Friday, January 7, 2011

December Averages

Time to look at December's polling. Seven national polls were released during this month (sameas last month), totaling about 11,460 interviews. Here are the results we get at the national level, with the difference from last month's average in brackets.

Conservatives - 35.3% (+2.0)
Liberals - 27.8% (-0.4)
New Democrats - 16.2% (-1.0)
Bloc Québécois - 10.2% (+0.6)
Greens - 8.8% (-0.8)
Others - 1.7% (-0.7)

The Tories made a big jump in December, gaining two points. It's the highest they've been since December 2009, when they were polling slightly over 36%. Perhaps the Tories get a December boost. The Liberals, on the other hand, have dropped 0.8 points over the last two months. The New Democrats have also dropped, while at 8.8% the Greens are at their worst since, coincidentally, December 2009.

The seat projection for these results is as follows, with the difference from last month in brackets:

Conservatives - 140 (+12)
Liberals - 90 (-3)
Bloc Québécois - 53 (unchanged)
New Democrats - 25 (-9)
Greens - 0 (unchanged)

A big jump for the Conservatives compared to November, coming mostly at the expense of the listless NDP. The Liberals aren't exactly barn-burners either, but seem to have been stuck in the 90 seat range for months.The regional results, with difference from last month in brackets:

BRITISH COLUMBIA (7 polls - about 1,140 people)

Conservatives - 39.6% (+7.8)
New Democrats - 23.6% (-4.1)
Liberals - 23.4% (-0.3)
Greens - 11.3% (-2.6)
Others - 2.1%

If you're looking for an explanation for the Conservative gain in December, look no further than in British Columbia. That's a massive increase in support. The NDP drop comes out of nowhere, as the party had been stable in November. The Greens should be worried, as they've dropped almost four points in two months in the crucial province. The Conservatives would win 22 seats (+6 from November), while the Liberals would win eight (-1) and the NDP six (-5).

ALBERTA (6 polls - about 840 people)

Conservatives - 54.4% (-3.7)
Liberals - 20.6% (+0.2)
Greens - 11.7% (+2.8)
New Democrats - 10.8% (+1.0)
Others - 2.5%

The gain in BC is slightly offside by the Conservative slip in Alberta. They still dominate the province, though. But it is remarkable that the Liberals have been holding at over 20%. The Greens, meanwhile, have gained more than what they've lost in the province from September to November, and are third in Alberta. Unchanged from November, the Conservatives would win 27 seats and the Liberals one.

SASKATCHEWAN & MANITOBA (7 polls - about 1,590 people)

Conservatives - 47.9% (+5.1)
Liberals - 23.4% (+1.1)
New Democrats - 21.2% (-2.5)
Greens - 6.2% (-3.2)
Others - 1.3%

The Conservatives rebound after a big drop in November, while the Liberals are up a little. The NDP has slipped back to third in the region while the Greens are back down. The Conservatives would win 21 seats (+1), the Liberals four (unchanged), and the NDP three (-1).

ONTARIO (7 polls - about 3,780 people)

Conservatives - 38.4% (+1.6)
Liberals - 33.4% (-1.1)
New Democrats - 16.7% (-0.2)
Greens - 9.9% (+0.1)
Others - 1.6%

Before anyone gets too excited about the Conservative gain, this just puts the party back to where they were in October. The Liberals have lost 1.1 points for the third consecutive month, which is a bit of bad news for them. The Conservatives would win 53 seats (+4), the Liberals would win 40 (-3), and the New Democrats 13 (-1).

QUEBEC (9 polls - about 4,870 people)

Bloc Québécois - 39.8% (+1.4)
Liberals - 20.9% (-1.7)
Conservatives - 17.7% (+1.3)
New Democrats - 14.0% (+0.7)
Greens - 5.9% (-1.6)
Others - 1.7%

This is the best Bloc result since July, while the Liberals have lost 3.2 points since September. The Conservatives are up for the second straight month, as is the NDP. The Bloc would win 53 seats (unchanged), the Liberals 14 (-1), the Conservatives seven (+1), and the NDP one (unchanged).

ATLANTIC CANADA (7 polls - about 880 people)

Liberals - 41.2% (+5.4)
Conservatives - 35.3% (+4.8)
New Democrats - 18.3% (-4.6)
Greens - 4.8% (-4.8)
Others - 0.4%

A big jump for the Conservatives, part of the two-ends-of-the-country jump I mentioned in The Hill Times last month. The Liberals are up even more in the region, the highest they've been since August. With gains like these, someone has to suffer. Both the NDP and the Greens are down almost five points. The Liberals would win 21 seats (+2), the Conservatives nine (unchanged), and the NDP two (-2).In terms of net gains and losses in the six regions, the New Democrats performed worst with a net loss of 9.7 points. Most came in Atlantic Canada and British Columbia, two regions the party has always counted on for seats. The gain in Quebec, however, is positive.

Next worst were the Greens, with a net loss of 9.3 points. They were stable in Ontario, however, which is one of the provinces they will focus on.

Middle-of-the-road goes to the Bloc, which had a net gain of 1.4 points.

