Final projections for Ontario 2014: A Liberal minority

Coup d'oeil sur cet article

After a long campaign and many, many polls, here are my final projections. It's a long post divided mostly in three sections: the polls, the seat projections and a conclusion. Since most people are most likely only interested in the end result, here are the projections with, in order, the voting intentions based on the most recent polls, the seat projections with the 95% confidence intervals and finally, the chances of winning. Both the intervals and the probabilities are estimated using 5000 simulations accounting for the uncertainty of the polls as well as of the electoral system.

Proj Ontario final white

The detailed, riding-by-riding projections are available here. You also have riding-level probabilities.

There are five scenarios possibles, from a Liberal majority to, in theory, a Conservative one. The latter would require a significant underestimation of the PC in the polls of course. Here are the probabilities of these four events happening:

Lib majority 25%
Lib minority 55%
PC minority 17%
PC majority 2%
Tie 2%
In other words, the most likely outcome (by far) is another Liberal minority. And actually, Kathleen Wynne is more likely to get a majority than to lose her job.

Remember that if you don't trust the polls, you can use your own numbers in the simulator.


Before I go into the details, here are a couple of important points everybody should read:

- My model and projections are meant to show the most likely outcomes based on the information available. This includes past elections results, the polls (both province wide and at the riding-level), historical and current regional trends, the effects of a long-term incumbency, etc etc. It's the best that can be done since we don't have 107 polls.

-At the end of the day, despite my best efforts, it's naturally highly dependent on the polls. If they are wrong, my most likely scenario will also be incorrect. However, the intervals and probabilities should take care of that (and have in the past such as in BC).

- For this election in particular, it seems the provincial swings will be small, with most of the parties being quite close to their 2011 results. But it doesn't mean there won't be regional or local changes. Ontario elections seem to have more regional swings than Quebec ones for instance. This kind of variations are harder to track and predict and errors are more likely. Models such as mine usually assume that if the swing is zero, then the results in a given riding will be the results of the last election. So if the poll average is right, expect to see potentially a lot of variations cancelling each other.

- Seats projections like mine have proven to work, especially if the polls aren't completely wrong as in BC last year. But, and I can't stress that enough, there will be surprises. Always. There is a limit in predicting 107 races with only 4 numbers (the provincial %).

1. The polls.

Ontario polls June 12th 2014

The recent Quebec election had polls that agreed with each other. And there were some clear trends throughout the campaign (specifically, the PQ declining and the CAQ having a late surge). Quite the opposite of what we got in Ontario. They've been all over the place. For the longest time, we couldn't even get two consecutive polls with the same party leading. At at the very end, we started seeing very different numbers for all parties. For instance Forum has the Liberals up by 6 (41 to 35) and the NDP a distant third (20), while Ipsos has a close three-way race. This goes way beyond the margins of error.

The situation is made even worse by the use of various (and very different) "likely voters" models. Ekos, Abacus, Angus-Reid and Ipsos all have such models which are supposed to correct for turnout (think: people over 55 vote a lot more than people under 35, so you need to adjust for that instead of trying to get a sample that replicates the census data). Most find that likely voters adjutsments are helping the PC over the Liberals (except Ekos). It makes sense if you remember that the Tories are doing better among men and older voters, two groups more susceptible to get out to vote. However, let's remember that the use of such models is relatively new in Canadian politics. And if you have the wrong model, you'll be dead wrong (ask Mitt Romney and the Republicans in the States). The table below shows the effects of using a likely voters model by firm. As mentionned, it usually boosts the PC and decreases the Liberals.

Likely voters adjustments
Angus-Reid -2 4 -2 0
Ipsos -3 5 0 -1
Abacus 1 4 -3 -1
Ekos 5 5 -2 -5
The question of course is: should I use the numbers among decided voters or among likely voters? Or, if you prefer, do you trust the likely voters models?

