Monday, February 1, 2021

Newspoll: Surprise 2PP Tie Takes Heat Off Albanese

The first Newspoll of this year had the potential to be a very significant one for Opposition Leader Anthony Albanese, but he has escaped it without major harm.  This was one of the more significant Newspolls lately because of the leadup, so I thought I'd briefly write about it and related themes.

Leadership speculation has been swirling around Albanese in recent weeks.  This started after forces within the CFMEU and closely linked to unionist John Setka spruiked polling claiming to show dire numbers for Labor in its Newcastle area seats of Shortland (4.5%) and Paterson (5.0%).  Setka was expelled from the ALP in 2019 at Albanese's insistence.  

The polling has not been published for scrutiny.  The polls were conducted by Chorus Consulting with Community Engagement.  No public polling for any election by Chorus is known to me, but their director is also a director of Redbridge, who have released other polls predicting similar doom for Labor in other seats including Dobell and Macquarie.  I have been told that Community Engagement did tracking polls for Labor at Victorian state elections, but as no results were published or given to me at the time there is no way to comment on their accuracy.  I am however aware of some polls by them that were published for the 2016 election and these were not that accurate overall (the worst a Higgins seat poll that had Liberal 42 Greens 25 Labor 20; the result was 52-25-15).

So we have a lot of the hallmarks of a poll-shaped-object media beatup here (which is not to say there is necessarily anything wrong with the poll): poll not released in full, poll by obscure pollsters with insufficient publicly checkable track record, and poll gives the sponsor the unlikely result that they want.  That the CFMEU's Elizabeth Doidge, a known Setka fan, then openly called for a change of leadership to Tanya Plibersek or anyone but Albanese really, should have given the game away. The Australian, which claimed the polling "showed" Labor was on track to losing these seats, should have developed some shame for once, an issue I'll return to later.  

Nonetheless in the following days some possible signs of positioning from Plibersek and Bill Shorten could be detected, a betting agency had Albanese at $1.70 to go before the next election (now $1.65 to stay) and some rather bad interviews by Albanese plus an Australia Day misfire added fuel to the fire.  

In one of his interviews, Albanese suggested the Australian look at its own Newspoll, and argued that he was doing well on three metrics: the Labor 2PP was up on the last election, the Labor primary was up on the last election, and Albanese himself was rating positively.  (In a later interview Albanese falsely claimed his net satisfaction had only been in positive territory "with only I think one exception".  He has in fact had eight negative netsats as Labor leader, including one in February 2020 and two in September and October.)

Whenever a politician uses polls to justify themselves in this way they make a rod for their own back by creating something that can be used against them should the polls turn bad, as Malcolm Turnbull found to his cost with his famous "We have lost 30 Newspolls in a row." So my advice to MPs would be never to do this.  Albanese was taking especially bad risks here as Labor's 2PP had most recently been at 49% and could easily drop to 48% (below the last election) by sample error alone, while his own net rating had most recently been at +3 and could easily fall below zero.  

Had the first Newspoll of the year been a shocker for Albanese, critics within the party would have used his own words against him, and this could easily have led to the kind of downward spiral that has featured in so many leadership spills.  However, that hasn't happened.  Pro-Albanese forces may well be refreshed, though no-one should be under any illusions that Labor's path to victory is easy.  Nor can we be sure this poll, using methods tested at one state election so far (but with pretty good results there) is accurate.  

The Newspoll

This week's poll came in with primaries of Coalition 42 (-1) Labor 36 (-) Greens 10 (-1) One Nation 3 (+1) Others 9 (+1).  In two party preferred terms this all looks like nothing to see, because the lost percent from the Coalition should be nearly cancelled out by the lost percent from the Greens, with the difference offset by the gain to One Nation (who preference the Coalition somewhat).  However, the 2PP did move, from 51-49 to Coalition to a 50-50 tie.

