2024 Tasmanian Polling Aggregate

Aggregate of all polls (not a prediction) Lib 36.9ALP 25.3 Green 13.2 JLN 9 IND 12.7 other 3

Seat estimate for this aggregate 15-10-4-3-3.

--

This article is part of my Tasmania 2024 state polling coverage.  Click here for links to my main guide page which includes links to seat guides and effective voting advice.  

--

An attempt at aggregating the 2024 Tasmanian polls has been long-coming amid a very distracting and busy campaign, but for what it's worth here goes.  For the second election running I have doubts about the value of this exercise, but for entirely different reasons.  In 2021 there was very little polling and the only campaign poll to be publicly released appeared to (and did) have large house effects, which I determined using EMRS as a benchmark.  Despite me talking them down, both my house-effects aggregate and my no-house-effects aggregate somehow worked, with the former nailing the seat estimate and the latter recording voting share misses of 0.5% or below on all four lines.   I don't expect to be that lucky this time, however I hope the journey of how I try to come up with a what the polls are saying number will make some sense.

If any more public polls are released before 8 am Saturday a fresh aggregate will be included in the article covering that poll, or in this one.

Which polls to use

I believe in only using public polling data for headline numbers for aggregation, because without knowing the pollster who did a mystery poll it is impossible to benchmark it.  Also neither of the mystery polls has been released in enough detail to compile a full set of primary votes.  So I am excluding mystery polls from the primary vote figures, but I will use them for the seat distributions, because they have large seat sample sizes that can be compared to overall averages.    

My practice in Tasmania has been to use previous EMRS polls and not just the most recent, the paucity of polling data meaning even polling that is half a year old is worth putting some weight on so long as it is not too much and nothing much seems to have happened in the meantime.  For all polls I need to take into account both the freshness of the poll and also what I think of the poll's track record, both in the state and nationwide.  

To this extent, EMRS is known to have the strongest track record in Tasmanian polling since Newspoll (so far!) ceased polling in 2014.  Of the others, YouGov and Redbridge have no track record on Tasmanian voting intention, while the 2021 uComms poll was a failure.  On the other hand, in recency terms uComms>Redbridge>>YouGov.  

After multiplying out some weightings for age and some weightings for quality and slightly downweighting polls with fewer than 900 respondents, I came up with the following base weightings:

.05 EMRS August

.09 EMRS November

.11 YouGov January

.35 EMRS February

.20 Redbridge February

.20 uComms March

However these numbers require a lot of massaging.  Only the EMRS and Redbridge polls have JLN excluded from Clark but with a full IND/Other option; YouGov restricted the IND/Other option to specific independents, presumably inflating the JLN vote among others.  I adjusted the YouGov poll by adding what I thought was the minimum reasonable tally for independents/others beside those named, estimated by seat, which came to 7%, and then took that total off the non-government parties on the basis of the strong general polarisation in attitudes between voters supporting the government and those not.  I also had to adjust the early EMRS polls for the absence of JLN, which I did by proportionally allocating them at the same level as the EMRS February poll.   I'm not impressed with what anyone has done with separating IND and Others so I've just assinged Others a blanket 3% deducted from IND.  

After all this the headline aggregated estimate I came up with for a non-house-effects model was Liberal 36.5 Labor 26.4 Green 13.1 JLN 9.4 IND 11.6 others 3.0.

Assigning seat effects

The next step is assigning seat effects.  Here the sources are:

* the EMRS seat breakdowns.  I have used the aggregation based adjusted model for the last three that I previously published a model based off.

* two mystery polls, to the extent usable

* the YouGov seat samples (but at a very low weighting as they are tiny and old).

