Bob Ellis is the Unskewed Polls of Australian politics.
A digression: once, I used to feel some sympathy for Ellis. Some may recall Bob Ellis cost his publishers $277,000 with basically a single multiply false sentence in the book "Goodbye Jerusalem", which defamed Tony Abbott, his wife, and Tanya and Peter Costello. When the full facts about Tony Abbott's abandoned "son" came to light, I thought that the discrepancy between Abbott's former personal life and Abbott's religious-morality attitudes as a Howard Government minister said far more to Abbott's detriment than any scuttlebutt falsely claiming he was sexually seduced into the Liberal Party ever could. There are some small things in that case that might have gone down differently had Abbott's full biography been known when it was heard.
However, the sympathy I had for Ellis from that has been wiped away many times over by his recent series of comically bad attacks on mainstream opinion polling in Australia. Quite aside from his the defamatory and tinfoil-hat nature of much of his work, a big problem with Ellis's comments is that what he writes about opinion polls is riddled with easily avoidable factual errors.
And yet, while the theme of this article is totally and painfully self-evident to most lefties who have heard of Ellis (to the point that a common response would be "Der!"), there are still some out there who fall for Ellis's writing on polls without themselves being as cranky as he is. Rather, they like Ellis and believe him because he tells them what they want to hear - that if a poll gives Labor bad ratings it is probably at least incompetent, and quite likely a Murdoch plot.
I am probably only going to do this once (together with any follow-ups required). I just want to put something out there that can be easily found and referenced as evidence of the grotesque cluelessness of the man's polling commentary. My aim is to produce one article, concerning one Ellis blog post, that shows that he makes such a large number of errors that his comments on polling cannot be taken seriously. If this article saves just one person from his sad mirror-image of US Republican psepho-kookery, it's worth it. Also, even among those who may not take Ellis seriously, a lot of the issues he raises (like mobiles-v-landlines, and KAP and PUP preferencing) have been common objections to recent polling, so there may be some use in addressing those.
The blog post I've selected is entitled "The Nielsen and Galaxy Polls". As I write, it's already buried eleven "posts" deep on Ellis' blog (two of which simply repeat furphies from the piece in question), but Independent Australia also did themselves no favours by publishing it here. I'm going to go through it line by line and stop every time I find a mistake or an erroneous argument that I haven't already disposed of (some occur multiple times). It's going to take a while.
Ellis is in bold, and my responses are not. Each beyond-doubt factual error by Ellis is marked with an F and the number of the error in brackets.
Ellis: NIELSEN AND GALAXY prove to be cheating also, and this is a pity.
They do not show the Preferred Prime Minister,
KB: (F1) Nielsen's "Preferred Prime Minister" scores (50-42 to Rudd) were common knowledge by shortly after midnight Friday (indeed seen in a tweet by @GhostWhoVotes at 10:03 pm) but Ellis was still questioning their absence repeatedly through Saturday. Galaxy released PPMs in their national poll released on Saturday, although none from their Queensland poll have yet been sighted.
Ellis: or how well Katter and
Palmer are doing,
KB: (F2) The Galaxy poll canvassed results for Katter and Palmer (4% each) which were obtained by Poll Bludger. A common error in Ellis's reasoning seems to be that if he cannot find the poll results in mainstream reporting or immediately on the pollster's website, they don't exist. In any case, as for Nielsen, KAP and PUP are barely blips on the radar nationally so amalgamating them under Others is hardly a major deal.
Ellis: or how the preferences were distributed,
KB: I cut Ellis slack here and assume he means how they were distributed by the respondent, not the pollster.
Ellis: or how old
their respondents were.
KB: (F3) While Nielsen does not publish the number of respondents in each age group, it does publish the maximum MOEs for each, from which the numbers can be fairly accurately determined. In this case, 18-24 approx 175, 25-39 approx 355, 40-54 approx 380, 55+ approx 490.
Ellis: They polled from Tuesday to Thursday, for the
first time in world history,
KB: (F4) Nielsen apparently normally polls from Tuesday to Thursday once campaigns start. Here at least is an example from 2010.
