Thursday, November 4, 2010

Geek Fight 2 - The Re-Geekening

Over on The Pundits’ Guide, Alice Funke has taken me to task (again) for what she sees as some of the errors in my analysis of and projections for the by-elections that will be taking place at the end of the month.

Personally, I think she’s a little jealous - she only gets to work with dusty old numbers, while I get to work with shiny new numbers.

But let’s get down to business.

First, Alice says that by-elections, by definition, have no incumbents, so using previous general elections as a starting point is incorrect.

Obviously, there aren’t any individual incumbents, but there are incumbent parties. About 70% of by-elections that have been held since 2000 were retained by the incumbent parties. Certainly that is no coincidence, and rarely do incumbent parties lose by-elections by huge margins.

But, in any case, we have no other point from which to start. There are no riding polls, and there are very, very few regional polls, and most of them can’t be used with any level of precision.

In 2009, only the by-election in Montmagny-L’Islet-Kamouraska-Rivière-du-Loup had a radically different outcome than the 2008 general election (or 2006 election for Cumberland-Colchester-Musquodoboit Valley), and even then it was only because of a swap between Conservative and Bloc Québécois voters. While that was going on at the top, the Liberal vote share fell by only 2.2 points and the NDP vote share by 0.7 points. In New Westminster-Coquitlam, the Conservatives gained only 3.1 points. The Liberals lost one point, and the Greens 2.9 points. In Hochelaga, the Bloc gained 1.5 points, the Conservatives lost 0.9 points, and the Greens lost one point. Those are pretty consistent results, demonstrating that previous general elections are a good starting point.

My analysis and projection was based on the differences that occurred in those four by-elections as compared to what would have been expected, based on regional changes in opinion polls. Thankfully, my projection model was in full swing in November 2009, which gave me a consistent basis of comparison for a uniform swing of voting intentions for the by-elections in 2009 and the ones taking place this year.

So rather than just using the 2008 general election as the starting point, I also used the rate at which parties managed to increase (or lose) their vote in 2009 as compared to what we would have expected with uniform swing. And as the 2009 and 2010 by-elections have some similarities (no real ‘star’ like when Bob Rae was elected, no local uproar like when Thomas Mulcair was elected), it was a good set of by-elections to use as a basis of comparison, along with being the most recent.

But I also stretched it back further to the by-elections that took place in 2006, 2007, and 2008. Along with some consistency for the other parties, they showed a more consistent ability by the Conservatives to grow their share of the vote in by-elections over and above what we would expect based on uniform swing. In 2008, the Conservatives averaged 23% more support than would have been expected. It was 24% in 2007.

Alice also brought up the lack of regional polling, and the complete non-existence of riding polls.

While, in Canada, we are not blessed with a lot of polling, we do have a decent amount. And my projection model is designed to get the most out of the polls that are released. I was not using an average of two or three polls for my analysis, I was using the projected vote my model gives me, which takes into account a plethora of factors and includes more than 30 of the most recent polls.

As for riding polls, there is nothing to be done for that. Either I can try to come to some reasonable conclusions with what is available, or I can do nothing at all.

Alice was also concerned that “the projections have not taken one single other factor into account”. That is not entirely true – I took a long look at voting behaviour, going back to the 1950s in one case. Those historical voting records support my claim that the Conservatives and New Democrats will hold their seats in Manitoba.

I could have attempted to take more into account, but it would have required the development of new projection models that don’t yet exist. They will – before the next federal election I intend to have a riding projection model in place that will take into account every factor that can be incorporated.

For now, I wanted to use a one-size-fits-all methodology for the sake of consistency. Adding a few points here because of a large aboriginal population and a few points there because of a candidate’s history would have been, at this point, arbitrary. I wanted to keep my own subjective view out of it.

Alice also said that “obviously they know nothing about the Liberal candidate Kevin Lamoureux”. I can brush off the slight, particularly because I know quite a bit about Mr. Lamoureux.

