Thursday, November 11, 2010

Conservatives and Liberal below 30% in new EKOS poll

The latest poll from EKOS is a little more interesting than it usually has been, showing the Conservatives losing to the benefit of the New Democrats. But the poll has some odd regionals, as all polls seem to nowadays.The Conservatives have dropped 4.5 points since EKOS's last poll two weeks ago, and have dropped below 30% again. The Liberals haven't really taken advantage, as they are up only 0.8 points to 28.6%.

The New Democrats, however, are up 4.2 points to 19.3%, thanks to big, and perhaps improbable, gains in the Prairies, Atlantic Canada, and Ontario.

The Bloc Québécois is steady at 9.3% while the Greens are at 10.7% (down 0.9 points).

Interestingly, the Conservatives lead among men with 34% to the Liberals' 29%. But Michael Ignatieff's party has the advantage among women, with 28% to the Conservatives' 25%.

The number of undecideds in this automated telephone poll (calling both landlines and cell phones) was 13%.

The Liberals now lead in Ontario with 34.8%, down one point from two weeks ago. The Conservatives have dropped eight to 32.9%, while the NDP is up six to 19.6%. None of these numbers are particularly suspect, though the variations are a little too large. Meanwhile, the Greens are up one to 10.3%. The Liberals lead in Toronto with 35.9% and in Ottawa with 42.3%, followed by the Conservatives with 34% and 31.2%, respectively.

The Bloc leads in Quebec with 37.2% and is holding steady. The Liberals are up one to 23%, while the Conservatives are up four to 15.7%. The NDP is down one to 11%. The Bloc leads comfortably in Montreal with 41.6% to 22.6% for the Liberals.

In British Columbia, the Liberals have soared ahead and lead with 30.3%, up 11 points from two weeks ago. While that looks odd, we have seen a few other polls with the Liberals leading or performing very well. So, at the very least, we can say that there is some real potential for Liberal growth on the West Coast.

The Conservatives have dropped three points in BC to 28.2%, and are followed by the NDP at 20.6% (down six) and the Greens at 16.5% (down three). The Conservatives lead in Vancouver with 33.7% to the Liberals' 33%.

After single-handedly saving Atlantic Canada from a flood, Jack Layton now leads there with 44.5%, up 29 points. The Liberals are down six to 30.6% while the Conservatives, who caused the flood in the first place, are down 17 points to 17%.

Or, rather, probably not. This East Coast result is yet another example of why people need to look at polls with a critical eye. It's also a demonstration of why rolling averages and projection models like the one here at are a better indication of what is actually going on. I get annoyed sometimes when individual polls are treated like gospel, when they are really just another dart on the board. For example, in Atlantic Canada and with the latest polls incorporated, the Liberals are projected to take 38.4% of the vote, compared to 31.2% for the Conservatives and 22.2% for the NDP. Which seems more likely?

In Alberta, the Conservatives lead with 55.7% to the Liberals' 21% (and in Calgary with 61.6% to 26.1%), while in the Prairies the Tories are ahead with 39.4%. The NDP has jumped 17 points here and is now second with 29.3%, while the Liberals are down seven points to 19.1%.

With this poll, the Conservatives would win 38 seats in Ontario (down 17 from the last EKOS projection), 27 in Alberta, 20 in Saskatchewan and Manitoba, 18 in British Columbia (+1), six in Quebec (+3), and five in Atlantic Canada (-4) for a total of 115. That is a drop of 17 seats from the projection based on the last EKOS poll.

The Liberals would win 49 seats in Ontario (+8), 20 in Atlantic Canada, 16 in Quebec (-1), 12 in British Columbia (+5), three in the Prairies (-3), and one in Alberta for a total of 103. That is a gain of nine seats from the last projection.

The Bloc would win 52 seats in Quebec, a drop of two from last time.

