Friday, February 27, 2015

Close race also seen by Ipsos Reid

The polls have spoken. The Ipsos Reid survey published Wednesday afternoon by Global News put the gap between the Liberals and Conservatives at just one point, exactly the same margin as recorded by Abacus Data and EKOS Research. Even the support levels are very uniform: 32% to 35% for the Tories and 32% to 34% for the Liberals. For all intents and purposes, the national landscape is now a tie.

Ipsos was last in the field between Jan. 6 and 11. Since that poll, the Liberals picked up three points to move into the lead with 34%, while the Conservatives were down two points to 32%.

The New Democrats slipped one point to 23%, the Bloc Québécois was steady at 6%, and support for other parties (including the Greens, which Ipsos does not prompt for) was down one point to 3%.

Total undecideds numbered 13%, down four points.

None of these shifts were outside the margin of error of similarly sized probabilistic samples.

Ipsos's last poll was hinting at a Conservative surge. No other survey at the time was suggesting quite the same thing, so it appears that it was just the product of normal sampling error and this new poll is a reversion to the mean. Apart from the sharp decline in Liberal fortunes since the fall, there is no discernible trend in Ipsos's numbers over the last few months.

Indeed, the 'surge' may have been the product of an unusual result among women. That January poll put the Conservatives ahead among this demographic, something that no other poll was showing at the time or has shown since. But this new poll shows a reset of those numbers, with the Conservatives and Liberals swapping six points. The Liberals now lead again among women and the Conservatives among men, putting Ipsos back in agreement with every other poll.

Oddly, Ipsos showed a big uptick in support among the youngest Canadians for the Conservatives, putting them up seven points to 29%, just five points behind the Liberals and one point ahead of the NDP (which was down eight points). This might seem like a counter-intuitive result, but in fact the recent Abacus and EKOS polls have also shown improving numbers for the Tories among young voters. Why this might be, however, is beyond me, as is whether this is a real trend or an anomalous blip.

The regional results are fairly typical, with some small variations.

The Conservatives led in British Columbia with 39%, putting them ahead of the Liberals, who were at 34%. The six most recent polls had put the Liberals ahead, so this is a bit of a reversal. But that other polls were in such agreement in B.C. was, itself, unusual. Ipsos has recorded a drop in NDP support in the province, to 21%, as others have.

The Conservatives led in Alberta and the Prairies with 54% and 45%, respectively, but Ipsos puts the NDP in much better form in the Prairies than other polls at 27%, one point up on the Liberals.

The Liberals and Conservatives were in a tie in Ontario at 37%, which is almost identical to the current polling aggregate. The NDP at 23%, however, is a much better result than the party has seen in most polls in some time.

And Atlantic Canada is the usual, with the Liberals way ahead of the NDP and Tories with 47% to 26% and 24%, respectively.

But what about Quebec? It has been the wildcard in federal polling lately, and this survey is no exception.

As most other polls have it, the Liberals were ahead with 31% to 27% for the NDP. This represented a sizable, but just inside the margin of error, increase for the Liberals. The Bloc Québécois was in third with 26%, while the Conservatives were well behind at 15%.

We've discussed the standing of the Conservatives in Quebec quite a bit, and this Ipsos poll would seem to argue in favour of their gains having been wildly over-stated. I think there is something to that. But if we look at the trends from each of the pollsters that have been active so far in 2015, there is a pattern that emerges. Virtually all of them have had the Conservatives gaining in January and early February, compared to where they were in December. But now the polls seem to be showing those gains tailing back or plateauing. Ipsos is well in line with that, with a three point drop.

The debate, then, is more about the size of the gain and the amount of support the Conservatives really have. The trends seem clear enough, however.

Another notable result here is the 26% for the Bloc Québécois. Coupled with the EKOS poll that put the party in second place with 23% last week, some were saying that the new Quebec narrative should be about the Bloc's rebound. I think that is premature.

This Bloc result is the best the party has had in any poll in a year. But that it comes from Ipsos is significant. The Bloc is only up one point from Ipsos's early January poll, so the number is far from unusual.

In fact, Ipsos has tended to have better results for the Bloc than other pollsters lately. Polls conducted within a week of this most recent survey have averaged 18% for the Bloc, compared to the 26% here. In January, polls taken within a week of Ipsos's 25% result averaged 17%, while polls taken within a week of Ipsos's November poll, which put the Bloc at 21%, averaged 15%. The Bloc had 20% in the Ipsos poll from September, while other polls at the time were averaging 14%.

