Monday, December 27, 2021

2021 Federal Polling Year In Review

 At the end of each year I release a review of federal polling results for the year.  See the 2020 edition here, or click the "annual poll review" tab for articles back to 2014. 

How many polls?

Again the business of counting poll results for 2021 has been complicated by Morgan releasing eight 2PPs as dots on a graph only, and by not yet having anything from Essential since mid-November. (Essential drops its voting intention results in a batch every three months though has said it will be releasing them faster as the election approaches).  

This year saw 16 federal Newspolls released, tying 2020 as the sparsest year since 1991.  Essential has so far released 23, with probably one more to come.  Morgan released 15 as full poll reports, plus another eight that have so far appeared solely as dots on a graph.  New player Resolve Strategic graced us with eight.  That's a total of 62 fully released and eight as 2PPs only, putting us up again to around a level of poll frequency last seen in 2018, but well below the years before.  


2PP voting intention

Of the four active pollsters, only Newspoll publishes 2PPs that are based on preference distributions from previous elections (though the exact formula is not released).  Morgan uses respondent preferences, Essential uses respondent preferences unless the respondent is unsure (in which case last-election preferences are used) and Resolve is routinely put in stocks and pelted with rotten tomatoes on this site for not releasing 2PP estimates at all.

In the published figures, the Government won one, tied four and lost 17 Essential 2PPs (by my last-election 2PPs this becomes three wins and 19 losses).  The Government tied three Newspolls and lost 13.  The Government lost all 15 released Morgans, but one of those becomes a win if last-election 2PPs are used.  They also won one of the eight 2PPs released as dot-points only (and may well have won more if full preferences were released).  However, things appeared rosier for the Government in Resolve Strategic's figures, in which my estimates have it winning three of the eight 2PPs.  

Overall the government averaged 48.9% from Essential and 47.9% from Newspoll.  The eight Resolve figures convert to an average of 49.1%.  The fifteen fully published Morgans convert to 47.9 by last-election preferences but 46.9 by respondent preferences.  With the partly published Morgans added the latter figure rises to 47.8.  Treating the part-published Morgans as if they were last-election preferences (which there is some case for doing given that early in the year the Morgans were not showing much preference method difference) the figure rises to 48.2.  

A possible cross-poll last-election average from all pollsters for the year, weighted for periods of not polling but not otherwise, is 48.4% for the Coalition.  Predictively, this means more or less nothing. It's a familiar story: just assuming that the 2PP will be 50-50 would have a lower average error, and since the mid-1980s, four of the six governments with worse averages than this in the year before an election have won, as did six of the seven governments with better averages (Howard 2007 a marginal exception).  

The government's best 2PP polls of the year were in January, when it polled a 51.5% 2PP in an unpublished Morgan and a 48-45 2PP+ (=51.6% 2PP) from Essential.  There was also a Resolve in September that converts to about 51.1, but it was probably an outlier even by Resolve's standards.  The government's worst by respondent preferences (and the worst headline 2PP from anyone for a sitting government since September 2015) was a 43.5 from Morgan in December.  That Morgan was also the worst estimate I have by last-election preferences (45.7) though it's possible that one or both Newspolls that came to 46 after rounding were as bad.  

I still haven't found the time to construct a proper aggregate for this term, but again in the meantime here is a pattern of crude monthly averages of all released polls, with last-election preferences used for Morgan and Resolve.


The lead the government held since April last year was lost in February.  The general pattern since then has been a further gradual 2PP shift to Labor (despite the friendlier Resolve readings kicking in from April) though with a marginal and fleeting improvement (0.5%) following a well-received Budget and a similarly short-lived rally in September.  I've also looked at the graph with Resolve taken out, but it doesn't look all that different.

Morgan pinpoints the February crossover point as coinciding with the Brittany Higgins rape accusation and subsequent fallout though obvious signals of that were not visible in polling at the time and there is a long history of government polling falling off the perch in February.  It's hard to distinguish specific events through the year from a long pattern of decline in the government's position (though the further sharp fall in December is only based on one and a half polls and should probably for now be ignored.)

Deluxe aggregates are visible at Bludger Track and Mark the Ballot.