Runner-up is the Liberal Party, with a net gain of 3.6 points. Most of that was in Atlantic Canada and the three Prairie provinces. Losses in Ontario and Quebec are far more worrisome.

And December's winner is the Conservative Party, with a massive net gain of 16.9 points. They jumped more than four points in Atlantic Canada, the Prairies, and British Columbia, and had modest gains in Quebec and Ontario. Dropping in Alberta is not really a big deal for the Tories.


  1. Hmm December best case scenario should be getting close to a majority I bet.

  2. I won't be providing any spoilers.

  3. Like 2009, the CPC ended 2010 on an upswing.

    Except, 2009 didn't really end on an upswing. December was looking really good for the CPC, and then they prorogued parliament and took a big hit in the polls come January.

    The CPC has not made a similarly bad move this December, so I'm very interested to see how they start the new year.

  4. Ira,

    Very true. I think they'll start off the New Year with the current advantage they hold now, and either we'll go for an election, or those by-elections in those three ridings will happen. If it's the latter, then I would suspect that it won't be until then that we see any major movement from any of the parties, depending on the outcome.

  5. éric

    would it be possible to include a chart or narrative on the composition of the 7 national polls that went into the December average?

    If there is an EKOS poll (2000 sample) more or less in a month it would influence the monthly comparisons.

    From the intro I understand that you do NOT include the CROP or Leger Federal Quebec only polls in the monthly summary. True or false?

    Does the "CANADIAN PROJECTION DETAILS" chart you have at the bottom have any meaning or value? I think it must. could you explain how to read it?

  6. The seven national polls in December were conducted by Abacus, Angus-Reid, Ipsos-Reid, EKOS (x2), Harris-Decima, and Nanos.

    I include all polls in the monthly averages, including province only polls from firms like CROP, Léger, and Probe.

    The Canadian Projection Details chart at the bottom has a lot of meaning. It contains all of the polls that are currently being used in the projection model.

    The first column is the name of the polling firm, the second column is the final date of polling for the poll, the third column is the sample size, and the fourth column is the weight of the poll in the projection. Weight is determined by the size of the poll, how old the poll is, and the accuracy rating of the pollster.

    Yellow polls are those that were newly added in the current projection. The highlighted polling results are the best and worst results for each party.

  7. Wild ass prediction: (as usual)

    Next election... Spring 2012.

    we shall see.

    Also, Canucks win the holy of holies.

    we shall see.

  8. I know some here aren't heavily into history but Eric what would a trend line chart which only used the December results for the last five years look like ??

  9. Don't know how well this will post, but here are the numbers Peter. (And Kudos for originality in trying to find numbers that to build your "tories sliding, NDP godlike" narrative around).

    I used a monthly Dec average, because 28 polls from 06-09 +the 2010 polling.. too much for this format. You can throw it into Excel and have a look if ya want. It mostly shows a small liberal fall, a small NDP rise, and a tory blip up (and others down) in 08 when they tried to end the $2/vote/year followed by a settling in a little higher in the last 2 years as opposed to the first 2.

    BQ Tory Green Liberal NDP
    Dec-10 10.2 35.3 8.8 27.8 16.2
    Dec-09 9.5 36.8 8.1 28.1 17.1
    Dec-08 8.9 44.2 7.7 24.8 13.9
    Dec-07 9.5 33.5 10.5 29.8 16.0
    Dec-06 10.0 32.7 7.7 35.5 12.9

  10. Barcs

    " (And Kudos for originality in trying to find numbers that to build your "tories sliding, NDP godlike" narrative around)."

    Actually I wasn't doing that at all. Now reformatting your numbers, thanks by the way, what I see is, aside from two and possibly a third anomaly we are all talking about a chimera.

    Tory in 08, Liberal in 06, and maybe NDP in 09? Those are actually anomalies.

    There is no indication that for any party there is the possibility of majority. S much for all the punditry !!

  11. Most pundits, including this one if I can use the title, have not been talking about a Conservative majority.

  12. Eric i'd disagree.

    In late '09 (before prorogation) there was talk of a CPC majority in the punditry.

    And just lately in the last two months or so that talk has started up again, usually premised on some kind of breakthrough around Toronto or the Nanos cluster theory.

    Since poll numbers can change during an election it would be, of course, totally foolish and ridiculous to dismiss the possibility of a CPC majority after the next election.

  13. totally foolish and ridiculous to dismiss the possibility of a CPC majority after the next election.

    Actually given the historical record even to suggest a CPC majority is the height of foolishness.

    The country is stuck and has been for years between which way to shift.

    It's time for a national narrative on what direction the country should take.

  14. I feel bad taking advantage of your generosity in knowledge sharing.

    I have found an obvious flaw in your projection model.

    I won't mention the pollster by name (as that is almost a sure way for my comment to not be posted) but your seat prediction are based 74% on two pollsters.

    One of the pollsters is by far the most prolific pollster with large samples done very regularly.

    Both pollsters are indicated to have a pro-green and very anti-CPC bias in comparison to other pollsters in your research.