Ellis: including Cheap Movie and Late Shopping
nights. They rang only landlines,
KB: (F5) Galaxy polls mobiles as well as landlines.
Ellis: which nobody under thirty-five uses.
KB: (F6) This is an exaggeration, of course, at any level. In fact, while young age groups are harder to contact by landline if living independently, significant numbers of young voters still live with their parents. Clearly plenty of under-35s were polled by Nielsen.
Ellis: And they both occurred before Peter Beattie came in.
KB: (F7) Not entirely. Part of each poll was taken on Thursday, with the news of the Beattie run breaking on Thursday morning. Even the Thursday respondents would not have all been aware of the news, and news of this sort takes some time to be fully processed by voters. But the sample was not entirely pre-Beattie as implied.
Ellis: And Nielsen, even then, could only squeeze Labor down to 48. Howard won in 1998
KB: Howard did win with 49 but had the benefit of a massive number of first-term sitting members with new popular votes to dampen swings, and a lot of room to move in targeting areas to sandbag. It is unlikely 49 will be enough for Labor, without those advantages.
Ellis: The results demonstrate that 68-year-olds think Abbott more
trustworthy, better at sums and better at frightening boat people.
KB: (F8) Ellis has tried before to create the impression that the average landline poll respondent is in their late 60s. But if he was any good at sums himself he would realise that in this case, this would require the 55+ age group to have an average age of at least 113.
Ellis: I bet
they do. And preferring, I guess, Rudd to Abbott as PM by 55 to 45,
which is why it is not up there.
KB: A ten-point lead to Rudd as PPM would actually be nothing to write home about anyway since the PPMs of all pollsters except ReachTEL skew to the incumbent. As it turned out Nielsen showed 50-42 and tonight's federal Galaxy 47-34. No PPM from the Queensland Galaxy has yet been seen; one may or may not exist.
Ellis: Nor does it say, as it used to, ‘second preferences allocated as in
2010’ – before the KAP and the PUP existed, and the DLP was resurrected,
and the Democrats imploded.
KB: (F9) The Democrats' support really imploded from 2004 with the loss of all seats and a vote below 1% in the House of Reps in 2007 and 2010.
Also, the DLP "resurrection" happened long ago, and Madigan's was not the first seat the new DLP have won - they also won a state seat in Victoria in 2006. It is not as if there is a massive known boom in primary support for the DLP at House of Reps level - mainly, they got lucky on Senate preferencing.
Ellis: Taking all these things into consideration, it shows, or indicates,
or forebodes, or promises, I think, Labor on 52.5, or 53 after Beattie
arrived, and Labor winning handily.
KB: These figures all appear completely out of nowhere even assuming Ellis's assumptions are sound.
Galaxy is wickeder.
It polls only Queensland, does not say how many Queenslanders it
polled, does not mention Katter or Palmer, does not ring mobiles, rings
on Cheap Movie and Late Shopping nights,
KB: (F10) Most of the above is addressed already, but the sample size of 800 Queenslanders was also well known.
Ellis: maldistributes the pro-Rudd
preferences of his fond friend Bob Katter,
KB: While Katter himself is friendly towards Rudd and was willing to support him til the election there is zero published direct evidence that KAP preferences will support the ALP just because Rudd is leader. In 2007 those preferring Katter himself over Labor or Coalition in Kennedy - 92.7% of whom voted 1 Katter - preferred the Coalition candidate over Labor by a margin of 73:27. That was with Rudd leading Labor; in 2010 with Gillard doing so it was 70:30. In the Queensland state election KAP voters whose votes did not exhaust were twice as likely to preference the LNP as the (admittedly hapless) state Labor government. Katter's attraction seems to be more to rogue rural conservatives.
Ellis: shows no Preferred Prime
Minister, did not wait till after Beattie came in, nor did it ring back
anyone after he did, and was paid for by Rupert Murdoch, who wants to
‘throw this mob out’ and sacked, this week, Gough Whitlam’s son-in-law Kim Williams.
