What is different about the newest EKOS poll out this morning is that it is the first one from the company to estimate likely voters. Their likely voter model, which I'll discuss below, increases the Liberal lead from five points to nine, which is at odds with what Abacus Data and Ipsos Reid most recently found. While the numbers themselves were different, both firms did find that the PCs improved their position relative to the Liberals among likely voters.
In any case, the projection has only nudged a little in favour of the Liberals, with a 1.6-point gain to 39.4% (or between 38% and 43%). The Tories fell 0.3 points to 33.5% (or between 32% and 37%), while the NDP was down 0.7 points to 19.6% (or between 18% and 21%). The Greens were at 6.4% (or between 5% and 8%).
The Liberals are now narrowly over the majority threshold, with 55 seats (up four from yesterday). Their range has them straddling the line, with between 50 and 67 seats. The Tories dropped four seats to 37, or between 25 and 42, while the NDP was unchanged at 15 (or between 13 and 16).
Now that EKOS is doing a rolling poll, its numbers take up a lot of space in the projection. So, a few words about how it is being handled. Each rolling poll is included in the model, but is reduced significantly when a new poll comes out. EKOS is now doing a three-day rolling poll, dropping one day from the sample with each new survey. The projection model is doing something similar. When a new rolling poll comes out, the previous one is reduced to just 33% of its previous weight (1/3 days), and this in addition to the decay that occurs due to the poll being one day older.
The model also caps the weight of any individual poll at 67% of the total weight, but in the case of a rolling poll it caps the combined weight of polls that overlap with the newest poll at 67% as well. In the case of today's projection, the EKOS rolling polls of May 30-June 2, June 1-3, and June 2-4 take up a combined weight of 67% of the projection. The most heavily weighted poll after the new EKOS poll is, thus, the Abacus poll from May 28-31.
Among likely voters, EKOS gives the Liberals 41.4%, a significant uptick from the tally among eligible voters. The PCs also gained, to 32%, while the NDP fell to 17.8% and the Greens to 6.2%.
Other polls estimating likely voter support (Abacus and Ipsos) have shown broadly similar effects for the PCs, NDP, and Greens. The Tories tend to gain while the NDP and Greens drop. But the Liberals have gotten either insignificant or no gains among likely voters, with the margin between them and the PCs either narrowing or increasing to their disadvantage. EKOS finds the gap to almost double. Will other polls bear this out?
The EKOS model is interesting, giving each respondent a 'score' based on whether they have voted in the past, their emotional engagement, their stated intention to vote, and whether or not they know where their polling station is. They then reduce the sample to just the top 50% of respondents, to mirror the turnout of 2011. It is an interesting approach, not dissimilar to Abacus's. We will see if it works - but what if turnout is significantly different in this election?
The spread in these Ontario polls seems really high. How does it compare to polls in other provinces?
ReplyDeleteThis comment has been removed by the author.
ReplyDeleteEkos is going all in.
ReplyDeleteIf it turns out to be a CP win of any type might it be the end of the road for them as a "trusted" pollster?
If it turns out they are the most accurate pollster then it just as easily could be "the end of the road" for others. Quite likely each pollster will end up knowing why they missed the mark after the election once they get chance to do a big reflective.
DeleteEkos had a weak Conservative minority with Layton as PM if he got LIberal support last day before the 2011 federal election.
DeleteIn 2011 ekos had the PC under supported by 4 points. once again a 1 time out of 20 MISS.
In the BC election EKOS had the NDP winning 40.5 - 34.5. instead of losing 44.1 to 39.7..... yest another sawing and a miss.
EKOS did not poll for the Alberta election. Maybe because they did not have a Left of centre horse in that race.
They are not new to huge misses. But somehow keep getting reported without riders saying how far off they have been.
The 308-aggregate gives me:
ReplyDelete57 OLP
34 PC
16 NDP
The reported EKOS numbers give me:
53 OLP
35 PC
19 NDP
And the likely-voters give me:
65 OLP
28 PC
14 NDP
These are the results that I'd like please and thank you.
ReplyDeleteCan I borrow some money?
