Saturday, November 20, 2021

Labor Dominates In Recent Victorian State Polling

Recently two polls have shown that the Victorian Labor government of Daniel Andrews is currently crushing the Liberal opposition led by the recently returned Matthew Guy.  A Morgan SMS poll showing a 58-42 lead to Labor raised some eyebrows (mostly because SMS polling has in the past been volatile and trashy) but the first branded-as-such Newspoll for the term has reported the same 2PP.  In fact, the Morgan poll's primaries (ALP 43 L-NP 31 Greens 11 others 15) were even friendlier to Labor than Newspoll's (44-36-11-9) and but for the use of respondent preferencing the Morgan poll may well have come out 60-40.  This follows a Resolve poll in late October that had ALP 38 L-NP 34 Greens 10 IND 11 Others 7.  Resolve overestimates Independents, perhaps especially at Labor's expense.  I estimated the 2PP for that poll at 55.5 to Labor.  

Potential Seat Outcomes

There is normally no real point projecting polls taken one year out to the next election, and even in this case it's fairly unlikely the 2PP will be very close to 58-42 on election day.  But it is worth looking at some seat modelling findings for the current Newspoll in the event that the 2PP is similar to the last election, which is clearly a possibility. 

The Australian's article says that the Coalition would be on track to lose seven seats.  Coming off Antony Green's pendulum estimates, a uniform 0.7% swing indeed takes out seven of the Coalition's 26 seats; they gain one from Labor and end up on 20.  However, the baseline 57.3% 2PP from the 2018 election is in fact misleading, because it omits Richmond (a strongly pro-Labor 2PP seat which the Coalition piked on running in to try to force a contest between Labor and the Greens).  With an estimate for Richmond included, the 2018 2PP becomes 57.57%, and the swing of 0.43% may well be mostly or entirely rounding.  On a uniform swing on that basis, then, Labor only picks up two notional gains. 

However in reality on a pendulum so stacked with notional Coalition ultra-marginals, two of them occupied by first-term Labor MPs, Labor would probably do much better than that: for a 2PP of 58-42, I estimate a net notional gain of four seats.  Labor  prospects also benefit from retirements in Hastings and Kew.  Assuming all the crossbenchers retained, I'd therefore translate this poll as something like 60-22-6.  Some of the crossbenchers are, however, on very slim margins.  Especially, on uniform swing off these numbers, Labor would narrowly win Prahran, but uniform swing is a very poor predictor of the fate of seats with very high Green votes.  Morwell also probably becomes a notional Labor gain over independent Russell Northe (if Northe runs again), though it is difficult to model because it is hard to estimate Northe's vote in parts of the seat he did not contest last time (see comments).  

Are Polls A Year Out Predictive?  Well, Something Is ...

This section is a little bit mathsy, about 3/5 on the Wonk Factor scale.

Disregarding Morgan for now, is a single Newspoll taken a year out predictive to even the slightest degree?  Here's what I got as a graph of the relationship between single Newspolls taken around a year out (where there was a choice I picked slightly more over slightly less) and state election results, for most elections since 1988 in the mainland states.  I have estimated both poll and election 2PPs as necessary.  I have also accepted Galaxy or YouGov polls at times when they were operating Newspoll, but I have not included elections where there was no poll close enough to a year out (Qld 1992, NSW 1991, SA 2018 and WA 2021 all missed the cut):

19% of variation explained is something I suppose but it isn't very good, and the 95% confidence level on this is +/- 7.6, which is OK if a government is polling 58 (they'll hardly ever lose) but pretty useless in the middle.  This graph was going rather nicely at around 30% explained until I included Queensland, which has produced a number of outliers including the worst outlier by far.  In the January-March 2011 Newspoll, Anna Bligh's ailing government briefly surged to 52-48 off the back of a flood emergency and general LNP uselessness; the flood bounce receded and Campbell Newman entered and Labor ended up drubbed 37.2-62.8.

Victoria has also seen some high-flying governments crash, though not as heavily: the Brumby government was 57-43 up in late 2009 and lost 48.4-51.6 in late 2010.  The Cain government in 1987 was way ahead (I estimate 55.5-44.5) and ended up losing the 2PP and hanging on with a mere four-seat majority.  On the other hand in late 2017 Daniel Andrews' government was only level-pegging yet it won with 57.3% 2PP a year later.  What might happen to the Andrews government this time?

It turns out that one of my pet themes, federal drag, is a massive influence on how well state governments perform compared to their Newspolls one year out.  Actually it would be more accurate to say drag is the major influence on how they perform and the polls are embroidery.  Here's the graph above again separated into dragged and non-dragged governments.  (A government is "dragged" if the federal government is the same party that it is - this is a big disadvantage, whether because voters like having different parties in power at the two levels or just because voters like using state elections to give the feds a kicking and/or keep them on their toes.)

There's very little crossover between the dragged and non-dragged governments.  In the crossover zone we find some of the better performances by federally dragged governments, such as those of the Court and Kennett governments in 1996 (first term state governments with a very new federal government, so less reason for drag than normal).  The Berejiklian government in 2019 is also well up there.  On the other hand, snoozy performances by non-dragged governments include the Weatherill government in 2014 (which was showing its age, though it still somehow managed to win) and the Fahey minority government losing in 1995.  

If the polls a year out are correct, then all the difference in support between the dragged and non-dragged governments appears in the last year.  The dragged governments poll 51.3% 2PP on average a year out and finish on 47.4; the non-dragged ones jump from 51.1 to 54.6.  In the last year of the average state election cycle, close to 4% of voters switch (on a two-party basis) from the dragged party to the non-dragged party.