The NDP smack up hard against the ceiling in Atlantic Canada, but would win 19 seats in Ontario (+9), seven in Atlantic Canada (+4), six in British Columbia (-5), five in the Prairies (+3), and one in Quebec for a total of 38. That is a gain of 11 seats.

The Greens lose the seat they were projected to win in British Columbia two weeks ago.

Also in this poll was a question about the F-35 purchase. Opinion is split, but 54% oppose the purchase of these new aircraft. Opposition is highest among Bloc Québécois (77%) and Liberal (64%) supporters, while support is highest among Conservative (70%) and New Democratic (47%) supporters.

But back to the voting intention numbers. Are the Conservatives and Liberals really both below 30%? I highly doubt it. Has the NDP surged forward in the Prairies, Ontario, and Atlantic Canada? Not likely. But what this poll does tell us is what every other poll has told us recently - Ontario and British Columbia are tight races, and a significant Tory advantage is far from in the cards.


  1. What we do seem to be seeing in these recent polls though Eric is a lot of "jitter"?

    As if the public hasn't really made up its mind? Sort of flailing around looking for something?

  2. Eric said - "After single-handedly saving Atlantic Canada from a flood, Jack Layton now leads there with 44.5%, up 29 points."

    Now that's definitely some of the best in-depth political analysis of the numbers that I've seen in a long time! :D

  3. These regionals do seem very weird. Something funky is going on with EKOS's polling. Did they just make the switch to using robopolls rather than human operators? If so that could be accounting for some of the situation. Nate Silver at Five Thirty Eight was not particularly fond of the robopolling that was conducted during the US midterms.

  4. Looking at the Ekos charts they have the numbers for Atlantic Canada fooled up. The NDP should only have the support of about 24% of respondents not 44%. The Conservatives should be about the same while the Liberals would be at 42%.

  5. RTL, possibly, but more likely it has to do with the weighting of demographic groups within the AC sample.

  6. Devon, EKOS has used this method for some time now.

  7. Apart from the two coasts, most of the regional results are pretty believable in this poll. There is nothing odd aqbout the NDP being at 19% in Ontario. They got 18% in Ontario in the last election and had 19% in the '06 election - so this is well within the range we see in other polls. The Quebec numbers are also pretty consistent with most polls we see. The other regions all have pretty small sample sizes. The numbers in Man/Sask. are not that "wonky". The NDP took 25% there in the '08 election and gaining a few points from there when you have a federal Tory government that people are getting increasingly fatigued with - is pretty predictable. Alberta is same old, same old. I would agree that the 45% for the NDp in Atlantic Canada is pretty odd unless they polled too many people in Halifax or St. John's. But BC is weird as well - I don't think anyone believes the Liberals are in first place there and that the NDP would be down from what they got in the '08 election.

  8. Ok Folks lets not forget that this is the Second poll that shows the tories @39% in Maskachewan. The potash thing seems to have had an effect. Jittery should the tories be. The good news for the libsandNDP is they are up in Ontario. West will still be a tory stronghold and Maritimes a lib one.
    Another nail in the election fever coffin tho

  9. why didn't ekos release the Oct27th - Nov 8th like they usually do?

    Is it because it doesn't agree with the second week?

    We've seen Ipsos, HD and Nanos all give the tories a substantial lead. And then ekos comes out releasing 1/2 the data they usualy do with a whole bunch of extra questions.... citing the NDP up 28 points in atlantic canada? An 8 point drop in ontario for the tories, a 17 point gain in the prairies for the NDP, and the liberals up 12 in BC?

    I think that something has happened this week where different demographics were home. (and still wondering about the week they didn't release).

    Based on the other agencies, and on the data.. this is a junk poll... 20th one out of 20.

  10. I have my doubts about this poll. It is out of sync with every other pollout there. Either Ekos is right and everyone else is wrong or well you get the drift.



