Monday, April 8, 2019

Poll Roundup: Well-Received Budget Yields A Kitten-Sized Bounce (Maybe)

2PP Aggregate: 52.5 to Labor (2016 preferences) (-0.5 in a week)
(51.9 with One Nation adjustment)
Labor would win election "held now", with seat tally around the low 80s

Equal best polling position for Coalition under Morrison so far
Budget reception among the best ever in some regards, but not all
Weak evidence that there may have been a "Budget bounce" of around half a point, but impossible to confirm


The 2019 Budget has been delivered with the election expected to be called sometime in the next week.  This year's budget was in most ways well-received, although the voters have not bought the Government's attacks on the ability of the Opposition to deliver the same thing.  There was remarkably little polling in the leadup to the Budget, so it's hard to say for sure if we have seen that rare creature the Budget bounce or not, but if we have seen it, it's probably only a little one.  We can't yet reliably conclude that the modest move to the government this week is caused by the Budget at all as opposed to other factors.  As it isn't statistically significant, it may even just be random noise.

Newspoll last polled four weeks ago.  Since Newspoll switched to roughly fortnightly polling in 1992, this is only the second time there has been a four-week gap between Newspolls that was not either as a result of holidays or in the wake of a recent election.  The previous such case, in 2011, resulted from avoiding a clash with the Nielsen poll and then avoiding Easter.  This case was caused by avoiding a clash with the NSW election (for which Newspoll was in the field) and then avoiding polling two weeks in a row.  The 2PP of the last Newspoll (46-54) was a little worse than the trend of other polling at that time.  Since the previous poll roundup we had seen just two more polls:

* Essential two weeks ago at 52-48 to Labor, which I aggregated at 51.9.

* A Galaxy at 53-47 to Labor with primaries reported by @GhostWhoVotes as "L/NP 33 ALP 34 GRN 9 ON 8 UAP 3 CON 2 Other 3 Don't Know 8".  On redistributing the Don't Know the major party primaries become more like Coalition 36 Labor 37.  This poll, released with attribute findings that were the subject of much spin (see below) was the first standalone Galaxy since the last election.  It was somewhat oddly released about a week after sampling (March 25-28) finished.  As this poll is clearly using similar preferencing methods to Newspoll, I ended up aggregating it at 53.7 by last-election preferences.  I think the Others vote in this poll is clearly too low, given that Independents alone got almost that much last time.  The United Australia Party (formerly Palmer United Party) vote is of interest because previously the UAP has been polling only 1% generally despite massive spending, but this is the second poll this week to suggest it starting to make modest headway (see seat poll section below).

The voting intentions for the two post-Budget polls are:

* Ipsos 53-47.  The poll has its characteristic skew to the Greens and against Labor in the primaries, the history on a 2PP basis being that these tend to cancel out.  I aggregated this poll at 52.8 to Labor.

* Newspoll 52-48.  The poll showed only a one-point loss for One Nation following revelations that (i) the party attempted to obtain money from US gun lobby interests for its Australian campaigns (ii) its leader flirts with Port Arthur conspiracy kookery among friends then tries to pretend she didn't to the public.  As One Nation voters are much more motivated by disgust with mainstream politics than whether their own party is even remotely functional, I am almost surprised that their vote even went down at all.  Newspoll 2PPs tend to be Coalition-friendly compared to last-election preferences because of their handling of One Nation preferences (mainly). After considering this, but also adjusting my methods slightly because of evidence that the gap is diminishing, I aggregated this poll at 52.5 to Labor.

Essential will have something to say probably tomorrow, but for now my aggregate is at 52.6 to Labor, a 0.5 point narrowing since last week.  Here's the smoothed tracking graph:

The Coalition is definitely doing better this year than it was at any time last year after the removal of Malcolm Turnbull.  It's still doing worse, however, than in the middle of last year in the lead-up to the Super Saturday by-elections.