What I aimed to do with these was come up with estimated differences from the statewide totals for each state.  Because some of the samples only include some parties and not others, the totals often ended up not adding up so there was the usual banging and clanging to get the polls to produce a balanced estimate.  This is what came out:

Some of these may look and probably are wrong individually, but they are my read of "what the polls say".  As applied to the primary vote figures, these then produce this:

I should note that the IND vote in Braddon looks a tad low here, mainly as a result of issues with the first mystery poll.  In this aggregate Bass, Clark and Franklin are all reasonably straightforward barring a weird split of the independent vote in Franklin.  

Three seats are in doubt in this aggregate.  Firstly while it looks like the Liberals should win four in Braddon here, they are susceptible to leakage and there are a heap of preferences spraying about.  In this scenario if Craig Garland gets nearly all the independent vote he can beat the fourth Liberal on preferences provided that there is a fourth Liberal; if one of the new Liberals manages after preferences to closely match one of the incumbents Jaensch or Ellis then he has no chance.  It is interesting that I have here found a scenario where Garland might win on as little as 5% - don't expect to see this happen though.  

There are two seats in doubt in Lyons where for some reason the aggregate has the Liberals short of a safe three quotas.  If Jane Howlett gets an incumbent-style vote alongside Guy Barnett and Mark Shelton then they get three comfortably but I have doubts about this occurring.  If John Tucker gets a reasonable slab of the 0.9 quotas that has gone to independents here (likely an overestimate in reality in my view) he could knock over JLN and the Liberals even from behind.  My suspicion is that the Liberals would start with a lead that JLN and Tucker would not both catch; Tucker could do well off Shooters preferences while JLN could gain off Labor's.  Overall the most likely single scenario on this aggregate appears to be 15 Liberal 10 Labor 4 Green 3 JLN 3 IND.  

... but should we believe it?

I've added notes about some individual results in the aggregate that I have doubts about above, but how about the overall picture?  Here are some reasons for caution about any aggregate of polling, or any one poll this election.

Firstly the data are old.   The most recent public polls with known dates are EMRS 15-21 Feb, Redbridge 16-28 Feb and uComms 4-5 March.  This is different, for now, to the 2014 campaign with two major polls in field in the final nine days and 2018 with a last-week EMRS.  Perhaps there will be more in the last 36 hours but if this is all then we have had weeks of campaign for things to change.  Quite a lot has happened in that time in a particularly noisy and in some cases bumpy campaign.  Often, campaigns have little effect, but it would be risky to assume that this one hasn't.  My biggest concern here is that the Lambie vote declined through the 2018 campaign and may do so here again.  The fact that Labor's worst two polls were the last two is also interesting.

Secondly and linked to the above, there could have been bandwagon effect.  There have been a few cases, most notably 2006 and 2018, when a government that showed no sign of winning a majority months out suddenly roared into life once it became apparent only it was in the race for one.  In both cases there are other possible alternatives (both were infamous elections for "dark money" outside-actor campaigns, and both featured bad policy decisions by oppositons) but nonetheless it's always a risk in Tasmania.  There are however cases like 1996 where bandwagon effect could in theory have saved a government majority but actually didn't, and also 2021 as a case where majority was in doubt but no bandwagon seems to have occurred.  

Thirdly, there may be house effects.  But noting that last time my no-house-effects aggregate was actually closer than the house effects one, and noting various doubts about them, I am not sure that they are, and at this stage I have not put out a house-effects version.  To go through the pollsters one by one here:

* EMRS underestimated the Liberals at four elections straight by an average 1.8% through 2018, then had them 3.3% too high in 2021, but the 2021 poll was over two months before the election and wasn't really a "final poll".  

* uComms underestimated the Liberals by 7.4% in 2021 - which was predicted by my benchmarking off EMRS, which suggested uComms was under by 6.6%.  But uComms has since switched from robopolling to an SMS/robopoll mix (I believe it's mostly SMS).  Furthermore uComms' April 2023 poll (taken in the month before the Tucker and Alexander defections) matched the February 2023 EMRS as concerns the Liberal primary.  (It had Labor and the Greens higher than EMRS and generic IND/others curiously low)

* For YouGov and Redbridge there is no past Tasmanian form to judge them on, so it is hard to know what to make of them having the Liberal primary considerably lower than EMRS and uComms.  