DeleteEkos consistently shows the Liberals doing well (and Ipsos consistently shows the PCs doing well). I have a hard time believing that Liberal support is nearly as high as the Ekos polls suggest. Somehow I think that the weighting scheme needs to be adjusted so that no one polling company is given a massive weight simply because it has the most recent poll. If you reduce the weight of the recent Ekos polls and increase the weight of the Ipsos poll from a few days ago it averages out and you get a tie between Liberals and PCs again. I find it hard to believe that support has changed massively in favour of the Liberals just in the last few days. I would have thought that Wynne's poor debate performance (being repeatedly attacked for the gas plant scandal) and the MaRS scandal would help the PCs and/or the NDP. Until an Ipsos poll comes out showing the Liberals in the lead I find it hard to believe that the Liberals are in majority territory.
ReplyDeleteThat's compensating for something simply on the basis your "gut feeling" - given that we haven't had the election yet, we have no idea which of these pollsters is closer to the mark. That is just something Eric has to deal with - it would be much worse to create a false balance just because you think Ipsos is more accurate on the back of, well, nothing.
DeleteExactly, gut feeling is pretty useless when it comes to determining accuracy. A more accurate way of saying something similar though could be that the polling method used by company x often has more accurate results. Then follow that up with some type of statistical analysis and you could be well beyond gut feeling
DeleteWhile I understand the irritation with the Liberals because of the scandals, the PCs have basically pushed away anyone who is not a true believer and now have to live with the consequences. Sad.
DeleteIt's just another example of the "preferred" pollster thing.
ReplyDeleteI like the results but will only believe them if they appear as here on the12th !!
Debate effects are generally overestimated. Viewers tend to be among the more engaged, and likely more decided, members of the electorate.
ReplyDeleteWell, the debate itself may not be watched by the general electorate, but the clips and knock-out lines on the news and the constant analysis of who won and who lost does make it to the casual voter. In this regard, a bad or good debate can greatly influence the results, even if most of the debate itself turns out to be "pointless".
DeleteI can't speak for others but I'm engaged but not decided. The only debate certainly confirmed my "not Liberal" stance but did leave me, for the first time in the campaign, considering Hudak.
DeleteI am surprised that Forum did not have riding based polls in this election cycle. Perhaps they are currently working on it.
ReplyDeleteIn 2011, I thought that Forum did a good job with those showing certain ridings where there are unique circumstance that may not be shown by overall projections.
Forum did riding polls in Thunder Bay and Kenora-Rainy River last week:
Deletehttp://www.forumresearch.com/polls.asp
I'm hoping Forum does a Kitchener-Waterloo Riding poll. Eric's model has had the NDP, Liberals and PCs in the lead at one point or another. I'd like to see a snap shot of what is going on here.
DeleteThere is something new in this election. Apparently you can go to the polling station and DECLINE your ballot !!
ReplyDeleteAnybody know how this works and what the effects may be ????
That is not new, it has been an option in Ontario for many, many years.
DeleteHey Peter, I'm a poll clerk in the upcoming election. To decline your ballot, you can either hand back your ballot as soon as you are given it, stating that you wish to decline your ballot, or you can take the ballot, go behind the voting screen, mark nothing, and go back to the table and declare that you wish to decline your ballot. Once you declare that, the DRO will mark "declined" on the back of your ballot, and it will be placed in its own envelope (specifically reserved for declined ballots). The poll clerk will then add a comment to the list of electors that you have declined your ballot. It is counted separately than spoiled and regular ballots. In effect, it's voting "none of the above."
DeleteJon thanks. While I've lived in Ont all my life I had never run into the "Decline" thing. Wonder what happens to that Decline number ??
DeleteThe media never report declined ballots, but DRO tally sheets do include a line for counting them. (Same for spoiled ballots.) I've been a DRO, a poll clerk and a scrutineer in various elections, and never once seen a declined ballot, but the option has always been there.
DeleteI like that Ekos has explained their likely voter model in such detail. Their scoring method (and the choice of a 50% cut) can obviously be criticized - as can everyone else's - but it is clearly defensible. And I'm sure they were as surprised as anyone else to see the Liberal lead increase once they applied the model. But, they are reporting the results they get from applying a defensible methodology, which is the best that they or anyone else can do.
ReplyDeleteThis is important for the conspiracy-mongers like BC Voice of Unreason. Ipsos, Ekos, Forum, Abacus, and anyone else you care to name, all do their best to provide reliable, accurate results from defensible, cost-effective methodologies. All of their methods are arguable. But none of their choices can be reasonably criticized on the ground that they produce results favourable or unfavourable to any particular party.
A question for you Éric, just out of curiosity: What would your seat projection model forecast if we asserted topline numbers of Lib/PC/NDP all at 32% each? I assume a tie in the vote would produce a minority scenario. But would I be right to assume it would be a Liberal minority? And how large of one, if so?