I will admit that if any one candidate can shake things up in these three by-elections, it is Mr. Lamoureux. But I’m not sold. We all remember the “star” candidates who turned out to be duds.

Remember Svend Robinson in 2006? He cost his party 700 votes in Vancouver Centre. How about André Bachand in 2006? He lost about 2,400 votes for the Conservatives in Sherbrooke. Marc Garneau in 2006 cost his party 4,000 votes in Vaudreuil-Soulanges. Even Michael Ignatieff earned 600 fewer votes in 2006 than the Liberals had gotten in the 2004 election in Etobicoke-Lakeshore.

There is no reason to believe that Mr. Lamoureux is immune. His provincial riding of Inkster (which he lost as an incumbent in 1999 when the rest of the party was punished) only occupies about 1/4th of the riding of Winnipeg North. And his renown (he was first elected to the legislature in 1988 and ran for leadership of the provincial Liberals twice) did not help him in 2000 at the federal level, when he actually lost more than 500 votes for the Liberals compared to their 1997 performance in Winnipeg Centre. That was a slightly worse rate of loss than the party suffered as a whole in Manitoba. Since he is as much of a stranger to 75% of voters in Winnipeg North as he was in Winnipeg Centre, I think his drawing power has the potential to be far less than people think. By-elections are not just about local candidates and provincial voting intentions can't be transferred over to the federal level completely.

Mr. Lamoureux received 3,899 votes in the 2007 provincial election. There are more than 52,000 eligible voters in Winnipeg North. That’s a big gap, and he will need all of those votes again, as well as a good portion of the voters who elected provincial NDP candidates at the provincial level elsewhere in the riding, just to get to 20%. While there is a chance that my projected 9% will be a little low, I’m confident that I will be within spitting distance which, after all, is my goal. And as I brought up in my article, the Liberal vote collapsed in 2008 because people stayed home. Motivating them to go out and vote again in a by-election after a municipal election just finished will not be an easy task.

Alice also brought up that “nor did they mention the murders that took place in” Winnipeg North. No, I didn’t – I’m not sure why I would have. Will that be changing people’s vote? If anything, it will help the tough-on-crime Conservatives or the NDP’s Kevin Chief, who works with youth in the riding. But Winnipeg has always had a relatively high murder rate - so I'm not convinced that anything will change.

Now on to Dauphin-Swan River-Marquette. Alice says that “the Conservatives are rated much too high at 66% for a riding that has two provincial NDP cabinet ministers and is one-quarter aboriginal.”

The provincial New Democratic government is not at its highest level of popularity at the moment. The latest polls put the government down about ten points from their 2007 election, and finance minister Rosann Wowchuk, one of those cabinet ministers from the region, has been a target of criticism of late.

As to the proportion of aboriginals, I’m not sure how to take that into account. Have in-depth studies been done of aboriginal voting behaviour? Are they a monolithic group that can be counted upon to vote a certain way? The Conservatives have managed to be elected with well over 50% in the riding in the last three elections despite its high proportion of aboriginals. And the last poll I've seen of the voting intentions of Manitoba natives indicated that the Conservatives perform relatively well among Métis, who make up about 2/5ths of the aboriginal population in the riding.

Admittedly, yesterday’s revelations about the brouhaha behind the scenes which got Robert Sopuck nominated were unexpected. That’s a risk when projecting this early in a race.

But the riding has a very long history of Conservative support (in all its variations), so Sopuck still enters as the clear and obvious favourite.

Could 66% be too high? Yes, as the party hasn’t done that well before. But turnout is key, and the Conservatives have been effective at getting out the vote in recent by-elections. It doesn’t hurt that the provincial Progressive Conservatives are looking good in Manitoba at the moment.

Finally, Alice brings up Nate Silver. I’m flattered to be mentioned in such company, as Nate was the inspiration for this site (which began a few months after his). But Nate has an easier job than I do, and even then he can be off (as with this week’s mid-terms and this year’s British election).