Overall there seems some room above to think that there may be a lean against the Liberals in this polling.  But (i) the evidence isn't conclusive (ii) overall the evidence doesn't point to them winning any more than 15 seats if these polls were accurate but for any house effects when taken.  To get their primary well up into the 40s and start winning more, they would have to have actually increased their vote share.

Overestimating the Green vote has been an issue in past polling, but seems less so recently with more polls moving away from live polling.  In the aggregate the Greens seem to be reasonably solid in Bass because the majors are not threatening for extra seats and the independent vote is modest, while the Lyons seat is closer because of the extra competition.  

Finally none of the polls have properly surveyed the independents.   YouGov listed only a single independent per seat and the other polls have offered just a generic independent option or a generic independent/other.  There is no poll that simulates the ballot by offering named independents, but one might also say that if it does that it should name all the other candidates as well, which is impractical.  (Hard to believe but EMRS used to actually poll named candidates once upon a time!)  The concern here is not just the total vote, which the polls may have a good handle on, but that crucial information about the split between different independents in each seat is missing.

There's a common problem in polling that canvassing "independent" on the ballot everywhere will normally overestimate the independent vote.  The voter picks this option and then finds there are no independents running, or only an obscure one or one the voter doesn't like.  But this election is different, if a voter is looking for an independent there is quite a good chance they can find one to match their philosophy in any electorate, and it's even possible it's one they've already heard of!  Not only is it not clear that the independent vote is being overestimated here, but also it's possible some of the early polls underestimated it before the full range of indepednents running was known.  It will be a fascinating thing to watch.  

Strange new world?

The numbers in the polls - collectively - are quite remarkable.  Supposedly something close to 40% of Tasmanians will not vote for a major party this election, up from 23% in 2021.  It's something that's not surprising in the Senate, but we have never seen anything like at state level, where the personal votes of incumbents and the restricted party mix tend to keep the major party vote higher.  There is always some scepticism about this sort of thing because of the history of polling bubbles for third parties bursting (eg SA-BEST in the 2018 South Australian election) but there are elections that have unusually high rates of non-major party voting where the polls get those rates more or less right (Queensland 2017) or even underestimate it (federal 2022).  So maybe this will happen, but election night is going to be a mess if it occurs!  Or maybe we will see something more mundane - or even more extreme?  It is always risky to attempt to second guess the polls!

Update 22/3: Freshwater Fully Added

I expect to update this aggregate again today as another poll is apparently coming [EDIT: no, didn't happen], but the main change since the article was written was that mystery poll 2 has been confirmed as by reputable pollster Freshwater Strategy, a fortnight ago and some more details have been published.  I have weighted the incomplete details I have of this poll at 30% and the rest of the aggregate at 70%.  I have also modified the seat breakdowns accordingly.

It should be noted that federally Freshwater is slightly (about 1%) more Coalition-friendly/Labor-unfriendly than most other pollsters, which is not to say that it is wrong.  Including it in the aggregate may help balance out other polls that could lean the other way.  

The following is the revised aggregate:

The seat outcome did not change.  (NB Overnight I made a very minor change, which is to reduce JLN by 0.2% per seat for the expected higher impact of informal voting on their ticket.)

Can a majority still occur?

A Liberal majority is still in theory possible.  The polls could be somewhat wrong, there could be change in the last two weeks with no polls, and they could get lucky on the distribution.  But to illustrate how hard it is, I added 8% to the above aggregate for them, taking 2% off each of ALP, Greens, JLN and IND, and they were still very marginal for seat 18 in Lyons.  So Premier Rockliff's claim they are within "a whisker" of winning outright is not true according to the public polling.

I believe that a Labor majority can be safely dismissed.