Monday, April 12, 2010

New IR Poll: 10-pt Conservative Lead

Ipsos-Reid has a new poll out, showing a big Tory lead but no changes outside of the margin of error of 3.1 points.Compared to Ipsos-Reid's March 16-18 poll, the Conservatives have gained three points and are at 37%, while the Liberals are down one to 27%. The New Democrats are also down, dropping three points to 15%. The Bloc Québécois is up one point nationally to 10% while the Greens are steady at 10%.

Note that this poll was taken between April 6 and April 8. While that straddles part of the Guergis affair, the real stuff happened on the 9th. We'll have to wait and see what other pollsters find this week.

In Ontario, the Conservatives are up six points to 39%, while the Liberals remain steady at 36%. Considering the low performance of the Liberals in this poll nationally, that is actually a surprisingly good result. The NDP drops five to 10%.

In Quebec, the Bloc gains seven points and is at 42%. The Liberals drop four to 23% while the Conservatives are steady at 20%. The NDP drops four to 7%, one below the Greens at 8%.

In British Columbia, the Tories gain three and have a commanding lead with 46%. The NDP drops two to 20% and the Liberals gain two and stand at 19%.

Elsewhere, the NDP and Liberals are tied in Atlantic Canada at 33% while the Conservatives have gained five points to reach 26% there. In Alberta, the Tories are down eight points but still lead with 58%. In the Prairies, the Conservatives are up 13 and lead with 59%, while the NDP is down 11 to 18%.

The Conservatives would dominate in the West, winning 80 seats (out of 95!). They also win 51 in Ontario, 8 in Quebec, and 7 in Atlantic Canada for a total of 146. Close, but no cigar. Generally speaking, everyone holding a Tory seat now is re-elected.

The Liberals win 10 seats in the West, 45 in Ontario, 14 in Quebec, and 18 in Atlantic Canada for a total of 87.

The Bloc wins 53 seats in Quebec.

The NDP win 5 seats in the West, 10 in Ontario, and 7 in Atlantic Canada for a woeful 22 seats.

This poll is a little out of step with what we've been seeing lately, which isn't new for Ipsos-Reid. But, they have sometimes been vindicated before so we'll have to wait and see what the other pollsters come up with this week.


I've placed comments on moderation. I'm generally at the computer most of the day, so there should be little delay in approving your comments. I was already receiving all of your comments in my inbox, and reading through them, so this is actually no more work for me than having no moderation at all. Believe it or not, I do get quite a bit of spam (particularly on older posts) so this is an easy way to avoid that. It also allows me to halt any of the flame wars that spring up now and then. Rejection will be used sparingly, so don't hesitate to post.


  1. Just a "little" out of step with other pollsters? Try "hugely." More accurate word, in my opinion.

    Anyways, There is two numbers I don't trust here. BC and Atlantic Canada. It's counter-intuitive to see a Conservative rise in one, and an NDP tie with another. Both areas for all other pollsters are much, much different. The Conservatives should be about 10 points lower in BC and the NDP should be lower than they are, just by virtue of Dexter's continuing decline.

    But alas, it's still an interesting poll. Even a crooked poll like this shows the Liberals making a good race out of Ontario. And the NDP collapsing like that so easily... Jack and Tom can't be too happy with this. At least not until they see the Leger poll.

  2. If, as you say, all conservatives will be re-elected, Lizzie must be devastated at that news.

  3. Shouldn't the Liberals be lower in Atlantic Canada since LIBERAL Shawn Graham of New Brunswick is vying with LIBERAL Jean Charest for the title of most hated premier in Canada??

  4. It will be interesting to see in CPC numbers also increase in other polls this week. With all the bad news surrounding Geurgis I'm surprised at the results. If true I think we will be waiting a long time before we have an election as the NDP would be decimated. It would suicidal on their part to to in any way force an election. Good news for the CPC.

  5. DL,

    Wrong - Ed Stelmach is the most hated Premier in Canada. That would be CONSERVATIVE Premier Ed Stelmach.

    And that's funny you mention, since NEW DEMOCRAT Premier Darrell Dexter isn't too far ahead of LIBERAL Opposition Leader Stephen McNeil in ratings of who should be the next Premier.