I'm a little bit in-between. I find Ipsos' definition way too restrictive and they lose all a big chunk of their samples. They argue that they should indeed lose almost 50% since only 50% of people vote nowadays. I don't agree with that as I believe that if you agree to answer to a poll, you are already more likely to vote. The simple fact you're not hanging up signals that you are more likely to vote (in average). Therefore, such adjustments should not lose 50% of the sample. Also, I'm not sure why we'd need models with such big adjustments. In 2011, polls did relatively well and they didn't use any model (they underestimated the PC by around 1-2 points; They also underestimated the Liberals but by much less). Other firms have less restrictive models, even though Ekos is the one with the biggest adjustments. I think Abacus is probably the one that got it right, but that's just my guess.

I tried to use only numbers among decided voters or only among likely voters. It does make a difference. The lead for the Grits is much more important if you look among decided voters only. At the end, I decided to use the likely voters numbers for the four firms with such a model. I find the Ipsos nymbers a little bit crazy (especially for the NDP which is having a late surge to 30%; Ipsos was dead wrong about the fourth party in Quebec last April, having it at 13% instead of 7%) but it's compensated by a firm like Ekos. I do believe the PC is higher than the 32-33% we'd find among decided voters, but not by 6 points. The numbers in the graph and in my projections therefore use LV whenever possible. It is combined with the numbers from firms who do not have such models, such as Oracle, Leger and Forum.

The NDP is really the party varying a lot, from 17% in the latest Ekos to 30% in the last Ipsos. A 13 points difference! The PC on the other hand has been dead stable at 35% (this stability makes me think the likely voters adjustments for this party do work). When we compare the polls, we see an almost perfect negative relationship between the level of the NDP and the Liberals. It could make sense as there are a lot of swing voter between the two. Unfortunately for the NDP, when the time comes to cast a ballot, I expect most of these swing voters to choose the Liberals. Especially with a race this close (or seen as close).

Finally, Forum has been doing incredibly well in the most recent provincial elections. And this firm has published two polls in the last 3 days with the Liberals way ahead. While I do think they overestimate the Liberals, this reinforces my beliefs that a Liberal majority is actually more likely than a PC minority.

Regionally, it's hard to really compare because of the small sample sizes and the fact that pollsters don't necessarily agree on the various regions in Ontario (for instance, Northern Ontario sometimes includes central Ontario as well; What is referred to as the "905" isn't the same region everywhere; etc). But overall, the Liberals are clearly dominating in Toronto (except according to Ipsos). In the 905 (including Niagara and Hamilton), it's close with a slight edge for the Liberals (which is similar to 2011). The Tories are first in the East (although the very recent polls don't show that anymore) while the Southwest is maybe the best battleground after the 905. As for Central and Northern Ontario, there is too much volatility to really tell.

Tim Hudak is clearly suffering from a gender gap. Or, as I like to call it, a "binders full of women" type of problem. His party is usually first among men but way behind among women (while the Liberals are mostly stable). The Conservatives are also usually more popular among older voters, especially over 60.

Finally, if we look at other variables than the voting intentions, we find that most Ontarians think it's time for a change. But at the same time, we find a lot of them don't want or even fear a Conservative government. And the people who want a change are split (not evenly) between PC and NDP. So I think we could have this weird situation where the Liberals are reelected despite not being popular.

To conclude this section, the various polling firms really don't agree with each other. It could be because of the methodology (firms using IVR technology such as Ekos and Forum find the Liberals much higher than firms using online panels such as Ipsos or Abacus). It could be because of the use of likely voters. Or a mix of both. But in average, the race is close but seems to favor the Liberals.

2. The seat projections

While the race is as close as it gets in term of votes, it's a little bit clearer if we look at the seats. As far as forming the next Ontario government, it is NOT a 50-50 race. This is simply not true. Yes, both Kathleen Wynne and Tim Hudak could be Premier after tomorrow night, but the odds are not even. Also, even though my model doesn't directly use the regional poll numbers, I do take them into account. And the observed swings do not indicate that the PC could win the most seats without having a clear lead in the popular vote. In particular, we don't  have any reason to believe that the Tories will gain many seats in the 905. I'm not saying it won't happen. 2 months ago in Quebec, the CAQ managed to win a lot of seats in the Montreal suburbs simply by winning every wingle close race this party was in. So vote efficiency can come into play. I'm just saying that the most likely scenario isn't that. The graph belows shows you the chances of winning of the Liberals as a function of the lead over the PC. As you can see, even if the two parties are tied tomorrow night (so a difference of 0), the OLP is clearly the favorite.

