Thursday, November 26, 2009

New Ekos Poll: 9.8-pt Conservative Lead

EKOS has their latest poll out. It is a big one, as they promised, but it doesn't reveal all that much in terms of change.This represents a 0.3-point gain for the Conservatives from EKOS's last poll (November 4 to November 10). The Liberals, however, are also up, by 0.5 points. The Greens are up 0.2 points and the Bloc is up 0.6 points (nationally). In contrast to some other polls we've seen, it is the NDP that takes the hit here - a full 1.5 points.

The Conservatives had a big gain of 10 points in the Prairies, and are now well ahead. Their Alberta and British Columbia results are still lower than they need to be, while they are still riding high in Atlantic Canada. Their Ontario and Quebec results are close to their 2008 electoral result.

The Liberals had a gain of 3 points in Alberta, but aside from that there isn't much to be happy about. Their BC and Prairie results are alright, the Ontario one isn't horrible, but it is bad for the Liberals to be behind in Atlantic Canada and they need to do better in Quebec.

The NDP saw an 8-point loss in the Prairies, and showed weak results elsewhere. No silver lining for them in this poll.

The Bloc is up a tiny bit, but is steady. The Greens are riding high at 17% in British Columbia, though that still doesn't give them a seat in my projection.

Speaking of which, these are the results for this poll:

Conservatives - 146
Liberals - 85
Bloc Quebecois - 50
New Democrats - 27

So the Tories don't really make any gains from what they currently have, while the Liberals are up a few. The NDP is still the big loser in terms of seats.

The Conservatives would win 70 in the West and North, 55 in Ontario, 10 in Quebec, and 11 in Atlantic Canada.

The Liberals would win 15 in the West and North, 39 in Ontario, 15 in Quebec, and 16 in Atlantic Canada.

The NDP would win 10 in the West, 12 in Ontario, and 5 in Atlantic Canada.

The demographic breakdown in this poll revealed a few interesting tidbits.

For one, the Tories are way, way ahead among males - 40.9% to 25.0%. But the Liberals are running a close race among females, with the Tories leading 32.8% to 29.2%.

The Conservatives lead in all age groups except those under 25. The Liberals lead there, with 24.7%, while the Greens are second at 23.4%.

All education levels are being led by the Tories, except that the Liberals are back in front among university graduates, 33.6% to 32.8%.

Finally, in Toronto the Liberals lead 40.0% to 36.9%. The Tories lead in Vancouver, Calgary, and Ottawa.

Nothing ground-breaking here, but some very small signs of life in the Liberal numbers. The NDP numbers are back down to earth, doing away with the story line that the NDP is replacing the Liberals as an alternative. The Conservatives are still doing well, but are not in majority territory anymore.


  1. I think Ekos should experiment with a split sample and not prompting for the Green Party for half - because we know (from byelection results, general election results etc..) that their methodology tends to drastically overestimate them.

    I know you can make the argument that on ballot papers there are Green candidates, but in most ridings there are also various other independent and minor candidates and yet as far as I know, Ekos doesn't prompt "Independents or candidates of Other parties". I'd wager that if they did a shocking number of people would "press 6 for "other""

  2. That is a good point. I wonder how many of the pollsters who prompt with party names also prompt with "Other".

  3. DL wrote:

    'but in most ridings there are also various other independent and minor candidates'

    'Most'? Are you sure?

    In the last election, in all 308 ridings there were only a total of 71 independent candidates.

    Even the two biggest minor parties ran only 59 candidates each and the size of the candidate slates drops off sharply after that.

    Eric wrote:

    'I wonder how many of the pollsters who prompt with party names also prompt with "Other".'

    I answered an Angus poll earlier in the year and in that one 'Other' was one of the available selectable options. Angus polls do not show shockingly high results for 'Other'.

  4. How important is the participation and the size of the under 25 demographic?

    If you factor in how small this group, and their likelihood of voting the Liberals will discover why the NDP, Greens don't hit their numbers.

  5. Hi Eric.

    I was just wondering if you read an article from the Globe & Mail today about how the Green's could take office:

    I found it to be quite entertaining and interesting to read, not that I support the Green Party. And yes, it is related to the information presented to this newly released Ekos Poll.

  6. People's priorities change as they got older, so it isn't as if in 20 years the Greens and the Liberals will be the front-runners.