So, it would appear in this case that we should not get ahead of ourselves in predicting a comeback for the Bloc. As far as Ipsos is concerned, the party is in a better position now than it was in the summer and fall, but is otherwise holding steady. This is the consensus among most other polls as well. Quebec is a confusing place for federal polls, so let's focus on points of agreement when they exist.

Wednesday, February 25, 2015

EKOS, Abacus polls show close race

With little over six months to go before the official start of the 2015 federal election campaign, the latest polls from EKOS Research and Abacus Data show the Conservatives and Liberals in a very close race.

The current projection reflects this. The previous update, done on Feb. 17 and incorporating polls in the field to Feb. 10, had the Liberals narrowly ahead with 34.2% support to 32.9% for the Conservatives. Now, that gap of 1.3 points has reduced further to just 0.5 points, with the Liberals at 33.5% and the Tories at 33%.

That, in turn, has widened the margin in favour of the Conservatives in the seat projection: 145 to 126, against 140 to 135 in the last update. The ranges have also become more favourable to the Tories, with 126-164 seats against 107-145 for the Liberals.

The NDP stands at 19.4%, with a range of 47 to 77 seats.

I go over the latest projection numbers and the regional breakdown in my CBC article this week. Please check it out.

Let's look at the two national polls added to the projection (CROP was also added for the Quebec numbers), by EKOS Research for iPolitics and Abacus Data.

Abacus was last in the field Jan. 26-28. Compared to that poll, both the Conservatives and the Liberals picked up two points to reach 35% and 34%, respectively. The NDP was down three points to 21%.

None of these shifts were outside the margin of error of similarly sized probabilistic samples. Indeed, Abacus's numbers have been wobbling back and forth for some time, with no discernible trend.

EKOS, which appears to be reporting on a weekly basis now, was last in the field Feb. 4-10. There was little change since that poll, with the Conservatives up 0.3 points to 32.3%, the Liberals down 2.2 points to 31.6%, and the NDP up 0.2 points to 19.1%. None of these shifts were outside the margin of error.

Both polls showed similar gender breakdowns, with the Tories up on the Liberals by a margin of 40% to 35% among men in the Abacus poll and 36.7% to 30.2% in the EKOS survey. Among women, the Liberals were ahead with 34% to 30% according to Abacus, and 33.1% to 28.4% according to EKOS. Both polls also showed the NDP doing better among women (25% Abacus, 20.9% EKOS) than men (17% Abacus, 17.4% EKOS).

The two polls were also in lock step in age breakdowns, with the Conservatives narrowly ahead among the youngest cohort (unusual as that is), the Liberals ahead among those aged roughly 30 to 65, and the Conservatives in front among older Canadians.

Regionally, there were few major differences in the two polls in most parts of the country. The Liberals narrowly ahead in B.C., where the NDP is faltering, the Conservatives well in front in Alberta, the Liberals with majority support in Atlantic Canada, and the Conservatives ahead in Ontario (considering the sample sizes, the differences between the Abacus and EKOS polls there are really marginal).

The two polls were in disagreement in the Prairies, where Abacus has the Liberals with 41% to 38% for the Tories and 15% for the NDP. But the sample size, at 95, would carry a margin of error of +/- 10 points with a probabilistic sample. EKOS has more familiar looking numbers, with about 44% for the Tories, 31% for the Liberals, and 15% for the NDP.

The other point of disagreement is Quebec, which has become par for the course in recent polls. It does seem to be a methodological difference, since the online polls are all in agreement and the IVR polls are generally on the same page.

Abacus put the Liberals ahead with 32% to 31% for the NDP, the Conservatives at 18% and the Bloc Québécois at 17%. EKOS, meanwhile, has something very different: 23.6% for the NDP, 23.1% for the Bloc, 22.7% for the Liberals, and 22% for the Conservatives.

That represents some exceptional results. For the Liberals, it would be their worst result in any poll since March 2013, or 124 polls ago. For the Bloc, only two of the last 53 polls have been better.

It is possible that EKOS is on to something, but it looks more likely that this is a bit of an outlier result, which is bound to happen from time to time.

The most recent polls from the province have some points of agreement, but also paint a very confusing portrait of what is going on in Quebec. Here are the latest results (all polls taken between Feb. 9 and 17) with their respective error ranges (assuming probabilistic samples in the case of Abacus and CROP).

For the Liberals and Conservatives, the four polls do not all overlap. Or at least, the three by Abacus, CROP, and Forum do, whereas the EKOS poll does not.

There is only a little bit of overlap with the NDP numbers (at 27%) and the Bloc (at 20%).