    But anyways, the NB Liberals in NB aren't so low as to wipe out fortunes in Atlantic Canada. Last poll had them around 35-40% I think. In 2008, 33% of NBers voted for the Liberals, despite Graham's popularity at the time. And all over the Atlantic, the Liberals only won 35% of the vote - not too far off from the 33% we see now.

    So these numbers probably aren't too far off for the Liberals.

  6. Looks like another example of XXXX-Reid favouring the Tories.

    Of course since the National Puke was paying for it we should expect this kind of bias.

  7. This poll was in field before the "shit hit the fan" about Guergis and it also seems to be a bit of an outlier with regards to Tory and NDP support - we have seen other polls in the last week that show quite a different story.

    I can assure you that to the extent that the NDP does or does not "push" for an election in the near future - the fact that one poll has them at 15% while others have them at 17, 18 or 20% will not be a consideration. The NDP went into the 2008 campaign with most polls showing them as low as 13 or 14% - and they got over 18%.

    Now if there was suddenly a string of polls showing the NDP in single digits like in early 1993 - that would be a different story. But there is virtually zero chance of that happening in the current environment.

  8. Volkov c'mon, your comments do you injustice.

    This poll is "crooked", the Atlantic and BC numbers rae off BUT it does prove the LPC is making a race out of Ontario.

    This isn't something you can pick and choose. To make the top line numbers work if you downgrade Atlantic/BC you'd have to increase support somewhere else, say Ontario!

    Anyways the top line numbers look bang on.

    I don't trust regionals on any poll w/o a larger sample size because of the large MOE.

  9. DL I'd expect the NDP to be lower in Atlantic Canada given the HST increase in NS.

  10. Eric

    The more I look at these polls the more predictable they become.

    Previous trends of course but slanted to provide the pollsters customer with the result they want.

    Sorry but it is becoming a real crap shoot.

    I'd bet that you can predict very closely what Ehos or HD or any other pollster is going to come out with based on knowing who commissioned the poll ?

    Remember "the customer is always right" ??

  11. Shadow,

    I'm more inclined to believe the Ontario numbers than the BC/Atlantic numbers simply by virtue of consensus. Doesn't mean I fully trust 'em, just that they seem more reasonable than anything else, except maybe the Alberta numbers.

    Besides, if you're going to talk about pick and choose, why is it I can easily find instances of you complaining about Ekos or Environics on the basis of their pollster leanings, yet with Ipsos, it's like these leanings don't even exist? That seems a tad disingenuous.

  12. I'd expect the Liberals to be lower in Ontario and BC due to the HST increases.

  13. Here is my opinion for what it is worth.

    I do not trust Darrell Bricker, and his numbers.

    They always seem a little out of step.

    I am going to wait for some other polls to come out, and see if they confirm his numbers.

    Timing on this poll seems a little fishy.

    Is the NP, trying to influence public opinion toward the tories?

    My guess is yes.

  14. I wouldn't say that this poll is out of step with the other pollsters, hugely or otherwise.
    The last three polls we've seen have the the Tories, alternatively at 37%, 34% and 32%, all with margins of error of roughly 3%.

    These polling results are all consistent with a level of support for the Tories in the population at large somewhere in the 34-35% range, for the liberals, somewhere between 26 and 30%, the NDP between 14-18%, and the Bloc and the Greens floating around 10%. None of those numbers strike me as being implausible (and, in fact, they're fairly typical of the numbers we've seen level for all five parties over the last 5 years).

    Of course, it's possible that either the HD or the IR poll is the 20th time out of 20 and bears no relationship to reality, but given that their margins of error overlap, I'm not sure I'd draw that conclusion.

  15. Why is it we cannot access the details of an ipsos-reid poll like other pollsters? I even registered and can't.

  16. Volkov i'm not being disingenuous.

    I've said repeatedly that I find pollsters to over estimate Green or "other" support at the expense of the Tories.

    Especially in somewhere like BC.

    IR and AR don't do that.

    I trust them, you trust pollsters that give your party comfort.

    It won't be hard to see who's right at the end of the next election.

  17. "Is the NP, trying to influence public opinion toward the tories?"

    NO !! You don't surely think that ??? I mean a respected media source??
    How could you possibly??

    Too bloody right IMO. The paper is so far off to the right Conrad Black comes across as a liberal !!

