Monday, April 20, 2015

Alberta projection updated with candidate slates

The Alberta projection has now been updated to take into account the number of candidates each party will have on the ballot on May 5.

Only the Progressive Conservatives and New Democrats managed to nominate candidates in all 87 ridings, while Wildrose had one candidate barred from running. The Liberals came up just short of a two-thirds slate, while the Alberta Party and Greens fell well short of a 50% slate.

The projection has accordingly been adjusted to reflect these vacant spots. The Liberals have been reduced by the proportion of candidates they have on the ballot, as past experience shows that polls are largely unable to account for this fact.

The table below is a list of parties that did not run a full slate of candidates in recent elections but were nevertheless included in most polls. Parties that had a slate of at least 90% were not included, as it appears that this has little effect on the accuracy of the polls.

All of the numbers have also been rounded, because when we're looking at numbers as small as these a few decimal points can make a big difference. The issue, though, is that most polls only report whole numbers.
As you can see from the chart above, in virtually all recent cases the polls overestimated the support of parties not running a full slate by roughly the same proportion as the number of candidates they were running.

The poll average column shows the average of all polls conducted within seven days of the vote. The expected result shows the poll average adjusted by the candidate slate.

Only in the recent cases of the Saskatchewan Liberals, the Manitoba Greens, and the People's Alliance in New Brunswick have parties outperformed the expected result. And in the case of the Saskatchewan Liberals, their slate was so small that the adjustment reduced them to almost zero. They actually took 0.6%, just above the level needed to round them down to zero.

It would seem that many poll respondents are unaware of whether or not a candidate for a particular party is running in their riding. How this affects their voting behaviour is a puzzle. Do they discover only in the ballot booth that their favoured party is not on the ballot, or do they discover this in only the very last days of the campaign? And what do these people do? Spoil their ballot, not show up, or vote for another party?

Or is this overestimation just a sign of a party's lack of funds, organization, and get-out-the-vote infrastructure?

In the context of Alberta, what will these Liberal voters do? They could be worth three percentage points. An argument could be made that they could go to any of the three major parties, or the Alberta Party in ridings where they have a candidate where the Liberals do not. But it is impossible to speculate with much confidence. Rural Liberals may be different from Edmonton Liberals and Calgary Liberals. The local race may play a very big role as well. They may cast a ballot for the NDP in ridings where they have a good chance, or for the PCs in ridings where the race is between the Tories and Wildrose. Or they could even vote for Wildrose to send the PCs a message.

In regards to the other parties, the projection has also been adjusted. Support for other parties is often overestimated in polls, and in the past I have found that an effective measure of their likely support is based on two factors: the average support a party received in the previous election in the ridings where it had a candidate, and the number of candidates it has nominated in the current election.

In the case of the Alberta Party, to take an example, it has nominated 36 candidates. The party averaged about 3% per candidate in 2012. Take into account leader Greg Clark's potential performance in Calgary-Elbow, and you end up with the party capturing 1.5% of the vote. That is what the projection assumes will happen, and added to that are the expected vote shares for the Greens, independents, and an assortment of fringe parties.


  1. Somehow I can't see the NDP even challenging Wildrose when the actual votes get counted ??

  2. Were these projections to actually be the result on election day, Alberta would be in a minority scenario. The question then: what coalition (be it de facto or formal) would be most likely to result?

    NDP-Wildrose seems least likely on general principles. While both obviously want the PCs out, I'm not sure there's anything else they could agree on. Then again, "politics makes strange bedfellows" ...

    Between Wildrose-PC and NDP-PC, the former seems more natural, but, precisely for that reason, hardly a lock. Arguably, Wildrose has more to lose by governing with the PCs than the NDP do.

    I'd be interested to hear from some Albertans about what they consider most plausible.

    1. I think whoever wins the most seats will attempt to have a crack at governing. The other two parties will take turns trying to prop the governing minority up.