I did a multiple regression of 2PP by Newspoll a year out and drag status and got this:

2PP = 24.85+0.582*Newspoll-7.25*drag +/- 2.62

("drag" is 1 if dragged, 0 otherwise).

Two variables explain around 58% of variation, which is rather good.  So applying this back to Victoria, if the Coalition is still in power federally, then the Andrews Government is projected to keep its current polling lead and slightly increase its majority (58.6% 2PP) with virtually zero (<0.1%) chance of losing the 2PP on election day. (95% confidence range c. 53.5 to 63.7, ie almost certainly at least a solid Labor win) On the other hand, if the feds get the boot then the projection is 51.35% with a 20% chance of losing the 2PP on election day (95% CI 46.2 to 56.5 - the government is favoured but almost anything can happen).  

The first finding there, I think, is overconfident - I've treated results as normally distributed but because of optional preferencing outliers they're not really (it might be better to convert OPV elections to as-if-compulsory-preferencing estimates), and the overall sample is only 43 elections.  There are also always black swans out there - Labor's possible dependence on its Premier, the current IBAC inquiry, the pandemic and so on - but I think it's quite a powerful example of how much difference federal factors could make.  It may be that the second estimate underestimates Labor's chances even if they do win federally, because the Albanese government would be new and the drag effect may be weaker with very new federal governments.  

Incidentally, this regression is unflattering for the Marshall Government in South Australia, which led 51-49 early this year hence is projecting to 47.3% and a probable loss assuming that the SA election is held before the federal poll.  My historic regression (based on federal drag and government age) also projects likely defeat for said government, though caution is required on both fronts because incumbent governments have on average outperformed in Australia during the pandemic.  

It's not new that federal drag is predictive, and Armarium Interreta have found that well out from elections it is more predictive than polling.  But the extent to which it predicts polling shifts was quite impressive to me.  

One final thing about this: one might think the pattern I've mentioned results because the party in government federally changes and a state government goes from being not dragged to dragged, or vice versa.  In fact there are very few such cases in the dataset and - being cases where the federal government in question was new - they don't support that conclusion at all.


Finally, the poll has Daniel Andrews, after everything, still on a robust net rating of +14 (56-42), however that's down quite markedly from +29 (64-35) in a somewhat smaller sample a couple of months ago. Matthew Guy is on a nondescript net -8 (34-42) and Andrews leads 54-33 as Better Premier.  Andrews' lead seems very modest given that Better Premier skews to incumbents and given that he is an incumbent with an enormous voting intention lead.  However, Guy has the unusual benefit that as a returning opposition leader he is both a known commodity and also in some sense new.  The modest size of the gap may also reflect the somewhat polarised reaction from voters to Andrews (those who don't like him tend to really not like him) and perhaps also methods differences compared to Newspolls of yore.  

I feel that political life would be deeply frustrating for people in Victoria who are opposed to their Premier.  The man seems indestructible!  While the Liberals might be able to improve their position if they stopped fighting each other and in cases associating with extremist cranks who don't seem to be getting polling traction, much of their fate is probably still out of their hands.  They can fiddle around the edges of Labor's dominance but there may be little more that they can do but wait for Andrews to go away, the feds to lose or something huge to fall into their lap.  


  1. Kevin This May Sound A Silly Question But What Do You Mean By NOTIONAL As I Have Read It A Lot And Have Never Been Able To Find The Right Person To Ask Till Now

    1. "Notional" refers to what would happen in each seat if everyone voted for the same party as last time, irrespective of who (if anyone) is actually now the member for the seat.

      There are some seats that were won by one side at the last election, but because some areas have been moved in or out of those seats, if everyone voted the same way as last time the result of the seat would be expected to change. In this case, Bayswater and Bass were won by Labor last time but are "notionally Liberal" (because if everyone now in those seats votes the same way as they did last time, the Liberals would now be expected to win those seats). Ripon and Hastings were won by the Liberals last time but their voter makeups have shifted towards Labor, so they are called "notionally Labor".

  2. Labor will be eyeing off Morwell because the recent redistribution added Moe and subtracted some rural areas, thus almost certainly overturning ex-National Russell Northe's slim majority. In 2018, Labor came first on the primary vote with 34% but Northe won from second on 20%.

  3. Yes - the Tally Room estimates still have Northe holding on by 1% but I haven't checked this for myself ( It seems counter-intuitive if the final changes have boosted Labor's primary by 4.3% presumably at the expense mainly of Northe - that said a tricky thing with modelling such seats can be estimating the performance of the independent in booths they did not contest in the previous election. I'll see if I can find enough data to have a go at it myself.

    1. Yes, and Morwell would be a particularly difficult example given the somewhat ... peculiar reasons why Russell Northe became an Independent. I remember reading somewhere that it was suggested a lot of people voted for him so they could be sure he'd have the money to pay back his (publicly admitted) gambling debts!

  4. Respected pseph tweeter @auspollsucks has modelled it as 50.6 to ALP:

  5. The 2022 Victorian state election will presumably be 6-8 months after the Commonwealth election. A little shorter than the gap between the 1996 Commonwealth and WA elections but a lot longer than the very short gap between the 1996 Commonwealth and Victorian elections. If the ALP win the Commonwealth election, the Vic ALP vote will likely be less effected by Commonwealth drag like the 1996 examples (all be it they were only seeking second terms, the current Victorian government is seeking a third).

    I suspect an Albanese-led Commonwealth ALP government could have a positive effect on the Green vote, both at state level and Commonwealth level, due to Commonwealth drag and Albanese`s Beazley-esque likely lack of appeal to potential Green voters (at least in government).