Monday, February 28, 2011

Conservatives gain in monster Angus-Reid poll

On Saturday, Angus-Reid released the details of a huge poll of 6,482 members of their online panel. This gave them the ability to delve into and report the voting intentions of every province, including Saskatchewan and Manitoba and the four Atlantic provinces. It's all great fun, and in the end Angus-Reid shows that the Conservatives have made a gain since their last poll in early January.Compared to that poll, the Conservatives have gained five points and now lead with 39%. The Liberals are down two points to 26%, while the New Democrats are up one to 18%.

The Bloc Québécois is down two points to 9% nationally, while the Greens are down two points to 6%.

Angus-Reid reports that this poll has a margin of error of +/- 1.2%, 19 times out of 20. However, according to the MRIA it is inappropriate to report a margin of error for an online poll. Other pollsters who use an online panel, like Abacus Data and Léger Marketing, always report the margin of error of a random sample of the same size as their poll, rather than claiming that margin of error for their poll. Angus-Reid should probably do the same.

In any case, the only statistically significant variation at the national level, assuming a random sample, goes to the Conservatives and the Greens. The others are probably statistical wobbles.

The Conservatives are leading among both men and women, though the margin is far smaller among the fairer sex. Among men, the Conservatives have 44% support, compared to 24% for the Liberals and 15% for the NDP. Among women, the spread is 34%-28%-21%.

In Ontario, the Conservatives are up five points to 43%, as the Liberals drop six points to 30%. The New Democrats are up three points to 19%, while the Greens are down three to 7%.

The Bloc Québécois has dropped four points to 39% in Quebec, while the Liberals are down one to 21%. The Conservatives have gained seven points, and now stand at third with 20%. The NDP is down one to 14%.

In British Columbia, the Conservatives are down two points to 40%, followed by the NDP at 28% (+3) and the Liberals at 22% (+1). The Greens are down three points to 8%.

The Conservatives have dropped five points in Alberta, but still lead with 60%. The Liberals are up 10 points to 21%, while the NDP is down three to 9%.

I'll delve into the sub-regional numbers in a bit, but regionally the Liberals are leading in Atlantic Canada with 40%, down nine points. The Conservatives are up 16 points to 39%, while the NDP is up seven to 17%. Note that in Angus-Reid's last poll the "Others" were at 16%. They are now at 1%. This accounts for most of the shift.

And in the Prairies, the Conservatives are up 11 points to 55%, followed by the NDP at 22% (-14) and the Liberals at 19% (+4).

It would appear that the only variations of real statistical significance are the gains by the Conservatives in Quebec and by the Liberals in Alberta.

Now, to the sub-regional results. As we are very rarely treated with these numbers, I'll compare them to the 2008 election results.There has been very little change in Saskatchewan, as the Conservatives stand at 56% (+2 over the 2008 result), the New Democrats at 24% (-2), and the Liberals at 16% (+1). For some reason, Angus-Reid surveyed about 1,200 people in Saskatchewan, a huge amount.

In Manitoba, there has been a little change. The Conservatives are up five points to 54% and the Liberals are up two to 21%, while the NDP has dropped four to 20%. It puts the Winnipeg North by-election win in November in some context.

New Brunswick has seen even more change. Perhaps unsurprisingly considering the Progressive Conservative win in September in the province, the Conservatives are up six points to 45%. But the Liberals are up even more, gaining seven points to stand at 39%. Oddly, the NDP has fallen precipitously, from 22% to only 11% in the province.

Meanwhile, in Nova Scotia there has also been a shift, with the Liberals picking up 11 points to lead with 41%. The Conservatives are up six to 32%, while the NDP is down eight to 21%.

In Newfoundland & Labrador, the recent Telelink poll seems to be confirmed. The Conservatives are up a big 25 points to 42%, with the Liberals dropping 11 points to 36% and the NDP dropped 13 points to 21%.

And in Prince Edward Island, there has been very little change. The Liberals lead with 51% (+3), followed by the Conservatives at 40% (+4) and the NDP at 5% (-5).