Why is his job easier? Virtually every election he calls is a race between two people, and it’s an either/or situation. Virtually every race I have to call is a three, four, or five-way contest. He also gets polls for almost every race, at the district level and state level. I get nationwide polls with samples of 50 people in Manitoba and Saskatchewan. He has many pollsters reporting – his short list of “10 or more polls” pollsters has 63 firms on it. My entire list of pollster ratings has 13 on it, and three of those report once in a blue moon and two of them are provincial-only. I have to work with what I have, or not work at all. It's a handicap that I can do nothing about.

Alice sums it up by saying she will eat her shoes if Lamoureux gets 9% and Sopuck gets 66%. It’s a safe bet for her to make. I’ll consider myself to have been right if Lamoureux ends up with anything between 6% and 14% and Sopuck 61% and 72%. What I do is not anything close to an exact science, so it is nearly impossible that I will be exactly right with all of my projections. As with any poll, my calculations have a margin of error and readers are, or at least should be, aware of that.

But as to the Conservatives and New Democrats winning comfortably in Manitoba with the Vaughan race being close, I’d stand by that.


  1. Best part about this geekfight is that one of you will be proven right. You should agree on rules (for example, you both make a prediction at exactly 1 week out) and whoever is closer gets to buy me a beer ;-)

  2. I guess my major point, Éric, is that numbers by and of themselves don't have a life, they are the external manifestation of the demographics and campaigning activity that goes on behind them.

    The previous results in a riding can be used as a starting point if it's demonstrated that historical voting patterns there have a demographic and/or organizational basis. Incumbency, particularly when acquired under unusual circumstances (the 1993 GE for one, and any by-election), can mask that sometimes. This is my point about Dauphin--Swan River--Marquette, MB, for example. There is no historical basis for saying that the riding votes for a conservative party federally, when very clearly significant parts of it vote differently provincially.

    That's not saying whether "the provincial vote translates federally" or not. More must be at play, and local party organization is a big part of it. So now that there is no incumbent, we need to look at all the variables that might come have a bearing on the outcome of a by-election.

    Also, because of the distorting effect of 1993, it's worth looking at older results and/or provincial results as a gut-check on that. In Vaughan that's harder to do, because of the incredible growth in some parts of that riding. Also, the voting patterns of individuals change with age.

    I think it's incorrect to assume that because Kevin Lamoureux couldn't knock off NDP incumbent Pat Martin in Stanley Knowles' old seat ... a seat that has voted NDP with just two exceptions federally since the Winnipeg General Strike, and also elects New Democrats provincially and municipally ... that he couldn't do significantly better in a very well-resourced by-election (i.e., no incumbent) closer to his home turf and an area of much more robust Liberal support. There have been two historic bastions of Liberal support in Winnipeg, even when things got incredibly tough for that party ... River Heights (now represented by Leader Jon Gerrard and formerly represented by Sharon Carstairs, but also inside the federal riding of Anita Neville, formerly Lloyd Axworthy); and the Inkster/Maples area. You're focusing on Lamoureux's seat of Inkster as only being part of the federal riding, but aren't taking into account the slightly weaker but not insignificant Liberal history in The Maples. Also, the Chrétien Liberals were not very popular in western Canada in 2000, while the PCs under Joe Clark ran a better-than-expected campaign that cut into urban western Liberal support a bit.

    ... (con't) ...

  3. ... (con't) ...

    Also, unlike Svend Robinson, who in his day was one of the best prepared and most effective campaigners in the country (which allowed him to survive the 1993 sweep), Lamoureux did not have a very public fall from grace before running. He will do well, not because he is a "star" candidate, but an effective organizer who knows those communities and is known in them. And, again, Svend was running against a Liberal incumbent Hedy Fry, who has a long history in her riding as one of the very few doctors who agreed to treat AIDS patients when no-one else would. That affection alone has allowed her to ride out a lot of other bumps along the road for a very long time.

    You're also extrapolating a swing factor from the set of ridings being contested in the 2009 by-elections to those being called this year. An interesting hypothesis, that will be tested. I wouldn't have done it, because they're from very different provinces, and all in very different situations, but the numbers will settle this all the same.