With these numbers, we can posit that EKOS was probably a little low for the Liberals and high for the Bloc, and that the Liberals' true support probably lies closer to the high-20s or low-30s in the province. The New Democrats seem to be in the high-20s.

The Conservatives are likely in the high-teens or low-20s, where as the Bloc is most likely in the high-teens.

The province is certainly in flux, and undoubtedly the smaller amount of interest in the federal scene has an influence on how fluid voters are in the province. But with the race becoming a four-way contest, an extremely wide range of outcomes is possible.

If we use these ranges to calculate the best case scenarios for each party, we get an enormous variety of results. The Liberal best case scenario with these ranges gives them 51 seats (14 NDP, 13 CPC, 0 BQ). The NDP could take 54 (16 LPC, 8 CPC, 0 BQ). The Conservatives could win 20 (31 NDP, 24 LPC, 3 BQ). Even the Bloc, topping out at 27%, could win 40 seats (22 LPC, 10 CPC, 6 NDP) if the other parties divvy up the remaining vote.

Consider how Quebec alone could cause a huge swing. If we use these seat ranges and apply them to the current projection, assuming all the other provinces go as projected, the Conservatives would be at 137-149, the Liberals between 118-153, and the NDP between 33-81. The stakes in Quebec are laid out quite clearly here: probably not too significant for the Tories, but the province decides whether the Liberals win or place second, and whether the NDP stays as a strong party or returns to pre-2011 days.

The polls did a good job calling the federal election in Quebec in 2011, better than anywhere else in the country. It might not prove so simple in 2015.

Friday, February 20, 2015

So are the Conservatives gaining in Quebec or not?

The poll that many of us were waiting for finally arrived yesterday, as La Presse published the latest numbers from Quebec-based pollster CROP. The poll would settle once and for all the question we've been asking for weeks: are the Conservatives really making inroads in Quebec?

The answer the poll provided was: maybe, but certainly not to the same extent as we've seen in other surveys. But that is a boring answer. The answer a lot of people saw in this instead was: "welp, so much for that idea."

I think that is a very simplistic way to look at the results of this poll. I'll get into that in more detail below, but first let's just take a look at the overall numbers.

CROP was last in the field between Dec. 10-15. They have not recorded any statistically significant shifts in support since then (at least, if these samples were probabilistic).

The Liberals dropped four points to 33% in Quebec, while the New Democrats were unchanged at 30%.

The Bloc Québécois was also steady, at 17%. The Conservatives picked up three points to hit 16%, and the Greens were unchanged at 4%.

The Liberals have been wobbling back and forth in CROP's polling for some time, as shown in the last six polls from the company: 38%, 34%, 37%, 32%, 37%, 33%. This recent drop would seem to fit into that oscillation.

But what about the Conservatives?

The increase of three points is within the margin of error (or would be, of similarly sized probabilistic samples), so it could just be a statistical fluke. I think, however, that with the gains we have seen in other polls it stands to reason that it isn't a statistical fluke. In CROP's previous 10 polls, for instance, the Conservatives averaged just 13%.

And the Tories made gains throughout the province. They were up among francophones and non-francophones, made a significant increase on the island of Montreal, and inched upwards in Quebec City and the regions of Quebec. Only in the 'couronne', where they fell by a single point, did the party take a step backwards. And 15% of respondents picked Stephen Harper as the preferred person to be prime minister, his best result in a CROP poll since at least June 2013.

Nevertheless, that 16% is well below the current aggregate of 21%, and the 23% to 26% we've seen in the last two polls by EKOS. That would seem to settle it, then, yes? Much ado about nothing in Quebec. A large sample poll from a Quebec-based company showing a marginal increase is a very strong argument against Conservative gains in the province.

But I don't think we should stop there. A lot of people have accepted this CROP poll as the be-all-and-end-all of polling Quebec. I think that is a bit much. CROP is a good pollster. But they are far from the decisive and conclusive voice.

Consider that CROP hasn't been tested all that much in recent elections. In the 2011 federal election, the final CROP poll exited the field on April 20, almost two weeks before Election Day. In the 2014 provincial election, CROP was out of the field on March 16, more than three weeks before Election Day. In recent contests, only in 2012 did CROP put out a late campaign poll. But it was one conducted over the telephone, which tells us little about the accuracy of their online panel.

Much is also made of the large sample size. In this poll, CROP gathered the opinions of 884 decided voters. A standard national poll would have less than 250 responses from the province.

But the last three Forum and EKOS polls that have shown the Conservative gains in Quebec sampled a total of 1,037 decided voters. So sample size is not the issue. The reliability of the sample is another question entirely, however. Whether it be some 900 panelists or some 1,000 Quebecers willing to take an automated telephone survey, if the sample itself is of poor quality it doesn't matter how many people are interviewed.