  18. At elast Ipsos-Reid is consistent. Let's compare the polling environment surround this poll with the one surrounding the last IR poll to show the Conservatives at 37 (Feb 22).

    In February, the IR poll was surrounded by Ekos results of 31.2 and 33.4, plus Environics and Decima numbers of 31 and 32 respectively.

    This poll is preceded by Ekos polls of 32.2 and 33.6, plus a Decima poll of 32.

    The differences are exactly the same.

    As Éric mentioned, we can't really know which polls are accurate until we see an election. History tells us that the most accurate pollsters are Angus Reid and Nanos, but of late those two have been producing wildly disparate results.

    Therefore, I suggest that Éric is correct to focus not on the numbers, but on the trends within apples-to-apples polls. If the Tories are improving, we can see it. Where they actually are remains unknowable, but that they've pulled away from their post-proroguation nadir seems likely (the Liberal decline over the same period is considerably clearer).

  19. Shadow - You shouldn't trust any of them. You should use the information they provided, yes, but choosing to trust one data source over another without any real basis for doing so... well, let's just say that's how religions get started, and we all know how well that works out.

  20. Interestingly, if you apply Eric's computation for correcting how the different polling companies' methodologies "favour" (unintentionally) one party over another, they all say more or less the same thing.

    So, HD tends to be less favourable for the Tories than the other companies (at -2%, all percentages are rounded off), Ekos is slightly unfavourable than the other companies (at -1%) and IR is more favourable for the Tories (at +3).

    Toss all those numbers into the pot and you get an adjusted HD poll result of 34% for the Tories, an adjusted Ekos poll result of 34.6% and an adjusted IR poll result of 34%. I.e., if you correct (admitedly imperfectly) for their estimated methodological differences, they're all saying the same thing (at least with regards to Tory numbers), namely that Canadians are parking their vote in much the same way they've parked their vote over the last 5 years or so.

    And I suspect we'll keep seeing this pattern until talk of an election heats up and people start thinking seriously about politics again, at which point we'll see big changes one way or another (as we did last fall).

  21. Carl, that's not really interesting at all. That's a necessary consequence of the way Éric calculates those leanings.

  22. Yeah Carl, they're all averaged against each other so they'll all end up saying the same thing when adjusted.

    People shouldn't make the mistake of thinking the average is REALITY.

    Its a relativistic arguement to say one pollster favours the Tories. Do they ? Or do the others simply disfavour them ?

  23. IRA: "Carl, that's not really interesting at all. That's a necessary consequence of the way Éric calculates those leanings."

    Which would be a devastating point if Eric only caculated the leanings based on these three polls (in which case, that would just be a mathmatical neccesity). But of course, he doesn't. He's calculated them previously based on prior polls, so if you applied them to the current batch of polls, you wouldn't neccesarily expect to have have them produce, more or less, the same numbers for the Tories on these three polls (any more than you might expect Carey Price to give up exactly 2.77goals in any given game - an unlikely result for obviou reasons).

    The point is, if you think that polling has any statistical validity, you have to believe that, if pollsters used the same methodology, the pollsters' estimates of the true level of support for the Tories should be more or less the same (i.e., they should occur in a normal distribution - a bell curve - around the true level of support in the population). Applying Eric's leaning computation seems to be a (rough and ready) way of normalizing the different poll results and controlling for the firm specific methologies of the various pollsters, since, in this case at least, it seems to produce results clustering around the low 34% (using an aggregate of the various polling methodologies of the various firms). This only highlights my earlier point that, despite the seemingly large variations in polling numbers, it doesn't seem like the pollster are really saying anything different from one another.

  24. Carl is correct. If IR had polled the Tories at 45%, it would have brought them down to 42%. That the leanings calculations puts them all at 34% is not because of how it is calculated, but because that is probably where the real number is.

  25. Shadow said...
    "Yeah Carl, they're all averaged against each other so they'll all end up saying the same thing when adjusted."

    That's just the point, these polling results aren't averaged against one another. Eric's leanings are computed using prior polling results (his latest one which I drew my numbers from used data from February), not these ones.

    "People shouldn't make the mistake of thinking the average is REALITY."