We're probably six weeks from the election, though it might be seven.  Looking at past elections to see how polling predicts outcomes at exactly this stage is a tricky exercise, because around this time in past cycles there was a lot happening that produced unusual readings - Rudd replaced by Gillard, Gillard replaced by Rudd, September 11. As a result variation in polling at this time only explains 37% of variation in results.  Polling that looks lopsided at this point has usually narrowed, but close polling blew out to lopsided results in 2004 and 2013.  The current projection gives the Coalition just a 13% chance of getting to 50.2% (my estimated win target, though I'll be recalculating that soon) though the chance of the Coalition winning based on that would be a bit higher.

The subjective case that the Coalition has no chance whatsoever comes mostly from it being in a state of constant chaos, infighting and lack of self-control following the removal of Turnbull.  The question then is why is the Coalition even vaguely competitive still.  Are the polls actually getting it wrong (at federal level they generally don't), or are voters actually content enough with the economy despite sluggish wage growth to be not responding very much to the noise out of Canberra?  Is Morrison himself (as a reasonably popular PM at the moment) keeping the ship more afloat than it should be?  Are partisan voters more forgiving of scandal and dysfunction on their own side (as has become the case in the US under Trump, and probably the UK as well)?  At this stage I don't have the answers to these questions.


Newspoll has three Budget questions that it has asked in a stable form for a long time.

On the question of whether the budget is good for the economy, 44% of voters said it was good and 18% said it was bad.  The percentage saying it was "extremely good" (15%) was the highest ever (the previous highest was 11% in 1987 and the long-term average has been just 5%).  This question usually scores a positive net result, but the net result in this case (26) was the highest since the same figure was recorded in 2008.  Higher net figures were recorded in 1986-1988, 1996, 1998, 2004, 2006 and 2007. The all-time high was 49% in 1987.  This indicator doesn't seem to be much of an election predictor as the second and third highest net results (48% in 2007 and 37% in 1996) were both election-year sweeteners that didn't deliver a lasting sugar hit.

On the question of whether the budget was good for voters' own financial circumstances, 9% said it was extremely good (this has never exceeded 5% before) while only 6% said it would be extremely bad for them (this has only been lower than this twice - 5% in 1994 and 2007).  In all 34% said they would be better off and 19% worse off.  Voters tend to be very pessimistic about the impact of the Budget on them personally, so the net rating of 15% was the second highest ever.  Ipsos, however, found the Budget not much different to last year on this front.

As usual there's a lot of detailed analysis of the Budget that hasn't sunk into public consciousness and won't, because the Budget will soon be history and will probably be vastly compromised as the campaign fires up (and if Labor wins it will be revised anyway).  Especially see Greg Jericho's sixth graph here which shows that the usual relationship between unemployment and wages growth has been busted since 2014 yet the Budget expects wages growth to start rapidly returning to the trend line.

Here's how this budget sits on the axis of net economic and personal impact scores:

But there's another score the Budget Newspoll measures on which this one has not done nearly so well, and that's the question of whether the Opposition would have delivered a better Budget.  37% said yes and 45% said no.  This question has a massive skew to incumbent governments, especially because if you consider that the Opposition would have delivered the same budget, you'd probably say no.  A net rating of +8 for the government (no minus yes) is the third weakest by a Coalition government behind only the +7 in 2014 and 2018.  It is also ahead of only five Labor budgets since 1987.

In terms of the expected impact on government polling, the predicted gain for this Budget  (based on my overall ratings, on which it ranks ninth because of the poor "opposition better" rating) is 0.3 2PP points.  That's consistent with what has occurred, as limited as the recent polling data are.  That said, as that article shows, the relationship between Budget ratings and polling relies heavily on the impact of bad Budgets anyway.


Newspoll also saw Scott Morrison recording an encouraging +2 netsat (45-43), his highest since October and equal third-highest so far.  His satisfaction rating of 45% was his equal best so far.  There is normally a strong connection between PM netsats and 2PP, but in Morrison's case there has been no correlation so far, perhaps because he isn't the problem.  Bill Shorten recorded -14 (37-51) and Morrison "led" as Better PM 46-35.  As often noted here, Better PM skews to the incumbent, so after deducting the average 16-point skew, the Coalition is effectively behind (though not as far as would be normal for this 2PP).

