|Porcupine Fish Award for Ultra-Fishy Polling (image credit)|
Clive Palmer has claimed to have internal polling showing the Palmer United Party on track to win the Tasmanian election. The claim that PUP polling shows any such thing should not be taken even remotely seriously because:
1. The level of detail concerning this "poll" that has been released is grossly inadequate.
2. The poll was not carried out by a neutral polling firm or even one with any demonstrated record of competence.
3. Many Palmer United "internal polling" claims in the leadup to the federal election were shown to be wildly excessive by the actual results.
This is the second time in a week I've been picking on poor old Palmer United; the sad thing being that I actually hoped I could like them at this election. The three main parties have made very determined attempts to forfeit any chance of my support over the last four years. To impress me over such incumbents a new party doesn't have to do much more than avoid preselecting anti-gay religious fundamentalists, and avoid coming across as worse than Bob Ellis in areas I actually know something about. But for PUP, even these simple boxes are much too hard to tick ...
The latest unusual output from PUP is "internal polling" that said the party was "on course to win government", "expected to win 13 of 25 seats", "heading for victory" and so on. The initial press release carried no details of figures, methods, questions, timing or anything else that would permit the poll to be scrutinised. The claims are already being widely ridiculed on social media, and quite deservedly so. Questions asked by Calla Wahlquist elicited these responses:
* that the sample size was 500
* that the party conducted the poll itself by telephone (whether live or robodialling not specified)
* that the party would not release polling questions, respondent-selection details, detailed polling data or other information
Another story here says the sampling was random. It also says the distribution of support was consistent across the state.
Perhaps there will be more information in responses to other journalists, but for a start, let's compare this to the Liberal internal ReachTEL just released. The Liberal poll was four times the size and conducted by a well-known pollster who was named. The polling method is known, the poll question released verbatim, and dates, sample information and electorate breakdowns were all provided. Furthermore the poll question was adequately designed. Internal polling releases are always at least slightly questionable, but as far as practices for the public release of such stuff goes, the Libs deserve a thumbs-up. Most party releases of internal polling details do not get near this standard.
In comparison, with PUP we've been told very little so far, and what we do know doesn't bode well. We know that the "poll" was conducted by staff who are certainly not neutral and are not necessarily competent at polling. And we know that the sample size was only 500. (If anyone out there was actually polled by PUP in this poll, let me know).
You really can't do an accurate Tasmanian seat projection from a poll with such a small sample size. The problem is not just the 4.4% maximum margin of error on the primary figures but that the relationship between primary figures and number of seats won is shaky because of the various tricks of Hare-Clark, especially when a party's support is unevenly distributed. I'd comment on what spread of seat numbers might actually arise from PUP's claimed level of support in their own poll, but they haven't even said what percentage they are supposedly getting, so I can't.
What I can comment on is the claim that PUP support is consistent across the state. Coming off a state election in which underlying PUP support (as best measured by Senate vote) was more than twice as large in Braddon as in Denison, I think that's very unlikely, and would remain so even if the level of support increased greatly.
Knowing Them By Their Fruits
Since PUPpoll won't release their figures, we can only judge their likely accuracy from past internal polling comments that they have released. Clive Palmer claims his methods were accurate last election, but let's see. Here are a number of PUP internal polling claims I found using Google:
* In the final week of the campaign, Palmer claimed to be polling 30% in Queensland and 25% nationwide; actual results were 11% and 5.5%. (Earlier in the year he'd been claiming 40% and 30%.)
* Days out from the election Clive Palmer claimed his polls showed PUP was looking at "about ten Senate seats across Australia, that's what our polls are telling us". He disputed Galaxy and Newspoll as "rubbish that Rupert Murdoch puts out" (Galaxy had PUP on 5% at the time.) The party eventually won two Senate seats and their claims to a third are before the Court of Disputed Returns.
*PUP candidate for Ballarat Gerard Murphy said PUP internal polling showed the seat was between him and the Liberals. In fact Catherine King (ALP) retained the seat with a 54.9% 2PP vote. PUP polled 3.63% and finished fourth with less than half the vote of the Greens.
* In August on Sky Palmer said "Our polling's showing that we'll win two senators in Queensland, two in Tasmania, one in NSW and two in Victoria." They won one each in Queensland and Tasmania and none in the other two states.
Every internal poll figure I can find, and every seat projection supposedly based on polling, that PUP released prior to the federal election was a massive exaggeration compared to the actual result. To say that the polling predicted what success PUP had is useless given that PUP's "internal polling" predicted so many successes that didn't occur. Indeed the track record is that if PUP "internal polling" says PUP will succeed to a stated degree, then they won't.
