Saturday, October 20, 2007

Not Buying

Maybe inner denial, but I'm just not prepared to buy the Ipsos Reid polls of late. The new offering:
Conservatives hit the magic majority number of 40 per cent for the second consecutive week.

The Liberals trailed by 13 points at 27 per cent, a spread that pollsters attribute to weeks of unrest with Dion's leadership, party infighting and ruptures within the Quebec wing of the party. The NDP garnered 14 per cent support and the Green party had eight.

Here's where I smell an odor:
The Conservatives' lead over the Liberals in seat-rich Ontario widened to eight points from three points last week, virtually guaranteeing they would score a "solid majority" victory if an election were held today, Bricker said.

Tory support was 42 per cent, up two points from the previous survey, while the Liberals dropped three points to 34 per cent. The NDP had 13 per cent, and the Green party 10.

The only polling outfit that has the Tories ahead in Ontario, now widens the gap.

For context, here are three other polls of Ontario, released in the last week:
Strategic Counsel:

Aug. 10-12 Oct. 11-14
Liberal 40% 40%
Cons 35% 33%
NDP 17% 14%
Green 8% 14%

Environics:

Libs 38(+3)
Cons 31(-6)
NDP 18(+1)
Greens 12(+2)

Decima:

In Ontario, the Liberals led the Tories 38-33...

Just to add, the Environics poll had a larger sample size.

Conclusion, based on other polls, not to mention the last year of polls, which consistently show the Liberals firm in Ontario, I'm considering Ipsos an outlier, both on the regionals and the nationals. Next.

22 comments:

Anonymous said...

got MOE's for the ontario breakdowns? I tend to only look at the whole picture, as many regional numbers are pretty meaningless given the higher MOE.

Given they were able to replicate 40% twice now does suggest the chances of it being an outlier are pretty slim, and this is from someone who thought the first 40% was an outlier.

Steve V said...

tori

Then why hasn't one other polling outfit come close to Ipsos, both nationally and regionally? Do we ignore all the other polls? Maybe a Con does....

The MOE is smallest for Environics, with the other three comparable.

Anonymous said...

no, not suggesting that at all. certainly, if more outfits show a similar trend, it would be good. But to dismiss ipsos' replication would be like me dismissing a poll that replicated good numbers for the libs.

you said that you doubt the ipsos numbers because of the regional numbers for ontario. remember, looking regionally needs to be taken with a gain of salt as the MOE's are always bigger than what is acceptable.

Steve V said...

The MOE's are always reasonable for Ontario, for all outfits.

I don't take Ipsos seriously, because EVERY other poll doesn't "replicate".

Anonymous said...

for the latest ipsos MOE's:

bc: 9.1
AL: 10.2
sask/man:13.0
ont: 5.5
que: 6.6
altantic: 12.3

liberazzi said...

This is the second straight poll, that was announced by Con supporter Don Martin with great glee, that is complete nonsense. The only poll to believe is SES, all others are garbage. Therefore, I will start worrying when SES comes out with a similar poll. However, if you average it out then its the Cons 33/34% and Libs at 28/29%. I find it strange that Harper has such a strained relationship with the media when most of them seem to be in love with him, like Don Martin.

Anonymous said...

"I don't take Ipsos seriously, because EVERY other poll doesn't "replicate"."

replication/convergence of evidence is the hallmark of scientific study.

900ft Jesus said...

Sreve V, if 5-10% (off the top of my head) swung green in recent polls and cons stayed the same, with most leaving Libs to go green, does this have any actual significance election-wise?

Swing voters tend to be the ones looking for reasons to go one way or the other and many, when actually going to vote, will decide against a party like green because they feel their vote will be wasted since that party won't win anyway, and they would rather a Liberal than a CPC. What do ou think?

Also, and here is where I'm not familiar with how these polls come up with their numbers, if say 1,000 people are polled, do the polls take into account that 5-10% may seem like a lot if voting wasn't by district, but that can mean a lot less when calculating seats. In a strong Lib riding, that percentage won't necessarily cost them a seat.

