Abacus, Outliers, And The Return Of The Conspiracy Left

On Outlier Data And Insane Theories

Sunday saw Abacus drop the first poll since the last election that has the CPC leading, which has led to some of the worst discourse in the world. It is incredibly predictable - anybody who lived through the last Parliament will know that both sides of the aisle are habitual hypocrites who choose their own adventure constantly when it comes to polling. When one’s side is up, any bad sign is dismissed as an outlier and/or irrelevant, and when one’s side is down, any positive movement is embraced with the passion of zealots. It’s a time honoured tradition, and it is both sides - Lord knows how many times Trudeau fans tried to claim things were fine and he didn’t have to do - but right now it’s Conservatives trying to defend the idea that one poll can be enough to trump the rest.

What’s interesting is I’m not entirely sure it’s a bad thing for the Liberals to get a couple of these airquotes “bad” polls.

..

There isn’t an election any time soon. The best utility polling currently has is to inform the issues Canadians care about, and give a general impression of how the government’s doing. But given how much the polls usually move as elections get close, let alone how much they just did in the run-up to April, it’s hard to see how important it is to know whether the Liberals would project for 140 seats or 190, because whatever that number is today, it’s not going to be the number in 2027 or whenever the next election ends up being. The problem with that statement is that Poilievre has a convention and a leadership review to get through.

It’s also worth saying that the conspiracy theories about Abacus Data and David Coletto are back, solely because the Liberals are down 2% in one (1) poll. The reason for it is, at least to my eye, very simple - Abacus weights to recalled past vote, which means that if the sample they got was low on recalled Conservatives, they’d have to adjust that. Their raw sample got only 31% Conservative, in a country that gave them 41%, they get up weighted to ensure a reasonable sample. Of the unweighted 1521 people who recalled voting in April, there was 46% Liberals and 31% Conservatives - an obviously unrepresentative raw sample.

Now, whether or not weighting by recalled vote is the right answer is another question - I don't think so, and in this context it’s pretty clear to me why it’s yielding friendlier numbers for the CPC. I don’t think an electorate that recalled voting for Carney by 3% is actually accurate - there’s plenty of academic evidence that people remember voting for the winner. In 2019’s UK election, wide gaps in recalled vote was used by some to claim that the polls were missing Labour voters. What actually happened was some Labour Leavers forgot they voted for Corbyn in 2017, because they remembered voting for a pro-Leave party and assumed it couldn’t have been Labour given Corbyn’s anti-Brexit rhetoric, votes, and stance after 2017. Shockingly, the polls were right.

Nate Cohn of the New York Times/Upshot tweeted “snapshots” of his polls often in 2020, with party registration and recalled vote, which they asked despite not weighting to it. Almost every time, the poll would show a recalled vote more pro-Trump than the 2016 result in that state even while Biden would have huge leads, because people remembered voting for Trump who didn’t. Had they weighted to the actual 2016 results per state, their numbers would have been worse.

The advantages of weighing to past vote are real - big, and what some might call phantom, swings caused by enthusiasm or getting a disproportionately active or disengaged partisan mix don’t happen in this method. The huge swings forwards, backwards, and sideways that some pollsters saw are smoothed out by ensuring a fairly constant sample. It’s not meritless, even if I wouldn’t do it. But a defensible choice I wouldn’t make is a very different thing than a conspiracy.

The point of these sorts of methods questions is to make them blind - to make decisions before you know what the data will say, and not wildly apply selective criteria as you deem fit day to day or week to week or poll to poll. Clearly after the election, Abacus decided that they’d continue to weight to recalled past vote, and so that’s the choice. While Mark Carney is more popular than the Liberal Party, that likely means some Conservatives who nearly voted for Carney remembering that they did vote for Carney. In reality, maybe the “right” weights would be a LPC +6-8% electorate by recalled vote, not +3%, but I pulled that fully from my ass and having pollsters adjusting based on their feelings is only going to induce bias and error.

More importantly, this very fucking weird thing where we invent a backstory and a history onto people based on their outcomes of these choices is comical, and corrosive. I’ve written before that I don’t know David Coletto’s politics and I wouldn’t even be able to guess, and I stand by it. To invent some weird conspiracy about him because you’re mad his polling didn’t show what you wanted, outside a writ and years before a likely election, means you shouldn’t be allowed around electricity.

Yes, Coletto’s been on my podcast a couple of times and yes, he’s given me a couple of pieces of data for free. If you want to think I’m shilling for him solely because of a BC column last year and a few pods, you’re free to, but you’re wrong. The reason I’m willing to defend him vigorously is the same reason I defended Angus Reid when their eponymous founder tweeted about residential schools in a column for this site and have routinely mocked the Mainstreet haters - our pollsters are simply very good. Dealing with the clusterfuck of US polling, the stability and the quality of the crew up here is worth remembering, and appreciating.

It’s also somewhere between possible and likely that Abacus/Coletto just helped the Liberals with his allegedly evil poll. Poilievre staying as Leader is good for the Liberals, and if Conservatives trick themselves into thinking everything is kosher because of Abacus, then they’re less likely to toss Poilievre. As a Liberal who dreams of getting to go against Poilievre whose shine is off and whose brand is damaged from losing 2025, this poll has done more to boost my internal sense of our chances of victory than any Leger or Nanos since the election. But that’s not the point, because all they did was report their numbers.

It’s also worth noting that Abacus were in the field from the 15th through the 19th - also known as the Long Weekend Of Mark Carney Fucking Up The Air Canada Shit. Run that poll again this week and maybe it’s back to a 2% Liberal lead. We’ll know soon enough, since it’s been mostly biweekly releases from Abacus since May. (The “mostly” is that they waited three weeks once so they didn’t release the Civic Holiday weekend. What a crime.) None of this is that important, because again, the next election will not be till 2027.

But what is important is that we don’t become what we claim to oppose. Liberals and progressives often oppose the conspiratorial nature of the right, claim to be pro-science and pro-innovation and competition, and claim not to be nasty people who make things up. We like to claim we care about facts and not feelings. It’s time we act like it. Here’s some facts. People can, in good faith, put out data that you dislike. None of these objections to Abacus’ measures were being quoted last week when we were using their poll on the Air Canada flight attendants’ demands to pressure Carney and Air Canada to do the right thing. No other pollster gave us data anywhere as useful on that key issue of public policy that hugely informed the debate. And it’s one of the great pleasures of this insane journey I’m on that I can consider David Coletto a friend.

Now stop sounding like RFK Jr when he talks about vaccines before it’s too late.