Saturday, March 21, 2009

New Poll: Nanos Research

Nanos Research released a poll today, conducted between March 13 and March 18 and involving 1,002 interviews. The national results:

Liberals - 36%
Conservatives - 33%
New Democrats - 13%
Bloc Quebecois - 10%
Greens - 8%

This is the best poll the Liberals have had since the election. Previously, 34% was their high-watermark, back in early January, also in a Nanos poll. The Ontario result:

Liberals - 44%
Conservatives - 31%
New Democrats - 14%
Greens - 10%

Again, this is a high for the Liberals, their previous high being 43% in a Strategic Counsel poll in early February. The Quebec result:

Bloc Quebecois - 36%
Liberals - 32%
Conservatives - 19%
New Democrats - 7%
Greens - 6%

This is on the high level for the Liberals, but not out of the ordinary. The Bloc result is the lowest since late January in an Angus-Reid poll, but the Conservative result is the highest since that poll as well. The NDP result, however, is the lowest result they've had since the 2008 election.

There is an Atlantic result with you can see in the detailed table below, as well as a West result which hasn't been entered into the model but I will present here for you:

Conservatives - 46%
Liberals - 30%
New Democrats - 15%
Greens - 10%

In addition, Nanos asked which leader would make the best Prime Minister. Stephen Harper led the pack with 33%, followed by Michael Ignatieff at 27% and Jack Layton at 12%. Harper was most popular in the West (48%) but least popular in Quebec (14%), continuing a worrying trend for Harper's personal appeal in that province. Ignatieff got his best result in Ontario (33%), but his worst in the West (21%). However, as is the case in virtually every poll, the Ignatieff result has far less variation than Harper, demonstrating he is a bit more of a "national" leader. Layton, surprisingly, had his best result in Quebec (19%) and his worst in the West (9%).

The projection has changed, as the Liberals (103) continue to creep up on the Conservatives, who are down one seat to 135. The national popular vote has also changed:

Liberals +0.2
Bloc Quebecois +0.0
Greens +0.0
New Democrats -0.1
Conservatives -0.1

In Ontario, the Liberals are starting to pull away, gaining 0.1 points while the Conservatives have lost 0.2. There is now a 0.4 point gap between the two. There are no significant regional changes, though it looks like the Liberals and NDP have traded 0.2 in Quebec.

This has to be the most worrisome poll for the Conservatives in a long time. But, there are a few positive points for them. Nanos usually has more favourable Liberal results, and the 19% in Quebec is a reversal in their downward spiral. The Liberals have much to be pleased about, especially their 13-point lead in Ontario and the 4-point gap between them and the Bloc in Quebec. This is another worrying poll for the NDP, and the 7% in Quebec virtually wipes out any hope of gaining a seat in that province and puts Thomas Mulcair in danger. The Bloc has been slowly dropping in support, with polls having them drop one or two points, consecutively. Their opinion poll trend is a steady but slight slope downwards.


  1. Eric

    Just so I understand. You use past election results, as well as 11 month old polls to come up with your projections? You plug those numbers into your weighted average?

  2. Yes, that's correct. The projection model isn't just about what the latest polls say and what opinion is right now. It is about what the result would be of an election. Polls are never absolutely correct, and past voting habits play a role in people's choices, so that is why I incorporated past electoral results and older polls (though older polls are reduced in weight).

    None of the polls are 11 months old, however. The oldest one currently in the model is mid-December. Eventually, a poll will have such a low weight that it will be dropped entirely.

    That gives the popular vote projection. The seat projection is based on that popular vote projection and historical results.

  3. I hear what you're saying, but past voting behavior will also manifest itself in the polls, so I'm sort of weary about using really old results to get an accurate read. I think you can gleam the trends, where we really are, by using the last result for every polling company. In that way, you get a better gauge of fluidity. A mid-December poll, during the height of this coalition question, wherein the Cons soared, the Libs down to 22%, is pretty much a meaningless indicator now. If you actually breakdown all the latest offering, from each pollster, you find a flip a coin electoral scenario, especially when you plug in the regionals. I appreciate your methodology, I'm just saying it might have a lag that doesn't appreciate the trends fully IMHO.

  4. I see what you're saying, but the polling companies are often wrong by a significant amount. If I worked simply from them I wouldn't have that great of an indicator either.

    And yes, you're right, the current model does have a lag. But, in the case of an election, polls would be coming out on a daily basis, and would quickly overwhelm older polls.

    It is a mix of everything which I think gives good results. Polling firms get it right sometimes, but they can't take into account whether anyone will actually vote and whether they are telling the pollsters the truth. Past electoral results aid in minimising those possibilities.

    The system also works from the perspective that this would be the result after the end of an election campaign, not RIGHT NOW.

    By the way, I don't have any polls with the Liberals at 22%, the lowest I have them is at 26%, which is within the margin of error of some more recent polls.

    And don't forget, older polls lose their weight quickly after two months.

  5. One final point, we've all been burned by a few good polls that didn't come through after an election. My model is weighted to be conservative in its estimates, swinging too widely doesn't give any semblance of consistency. Polls are part of my model, they aren't the entire thing.

    My system worked well during the Quebec provincial election, when I did better than any other projector. One of the issues was that the ADQ was polling very low, but eventually got a higher electoral result. If I had used only the polls and not past electoral results, I would've under-estimated the ADQ vote even more than I did.

  6. I very much like the weighting, based on past accuracy. Some of these pollsters have a very poor record, so to weight them the same creates a false picture.


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.

Note: Only a member of this blog may post a comment.