Tuesday, April 7, 2015

Alberta PCs favoured as election kicks off

It couldn't be avoided indefinitely. The time has finally come for the polls, and ThreeHundredEight.com, to tackle a provincial election in Alberta once again.

The new projection for the Alberta provincial election is now live, and you can see all the details (as always) by clicking on the chart at the top of the page. Before getting into the intricacies of the model and the difficulties this election poses, let's quickly take a look at the numbers.

The Progressive Conservatives under Jim Prentice lead the pack, with between 30% and 34% support. This should be enough to give the Tories between 35 and 59 seats, straddling the line of 44 seats needed to form a majority government. At 46 seats, the model narrowly favours a PC majority at this stage.

Wildrose, under newly minted leader Brian Jean, is second with between 27% and 30% of the vote and between seven and 25 seats. The maximum range does put Wildrose in a position to win, but that would require a, well, Alberta-esque miss by the polls.

The New Democrats, also under a new leader in Rachel Notley, are in third with between 18% and 21% support. This could net them between 11 and 15 seats, making the Official Opposition role well within their grasp.

The Liberals, under new/old leader David Swann, round out the table with between 15% and 17% support, enough to give them between nine and 12 seats.

Support for other parties (and more on that later) sits at between 4% and 7%, and could result in one seat for the Alberta Party.

Those are the broad strokes, and you can take a look at the Alberta projection page to see the regional and riding breakdowns.

The model itself is identical to the one being used for the upcoming federal election. The full methodology is explained here, and the thinking that has gone into the model is summarized here.

Haven't we learned our lesson?

From a polling perspective, the 2015 Alberta provincial election is going to be a rough one. It will probably be impossible to say anything about the polls without referencing or being heckled about the debacle in 2012.

So, what happened?

There are as many theories about what happened in 2012 as there are about the Kennedy assassination. Some of those theories seem to explain most of what happened, but we are still left with a question mark.

There is no doubt among pollsters that for much of the 2012 campaign, Wildrose under Danielle Smith was leading in the polls. Had an election been held at midpoint, and had voters cast their ballot as they said they would, Smith would be premier today instead of having lost her bid for the PC nomination in the riding of Highwood.

But something certainly happened in the last stage of the campaign. Changing government after 41 years is quite a big decision, and handing the reins over to the untested and relatively unknown Wildrose was, for many voters, too much. When the views of a few candidates were disseminated, a lot of the worst fears were confirmed: Wildrose was not ready, and voters swung back to the PCs. Coupled with Liberals and New Democrats voting PC to block Smith from the premier's office, as well as the well-oiled machine that was the PC Party, it was enough to give the Tories another big victory.

Why didn't the polls pick this up? For the most part, they couldn't. The election was held on April 23, a Monday. The last polls by Abacus Data and Campaign Research wrapped up on April 19, the Thursday before the vote. ThinkHQ was done on April 18, the Wednesday. Léger's polling was out of the field on April 16, a full week before the vote, and Return on Insight was done two days before that. If there was a late shift going on, these pollsters were out of the field too early to catch it.

This was particularly the case for Léger, which was doing the traditional live-caller polling. Its final poll of the campaign put the gap at just six points between the PCs and Wildrose, the narrowest margin anyone was showing at the time. If Léger had polled that final weekend, what would it have found?

Complicating matters, however, is that not every pollster was out of the field early. Angus Reid did its final polling on the Friday and Saturday before the vote, yet gave Wildrose a nine-point lead. Forum Research, also polling on the Saturday, also put the gap at nine points.

It strains credulity a little to believe that in a period of 48 hours, the electorate swung from a nine-point Wildrose lead to a 10-point PC victory. The final Forum poll done on April 22, the eve of the election, did show the gap narrowing to just two points, hinting at the surprise that was to come. But that swing should have been picked up a little earlier, as the internal polling in Alberta was reportedly picking up. And the PCs won the advanced vote by a comfortable margin as well (44% to 37%, vs. 44% to 34% among votes cast on election day).

That does leave some uncomfortable questions about what happened with the polls in those final days. Was there a bit of herding going on? Were the online panels and IVR methods being used ill-equipped to reflect the voting intentions of Albertans? Were opinions so weak that, even if the polls were right at the time, they were unable to make any real guess about what would happen in the voting booths?

These questions are impossible to answer, but they do lead us to approach this campaign (and every other one since) with a good deal of caution.

