Tuesday, October 4, 2011

Final Manitoba Projection: Selinger's NDP wins majority

With hours to go before the polls close in Manitoba, ThreeHundredEight.com projects that Greg Selinger's New Democrats will win a majority government, with Hugh McFadyen's Progressive Conservatives forming the Official Opposition. Jon Gerrard's Liberals will be the third party in the Manitoba legislature.

This comes after the addition of Angus-Reid's Manitoba poll released yesterday, conducted between September 30 and October 2, and surveying 800 people online. The poll was identical to Probe Research's numbers, with the New Democrats at 46% and the Progressive Conservatives at 43%.

This is a margin that, however, makes me quite uncomfortable, particularly in light of yesterday's results in Prince Edward Island.
The New Democrats are projected to take 45.6% of tonight's vote, with the Progressive Conservatives taking 42.9%.

The Liberals are projected to take 8% of the vote, while the Greens take 3.1%.

That is, in any case, what the polls are suggesting.

Since the projection of September 30, this is a gain of 0.9 points for the New Democrats. The Tories and Greens have each lost 0.4 points, while the Liberals are down 0.1.

The New Democrats are projected to win 36 seats, with the Progressive Conservatives winning 20 and the Liberals one.

The New Democrats are projected to win 25 seats in Winnipeg, four in the north, four in the southeast, and three in the southwest. The Progressive Conservatives are projected to win 10 seats in the southeast, five in the southwest, and five in Winnipeg. The Liberals win their one seat in the provincial capital.

Four polls were released in this campaign, all in the last two weeks. Only one of them showed a Tory lead, while the last three all put the margin between the NDP and the PCs at three points, inside the margin of error.

Averaged out, however, the New Democrats have been tied or have led the Tories for the entirety of the campaign. But the race remains incredibly close, and dangerously close for a projection.

The New Democrats are leading in one close race and trailing in two others, all implicating the Tories. This puts the New Democrat range at between 35 and 38 seats, and the Tories between 18 and 21 seats.

I suspect, however, that more races will be quite close.

But it should not be forgotten that projections are not a crystal ball looking into the future - they are based on the available polling data, which currently shows an NDP lead.

Though the ridings in Manitoba are larger than that of Prince Edward Island, they are still small and local factors will undoubtedly skew some of the riding projections. Also problematic is that the boundaries have changed, and this has the potential to greatly change how voters behave. For many voters, their "incumbent" ran in a different riding in the 2007 election, so there is a good chance that there will be a few surprises tonight.

Adjusting the projection for the polling error in the 2007 Manitoba and 2011 federal elections, however, changes little in the outcome.

If the error is the same as it was in the 2007 provincial campaign, the New Democrats should take closer to 49% of the vote and win 37 ridings.

But if the error is the same as it was in the 2011 federal campaign, and as we have seen in many cases the Conservative vote is often under-estimated, the Tories would win more of the popular vote - but only just. The projection model would still give the NDP 36 seats to the Tories' 20, but the PC range would be extended to as many as 26 seats, with the New Democrats taking the 31 others.

If we use the riding projection accuracy rating of last night's projection, we could expect as many as eight ridings to be called incorrectly tonight. That is a potentially significant number. If the projection tonight is as accurate as last night's, the New Democrats could win between 28 and 44 seats and the Progressive Conservatives between 12 and 28 seats. In other words, the legislature could be split down the middle at 28 seats apiece for the NDP and the Tories.

If the polls have it, though, the edge the New Democrats have in Winnipeg (roughly 20 points) should be the clincher. The city contains 54% of the province's ridings, and the New Democrats are strong enough in the north, Brandon, and a few other rural pockets to make up for any Winnipeg ridings that side with the Progressive Conservatives.

But the outcome is far from pre-determined. With the margins of error of the polls, the Tories could actually come out with more votes. And if the changes in riding boundaries has a particular effect, they could come out with more seats as well. I suspect it could be a long night in Manitoba, as it will be for myself here in Ottawa.


  1. Hopefully the prediction is more accurate

  2. "This is a margin that, however, makes me quite uncomfortable, particularly in light of yesterday's results in Prince Edward Island."
    Not to be rude, but taking in consideration your track records from the Fed elections and PEI I'm not sure that "comfortable" is the right word.
    The results in ON and MB will probably let us know if there is any bias/error in the projection model. If Conservatives will do better in both of those elections, then some will have very good reasons to ask if there is any left bias in these projections.
    On the other hand, if the results will match the projections, then finally we would have a Canadian Nate Silver.
    In any case, thank you for the great work. This blog is a must for all Canadians political junkies.

  3. Based on the number of PC signs in my newly created riding (Ft. Richmond) I think the final tally here may be much closer than your projection. Will the NDP lose the riding? No, probably not. But I do think the PC's may make significant inroads.

  4. Anonymous 11:29, the problem has never been Eric's model. The problem, most transparently in the federal election, was the polling under-counting Conservative voters.

    That undercounting may be coming into play in the final days of the Ontario election. The PCs are pulling out all of the dog-whistle politics stops in an effort to win, and this could lead to people not wanting to admit to a pollster that they are planning to vote for the PCs.

    As Eric demonstrated after the federal election, if he'd had the correct numbers to work with, he would have pretty much nailed it. It is the polling models that are flawed, and producing inaccurate results somehow. The model can only work with the numbers available to it.

  5. "this could lead to people not wanting to admit to a pollster that they are planning to vote for the PCs."

    FYI: Almost all the polls we are seeing are either online or are IVR push-button robo-calling - there no reason to not want to admit to voting PC to a robot or a computer screen.

  6. But robots can be very judgmental.

  7. "But robots can be very judgmental."

    This might be funny. The latest work done by Canadian pollsters, not so much.