      I think the Wildrose and NDP may have some differences on the role of government, but I think they can work together on accountability/ethics issues. If the federal NDP propped up Harper's CPC at certain points during the minority period, I am sure something could happen here.

    2. An NDP-WR coalition (which I don't expect to be necessary) could go really well, or really terribly.

      Personally, I'd love to put the NDP in charge of all of the social issues, but have WR running the budget. It would be a sort of perfect libertarian combination.

      If they did it the other way around, it would be like the Republic of Gilead in the Handmaid's Tale. That would obviously be awful.

    3. That is an interesting thought, I like it actually, but I am not sure a coalition like that is practical.

    4. If it turns out to be a close split hung parliament I could see an issue by issue consensus government emerge.

      I don't think any of the front runners can join in even in an accord, the PC have burned a lot of good will in the last 44 years hard for the NDP to team up with them and impossible for the WR work with them without looking like sellouts.

      The WR and NDP working together would be governmental equivalent of Schizophrenia.

    5. Chris,

      Prentice remains premier until he resigns. If the results were identical to the projection Prentice would remain as premier and would be given the chance to present a Throne Speech.

      I don't think a WR-NDP Government likely even possible. WR supporters would be up-in-arms if it happened and frankly I'm not sure enough of a common platform exists between the two parties to make it work.

      WR-PC on the face of it seems more likely but, from a political standpoint I'm not sure it works-both are essentially going after the same voters and that means "one has to die for the other one to thrive". Could a handful of rightwing PCs prop up a WR Government? Yes, but, a formal coalition would probably require a merger of the parties.

      Having said all that I think a WR majority Government is the probable outcome.

  3. WR was never going to win many votes in Edmonton-Strathcona, so them not having a candidate there will have no marginal effect.

    It will probably inflate the PC vote count there, probably getting them above 5%. I don't know if that's a meaningful threshold in Alberta (it has campaign finance implications federally).

  4. Ok, now the NDP leaders the PCs by 4% and trails Wildrose by less than 2% and you're still giving them less seats than the PCs? I'm starting to think you just don't like the NDP, Eric.

    1. Sigh.

      All I did was remove some support from the Liberals to reflect their lack of candidates. The proportional margin between the other parties has not changed whatsoever. Why would the seat projection, which is a proportional model, change?

      The model spits out numbers. I do not adjust the numbers whatsoever. I do not care how any of the parties do.

    2. What does Eric have to benefit by skewing the NDP seat numbers?

      This is the reality of the FPTP system. There might even be a scenario where the PCs are third in popular vote but first in seat count. Look at the 1919 election in Ontario.

    3. I am sure for as long as you run a blog on polls, election predictions and breaking down model performance, occasionally people will accuse you of skewing things deliberately to favour a particular outcome.

    4. Oh yes, it happens all the time. It is no less exhausting.

      And if I wanted to do that, it would be a fool's errand. This goes to a lot of the tin-foil-hat conspiracy theories leveled at pollsters.

      How would skewing things favour an outcome? Skewing the NDP downwards, for instance, might make people less likely to jump on the bandwagon. Or, they might make them more likely to vote NDP to ensure they get good numbers (or, say, feel that voting NDP is safe since they likely will only form the opposition).

      If I had the NDP very high, maybe their turnout would be depressed (why bother, they'll win anyway). Or maybe their supporters would be excited and vote in droves. etc. etc.

      If you were a partisan for Wildrose, would you want their numbers to be sky-high, or would you prefer them to be in a close race? If I was thinking strategically, I'd want the close race to reduce the potential for Wildrose opponents to be scared into voting PC. But if I was a wild-eyed partisan, I'd want the polls to be 80%, I guess.

      If anyone had any incentive to skew the numbers for a particular outcome in polls, they might instead have unintended consequences.

      But what should I expect? When I'm writing about how party X is doing well in polls, to some partisans it is a clear sign that I love party X. When I then write that party X is doing badly in the polls, it is a clear sign to others that I hate party X. The idea that what I'm writing is only dictated by trends in polling escapes them.