With the results of this poll only, the projection model gives the Conservatives 19 seats in British Columbia, 27 in Alberta, 22 in the Prairies, 62 in Ontario, nine in Quebec, and 10 in Atlantic Canada for a total of 150. That's a gain of 19 seats from Angus-Reid's last poll.

The Liberals win six seats in British Columbia, one in Alberta, two in the Prairies, 29 in Ontario, 14 in Quebec, and 20 in Atlantic Canada for a total of 74. That's a drop of 23.

The Bloc wins 51 seats in Quebec, down four.

And the New Democrats win 11 seats in British Columbia, none in Alberta, four in the Prairies, 15 in Ontario, one in Quebec, and two in Atlantic Canada for a total of 33, up eight.

For regular readers, note that the riding-level projection model is in place for British Columbia, Alberta, and the Prairies. Ontario is close to being completed.

Also, I will be using this poll as a model to properly distribute support in Atlantic Canada. While the Conservatives are up in each province, the bulk of their gain comes in Newfoundland & Labrador. This will need to be taken into account in the projection model.

This poll serves to confirm what the others showed as the situation in mid-February: a double-digit Tory lead. But the Atlantic Canada numbers tell us something that other polls cannot: the Liberals still have very good numbers in the Maritimes, while the Conservatives will be the front-runners in Newfoundland. That changes the game quite a bit, and paints a very different picture of the situation in Atlantic Canada than the one we expected before.


  1. Your proof reader should correct a couple of things in the text. Tory support among women is 34% not 38%. NDP support in Ontario is 19% not 13%.

    According to this poll in Ontario compared to the last election, Tory support is up 4%, NDP support is up 1% and Liberal support is down 5%. By my reckoning, if we assume a uniform province wide swing (which is all we can do at this stage), i suppose the NDP would lose Welland to the Tories, but should gain Parkdale-High Park and Beaches-East York from the Liberals - to net at 18 seats in ontario.

  2. I give the CPC 4 additional seats in NL and PEI total. so i come out with 154.

  3. DL,

    My proof reader? Do you think I have a staff?

    You're right, thanks. Corrected.

  4. I'd say the Liberals and the Consevatives will pretty much tie in NL. That's the way it has pretty much always been even in the 90s and early 00s the PC vote with the Reform or CA was pretty equal to the Liberals.

  5. "Angus-Reid reports that this poll has a margin of error of +/- 1.2%, 19 times out of 20. However, according to the MRIA it is inappropriate to report a margin of error for an online poll. Other pollsters who use an online panel, like Abacus Data and Léger Marketing, always report the margin of error of a random sample of the same size as their poll, rather than claiming that margin of error for their poll. Angus-Reid should probably do the same."

    Bravo, sir, and thank you.

  6. Its very hard to know how the "generic" vote figures in NL would translate into seats. Newfoundland is one place where people tend to be very loyal to incumbent MPs and its very rare for incumbents there to be defeated. Four Liberals in NF have been there through several elections and were winning their seat by very big margins even before Williams did his ABC campaign. Jack Harris is almost certainly unbeatable with his 75% of the vote last time and his very high profile and his backing from his old law partner Danny Williams. The Tories have a shot at winning back Avalon and a longer shot in St. John's South-Mount Pearl - but there again, the Liberal MP seems to have consolidated her support and has a high profile.

  7. DL: "Its very hard to know how the "generic" vote figures in NL would translate into seats. Newfoundland is one place where people tend to be very loyal to incumbent MPs and its very rare for incumbents there to be defeated."

    That may be true but...

    First, the "generic" vote figure probably reflects that loyalty to a fair degree. I mean, the AR polling question asks which party you'll support in your constituency. To the extent you're going to vote for the local, say, Liberal incumbent (if you're loyal to him, you presumably know who he is), despite his party, you aren't likely to answer "Conservative".

