Final Poll And Campaign Roundup: Was It Really So Bad Now?

2PP aggregate smoothed tracking graph. Probably final

EXPECTED FINAL 2PP Aggregate: 50.8 to Coalition
CAUTION: Final polls appear herded
Seat Projection if this is the actual 2PP: Probable Coalition majority win (approx 79-65-6)
Subjective prediction 78-65-7 (plus or minus lots!)
NXT seat outcome in SA is unpredictable - could win none or lots



Newspoll: We Have To Talk About Herding

The final Newspoll has come out at 50.5 to Coalition, off primaries of 42-35-10-13.  The expected 2PP for these primaries is exactly 50.5 so there is no evidence of rounding in either direction.  Those hoping for a blowout in the Coalition's favour might take some heart from the new Newspoll having skewed to Labor by close to a point in my aggregate since Malcolm Turnbull became PM.

However all the leans and skews of various polls through the term (save for Ipsos' skew to the Greens of course) appear to have magically disappeared leaving us with the most blatantly herded-looking set of final 2PPs at a federal election in at least the last 20 years.  Six polls have landed with headline 2PPs in a one-point range - one 50, two 50.5s and three 51s.  The last 20 polls by all pollsters have had headlines in the range 49-51.  There also appeared to be herding in the Victorian state election where a result of 52-48 to Labor was produced by pollsters an extremely large number of times, but it turned out to be absolutely right.   An apperance of herding can happen by chance, but this appearance appears more times than chance would suggest!

When herding happens  no particular pollster looks bad by having released a final-poll shocker, but there is an increased chance that everyone is wrong.  However, there's still a good chance they're all right. It doesn't especially feel like they are wrong - there are no vibes coming from the parties of a seat outcome radically different to what a 50-point-something Coalition 2PP implies.  You can argue Labor's campaign was a shambles and they "deserve" to lose by more, you can argue that Labor successfully raised major questions about health policy and that they might do better.  We don't have a well-developed science of fundamentals-based prediction in this country (and I'm not sure it really exists anywhere.)  So maybe they're just right and the Australian polling miracle continues.  Without much caring personally who wins, a thumping Coalition win or any kind of Labor win would be a bad look for polling after such a set of final figures.

Malcolm Turnbull records his best satisfaction rating since early March but his netsat of -7 (40-47) is still nothing to write home about, and nor is Bill Shorten's -15 (35-51).  Turnbull has stretched his better-PM lead to 48-31

I believe there are no more polls and so my final aggregate, if that is right, is 50.8 to Coalition.

Seat Projection and Predictions - final comments

My final polling-based seat projection barring startling data in the next 24 hours is 79 Coalition 65 Labor 6 Others.  The 79 is rounded up from 78.501.  The projection differs from a prediction because there are some seats that simply cannot be forecast by the methods I've been using because of the lack of objective data.  At the 2013 election the polls were right.  This projection assumes the polls are more or less right again.  If the polls are seriously wrong, the projection off them won't work either, and the search will be on for other ways to predict elections in this country!

To give a notional one-point error bar on the projection (assuming errors in the 2PP are caused by sample error only), for a 2PP of 49.8 for Coalition the model gives 74-70 as the major party seats.  For a 2PP of 51.8 for Coalition the model gives 83-61 but I would expect fewer if any NXT wins from the Coalition in that case, so maybe more like 85 Coalition seats.

In terms of Coalition vs Labor seats, I use a base of 92 for the Coalition (including the three notionally Labor seats it occupies, Fairfax vacated by Palmer, and the notionally Liberal seat of Burt).

My seat model expects the following losses:

Barton
Paterson
Petrie
Capricornia
Solomon

- these five have been losses at all stages, with no polling evidence they'll be retained, though it's a bad sign of how shaky things are for Labor that the retention of three of them has been seriously canvassed at stages during the campaign

Dobell

- this is below the waterline of the expected swing though a 50-50 seatpoll means that Labor are only narrow favourites. Again here there is serious doubt.

Lyons

- this is on a slim margin and had a very adverse seatpoll, though it's always possible the seatpoll could be wrong.  I believe this seat is way too short in betting based on the objective evidence but that Labor should still win it.