Assuming an 8% primary vote for Katter, a 4% for Palmer, a 6% for the
Greens, a base Labor vote of 36 and a base Coalition vote of 46, the
two-party preferred vote comes in at 50-50,
KB: Here Ellis is not only setting Ind/Other excluding KAP/PUP at zero but also crediting the Coalition with only a c. 22% preference flow from KAP, PUP and Green combined. That would suggest KAP and PUP supporters were about as pro-Labor as Greens, if not more so!
I add that the only resemblance those primaries bear to the real Galaxy Queensland ones is that Ellis somehow guessed one right. PUP really are on 4. But I will cut him slack again and assume he knew the Labor and Coalition polled votes were really 34 and 48, and changed them because he thought they were not accurate.
Ellis: a 16% swing to Labor, and a
Coalition loss in Queensland of twelve to seventeen seats.
KB: (F11) Here Ellis blunders by calculating the swing as the difference between his own claimed 2PP (based on a ludicrous distribution from concocted primaries anyway) and the primaries from 2010. But actually the swing is calculated 2PP to 2PP and even Ellis's bogus 50-50 in Queensland is only a swing of 5.1%. A swing of 5.1% in theory wins 9-10 Coalition seats, but in practice would not win the fake marginal Fisher and would likely fall short in one or two others because of sophomore effect too. So, say, seven seats. Not that that swing is happening at present anyway.
Ellis: There is no other reason why there were no Katter and Palmer numbers listed, no Preferred Prime Minister, and no Greens.
[one defamatory sentence of Ellis's redacted; I don't have a spare $277K - KB]
Ellis: Morgan on Monday will tell the truth. And it will be, I think, to judge by these fabrications, Labor on 52.5.
KB: Prediction noted. It could be the truth, or 52.5. It certainly won't be both of those things. :)
(Update on that score: predicting poll readings is a bit like derivatives trading and anyone who's tried and failed has my sympathy. That said Ellis was wrong by 2.5 points by respondent-allocated preferences, which I would expect him to prefer. Morgan has just come out at 50-50 by that method.)
Updates may follow. I thank my partner for the comment that unintentionally inspired the title. I want to make clear that this article isn't just "pedantry" in which one snips around the edges of a basically sound or plausible case without overturning the main point. Rather, I have shown that Ellis's piece is so hopelessly riddled with major errors that once those errors are corrected, the original is gutted and the carcass isn't even fit for crab bait.
After that light entertainment, normal service will resume with the weekly Poll Roundup and Seat Betting Watch, which I intend to deliver the day after Newspoll is published in the Australian.
Addendum: Twitter Silliness (12 August): To my knowledge, word of this piece has yet to reach the shores of the great man himself (or he's ignored it) but the following insight into the mindlessnessset of his followers was amusing. Here is one of the original uncritical tweets of Ellis's article:
After I published mine, the following discussion with this person resulted:
GP: What do you say to the proposition (made to me by a cranky statistician) that any poll under 3000 is not worth it
KB: A poll under 3000 can be very useful data, it just won't tell the whole story by itself.
GP: That is not an answer. Suggest you pursue the mathematical proof of the 3000 hypothesis with relevant academics...
KB: The relevant academics can publish it somewhere if they expect it to be taken seriously for >0.1 seconds :)
GP: Thought that might be yr response. Was happy to give you email contact. Cannot take you seriously now. Blocked
The first of doubtless many! Now quite aside from the comic gracelessness (pun intended) of trying to convince someone of nonsense then blocking them because you don't like their response, there's a little bit more worth mentioning here. There's a 98% chance that GP misunderstood whatever the "cranky statistician" was saying in the first place, but even if she didn't, a statement that's plainly absurd becomes no less so for having been made by someone with a qualification. Debates on these things proceed best through the public exchange of ideas. Not through exclusive twitter invites (apparently gained only through excessive forelock-tugging towards rumours of a not even remarkable level of authority) to email someone who someone knows.
Note added 2015: See also: What Is Independent Australia Independent Of?