Strange Times In Queensland Polling

Galaxy Queensland: 52:48 to ALP based on composite preferences
Based on last election preferences 54:46 to ALP
Result based on this poll if election "held now": Easy Labor win (approx 54 ALP, 32 LNP, 2 KAP, 1 Ind)
However, most other polling so far has suggested little change since 2015 election

Update added for Aug-Sep Newspoll (53:47/55:45)

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

A new state government has won a remarkable victory, ousting the regime that had thrashed it three years earlier after just one term despite having been almost wiped off the political map.  Six months into its term, and despite a significant scandal that has endangered its already fragile hold on power, its primary vote has been polled at two and a half points above its election result, the Opposition is down a similar amount and the most supportive minor party's share of the third-party vote has improved.  Not the greatest honeymoon in polling history but still, all things considered, pretty good?

Well, supposedly not.  According to reporting (?) of a newly released Galaxy of Queensland state voting intention, the Palaszczuk Government has "stalled", "stagnated", "received no bounce from handing down last month's state budget".  Apparently it "hasn't been much of a honeymoon period [..] in stark contrast to the burgeoning support being enjoyed by governments in southern states".  The Premier's popularity has "failed to prompt any new love for Labor".  Her government "could scrape over the line" but "would rely heavily on votes flowing strongly from preferences".  The government apparently should be concerned that it's been "unable to convince more Queenslanders they're a competent administration" and unless it can prove it is doing something then "the patience voters have shown will run out".

What is going on here?  What's going on is what happens when you take the Courier-Mail's oft-noted love of curious poll-spinning and combine it with an understandable, but nonetheless unusual, preferencing practice.  This article looks at how the Palaszczuk government is really going in terms of known public polling.



August Galaxy 

The August Galaxy has primaries of (with changes since the election) ALP 40 (+2.5), LNP 39 (-2.3) Greens 10 (+1.6) KAP 4 (+2.1) Others 7 (-3.9).  The drop in Others makes complete sense since Palmer United, which polled 5.1% at the 2015 state election, has now withdrawn from state politics.

Given that the 2PP at the last election was 51.1 to ALP, and given that there is about a 2.4 point swing on the primary and also a swing to the Greens from Others, it's to be expected that Labor would be up a few points based on last-election preferencing.  My estimate is 54.1% 2PP to ALP by that measure, up three from the election.

However, following the dramatic preference shift seen in 2015, in which Labor suddenly made an unprecedented (under optional preferencing) preference gain of close to 0.3 votes per third-party vote, Galaxy has assumed that that wouldn't happen again.  As a result, Galaxy's preferences are not purely last-election preferences, and appear to be determined using a combination of previous election results.  The April Galaxy was reported as based on preference flows from the 2012 and 2015 elections combined while Poll Bludger suggests the formula may now even include three elections.  If any confirmed information comes to hand on exactly what method Galaxy is now using I will note it here.

It may very well - I think it probably will - turn out that some such multi-election figure is a good method for the next election. Third-party voters will have had three years to try to forget the Newman government, and it's unlikely that the 2018 Queensland election will be held with the Abbott Government polling so awfully federally (assuming it is even still in office).  A "Put The LNP Last" campaign should have a lot less bite.  Moreover, it's impossible to govern Queensland without annoying either the greenies or the north, and so Labor's third-party preference flow is likely to come down from one side or the other, if not both.

However, assuming this kind of preferencing shift so soon after the next election creates severe problems for comparison with other post-election bounces.  Other 2PP figures generated in the months after an election have pretty much always used respondent preferences from that election (once they have become available), and that is the method being used to demonstrate bounces in other states. So to award the Palaszczuk Government only a one-point bounce and compare that unflatteringly to 2PP bounces elsewhere is simply not comparing like with like.

Indeed Galaxy's method describes what the 2PP would be like if voters voted today but distributed their preferences the way the pollster expects them to do in two and a half years' time.  Assuming (as seems likely) Labor's preference share does change by the next election, then this isn't something likely to have happened to the full extent right away.  Rather, the flow would change through the government's term.  The one published poll with respondent-allocated preferences so far suggested only a minor change.  Assuming preferences for the next election now might make sense in terms of comparative tracking in subsequent years, but it cannot describe current voting patterns adequately.  (I should also add that the question of an election "held now" is far from academic in Queensland, given the risks to the government's existence.)

A three-point honeymoon effect for Labor in Queensland is modest compared to Victoria (where the bounce runs at 4-6 points varying by poll), or to the similar improvement for Labor in South Australia, but it is similar to the bounce in polling for the returned Baird government in NSW (2-4 points) while the new Hodgman government in Tasmania got no honeymoon at all.