Ipsos tends to be more benign for leader ratings than Newspoll and found Morrison with a net rating of +10 (48-38) but wasn't so kind on Shorten (-15, 36-51, so actually no better than Newspoll).  It also found the same

Some News Corp reports made a very big deal of Bill Shorten being judged "untrustworthy" by 34% of voters, a statistically meaningless point above Pauline Hanson and four points more than Morrison, with Miranda Devine even calling it "a fatal character assessment."  "Untrustworthy" is a rarely polled item in Australia, but there has been a lot of polling of "trustworthy". On this front Shorten hasn't been polled for a while but his ratings late last year (33-39%) were in about the same range as Tony Abbott's were at this time in the 2013 leadup.  Given the low general regard for politicians and the fact that one only has to have one's party preferred to the other by around half the voters to win, the idea that a 34% "untrustworthy" score spells game over can only be considered to be clutching at straws.  Especially when we already know that about half the voters don't approve of Shorten for whatever reasons, and that Opposition Leader approval scores have very little relationship with voting intention.

Seat Polls

No reputable seat polls have been sighted in the last four weeks, so we have had to be content with commissioned poll results.

Liberal Party commissioned MediaReach polling of Bass was released with nearly full primary detail  through the northern Tasmanian Nine papers. This mysterious firm has a spotty record in Tasmania, accurately tracking the 2018 state leadup (perhaps with some help from a bandwagon effect) but also producing a wayward narrative of the Braddon by-election.  (Especially the Liberals claimed the poll had shown their attacks on indie Craig Garland had cut his vote in half, and then he actually got more than they had found him to be getting in the first place.)

This poll claims a 50.40-49.60 lead to the Liberals' Bridget Archer off primaries of Liberal 39.01, Labor 38.84, National 1.97 (yes they are running), Greens 4.23, UAP 4.46, PHON 4.36, JLN 5.40, Other 1.49, Don't Know 0.24.  The sample size was only 545, which given the issues with seat polls recently translates to effectively below 100.  The result is vaguely consistent with a weakish sample for Labor in Bass in an EMRS poll but that sample was even smaller.  The most obvious thing wrong with this sample is that the Greens vote is patently too low.  The most interesting thing (in view of Tasmanian Senate implications) is the surprisingly even split between UAP (known to be running candidates in Bass) and PHON and JLN (not known to be running candidates in Bass).  There is no evidence that individual candidates were named.   Bass is on a pretty large margin, but the seat is notorious for turfing incumbents and the 2016 vote was fuelled by a massive GetUp! campaign against incumbent Andrew Nikolic.  The Liberals will be hoping to scare Labor into expending effort on retaining Bass so it has less to throw at endangered Coalition marginals.

Unnnamed Liberal sources made internal polling claims including that they were winning Wentworth and Lindsay (no figures), that Queensland was "messy" (no figures) and that polling in Tony Abbott's Warringah was "diabolically bad".  The 12% swing against Abbott in the report would presumably be on the primary vote (since if it was on 2CP he would be almost retaining.)

Channel Nine put out a rather bad report of internal polling said to have been "leaked".  So-called "leaked" internal polling is nearly always deliberately given away by the commissioning source which is using the outlet to publish the results, so networks claiming to have "leaked" polling should say whose polling it was, or at least explain the circumstances if they don't know (I received a genuinely leaked poll that had been bounced around multiple sources like this once.)  If polling was leaked by disgruntled sources within a party against the party's will, that also needs to be explained.  The inadequate details of the polling that were reported did include that Labor was trailing 45-55 in Lindsay, that Labor was behind on primaries 36.5-48.2 among voters aged 65+ in the seat, and that Labor was also behind in this age group in all of a dozen seats polled in the state.  Trailing by such a margin in Lindsay is bad (though there are good reasons to consider it one of Labor's most endangered seats), but the figures on voters over 65 are actually business as normal.  At the last election, over-65 voters skewed very strongly to the Coalition, and voted quite differently even to voters in their early sixties.

In one of the funniest poll results, a Liberal source was up to no good in telling James Campbell that Pauline Hanson was polling net ratings like -63 in Victorian marginals (Victorians have never liked One Nation anyway) but mischievously slipping in that Peter Dutton had been as low as -50 in some as well.

