Wednesday, October 12, 2011

Liberals defy trends in Newfoundland and Labrador

Last night's election in Newfoundland and Labrador played out generally as expected, but around the margins the Liberals managed to cling to the role of the Official Opposition by a handful of votes.

A full analysis of the results of the Newfoundland and Labrador election can be read on The Huffington Post Canada website here.

For the projection, it was a bit of a rough ride. Though the final results were not significantly different from the final projection in terms of seats (at an error of 3.3 seats per party it ranks as the third best projection of seven), in terms of outcome it was greatly different. Yes, the Progressive Conservatives won a majority and the NDP made a historic breakthrough, but the Liberals outperformed expectations and won six seats instead of two.

And they did so in a slighlty bizarre fashion. One incumbent was re-elected, one Liberal seat was held, one Liberal incumbent was defeated, one Liberal seat was lost, and four completely new seats were won at the expense of the Progressive Conservatives. The Liberals dropped from 22% to 19% but gained two seats, and it is difficult to attribute those seats entirely to vote splits or the rise of the New Democrats. It was just a somewhat abnormal performance.

The New Democrats also bucked provincial trends, doing much better in St. John's than was expected, winning a seat in western Newfoundland but losing what was supposed to be their best shot outside of the capital in Burin-Placentia West.

The regional divisions that the election indicated (New Democrats in St. John's and eastern Newfoundland, the Liberals in western Newfoundland and Labrador) point, again, to the need for a regional model for even a small province like Newfoundland and Labrador. This election did provide some regional data, so perhaps that is something to look at for the future.

Another thing was the shift in votes from the last polls of the campaign, released a week before the vote. The New Democrats dropped by about five points and all of it went to the Liberals. The chart below shows the change.

 Had this sort of shift been anticipated, the projection model would have given the Tories 41 seats, the Liberals four, and the NDP three. Though the level of error would have been only a little better, projecting the Liberals as the Official Opposition would have been key. However, the riding-level accuracy would have been even lower, demonstrating the difficulty in projecting smaller provinces.

The Electoral Track Record has been updated, so you can see how the Newfoundland and Labrador projection stacks up to others.

Moving forward, we have one last election left this year. Saskatchewan should not hold many surprises, as the race is between two parties and Brad Wall has a massive lead over the NDP. It could play out more predictably, as the Manitoba election did. The site will be updated soon to give the Saskatchewan election more prominence.


  1. Looking at the top of the page for 308 we get four projections for provincial elections and four results.

    In three cases Eric your projection has given more seats to the winning party than they actually received in the vote.

    I think you need a bit of rework ???

    Proj. Actual
    Ont 58 53

    Man 36 37

    PEI 26 22

    NL 42 37

  2. While the incumbency effect did seem to help elect every government, at the riding level it appears to have been less pronounced.

    As always, the projection model will be tweaked and improved with every new set of data.

  3. could it be that there are simply temporary conditions that favor incumbents at the moment?

  4. Or for that matter just random noise. If you flip a coin four times, what are the odds that it comes up heads at least three times? Roughly 31%. That's not evidence that the coin isn't fair.

  5. What seat did the NDP win in Western Newfoundland? The Straights-White Bay North? That's on the north-eastern corner of the mainland.

  6. Anonymous 12:52,

    If you are looking at Newfoundland in terms of west/east, that riding is in the west.

  7. again this shows the flaw in aggregate polls
    the problem is your methodology in aggregating and weighing when you should publish the polls as is and project seats on that basis and just offer a range or median point.

  8. Have you thought about doing a more rigorous model a la 538? IE establish what the probability is that party A > party B on election day if your models shows party A with an X% lead based on previous election results?

    I guess that wouldn't work entirely though, because in order to establish the odds of party A winning you'd need to assume that the probability that B > C is independent from the probability that A > B, which it isn't. Hmm...

  9. Yes, 538's work is aided by the fact that in most cases there are only two parties. When he tried to do the last UK election, he used a system similar to what I use.

    There is always more work to be done and improvements to be made. It is an on-going project.

  10. I'm wondering if the polling firms were oversampling people from St. John's. That might account for the higher percent for NDP and lower for Liberals (since NDP did very well in St. John's and Liberals did very poorly there). It also makes you wonder if there is a sort of "herd" mentality involved that influences choices at a regional level.

  11. There were some interesting techniques in this paper by the way:

  12. Hi,

    When is the final post-mortem supposed to come out?

  13. I have to do final post-mortems for all the provinces, and I'll try to do them as soon as possible.

  14. "I'm wondering if the polling firms were oversampling people from St. John's."

    No, any reputable polling company sets regional quotas and/or weights by regional population etc...I think that in NL it was always clear that a lot of NDP support was very soft and since turnout was apparently quite low - it was always likely to not show up. Also, outside of half a dozen ridings in St. John's and Burin-Placentia, the northern peninsula seat and parts of Labrador, the NDP had no ground campaign whatsoever. It will be a very different story in 2015 after 4 years of having a much bigger budget and a bigger profile and in all likelihood active campaigns in all 48 ridings.

  15. The last poll conducted before the election by CRA was within the MOE and spot on with NDP support. They had the PCs at 59%, the NDP at 25%, and the Liberals at 16%. The NDP ended up with I believe 24.6%, while the PCs won 56% and the Liberals 19%.

    MQO, who is new to polling, had NDP support higher and Liberal support lower, but had high undecided rates. MQO I believe surveyed using both the internet and phone, which may have underrepresented those living in rural areas.


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.