    No one's suggesting (I hope) that an an average of the poll results reflects reality (that's certainly not what I said earlier - I said that the latest round of polls had overlapping margins of error. What I'm saying, in this context, is that, if you take out firm specific methodological differences, you would expect (if the pollsters are honest and competent) for them to generate more or less the same number. After all, thati s what you would expect to see if all pollsters used the exact same methodology (since, in that case, you would expect to see the pollsters produce numbers which all cluster around the true level of support in the population at large). Eric's adjustment for leanings strikes me as a rough and tumble way of taking out those methodological differences. And, what do you know, they all produce the same results.

    It's true, I suppose that, that doesn't mean that that's the true level of support in the population (although, if I were a betting man, I wouldn't bet against that. However, if you assume, as I suppose I implicitly am, that, on average the pollsters' methodologies should produce results consistent with the level of support in the population at large (which, if you think pollster's aren't drooling idiots, hardly seems like an unlikely proposition) applying Eric's leaning adjustments to mutiple pollsters results may be a better approximation of the true population level of support.

    Put it another way, unless somewhat can tell me catagorically that one pollster's methodology is "right" and the others are "wrong", I'd stick with the assumption that, on average the pollster's methodologies are right and apply the correction from there.

  26. Ahh, yes. I was just presupposing the statistical validity of polls.

    If we don't do that, why are we even paying attention?

  27. Carl: Interestingly, if you apply Eric's computation for correcting how the different polling companies' methodologies "favour" (unintentionally) one party over another, they all say more or less the same thing...

    Thanks for delivering the best analysis of this poll in all of the comments. The message ("move on, nothing to see here") may not be exciting, but it's the right one.

    Also, a bow to Éric for his fudge factors. (Pardon me-–polling firm leanings.) The recent polls have validated them very nicely.

    The takeaway is that nobody should be excited by changes in the electorate's views, or at least not excited enough to force a snap election. It's 75 days to Pen Lake, which will quench the PM's desire for a non-confidence interpretation of the contempt motion before then; it's five or six months to the Liberals' suggested march to the polls. The Grits will only be prevented from forcing a fall election if the Dippers or Bloc totally crater before then. There's clearly no sign of that happening.

  28. Ah but Carl didn't you just do exactly what you said you wouldn't ?

    Which is to say that the leaning adjusted results ARE probably closer to reality then the unadjusted results.

    I don't think this is the case. You can break down the pollsters into 2 lean CPC (AR/IR) and 4 or 5 anti CPC.

    The assumption that the truth is in the middle seems the wrong one to make.

    Since we are dealing with different methods I think its safe to say one is "right" and the other is "wrong", as opposed to ALL of them being somewhat wrong.

    Either the anti CPC group is right or the lean CPC group is.

    Which means applying adjustments will get you CLOSER to reality with one group of pollsters but not the other, doing so will in fact take you further away from reality.

    And since they are being averaged against each other the incorrect samples are skewing things into the wrong direction. In fact the adjustments aren't correcting ENOUGH or perhaps TOO MUCH.

  29. Speaking of "polling firm leanings", does anyone have a better name for it?

  30. Éric: Speaking of "polling firm leanings", does anyone have a better name for it?

    Technically it's a bias, an error or a residual, but the first two carry too much negative baggage and the third is only meaningful to geeks. I like "leaning". If you really want to change the term, call it an "adjustment" or "normalization".

  31. Eric they use the term "house effect" over at 538 and at Pollster.

    It seems to be the quasi-official term for it.

  32. Shadow,

    What you have to keep in mind is that, because the polling firms are using different questions, the true population level of support for a given party may (and likely will) vary depending what question is asked and how it is presented. Given that it's possible for polling firms to produce numbers that are different, yet which accurately reflect the population's level of responses to their question.

    For example, Ipsos Reid asks: "Thinking of how you feel right now, if FEDERAL election were held tomorrow, which of the following parties' candidates would you, yourself, be most likely to support ?". Harris Decima, on the other hand, asks:"If the federal election were held today, and you had to make a choice, which party's candidate would you vote for? (If necessary) Which party's candidate do you lean to slightly?"( has a list of the questions asked by different polling firms).

    Notice the difference, one asks who you would likely support, the other asks who you would vote for. It's subtle, but because they're asking different questions, you would expect them to get (slightly) different results, because people may respond to the questions differently. In contrast, if each pollster asked people the same question (i.e., used the same methodology), you would expect them to all cluster around the true level of population support (i.e., how the population as a whole would answer that specific question).

