Abacus, Byelections, And Maddening Metrics

On Two Data Points

Two things of interest have happened this week outside of the Cabinet Shuffle that’s never ending (which I have no intent on writing about, but I assume is why I’m doing the Charles Adler Show today) – the Calgary Heritage byelection was Monday, and Abacus released a poll this morning with the CPC up 10. Would I usually write about an individual poll? No, but I’m up early and I know it’ll do a lot of traffic, so why not.

In specific, the Abacus poll is a disaster for the Liberals, from a credible and sensible pollster. A six point Tory lead in Ontario and a tied Atlantic sample (and even an oversample, because thankfully Abacus understands that that region is too important to subject to tiny samples) means the Liberals would be lucky to walk out of election day with much more than 100 seats. It is a clear disaster.

The problem?  I kind of can’t believe it, because of some old research I did in my University days that I finally dusted off and posted on Monday night – byelections have been predictive in recent Parliaments, at least in terms of the decline in the government’s vote. And right now, the median byelection change in Liberal vote is a 2% gain.

So, how do you square 19% thinking the government deserved to be re-elected and a byelection profile that points to gains in both seats and votes? I have no idea.

…

When I was in school, I went through the last 10 completed Parliaments, lining up change in vote in every byelection. Take the median of all of those results for the government, and compare it to the eventual change in national vote for the incumbents. It was a check at the time as to whether the gaudy results the government was putting up in the byelections in the Trudeau Majority Parliament were meaningful or not, as someone still intending to vote Liberal (this would have been Summer 2018, I think, or 2017). And the thing is, most of the time they do predict it.

In 7 of the last 10 elections, the median change in vote at byelections has been within 2.5% of the change in vote at the next election, with three outliers – 2021, 2006, and 2004 – that all have reasonable explanations. 2021 had two byelections, one of which saw the Greens jump to the low 30s in vote share, 2006 had only 1 (and it’s Labrador, not exactly typical), and 2004 saw all the byelections happen before the change in Prime Minister and before revelations about AdScam and the Reform-PC merger. (No, the “Canadian Alliance” doesn’t exist, it’s a psyop.)

So, where does that leave us now? The Liberals have gained vote share in 3 byelections, lost it in 3, but they’ve yet to lose more than 3% in any of them. On the other side, they’ve gained 6% or more in all three seats they got a swing to them, which means that whatever way you do your average, you get a positive Liberal swing. (I found median swing does the best job predictively, working backwards.) Median swing’s +2.09% right now, a number that will be interesting to update whenever the Durham byelection ends up coming in.

It's not the world’s most robust sample – it’s only 6 byelections – but it’s the case that these results usually do have a global value, beyond specific seats in specific instances. It’s a claim that doesn’t have a lot of robust discussion in the press – by the time of the election, we all fall into the familiar rhythm of panicking or celebrating about the latest from Frank or Quito or Nik. But it’s pretty clear that right now something weird is going on that can’t be explained.

In the Harper majority, the byelections and the polls mostly said the same thing – even when the Tories had good polls, and even leads, it was almost metronomic in how consistent they would be. Never below the high 20s, never above the low 30s, which is exactly what the byelections said. Yes, the Conservatives got the post-Ottawa shooting bump, but outside of occasionally popping a few 35s in the six months after October 2014, the polls and byelections said the same things.

Even the three outliers aren’t exactly hard to reckon with – two of them had limited data, and even 2006 was still directionally correct, just a bit off on the extent, whereas 2004 is so contextually distorted as to be worth entirely dismissing – which makes this interesting. Are the Liberals in deep shit, or is Abacus a mere July oddity of a nature to be dismissed? Not that I want to be mean to David Coletto, but I’m pretty sure it was a July Abacus that had the CPC falling to 25% in 2021, pre-writ. Even if Abacus is correct now, there’s no guarantee the act of an election being called wouldn’t change things, as 2021 showed.

What’s all of it mean? I have no fucking idea. I want to saw that the poll’s wrong and that the byelections are right, or at the very least that when the chips are down, that polls can’t seem to square the theoretical prospect of voting for Poilievre with people’s actual fears about him, but it’s speculation driven by the fact I desperately want that to be the answer. I don’t want Poilievre to win both as a pundit and a citizen, and pretending that Abacus doesn’t rock my confidence at least a little bit would be dishonest, even as I intellectually know I shouldn’t be letting single polls in the doldrums move my priors.

I maintain what I said on Monday – I don’t think this is a government that’s on the way out. This feels much more like Australia 2004 than Australia 2007, mostly because Poilievre isn’t actually an electoral asset. Even so, I can’t find myself doing anything other than nodding along to much of the criticism of this government’s comms approach – as I wrote nearly six months ago at this point. The government needs to weave together their disparate efforts into a coherent and cohesive narrative, or they will risk a Poilievre government.

The weird thing is this poll may end up, in a funny way, saving the government from its worst impulses. Trudeau has always found himself strongest when he’s weakest – a fighter who weirdly seems to need his back against the wall to get his head out of his ass. The campaign he ran after blackface was a masterstroke, and has been lost to history at this point. It shouldn’t have. Whether or not Abacus is correct or not, polls have consequences.

In 2018, it was an Ipsos poll showing a huge spike in Labor support that brought the possibility of a challenge to Malcolm Turnbull’s leadership from quiet talk to active plotting. Even polls that are wrong (and I’m not saying Abacus is!) have impacts, because whether the Liberals believe this poll and internalize the message is more important than if it’s true.

At the end of the day, Abacus and the byelections point to two very different views of this government right now. It might be that Trudeau’s hated but people can’t actually vote for Poilievre, or one of them is wrong. Guesses as to which are true are less likely to be predictions and more hopes. But what we do know is simple – if the Liberals want to make sure Abacus isn’t replicated at the ballot box, they have all the power. Let’s see if they take advantage.