  7. To assume the Green vote is dramatically higher than it would be in an election is not a good idea to do as they are showing consistent levels support. In any case where their relative support is at is not that important because it has no really impact on the other parties.

    The problem the Greens have is that their support is from the younger voters and they are less likely to vote. The Green support simply does not vote in the election. People not voting when they expressed an opinion in a poll is a much bigger impact than any vote switching. Should the youth actually vote in a higher proportion than they have done, this would be a huge improvement for the Greens over all others.

    The pollsters make a huge error in their sampling process, they do not exclude all the people that will not vote. Ekos got an 82% voter turn out in their survey, this is much higher than would actually happen in an election, in the 2008 election turnout was 59.1%.

    This leads to the issue of the margin of error in polling due to lies - clearly a result of 82% saying they will vote and less than 60% will vote means that one in four respondents lied to the pollster. The margin of error reported for any poll is a mathematical measure, it is not a measure of actual errors in the poll. No one has calculated the error in polling do to people not telling the truth to the pollsters.

    To date the assumption has been that the degree of untruthfulness will be equal for all the parties and in all the regions of the country. There is no data to back up this assumption.

    A few other interesting results in the poll:

    More responses from people in Ottawa than in Vancouver even though Vancouver has more than half the population in BC.

    In the prairies the percentage of the Greens overall (7.7%) is lower than their support with men (8.5%) and women (9.4%)

    Support for the Bloc seems to over represented nationally. The Ekos sample is about 10% too high for Quebec but in the national result this is not reflected, working with this, I get to 8.7% for the Bloc.

    In fact the national sample size of 4744 does not seem to be modified given the fact that Quebec is over and Ontario is under.

    In the overall sample there seems to 1750 Conservative responses. Taking the regional results you get 1742.

    If you modify the Ontario results for the Conservatives to their relative weight, you get 728 as opposed to the raw number of 673.

    Quebec is over represented and that gives a modified number of 244 versus the raw one of 271.

    Alberta is 5% low and therefore 278 modified and 265 raw.

    The Prairies are 6% high - 169 mod, 180 raw.

    BC (234) and Atlantic (118) are on the mark.

    Putting this all together gives us:

    Total aggregate raw Conservative support of 1741. Modifying the results to reflect provincial population gives us 1771, a difference of about 30.

    This may not sound like a lot, but it would boost the Conservative support nationally to something around 37.3%.

    Looking at things quickly, it looks like the Conservatives and NDP are under reported nationally and the Bloc is over reported.

  8. I suddenly want to raise the voting age.

    Ekos's methodology is the thing that allows them such large sample sizes. So while I can see why they do it (basic statistics tells us that a larger sample will have higher confidence), if the methodology has bias (and we know it does), is that added confidence really worth anything?

    If we look at the Ekos polls prior to the 2008 election, we can fairly accurately predict the election results by stealing 3 points from the Greens and handing them to the Conservatives. So what does this mean? Are we becoming more confident in the results, or in the bias?

  9. Bernard, EKOS re-weights the polls. So, Ottawa didn't occupy a larger proportion of the survey than it should have. They just polled more in that region to get a useful result, but then re-weighted the result to fit into the provincial and national numbers correctly.

    We don't see all of their behind-the-scenes calculations, so we don't know exactly what they did. We can assume, however, that they did it correctly and any speculation on our part would be from a position of ignorance.

  10. How do you explain the obvious error with Green support on the prairies? How can more men AND women support the Greens than the whole population combined?

    How do you deal with the numbers I am worked out with the Conservative support and that their national numbers are not accounting for over represented areas in their results?

    I have had issues with most polling companies for years because I have been less than satisfied with their understanding of statistical mathematics. I have been even less impressed with their lack of ability to report on sources of errors in their studies as one does in every experiment.

    Back in the 1990s I looked at starting a polling company with a mathematician because of the lack of statistical rigour in the work.

    So I ask again, how do we have confidence in these numbers when I easily could show a statistically significant error in the reporting of Bloc and Conservative support? How can they claim to have been able to get an 82% response rate when we know that less than 60% will vote? Have they adjusted their numbers to reflect that 1/4 of the respondents will not be voting?

    We can not assume it was done correctly if we can find errors easily and they do not report their full results and their full methodology.