    What Eric's correction does is take out the house effects for each firm. This should produce (in theory) the results that every firm would get if they use a standardize methodology (which, in this case is an amalgum of the methodologies of all the firms).

    IF you want to take the position that IR's methodology or HD's methodology was "right", you could normalize it to those methodologies. You'd probably get results which cluster around different population levels of support, but that's what you would expect if you think that the population as a whole would answer slightly different questions in slightly different ways.

  33. Carl I was thinking more about two issues that seem to be hotly debated:

    1) Robo or human interview ?

    2) To prompt or not to prompt

    I think that would have a more noticeable effect then subtle question differences.

  34. Shadow,

    Fair enough, but that just highlights the point that the different firms are asking different questions in different ways.

    What I'd be curious to see is whetehr these differences tend to disappear in the run up to an actual election as voters make up their minds as to who they're going to vote for (or at least give some thought to the issue before they get polled)so that the effect of different methodologies would tend to be diminshed.

  35. Carl - we can check that by comparing poll results during a campaign to poll results prior to that campaign, and see if they're more tightly grouped as election day approaches.

    Wikipedia has an archive of all the public poll results prior to each election (even pre-writ polls).

  36. Ira, Carl I believe someone on here a few months ago posted the results of a study that said Tory election results tend to be higher then pre-writ polling results.

    So either they're good at campaigns or there's something about pre-writ polling that fails to capture their true level of support.

    Its probably because our polling companies put ZERO effort into creating likely voter screens and older voters, who are more likely to actually show up and vote, tend to vote Conservative.

    Even the Green guy on here admits that actual Green results will be lower then the pollsters show it because young people have lower turnout numbers.

    Polling all adults to see what parties they support is a truly useless exercise.

    Not all adults vote.

    We should be finding sophisticated ways to poll VOTERS like the American companies do.

  37. I know I've made comments to that effect here in the past (though, it was based only on a review of the data posted on the Lispop website, so I wouldn't hold it out to be a particularly rigorous study).

    I agree that polling results, particularly polling numbers outside of an election campaign, shouldn't be taken as being predictions of election results. That's true in part because election campaigns themselves matter (so pre-election polling, while not utterly irrelevant, shouldn't be given all that much weight). Historically, I suspect the problem hasn't been so much that pre-election polling has overstated Liberal party support because, between election campaigns when most people aren't paying particular attention to politics, they tend to be the easy default option (in part because they've always been a suishy middle-of-the-road kinda party and in part because, historically they've usually been the government which means (a) they have much higher visibility between elections than the opposition and (b) the local MP, i.e., the only local candidate most Canadians can identify between election campaigns, is more likely to be a Liberal). Once the campaign starts, its a whole new ballgame (and, as you suggest, in recent years this trend has probably been aggravated by a collossally poorly organized an inept Liberal party facing a well-founded and, relatively, efficient Tory party).

    This might also explain the predictable collapse in Green support between pre-election polling and E-day. Outside of an election campaign, saying that you're going to vote for the Green's may be a cheap way to send a message to the powers-that-be. But come election day, voters may either decide they want to vote for a potential winner (for instance, if they're engaged in strategic voting) or they may actually look at the assortment of often nutbag policies the greens are running on (and the often equally nutbag candidates - Sorry, Greens, it's true) and decide that maybe they're not interested.

    But the thing is, if that's true, the problem isn't with the methodology used by the polling(although I take your point that measuring actual voters vs. potential voters is problematic, and that may explain some portion of anydiscrepency between, say, the last poll before the election and the electio result - certainly differences in turnout is probably why most of the pollsters except AR understated the likely Tory results, and overstated the Liberal results the last time out) but that the opinions of Canadian voters actually do change between pre-election polling and E-day.

  38. I'm surprised the Conservative party isn't blaming the Liberal sponsorship scandal for the Guergis/Jaffer spectacle.


COMMENT MODERATION POLICY - Please be respectful when commenting. If choosing to remain anonymous, please sign your comment with some sort of pseudonym to avoid confusion. Please do not use any derogatory terms for fellow commenters, parties, or politicians. Inflammatory and overly partisan comments will not be posted. PLEASE KEEP DISCUSSION ON TOPIC.

Note: Only a member of this blog may post a comment.