lib: 6
People's Alliance: 1
NDP: 0
Green: 0

- The PC's should be getting a landslide of seats.
- Liberal keep only a few in the northeast, and Shawn Ghaham will lose his seat.
- Kris Austin of PA may very likely win the Oromocto seat, because he may be the one protest seat.
- The NDP will win a seat again.

8. Éric,

Outstanding work. We are much in your debt. I want to pay you the highest compliment I can think of -- I hope that a sovereignist can one day see his way clear to work for federalists (without necessarily changing his personal convictions).

That federal government, institution or party would be so much richer if it counted you among its "numbers"!!!

Merci beaucoup.

9. It looks as though Eric's "Adjusted" numbers were the closest thing to an accurate prediction of the actual results.
Especially in low-energy campaigns, the momentum of the final week seems to make a huge difference.

Granted, I don't think anyone understands sociology or political science works well enough to explain why people for person X just because they think other people have switched to voting for person X, but there's no doubt that people do it.

COMMENT MODERATION POLICY - Please be respectful when commenting. If choosing to remain anonymous, please sign your comment with some sort of pseudonym to avoid confusion. Please do not use any derogatory terms for fellow commenters, parties, or politicians. Inflammatory and overly partisan comments will not be posted. PLEASE KEEP DISCUSSION ON TOPIC.