Rolling Poll Roundup And Seat Betting Watch: The Final Days

2PP Aggregate 51.5 to Labor (by last-election preferences) (-0.3 since last week)
51.1 with One Nation adjustment
Current seat projection approx Labor 79 Coalition 66 others 6 assuming polls accurate.

The end, it seems, is nigh.  If the (ludicrously herded) national polls are right, the Morrison government would need a huge amount of luck to survive a swing of somewhere around 1.3-2.3%, at an election at which it can afford perhaps two net seat losses.  If the polls on the whole are just modestly wrong in one direction, the Government's chances of survival improve greatly, but the other direction leads to a decisive loss.  The most likely path to survival is a Donald Trump style path - a combination of a modest national polling error and a fair amount of luck with the local breakdowns, but if it happens in this case, after everything, it will be very surreal.



If you wanted to look for an argument for the polls having Labor too high, you might look to the history of the Liberals outperforming even their final 2PP polls, though that's really only happened significantly three times (the last was 2004) compared to Labor's once in the last ten elections.  For them having Labor too low, there's been a recent spate of blowouts from close polling in state elections (a severe case being Victoria) and some federal by-elections, but perhaps that's just coincidence.  All will be clear soon enough.  In the meantime, I'm nervous about relying on a herded-looking set of polls, that come from a mere four polling companies, and that in two cases I have serious methods concerns about (Ipsos mainly with the Greens vote, and Morgan, well ... I do not have all day).  But I was also nervous last time and in that case the national polls, herded or not, were right.

This post will be a rolling post to cover off on the current state of polling and to deal with polls as they come out, up to whatever point I feel like it's time to do a new one (which will probably be Friday sometime).  The 2PP and seat projection will be updated at the top, but the tracking graph won't be.  There will be a delay in the seat poll coverage through the middle of the day but I will try to get them under control by mid-afternoon.  After the Galaxy seat polls have all arrived I will be adding my current seat poll model output (a spoiler: at the moment it's being a lot more generous to Peter Dutton's chances than anybody else is, perhaps wrongly.) More updates will follow at the bottom.

This week's polls

Not much to add since last edition.  We had another 51-49 to Labor Newspoll, the third in a row.  By previous-election preferences the primaries would on average have given a 2PP of about 51.7, whereas by Newspoll's current method (with adjustments for UAP and One Nation) it would be about 51.1.  A number of people on Twitter have confused my comments on this with my frequent comments about rounding, and concluded that Newspoll has been rounding from the high 51s down to 51.  This isn't the case - it has simply been getting a number that correctly rounds to 51 because it is using an altered preference method.

Newspoll was most notable for the leadership figures, in which Bill Shorten's netsat improved to -10 (39-49), his best since March 2015 and the end of almost four years to the day with a dissatisfaction rating of 50% or above.  Shorten also closed the gap on the skewed "better Prime Minister" metric to seven points (45-38).  Apart from the reading taken during Morrison's first week in the job, the 38% was Shorten's best score since Tony Abbott was Prime Minister.  On the strength of these numbers, anyone who thinks Shorten's unpopularity will lose Labor the election would be well advised to belatedly get off that horse.  (They should not have been on it in the first place, given the weakness of evidence for opposition leader ratings affecting voting intention.)  Indeed, Morrison's lead is actually unusually small given the 2PP result. Whether Labor's policy or campaigning decisions could lose them the election is another matter.

A Morgan was also sighted which, after much confusion, proved to be another face-to-face poll with a 52-48 2PP.  This came out to 51.6 by last-election preferences.

The net impact of these two polls on my aggregate was virtually nothing, with the Coalition gaining another tenth of a point.  Here's the smoothed tracking graph, which is going to look very sorry indeed if the Coalition do somehow pull this one out of the fire:

Essential has not yet emerged so is presumably working on its final poll.  Ipsos is in the field - I know as I was actually polled!  It was rather disappointing to find they still had the Centre Alliance in the readout in Tasmania when it is only contesting three seats nationwide, all of them in SA.  Also One Nation are in the readout (not contesting my seat either.)