Burt

- this is notionally Liberal with a large margin but is a new seat made of bits of seats where the Liberals lose personal votes, including of the late Don Randall.  The ALP candidate ran in the Canning by-election and the result is buttressed by a seat-poll.

My seat model expects there to be five more Coalition 2PP seat losses (in seats where it has the Coalition favourite) and one by Labor the other way.  However when Labor seats are examined on a seat-by-seat basis there is no seat at obviously high objective risk of falling - just a long list of maybe-but-probably-nots.  (There are many more of them on the Coalition side).

The model has the Coalition narrowly favourites (between 50 and 70% based on the 2PP and expected seat swing) in Braddon, Banks, Eden-Monaro, Macarthur, Brisbane, Forde and Cowan.  The latter two are practically 50-50, though I don't think Forde is considered to be at that much risk.  Page would be included in this list if I included a feature to damp sophomore effect when a former member runs against the person who defeated them.  This will be included next time!

Of these "the vibe" is that Eden-Monaro is about to lose its bellwether status and I agree with this.  It was also a heavy loss in commissioned seat polls and perception is the local member is ordinary.  Page has also been a loss in two commissioned polls and nobody has bothered polling it neutrally either.  I am inclined to override both of these.

Macarthur is a favourite in seat-betting but is clearly very close, and I find the objective evidence available to me favours a retain (just), and I don't know any better.  So let's not override this one, and see what happens.

Cowan has been the most interesting 2PP contest.  In a rare case of me caring about the election, I am hoping Dr Aly wins because her expertise would be an asset to the parliament.  I am not convinced there's any objective reason to override my model though, other than to note that it is very close.

I cannot identify any other seats to override, so there is at least a two-seat difference between my model's total and the number of seats I think are Coalition wins.  This may be a smoke signal that the Coalition has done better than projected in shoring up the marginals, and that a 2PP of 50.8 could be good (as Metapoll projects) for more than 80 seats - unless there are too many crossbench raids.  Or more likely it is just that the Coalition has more seats in the vague-risk category and that some of those will fall. As well as those mentioned above, there are the usual crowd of realistic chances.

Onto the non-classic seats.  I believe NXT will win Mayo as shown by all recent polls, and based on the intense effort on the seat, the seat's history of third-party challenges, and the scandal that caused Jamie Briggs to leave the ministry.  I am not super-confident of that though as the polling is old and Rebekha Sharkie has had the odd scrape on the campaign trail herself.

I cannot say with any confidence where the NXT surge will land - they have real chances in Grey, Boothby and Barker at least (also Kingston as well, though their candidate there is a liability), and could even win several seats. My limited read of their polling statewide since the Grey poll is they might have slipped a little.  I suppose I should tip Grey based on it having a poll and a clear reason to fall but Barker is actually the one I have stronger modelling evidence on, so both of these will be very interesting.

Batman is at major risk from the Greens.  Ten points seems a huge swing but David Feeney has had an awful campaign and inner-city intensification and seat-targeting is something the Greens are good at.  The objective evidence is not quite there, but subjectively I think it might just go.

I believe the Nats have done over Tony Windsor in New England with one of the few creatively clever attack-ads of this campaign plus some very old dirt, and that his campaign has also had a few too many speedbumps, so I'm picking Joyce to win.  I am much less confident re Cowper where Oakeshott's opponent lacks the status of Joyce and the Nats have been caught with their pants down by a sneaky challenge.  It still seems to me an extraordinary proposition that Oakeshott would return (which has been a difficult prior for me to move off), and hard to tell if the betting dive is based on evidence or just a run.  I think I have to wimp out and say "who knows".

In the roughie department, I'm keeping an eye on two other Green targets - Wills (no polling I've seen and the Labor candidate is controversial) and the three-cornered Melbourne Ports (though I don't think Greens-commisioned polls are convincing evidence that Michael Danby will be pushed into third) and Higgins (not falling in reputable polling).  It has been assumed Adam Bandt in Melbourne is completely safe but that's another one we've seen no polling to confirm, and I have heard the odd not especially credible muttering about it.  Oh and Menzies is vaguely interesting - Stephen Mayne presumably will not get near beating Kevin Andrews, but someone has to some day, for the good of the Liberal Party and the nation.  The circus running against Abbott in Warringah will mostly take votes off each other and I don't see anything changing there.