    Second, the people of Newfoundland aren't that loyal to their incumbents, since they turfed all the Tory incumbents back in '08. Those were unique circumstances, but then, so were the elections of the candidates who replaced them.

  8. Not sure to understand, pollsters shouldn't report MOE if this is an online poll.

    I understand that the formula for the standard error (and therefore for the MOE) would have to be different, but it is possible to calculate this. Maybe they couldn't have a closed form formula as simple as for the random sample, but they could always bootstrap or use other methods.

  9. Actually, people in Newfoundland only turfed one incumbent in 2008 - Fabian Manning. The other two Tory incumbents in the two St. John's seats did not run for re-election. Those seats were open.

    In ridings where you have a popular MP actually makes a HUGE difference whether people are just asked a generic party preference as opposed to being given names of local candidates etc... Let's not forget that levels of political literacy in Canada are so low that a lot of people probably think Jack Harris is a Tory because he's friends with Danny Williams!

  10. There was one CPC incumbent in NL in 2008 Carl.

    Siobhan Coady will lose her seat to the Conservatives and Avalon should be as close as a race as it was in 2008.

  11. Bryan,

    The margin of error is based on a random sample. Online panels are not random, and so industry bodies both in the USA and Canada have stated that claiming an MOE is inappropriate.

    Other Canadian pollsters who use online panels are very careful about this. It is an industry standard. Angus-Reid does good work, their track record is very good over the last few years. But they still shouldn't be reporting an MOE without the caveat that the MOE is based on a random sample.

    It is a matter of reporting. If they did like the other pollsters, and stated what the MOE would be with a random sample, then there would be no issue.

  12. DL there was a 9% drop in turnout in Newfoundland in '08 compared to only a 6% drop nationally.

    So a lot of those huge margins, when compared to the last "normal" cycle, are misleading.

    "Newfoundland is one place where people tend to be very loyal to incumbent MPs"

    Any proof of this ?

    You could say the exact same thing about Alberta.

    Except in both cases its not loyalty, its one party dominating certain areas that's confounding your claim.

    Since you have no proof that people are especially loyal in Newfoundland as opposed to other provinces then we must treat this as a non-factor when looking at national polls.

    And since there's no systematic in polling (ie. people aren't stupid like you're saying) we can know pretty darn well how these seats will translate.

    I don't see Eric having any trouble making a model for the province.

  13. Eric, I understood that, I'm not stupid. Again I know stats.

    What I'm saying is that there IS a way to calculate the standard errors for an online (non-random) panel and I don't understand why they don't do that. And as I've said, maybe they would have to use bootstrap or another statistical method. But the fact remains that it's totally possible, just more complicated. I have to adjust my standard errors for non-random issues such as time correlation, spatial correlation, etc all the time.

    At the end, reporting MOE as if it was a random sample is still only an approximation. So I was just wondering if you knew why Angus-Reid or other pollsters using online panel don't calculate the standard error properly. I'm not the only one wondering that:

  14. I don't have the answer to that. The most likely reason is that online pollsters don't want to appear different from the telephone pollsters.

  15. "Actually, people in Newfoundland only turfed one incumbent in 2008 - Fabian Manning. The other two Tory incumbents in the two St. John's seats did not run for re-election. Those seats were open"

    Fair enough, though would the two other tories have run again if they had a hope in hell of winning in 2008?

    In any ways, again, I have a hard time buying that an incumbent can be, one the one hand, hugely popular, and on the other hand, that local voters don't know which party he represents. There's a disconnect there.

  16. There is no comparison between Newfoundland and Alberta in terms of the role of personality vs. party. I would say that in about 25 out of 28 seats in Alberta - you could run a door knob as a Tory candidate and that person would get something like 80% of the vote. NL is a different story. Look at the riding of Avalon - when Liberal John Efford was the MP he was routinely piling up mammoth margins. Then as soon as he retired in 2006, the seat went Tory quite comfortably. You see a lot of that in NL - ridings will stay Liberal or Tory as long as the incumbent is there - then as soon as they retire - its anyone's game.

