Tuesday, June 7, 2016

Poll Roundup And Seat Betting Watch: Fear Of A Hung Parliament Edition

2PP Aggregate: 50.2 to Labor (unchanged)
Seat projection for this 2PP: probable narrow Coalition win, not necessarily with majority (estimate 76-69-5)

As usual, this roundup is quite long so feel free to just read whichever bits interest you.

It may as well be a recorded message: for the eighth (!) week in a row national 2PP voting intention has been around 50:50 and there has been no significant movement in the 2PP at all.  Excitable noises are made about moves of 1-2 points this way or the other in this poll or the other (typically by the media those polls are associated with) but it is all meaningless babble as nothing has actually happened.

This week Newspoll came out at 50:50 following four weeks of 51:49 to Labor.  Ipsos, which has tended to lean slightly to the Coalition but is bouncy because of its smaller sample size than other pollsters (and lack of artificial bounce-retardants, I suspect) raised the odd eyebrow with a 51:49 to Labor (by both kinds of preferences).  Essential went to 50:50 after being 51:49 to Coalition last week and ReachTEL went to 50:50 as well.  The two-point move in ReachTEL was mostly caused by volatility in their respondent-preference sampling; by last-election preferences the Coalition improved by only 0.7 points.

I have removed the house effect assumption for Essential again as its mostly pro-ALP skew since Turnbull became PM appears to have at least dropped below a point in recent readings, if not gone away.  However I've also increased the "global house effect", which adjusts for there being more apparently ALP-friendly polls in the mix than Coalition-friendly ones.  The combined impact of these two decisions was zero. After considering the primaries I entered the Newspoll and Essential both as 50.2 to Labor, the Ipsos as 51.0 to Labor and the ReachTEL at 50.1 to Coalition.  The current aggregate (noting that this doesn't just include this week's polls) is 50.2 to Labor, which was also the end-of-week result for both the previous two weeks:

BludgerTrack is on 50.1 for Labor, Phantom Trend 50.2 (pre-Essential), Luke Mansillo 50.37 Mark the Ballot 49.3 (using house effects anchored to last election), and I'll add any others I come across.

From next week I switch on the special weightings for the newest polls released in the final weeks of the campaign.  If there is change then all this 50-50 stuff should be downweighted pretty quickly.

How Little We Know

In terms of where it might go from here, I track two historic projections, one of which ignores which party is in government and the other of which includes a party in government term.  The simple version expects the 2PP to on average go nowhere and gives the Coalition a 63% chance of winning.  The advanced version expects the Coalition to gain (on average) half a point and gives it a 71% chance of winning.  But what is most notable is that these projections still - only four weeks out - explain only 40% and 44% of variation in past 2PP results respectively, and the average error on these projections is large (1.9 and 1.7 points respectively).

What's more, the errors on these projections are not even normally distributed - about 40% of elections see blowouts of at least 2 points and in cases up to 4 points from the projection, and it is also not the case (as might be thought) that the poll-based projections work better when the polls are close.  I even tried taking the old pre-1987 Morgan results out of the regressions to see if using more recent elections with richer data made for better predictiveness.  Predictiveness got worse.

At this stage in 1993 the Keating government was being smashed at below 47% 2PP (it won).  The Fraser government in 1977 was narrowly winning (it won massively).  The Hawke government in 1987 and 1990 was winning comfortably (both those elections were close in 2PP terms).  The Howard government in 2004 was struggling (it increased its majority) and so on.

Four weeks out, all the national polls are saying in the light of history is the following:

1. The election will very probably not be a massive 2PP win (54+ for Coalition or 53+ for Labor)
2. The Coalition has the better chances because Labor is not far enough ahead.

And that is all.  And because each election has unique properties this could always be the one - though there's no reason yet to believe it is - that does something right off the scale.

There's a view in some quarters that if there is to be a blowout it will be to the incumbents.  There are a few ways this could happen.  Firstly, through the campaign Labor has been making far greater spending promises, as Oppositions often do, but has not explained how it will pay for them.  Suppose that when that explanation shows up, it is not convincing (either voters don't believe it, or they don't like how it will be achieved).  [UPDATE: It seems they're going to pay for them by blowing out the deficit.  That seems like easy meat for scare campaigns.]

Secondly, suppose that polling continues to point towards a major risk of a hung parliament, but with the Coalition best placed to capture a majority.  Then the argument becomes that only the Coalition can deliver a stable government (give or take the odd PM or two) and anything else will be 2010-3 all over again.  This argument works especially by analogy with the UK election, in which the choice was between a Conservative government (at worst a continuation of the coalition with the Liberal Democrats) or a Labour government that might see the whole UK in thrall to the Scottish Nationalists.  For Australia, replace "Scottish Nationalists" with "Nick Xenophon Team", ie South Australia, or just the Greens as always.