    "Almost all the polls we are seeing are either online or are IVR push-button robo-calling - there no reason to not want to admit to voting PC to a robot or a computer screen."

    Yes, but pollsters might want to adjust their models a little better, if they constantly underestimate Cons support. I don't know... maybe Cons supports are not too inclined to take on-line surveys or answer the phone. Hopefully, pollsters will end up actually underestimating Libs and NDP support this time, so no one can accuse them of leftist bias.

  8. Could it just be that Conservatives and New Democrats just focus their efforts more efficiently? Perhaps, rather than running a single uniform campaign with across-the board messaging, they're running separate, distinct campaigns in those ridings they have the best chance of winning.

  9. Polling and Estimating are both black arts, which are little helped by political parties which lie about what they are. I haven't been paying close attention to the other provincial Conservative parties, but the Ontario Progressive Conservative Party, like the Conservative Party of Canada isn't Conservative. I probably should have written about them as well, but they aren't likely (in my opinion) to win, and while the Federal Conservatives did win by lying about what they are.

    Thus this post:

    Canada Needs a Conservative Political Party


    I have a suspicion that it is going to annoy a few people, but hey, I'm good at annoying people.


  10. "Could it just be that Conservatives and New Democrats just focus their efforts more efficiently"

    There is some evidence that Conservatives did exactly that in last federal election. However, not sure if it's enough of an excuse for pollsters.

  11. The polls in 2006 (federally) underestimated the Liberal support.

    If I had to guess it'd be that polls in general, and IVR polls especially, tend to skew towards the angriest / most emotional voters. Those usually aren't the incumbent party.

  12. Early days yet, but so far the model seems almost dead on actually....

  13. NDP leading or elected in 37, Cons 19, Libs 1 at 10:12 eastern. This could be a good one

  14. As of 10 PM EDT CBC says the following. Looks a lock to me ??

    NDP: 37

    PC: 19

    LIB: 1

  15. Seems to be settling on a dead on correct projection actually, outstanding work Eric.

  16. I know this is an anecdote, not data, but....my wife grew up in BC in the 60s and 70's. She says she never met anyone who ever admitted to voting Socred, but the Socreds got elected time after time.

    I wonder if people's self-image (and therefore the image they want to project to others, including robo-callers) is slightly different from their reality (and self-interest) when they get into the voting booth.

    Seems to me that we might want to consider changing the methodology for evaluating poll results by adding some percentage to the right-wing numbers as reported in the late polls. Maybe the trick to the methodology is figuring out the ratio between the late-poll total for the Conservatives and their allies and the actual voting percentage.

  17. One anonymous said: "If Conservatives will do better in both of those elections, then some will have very good reasons to ask if there is any left bias in these projections. "

    The projections are mathematical, therefore the idea of subjective bias in them is hard to consider seriously.

    More likely, if there is a bias somewhere, it is likely in polls. Since polls are the input of the model, if they tend to underestimate Conservative support, then the results will tend to underestimate Conservative predictions, through no fault of the projection model. This is the "Garbage In, Garbage Out" principle.

    If there is a bias in polls, it likely results from the fact that voter turnout in Canadian elections is low, especially amongst the young population. The young tend not to vote Conservative, so if they answer as much to polls as older voters but do not bother to vote on election day, polls will indicate lower Conservative level of support than will actually appear in the election results. Thus, polls can be biased without anyone inserting their subjectivity or personal bias in the system.

  18. Social Desirability is a real thing that questionnaire constructors always worry about. There are ways around it, though they are not easy to do with a political poll. It may be the case that it is not socially desirable to say you are voting for an incumbent, rather than say it being a left or right thing.

  19. Nearly Dead-on! Good Job Eric. 35 NDP, 21 Pc, 1 Lib.

  20. Hey Eric...bang on.

    Now if you get the Ontario one close, then you'll be a hero among men...well non-Tory men anyways. :-)

  21. Spot on projection Eric - nice piece of work!

  22. Thanks everyone! That'll buy me some peace, at least until Thursday night.

  23. Even better, 2 of the PC switched bringing the total to 37NDP to 19 PC, 1 Lib. So there is no leftist bias in the projection after all. Great job Eric. The only miscall is St. Norbert in Winnipeg which the NDP won by 157 votes. It could go to PC after a recount :)

    On another note, i would like to comment is that your projections do very well when the results or polls are close to the last election, once there is significant movement( like a party doubling up or tripling up support), your projections seem to falter. I think this is expected due to the uncertainty of where the increase of the support comes from. Maybe you should have another model for swing elections.

  24. Unless recounts happen there is only one seat left, and if it doesn't change it will be the only wrong one (St Norbert). Right now %s look like .3 off for NDP, 1 off for Cons, .5 for Libs, and .6 for Greens.

    Awesome, looking forward to Thursday

  25. New ipsos reid poll out today.


  26. Well done Eric. Virtually dead on !! Great job.

    Now let's get tomorrow's Ontario right ??

  27. Well done, 308. You nailed it.

  28. Ontario will be a difficult one. We got a whole variety of last minute polls show everything from Liberal majority (Ipsos, EKOS) to PC minority (Angus).

    The Tories aren't helping themselves with last minute blunders. With homophobic comments, war or public transit and a coalition scare, it looks like the party just wants to get its party faithful out to vote. If Ipsos and EKOS are accurate then Tim Hudak will bring the popular vote lower than Ernie Eves and John Tory.

    - Maple

  29. Excellent job in the forecast!

    Its very interesting to see how even with a substantial swing in popular vote, you can end up with almost no seat change. Such is the way with First Past the Post. Now, I hope Erik sends updates to Craig Oliver at CTV so he doesn't get so giddy about his party taking over the remainder of Canada.


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.