  11. That's crazy, crazy regionals, man.

    However, and call me crazy as well, this has to be the third or fourth poll showing, if not a Liberal lead, still a very close race in BC. I wonder why. Is it just Vancouver starting to come back into the Liberal fold?

  12. EKOS told me there was a computing error with their first week of polling that made the results invalid.

    No conspiracy, folks.

  13. Well this weeks EKOS was officially a waste of time.

    Lets let them reboot their computers.

    See ya in two weeks folks so we can get some sane results to sink our teeth into.

  14. "No conspiracy, folks."

    ... was watching stargate sg-1 the other day.

    The old man, Daniel's "grandpa" was in a mental hospital. Daniel was out of phase, no one could see him. Except the old man who had gone through the same experience once.

    Conversation: "It doesn't matter what you say, you are just an illusion." .... "No I'm not, I am real" .... "Yes I know, all illusions say that to me"

    The lesson is. Just because you are paranoid, it doesn't mean they are not out to get you.


    Seriously tho. Did they check the second week data? Because it looks like their "computation error" extends into that data too.

    (And if it was just a computing error... doesn't any real scientist or in this case statistician know enough to keep a raw data set that is pure and unchanged as a baseline?)

  15. BC: n = 96
    MB/SK: n = 97
    Atlantic: n = 109

    ""Anything less than a 300 sample ... should be viewed with a great deal of skepticism," says the president of Strategic Communications, 24 hours newspaper's official pollsters."

  16. That's breaking news, Jonny Quest!

  17. I don't understand Eric, Johnny.

    If for example 100 people in the atlantic are mispolled and end up 50% liberal, 50% Conservvative 0% NDP.

    Would that not skew the NDP number on the national stage too?

    Even this week with the NDP up 28%. in the atlantic. The MoE is only 10% in comparison. If we assume the last week polling and no change. (what happened to change the atlantic number in the last week?) add in the margin of error. That is still probably 20% off (outside the margin of error).

    Using their weighted sample of 121.. that is 1 in 6 people mispolled. On the sample of 1587.. that is still more than 1.2% mispolled.

    Mispolling jsut that region accounts for more than 1/4 of the NDP movement.

    Simply equating them with another sample from another region doesn't make them more accurate unless the other areas are as badly polled.

    The Samples of the Atlantic, Alberta, BC, Ontario, etc are mutually exclusive. Increasing the sample size in Alberta for example is not going to give you a better reading on the Atlantic no matter how you weight it.

    The sample size from the small region has to be made made more accurate, not just accepted as fact, and added in to skew the numbers from other regions.

    (That all assumes that the 44.5% in the atlantic is infact wrong. But then.. is there anyone willing to bet that 44% is the right number? That next week will be within the 9% MoE of the current 44%?)

  18. "RTL, possibly, but more likely it has to do with the weighting of demographic groups within the AC sample."

    In their Atlantic Canadian regional breakdown they showed the overall NDP support at 42% but between mlaes and females their support was only in the low twenty percent range and in the age demographic breakdowns it was only the <25 that was in the forties and the rest were in the twenties.

  19. What happened to Oct 27 - Nov 2? Did EKOS take a week off? Normally their biweekly press release contains data for both weeks, but I don't see it.

  20. Or, if I read the discussion first before posting I get my answer.


  21. In this morning's version of the release, the NDP was at 44.5% in Atlantic Canada while at the same time being below 26% among both men and women in Atlantic Canada.

    Logical conclusion: Atlantic Canada has a very large proportion of people who are neither men nor women.

    This evening, the sex breakdown for Atlantic Canada is now 22.3% support for the NDP among men, and 55.8% among women. But, intriguingly, among Atlantic respondents under the age of 25%, the NDP is polling at 91.3%, while the Conservatives are polling at 0%.

    Logical conclusion: The NDP-voting third sex in Atlantic Canada is a very young demographic.

  22. My records show that 122 different federal polls have been released so far in 2010, and of those, 7 have shown CPC support under 30 points.

