Saturday, April 24, 2021

What's This Then? Commissioned Poll Claims Liberals In Trouble

 Australia Institute uComms: Liberal 41.4 Labor 32.1 Greens 12.4 IND 11 Other 3.1

If accurate Government would probably lose majority (approx 12-10-2-1 or 12-9-3-1, perhaps more INDs)

Handle with caution - commissioned robopoll, IND figure looks exaggerated

Aggregate-based model of vaguely recent vaguely public polling: approx 13-8-3-1

A week out from the 2021 Tasmanian state election (link to guide main page) there remains no public polling that is less than two months old.  This is a parlous situation and I feel that relying on the media to commission useful polling at a useful time has failed and that for future state elections it will be necessary to look at crowdfunding reputable polling to fill the void a few weeks out.  If, that is, such a thing proves viable.  


Mercury polling that was said to be coming last night and is being hyped as an "Exclusive Poll" unfortunately turns out to be not a poll at all, leaving me feeling that April Fools Day has come 23 days late after getting up a number of hours too early to cover its results.  It's an opt-in reader survey from which no useful conclusions can be drawn.  It reflects only the views of those readers who have chosen to participate (likely to be highly politically engaged), and may have been spread through social media by some political groupings more than others.  It would also be likely to be Hobart-centred.  All up Labor should not lose sleep about being in (gulp) fourth place in it, though from the ground campaign evidence in my neck of the woods I'm not entirely sure the ALP exists.  

We do however have voting intention polling in the form of a report of voting intentions in an Australia Institute uComms of 1023 voters statewide.  The raw numbers are Liberal 40.2 Labor 27.4 Greens 12.4 IND 10.2 Other 2.8 with 7 "undecided".  However "undecided" voters in uComms are actually voters who are leaning to a party and would be included in top-line results by other parties.  The redistributed results are Liberal 41.4 Labor 32.1 Greens 12.4 IND 11 Other 3.1.

Polls that give "Independent" as a standalone option and don't also include a list of minor parties are prone to overestimate the IND vote.  A previous uComms in December 2019 had the Independent vote on 11.7% (higher than at present!), and this was at a time when no independents were actually known to be running.  

The list of cautions about this poll is very long:

* While it would be assumed that the coverage was evenly distributed statewide, this is not known for sure.
* uComms is untested at Tasmanian state elections (though ReachTEL robopolls in 2014-8 on the same platform were quite good. uComms federal polls of Bass and Braddon in 2019 at least had the right winners and were only a few points out on the 2PP, but the primary votes were all over the place).
* The poll only weighted by age and gender, a dinosaur-age approach given the 2019 polling failure.
* uComms polling for the WA state election had an 8.7% miss on the eventual two-party preferred - that said it was taken six weeks out from the election and voting intention may have changed a lot in that time.  
* uComms polls have frequently shown implausibly conservative results in the 18-34 age group, presumably because they are hardly reaching anybody in it.
* The poll is commissioned by a left-wing interest group.
* uComms is a controversial poll because of its union links.
* The Labor vote in this poll is particularly soft.  Labor has only 27.4% before the pressing of voters who are initially reluctant to name a party.  It is rare to see these so-called "undecided" voters splitting so strongly to one party, and probably partly a product of small sample size.  
* uComms may have house effects (see added Benchmarking section).

These cautions noted, I have in the past found that voting intention components of these TAI polls tend to be useable data, unlike the commissioned issues questions which frequently have skewed wording or are otherwise badly designed.  (In this case the poll uses agree/disagree questions (always dubious) to soften up respondents before asking if they support a proposition regarding salmon farming that most would have never thought about.)

An attempt to convert this poll to possible voting intentions statewide if it is correct is below.  As there is no report of electorate by electorate breakdowns, I have assumed that the Clark others vote is about double the state average and the Braddon others vote is on the state average.  I have also assumed that the increased Others vote in Clark draws 80% from the Greens and Labor.  (Although Sue Hickey is a former Liberal, her campaign is more or less entirely left-independent in content.)



In this scenario the Liberals win three in Bass and Braddon.  They probably also do so in Lyons because Others preferences in this division tend to favour them.  However they would be in trouble in Clark where their surplus after quota is in seventh place and candidate effects would be unlikely to change that.  The Clark result in this model (Liberal-Labor-Green-Independent) would be 1-2-1-1, 1-2-0-2 or 1-1-1-2.    

