2020 Federal Polling Year In Review

 At this time I normally release a review of the year in federal polling that, among other things, states the number of federal polls released that year.  See the 2019 review here. Things have got so weird in Essentialville and Morganland that for 2020 I can only report on how many have been released so far (since others with 2020 data are likely to belatedly appear next year), and in both cases, "released" becomes a stretch in the case of 2PP readings that have no public existence beyond a dot point on a badly labelled graph.  Such is life in the days after the 2019 polling failure.

How many polls?

This year saw 16 federal Newspolls released, the fewest since 1991.  Essential released 13 "2PP+" results (see my comments on this method here) as figures, and a further sixteen as dots on a graph. Morgan provided this mess:

* five polls for which it issued polling reports

* two further polls on its voting intention table

* at least one further poll result where the poll was not published but the 2PP can be inferred based on the stated poll to poll changes

* fourteen graph dot points (giving the impression of being weekly breakdowns from fortnightly polls) that include:

- eight readings that align exactly in time with four polls for which fortnightly 2PPs were public

- two consecutive readings that are out of whack with one of the public 2PPs by a week

- four readings that don't align with any of the otherwise public polls. 

Thus the total number of Morgan 2PP readings of which some kind of evidence was available publicly is at least 22 (to date) but this includes at least four double-counts.  

Applying a minimum standard that a pollster must publish a 2PP (and not just by reference to changes from a previous poll) to be included, there were 16 Newspolls, 13 Essentials and seven Morgans, for a total of 36 polls.  If this figure is used it continues the downward trend in the number of published polls per year.  However the number of 2PP readings compiled by the pollsters but not published other than as dot points is considerably higher; there might be as many as 63 independent readings, with others still to be (maybe) retro-released.

The mess created by Morgan in particular does no favours for the image of polling and I hope the new Australian Polling Council will recommend to its members standards that will avoid and discourage ad hoc decisions about when to retro-release polling data.  If a pollster wants to release voting intention polls in batches, as Essential are doing, that is one thing, but the pollster should then at least be transparent about when they will poll and ensure that all the readings are retro-released, not just some.

2PP voting intention

Although it is not possible to determine exact 2PPs from most of the graphed Morgan and Essential dot points, it is possible to say who was in front.  The government won ten, tied two and lost four Newspoll 2PPs.  Essential had it winning 15, losing 12 and (I think) tying two, and Morgan had it losing five graphed dot points and one retro-published 2PP and winning the other 16 readings of various kinds.  

The Coalition's largest published 2PP lead was 54% in an August Morgan.  Its worst published-poll 2PP was a 47% in a Feb-March Morgan though the Morgan graph refers to a 45% result attributed to "Jan-Feb 2020", presumably a one-week breakdown.  

A national ANU poll of dubious quality was also released (with incomplete primary breakdowns, no 2PP and a strong appearance of overestimating the Green vote).  

Because of the mess created by Essential's retro-releasing and Morgan's ad-hoc partial retro-releasing I have not attempted to conduct a running aggregate during 2020.  However the following is a simple unweighted aggregate of released 2PPs (polls that include data across two months are split between those months):

The year splits into two parts - the "bushfire" polls of January through March in which Labor generally led, and the "pandemic" polls of April through November in which the Coalition nearly always led, but usually not by much.  The Coalition's average result per month was a 50.4-49.6 lead, which makes 2020 the first year since 2010 (Rudd/Gillard) in which the government has led on average through the year without a mid-year change of government.  Previous years in the Newspoll era in which I have the government on average winning the 2PP are 1986-1989 (Hawke), 1997, 1999, 2002-3 and 2005 (Howard, also that part of 1996 he was in office for) and 2008-9 (Rudd). So overall this happens a bit over a third of the time.  It isn't a feat to be sneezed at given that seven governments have been re-elected, mostly comfortably, in years in which they didn't win the average, but that might also raise the question of how much it actually means.  

Following on from the 2019 polling failure in which all active pollsters overestimated Labor's 2PP by around 3%, it may well be asked whether this picture is accurate.  Both Newspoll and Morgan underestimated Labor at the 2020 Queensland election, but not grievously so, and this isn't much use because the history of state election polling errors is that most polls err on the same side for any given election, but that side changes from election to election.  Also, that history isn't much use in predicting federal errors.  There's a lot of room for scepticism about whether the pollsters have really cleaned up the mess, and only one of the three pollsters (Newspoll) appears to have made major targeted changes of a logical and promising variety.

