To the surprise of nobody, Newspoll number 30 tolled for Malcolm Turnbull overnight.
By the logic Turnbull employed as he announced his leadership challenge in September 2015, it can only be concluded that the “people have made up their mind” about his prime ministership.
There are actually numerous other points of polling evidence to suggest his position is not quite as bad as that, including the weekend Ipsos poll for the Fairfax papers, which found 62% supporting him remaining as Liberal leader, and only 28% wanting him gone.
Nonetheless, Newspoll’s two-party measure was the petard on which he chose to hoist himself, and there he hangs today.
As The Australian‘s Jack the Insider put it last week, Turnbull no doubt chose his words in full confidence that, once ensconced in The Lodge, “the old Turnbull charm would rub off on the Australian people and, enamoured by that cheeky grin, we’d pump out the love for his leadership”. But no matter how healthy Turnbull’s self-regard, he had been around the block long enough to know that rough spells in the polls were a fact of political life even for the most popular of leaders, having been on the wrong end of a few himself (in his ill-fated spell as opposition leader in 2008 and 2009).
One reason he might never have envisioned this translating into a losing streak to equal Tony Abbott’s is that the signals transmitted by polls are, by their nature, distorted by statistical noise. What Turnbull had no cause to anticipate was how muted this noise would become in the poll series he identified as the only one that mattered.
In drawing upon roughly 1600 respondents to represent a nation of over 12 million voters, Newspoll should be subject to the vagaries of sampling error, causing its polls to spray around at least 2.5% on either side of the “true” position.
The way this should look is illustrated in the histogram shown below (left), which is based on a large number of randomly generated hypothetical poll results.
These assume a permanent underlying Labor lead of 53-47 — which, implausible as it may seem, is remarkably close to the true situation since the start of last year, according to an aggregated poll trend measure.
On this basis, Turnbull should have had at least some chance of lucking out with a 50-50 or better at some point since his horror run began in September 2016 (at the considerable cost of suffering the odd 56-44 blowout on the downside).
However, the histogram on the right, which shows the actual spread of Turnbull’s 30 defeats, finds Newspoll behaving as if its margin of error was around 1.5%, rather than 2.5%.
All but two of Labor’s 30 successive Newspoll wins have been in the range of 52-48 to 54-46; fully half have landed right in the middle; and at no stage has sampling error been so kind as to yield Turnbull a 51-49, never mind the tied result that would have spared him his present embarrassment.
That Newspolls have been as stable as this can only be regarded as a mystery.
The margin of error represents the inescapable limits within which any poll operates, no matter how well conducted — and Newspoll is actually coming off a rare miss at the South Australian election, when the Liberals were credited with 34% of the primary vote in the pre-election poll, but actually came through with 38%.
As luck would have it, Newspoll’s stability dates from around the time Turnbull trumpeted the pollster as solid-gold proof that Abbott had to make way for him.
This happened three months after the Newspoll market research company was wound up, after which Newspoll became a mere brand name for polling conducted for The Australian by the company now known as YouGov Galaxy.
The change was more than just cosmetic — whereas old Newspoll was conducted through interviewer-administer surveys targeting only landline phones, new Newspoll uses a combination of online surveys and “robopolling”. Old Newspoll had the opposite issue to new Newspoll, in that it became increasingly bouncy towards the end of its lifespan, most likely due to its difficulty in contacting sufficient samples of young respondents through its landline-only approach.
Whereas the average change from one poll to the next over the 30 most recent Newspolls has been 0.7%, the equivalent figure during Abbott’s prime ministership was fully 1% higher, very little of which would have reflected real change in voter sentiment.
If Turnbull had been privy to that level of volatility, the odds of him finding himself in his present quandary would perhaps have been as low as one in four. The reason that did not avail Tony Abbott is that the Coalition was clearly in a worse position under his leadership than it is right now.
In happier circumstances, all this minutiae about the performance of opinion polls would be of no imaginable interest to anyone outside a small coterie of psephological trainspotters. Perhaps nothing better illustrates the King-Midas-in-reverse touch of Turnbull’s prime ministership than that it now stands elevated as a matter of major national importance.