Of the others on these hypothetical breakdowns the Greens might win the final seat in Bass, though a sufficiently even candidate split in Labor would defeat them, and Craig Garland could win in Braddon though he would not get all the Others vote and Labor's even candidate split would still help them.  I am not saying any of the things implied in this breakdown will or are even likely to occur, just that this is a glimpse into what the poll could produce if accurate.  It is still possible even on these statewide numbers to find a path to 13 Liberal seats but it would be a very flukey one.  

The very high Independent vote is one reason for caution here.  It is plausible that independents in Clark could get 25% (I've heard of a party poll at this level, but polls could be oversampling politically aware respondents) but aside from Garland who could get several percent, there are no Independents with any profile to speak of elsewhere.  So 11% statewide seems excessive. 8% maybe.  Given that the independent vote was at the same level when there were no known independents, possibly a lot less.  

State polling in Tasmania has a history of favouring the Greens and it is worth flagging here that a very high Independent vote in Clark could threaten Cassy O'Connor's seat.  I have also heard of some party polling that if accurate raises the same possibility.  

The major question is whether a Liberal to Labor gap of only 9% is possible.  It's very different to previous polling months ago that had gaps of 18-30%, but a lot has changed since then (the early election call, the less COVIDy than expected nature of the campaign and so on).  It's also inconsistent with what noises about party polling I've heard, which report a gap around the 20s, more consistent with the now outdated public polling.  And, again, it's a commissioned robopoll.  There's a fair chance that it's nonsense.  Still, for the moment, we know nothing reliable about the current state of voting intention, so we shouldn't dismiss it out of hand.  At the very least, it is some sort of evidence against a WA-style landslide, which is unlikely anyway, despite COVID, in view of federal drag and the age of the state government.  Indeed, if I knew nothing about this election but those things, I would expect the Liberals to lose their majority.  

What would the Liberals falling just short mean?  If we take the leaders at their word Peter Gutwein would resign as Liberal leader while Rebecca White would be unable to form majority government but might try to continue as Opposition Leader if necessary.  A different Labor leader could take over and become Premier supported by Greens and/or independents, or the Liberals might come to an arrangement with an independent and continue with a different Premier.  From the perspective of the Liberals, whether they governed or not, the early election call would be a tragic waste.  

The following is a possible aggregate-based projection weighting this poll at 20% and the old data at 80%:  I feel that both these values are way too generous but it has to add up to 100 somehow!  


The major uncertainty here would be the final Bass seat which could go to the Liberals, Labor or Greens.  Although Labor would be a very long way behind on raw quotas, if they could split that evenly between their top two, they might be able to keep two candidates ahead of the fourth Liberal and win on Liberal preferences.  If they don't manage the even split, on these numbers the Greens win, but the usual caution about Greens and polling applies; four Liberal would be possible though the recent behaviour of Labor preferences (see Franklin 2018) suggests Greens win.  (I'll believe it when I see it!)

I have heard of a set of party polling that according to the source finds broadly similar results to the above and that also gets Clark as quite a mess, and that has Sue Hickey closer to Kristy Johnston than the general narrative has been believing.  On the numbers I have above the Greens would hold, but on slightly worse numbers they could be at risk depending on the split between the Independents and which one (if either) is knocked out first.  

There is a widespread expectation that the result of this election will be something like the status quo 13-9-2-1 (eg the entire Fontcast crew and the Not-A-Polls in my sidebar, which should make the former nervous) or perhaps 14-8-2-1 (eg Sean Ford in the Advocate).  I think we actually still know very little about this weird election and that while 13-9-2-1 is hard to argue with, the Liberals could still do anything from fall just short (as in this commissioned poll) to win by an enormous margin.  

Perhaps we'll get some real and public polling soon!  A uComms in the field last Monday sounded like a media poll but at this stage has yet to show itself.  If nothing else, we could get EMRS.  

Added 27 April: Benchmarking uComms off EMRS

Something that's occurred to me to do is to benchmark previous TAI uComms polls by comparing them to EMRS polls around the same time.  

TAI uComms polls were taken in November 2019 and November 2020.  We can compare these polls to EMRS polls taken around the same time:

November 2019: The Liberal vote was 8% lower than EMRS, Labor 0.4% higher, Greens 3.8% higher, IND/Other 3.8% higher.

November 2020: The Liberal vote was 1.7% lower than EMRS, Labor 6.8% higher, Greens 2.3% lower, IND/Other 3.8% lower (NB EMRS added to 101%).