This week we had Newspoll primary vote breakdowns (see Poll Bludger review), which were based off the last five Newspolls.  However some of these deserve some comment.  Firstly for Victoria the One Nation vote of 4% is too high as they are contesting only 5 seats there, and it is one of their weakest states.  One would expect the lost votes to spray to the major parties and UAP, but there won't be any effect of them following the how-to-vote card.  Also for SA, the published 2PP is 48-52 off primaries of 38-all for the majors, 9 for the Greens, 3 for One Nation and the rest for others, but that seems a bit stingy to me, given that the others would have until recently included more votes for Centre Alliance (whose preferences split 60-40 to Labor last time).

Newspoll Naughtiness (And Other Seatpoll Silliness)

One assumes it was the Australian's decision to commission seat polls of Bass, Herbert, Lindsay and Corangamite (and no other seats) to come out this week, but if so this was not very good behaviour.  Newspoll's national polling has persistently suggested that more Coalition seats are in danger than Labor, so choosing to poll three trouble-spot Labor seats (two of which had been polled already) out of four seemed like the sort of thing you would do if you were more interested in geeing up the faithful than in providing a read on the band of seats that would be expected to decide the outcome.  The polls had Labor ahead 52-48 in Bass, behind by the same in Herbert and Lindsay,  and ahead 51-49 in Corangamite.

Also seen this week were some Greens seat polls of the target seats of Kooyong and Higgins.  These were by Environmental Research and Counsel, a little-known firm which is within the Essential Media group of companies (along with Essential Research which does the Essential polls and also UMR, a well-known ALP internals pollster).  However it isn't actually the same pollster as Essential.  The Kooyong poll had an enormous sample size of 1741 and found Julian Burnside (Greens) trailing 48-52.  (Josh Frydenberg's 41% primary rises to a reasonably safe 44.5% after reallocating undecided).  It will be interesting to see how this one goes, but it has already had a considerable effect by casting Burnside as the challenger and killing off most speculation about Oliver Yates' chances, causing Yates to suggest he had been stiffed by the Greens finding such a high profile candidate.

Far less promising in sample size terms was the Higgins poll which reached a mere 400, apparently by computer-assisted live calling.  This apparently had the Liberals on 36%, Labor 30%, Greens 29%, but this would require a 16% primary vote swing against the Coalition and I don't propose to take that seriously at all.  The 2CP was 54-46 to the Greens (assuming they snuck into second) - off those primaries it would very likely be more than that.  However the sample size is so small that had an established pollster done this one in 2016, it would have had effective error margins in double figures on the primary votes.  I really don't think that party polling with sample sizes this small should be published in the media, but plenty of the 2PP figures for internal major party polling flying around are probably from tracker polls that are even smaller.

Although many people are more interested in it than the rest of the election, the great seat of Warringah has still to attract one single neutral seat poll attempt.  There was one story with inadequate details about the Liberals having Abbott at 50-50 in their polling but it was almost immediately contradicted by other reports that he was still way behind.

Late tonight a 51-49 to Liberal Galaxy of Deakin was released (Newspoll, ie the same pollster, having earlier found the same).  Over Thursday seat polls of Reid, Gilmore, Macquarie, Higgins, La Trobe, Flynn, Herbert, Forde and Dickson plus the full details of the Deakin one will also drop (one per hour from 10 am, possibly with silly bonus questions) and I am also aware of a Galaxy of Hasluck that has yet to be released.  These seat poll results will be listed in the space below with comments:

-----------------------------------------------------------------------

Flynn 53-47 to LNP. Consistent with various claims of north Queensland having gone soft on Labor as it frequently does at election time.
Macquarie 53-47 to Labor.  Less for Labor than would be expected off the national swing, and this seems to be a general pattern in Labor seats that have had seat polls.
Forde 50-50
La Trobe 50-50
Reid 52-48 to Liberal
Higgins 52-48 to Liberal over Green off primaries "LIB 45 (-6.6 since election) GRN 29 (+4.8) ALP 18 (+1.5) UAP 4 (+4)"  Swings quoted from @GhostWhoVotes - poll shows a surprisingly small increase in the Labor vote.
Gilmore 52-48 to Labor
Herbert 50-50.  This one has a massive non-major vote and a particular issue is the flow from Katter's Australian Party who have 14%.  In 2016 their preferences statewide flowed 53% to Coalition.  It appears the same is being expected by Galaxy again.