  11. We don't know less than 60% will vote. We think that will be the case, but we don't know.

    I suggest you contact EKOS directly.

  12. I meant to comment on Ottawa.

    Since we know from the data released that they are not re-weighting the provinces nationally, how do we know they are re-weighting within the regions of the provinces?

    There should have only been 162 responses from Ottawa - they had 252. If we factor in the under reporting of Ontario, there should only have been 146 responses from Ottawa. Did they reduce the value of each response in Ottawa or not? We do not know and we have some evidence that they are not doing this in other parts of their work. A high Ottawa result boosts the Conservatives in Ontario.

  13. Looking at the male/female results in the Prairies, we should get the following results overall:

    CPC - 53.0%
    NDP - 20.8%
    LPC - 17.2%
    GPC - 9.0%

    Some of those discrepancies could be in the rounding, but it seems that the Greens should be closer to 9% and the Liberals closer to 17%. Not sure why. Not really important, though.

    --- "Since we know from the data released that they are not re-weighting the provinces nationally, how do we know they are re-weighting within the regions of the provinces?"

    They told me they do both.

  14. We can be quite certain that voter turn out nationally will be roughly in the same sort of range as the last elections. An assumption of 60% is very reasonable.

    I raise this issues to simply point out that there are some very large inherent errors within the polls that are not reported within the margin of error.

    Given my past experience of raising these issues with pollsters in Canada, I have no faith that any of them have the staff on board that understand the math behind what they are doing, let alone basic underpinnings of experimental procedure.

    Given all the external sources of error other than the statistical measure of the bell curve, which has its own issues I could go about it if it applies to these sort of results or not, one can safely assume that the any numerical result is within a range or +-10% of the reported number. By this I mean the 36.9% Conservative result in the Ekos becomes 33.2% - 40.6.

    The statistical margin of error is not nearly as relevant.

    I enjoy polls because of the fun of what they might mean, but at the end of the day I have a lot less confidence in them than most people do because I understand the incredible weakness of the methodology and the assumptions behind the methodology.

  15. Bernard,

    One point to remember is that the "decided voters" may not have been re-adjusted in terms of "total sample size", but may have been re-adjusted to meet the national proportion. It is quite possible that all of the numbers, in terms of sample size, are not re-adjusted on the table but were in the calculations. And we don't know if some of the numbers are adjusted and some aren't.

    In other words, we don't have enough certainty of what means what in the poll to make our own calculations.

    --- "We can be quite certain that voter turn out nationally will be roughly in the same sort of range as the last elections. An assumption of 60% is very reasonable."

    Technically, no, we can't be certain of that. And if something big happens, turnout could skyrocket or plummet. We just don't know.

    I have the same concerns as you, which is why I use the sum total of all polls and historical results to reach my projections.

  16. You tell me by what sort of modeling can you make two smaller samples with larger numbers end up with one larger one that is smaller than both? Can you give me a mathematical example of how this could happen?

    They may tell you they are re-weighting, but their data does not show it. How do you explain the national Bloc number being 9.4% when 37.4% of Quebecers is only 8.5% of the nation?

    Show me how the math works for either one of these situations.

  17. Eric, if we can not work with a 60% turn out being a reasonable certainty, then all the polling done by anyone is even less certain.

    We have to assume turn out in the next election is a reasonable reflection of the current trends. 60% is optimistic.

  18. Bernard,

    --- "Show me how the math works for either one of these situations."

    I can't. But I'm also not convinced we have all of the information. All of the results add up to 100% in the Prairies. Could it have been as simple as typo that became part of the calculations?

    --- "We have to assume turn out in the next election is a reasonable reflection of the current trends. 60% is optimistic."

    Alright, but we don't know if the 22% of Canadians who say they are decided voters but will not vote on election will breakdown significantly differently than the 60% who will vote. Without better information, we must assume they will breakdown the same way.

  19. The fact that 1/4 of the respondents were recorded as having an opinion when we are more certain they will not be voting than the level of the support for any or the parties means we have a huge error in the results. We do not know how these people impact the results and to assume they have no measurable impact is dangerous.

    The turn out issue matters a lot if there is variation of support for parties in different categories.

    As an example all the work out there says a younger voter is less likely to vote. If your support is among the young, your support will be over reported. This means the Greens are likely over reported. The opposite is true of older voters and would indicate the Conservatives are under reported.