So while my seat model expects 79-65-6, I can see a case for throwing, say, one from the Coalition to the crossbench, hence my "subjective prediction" 78-65-7.  On the other hand there are no solid and clear crossbench gains, so if NXT fall over we might be left with just the present four.

There are many more competitive seats than last time, and there will be some surprises.

======================================================================

The Campaign

It's almost over but the voting and the counting, and the counting is going to take a while.  A very long while in the case of the Senate, but for the Reps we usually more or less know the result on polling night. If we don't, that will probably mean something interesting is happening.

It's obviously been a long campaign - even longer than its formal length since the election date was a foregone conclusion before the campaign formally commenced.  Many people have found it boring, and I partly agree and partly don't.  The policy debates have often been dry and bland, and the leaders less than entertaining, but haven't people said they want more policy, less trivia?

This hasn't been the "campaign about nothing", for that was 2010, a mindless political celebrity-fest in which both leaders strove to efface their personalities as much as possible and endlessly repeat meaningless slogans.  The 2010 campaign became preoccupied with its own mechanics and all kinds of mindless trivia and was easily the worst federal campaign I have seen.  At least this time the leaders have tried to flesh out how their three-worders connect to their policy priorities, and on the whole done a half-decent job of it.

At least this time we've also seen the ALP reject both the "small-target" Opposition strategy that left voters with no actual reason to switch to Labor in 2001 (so they didn't) and, mostly, the "selective differentiation" rebadge that worked fine in 2007 (but only works if you are winning).  We've seen Labor refreshingly take on the issue of negative gearing, which was thought to be untouchable, without any obvious harm, and generally be willing to take risks to present something more authentic to the voters.  We've seen Labor offer spirited opposition and debate.

On the downside, if Labor lose, a reliance on implausible scare-campaigns and self-declared gotchas, a pitching of the campaign stridently to the left (to compete with the Greens or because they had read too many Australia Institute polls?) and at times a failure to even take simple economic competence questions seriously, might all be marked down as factors.  Labor have also made hardly any attempt to counter the Coalition's claim that only they can win a majority and everything else is chaos, although it would not have been so hard to do.

But let's not call the result certain yet, because the polling is still close, and could be wrong, and indeed there is a responsibility for media and analysts to avoid overconfident calls lest they impact on voting behaviour.  Not that I got any feeling from Bill Shorten's dispirited interview on 7:30 that he thought there was any hope left at all after a bad final week for Labor, but maybe he was totally exhausted.  In search of reason behind Labor's methods, I wonder if they feel they've tapped into a similar vein of left-wing enthusiasm as propelled Jeremy Corbyn and Bernie Sanders, but then I remember that Australia has compulsory voting, which makes such an idea double-edged.  My own belief is that the Coalition will win a majority somehow, and maybe even a comfortable one, but the objective evidence says that this is no sure thing.

On the Coalition's side, while presenting Malcolm Turnbull as a modern leader for the innovation age doubtless seemed like a good idea given his business experience (at least so long as no-one talked about the NBN) the sales pitch seemed like "everyday voters" who were not all that high-tech would have wondered what he was going on about.  There was even a risk that all this talk of rapid change would frighten voters (who according to ALP feedback took it as implying they would lose their jobs).  So out came the infamous and ridiculed Tradie ad, which just seemed like a misfire to the majority of election-watchers, but I think that it was actually defensive.  If grumpy blue-collar voters are wary of losing their jobs in the new economy, scare them back with the idea that they'll actually lose their jobs under Labor.

In the last week the campaign has often been distracted by debates about a plebiscite on same-sex marriage.  Tony Abbott - whose political existence these days seems to have little purpose other than an endless search for veneration of his dubious legacy - rather cheekily claimed this was because the Coalition hadn't been banging on about borders and "budget repair".  In reality the distraction happened because the Coalition has people like Tony Abbott in it, and these people put their religious prejudices ahead of passing a reform that most voters support and that has done no harm at all to conservative governments that have passed it elsewhere.  However, evidence repeatedly finds that same-sex marriage has almost no "salience" - it influences a small proportion of voters whose votes are often not mobile anyway.   Some voters would be more frustrated with the media and Labor raising the issue than with the Coalition's failure to resolve it.