That election was the first one in which I relied heavily upon ranges, and that turned out to be a good idea. On election day, the Forum poll and these ranges led me to be one of the few commentators suggesting that Wildrose might actually fail to win. And after the election was over, the model showed that it could do pretty well with the right poll numbers, awarding the Tories 65 seats to 18 for Wildrose and five for the NDP. The actual results had been 61 for the PCs, 17 for Wildrose, five for the Liberals, and four for the NDP. The model missed on the amazing resilience of a handful of Liberal MLAs. The party's vote tanked everywhere, but those five MLAs retained far more of their vote than they should have.

So I'm confident that the model can do the job again in 2015, if the polls can do their job. Just because they missed in 2012 does not mean they will miss in 2015. There is nothing about Albertans that makes them more unpollable than other Canadians. The miss in 2012 was primarily caused by the dynamics of the 2012 race. Those dynamics have changed. If the polls miss in 2015, it will be because of the dynamics of this race, and not because pollsters were foolhardy enough to try polling Albertans again.

The 2015 model and the dynamics of the campaign

In some ways, the model should do a better job projecting seats in 2015 than it did in 2012. The data is better, as we now have results for Wildrose in all ridings throughout the province. But there are a few factors that complicate things:

A new PC electorate. The PCs in 2012 were a coalition of Red Tories and Liberals/New Democrats who disliked the idea of a Wildrose government more than they did a PC government. It seems that these fears, for the time being, have gone away. The Liberals and NDP took 20% of the vote in 2012. They are currently projected to take 35% of it, and almost all of that extra vote is coming from the PCs. The model may not be able to capture all of these very specific shifts.

Wildrose in a different light. In 2012, Wildrose was a relatively unknown party with a charismatic leader who did not seem to align with all of the politics of her party. It made for an odd combination. Were Wildrose supporters voting for Smith, the Wildrose platform, or against Alison Redford? Now, the party seems quite a bit more monolithic. But compared to Smith, Jean is an unknown. And compared to 2012, Wildrose is a party that is no longer looking like a winner, but rather a party in the midst of political post-traumatic stress. Will this shift its electorate in unexpected ways?

The rise of the NDP. The New Democrats are currently polling at a level about twice as high as their results in 2012. If this surge occurs proportionately to their 2012 results, the model will have no trouble accounting for it. If it does not, it could miss some things. The party is doing very well in Edmonton - will that lift all boats equally?

The oddity of the Liberals' 2012 results. One of the strangest things about the 2012 election was the ability of five Liberal MLAs to resist the wider shifts that almost destroyed the party. And three of those MLAs are not running again in 2015. Though the model does take into account the lack of an incumbent, it is based on a generic incumbent and not these super-incumbents that the Liberals had in 2012. That means that the ridings these three MLAs are vacating could be at play to a greater extent than the model suggests, and that the Liberals, if they take as much of the vote as the polls give them, could win some ridings in unexpected places.

There aren't too many other oddities for this campaign. The riding boundaries are the same as in 2012, and the four major parties that ran full slates in 2012 will do so again in 2015.

One potential wildcard is the Alberta Party. They aren't polling at a level that makes them much of a factor yet, but in the fall's by-elections the party did show it can pull in a big share of the vote when it puts in an effort. Leader Greg Clark is running again in that Calgary riding he almost won, and the model does think he has an outside chance of winning it this time.

Nevertheless, I have made the choice not to include the Alberta Party as a separate entity in the projection model. Many will undoubtedly question that choice. But the party only fielded 38 candidates in 2012, or 44% of a full slate, and does not seem to be planning to run a full slate in 2015. Generally, I do not have a party listed separately in the projection if it did not run a candidate in at least two-thirds of ridings in the previous election. Less than half is really pushing it.

Another requirement is that a party needs to be listed separately in the polls. Granted, the last two polls did include the Alberta Party separately, but the vast majority of polls conducted since 2012 have not. That would indicate that the vast majority of polls in this campaign will not include the Alberta Party separately, and it would be a bad idea to try to estimate this party's support going forward if most polls do not have any data for it.

If the evolution of the campaign warrants the Alberta Party being added to the projection separately, I will do so. For now, it doesn't make the cut.

But let's hope that the polls will. Another miss in Alberta would be disastrous for the industry, particularly in a province with a lot of business interests which might be keen on some market research. Hopefully that will galvanize pollsters to do a good job, and thus give voters a clear idea of what is going. I'll do my best here.