Chances of winning Ontario, June 12th

This factor alone explains why I don't find this election that close. No matter if we use the numbers among likely or decided voters, the Liberals are either tied or ahead. In both cases, they should win the most seats. And if you think I'm making that up, remember that in 2011, The Liberals almost won a majority with only a 2 points lead. A party that has an electoral advantage with the map and/or the system can't really lose it in just three years. And as I've said, the regional swings do not indicate that either.

With that being said, the Conservatives won more seats than expected in 2011 (even by using the right percentages). So again, expect surprises. I just don't think these surprises could be enough to make Tim Hudak Premier. His only chance is really to be underestimated in the polls. Or, alternatively, that the polls that put him ahead (usually using likely voters) are right and the other are wrong.

Another reason I don't believe this election is that close is because most ridings aren't in play. The table below shows you the probability of winning of the candidate projected first. In some ridings, the model is 100% sure of the call and the probability of winning is 100%. But in others, such as in the close three-way race of Kitchener-Waterloo, the Conservative candidate is only projected to win 38% of the time.

Probability of winning of the frontrunner # of ridings
30-39% 1
40-49% 0
50-59% 6
60-69% 7
70-79% 3
80-89% 16
90-99% 52
100% 22
The model has never made a mistake while calling a riding with a 100% confidence, so that's 22 ridings right there where there is no uncertainty at all. The ones between 90 and 99% can in theory end up different than m projections, but it'd require big mistakes from the polls and/or important local effects. Realistically, the ridings below 80% can be seen as unsafe. There are 17 of them. It's relatively typical of elections with our electoral system. Out of these 17, the PC is projected to win 6. This party is projected with 0% chances in 4. This leaves 7 ridings who could be the key to a Tory victory. A good turnout in these ridings and a surprise becomes possible. We're talking about Glenngarry-Prescoott-Russell, Brant, London-West, Cambridge, Windsor-West (a very long shot), Niagara Fall and Brampton-Springdale. Winning them all would make Tim Hudak the enxt Premier. But while it's possible, it's highly unlikely. Because remember that it'd require the PC to win 13 close races out of 13 (cause the 6 ridings currently projected to go PC aren't safe either since the probabilities of winning are low).

3. Remarks and conclusions

I honestly think I've done my best. I collected all the data I could, crushed the numbers as much as I could and left my intuitions or subjectivity out of it. At the end, I'm pretty confident in projecting a Liberal minority. If I used the poll numbers among decided voters, the situation would be very clear. Using the Likely voters makes it harder. However, the Liberals hold a certain advantage with the electoral system. Therefore, they should win the most seats. There also are a couple of factors that, according to me, will help them. First of all, while PC voters are more committed and likely to vote, I believe undecided or swing voters will ultimately choose the Liberals. People seem very polarized on Tim Hudak. Every poll I've seen has shown that outside of Conservative voters, people just don't like him. And the idea of a PC government isn't welcoming to many voters. In particular, a lot of swing and NDP voters seem to really want to avoid that. Thus, it's not much of a stretch to imagine that some NDP voters and undecided will ultimately vote for Kathleen Wynne tomorrow. This could mitigates the effects of the likely voters. So in conclusion, here is what I think:

- While the Conservatives have more commited and likely voters, the Liberals are probably not overestimated among decided voters. Similarly to what happened in 2007, I wouldn't be surprised if both parties end up higher than the simple poll average. With the PC getting a bigger boost. Therefore, I do not believe all the models that change the race from a Liberal lead to a PC one. Of course, if I'm wrong, make sure to remind me of this on twitter (@2closetocall). Altough to be honnest, starting tomorrow, I'm in world cup mode :)

- While there is considerable uncertainty as far as the vote is concerned, I truly feel that the situation is clearer in term of seats. The Liberals are clearly the favorite and should remain in power. I'm not saying Tim Hudak won't become Premier, I'm simply saying this isn't the most likely scenario. But Christy Clark wasn't supposed to remain Premier in BC last year. And the LA Kings have shown this year that probabilities sometimes don't mean much. So makes of my probabilities what you want.

