Monday, November 29, 2010

Updated by-election projections, and their likelihood

UPDATE: I'm following the by-election results live on Twitter.

Regular readers of this blog will remember my by-election projections which appeared in The Globe and Mail at the beginning of the month. What was intended as a small exercise and an attempt at projecting today's by-elections with numbers alone became an article which raised the ire of a few commentators and was referenced as a "poll" for much of the campaign. Some have gone so far as to allege that I fudged the numbers for effect, something that is both quite insulting as well as being completely wrong.

As today is voting day for the three by-elections in Manitoba and Ontario, it seems like a good day to update the projections that I made in early November, using the same methodology. I will, however, add the "star candidate" factor I identified in a recent post.

First, I will explain the methodology that will be used this time and was used last time. To start, I've taken the election result from 2008 and adjusted it according to the proportional change that has occurred in the province (or region for Manitoba) since that election.

For example, the Liberal received 33.8% in 2008 in Ontario and are now projected to be at 36.6% in my model (including the latest EKOS poll, which the projection at the top of this page does not include). So, 36.6/33.8 = 1.0828402. The Liberals received 49.2% in the 2008 election in Vaughan, and 49.2 x 1.0828402 = 53.3.

I then take these results and adjust them by the "by-election factor". This was determined by taking the average difference in the 2009 by-election results from what was expected using the same formula as the one described above. As I have my projection data from November 2009 I was able to replicate what I am doing for the 2010 by-elections.

On average, the Conservatives increased their vote share by 23% (vote share x 1.23), while the New Democrats increased theirs by 22%. The Liberals only retained about 77% of their vote share, while the Greens retained about 48%.

Applying these factors to the expected result gives me my by-election projection. I then adjust the numbers proportionately to total 100%. When doing this calculation in early November, I ended up with the Liberals and Conservatives tied at 42% in Vaughan. Before the adjustment to total 100%, they were even tied at 39.4%. No fudging, it was really that close.

I then took these numbers and compared them to the historical election results in the three ridings. After looking into each of the ridings in-depth, I felt comfortable that they were plausible and went ahead with writing my article for The Globe and Mail.

But for this update, I am adding the "star candidate" factor to Kevin Lamoureux in Winnipeg North and Julian Fantino in Vaughan.

The results are that the New Democrats and Conservatives would hold on to their Manitoba seats, while the Conservatives gain Vaughan.In Dauphin-Swan River-Marquette, Robert Sopuck of the Conservatives would win with 63%, down three points from the earlier by-election projection. The NDP would finish second with 18% (up one), the Liberals third with 15% (up two), and the Greens fourth with 4% (up one). I consider this projection to be likely.

In Vaughan, the projection gives Julian Fantino of the Conservatives 45% (up three), and the win. The Liberals finish a close second with 41%, down one, while the NDP and Greens finish with 10% and 4%, respectively. This seems like a very plausible result to me, though personally I would not be surprised to also see the Liberals win by as much as 10 points. It really is a toss-up.

In Winnipeg North, these calculations would give Kevin Chief of the NDP the win with 64%, down one from the earlier projection. The Conservatives would finish second with 22% (down two) and the Liberals third with 11% (up two). The Greens would place fourth with 3%, down one.

Winnipeg North shows the limitations of projecting by-elections. Unlike a federal election, where these sorts of calculations would have a better chance of proving accurate because of the over-riding influence of a national campaign, by-elections are subject to local issues and fluctuations that I cannot project. I am unable to give Lamoureux more of the vote without adjusting the numbers by my own judgment. I think Lamoureux will probably finish second to Chief, but even applying the "star candidate" factor cannot get his numbers close to the Conservative candidate.

We shall see tonight what the results will be. I think that these three winners are the most likely trio. As to the implications, I think people are blowing it way out of proportion. The results in Manitoba won't have any leadership or national implications unless there is an upset. If the Liberals win in Vaughan, it will be because of a strong Genco campaign and a weak Fantino campaign, and because Vaughan has been Liberal for so very long. If the Conservatives win, it will be because Fantino was a good catch for the Conservatives, and because Vaughan was more of a Bevilacqua riding than a Liberal one. People aren't voting for Stephen Harper or Michael Ignatieff in the Vaughan by-election. While losing it would be bad news for the Liberal leader, it wouldn't be his fault anymore than an upset in Winnipeg North would be Jack Layton's.

I won't be doing any live-blogging of the by-election results, but will probably be commenting on Twitter.

17 comments:

  1. Interesting results, Eric. It'll be fun to see how they compare to tonight's results.

    If you want my opinion, I think you're going to be very far off the mark. Not because of any flaws of your system, just that riding-by-riding projections, even more so for by-elections, are notoriously hard to project. I think Nate Silver did up a post about how guessing "special elections," aka our by-elections, for congressional districts is a fruitless endeavour, simply because of the factors involved, and how nationally-based polling being fitted down to riding-sized bits doesn't really work out.

    But, we'll see.

    ReplyDelete
  2. Doesn't mean we can't try.

    ReplyDelete
  3. Hi Éric,

    I still believe you're using the wrong starting point, namely the vote-shares obtained by incumbents of 22, 17 and 13 years standing.

    The Liberals will obtain significantly more than 11% of the vote-share in Winnipeg North. And the Conservatives will obtain significantly less than 63% of the vote in Dauphin--Swan River--Marquette, MB.

    Or I'm eating shoe leather.

    I believe the best baseline for Winnipeg North would have been 2004, when two incumbents ran against each other on the new boundaries.