The Latte Set

In a very interesting and excellent recent poll, the Australia Institute polled voters on drink preferences by party - a poll significant in terms of the tendency of some conservatives to deride left-wingers and particularly greens as latte sippers or chardonnay swillers.  The poll, albeit with the usual sample size issues, did find Greens voters were disproportionately a bit more likely to drink latte and less likely to drink cappucinos and flat whites.  However, most latte drinkers were major party supporters.   (The stereotypes of Green voters worked better when it came to green and herbal teas, or anything with soy milk.)  In terms of alcohol, Coalition voters were disproportionately the most likely to often drink almost anything - especially red wine - except for spirits (One Nation, one suspects usually mixed with soft drink) and light beer (ALP).  (Surely if it gets out that comrades let comrades drink light beer the Labor campaign will be finished!)


Election betting is not reliably predictive, but it's interesting to keep an eye on anyway.  Punters weren't impressed by the lukewarm poll recovery following the Budget, possibly because a big recovery would have been a big signal of chances, and more likely because they overestimated the chances of a big recovery occuring.  So the government blew out from $4.75 to $5.50 on at least one market before cooler heads dragged it back to $5 today.

As for seat betting (see here for methods) these are the current market expectations:

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, Page, Boothby, Dawson, Bonner, La Trobe, Pearce, Swan, Leichhardt, Brisbane, Deakin, Stirling

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

Expected Coalition gain from IND (close): Wentworth
Expected Coalition holds against IND (close): Mallee, Farrer, Warringah
Expected Coalition holds against Shooters (close): Parkes
Expected IND hold vs Coalition (close): Indi (tied in one market)
Expected IND gain from Coalition (close): Cowper

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

The Coalition are currently not favourites in 25 of their own seats, but are favourite in one seat they do not own, so they are favourites in only 51 seats.  The balance of close seats is in their favour but only just, as the market has doubt about a lot of Labor's gains. Relatively few seats have flipped since last time.  Here's the colour coded 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 (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

The seat betting markets are at present expecting the kind of cataclysm that would be expected with a 2PP of around 54.5% to Labor.  This flies in the face of both the present polling and the history of polling usually narrowing as election day gets closer.  Betting on the number of individual seats places Labor on about 84 seats yet even with only very minor adjustments for longshot bias, the seat betting markets collectively think Labor will win more than this.

I will post updated comments (but not update the graphs) for Essential, and apart from that, back next week when we may have an election on our hands!

Morgan update (Monday): Morgan has released (via an email list) a face-to-face poll from April 6-7 with a respondent-allocated 2PP of 52.5% (down 2.5 from an unstated date pre-Budget) and a sample size of only 809.  Primaries were L-NP 37 (+2.5), ALP 35 (-1.5), Green 13.5 (+1). PHON 4 (+0.5), UAP 1.5 (-0.5), Conservatives 1 (-), KAP 0.5 (-), Ind/other 7.5 (-2).  By last-election preferences I get the 2PP at 53.3 to Labor (compared to 55 for the previous one).  Morgan skews to Greens and indies and against right-wing micros, because of social desirability bias.  I use these polls only at a very low weighting because of their accuracy issues and small sample size.

Essential update (Tuesday): Essential has stayed at 52-48 off somewhat interesting primaries of Coalition 38 Labor 35 Greens 11 One Nation 5 others 10.  This moved my aggregate down by 0.1 points to 52.5.


  1. Given the budget itself was so widely liked, doesn't it seem odd that the budget bounce, if it exists, was so small at only half a point?

    Is it more likely to mean that the polls were simply producing results within the margin of error than an actual bounce?

    1. First question: no. Budget bounces are rare and the regression suggests that even the best Budgets (which this wasn't, because of its poor rating on whether the Opposition would do better) should only on average be worth a point or so. The average post-Budget move is -0.4

      Second question: yes the small size of the apparent bounce means it's impossible to say if it is real or random noise. I haven't tried to estimate the probability that it's at least partly real (would be an interesting exercise) but my guess would be about 50%.

    2. It's a tricky statistical question. Do well-received Budgets always move voting intention positively but the effects are often drowned out by noise and/or irrelevant events, or do they only sometimes move voting intention positively and sometimes not? Probably impossible to answer.