The Coalition has also been hamstrung by its messy past and by the inflated and soon disappointed expectations that Turnbull would drag the Liberal Party further back into the centre.  If we are talking about strong, stable majority government it is clear that the Coalition, with a huge majority, just spent a term providing anything but.  Somehow though, I suspect that enough voters swallow the line that under Turnbull it will all be different.

Bill Shorten claimed an alleged gaffe by PM Turnbull was the defining moment of the campaign.  This smacked of desperation and the media didn't buy it and generally declared the statement had been taken out of context.  If Turnbull wins, even narrowly, then my nomination for the defining moment is Turnbull singing alongside volunteer firefighters in Victoria.  Like John Howard addressing cheering loggers in Tasmania in 2004, it was a case of a PM in an election campaign appearing in solidarity with an unexpected crowd.  Unless the seat polls there are very wrong, the CFA controversy has badly dampened Labor hopes of seat gains in parts of Victoria, and placed a few seats there at risk.

On to those polls ....

Final Ipsos and Galaxy

The final Ipsos was taken from Sunday to Wednesday (26-29) and returned a 2PP of 50:50 by last-election preferences (51-49 to Labor by respondent preferences).  This was from primaries of 40 Coalition, 33 Labor, 13 Green and 14 Other.  The 14 Other has been greeted with the usual scepticism - who will these Others be?  I aggregated Ipsos at 49.8.  The Ipsos writeup in the SMH not only declared the Coalition was facing a "shock loss" but also carried the following:

"The final Fairfax-Ipsos poll of the 2016 election shows that with most Australians preparing to cast their votes Labor and the Coalition are locked in a 50-50 embrace based on second and subsequent preference flows recorded in the 2013 poll when Labor's electoral support had tumbled."

This is a nod to the claim fed by Labor to Barrie Cassidy and since repeated by many lefties on social media, none of them bothering to investigate whether it was true.  It was debunked here weeks ago and yet it still does the rounds as if its life ambition is to be included in an ABC Zombie Fact Check.  The fact is that historically increases in 2PP vote and preference flow have not gone hand in hand; if anything the opposite.  This election might be different, but the past is no valid reason to believe it should be, especially not when the most recent election showed Labor's preference share rising as its vote fell through the floor.

The left had about five minutes of hung-parliament-maybe sunshine before Galaxy came along and spoiled the party with a 51-49 to Coalition, although judging from the 43-36-10-13 primaries, the 2PP was likely rounded up.  I aggregated the Galaxy at 50.7.

With these two included, my aggregate has gone to 50.6% 2PP  and an estimated 78-66-6 seat breakdown.  I stress again that there is significant uncertainty about the quality of polling in this election because of the rapid turnover in pollster lineup and methods, and the consensus of the polls could be wrong.

In all (when the polls are read with personal votes and primaries taken into account) we have the most recent Newspoll, Galaxy and ReachTEL (and final Essential, see below, though their final two-week average points to a Labor minority win) all pointing to a very likely Coalition majority, the Ipsos pointing to either a small Coalition majority or more likely a Coalition minority government with a large crossbench.  The polls with the best past predictive record are all saying the Coalition will win (though the Newspoll methods changes mean it is really just a variant of Galaxy).

The Nick Xenophon Team situation is a great cause of uncertainty.  How many seats NXT get is very sensitive not only to their own vote but also the balance of major party votes.  We have one seat (Mayo) for which all available recent polling has showed an NXT win, but none of it has done so by a big margin.  For other seats the only recent-ish seatpoll evidence of them falling is the 54-46 ReachTEL in Grey (a while ago now and a hard seat to poll) and a commissioned poll of Barker.  On the other hand a poll of Boothby showed them not quite into second but competitive, and various polls of Sturt have all been pretty unconvincing.

Another target seat, Labor-held Port Adelaide, has just now been polled by Galaxy, and while only the Labor-Liberal 2PP of 67-33 has been reported (actually misreported as 76-33) together with Mark Butler's primary of 48, it can be assumed NXT aren't competitive there on that sample.  (Galaxy also found Labor up 53:47 in Adelaide, which has been considered at risk to the Liberals.)