  17. Eric any chance of updating this post when your model is 100% complete ?

    It seems like the perfect poll to use your new system on - large sample sizes, long time in the field.

  18. One thing I think with NL is that where it is small a high profile candidate is probably better known here then say a high profile candidate in Alberta or a larger province.

  19. I think it depends on the region. It's probably the same in northern Ontario, on Cape Breton Island, or in the Gaspésie.

  20. There are also anomalies like that annoying Dr. Keith Martin in BC who for reasons I cannot fathom seems to have some sort of a personal following...but it's pretty common knowledge that without him running the Liberals will almost certainly finish a distant 3rd in Esquimalt-Juan de Fuca.

  21. DL - TO be fair, there are only a handful of ridings in Alberta where the Conservatives actually get 80% of the vote.

    But yes, a shovel could be elected MP if it ran under the CPC banner in much of the provice.

  22. Ira I think the question is whether there is anything UNIQUE about Newfoundland in terms of incumbency.

    Wascana keeps electing Ralph Goodale. Alberta kept electing landslide Annie and Kilgore. Judy W had a lock in Winnipeg North until she retired.

    Vaughan used to be Liberal.

    There's examples all across the country of a region or riding sticking with an incumbent.

    Carl is right though. That's already factored into polls.

  23. So now we have two online polls indicating a double-digit CPC lead. Interesting.

    I may have missed it, but what was the sampling period for this poll?

  24. This should put to rest the argument that the Tories were using "conspiracy theories" a couple of weeks ago when Ekos came out with the poll showing a massive lead. They said that his next poll would show a big drop in support ... and it did. Not only that, it's looking more and more like the newer Ekos poll is an outlier ... surprised nobody has brought this up yet.

  25. EKOS ended polling on February 22. Angus-Reid, despite being released after EKOS's poll, ended polling February 18.

    It's too early to declare it an outlier, as aside from a one-day Abacus poll we haven't had any other polls conducted in part after February 18.

  26. "Online panels are not random"

    Not to rehash an old argument, but phone polls are not truly random either.

  27. But they can be random - if you call both land lines and cell phones, everyone has an equal chance of taking part in the survey.

    With online polls, only those with the internet and who signed up for the panel can take part.

    Non-response rates and difficulty in reaching non-land line telephones reduces the randomness of telephone surveys, but they are based on a solid foundation.

  28. Eric,

    There are a number of problems with your claim.

    The most basic problem is that not everyone has a telephone at all -- while a small percentage, people without a phone at all do exist and that invalidates any claim to true randomness.

    "land lines and cell phones"

    The techniques for handling this issue are far from foolproof. Do you call all numbers with equal frequency? Does the pollster even have such information (including unlisted numbers, etc.) If equal frequency, what if the same individual has 2 or 3 phone numbers?

    You mention non-response, and indeed that is a huge issue. Even if one ignores the issues with which numbers to call, there is no sure way to know that those who respond to the survey are random (both answer the phone and agree to participate). One of the knocks against online panels is they attract those who are more interested in politics. No kidding! But the same issue exists with phone polls.

    Pollsters report margin of error based on the size of the response set. The response set is most definitely not random.

    "based on a solid foundation"

    I think that is a debatable claim.

    But even if we accept a modified version of that statement, it is a gross mischaracterization to simply claim that online panels are "not random" but that phone polls are.

  29. Henry,

    The number of people who do not have a phone is very small, and really not worth worrying about. Your other examples are completely valid, but you are picking nits.

    A perfectly random sample is, obviously, impossible. But some samples and methodologies approach true randomness more than others, and that is the point.

    In any case, I defer to the standards of the industry watchdogs on these issues.

  30. Eric,

    "But some samples and methodologies approach true randomness more than others"

    I would word it that some methodologies _simulate_ true randomness better than others.

    My main point was simply to point out that neither technique is truly random and using caveats for online surveys and not for response-sets from phone polls is problematic at best.

  31. New Ipsos Reid poll out has Tories at 43% ...


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.