Adventures in respondent preferencing

The year in polling can be divided into two phases: the first third or so with three pollsters giving pretty much the same story, and the last two-thirds with four pollsters each saying something rather different.  Resolve has of course been off on its own both on the independent vote and the implied 2PP.  An excellent Armarium Interreta piece has showed that historically polls that have outlying results as concerns specific parties or groupings at election time strongly tend to be wrong.  It will be interesting to see whether Resolve's oddness with the independent vote persists, herds or is joined by other pollsters as the election approaches.  However during the latter part of the year Essential has tended to find weaker ALP leads than Newspoll, while Morgan has tended to find bigger leads specifically because its respondent preferences have become very skewed to Labor.  The preferencing divergence in Morgan's estimates for the Coalition 2PP can be seen to have increased through the year:


The share of Others preferences going to the Coalition in Morgan has dropped from about 38% early in the year to 31% now - which is not very believable given that the Coalition got 34.5% of all preferences in 2010 when the Greens vote was nearly two-thirds of all available preferences (compared to around 40% in recent polling).  In contrast in Essential, this shift hasn't happened at all:


Respondent preferences have a history of skewing to Labor and are unreliable for that and other reasons, but there is more going on here. The recent breakout of United Australia Party votes for all the Essential polls and the most recent Morgan shows that Essential has had Others excluding UAP and One Nation on around 4.5-6.5% all year, while Morgan's most recent poll had such Others on 12.5%.  In 2019 this category polled 8.31%.  It may well be that Morgan are picking up - and perhaps oversampling - intending voters for what I call "climate independents" (whose preferences flow about 70% to ALP) while Essential aren't picking them up at all.   

Unfortunately because Morgan persists in not joining the Australian Polling Council there is no information about how Morgan designs its voting intention readout, while for Essential we know that "Independent or other party" is on the list of options everywhere, but always near the bottom.  What is known about Morgan is that it currently samples voting intention as part of its lengthy Single Source surveys and it would not be surprising at all if the length of those surveys alone was producing some skew.  When Morgan was conducting this survey face to face it overrepresented graduates; it is unclear if this is still the case, and also if Morgan weights or sets targets based on education.

Leaderships

This year Scott Morrison averaged a Newspoll netsat of +9.6%, but his old stuff was better than his new stuff and he had declined to a (still reasonable) -8 in the last two polls of the year after being as high as +32 in February.  Anthony Albanese averaged -6.3%.  Morrison's average Better Prime Minister lead was 19.1% but this also fell from as high as 35 points in February to 8 and 9 points in the last two polls.  In historic terms, the pattern of Morrison performing well on this skewed indicator given the state of voting intention at the time has continued.

My federal Newspoll records page was little troubled for updates during the year.  Top-five streaks of positive netsats for the PM and the PM having a higher lead than the Opposition Leader both came to an end in August and November respectively.

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).  Current odds have Labor as a modest favourite to form government with an implied chance in the range 55-58%.  At the moment there is an arbitrage in Sportsbet odds about whether or not there will be a hung parliament, since the yes/no odds imply a 73% chance of a majority government while the odds broken down by party that wins imply only 53%.  Individual seat odds betting I have seen predicts the Coalition to lose Boothby, Pearce, Chisholm, Deakin (!), LaTrobe (!!), Swan and Bass with a split market on Longman but some seats being touted as possible Coalition gains in NSW are not even on these markets yet.  While the Deakin incumbent is a liability because of his factional involvements, for seats like Deakin and LaTrobe to be in play as suggested by TAB off a uniform swing, Labor would have to be getting over 58% 2PP in Victoria, which it only ever got in 1929.  It is not even on that in the current Newspoll quarterly.

The road ahead

The election will be held by (and in my view probably in) May.  At the moment the state of polling is not much of a guide as to what will occur.  Historical regressions off current polling mostly lack predictive power, and to the extent they do predict anything they are pointing to a roughly even contest (in general the predicted Coalition 2PP lands within a point of 50).  This is because governments that are trailing in lead-up polling tend to recover.

The drift in polling towards Labor over the last few years seems consistent with the idea that the Morrison government won an election too many and that it would have been downhill most of the way except for the pandemic bounce, much as the Keating government 1993-6 was only competitive when the Liberal Party was having leadership disasters.  However we are still to see any of the large Opposition leads that have presaged most previous changes of government, so it's hard to see any objective basis for saying Labor are clear favourites.  The consensus of those engaged in insider chatter about seats is of another seat-by-seat contest with little turnover or 2PP change and a fairly high chance of a hung parliament, but there remains plenty of time for that to not be true.

1 comment:

  1. Morgan's respondent preferences have become very skewed to Labor? There could be a reason for that - that people favouring the minor parties have seen through the Flim-Flam Man's bullshit* and really intend to prefer Labor regardless of how they have voted before. Time will tell whether they stick with that till election day. I know that in the past people have reverted to their historical preferencing behaviour but maybe not this time.

    *See Harry Frankfurt on bullshit: a liar knows what the truth is and that he/she is lying; a bullshitter simply doesn't care whether he/she is telling the truth as long as it will persuade someone. A background in marketing is great training for a bullshitter.

    ReplyDelete