    EKOS has done six of those seven polls.

  23. I'd like to know, what seat do you think the Libs would pick up in Alberta? Being on the ground here in Alberta, I really can't see them winning a seat... even in Edmonton. The NDP are much more mobilized.

  24. Tyler, taking this poll result, the Liberals have almost doubled their level of support in Alberta.

    That has the potential to make them very competitive in Calgary Northeast, Calgary West, and Edmonton Centre, all ridings in which they had 20% or more support in 2008.

  25. "Well this weeks EKOS was officially a waste of time (Shadow)."

    OK, we get it. You don't like the results of the poll therefore you go into denial and dismiss is as a waste of time. What will your "coping mechanism" be when you get the results of a general election that you don't like?

  26. EKOS every now and then produces a poll with inconsistent results. Inconsistent with other polls conducted during the same time frame. This is one of them.

    The confidence level in this EKOS poll can only be graded as low.

    NANOS and IPSO polls during the same time showed completely different results.

    EKOS continues to use extremely small sample sizes to produce their national numbers. As a result data aberrations and inconsistent results creep in. This poll is a prime example.

  27. DL the majority of commentators seem to also have problems with this poll.

    In fact you seem to be the only one in agreement with it.

    So we get it. This poll is good for the NDP. You like the results.

    So you go into denial and ignore the problems with it. What will your "coping mechanism" do when there is an actual election and the results are nothing like this ?

  28. The sample of this EKOS poll was greater than both the most recent Ipsos and Nanos polls.

  29. Eric I think you should look at doing a Globe and Mail piece on voter preference by education level.

    It's amazing, especially in this poll, to see the large difference in voter opinion based on education level. In Alberta amongst the university educated the Liberals are actually at 34% compared to 44% for the Conservatives. Then at the high school level the Liberals are a 6%.

  30. Can I assume DL that you are dismissing the rest of the pollsters this month that predict for the most part a 5-6 point gap in favor of hte poll you do like?

    The one that has been pointed out has the NDP, 3rd sex, young voters voting in droves?

    Did you know that 0% of under 25's in Atlantic Canada prefer the Tories??

    How bout Ira's revelation about 6 of the 7 polls that show the tories under 30% in the last year are ekos? out of 122 polls?

    Shadow is right, taking this particular poll was a waste of time (at best they can claim that their computer error extended into this week, at worst it is more manipulation.)

    It would seem that Shadow is coping just fine with junk results from one pollster. Why are others waving their arms and cheering about how this is gospel in the face of polling form other pollsters that agree with each other but not this one?? What is their coping mechanism going to be when parliament returns with the same seat numbers (Again)??

  31. I'm going to postulate something and then everybody can try taking my head off.

    What all these polls are showing is the core support of the parties, not the real way it would come out in an election.

    Given that, then the undecideds do become a big factor and not the small one all the pollsters seem to make them.

    Also if seats are the issue then the only real increase in core is the Bloc !

    Let's try another:


    Ekos 30 MOE 3
    Nanos 36 MOE 3

    3+3 = 6. Do we see a pattern ?

  32. Peter the odds of two polls reaching the exact outer boundary of their confidence interval is unlikely.

    More likely this is the 20th time out of 20 (refering to the 5% chance pollsters usually give of being completely off.)

    Perhaps the F-35 question skewed this poll ?

    Perhaps their computers need another look or their weighting method got skewed or they have a new intern working for them.

    I don't like EKOS to begin with.

    But every week I check to see their % change in party support.

    Usually its reliable even if their top line numbers are not.

    This time around a 4.5% drop in CPC support just seems at odds with reality. What event happened in that time span ??

  33. I think what this poll is telling us is that we need a clean sweep of all the leaders.

    It's ridiculous that after 6 years no party has had enough support, for an long enough period of time, to win a majority government.

  34. Red Tory Liberal that doesn't make much sense.

    For 3 elections in a row now Harper and Layton have increased the seat counts for their party.