      When the Liberals were doing well in the polls last summer, I was a clear Liberal. Now that they're doing badly, I'm a Harperite or Dipper. It is all just so dumb.

    5. Ryan Foster,

      You are being a bit hasty in your assumptions, The NDP leads the PCs by 4% but, where is that lead? It's not in the rural areas and to a lesser extent Calgary where the PCs are polling "well". If one believes the polls (the basis for Eric's seat projection) the NDP's lead is housed in Edmonton, a widening lead in the capital will not give the NDP any more seats! First Past The Post simply does not work that way. A widening lead (in Edmonton) will only give the NDP a less efficient vote in general.

    6. Eric,

      I am not accusing you of being partisan in any way, however, one must be willing to at least entertain the possibility at times that a particularly model or methodology may favour Party A over Party Y. In this particular circumstance I don't believe that is the case. However, perhaps at a later date it would be beneficial to explain how you and your model prevent against this outcome.

      Ryan Foster is correct when he writes the gap between the PCs and NDP is four points but, he needs to take the second step and determine where that lead is located. If he had he would deduce that it is mainly contained to Edmonton (at least if he believes the polls, which one assumes he does since, he visited your site whose numbers are poll based). As others have pointed out the reality of the Westminster system is; localised increases in popular vote tend to lead toward wasted or an inefficient vote pattern and do not usually increase the number of parliamentary seats to a particular party.

    7. Obviously, Eric is a flip-flopper, who can't make up his mind who he prefers for his positive bias, but his negative bias is to suppress the Green-BQ coalition from forming the next government!

      As for the intrinsic bias of the model, there are three possible sources in my mind. First, there is the three part classification system of Government, Opposition, and Other. I think the model does quite well on the first category, but the fit to reality might be harder for the others. Specifically, in the case of significant third/forth parties, who get lumped in with the fringe. It would be interesting to see how the model might differ with such an adjustment. Next, I'm not sure if the model only accounts for the most recent election results, or like the polls uses a weighted history. This could induce a bias in either direction either propping up numbers for a party in decline or holding back numbers for a party on the rise. Finally, the incumbency bonus might actually differ across parties (for example, Conservatives might be more likely to vote for their incumbents, but Liberals/NDPs might be more open to switching between the two). Overall, every assumption of the model is a potential source of bias, it’s just a matter of trying to make the overall bias match reality.

    8. There are two important reasons to expect that Éric is not intentionally biasing the results.

      1. His livelihood rests on his credibility as an analyst of polls. If he says stuff that proves not to be an accurate reflection of electorail reality, and he can't adequately explain what happened later, he loses that credibility.

      2. Éric has nothing to gain or lose from the outcome of an election in Alberta. Unless we think that some Alberta partisan is secretly paying him, why would he do that?

      Éric benefits from being right. It is unreasonable to expect him to abandon that as an objective.

  5. Eric, I know you don't include these polls in your analysis, but what is your take on the newest numbers from 1abvote?

    11 point lead over the PCs, Wildrose down 8, 1 point lead over Wildrose in Calgary, and still 29% in Rural Alberta. In addition to the poll showing Joe Ceci leading in Calgary-Fort, this would mean NDP would likely walk away with around 1/3 of Calgary seats, and a significant number in Rural Alberta too (the better question is - outside of Lethbridge - where?!).

    1. I'm looking forward to seeing how the poll performs by election day. It is refreshing that Singh is so open about the potential biases of the methodology, but that is also quite concerning.

  6. Of the 11 seats at 55% or under, 2 are WR over PC, 6 are PC over WR and 3 are NDP-PC races. With a little bit more of a PC meltdown, WR will likely be in a strong minority position where even the Liberals might prop them up, perhaps even at a point where Liberal abstention would suffice.

  7. I like the new projection. The NDP is now forecast to sweep Edmonton, as I predicted.

  8. Those three way races also add some volatility. Ridings like West Yellowhead and Wetaskiwin-Camrose should be fun to watch.


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.