Please, if you live in Ontario, go out and vote. This is important. It's always sad to see. And thanks for reading me.

3 commentaire(s)

Thierry Soucie dit :
12 juin 2014 à 7 h 24 min

Je ferai ma prédiction finale en utilisant la moyenne des sondages ci-haut:

47 OLP 41 PC 19 NDP

La projection circonscription par circonscription se trouvera ci-bas. Les sondages ne rendent pas cet exercice facile vu leur volatilité. J'ai l'impression que les résultats seront plutôt surprenant, contrairement, notamment, à l'élection québécoise.

Thierry Soucie dit :
12 juin 2014 à 7 h 26 min

Ajax—Pickering OLP Algoma—Manitoulin NDP Ancaster—Dundas—Flamborough—Westdale OLP Barrie PC Beaches—East York NDP Bramalea—Gore—Malton NDP Brampton-Ouest OLP Brampton—Springdale OLP Brant PC Bruce—Grey—Owen Sound PC Burlington OLP Cambridge OLP Carleton—Mississippi Mills PC Chatham-Kent—Essex PC Davenport NDP Don Valley-Est OLP Don Valley-Ouest OLP Dufferin—Caledon PC Durham PC Eglinton—Lawrence OLP Elgin—Middlesex—London PC Essex NDP Etobicoke-Centre OLP Etobicoke—Lakeshore OLP Etobicoke-Nord OLP Glengarry—Prescott—Russell PC Guelph OLP Haldimand—Norfolk PC Haliburton—Kawartha Lakes—Brock PC Halton PC Hamilton-Centre NDP Hamilton-Est—Stoney Creek NDP Hamilton Mountain NDP Huron—Bruce PC Kenora-Rainy River NDP Kingston et les Îles OLP Kitchener-Centre PC Kitchener—Conestoga PC Kitchener—Waterloo PC Lambton—Kent—Middlesex PC Lanark—Frontenac—Lennox and Addington PC Leeds—Grenville PC London-Centre-Nord OLP London—Fanshawe NDP London-Ouest OLP Markham—Unionville OLP Mississauga—Brampton-Sud OLP Mississauga—Erindale OLP Mississauga-Est—Cooksville OLP Mississauga—Streetsville OLP Mississauga-Sud OLP

Thierry Soucie dit :
12 juin 2014 à 7 h 26 min

Nepean—Carleton PC Newmarket—Aurora PC Niagara Falls PC Niagara-Ouest—Glanbrook PC Nickel Belt NDP Nipissing PC Northumberland—Quinte West PC Oak Ridges—Markham OLP Oakville OLP Oshawa PC Ottawa-Centre OLP Ottawa—Orléans OLP Ottawa-Ouest—Nepean PC Ottawa-Sud OLP Ottawa—Vanier OLP Oxford PC Parkdale—High Park NDP Parry Sound—Muskoka PC Perth—Wellington PC Peterborough OLP Pickering—Scarborough-Est OLP Prince Edward—Hastings PC Renfrew—Nipissing—Pembroke PC Richmond Hill OLP Sarnia—Lambton PC Sault Ste. Marie OLP Scarborough—Agincourt OLP Scarborough-Centre OLP Scarborough—Guildwood OLP Scarborough—Rouge River NDP Scarborough-Sud-Ouest OLP Simcoe—Grey PC Simcoe-Nord PC St. Catharines PC St. Paul's OLP Stormont—Dundas—South Glengarry PC Sudbury NDP Thornhill PC Thunder Bay—Atikokan OLP Thunder Bay—Superior-Nord OLP Timiskaming-Cochrane NDP Timmins—Baie James NDP Toronto-Centre OLP Toronto—Danforth NDP Trinity—Spadina OLP Vaughan OLP Welland NDP Wellington—Halton Hills PC Whitby—Oshawa PC Willowdale OLP Windsor-Ouest NDP Windsor—Tecumseh OLP York-Centre OLP York-Ouest OLP York—Simcoe PC York-Sud—Weston OLP