The swings between the November 2020 poll and the current uComms are Lib -8.9, ALP +0.3, Greens +1.7 IND/Other +6.9.  

There is a suggestion here that uComms may skew against the Liberals by quite a bit compared to EMRS, which has itself underestimated the Liberals in its final polling at the last four consecutive elections (by an average of 1.8%).  This would suggest that as an extremely rough estimate the current uComms could be underpredicting the Liberal vote by several percent (I get 6.6%) which would put the Liberal vote at about 48 (= c. 13 seats).  But with so few data points to benchmark uComms off, the underestimation could be much lower, much higher, or might not even exist at all.  Also, we don't know that EMRS still underestimates the Liberal vote slightly, as it changed its polling methods during the present cycle.  

In any case, the existence of any degree of similarity between the two at least suggests that the uComms might not be ridiculously out.  The Liberal vote might be much higher than 41, but is less likely to be, say, high 50s.  Or maybe uComms has no house effect at all but is just really volatile and the Liberals could do terribly!  We will see ...

11 comments:

  1. It would be interesting to if the pseph community could come together to crowdfund a poll from their respective readerbases. It might be one way to get polling transparency started (lead by example and all that jazz).

    I also wonder if it's possible to incentivise anti-herding from polling companies. e.g. a small bonus if the poll commissioned is closer to the final result than an average of their competitors, but no penalty if they're further out. Would you happen to know how much a standard election-week poll would cost?

    ReplyDelete
    Replies
    1. I've seen costs of a few to several thousand for the ReachTEL style robopolls but I haven't asked around yet. If I get serious about the idea then I will be seeking quotes closer to the time.

      Delete
  2. Regarding the independent option, I'm seriously astounded to hear, both in this case and with Resolve, that Australian pollsters still regularly offer "Independent" as an option on polls. The problem is that everyone wants an independent (or from an American context, everyone wants a 3rd party), but no one can agree on the kind of 3rd party they want. When you say "will you vote for an independent", some people will imagine a rural conservative a la Katter, some will imagine a leftist, some imagine a small-l liberal, some imagine a moderate etc etc. Lots of voters are willing to vote for the "right kind of independent", but not the independents who actually contest their seat (many of whom they likely have never heard of).

    In an environment like Tassie, where nominations have already closed and there's a relatively limited number of independents, I think pollsters should try to tailor the options to the electorate which the respondent lives in (I know online pollsters such as YouGov already ask for your postal code, so it's not impossible). For example, instead of asking "will you vote for Labor, Liberal, Green, Independent or Other", they should ask "will you vote for Labor, Liberal, Green, Independent Kristie Johnson, Independent Sue Hickey, or Other" (using Clark as an example). Reasonably speaking, if a respondent hasn't even heard of the independents running, they're not very likely to vote for them.

    (of course that might bring up some issues of which independents should be read out and which should not, but as far as I can tell pollsters already have to deal with such issues when deciding which minors to break out from the Others grouping [e.g. early days of the Greens, PUP/UAP in 2013/2019, One Nation and its boom-bust cycle of popularity])

    ReplyDelete
    Replies
    1. I agree. Looks in this case that cost issues and the small coverage per division meant they went with a generic readout statewide. I would have done different readouts for different seats and given Ind (Johnston) and Ind (Hickey) as options in Clark, as at least one party poll is doing. I might have also given Ind (Garland) as an option for Braddon.

      Delete
  3. Regarding the high independent vote and relatively small Lib-Lab gap, I wonder if this poll (intentionally or otherwise) has over-sampled Clark at the expense of the northern seats?

    ReplyDelete
    Replies
    1. Yes that's something that's possible but because of the lack of transparency I have no information on it.

      Delete
  4. Bummer about EMRS the cowards

    ReplyDelete
  5. Governors General, not Governor Generals ;-)

    ReplyDelete
    Replies
    1. That's annoying, I already noticed and fixed that one, it must have reverted somehow or I must have saved an unfixed version on top of it. Fixed again now.

      Delete
  6. Hi Kevin. What are your predictions for candidates most likely to be elected?
    Greg

    ReplyDelete
    Replies
    1. My various electorate pages have some thoughts on which candidates might be in the mix but I am mostly avoiding firm predictions because there have not been enough polls.

      Delete

The comment system is unreliable. If you cannot submit comments you can email me a comment (via email link in profile) - email must be entitled: Comment for publication, followed by the name of the article you wish to comment on. Comments are accepted in full or not at all. If you submit a comment which is not accepted within a few days you can also email me and I will check if it has been received.