I can see the poll being more significant if the swing happens between CPC and Lib, but not when the main change is Lib to Green. Am I missing something? Thanks

Anonymous said...

I too want to take the opportunity to chime in on the comments that the Ipsos Reid polls are off the Mark.

First, the fact that Ipsos Reid, for the past two weeks has had numbers of 40% is quite telling. That means that over the past two weeks they have had a sample of over 2,000, with nealry 800 in Ontario. Combine the two polls if you like, take an average of the two, and you've still got a solid lead for the Tories in Ontario, with a margin of error based on about 800 responses, which I think is around 3.5% percent.

Second, Angus Reid can be ignored because it is online. Not that I don't have faith in the online methodology, but Angus' numbers have been quite far off (particularly in the Ontario election). Online numbers only work for federal politics if you've got a really sophisiticated weighting scheme to account for the inherent biases of individuals who are online versus individuals who are randomly phoned. Your universe then becomes 'online' folks, and they do behave differently than the General Population. This affects not just the vote intention quetsion, but also the other commentary questions such as who makes best PM, etc. Online people tend to be more progressive, left-wing, and so it doesn't surprise me that a significant proportion say 'none of the above' when it comes to best leader. It's a protest, green thing. THe NDP and Greens will be oversampled; but since Angus has no phone numbers of his own to compare it to, how can that firm be comfortable with its weighting scheme?

Third, you mention the larger sample size being important to you in the environics poll...did you take a look at teh dates that that poll was in field? It was in field for a month! Attitudes can change quite drastically in a month. But it's not just that. They were in field during the provincial election, and asking about federal politics questions. This poses the chance for confusion and cross-contamination of the numbers from the Ontario sample. It's no wonder that they have strong Liberal numbers...that's how the provincial eletion played out.

Fourth, Ipsos Reid's numbers appear to be the freshest, in terms of field dates, reportedly coming out of field two days ago apparently.

Fifth, I get the sense that nobody knows the Ontario electorate better than Ipsos Reid...did you see how close they were in predicitng the provincial election? I think their last telephone poll had the Libs 43%, Cons 32%, NDP 17%, Green 7%...which is damn close to what the final outcome was. Not just that, but they ran an electin day online poll of thousands of respondents which basically came out with the same numbers. These people seem to know Ontario quite well, and if they weren't comfortable with the numbers that they got in their poll, they simply wouldn't have released them...at least I would hope not!

Sorry for the long rant, but it is for these reasons that I don't think you can discount the Ipsos polls as of late...

Anonymous said...

as well,

a cursory glance at the numbers from various polls suggest around a 34-35% in conservative favour, with an MOE of around 3.

MOE talks to precision. If you have a number of 35% with a MOE of 3, you are saying you are confident that the number falls within 32-38%, 19 times out of 20.

Ipsos has a 40% with an MOE of 3. Given that, the range call fall anywhere between 37-43%. Even as a "con", I would never think that the number is above 40%, but there is overlap between the polls at around the 37-38% mark.

Anonymous said...

Yes, you've got it right with the MOE. However, the MOE is the MOE of the results within the methodolgoy specified. And so, a field window of one month gives results of how Canadians felt up to a month ago, or, at some point during that field window. An MOE for an online poll is only within the greater context of a not perfect methodology...

A random phone poll, with a narrow field window, with results reported quickly, and with results weighted to reflect statscan demographic data is about as solid a methodology as you can get...then the margin of error is accurate

ottlib said...

All of the polls are saying the same thing, which is all of the parties are rather stagnant in their support.

The two Ipsos polls have estimates where the change in the estimates from one poll to the other is within the MOE, both nationally and regionally.

The other polls have the similar patterns.

That is what is so interesting about the polls from the last week-and-a-half. Canadians have parked their votes and not much seems to move them one way or another.

As for their individual results, I would not read too much into any of them. Every polling company has a sponsor that writes the cheques and those sponsors want polls that will suit their purposes. So the polling company will develop a polling methodology to get them the answers they want.

That is why polls are never to be confused with surveys.