-----------------------------------------------------------------------

Seat Betting

Seat betting isn't especially accurate, but it's interesting to keep an eye on it given the ongoing debates about the accuracy and predictive use of betting.  So here's the results of my Wednesday night run around the markets (see here for methods):

Expected ALP classic-seat gains (not close):  Dunkley

Expected ALP classic-seat gains (close): Dickson, Petrie, Gilmore, Reid, Robertson, Chisholm, Hasluck, Swan, Corangamite, La Trobe, Forde

Split market in Coalition classic seats: Flynn, Stirling

Expected Coalition classic-seat holds (close): Brisbane, Bonner, Dawson (tied in some markets), Leichhardt, Banks, Page, Casey, Deakin, Flinders, Higgins, Pearce, Boothby, Grey, Capricornia (tied in some markets)

Expected Labor loss to Coalition (not close): Herbert
Expected Labor loss to Coalition (close): Lindsay
Split market Labor held vs Coalition: Braddon
Expected Labor hold vs Coalition (close): Bass, Solomon, Lingiari

Expected Coalition loss to IND (close): Cowper, Warringah
Expected Coalition gain from IND (not close): Wentworth
Expected Coalition gain from IND (close): Indi
Split market Coalition vs IND: Farrer

Expected Labor hold vs Green (close): Macnamara

And here's the colour-coded favourites tracker:

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 - 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

(Note: there's an error in Dunkley; it should be Liberal -1.0)

Seat betting markets are pointing to totals in the low 80s.  Overall the betting markets have just about fallen in with what the national polls are saying, after initially expecting a blowout from the less close polling at that time.

Updates Thursday night:

Essential has come out with a 51.5% 2PP and Ipsos with a 51% 2PP.  I've aggregated these at 51.6% and 51.0% respectively, and they've taken another two-tenths off my aggregate, and another seat from Labor in the projection.

The Ipsos preferencing mystery continues with the following statement appearing:

The two-party result is based on preference flows at the last election, allocating second preferences from One Nation and United Australia Party using a split of 53 per cent to the Coalition and 47 per cent to Labor.

Perhaps the word "except" is missing somewhere, to indicate ON and UAP are not distributed based on last-election preferences?  Neither ON nor UAP delivered such a split at the last election, though UAP delivered close to that (as PUP) at the one before.

Ipsos also reports that voters who have already voted are splitting 41-33 on primaries compared to 39-33 overall (Ipsos has the Labor primary low as usual).   This doesn't suggest a big difference to last time as pre-day voters were also slightly more Coalition-friendly in 2016.

Here's some seat poll model output (click for larger clearer version).  The current version is off a 2PP of 51.5 so it aims to estimate results and probabilities on a 2PP basis assuming that that's the 2PP.  (If the 2PP is different, then the probabilities change, so a 99% probability in this model doesn't necessarily mean the same thing in reality.)

The model takes into account existing margins, national swing, personal votes and (at a rather low weighting) seat polls.  In total it expects the Coalition to suffer a net nine 2PP losses, and has eight specific seats where it expects a loss: Dunkley, Corangamite, Capricornia (hmm), Forde, Gilmore, Robertson, Banks (hmm) and Petrie.  Some of those are at close to 50-50, but it is also cautious about Flynn, Dickson, Hasluck (a widely expected loss) and Page.  It doesn't have any adjustment for state differences, so I'd think Chisholm is a more likely loss than some of those it does expect to fall.  The Poll Bludger version, which has similar totals and much swishier presentation, includes state breakdowns but not seatpolls.  As a result of using state breakdowns, its casualty list is quite different.  (I don't use them because I don't have enough data.)