    They are clearly not weighting for this in their data from anything I can see.

    I suspect that much of the variation in polling numbers is mainly down to people that will not be voting in the election. Can I prove this? No.

    This is work the polling companies should be doing and are not doing.

    It also bothers me they do not publish the details of their polls so that it can be fact checked. As an example, what population numbers do they use? How do they account for differing levels of voter participation in different parts of the country? And many more questions.

    We have no proof of anything they are doing. With the lack of evidence you have to assume the worst to have the most accurate result.

    I find the polls a great source of entertainment, sort of like hockey. Neither of which really matters much.

  20. Woah woah woah.

    Hockey matters.

    But really, pre-election polls measure public support, not electoral results. Only exit polling measures electoral results.

    We have to extrapolate what we can from public opinion polls. This is about opinion, not about voting.

  21. What is you only used General Elections from 2000-2008 for a basis of statistical analysis?

  22. Eric,

    The general assumption in recent years is that high turnout benefits left wing parties and that low turnout benefits right wing parties.

    (Which is one of the reasons why I get annoyed at the media urging on higher and higher turnout - if everyone showed up we could elect the first green MP!!)

  23. Diverse opinions, public participation in democracy - those are intrinsically good things, not a "let's elect left-wing parties" conspiracy by the MSM.

  24. Eric,

    "those are intrinsically good things"

    No, not really.

    Most opinions are probably counter-productive. And votes made based on ignorance or without attachment probably tend to lead to things other than the best course for our country.

    I'd rather only sincere votes that added up to 60% of the population, than throw away votes for joke parties or whatever party is cool with the young people that added up to 100% of the population any day.

    True participation is based on being informed, not on simply the act of voting.

  25. Voting for any of the major five parties is a reasonable decision. No one is suggesting the CHP or the ML need a seat.

  26. Eric,

    I hold the Greens in the same esteem as the CHP - fringe, one issue parties that have no bussiness being discussed by serious people.

    But even if somebody votes for a serious party, isn't is something of a happy accident if their reasoning is:

    "well, my family is NDP so that's who i'm going to vote for."

    I think the rate of voter turnout actually sends an important signal, either a warning sign or a vote of confidence in the political system as a whole.

    Mandatory voting or public shaming that people vote would rob us of that important information.

  27. My experience of people that do not vote is that many of them chose not to vote because they do not know enough to make a considered decision.

    The other common reason people do not vote it that their preferred party is pissing them off and they are not about to vote for anyone else.

    Low voter turn out is not inherently bad, but it is a reality and it is not being factored into public opinion polling in any sort of a meaningful manner.

    At a minimum the pollsters should be measuring how likely people are to vote and if they voted in the previous elections. Without this data, the rest of the polling is of much lower value in showing anything useful.

  28. --- "well, my family is NDP so that's who i'm going to vote for."

    That sentiment is more likely one that benefits the two major parties.

  29. Didn't the electoral study that came out for the 2008 election show that 75% had said they had voted?

  30. Bernard,

    American polling is much more sophisticated in that regard.

    The polls will usually specify which group is being polled - Adults, Registered Voters, or Likely Voters.

    Adults are obviously the largest group but the least likely to vote. RVs smaller group but more likely to vote. LVs smallest group but most likely to be voters.

    The differences are pretty dramatic. The major news networks poll Obama's job approval amongst adults and find him at 55% approval, where as Rasmussen or Gallup use a likely voter screen and have him down at 47% approval.


    before you suggest i'm suggesting its a MSM conspiracy to use a polling method that's the most favourable to lefties i'm going to say its just a total coincidence!

  31. One of these days, someone will write a story about how one of the reasons that we are even talking about the Green party is that starting in about 2003, they decided to aggressively lobby polling companies to start reading out "Green party" when they polled people - first they got Ipsos to do it and suddenly Green went from 1% to 5 or 6% and then the others followed suit. If the CHP was smart they would spend every penny they had taking pollsters out to lunch and trying to cajole them to start to read the Christian Heritage Party.

    The Green party in Canada was essentially a party that was fabricated through a change in polling methodology that became a self-fulfilling prophecy.

  32. Maybe people do not want admit they did not vote?

    So why do people lie to pollsters or on surveys?