And this brings up another interesting question. Is there a methodological difference in the results? Since the beginning of 2015, the Conservatives have averaged 23% support in Quebec in six IVR polls. In four online polls, they have averaged just 17%. Could this be a sort of 'shy Tory' effect in Quebec, an issue Joël-Denis Bellavance mentioned last week on CTV's PowerPlay? Is there a reason that online panels would be influenced by that while IVR polls would not be? In 2012, the closest thing to that would have been a 'shy Charest' effect, and the online polls did post lower Liberal numbers than Forum's IVR surveys.

All of this is not to say that CROP's results should be discounted - far from it. But it should make you ponder whether one poll really has such a monopoly on the truth. Instead, each poll adds to what we know, and CROP gives us a very good piece of information. It tells us Forum and EKOS have probably been too high for the Tories. But CROP might still be too low. Methodological influences and house effects can be very important.

Take, for example, the last time CROP was in the field in December. At around the same time, six other polls had been conducted. Look at the differences between the consensus of those six polls (the average) and CROP's findings:

There was little real difference that couldn't be explained by normal sampling error for the New Democrats, Bloc Québécois, and Greens. CROP was high on the Liberals, however, and low on the Conservatives.

Does this mean the exact same thing could be happening with this latest poll? Not necessarily. And does it mean that CROP is wrong while the others are right? Again, no. What it does mean is that each pollster has what is called a 'house effect', a methodological bias caused by any number of sources: mode of contact, the people sampled, the questions asked, the weightings applied, etc.

This is the benefit of using an aggregate of polls. It can iron out these differences, and get us closer to what might be the truth. CROP's polls are very valuable for their large samples, regional breakdowns, and local knowledge. But that doesn't mean other polls are clueless - in fact, the non-Quebec-based pollsters did very well in Quebec in 2011.

The regional breakdown

Let's get back to the poll itself.

The New Democrats held on to the lead among francophones with 32%, narrowly ahead of the Liberals at 29%. The Bloc was at 20%, while the Conservatives were at 15%. Note, though, that for the NDP they are polling quite a bit lower than the 35% to 39% recorded by CROP in polls done between June and November of last year.

Among non-francophones, the Liberals had a 13-point drop to 49%, with the NDP steady at 19% and the Conservatives at 18%. The Bloc made a big jump of 10 points to 10%, suggesting that in this case we may be looking at a statistical anomaly since it seems unlikely that about 1 in 6 non-francophones surveyed in December decided to de-camp from the Liberals and head to the Bloc.

The Liberals led in and around Montreal, with 34% on the island and 44% around it. The New Democrats were not far behind in Montreal with 31%, though they dropped eight points to 24% in the 'couronne'. The Conservatives placed third on the island of Montreal with a jump of 11 points to 18%, but were in fourth behind the Bloc (20%) with 9% in the suburbs.

The Conservatives held the lead in Quebec City, though, with 38% support. That was virtually unchanged from December, and the Tories have not polled so highly here in two consecutive CROP surveys since the beginning of 2012. The NDP was at 33%, while the Liberals were down to 16% support.

In the rest of the province, the NDP was narrowly ahead with 31% as the Liberals dropped to 30%. The Bloc was steady at 20%, while the Conservatives were at 14%.

On who would make the best prime minister, Thomas Mulcair was in front with 25%. Justin Trudeau experienced a big tumble, dropping five points to 23%. As mentioned, Harper was up to 15%, a three-point increase.

If we ignored what other surveys have been saying, how would we look at this poll? For the most part, we'd consider it par for the course. These are the sorts of numbers CROP has been putting out for months. A close race between the NDP and Liberals, but with the NDP ahead among francophones. The Conservatives up slightly, which we'd consider just a wobble, but good numbers in Quebec City. The Bloc continuing to flounder.

Overall, we'd probably consider it a decent poll for all three federalist parties. The Liberals still lead, and look well-positioned for gains in and around Montreal (the drop they experienced on the island was due to those odd results among non-francophones, which are sure to be reset with the next poll). The NDP still leads among francophones, and so should retain the bulk of their seats. The Conservatives could make big gains in Quebec City.

But in light of other polls, we can look at this in two ways. The first is to consider that the Conservatives may not be making the gains other polls have suggested they are, and that their hopes need to be tempered. The second is to see in this the same trends that other polls have recorded, and that Quebec will indeed be a battleground for all three national parties. As usual, time will tell.