Also, whatever might be said about future preferencing, to argue that the government merely "could scrape over the line" and "would rely heavily on votes flowing strongly as preferences", as maintained in the Courier-Mail piece, is bunkum even if the 2PP is considered to be only 52:48.  At the 2015 election, the new government received remarkably poor bang for its 2PP buck, failing to win a majority despite polling 51.1% 2PP.  The main reason for this was double sophomore effect - the huge number of seats won by the LNP from Labor in 2012 at which Labor had lost a personal vote and the LNP gained one.  In 2018, that works the other way around, making all Labor's marginals much safer than they appear on paper, while the LNP enjoys only the weaker "single-sophomore" benefit in just one of the nine seats it holds over Labor by less than 2.5%.

The most likely outcome of a 0.9% swing to Labor would be a gain of at three or four LNP seats plus Cook, and majority government with 47-48 seats. And, starting with a primary vote lead based on this poll, Labor wouldn't need a heavy preference flow at all.  Even an even split of third party preferences (something Labor was not reduced to even in the dire 2012 election) would mean a zero 2PP swing, based on which Labor would remain in office and be more likely than not to gain one or two seats based on swing variation.  So, there should be no doubt at all that the correct read of this Galaxy is that Labor is still winning.  Indeed, if the preference flows of the last election are still applicable now, then the Galaxy should be read as pointing to about ten Labor gains, for a result around 54-32-3.

The general run of polling

The general run of polling is not so positive for the new regime.  The following table shows primaries, published 2PPs and my estimate of last-election 2PPs for all polls published for Queensland since the 2015 state election (click for larger version):

(Election result was ALP 37.5 LNP 41.3 Green 8.4 KAP 1.9 PUP 5.1 Others 5.7.  ReachTEL "Other" includes KAP.)

By my estimate of 2015 preferences, the current Galaxy comes out as the government's best poll so far, and the odd thing is that the Courier-Mail's comments are closer to accurate (but still not very close) concerning the rest of the polling. The remaining polls to date have been dominated by Morgan SMS polls, a series which so far has been very volatile and not very accurate, and certainly less trustworthy than Galaxy and ReachTEL.  It's only during the months of April and May that we previously had non-Morgan polling, and in May the Galaxy and ReachTEL strongly contradicted each other.  A fair summary though would be to say that prior to this poll, the government had been somewhere around its election 2PP for at least the previous four months.

As we have had nothing except Morgans for nearly three months, the new Galaxy should be weighted rather heavily and, pending any other polls in the near future, should be taken as a sign that the government is at least somewhere in the high 52s by last-election preferences, which probably means at least the low 52s if an election was held "right now".  (The ReachTEL supports some shift in the preferencing behaviour of Others voters, but not enough to make anything like the two-point difference implied by Galaxy's methods yet.)

Leaderships

The other trend we have seen clearly in early Queensland polling is that while her government hasn't soared to the heights of honeymoon success, Premier Palaszczuk is off to a rather good start.  In Galaxy she's recorded successive netsats of +29 (53-24), +32 (59-27) and +29 (57-28).  Her ReachTEL ratings in May were less stellar but this is probably another case in which the middle "satisfactory" option contains a lot of soft positive responses, and also that was a sample with a poor result for Labor generally.  Opposition Leader Lawrence Springborg has been polling middling ratings (netsats of +2 in the last two Galaxys).  Preferred leader scores tend strongly to favour incumbents, and so Palaszczuk's 52:32 lead as preferred Premier tells us nothing that we didn't already know.

Where's this going?

It might be thought that an "accidental government" that falls across the line in minority having never been expected to win has poor survival prospects.  However, some of Australia's more successful state governments started out this way, including the Bracks/Brumby government in Victoria.  A lot has to do with whether the government can be stable enough to see out its term, and avoid a premature demise for which it would be blamed. The other issue, as always, is the impact of the party in power in Canberra.  The short-lived Borbidge minority government did not enjoy the success of many Labor governments that have come to power in minority, but that would have been partly because the Coalition was also in power federally when it went to the people at the end of its term.

At this stage of a government's term, polling has very little predictive value for any government, but that is probably even more true for this one than most.  It is worth keeping an eye on strange interpretations of what is really going on in polling, but the fact that the government's polling is better than the Courier-Mail is saying tells us nothing about the future.

Update: August-September Newspoll

The first Newspoll for the Palaszczuk Government is out and it is a similar story: the released 2PP based on "previous elections" is 53:47 to Labor (ALP 41 LNP 38 Green 9). By 2015 election preferences I get it at 55.0 to ALP, which would be a crushing win.  Leadership figures are somewhat less flattering than the earlier Galaxys for both leaders with Palaszczuk +20 (53-33) and Springborg -9 (34-43).  This might be a poll-methods thing (standard Galaxys have one half that uses live interviews and call mobiles, while the same half in Newspoll calls landlines via robopolling) or it might just be a natural shift.  The preferred-Premier score doesn't add a lot to what we know; indeed for the given set of voting intentions 49:28 might seem like a modest lead, except that Palaszczuk is still a relatively new Premier.  These results were collected over two months.

The general picture seems to be one of a delayed honeymoon for the new government, that is now in full swing but with the extent of it masked slightly by pollster preference-modelling decisions.