Poll Roundup And Seat Betting Watch: Ellis Impersonator Edition

2PP Aggregate 52.5 to Labor (unchanged)
On most recent polling Labor would win a majority if election held now, with around 83 seats.

There have not been many polls in this week of an Easter-fragmented campaign, and the one national poll that has come out hasn't changed the national picture all that much (except for One Nation).  Essential will be polling over Easter (more likely to work for an online than a phone pollster), Newspoll probably won't, and we'll have to see what else we might get.  It's been a scrappy start to the campaign with both major parties losing candidates from uncompetitive seats (after some dumpings foreshadowed tonight, the Coalition will have lost six!), and gaffes by both sides including some troubles for Bill Shorten on superannuation policy and climate change policy costs, while Scott Morrison can get away from any question he doesn't like by just declaring it to be "bubble stuff".  I don't know if anyone's paying attention to any of this at the moment out there in voter-land.



This week's Newspoll came out 52-48 to Labor off primaries of 39 each for the majors, Greens 9, One Nation 4 and others 9.  The One Nation primary finally crashed; whether because of the Al Jazeera guns sting, or from competition from the Coalition or UAP is still unclear.  By 2016 preferences these numbers would typically come out to 52.9% 2PP, and One Nation preferences can't be having all that much impact while they're only getting 4%.  Therefore it is extremely likely the 2PP was rounded from something just below 52.5.  My aggregate uses last-election preferences as a headline figure and I aggregated this poll at 52.8 (the information that the published 2PP was 52 affects the range of probabilities slightly).  This wasn't enough to shift my polling aggregate which remains at 52.5, in accord with those of Bludger Track (also 52.5) and Mark the Ballot (52.6).  You can see their tracking graphs on those links (Bludger Track uses much longer-term smoothing than Mark's or mine) and here's mine:

I have also recently fired up my usual conditional probability model, which I will post some output from when I get a quiet moment in the next few weeks and there are more polls.  For the current estimated 2PP it has Labor winning 83 seats (assuming - which is by no means sure - that Labor doesn't lose Macnamara to the Greens.)  The principal difference between my model and Bludger Track is that I'm not including state breakdowns at this stage, and therefore my model is "only" throwing Labor five seats in Queensland rather than nine.  Actually, for a 2PP of 52.5 and given the remarkable history of the Labor vote softening in Queensland at federal elections, I doubt there would be even that many gains.

Very little changed in the leadership polling this week, so I see no need to say anything new about it.

Bass Poll

The only new seat poll I'm aware of was an Australian Forest Products Association commissioned uComms poll of Bass, with the apparent shock of a 54-46 lead to the Liberal candidate Bridget Archer.  This is less of a shock if you know that in 2016 every one of the four published polls for this seat underestimated Labor by at least five points on a 2PP basis.  As usual we have the problem with ReachTEL-platform polls of the primaries being released in a raw form without redistributing what they call the "undecided" vote.  The raw primaries were:

Labor 32.6 Liberal 42.8 Greens 10.0 Christian Democrats 1.9 (unclear whether running), Nationals 1.2, "Palmer's United Party" [sic] 2.7 Other/Ind 2.5 Undecided 6.3.

The breakdown of the "undecided" was:

Labor 31.4, Liberal 13.7, Green 5.9, CDP 7.8, Nat 11.8, "PUP" 29.4 Other 31.4

Which is handy - but sums to 131.4%.  Probably one of the two 31.4s is meant to be zero. If I assume that's the case then the primaries in the form in which other pollsters report them are:

Labor 34.6 Liberal 43.6 Green 10.4, CDP 2.4, Nat 1.9, UAP 4.6, Other/Ind 2.4

This comes to about 52.5% by national last-election prefs and 51.5% by preferences recently seen in this seat, so the flow of respondent preferences was weaker to Labor than it usually would be.

This result has of course led to a lot of searching for reasons as to why Labor might be struggling in Bass, although it's entirely possible they are not and that this is simply another rubbish seat poll.  Possible reasons do exist though, including the poor baseline of the 2016 Liberal campaign, recent funding announcements by the Liberals, and Malcolm Turnbull having played badly in service-vulnerable regional seats.  See my Tasmanian House of Reps guide for more on the seat.

At least those who didn't already know it should know from this poll that uComms doesn't always skew to Labor.  It does seem, however, that its results are really all over the place.

Ghosts Of Campaigns Past

The Australian this week ran an op ed from Dennis Shanahan (don't mention the blogwar) which showed examples of shifts from the first Newspoll of the campaign to the final result.  In the last five campaigns these all favoured the Coalition.  Shanahan opened the article with:

The first Newspoll survey of the 2019 election campaign has the Coalition in the same position as the past five governments: ­behind in two-party preferred support when the election was called, and that includes three that went on to win.

Except in yet another example of an op ed not being adequately edited for factual accuracy, this was in fact incorrect; the Gillard government in 2010 was way ahead in Newspoll when the election was called (an outlying reading compared to other polls) and the paper's own graphic showed it.  So in fact only four of the last five governments were behind when the election was called, for two wins and two losses.