I don't think this argument is actually all that strong.  Unlike in the UK, Labor can be argued based on current polling to have a real chance of majority government, since the number of crossbenchers tends to be small here.  But the issue is not whether the argument is actually strong; it is whether it will resonate and see voters flocking back to the government in the final weeks in fear of what a mess might do to the economy.

Anyway, the total minor party vote continues to creep up in this week's polling, with Newspoll recording it at 25% (the 15% Others including 3 points for NXT, an optimistic 3 for Family First, and 1 each for One Nation and - ho ho - Palmer United), Essential having it at 24% (NXT 4%), ReachTEL at 23.7% (NXT 5%) and Ipsos at 23% (with 19% for Others in SA/NT combined).  On this basis, as foreshadowed last week, my aggregate is giving one Coalition seat to NXT as a "nowcast" reading of current polling.  Whether they'll still be positioned to take that seat (with Mayo the leading candidate) by polling day remains to be seen.

The big question is whether these high third-party readings are real.  They may be just a bubble (with voters saying they will vote for third parties as a coded way of saying that the campaign is boring their socks off), or it could be that the great malady of modern polling, the over-capturing of politically engaged voters, has infected others besides Morgan.

There was a possible hint on the latter in this week's Essential questions, in which 22% of respondents said they had "watched one of the party leader debates on TV".  (Especially note the "on TV" bit - this should, in theory, exclude watching bits of it later on social media.) I assume nearly all respondents who said yes to this question watched the ABC debate, but TV ratings published by Crikey gave it an audience share of only 880,000 (less than 7% of voters who will actually vote).  I don't know how the audience ratings deal with time-shifting or multiple viewers in one household. But in any case it seems that either the poll respondents are more likely to be highly engaged compared to the population at large, or else some are lying ... to a computer!


This week's Newspoll saw Malcolm Turnbull at a new lowest netsat, by a point, of -14 (37-51) while Bill Shorten somewhat surprisingly shed all the seven points he gained last fortnight to be back at -19 (33-52).  Nothing to see in the better-PM department with Turnbull holding a normal lead for an incumbent (45-30).  Ipsos had Turnbull in the red for the first time (44-46) with Shorten also down a touch at -8 (42-50) and a rather large preferred-PM lead for Turnbull (49-31). ReachTEL has only released its better-PM figures, which show an insignificant move to Turnbull (who has 55.6% by forced-choice).

ReachTEL also found that Shorten was perceived, by a narrow margin (51.9:48.1) to have campaigned better than Turnbull so far.  This margin was mainly driven by Green voters, since Labor voters were, oddly, three times more likely than Coalition voters to think their leader had been outcampaigned.

Essential found that the Coalition's trust to handle issues is down an average of about four points in its areas of dominance, but if anything improving slightly in Labor's.   There is also the usual finding that voters think the Coalition is better at handling the economy for businesses large and small, that Labor is better at handling it for "working families", but that the Coalition is best at handling it overall.

Seat polls

ReachTEL released a seat poll of Longman (Wyatt Roy, LNP, 6.9%) which had the seat at 50:50 by respondent preferences, though it appears this was about 50.4 to LNP prior to rounding.  By 2013 preferences it would probably come out as about 51.2 to Roy, but it's hard to place much value on this when PUP polled 12.8% in 2013.

That was the only "public seat poll", but there were several commissioned polls.  For my view on the GetUp! Bass ReachTEL see my Tasmanian seats guide.

The GetUp! ReachTEL figures for Mayo and Sturt were reported without including voters who are initially "undecided" but leaning to a party.  (This is a big problem in making sense of a lot of the commissioned ReachTELs and is leading to a lot of confused reporting).  Once the undecideds are included (figures here) the primaries for Mayo are Liberal 40.5 ALP 19.5 Green 10.8 NXT 25.4 Other 4.2.  For Sturt it's Liberal 43.5 ALP 21.5 Green 9.2 NXT 22.4 Other 3.4.  On these exact numbers NXT would win Mayo (with only a 73% preference flow needed from ALP and Greens voters) but Sturt would be harder, with an 81% flow required assuming the NXT candidate could stay in second place (which in turn is highly doubtful).  Seat-polling of these sorts of seats would be unreliable at the best of times, let alone when commissioned by an activist group, but these results are not greatly surprising given the general evidence concerning the NXT vote.  I'm surprised though that GetUp! seem to have missed the opportunity to name the Mayo candidates, as it's quite possible Jamie Briggs' name would now be a negative for his party.

Cowan is an interesting seat because it is well below the waterline in terms of the size of swing on in WA, and yet it took a long time to move on the betting markets.  It may well be that punters were feeling that the ALP candidate (high-profile counter-terrorism expert Dr Anne Aly) might be a risk in the electorate because she is an academic and a Muslim, although she is so obviously well credentialled to make evidence-based contributions on a key political issue.