    So, back to the exact predictions that will cause me to dine on shoe leather: I can live with the 6%-14% range for the Liberals in Winnipeg North and 61%-74% range in DSRM. If either of those results comes to pass, I'll consume cobbling.

    The likely winners in each seat, you may be right on, for now. But methodologically it's not whether you came up with the right answer, but how you did it. I have nothing against an iterative model-building exercise that incorporates more date from each round of elections, but I believe you will be discovering more factors to include this time, rather than being able to count on those previously discovered to predict the shape of these outcomes.

    Oh, and Dan F ... nice try there, buddy ! ;-)

  4. Gosh, I hope that worked. I had a terrible time getting that comment published. What has happened to Blogger?

    Anyways, there should have been a two-part comment. Luckily, I've saved my work.



  5. Oh, finally, I meant to answer the question about why the shooting deaths would be considered significant to mention. Because it's a highly emotional issue that inserts itself into the middle of a campaign and might tend to be more identified with one party than another.

    Parties have good and bad positioning on various issues. If an issue becomes a vote-determining issue and the party is well-positioned on it, that can motivate its supporters, demotivate the supporters of other parties, and actually switch votes. Think HST in New Westminster-Coquitlam last time.

    Campaigning is more than simply "getting out the vote", as many people write in short-hand. It is about persuading, identifying, and then getting out the vote, and a bit of discouraging your opponents in the process. So that expression is overused, and under-understood, in my view.

  6. I don't disagree with virtually all of what you say. Where I think we disagree, however, is on the degree of importance to place on all of these factors.

    Local campaigns matter, yes, but not all the time and often not nearly as much as people think. A vote for Lamoureux is also a vote for Ignatieff, one for Sopuck is one for Harper, etc. Did the average Manitoban voter pay more attention to Lamoureux's campaign launch or the potash decision? I'd wager it's the latter.

    People will keep federal politics in mind when they head to the ballot box, and since the federal and provincial voting intentions of Manitobans are quite different, people may be over-estimating their inability to separate the two.

    In a perfect world I would have incorporated a dozen factors rather than two or three, but time and the information that is actually available is limiting. At present, I have no way to track organizational strength and incorporate it into the model. And it would take a great deal of study to determine whether demographics are a predictor of voting behaviour. I'd wager that regional variations are so great that it would be difficult to predict the voting behaviour of Canadians when faced with four or five options that, often, aren't very different. Nate Silver has the advantage of having only Republicans and Democrats as the two options, and it's much easier to track demographic voting patterns because of it.

    This is still a learning process. Sophisticated projection models are relatively new, and are even rarer in Canada. But debates like these help.

  7. Fair enough, although when people are shooting at your neighbours two blocks over, you're not voting based on the Potash decision.

    Anyways, once again, all in all a very worthy debate.

    Thanks for being a good sport, Éric.

  8. Sorry, my response was written before your shootings response.

    Certainly, it will be in the voters' mind. How to take it into account, however, is a little murky.

  9. I think the 6 to 14% range for Winnipeg North will be way off. I personally believe the low end of final results for the Liberals will be the 2006 result, which saw the Liberals take 21%. At the high end, we could see the Liberals come close to winning the seat (if not actually winning). As one commenter noted, Kevin is winning the sign count by a huge margin, so he obviously has big support in the addition, The Winnipeg north Liberals recently committed to putting up another 1000 signs before election about an advantage!

    As someone who is actually on the ground in Manitoba and knows Winnipeg, I would encourage you to re-evaluate your projection for Winnipeg North. I really like your political please take this as local knowledge, as opposed to criticism.

    P.S. In terms of the D-SR-M riding...I think your projection will be fairly accurate. I believe the Conservatives will win by a large margin as there will be very little competition.

  10. It doesn't really matter, Eric provides fantastic infotainment! I am glad to read him in the globe, because I can read about your predictions at work.


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.