The Coalition has been trying to trash the NXT brand by portraying it as a politically incoherent, candidate-malfunction prone party that is personality-based and will go the way of PUP.  PUP was never much of a thing in South Australia while NXT is an established brand.  Will the attacks succeed?

At the moment I have NXT on an expected two seats, but the scarcity of published data on their vote makes it hard to know if they have peaked and could wipe out, or if they are still in the mix for several seats.  It will make for a fascinating and confusing count!

On other possible crossbench wins the most promising prospect is Batman. I take the Greens' chances here very seriously despite the 10% swing needed and the Liberals still preferencing against them, and am not sure who should be my final pick for this one!

Really Final Essential

Reports that the Essential published earlier this week is the final one were incorrect; they would have been dangerously exposed on those figures and have produced a last-week poll off a sample of just over 1200. (This is actually normal; Essential has often produced a last week sample in the past).  The 2PP is 50.5 to Coalition and the primaries are 42.5-34.5 with 11.5 for Greens, 1.5 for NXT and 10.5 for Others.  I assume some method shift for controlling NXT overstatement has been applied here as is also the case for all the other major polls.  There is rolling state data which provides the information that NXT has an average 21% in SA over the campaign, while the Others vote in Queensland is 15% (consistent with a Hanson threat in the Queensland Senate; Ipsos has the same.)  I aggregated the Essential at face value of 50.5.

Adding this to my aggregate has little impact; it goes to 50.7 to Coalition, and the Coalition picks up an extra seat.

Final ReachTEL

ReachTEL were kidding us; their final poll is yet another 51:49 to Coalition.  I get 50.8 by last-election preferences, providing more evidence that there is little if anything to see in polling on the respondent/last-election preference gap. The poll did nothing whatsoever to my aggregate, as its appearance resulted in the removal of the ReachTEL-before-last, which had been great for the Coalition. Primaries in the new one are 42.8 Coalition, 34.6 Labor, 10.7 Green, 1.5 NXT, 10.5 Others, with movement from Others to both major parties and nothing to see besides that.  The results however are otherwise good for Bill Shorten, who chalks up his best ReachTEL net rating ever (-3.8), is ahead of Turnbull's net rating (-5.4) for the first time, and has closed the better PM gap to 52.9-47.1, the closest so far (bear in mind ReachTEL does not skew to the incumbent.)  Alas for him opposition leader ratings are barely if at all predictive.

As with Essential a question about Brexit finds a very small proportion of voters saying that the issue has affected their vote, but shows that nobody supporting the Coalition has left the fold while a small minority of other party supporters are tempted to stray to the dark side.  The economy and health have strengthened as key issues at the expense of job creation.

ReachTEL also have a welcome seat poll of Chisholm, which has Labor hanging on barely at 51:49, so the seat seems very much in play.

Preferencing

Essential's data on respondent preferences in this sample finds an 80% flow to Labor from the Greens and 56% from Others if undecided preferences are excluded, or 75% and 54% if they are assumed to split equally.  The truth is probably somewhere in between.  The former would give Labor a 0.7 point boost compared to the last-election 2PP and the latter would give Labor nothing.  The previous Essential poll showed a similar range of impacts but with a higher flow from the Greens and smaller from Others.  The Ipsos poll shows a one-point difference, but this is again expected to be higher because of its use of "batched" last-election preferences.  So these data points again support my view that the difference in respondent and last-election preferences may provide Labor with a modest boost  compared to the aggregated 2PP but is unlikely to cause a radical shift.

Herding

The polls look suspiciously herded with everyone's latest headline 2PP now on 50, 50.5 or 51 (although Newspoll and Galaxy are essentially the same poll and both are noticeably static.)  It will be interesting to see if this is still the case post ReachTEL (presumably) and Newspoll tonight.  The last nineteen polls across all pollsters have all produced headline 2PPs in the range 49-51 though two of the ReachTELs would have been 52 by last-election preferences.

Community Engagement poll

The apparently unknown Community Engagement firm has produced a commisisoned national poll with primaries of Coalition 40 Labor 31 Green 11 NXT 2 Others 16 (way too high), to go with some dodgy commissioned foreign aid questions that I lambasted in the case of their Higgins poll.  On the bright side the pollster has done something about overestimation of NXT by only asking the NXT question in South Australia but I can still find no information on what kind of pollster it is.  It also shows Christopher Pyne easily winning Sturt but I don't think we can take too much notice of any findings by a new pollster with inadequate methods details.