    Why on earth would CPC and NDP members put that trend at risk ?

  35. Shadow what you missed is that if you take the MOE and stick it on the same numbers you get Ekos at 30% and Nanos at 36%

    In other words they both read 33% but for reasons known only to them they add or subtract the MOE.

    In other words all these polls are LIES !!

    They are giving their clients what those clients want !!

  36. "Can I assume DL that you are dismissing the rest of the pollsters this month that predict for the most part a 5-6 point gap in favor of hte poll you do like?"

    No, I don't "dismiss" any single poll. Ekos has as good a record of accuracy as anyone else. They were the only company that gave Rob Ford a big lead in the Toronto election and they did well in the last federal election too. My only criticism of them is that they (like others who prompt) tend to inflate the Green vote a lot. So maybe the Green vote is more like 5% than the 10% that Ekos says. Most people who park their votes with the Greens are people who hate the tories and don't know who to support yet. Let's assign them 2% to the Libs, 2% to the NDP and 1% to the Tories and the Ekos poll would yield Liberals 31%, Tories 30%, NDP 21% - sounds good to me. The NDP vote can only go up even more now that the Liberals have folded over extending the war in Afghanistan - leaving the NDP as the only anti-war party in English Canada

  37. Shadow: Well this weeks EKOS was officially a waste of time.

    Not a waste of time; just somewhat off. Getting quantitative, two points too low for the Conservatives, three points too high for the NDP, not bad for the Liberals and Bloc Québécois, and half a squinch low for the Green Party. As always, that's relative to EKOS polls only, not the bistromathematical Truth. The source is the Mark I Eyeball Rolling Average Computer applied to the EKOS weekly graph on page 3.

    So where did the Dipper jump come from? Statistically, each reported level of support is within the margin of error 19 times out of 20. Acknowledging that the results for the parties are not independent and that the math is therefore somewhat involved, we can observe that we're looking at five top-line numbers each poll. The oversimplified conclusion is that, on average, we'll see one party's support outside the MOE every four polls. Extending that line of reasoning, the probability that every figure in any EKOS poll is within the MOE is... let me run the numbers... not very much.

    Sometimes a tea leaf is just a tea leaf. It would be more fun if it Portended the Future, but hey, that's life.

  38. Since Canada is not voting in a general election today - we will never know whether the Ekos poll in early Nov. '10 was "off", "average" or "dead-on". There is certainly nothing at all "odd" about having the NDP at 19% - that is inline with what the latest ARG poll says and is 2% below what the Abacus poll says and is only 1% above what the NDP got in the last election.

    I think the 19% in the Ekos poll is probably an underestimate. Research in the US shows that these "demon-dialler" robo-call polls that Rasmussen and SurveyUSA use along with Ekos tend to systematically favour Republicans. And any methodology that favours the GOP in the US probably also favours the Tories in Canada since they are essentially the same party. On top of that Ekos only has the NDP at 11% in Quebec when every large sample poll in Quebec by a respected Quebec-based pollster has them at 17-18% just bumping up NDP support in Quebec by 6% adds about 1.5% to the national numbers.

  39. If DL is right that these robo-dialer programs do indeed favour the right then all polls using them become suspect.

    Eric can you do some kind of results camparisons between pollsters on the robo-dialer and those who use a more conventional approach ??


  40. There aren't enough pollsters to make any serious comparison. The evidence would be anecdotal.

  41. WE know that robo-dialling in the US systematically overestimates GOP support by 3 or 4 points. It remains to be seen if a similar pattern exists in Canada given that Ekos is the only company we have that uses that technology.

  42. That's all well and good, but Republican does not equal Conservative, and Democrat does not equal anything in Canadian politics.

    And assuming you are right, then we have to bump EKOS's CPC findings down to 26%. That doesn't seem very plausible, and even then how do we distribute those three points?