A disconnect between sky-high leadership ratings for Scott Morrison and lukewarm voting intention leads for his Coalition has been present most of the time since April.  This disconnect was also seen in Queensland, and there the leadership ratings proved the better portent of Labor's convincing win. The Eden-Monaro and Groom by-election results, in which the government outperformed the historic average 2PP swings in Opposition and Government seats respectively, were both consistent with the government being in front (it would be a long bow though to conclude that they showed the polls to be wrong and the government to have a bigger lead).  

Leaderships

The year in Newspoll leadership polling statistics also falls neatly in two parts - the bushfire phase up to mid-March, and the COVID pandemic phase thereafter.  In the bushfire phase, Scott Morrison was unpopular, reaching a net satisfaction minimum of -22, and Anthony Albanese won the first three out of four Better PM polls, a relatively rare feat for an Opposition Leader.  In the COVID phase, Scott Morrison polled netsats in the range +26 to +41 and led as Better PM by between 24 and 35 points (noting that around 15 points of this lead is the incumbent's house advantage).  Morrison has been above net +30 now for longer consecutively than any other PM in Newspoll history.  Overall the averages for the year were:

Morrison netsat +22, the highest for a PM who lasted the whole year since Kevin Rudd (+34) in 2009

Albanese netsat +3, the highest for an Opposition Leader who lasted more or less the whole year since Rudd in 2007, or if Rudd is disqualified because he was elected PM after the final Newspoll for the year, Mark Latham in 2004

Morrison Better PM lead 22.1, also the highest for a PM who lasted the whole year since Rudd (+44) in 2009.

Morrison set all-time records for the largest poll-to-poll net satisfaction increase and Better PM lead increase early in the pandemic, but this was partly because he was coming off a low, bushfire-singed, base.

Albanese: the consensus and the data

There is a consensus among commentators that the Government has Labor's measure and Labor's leader's measure in particular.  Anthony Albanese is often seen as lacking cut-through, as having failed to remedy the Opposition's difficulties in connecting to its traditional base, and at the same time as being both a captive of the left and a captive of the Government.  It's hard intuitively to disagree with any of this or to see that Albanese has any kind of plan with any promise of success.  He is even being compared to Simon Crean by prospective knife-sharpeners, but at the same stage of the cycle Crean was polling -35 netsats, losing Better PM by almost 50 points and usually losing the 2PP 47-53 or worse.  Albanese is so far, mostly, holding up much better in polling, in the context of a pandemic that has seen many Opposition Leaders polling appallingly, with some oppositions (as in NZ and WA) ending up in a shambolic condition.  The Labor primary vote is about as low as during the comparable Crean phase, but this is offset by a higher Green vote and smaller Coalition lead, leading to a more competitive 2PP. 

The problem for those who want to make poll-based arguments for ditching Albanese and replacing him is that the data-based arguments are not that strong.  Yes the government is ahead mid-term, which is normally not a good sign for Oppositions, but then again almost everything in polling is a bad sign for federal Oppositions for the simple reason that they usually lose.  Even the kind of polling seen from governments that have lost was seen in the last cycle for a government that won.  It's not clear that there are any polling indicators that would identify a mid-term government as more likely to lose than win with any sort of reliability (because the sample of losing governments is so tiny).  So if an opposition leader is not polling appallingly, can any assessment of them as a probable loser - even if sound - tell us anything?  It seems that those using poll-based arguments for disposing of Albanese have made the judgement that he should be rolled first, and then started seeing the data as supporting that claim.  

Betting

Betting is not a reliable predictor, representing mainly the collective opinions of people who are losing money (or, early in the term, the judgements of bookmakers).  Odds I could find had implied chances for the government between 54.5% and 64.5% with an average of 61%.  This isn't overwhelmingly high, but note that these odds price in the possibility of Labor changing leaders.  As to whether they will do that, one market has Jim Chalmers as a slightly better than even favourite to be the next Labor leader, but I haven't seen betting on whether or not Albanese lasts til the next election.  Markets on when the election will be held still see about a two-thirds chance of it being in 2022.

The road ahead

After several months of polling stasis, things might get more bumpy in 2021.  We'll see the winding back of government assistance programs including the end of Jobkeeper after March, we'll see the rollout of COVID-19 vaccines through the year, and at some point in either the second half of 2021 or the first half of 2022 comes the election.  How much economic or other turbulence is to come is one question, and another is how much voters will give the government a free pass for anything that does happen.  The difficulty for Labor in waiting for the pandemic boost to go away is that if Australia's relative success in combating COVID-19 continues, it could last all the way til the next election.

PS: BludgerTrack relaunched

William Bowe has relaunched a voting intention version of BludgerTrack, showing rhe Coalition currently 50.8-49.2 ahead (if the polls are to be broadly believed), and a very similar if naturally smoother pattern to my monthly averages above.  The data used for the two are almost entirely the same (I have used two more Morgans).