Seat Model Output (original section - update at top of page)

Here is some sample current output from my seat model for most of the Coalition seats.  The model aims to find the probability that the Coalition would win the seat if my assumptions about both the national 2PP and the state swing are correct.  The state swings are based on BludgerTrack, which has access to some seat by seat data that I don't, but I have applied some modifications.  One in Tasmania specifically is that I adjust the Tasmanian 2PP by 1.2 points in Labor's favour to avoid double-counting sophomore effects.

The model includes neutral seat polls, and more heavily weights seat polling in a seat if it has been polled repeatedly.  However the weighting on single seat polls is low and there is also a constraint to prevent any systematic skew in the seat polls from swamping the sample.  Including seat polls made little difference in 2013 and if this is the case again I might save myself the trouble in future.  The MediaReach poll of Solomon is not included because it does not meet the due diligence requirement of being clearly neutrally commissioned.  (The word "independent" in media reports is insufficient.)  If it is included the Coalition's chance in the seat drops to a few percent.

The first column is the swing the seat is on, the second column is the adjusted swing after allowing for state effects and personal votes, and the third is the current estimated probability of the Coalition beating Labor in the seat 2PP, given the national 2PP and seats swings entered in.  So the model is not saying that the chance of the Coalition winning the 2PP in Canning is 99.2%; rather it is saying that if the 2PP is 50.6 and I have the WA swing right, then there is just a 0.8% chance of Andrew Hastie having to take up a new career as a mullet hairstyle modeller.

The model is pretty confident about the "usual suspects" (the seats that Labor has been favourite in all through the campaign) - Barton, Paterson, Petrie, Capricornia, Solomon.  That said there has been plenty of talk about Labor not necessarily winning those Queensland seats.  The model is also suddenly gung-ho for Brian Mitchell in Lyons because of a very strong ReachTEL sample, but that sample is hardly game over.  It is leaning to Labor in Dobell and Burt and has Cowan on a neat 50.0% chance.  It has the Coalition very shaky in all of Braddon, Banks, Macarthur, Brisbane, Forde.  I also mentioned before that the model differs from the markets and commissioned polls in Page and Eden-Monaro, mainly because it ignores commissioned polls while the markets may consider them.

Of course local factors in seats that haven't been polled can often mean the output is incorrect for particular seats.  Indeed the model predicts it will get about ten seats wrong even by chance without this.  There's also a possibility that in Victoria the CFA saga might be having a localised impact that brings a small number of Labor seats to much higher risk levels, so the model's projection of one ALP loss to Coalition could be optimistic.

Other Models and Predictions

I'm posting a list of other models and predictions around the traps here as well as adding my own final best guess sometime tomorrow.

The Psephosphere:

BludgerTrack 50.9 and 80-66-4 prior to Ipsos and Galaxy, however BludgerTrack does not attempt to project NXT seats or other extra crossbenchers.

Metapoll 50.8 and 82-63-5 including betting market data (headline rate) but 51.2 (Coalition 41.7 Labor 34.6 Green 10.3 Ind/Other 13.4) without including betting market data.

Alizarin Indigo 79 Coalition seats excluding crossbench losses.  Expects at least Mayo to go to NXT.

Phantom Trend 50.5 and 82-64-4

Mark The Ballot 51.6 and 79-65-6

Guardian Poll of Polls 50.71 2PP (last updated June 26)

The Election Analyst 78-64-8 with NXT winning Mayo, Boothby, Kingston and Sturt.  I have not seen this site before.

Malcolm Mackerras 80 Coalition with some rather disturbing Senate ideas

Nick Casmirri (on Twitter) 75-67-8 on a seat-by-seat basis but "It seems most likely Coalition will get at least 2 or 3 more than that)

Yale Stephens (Video): 77-79 Coalition with most losses in NSW

Media:

Mark Kenny (SMH) 79-81 Coalition Dennis Atkins (Courier-Mail) 90 Coalition  (please report any other forecasts)

Finally for discussion of seat betting predictions see here.