  43. Regarding the NDP's numbers, I wonder to what extent provincial matters have affected federal ones: whether his handling of the flood has put Dexter back in favour in NS, and whether that's translating to federal NDP support, whether this poll was taken before Harper squashed potash and whether uncertainty there has helped the NDP in Saskatchewan, and even whether Howarth's relatively good numbers in Ontario are transferring to Team Layton. Doesn't explain BC's regional numbers, though if we disregarded them, the Campbell saga could also throw a few votes at the federal NDP. Shrug.

  44. DL we know nothing of the sort. Here is the conclusion of Nate Silver from 538:

    "On average, the robopoll firms have a 2-point Republican-leaning house effect, whereas the live interviewer polls have a 0.7-point Democratic-leaning house effect. The difference between the two, then, is 2.7 points."

    2 whole points !

    And that's house effects, not a measure of ACCURACY which is totally different.

    SurveyUSA, which shows the highest average lean, was also one of the MOST ACCURATE pollsters in 2010.

    Rasmussen actually shows a smaller lean but was less accurate on average because of wonky results in specific races.

    Nobody is "over estimating" anything. That would suggest a poll is overstating a party's actual support.

    That's different than a poll showing a higher level of support for a party than the average of other polls.

    The right way to think of this, therefore, is that non-robocalls UNDERESTIMATE Republican/Conservative support.

  45. Excellent piece by Susan Delacourt on the malaise in Ottawa


  46. In the US mid-term elections the final popular vote spread between the Democrats and the GOP was 53-46 - a seven point spread. Rasmussen with their "robo-dialling" was projecting a 12 to 14 point spread. That is a huge error.

    In fact Nate Silver's analysis shows that polling done by robo-calling invariably projected higher Republican votes than the results show.

    Part of the problem with polling in the US is that they try to use these "likely voter" screens that are often very crude and tend to overestimate what proportion of the electorate will be Republican because of all those cranky old men who go to tea party rallies and seem more "enthusiastic" about voting. Of course when the votes are counted the Republican vote of some "enthusiastic" tea party crackpot - has exactly the value as the Democratic vote of an UN-enthusiastic single mother in Chicago.

    Who knows what will happen next election - in 2004 and 2006 - the Tories did quite a bit worse when the votes were counted than the pre-election polls predicted. In 2008 they did a bit better.

  47. DL the firm that most over estimated GOP support was GALLUP at +15. CNN was +10

    They do live interviews.

    "Part of the problem with polling in the US is that they try to use these "likely voter" screens"

    Polls that use likely voter screens also give results for either all adults or registered voters.

    IN almost all cases the likely voter screen creates more accurate results.

    Its why Nate Silver uses LIKELY VOTERS in his model wherever possible.

  48. You are confusing two different issues. But Rasmussen (which is known for being a ridiculously biased pro-GOP polling company whose president is a teabagger) is the father of robo-calling and their polls were totally off.

    Its not as simple as just "likely voter" vs not. Its a question of how you determine who is a "likely voter" and Gallup in particular made the mistake of using too crude a measure and that's why they drastically over-estimated GOP support. Every company in the US has a different formula for calculating "likelihood of voting" and the formulas that were the least "restrictive" tended to be the most accurate.

  49. "the formulas that were the least "restrictive" tended to be the most accurate."

    Besides CNN, Rasmussen, and Gallup EVERY polling company understimated the generic congressional ballot spread.

    Go to real clear politics and check their archive.

    There's about a dozen firms who needed a MORE RESTRICTIVE likely voter screen.

    DL you appear to be making conclusions without any evidence to back yourself up.

    Survey USA is all about robopolling and Nate Silver gave them a big thumbs up!

    Rasmussen was off this cycle, decent in '08, and excellent in '04 and '06. So their method can work at times.

    House effects and accuracy are two different concepts. One's relative to other pollsters while the other is relative to reality.


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.