Of course the party can claim the election of Jacqui Lambie proved its internal polling numbers to have been close to the mark, but this is not remotely true for any internal polling they made public that is easily findable. For any polling they didn't make public at the time, the claim is probably now untestable.
The Australian's compilation here also showed that Palmer fast gets in hot water when quizzed on his understanding of polling. The interview with Peter van Onsolen on Sky is revealing; Palmer initially reckoned that every poll that isn't a Murdoch-press-commissioned poll showed his party romping it in, although in fact no mainstream poll did so whether it was a Murdoch-press-commissioned poll or not. Then he dropped back to referring to a newspaper-run "online poll" (a useless source of evidence because of its opt-in nature and reader bias) and even then couldn't characterise that "poll's" question properly.
Palmer also claimed that with ALP preferences his party could have beaten Tony Abbott and Malcolm Turnbull. In fact Wentworth, as a highly educated posh inner-city seat, is exactly the sort of place PUP were never going to get any traction on primaries under any circumstances (they finished second-last on 1.08%) and in Warringah Tony Abbott easily cleared an outright majority (as did Turnbull) with PUP polling 2.2%.
Bucking The Experts?
It's true that the strength of PUP's performance on federal election night surprised plenty of analysts (me included). But while PUP outperformed their mainstream polling on the day, they did so on the back of ferocious advertising spending in the final week, and even then their results were still in the ballpark of the published mainstream polling, just on the higher side of mainstream expectations. That PUP's mainstream polling weeks out was lower than the final result mainly indicated that that was around the size of their support at that time, not that the polls at that time were "wrong". There may also have been some polls that slightly underestimated PUP because of design issues. Single-order robopolls with PUP way down the list could be an example of this.
Indeed if PUP actually believed their claimed "internal polling" they would have been disconsolate at the difference between it and the national outcome; instead they came across as very pleased with their handful of successes. And there were plenty of analysts who took the chances of Clive Palmer winning Fairfax, or of the party getting at least the Queensland Senate seat, quite seriously.
The party have denied that this poll release is a stunt. While I would not recommend accepting that denial, here are two other possible reasons why they might have chosen to make these claims about their support level now:
1. To raise supporter expectations about the party's chances after its poor results in the Liberal ReachTEL and in recent national polling. Palmer United voters don't seem to be especially critical in their thinking about their own party for the most part, but they would be likely to believe a poll result showing they were struggling unless Clive came out and claimed to have evidence it was all nonsense.
2. To create a distraction to take the heat off Mark Grewar and other PUP candidates coming under scrutiny over social media posts, alleged past signed petitions and the like.
One slight disappointment with a poll like this is that some media outlets will run it with no real indications of doubt about its validity. It is newsworthy (in the entertainment section at least) that the claim is being made, but if Clive Palmer released a press release stating he had developed a cure for cancer that worked by banging spoons together, you would expect newspapers to run it past a doctor before printing it. Likewise if Palmer claimed to have set a new world speed twerking record you would think the folks at Guinness would be asked if it was true. The reports published so far have varied in their vigilance levels and there are a couple of tweets I think are worth reproducing here:
Since this piece was written, Chris Pippos in The Advocate has treated the PUP "internal polling" with the contempt that it deserves.
Update 13 Feb: At the National Press Club Palmer was singing a very different tune:
"Our polling is showing that we will win between three and five seats because, broadly speaking, you've got the Greens and Labor occupying about 40 per cent between them, the Liberals about 40, 43 per cent, which is what our polling is showing"
Given that this brazenly contradicts the internal polling claims made by Palmer a week ago and that Palmer has a history of pushing "internal polling" claims that fail when tested at elections, none of this has any credibility either unless (at minimum) full figures and details of polling methods (including the identity of the pollster) are released.
Anyway, there is a real poll out today!
Further Update 17 Feb: Since this article we had a small-sample EMRS that didn't give PUP enough support for any seats, followed by a ReachTEL from several days later that had them in the mix for one or two (but at this stage no more). I found a full video of Palmer's NPC comments, which say that according to their polling Labor and the Greens would win 10 seats, the Liberals 10-11 and PUP 3-5.
A revealing sign that Palmer doesn't know too much about Hare-Clark in Tasmania comes when he suggests the quota is about 20% (a common mistake, the quota for five seats is a sixth (plus a vote) not a fifth). Another is treating Labor and the Greens as a combined party. Because preference flow is weak in Tasmania this doesn't necessarily work; two loosely allied parties can do either better or worse in seat terms on a given vote share than if the same vote share went to a single party. Of course, if the big three had only 83% of the vote between them and PUP had most of the rest, PUP would win seats, but no figures were provided to justify that.