This is true of all of the companies, including the much vaunted SES. If you want a clue on how to interpret and how much faith you should put into any poll result, look at who paid for it.

Anonymous said...

call it a gut feeling, an instinct, a view on how the har will behave in an election, arongantly, with foot in the mouth sickness, have an election and he will be defeated...but thats just me, no one likes him, just no streght seen on the other side for now...

Steve V said...

"replication/convergence of evidence is the hallmark of scientific study."

Why yes it is Tori, which is why we take the three polls with similar results, not the one that has no relationship.

BTW, those MOE's you posted are the same, if not lower, for the others. And, the national MOE's are basically identical. Ipsos is the outlier, seems pretty obvious. You have three at 5% and one with 13%, only a Tory would argue we should look at the with the wide margin as accurate.

Steve V said...

Another poll:

http://translate.google.com/translate?hl=en&u=http%3A%2F%2Fwww.cyberpresse.ca%2Farticle%2F20071020%2FCPACTUALITES%2F710200567%2F6488%2FCPACTUALITES

Libs up one in Quebec, Harper not in majority terrority.

Anonymous said...

Is it me or does it seem that when Harper makes a speech or the budget is read - polls go up for him and then in a couple of days when people mull it over - they go down again.

Maybe it's just the show and tell, hyped up State of the Union Address - whoops I mean Throne Speech.

Anonymous said...

"BTW, those MOE's you posted are the same, if not lower, for the others. And, the national MOE's are basically identical. Ipsos is the outlier, seems pretty obvious. You have three at 5% and one with 13%, only a Tory would argue we should look at the with the wide margin as accurate."

steve, I know I'm tired, but where did I say that I put stock in any result with an MOE of 13???

And can we stop with the "only a tory" and "only a con" remarks?

All I am saying is that the other polls that you seem to like have a conservative spread of 32-38% (35% MOE 3). The Ipsos has a conservative spread of 37-43% (40% MOE 3). There is some possible overlap there.

Of course, looking at the other polls, one could not view some of the questions asked, and some were not clear on methodology (telephone vs online, weighted vs not, etc), although I assume these were covered.

Tomm said...

Steve,

The polls are worthless if the parties aren't using them as guides.

If Harper dares the opposition to topple him, he is not just ignoring all of these polls. He is leading the public not letting the public lead him.

He's telling the public what to think, and it looks like some of them at any rate, are listening to him. This makes the Liberal position weaker and weaker.

I don't know how many times I have to keep saying this, but the Liberal's should have quit propping up the CPC the day after electing Dion. They are bleeding support by every day they wait. Initially they bled away some of the Jewish and hardcore Rae/Ignatieff support. Than they bled away some of the fierceness around the environmental movements support, now they have watched both the environment file and the Afghanistan file he neutralized. The Grits are having each and every tooth pulled without fighting back.

They have to force an election, even if they think they will get slaughtered. Because the longer they leave it, the worse the carnage will be.

People are getting comfortable with Harper and the direction of his government.

Tomm

ottlib said...

tomm:

Hehehe.

I always find it incredibly funny to hear people say the party they support does not care about polls.

What BS.

I can virtually guarantee you that the Conservatives began doing nightly tracking polls just after Labour Day. The results they received daily were part of the decision making on their approach to the Throne Speech.

The Liberals have been doing the same for the same period of time. Again, it goes into their decision making.

Politicians of all stripes routinely use polls as part of the decision making process and to think otherwise is naive.

Steve V said...

"steve, I know I'm tired, but where did I say that I put stock in any result with an MOE of 13???"

Sorry, I meant the 13% spread in the Ipsos poll between the two parties :)

Anonymous said...

Hmm,

in terms of who pays for the polls, I don't think anybody pays for their political polls. Their release only says "for CanWest News Service and Global Television", but I wouldn't think that would be biased in any way.

Also, I heard that Ipsos Reid is the only firm who doesn't do political party polling...

Steve V said...

"Their release only says "for CanWest News Service and Global Television", but I wouldn't think that would be biased in any way."

Nope, no bias there ;)