However, one of those (John Howard's in 2004) wasn't actually behind at all.  That was the year of the old Newspoll's fling with respondent preferences.  The poll that had Howard behind 48-52 as he called the campaign should really have been 50-50.  Using the full history of Newspoll and converting 2PPs where none were published, the record is as follows:

* Four governments were ahead on 2PP as the campaign started; they all won
* Six governments were behind on 2PP as the campaign started; three won and three lost
* One government was square on 2PP as the campaign started; it won

Of the three that started behind, all were behind only 49-51.  One of these (Howard in 1998) won despite losing the 2PP by that margin, so it didn't make any actual gains.  The two that were behind, made gains, and won, were Keating's in 1993 and Turnbull's in 2016 - but everyone keeps telling us what a bad campaign Turnbull ran?

The point that the Coalition has gained compared to the first Newspoll of the campaign in all five campaigns since 2004 is a valid one, but in the four campaigns before that by the same yardstick the Coalition went backwards and lost in 1993, backwards slightly in 1996, broke even in 1998 and went backwards by over six points in 2001.  Of course, 2001 doesn't count because of September 11, but then neither do 2010 or 2013, in both of which the early-campaign polls were influenced by Labor having just changed its leader.  Historically there is quite a tendency for the Coalition to, on average, do better relative to its leadup polling (over pretty much any timeframe) than Labor, but it is not as reliable as a string of five in a row suggests.

Hey Bob, Is That You?

Nationals leader Michael McCormack is well known as an Elvis impersonator but unfortunately he's lately seemed more like an Ellis impersonator - Bob Ellis that is.  Unfortunately McCormack's rich seam of opinion-poll denialism has way too much of the late lamented literary figure (and late unlamented misogynist and polling crank) about it.  So apparently polls are "usually done in metropolitan areas" (well funny that, that's where most of the people live, but of course they survey in the bush too), they ring up at dinnertime (how does he explain online surveys that don't ring up at all getting the same results), and had they been right then:

* "Steven Marshall would not be the premier of South Australia" - final SA 2018 polling was line-ball as to who would form government
* "Will Hodgman would not be the premier of Tasmania" - polls in the last fortnight had Hodgman winning. The last poll he wasn't winning in was three months before the election (plus Tasmania has a particularly strong bandwagon effect at some elections caused by desire to avoid minority governments.)
* "Gladys Berejiklian probably wouldn't be the premier of New South Wales" - Berejiklian's government was on course to govern (but its majority prospects were unclear) by the final Newspoll, and had never been clearly losing before that (at worst about a tossup) even when slightly behind on 2PP.

Of course I will pay such examples as Brexit and Trump but these concern voluntary-voting elections that are irrelevant to Australia.  And in the case of Trump, national polling had him losing the national vote (which he did, by not much less than expected); the fault with polling was mainly in a number of critical states.

Unlike Ellis, McCormack doesn't claim that pollsters only call elderly voters, because that wouldn't help his cause.  But it's a similar sort of thing, albeit a polite and non-cranky version and in a conservative mirror image.  If polls were never "representative of all Australia" then why, in 2013 and 2016, were they right?

Seat Betting

Election betting is not reliably predictive, but it's interesting to keep an eye on anyway, especially as people still keep claiming it is.  At this election there is an ongoing difference between what the seat betting predicts and any reasonable extrapolation from current polling.  This doesn't mean the seat betting markets are saying that the polling is wrong about current voting intention, but it may be that bookies/punters expect it to blow out later.

Here are the current expectations (see here for methods):

Expected ALP classic-seat gains (not close): Corangamite, Gilmore, Flynn, Robertson,  Chisholm, Reid, Dunkley

Expected ALP classic-seat gains (close): Banks, Capricornia, Petrie, Dickson, Forde, Hasluck, Boothby, Dawson, Bonner, La Trobe, Pearce, Swan, Leichhardt, Brisbane, Deakin, Stirling

Expected Coalition classic-seat holds (close): Casey, Sturt,  Canning, Bowman, Aston,  Hinkler, Hughes, Flinders, Ryan, Menzies, Grey

Expected Coalition hold (three cornered) (close): Higgins

Split market Coalition held vs ALP: Page

Expected Coalition gain from IND (close): Wentworth
Expected Coalition holds against IND (close): Farrer, Warringah
Expected Coalition holds against Shooters (close): Parkes, Calare
Expected IND hold vs Coalition (close): Indi 
Expected IND gain from Coalition (close): Cowper

Expected Labor hold vs Green (close): Macnamara
Expected Labor hold vs Coalition (close): Lindsay, Bass, Herbert

The only change in favouritism this week is that Page became a split market (just). Here's the chart of changes in favouritism:

Key to colours (more may be added):

Red - Labor favourite in all markets
Orange - Labor favourite in some markets, tied in others
Dark blue - Coalition favourite in all markets
Light blue (none yet) - Coalition favourite in some markets, tied in others
Grey - all markets tied or different favourites in different markets
Purple - IND favourite in all markets
Pink - IND favourite in some markets, tied in others