    This is a serious issue that calls into question much of the assumptions in public opinion polling. It is an area of sociological behaviour that needs significant study so that we can have understanding why people lie and what the implications are to the data from the liars.

  33. Maybe people don't "lie" about their voting behaviour. maybe many of the people who don't vote also refuse to take part in public opinion surveys?

  34. He refusal rate in polling is much lower than the number of people that do not vote.

    In analysing any polling data, you need to consider who the liars and what their most likely answer will be

  35. DL,

    There's probably a social desirability bias at work.

    Eric's view that voting is an intrinsic good is shared by a large segment of society and is constantly being reninforced by media, politicians, and civic figures.

    Its not out of the question that people would feel shame about not voting and tell people they did.

  36. "He refusal rate in polling is much lower than the number of people that do not vote."

    You are 100% wrong about that. In almost any public opinion poll, the response rate will typically we about 10% - in other words for every one person who answers the questions there are ten who either refuse or who never pick up the phone when their call display says "Ipsos" or "Nanos" etc... About 40% of Canadians don't vote. About 90% of people contacted to do polls do not take part.

  37. The Green vote always collapses come election day and most of that vote goes to the NDP. Iffy has a long way to go to be a threat to the Tories but the Tories need to boost their numbers in BC and Ontario to stand a chance of forming a majority. Could be tough as there are too many Canadians who think its Harper who wants the HST, its BC and Ontario provincial governments that want HST. I always vote Tory federally and Liberal in BC but Campbell is not getting my vote next time. The man is a lying crook! NDP next time!!

  38. Bernard, I sent your questions to EKOS and here is the response they sent to me. This is in response to your November 26, 12:50 comment:

    I’ve read over your reader’s concerns and there appears to be some confusion over how the data is weighted. I will do my best to explain the process.

    When we get our “raw” results, we weight them according to several factors, including region, gender, and age. For example, BC residents make up approximately 13 per cent of the population, youth (ages 18-24) make up approximately 12 per cent of the population, and males make up about half (these are not the exact numbers we use). Therefore, male youth in BC account for about 0.8 per cent of the population and, logically, their responses should count for 0.8 per cent of the overall results.

    However, since we use a stratified random sample, we are bound to oversample certain demographics and undersample others. For example, in their first statement, your reader is correct that we oversampled Ottawa. To compensate for these fluctuations, we weight the data according to their proportions in the Canadian census.

    For example, let’s use BC again. Say we sampled 1000 Canadians at random. We should logically get 130 responses from BC (13 per cent). Say, however, we get 260 responses from BC. Ignoring age and gender for the moment, respondents from BC are being counted twice as much as they should be. Even though they represent only 13 per cent of the population, they are accounting for 26 per cent of responses. To compensate for this difference, we would multiply the weight of each response from BC by 0.5. In short, if all respondents from BC stated that they would vote for Party A and the rest of Canada stated that they would vote for Party B, the results would still come out as 13 per cent for Party A and 87 per cent for Party B.

    Your reader’s second statement is difficult to explain without a strong background in statistics. In short, the greens in Sask/Man showing at 7.7 per cent, yet the male/female findings being higher is a direct result of the weighting of the relatively small sample size, in concert with the display of decimals (meaning, the rounding).

    As for your reader’s third statement, I believe they are looking at total population, rather than the population of eligible voters (we do not survey individuals under the age of 18). According to the census, Quebec accounts for about 24.8 per cent of the eligible voters; thus, 0.248 x 37.4 = 9.3, which is approximately equal to the reported figure of 9.4 per cent, save for rounding.

    In response to your reader’s fourth statement, the figure of 4744 is not weighted (or “modified”); 4744 is the actual number of decided voters who responded to the survey.

    For statement five through ten, the reader appears to be taking weighted proportions and multiplying them by the un-weighted sample sizes. For example, in the fifth statement, the reader takes the figure of 36.9 per cent and multiplies it by 4744 to get 1750 and concludes that we surveyed 1750 CPC voters. That is not correct; the figure of 36.9 +/- 1.4 per cent is the estimated proportion of the population that intends to vote for the CPC if an election were held tomorrow. All of the findings that we reported are based on weighted proportions.

    Anyways, I hope this response helps.

    Best regards,
    Jeff Smith
    EKOS Research Associates


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.