    In DSRM, the previous incumbent is now campaigning for a different candidate than his old party. I don't know what kind of baseline one would take in this riding, where Inky Mark's attraction had always been his maverick nature, and willingness to vote against his party. But his own previous vote share wouldn't be it.

    Thank you for publishing your methodology, however.

    I'm also sorry that you've felt insulted. However the current obsession with polls has driven media coverage of elections to the point that people believe there have been polls taken when there haven't been any, and believe that the earlier projection published at the Globe and Mail was a poll. Even Gloria Galloway wrote this morning that it was, for heaven's sake.

    This creates a bandwagon effect that I believe is truly unfair to people participating in elections, and especially to voters themselves. I apologize for leaving the impression that your earlier calculations did not follow your stated methodology, but I'm afraid my concerns about the methodology (and perhaps even the by-election endeavour itself) stand.

    On the other hand, it might give you some data for future by-election use in assessing any incumbency effects.

    very sincerely,


    alice

    ReplyDelete
  4. I have no problem with people criticising my methodology or findings. That's fair game. Not to be too melodramatic, but my integrity is another thing entirely. Every number on my site and in my articles are the numbers that show up on the calculator.

    The projection for Manitoba is relatively unchanged from the 2008 result because the changes in voting intentions in the province since 2008 were canceled out by the "by-election factor".

    That was not the case in Vaughan, which is why I ended up with a very different result in this riding compared to the 2008 election.

    And your point about incumbency is well-taken. My list of to-dos includes look at incumbency from as many angles as I can think up.

    ReplyDelete
  5. I'm really skeptical of the methodology, since it seems like you're using too general numbers to predict narrow outcomes, but a quick back of the envelope calculation for the Outremont byelection pleasantly surprised me.

    I used the October 2007 CROP poll as a basis (that was the nearest poll a quick google gave me, a star candidate factor of 1.21 for Mulcair, and a byelection factor of 1 for the Bloc (which I believe is probably higher than the one you've found), and that model gave me a share of NDP 42%, LPC 20%, Bloc 18.5%, CPC 17.5%., GPC 2.

    The actual result was NDP 47.5, LPC 29, Bloc 11, CPC 9, GPC 2, so the model wasn't very accurate, which I expected, but I really didn't expect it to give as close an approximation of the actual situation as it did.

    I didn't really take your by-election projections very seriously before trying it out on Outremont (and yes, I realise that Outremont was part of the dataset you used to calibrate the model in the first place, but that shouldn't on its own be this significant), but now I do at least think they're useful to get a general idea of the lay of the land.

    ReplyDelete
  6. Interesting, thanks Joffré. Outremont was not taken into account, I only used the 2009 results.

    ReplyDelete
  7. Eric,

    The fact that you only used the 2009 results explains why you have such a low Green result, given that you'd expect at least some sort of positive bump, considering the huge Lizzy May result back in 2006's London-North Centre result.

    ReplyDelete
  8. I, for one, quite like Eric's methodology as liberals will come away empty handed. Love this blog, Eric. I take a look at it every single day.

    ReplyDelete
  9. Should be an interesting night. Sun Media predicts a CPC sweep. I think that highly unlikely as I can't see the the CPC taking the Winnipeg seat. Would shake things up though as would the LPOC taking two of three.

    ReplyDelete
  10. Looks like your projection of the Liberal vote in Winnipeg North was off by 35.8 points. Back to the drawing board?

    ReplyDelete
  11. Yep. I thought Lamoureux would do better than I projected, maybe 30% or something, but I didn't expect this.

    The calculation performed well enough for the other two ridings, though, so not too bad. I'll be able to use this data to do a better job next time.

    ReplyDelete
  12. If Volkov is around I should take the time to acknowledge that he was right.

    (We didn't make a bet so I don't have to eat a hat or anything.)

    But my prediction of NDP in 1st, CPC in 2nd, and LPC in 3rd for Winnipeg North was horribly, horribly off.

    ReplyDelete
  13. Well, based on the results, it seems basing numbers of regional and national polls mean absolutely nothing. However, your Vaughan projection does kind of match last night's numbers.

    ReplyDelete
  14. the pollsters are wrong.

    The Liberal support is not going up in Ontario.

    A 17 point swing from the Liberals to the CPC does not show Liberals gaining votes or seats.

    If the CPC run credible candidates the Liberals will be down to 21 seats as Vaughan had the 22nd largest margin of Liberal victory last election.

    ReplyDelete
  15. Departing incumbants and bye elections can make for pretty unpredictable results I suppose. Though I guess we all expected that much. Back to the drawing board.

    ReplyDelete
  16. Shadow,

    Hello, sir. :D

    In all honesty, I would have guessed NDP 1st, Libs 2nd, Cons 3rd. We were all kind of caught by surprise, especially Eric's poor projections...

    BC VoR,

    Hey, dude, Liberals only dropped 2 points from their 2008 result.

    So how could there have been a 17-point swing from the Liberals?

    ReplyDelete
  17. You obviously have no understanding of the law and order issues within the North End of Winnipeg, and the racial-ethnic mix.

    Yes, Wpg North is a ratsnest of old Marxist-Leninist-Trotskyite brands of communism/socialism .. but it’s strictly a racial issue when a Liberal wins in that riding.

    Winnipeg North voted Liberal to stop the FN candidate from representing them .. believe it.

    They voted for a neocon Ignatieff Liberal candidate to stop the NDP FN candidate ... plain and simple ... just check the polling station results!!!!

    The NDP will not field another FN candidate in the riding for the next election ..!!!

    We must not suppress the truth about the Winnipeg North End vote ..!!!!!

    ReplyDelete

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.