    That goes a long way to explain how you are predicting a Liberal gain of 20 seats where there does not seem to be as many CPC ridings on the bubble as Liberal ridings.

    The most accurate (the Gold standard in Polling: AR) accounts for only 12% of the weighted data going into your model.

  15. They are but one polling firm, and as the last Alberta election has shown, they can be just as wrong as anyone else.

    EKOS did very well in the Toronto mayoral election, however. Their methods should not be discounted.

  16. They are but one polling firm

    as is your preferred polling firm who accounts for 56% of the input to your seat projection model.

    Before peeking behind the curtains I was an advocate for your analysis which I believed to be an unbiased aggregation of results built with a model that accounted for bias, reliability and accuracy.

    Now I am afraid that it is volume driven and dominated by the most biased partisan polling source.

    There may not be a CBC/Liberal conspiracy to influence the electorate with negative CPC, positive Green polling results but the systematic onslaught and dominance of polls released suggests otherwise.

    I hope that you are able to step back and evaluate if your current weighting is the most accurate seat projection model.

    otherwise you need to add a disclaimer that your seat projections are based on the assumption that EKOS is 4.7 times more accurate than AR.

  17. "In late '09 (before prorogation) there was talk of a CPC majority in the punditry."

    There was also talk of a CPC majority in December '08 - which may explain why Iggy extracted his party from the Dion coalition.

    A Tory majority is possible, if they play their cards right and their opponents don't.

  18. EKOS is not my "preferred pollster". They supply the most data, and so they get a lot of space in the model. If I was heavily relying on month-old Angus-Reid polling all of the time the projection would be out-of-date. And because Angus-Reid hit the target on the dart board in the last election does not mean that they will in the next election.

    EKOS is no more "anti-CPC" than Ipsos or Angus is "pro-CPC". There is no more proof that EKOS systematically under-estimates Conservative support than those two pollster over-estimate Conservative support. The loudest concerns on "pollster bias" that I see in my comments section are almost always partisan-driven, and I take them into consideration accordingly.

    In an actual campaign, all pollsters will be reporting, likely on a daily basis, which will mitigate the factor that concerns you so.

    But there is no "conspiracy", and I'm afraid your mention of one makes it difficult for me to take you very seriously.

  19. In addition, the projection is adjusted to reflect the difference between polling and results over the last three elections. Because of this, the Green vote is reduced. So even if the methods of some pollsters inflate Green support, it is not reflected in the projection.

  20. Peter how did you enjoy Kim Campbell's big majority win in '93 ?

    Campaigns clearly matter.

    Just because the CPC haven't consistently been in majority territory before doesn't mean their numbers can't go up.

    Frankly its just wishful thinking to rule out the possibility of a Tory majority.

  21. December weighted average of Nanos, AR, IR and Abacus

    CPC 38
    Lib 28
    NDP 17
    Blo 10
    Green 8

    These are all relatively close using different methodologies.

    EKOS with a 5 point spread between CPC and Liberal rather than the 10 % must from all these independent pollsters must be correct as they sample more?

  22. totaling about 11,460 interviews

    That is stretching the meaning of "interview"..... press 1 for correct 2 for incorrect

    You have just been interviewed :)

  23. I will make some minor changes to lower the influence of volume, particularly at the national level.

  24. Eric,

    I take your points, but I do think there is a problem with weighing polls by number of interviews. The obvious point is that a poll with a 2000 person sample isn't twice as "good" as a poll with a 1000 person sample. True, the first poll has a smaller margin of error, but it's margin of error isn't half the size (its 2.2 vs 3.1) and, in any event, they should cancel out in a large dataset.

    Statistically, there's no basis for using sample size (at least not beyond a certain sample size) is a meaningfull measure of poll quality. I won't comment on your other weighting elements (i.e., age and accuracy), but I think weighting by poll size can't be justified.

  25. Yes, that is why I will be making some changes, putting the cap at 1,500 people. Beyond that, polls will not get any weighting bonus.

    As to why 1,500, a poll of 1,000 people can often only have 750 decided voters, making it less valuable than a poll of 1,500 people with 1,125 decided voters.

  26. "As to why 1,500, a poll of 1,000 people can often only have 750 decided voters, making it less valuable than a poll of 1,500 people with 1,125 decided voters"

    I don't agree. The margin of error for a poll of 750 is 3.5%, for a poll of 1125, it's 2.9%. So the larger poll isn't 1.5 times more useful than the smaller one (at least if usefulness is measured as margin of error) nad arguably isn't materially more useful at all.

    But hey, it's your model, and your proposed change would at least be an improvement.

  27. That's very gracious of you.

    This isn't the first time I've grappled with this issue. I haven't been able to figure out a formula that could be input into the model easily to do a more accurate weighting. I'll get there eventually.

  28. For those interested, I've changed the weighting system for sample size. The weight given to a poll based on its sample size is now proportionate with its MOE. So, for example, a poll of 1,000 people has a margin of error of 3.1. A poll of 2,000 people has a margin of error of 2.2, so it has a weight 1.41 times higher than the 1,000-people poll, rather than being weighted twice as heavily.


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.