The issue here is rounding.  While changes are described as a whole point, for all we know the Coalition might really have come down from 43.4 to 41.6 (a 1.8 point loss that appears as one point after rounding at both ends).  The Greens might really have come down from 10.55 to 10.45 (a 0.1 point loss that appears as a point loss after rounding at both ends).  Those are extreme and unlikely scenarios, but what's more likely is that the Coalition 2PP might have come down from just above 50.5 to just below 50.5, ie, the change isn't really a point.  Indeed, both sets of primaries as published imply a last-election 2PP of around 50.5.  More likely than not, the Coalition was actually still ahead in this sample, and would be expected to win an election "held now" if the sample was accurate, but Labor forces bunkering down for 48-52 or worse would take it with much relief.

On leadership statistics, PM Scott Morrison drops six points on net rating from +36 to +30 (63-33).  This represents his lowest net rating, highest dissatisfaction, and lowest satisfaction since his first COVID-boost Newspoll (+26) back in April.  It's still a strong rating for him.  

Albanese drops five points from +3 to -2 (41-43).  This is bad given that he was just boasting of being in positive territory, but probably won't hurt him nearly as much as a 47% 2PP or a 33% Labor primary would have done.  

PM Morrison's lead on the skewed Better PM metric is also down from 32 to 28 (57-29), Morrison's narrowest lead since May, but again, still rather good.  

Way Past Time For Benson To Go

Useless commentary has as usual come from Simon Benson at the Australian.   I have been calling for Benson to be removed from Newspoll commentary for a long time because of the large number of extremely basic errors that he makes. It now seems that even the revelation that he is in a relationship with a Nationals senator cannot shame The Australian into even disclosing that conflict of interest, let alone giving someone more competent a go.  In the latest total howler defacing the national broadsheet he writes that "While maintaining an unbroken 12-month record of popularity and approval ratings in the 60s, the last time Mr Morrison suffered a similar fall in satisfaction was during the height of the Victorian lockdown of last August."  In fact it is just under ten months since Morrison poked his head above net zero and recorded his first 60+ positive satisfaction rating (before that he was on net -12 and 41% satisfaction).  This follows another blunder in which Benson claimed the government had a "one-seat majority" (see previous article). 

In this case Benson attributes the ratings change for Morrison to voters "venting a broader frustration over Christmas border closures and the disruptions to holiday plans imposed by premiers" but a poll with the PM and LO on a combined +28 is hardly one that suggests a furious voting public.  More humdrum explanations could include that the Morrison bounce is gradually deflating as voters get used to the COVID situation and start thinking about other things, and also that the change is mostly sample noise.  (It's just outside the published margin of error, but the real margins of error of polls are often larger than published).  

Can't Get There From Here?

There continues to be vast scepticism about Labor's chances of winning the next election from its current position.  This arises not only from the pandemic boost that governments get if they succeed in containing COVID-19 but also from the generally bleak narrative about Labor: that it stands for nothing much, lacks an ideal leader, has no plan to win, and has lost blue collar workers/the outer suburbs/the regions while at the same time failing to enthuse the left.  In fact the swing against Labor in the outer suburbs in 2019 was just consistent with the national swing after a good performance there in 2016 against Turnbull; it is in the regions and the bush where Labor returned to 2013 level awfulness. 

The counter-case is that the last election was reasonably close and Labor really should have won it, so if Labor can just avoid the major mistakes it made in 2019 (including having such an unpopular and distrusted leader) and come up with a plan for post-COVID rebuilding that draws on the party's historic strengths, then, who knows, maybe there'll be a surprise.    

However, if the election is going to be late this year, history doesn't bode well for Oppositions that weren't well ahead at this point.  Andrew Clennell overstated this slightly when he said Oppositions usually have 53% or 55% 2PP by this stage if the government is going to change, but generally, Oppositions that won have been well ahead around nine months out.  The Coalition under Abbott were up about 52-48 on aggregate in late 2012 (and even that was a lead that had narrowed during the Gillard government's short-lived resurgence), Labor were up about 55-45 in 2007, and the Coalition under Howard were ahead about 52.5-47.5 in mid-1995.  To go back further using old Morgans, there were large leads for eventually winning oppositions for Labor in the leadup to 1983 (yes even under Hayden) and the Coalition in the leadup to 1975; I get these at around 55-45 and 58-42 respectively.  In 1972 Labor had a slim lead, about 51.5 on my conversions, in a poll taken in February.  For the only counter-example we need to go back to 1949 where Labor was just ahead nine months out from losing, and that's not too much of a precedent for modern polling!