ReplyDeleteI can't speak for Eric's model, but if I do as you say in mine, I get:
Delete40 PC
37 OLP
30 NDP
Not exactly split evenly, but not bad at all. I guess a strong NDP hurts the OLP more than a strong PC, so the latter slips in between and ends up in power.
Using the same type of scenario (24% for the 4 major parties), Québec would give:
45 PLQ
38 PQ
32 CAQ
10 QS
And for Canada, with 32% for the 3 major parties in all regions (except Québec where the 4, including the BQ, get 24%), the results would be:
113 CPC
88 NDP
87 LPC
19 BQ
1 GPC
Thanks Thierry. So, not a Lib minority. Interesting. The Quebec example is fascinating - QS could get 24% of the vote and only get 10 seats?!? Whoa.
DeleteI just want to correct a small mistake in my post above, I meant to say that a strong NDP hurts the OLP more than it hurts the PC.
DeleteAs for the Québec example, it is quite surprising, but keep in mind that QS's vote is concentrated in Montréal, so less possible gains outside of the island. In my mind, it clearly shows how broken our system is when a party can collet 24% of the vote but only 8% of the seats...
If I use my 2018 projector insterad of my 2014 for Québec, there's not much of a difference, but it would be:
Delete42 PLQ
38 PQ
34 CAQ
11 QS
So QS wins only one more seat, which goes to show that they still have a hard time convincing people outside of Montreal (7 seats would be on the island, 3 in Québec City (!) and the last one would be Sherbrooke). The PLQ would lose 3, 2 of those going to the CAQ, which is disappointing considering their rise of the phoenix in the last election. Strangely, to me anyway, the PQ stays at the same number, even though they plumetted. It should be a good sign for them that the CAQ didn't win any seats frmo their "core", so to speak. The CAQ would probably be a bit disappointed to not say more gains, but being within 8 of the lead in a three way tie is better than 13 away, so it's still a victory for them there.
I wonder if the debate effect is more profound if there are relatively new candidates versus candidates that the public has seen for a while. Also wondering if the timing of a debate (later in the campaign versus earlier) also makes a difference to potential impact.
ReplyDeleteI'd just point out that Abacus has gone back and forth on who their likely model favours. 2 polls back or so applying their likely voter model grew the Liberal lead.
ReplyDeleteI don't really see the Liberals or PCs having a significant advantage with likely voters. I think both parties have their bases motivated to vote (anti-Hudak or anti-Liberal sentiment).
DeleteOn a side note, interesting data from EKOS which claims that voters born outside of Canada support the PCs over OLP 35-31, while those born in Canada support the OLP over PC 37-30. University educated voters support the OLP by 20 points and voters with just a high school background support PCs by 7 points.
On the born-outside-Canada data from Ekos, Big Jay, note that the margin of error is 6.6 points. That quite swamps the observed difference of 3.4 points. Which matters in this case - previous day's Ekos releases all show the immigrant demographic breaking for the OLP (39-29 yesterday, 37-31 the day before, 44-25 the day before that ...). In fact, there needs to be a difference of over 13 points (over 15 for previous days) in order to suspect that the difference is anything other than noise.
DeleteThe same applies to most of the demographic breakdowns. The OLP and PCs are statistically tied amongst men, for example ...
Sorry, but it's getting hard to believe the polls. Way too much variation. Eric, can you address why there is so much variation in these polls over the past few weeks? US polls don't typically show this kind of range. Why is that? Is it that US polls use bigger samples? Don't think so. Or there are more US polls hence the variation gets smoothed out? Something seems wrong here.
ReplyDeleteFirst, Ekos is a liberal think tank and has over estimated the liberal advantage over the conservatives by over 5-8 points in the last 2 provincial elections and federal election. (So adjusting for that, it is a slim minority liberal governement) Only 3 data points, but thats a systematic error.
ReplyDeleteThe other signifncat issue is that Hudak's entire campagin is based on getting disenfranchised Conservative voters back to the polls (ie increasing turn out, among those who opted out last few elections)... so if he's successful based on Ekos poll it could easily be a conservative minority.
i.e Toss up election. Polls will tell you nothing about how successful Hudaks campagin has actually been.
The ipsos one is interesting because because it actually tries to measure this. Problem is there is no history to calibrate it, so maybe conservatives are always a prickly bunch that say nothing will stop them from getting to the polls, and liberals are always wishy washy on that question.
Regarding the Ekos performance: in the only other Ontario general election where they were a participating polling firm (2011), their last opinion poll of the election had the Liberals at 37.7%.
DeleteIn the actual election, the Liberals got 37.7% of the vote.
yep but the had the conservatives under by 4 points.... outside the margin of error. A 1 time in 20 occurrence.
DeleteSigh. Ekos is a commercial public opinion and market research company, not a think tank of any partisan persuasion. And ALL the pollsters have ways of estimating who is going to get out and vote. They all have different ways of doing so, all of which are debatable, but it's hardly just Ipsos. But, you like what Ipsos is telling you and don't like what Ekos is telling you, therefore Ipsos is credible and Ekos isn't, right? Really, that says more about you than it does about either firm.
DeleteWhy do I get the same vibe from you as I did from that Dean Chambers of "unskewed polls" fame, the guy who insisted on an industrial bias in 2012 against republicans?
DeleteDan, because that's exactly what's happening
DeleteI tracked down EKOS poll numbers for their final poll in the 2011 Ontario Election.
ReplyDeleteAll Eligible voters by EKOS in 2011:
LPO - 37.7%; PC - 31.5%; NDP - 23.3%
Likely voters by EKOS in 2011
LPO - 38.6%; PC - 34.0%; NDP - 21.1%
What actually happened in 2011:
LPO - 37.65%; PC - 35.45%; NDP - 22.74%
Considering the error associated with the sample in EKOS, they nailed it.
That was an exit poll they took in the afternoon on the 5th, and they still missed con voters by 4 points. look at their polls 5-days prior.
DeleteYes, the poll was taken on the 5th. The election was on the 6th. As said, this was their final poll. I'm not sure why I'd look at polls 5 days prior when we have a poll taken closer to the election. We have no way of telling what happened in those 5 days.
DeleteAnd, yes the poll was off by about 4 points for the PC in eligible voters. But, their likely voter model was only 1% off. Why look at eligible voter numbers and not consider a likely voter model, which again, was right on?
I looked at the ages of the responders to the EKOS tracking poll and noticed they have under 45 years at 20% of population, and 45-65 and over 65 at 40% of populations. This just does not seem right to me. Are the age splits correct for the population of Ontario? If not, the larger numbers of older voters making up the entire sample would seem to bonus the PC and suppress the NDP vote. This might explain why EKOS has consistently had the lowest NDP numbers. Eric, I would be interested in your views on this. Please comment.
ReplyDeleteNo, those numbers are not correct but they would be corrected when weighting, which would ensure the sample is not skewed towards one age group or the other.
DeleteIt does make it more possible, however, that due to small sample sizes the younger cohort will be more prone to errors, which could be amplified when the weighting is applied.
I feel that given what today is, D-Day, we all need to step back and remember that if it was not for the sacrifices made in our names 70 years ago we would not be here arguing politics.
ReplyDeleteSo possibly a breather would be a good idea for all of us.
Today is not D-Day-it is the 70th anniversary of D-Day.
DeleteThe Second World War raged for 6 years with soldiers and civilians dying or becoming injured throughout. I see little reason to elevate one day above the others. If any day should be so commemorated it is probably best saved for May 7th the day the War ended.
New EKOS polls shows PCs have moved slightly ahead. I would say the Libs need a miracle now as the post-debate bounce may become real momentum for Hudak. And the OPP investigating McGuinty is definitely not helping their cause.
ReplyDeleteI wouldn't start counting the chickens unless this is reflected over the next few days. If it is though, this could be significant, especially since EKOS has been showing a consistent and statistically significant lead over the Tories during this entire campaign. It will also be interesting to see if other pollsters show a similar swing.
DeleteKeep in mind that the EKOS likely voter numbers still favour the Liberals, but it will be interesting to see how those numbers are reflected if the PC lead in eligible voters continues into next week. Haven't seen anything from Ipsos or Abacus lately to put likely voter numbers into any sort of perspective, either.
I have trouble believing the latest EKOS numbers and their effect on the model.
ReplyDeleteIt is an article of faith with me that Liberals will not win again in a rural, agricultural riding where wind turbines are an issue, Yet the model projects a Liberal win in Perth-Wellington, a seat where McGuinty's cabinet minister was defeated last time. In contrast, the PC numbers seem to be exaggerated in Ottawa, especially Ottawa-Orleans which would be a breakthrough.
I know the margin of error is much greater for ridings, but I find both these projections unlikely.