There was also a "leak" of alleged internal ALP polling said to show Labor behind 55:45 in Lindsay (LNP 3%) and only up 53:47 in Werriwa (ALP 6.5% but with the sitting member retiring.)  Grim stuff if true, but gloomy internal polling often surfaces before elections.

Finally the most derided commissioned seat-poll of the week was one sponsored by Malcolm Turnbull's Labor opponent in Wentworth, said to show a 10-point swing against the PM on both primary and 2PP votes, and much general disappointment with Turnbull's performance (I haven't seen the text of this poll's questions yet).  But he might be onto something here, because it's not so rare for party leaders to poll badly mid-campaign (some but not all non-commissioned polls had Kevin Rudd losing Griffith last time).  A leader's electorate can often feel neglected if their MP is spending all his time gallivanting round the country and is not focused on his own seat.  As for the disappointment, Turnbull had been selling his supposed progressive soul little piece by little piece for months before his elevation in order to secure the leadership (it made me think of The Picture Of Dorian Gray).  Anyone who didn't notice this and was expecting him to emerge as a proxy for GetUp! just wasn't paying attention.  It does still seem that there were many such people.

This could well be a very good election to assess the accuracy of commissioned polling compared to neutral polls, since commissioned seat polls are so common at it (mainly because robopolling is cheaper than commissioned polling used to be).

Amusing Poll Of The Week

I haven't seen the full wordings and results of this one yet, but this article re an Australia Institute poll on Senate crossbenchers is hilarious.  Apparently 33% of respondents said they wanted more Senate crossbenchers (whether this includes Greens is not clear), 28% said they wanted fewer and the rest didn't know.  That's not the funny bit. A remarkably small 52% of respondents said it was normal for governments to require crossbench support to pass legislation, while 24% thought it wasn't.  A small hint here: of course it's normal; only 2 of the last 20 parliaments have had majority Senates.

The point being missed here is that it's normal for governments to have to deal with a small number of balance-of-power forces elected by the will of the people (historically the Greens, Democrats or DLP or the odd state-based independent), not a random gaggle of unstable and unaccountable one-person-band beneficiaries of preference-trading and ballot-flooding.  If voters want lots of crossbenchers and now go forth and elect them fairly then that will be fair enough.  But the irrelevance of the question and the cluelessness of the answers still isn't the funny bit.

No, the funny bit is Ben Oquist's comments:

"Given that the constitution created the Senate to be a check on prime ministerial power, the frustration of prime ministers shouldn't really come as a surprise."
"The Senate is there to protect the public from policy ideas that come as a surprise,"
"The constitution makes it difficult to pass laws. This is not chaos, it is stability."

Spoken like a true 19th century conservative!  One can hardly imagine TAI wheeling out this classic Victorian (the time, not the state) defence of obstructive upper houses when the left next finds itself in power and having, say, effective action on climate change blocked by minor-league righties.  (Which is less likely to happen under the new system anyway, but I digress).  But what's really funny is to have TAI arguing that the public should be protected from surprising policies, when most of the time its business model consists of putting surprising policies to the people in silly pony-polls and breathlessly reporting 73% "support" for ideas scarcely anyone has heard of, let alone thought through.

Seat Betting

Myriam Robin's article in Crikey (paywalled but you can get free trial subscriptions) had some very useful comments about how seat betting markets are framed.  According to the article the seat odds are all along (not just at first) set subjectively by an employee who is guided by how the punters are betting, but not bound by it.  What we're seeing is not just the undistilled wisdom of betting crowds.

(In the eternal and invariably confused battle of betting odds vs polls, I'll point out that the reason why the markets more strongly expect a Coalition win than poll respondents - when the latter are asked a "who will win" type question - is not the closeness of the polls but partisanship.   Voters with strong leanings to a party are more likely to believe their party will win.)

At present these are the seats considered on the Sportsbet and Crownbet markets to be changing hands or else reasonably close.  If a seat is tied on one market I use the other.  In cases of conflict I call it a tie, but there are currently none of those.  If a seat is close on either market (usually both parties inside $3) I class it as a close seat.

Loss (Coalition to Labor): Barton*, Paterson*, Petrie
Close Loss (Coalition to Labor): Solomon, Dobell*,  Page, Eden-Monaro, Macarthur,  Capricornia, Hindmarsh, Burt, Hasluck, Cowan
Loss (PUP to Coalition): Fairfax
Close Loss (ALP to Green): Batman

ALP Close Holds: (none)
Coalition Close Holds: Banks, Reid,  Gilmore, Lindsay, Robertson, Macquarie, New England (vs IND), Bonner, Brisbane, Forde,  Dickson, Leichhardt, Lyons, Braddon, Deakin, Corangamite, Dunkley, La Trobe, Stirling, Swan

Here's the colour-coded tracker.  A pale shade indicates the seat is tied in one market, and grey indicates an overall tie (there are none at present).