None of this says it can't be done though - the problem is that winning from Opposition is such a rare feat that we don't know a lot about the circumstances under which it occurs.  

There will probably be another dump of voting intention results from Essential shortly, and if that happens this week I will update this article with comments on it.  

Essential Quarterly Dump

Essential has now released its quarterly set of voting intention results.  As discussed when Essential adopted this method, the idea is to get away from horserace politics in which too much reliance is placed on individual poll results, but Essential contradicts this by leaving the most recent single poll result flapping in the breeze on its website for months as if it is the current state of play.  Displaying a quarterly average in the sidebar would be more consistent with Essential's aims in that regard.  

The current Essential sample has Labor ahead 47-44 (= 51.6% 2PP) on Essential's 2PP plus measure but the previous fortnight's sample had the Coalition ahead 48-45 (also 51.6% 2PP). These samples are only just over 1000 voters each, and are reasonably volatile.  Through the quarter the average result has had the Coalition slightly ahead (I get 50.3-49.7), with three Coalition leads (one of them large) and four small Labor leads.  Even that lead disappears if the primary votes are converted to last-election preferences (I then get 50.1% to Labor), suggesting that the Coalition (!) has done well on respondent preferencing in this sample of seven polls.

The Coalition did somewhat better in the three published Newspolls during this quarter, so time to have a look at the Newspoll-Essential comparison graph, for which I find the closest Essential reading (in normal 2PP terms) for each Newspoll that comes out:

Obviously Essential is much more volatile than Newspoll on this graph, but the graph does not include all the Essential readings, just those closest to each Newspoll.  The average difference across the term is that Essential is 0.5 points more Labor-friendly than Newspoll.  These sorts of differences tend to chop and change from time to time but I think it is notable that Newspoll has implemented targeted changes designed to address the 2019 polling failure.  Essential, however, does not seem to have done much to alter its sampling or weighting, and has mainly altered the presentation of its data.*  For this reason if I had to guess which was more likely to be accurate, I would go with Newspoll (which has also been tested at one state election whereas Essential hasn't been tested since 2019), but all polling needs to be treated with extra caution until pollsters have gone through a couple of federal election cycles without a repeat or near-repeat of 2019.  

Essential has found Labor improving in recent months among voters aged 35-54 and among voters with low household income (<$52K).  The latter has an obvious potential explanation (windbacks of COVID assistance) but the patterns from the graphs in their Zoom session today are pretty messy and sometimes patterns like this in subsamples can be just sample noise.  

(* The only weighting change I can find a public reference to refers to use of state-based demographic data rather than national.)


  1. Seriously? A voting intention poll for the federal election which hasn't even been announced yet?

    There's an election in about a month for WA and there haven't been any state voting intention polls of that for 2 years, for crying out loud.

    (Not to mention, the outcome of the WA state election barring some massive upset is pretty clear. Sounds like an easy way to "predict" a correct result; why aren't our pollsters jumping all over this?)

    1. Will be interesting to see what (if anything) actually comes out by way of polling for WA. I'd assume there'd be at least one public poll by somebody (probably YouGov/Newspoll) but will be interesting to see if there is more.

    2. My WA state election prediction; the Libs won't need more than a Holden Barina for party road trips.

  2. I’d say the skepticism about Labor is more based on distrust of the polls than any existential issue. “The polls showed Labor well up last time and they still lost...can’t be much hope if they’re barely neck and neck!” kind of thing.

    1. Yes there's quite a bit of that about.

    2. Which is probably not justified if Newspoll have altered their methodology in a way that reduces Labor's polling advantage compared to the last election cycle.

  3. Given Newspoll usually rounds to the nearest 0.5% (why they would unnecessarily reduce the precision of their polls is beyond me), I'm surprised this result doesn't round to 49.5% instead. Using last-election preferences, I get about 50.3% 2pp, which should round to 50.5% instead of 50%.

    1. Newspoll doesn't round the 2PP to halves except in the final poll before the election. The formulae I use for last-election 2PPs (depending on the data available) are given in this article: