Friday, November 23, 2018

2018 Victorian Final Polls

Galaxy 53-47 to Labor, ReachTEL 54-46 to Labor
Current primary vote aggregate ALP 40.7 L-NP 39.4 Green 11.0 Other 8.9
Polls could be underestimating Labor 2PP vote slightly and may be overstating Greens primary
Seat projection estimate ALP 48 L-NP 36 Green 3 IND 1

This post will update all polling news in the final 24 hours of the Victorian campaign.

If the latest polls are right, yesterday's token post about Lower House modelling might not be quite so token after all. What we've seen in the Herald Sun's YouGov-Galaxy (53-47 to Labor) and Fairfax's ReachTEL (54-46 to Labor) suggests that the net effect of the last few weeks of campaigning has been more or less zero.  The polling would not have to be wildly wrong for an unlucky distribution of seats to leave Labor short of a majority, but it would have to be very wrong indeed for the Coalition to win the election in any way.

In response to this, Liberal sources have been citing precedents including South Australia and Tasmania for the polls being wrong.  (They say polls pointed to a Labor win in Tasmania and either a Labor or a Xenophon win in SA).  But the only polls pointing to a Labor win in Tasmania were months old by election day and in the final week all polls pointed to a Liberal majority.  In SA the Xenephon bubble had well and truly burst by the final weeks and the final polls (ordinary as they were) did not point clearly to a Labor win but rather had it too close to call.

Indeed, if there's a risk side in predicting the election from these polls, I feel it's that the Labor result could well be stronger than the mid-53s that is the average of the recent 2PPs.  Looking at the primaries, the Galaxy has both majors on 40, the Greens on 11 and others on 9.  The fragmented detail of the ReachTELs hasn't yet included full figures for the primaries without rounding, but I'm currently estimating that after redistributing "undecided", Labor is on about 40.9, Coalition 37.7, Greens 10.9 and others 10.5.  With a small weighting for polls from three weeks or more ago that are old rope now, and weighting the Galaxy slightly higher than the ReachTEL, I get a primary vote estimate of:

ALP 40.5 (+2.4 since 2014)
L-NP 39.1 (-2.9)
Greens 11.0 (-0.5)
Others 9.5 (+1.0)

The thing about this estimate is that by 2014 election preferences, Labor comes out at about 54.6%, but off polls with released 2PPs that are either 53 or 54.  At the 2014 election Labor got nearly 70% of preferences, but the polls have that share dropping into the low to mid 60s without much change in the percentage of the non-major-party vote that is Green.  There are reasons why this could be so (Labor-Green fights in the inner city, general government wear and tear and also the likelihood that the Others column is hiding One Nationy types who will tend to preference the Coalition), but that doesn't make it certain that it will.

It is also possible the Others vote could be overstated and that One Nation and Christian right voters will rock up at polling booths on Saturday, find their parties not on the ballot and (mostly) vote Liberal.  In 2014 the final polls slightly overstated the Green vote, understated the Coalition vote, and got away with it on a 2PP basis because of the increased preference flow to Labor.  The 2014 Victorian election was notable because every one of the 17 polls taken during the election had the Greens too high, but state polling results for the Greens since have been more accurate, and the worst offenders (Morgan and Ipsos) haven't been in the field for this election.

The ReachTEL requires caution because of aberrant gender breakdowns.  Women were 7 points less likely to support the Greens than men, which some outlets have blamed on the two Greens candidates who have past histories of singing or liking sexually problematic material.  However if you buy this you have to also buy that the Green vote among men has sharply increased at the same time, which seems unlikely.  A bigger problem is that the poll has women 14 points more likely to vote Coalition than men, which is simply impossible, and means that all the gender breakdowns in the poll are rubbish.  Unless figures have been transposed, the poll must have accidentally sampled a very conservative bunch of women and a very progressive bunch of men, and we can only hope these errors cancelled each other out.

2PP Complications

One tricky aspect I hadn't yet included in my seat model for this election is population growth.  On average, population growth in Labor seats has been 6.5% higher than in Coalition seats.  This means that all else being equal,  we should expect the average swing per seat to Labor (assuming there is one) to be smaller than the statewide swing.  I've now included this in my seat model by docking Labor 0.28% in every seat.  The other possibility is that pollsters are not adjusting for this, in which case the adjustment is unnecessary and what should really be expected is that Labor overperforms on the 2PP, or that this issue helps to cancel out preference shifting.  

I am also not adjusting for the fact that Richmond has no Liberal candidate.  This one is a real pain in the neck for modelling.  It's not clear yet whether official figures (and also ABC modelling) will treat Richmond as 100% Labor or as not existing for 2PP purposes.  I'm also not clear what the polls have done, though it's likely ReachTEL would have polled as if the Coalition were running (which is equivalent to a third solution that official sources won't be using: give Richmond an artificial figure based on the state swing.)  If Richmond is treated as 100% Labor then that means Labor overperforms on statewide 2PP by another 0.3% compared to its state swing.  If it is excluded, Labor underperforms by about the same amount.  

All up the effect of the Galaxy and ReachTEL polls was to move my seat model only to about 48 Labor, 36 Coalition, 3 Green and (subjectively) 1 Independent, though I am quite shaky about the numbers for the minors.  The Green vote could be being overstated a bit in polling again and their seat success is wholly dependent on how well they go in the inner city, a wipeout with zero or only one seats being a possibility.  Also I usually find that whichever way I jump on trying to subjectively estimate independent seats is wrong.  I'm very doubtful about Russell Northe holding on in Morwell, and while there are a lot of half-chances elsewhere, none of them have been nourished with even seat polling evidence.  

I will update this article if there is a Newspoll overnight, which I expect there will be.

Newspoll and Morgan

Newspoll has come out with a 53.5% 2PP (final Newspolls are often rounded to half a point instead of a full point) off primaries of Labor 41 Coalition 40 Green 11 Others 8 - almost identical to the Galaxy.  After adding these to my aggregate with the Newspoll weighted most heavily I get Labor 40.7, Coalition 39.4, Green 11, Other 8.9.  My last-election 2PP off these numbers is 54.4%, noting that that may be an overestimate.  For my seat model I am splitting the difference between last-election preferences and those used by the pollsters, and therefore using 53.9 as my estimate.

I am wary of the Greens vote because of what happened in 2014 (as well as in SA this year) but in Queensland 2017 the pollsters included were accurate concerning the Green vote, and this was also so at the 2016 federal election.  So I have resisted the temptation to adjust it slightly.

Morgan also released an SMS poll which provided results for voters who have voted and those yet to vote but curiously didn't amalgamate the two.  "Already voted" had Labor 40 Coalition 35 Green 12 Ind/Other 13 and a 2PP of 53-47 to Labor.  Yet to vote had 39-32-13-16 and a 2PP of 55-45.  So, by averages, we might expect 54-46 to Labor, though with typical Morgan wackiness the poll doesn't bother publishing a headline reading.  As with past Morgan-SMS releases the minor party votes are extremely high compared to others.  Another notable aspect is that the preference flows to Labor here seem remarkably weak, only 52% in the "already voted" and 55% in the yet-to-vote, and I just don't think this is credible (presumably these were respondent-allocated.)  I will be checking how this poll went in my review of the final polls but I have decided not to include it in my aggregate.  It is too long since the last such Morgan SMS release and if I do include them I will end up downweighting and benchmarking them almost out of existence anyway.

Overall, the final poll 2PPs are tightly clustered (a 53, a 54, a 53.5 and a 54-ish) but a similar thing happened in 2014 when many polls had 52-48 which was the correct result.  There are not enough polls to provide strong evidence of herding.

At the moment, I still find it hard to find seats that will change hands. Assuming the polls are about right then there could be a Labor-Green swap of Brunswick for Prahran, a few low-hanging Liberal fruit might fall (Ripon, South Barwon, a few others are shaky), Morwell nobody has the foggiest and there might be the odd indie bolter against either of the majors.  Labor might lose a seat to the Coalition but it's hard to say where (I should mention Albert Park as a seat worth checking in on because of a theory it's been moving to the Liberals demographically.)

Seat Betting

A final roundup of seat betting, which is not necessarily predictive.

Expected ALP gain from Coalition (close): South Barwon

Split market - possible ALP gains from Coalition: Bass, Ripon

Close Coalition holds - Burwood,  Eildon, Mildura (vs Ind), Ovens Valley (vs Ind)

Close Labor holds - Carrum, Frankston, Eltham, Pascoe Vale (vs Ind)

Expected Green gain from ALP (close): Brunswick

Split market - possible Green gain from ALP: Richmond

Close Green hold vs ALP: Melbourne, Northcote

3-cornered Green held no clear favourite: (Coalition odds-on favourite on two markets): Prahran

3-cornered IND held no clear favourite: Morwell

Four seats are expected to change hands (though in two cases, unclear to whom) and another three have split markets.

[Update: as of midday election day the two markets still open both have Labor winning Bass and Ripon.]

Live coverage here which will start from 6 pm unless there is important news earlier.  Note that sorting of prepolls is starting at 4 pm and there may be scrutineering samples reported from this process although the VEC is impounding mobile phones during it.

No comments:

Post a Comment