This week saw the seat betting markets flip two seats for the first time - Cowan (for reasons that are entirely obvious) and Page (perhaps someone knows something I don't).  Overall these markets are about as pessimistic for the Coalition as they were four weeks ago.  If you want a detailed drilldown for all seats since May,

The case of Petrie is interesting.  The markets give Labor a 25% chance here, but my seat model (which is blind to local factors unless they are captured in a seat poll) has it at 40% and the current edition of BludgerTrack has the LNP losing only one Queensland seat.

There hasn't been any notable change in the expectations of overall totals markets (as noted last time) so it's worth rounding this week's efforts out with a look at a few other market offerings.  Markets on the number of Lower House seats that NXT will win have "none" favourite in the $1.80s range, which comes out to a bit below 50% implied probability, or it may as well be 50-50 if adjusting for longshot bias.  Sportsbet has Pauline Hanson at an implied 58% chance of election to the Senate, while Derryn Hinch has drifted from $2 to $3.  A market on the number of Senate crossbenchers including Greens has 16 as the median number (compared to the current 18), while the most expected number for the Lower House is 5.

By next week we'll have seen a major milestone with the close and release of nominations.  Nominations close at midday on Thursday so we'll soon know who is running for what and indeed whether PUP are fielding any Lower House candidates at all! I will be busy through the long weekend so further coverage may have to wait til this time next week.  As well as the usual roundup, at some stage next week I expect to post a guide for the Tasmanian Senate.


  1. Hi Kevin,

    Excellent post. One correction though, the seat of Page (NSW) is held by the Nats (not the Libs), as per your table.

  2. Any commentary on Reid, it seems there are lotsa people angry with the NSW Baird state gov over Westconnex and/ or council amalgamations, not that the other side's rep has gotten rid of that ICAC smell as yet, will this carry over into the fed election vote?

    1. Some general comments on Reid - although it is on a 3.4% margin the Liberal incumbent has the benefit of sophomore effect in most of the seat (and the remaining parts were transferred in from Labor electorates, meaning Labor loses personal vote share there) meaning that the real margin is more like 5%. There was a Galaxy seat poll of Reid in mid-May showing 51-49 to Liberal, which is a couple of points closer than the margin and the state swing suggest, but not enough to be much evidence of anything.

      There is quite a difference between the odds in my seat model (which give the incumbent a 77% chance of retaining but don't take into account any local factors unless they are reflected in a poll) and the odds from the bookies (which have it at only 60%) - so I suspect the bookies are taking into account the issues you mention. State issues can have some impact on federal campaigns - as we are currently seeing to a far greater degree in WA.

  3. It's interesting to hear the discussion about the accuracy of these commissioned seat robopolls. I've been wondering about their reliability ever since I got polled by the Labor campaign in Swan even though I'm enrolled in Curtin a while ago...

  4. There have been a number of reports of this happening on social media. ReachTEL tweeted in response to this: "Here's some context. In the last 72 hours we have made >2 million phone calls. The error rate is well less 1%."

  5. The recent poll in Grey in South Australia is an interesting one showing the NXT with a real chance. What is the highest number of lower house seats NXT could realistically take from the Coalition on July 2?

  6. Grey is actually the worst SA seat for NXT in my NXT seat model so if they can win Grey they can win anywhere in SA. That's a pretty big "if" though just based on a single seat poll, and I'd expect that even if it was on, the Coalition would manage to put out a few of the fires before election day. It's been noticed that that Grey seat poll has a Lib vs ALP 2PP swing of 10+ points, which seems a bit surprising.

  7. It seems Malcolm Turnbull believes in the idea of fear of a minority government. He trotted out the "a stable Coalition government versus a Labor-Green-Independent shambles like the Gillard days" argument when rejecting the deal with Greens.

  8. Interested in your thoughts Kevin as to what percentage of voters follow their party recommendation for preferences? Is there any pattern as to what sort of seats have a greater number of voters who stray from their party how to vote card

  9. The proportion for Reps HTV cards is high for major parties (as a very rough guide around half, you'll probably find more accurate estimates on Antony Green's site somewhere) but for Greens voters it is so small that a Greens preferencing decision typically makes 5% difference to the flow of preferences from the Greens if that (eg a 80% flow for no card might become 85% with card). For small parties it depends on the amount of organisation of the party, but my impression is that the flow is not strong anyway. If a party has no booth presence then it may have an HTV card online but hardly anyone will follow it.

    Voters are more likely to stray when a preferencing suggestion doesn't make political sense (eg Labor preferencing the Liberals ahead of Wilkie in Denison last time). I haven't looked at variation in card follow rates by seat more generally. I would expect follow rates to be higher in areas with low proportions of English as a first language, and lower in areas with high average education (for example).