2023 Federal Polling Year In Review

2PP Aggregate Average For 2023: 54.8 To Labor (2023 Preferences)
Labor won almost every poll this year
Labor's lead declined in second half

What I think may well be the last federal voting intention polls of the year have come out and at this point it's time for a regular site feature, my annual review of federal polling.  The 2022 article was here and for earlier articles back to 2014 click on the annual poll review tab.  If any more polls come out I will update this piece accordingly, but perhaps not very quickly.  

2023 was another successful year for the Australian polling industry.  Final polls were rather good (if mostly a bit light on for Labor) in the NSW state election but the biggest test came in the October Voice referendum.  In the face of poll denial levels so out of control that I wrote an article about it, the industry recorded an outstanding result (especially by referendum standards), although a minority of polls had big misses on the Yes side.  

Even without the very richly polled referendum, there was a lot going on in Australian polling this year.  The most dramatic event was the YouGov breakup in which Campbell White and Simon Levy left and started Pyxis Polling and Insights, with the former YouGov continuing with Newspoll for just one poll before it was transferred to Pyxis.  Both Pyxis and YouGov were able through the chaos to be among the best pollsters on the Voice, and polling has been bolstered for now by the addition of a YouGov series that is a Newspoll clone in wording but uses an increasingly different weighting and targeting structure (now even including Voice vote).  

There was also a change at Essential, which belatedly added education to its weighting frame after big misses in the Voice and New Zealand.  This appeared to have quite an impact in the first poll after it happened, but less so since.

The current batch

It is worth a quick comment about the current batch of polls before I move on to the yearly statistics.  Things had been closing up in late November, especially with a 50-50 Newspoll, but in the last few weeks it seems that things have at least settled and perhaps even that Labor has a little tinsely uptick of the sort that's not uncommon at this time of year.  (It often doesn't last.)  Starting this week at an estimated 51.4 to Labor in my recently restarted aggregate, we had a pollster-released 52-48 to Labor in Newspoll, 50-50s in Freshwater and Morgan and 52.8-47.2 to Labor in Redbridge.  Freshwater and Redbridge respectively tend to produce good results for the Coalition and Labor compared to most other pollsters specifically, and Morgan's numbers are lately better for Labor when last-election preferences are applied (I get 51.3 in this case).  My aggregate also weights Newspoll most heavily so the net impact of all these changes was a slight rise to 51.7.  Things were heading downhill fast for Labor four weeks ago but the rot has for now stopped.

How many polls?

This year Morgan has been much more proactive in terms of releasing or at least back-releasing full primary vote data.  As a result there have been 50 Morgan readings, 21 Essentials, 15 Newspolls, 11 Resolves, 4 post-split YouGovs, 4 Redbridges and 3 Freshwaters for a total of 109.  Largely thanks to Morgan's weekly readings this is the most federal polls in a year since 2015 which had 115 polls.

2PP Voting Intention

There were 98 pollster-released or back-released 2PPs this year, of which Labor won 92, tied four and lost two.  Resolve does not release 2PPs outside election times but on any sensible reading Labor would have easily won all those.  The four ties were two Morgans, one Newspoll and one Freshwater.  The two "losses" were both 49.5-50.5s from Morgan, one caused by a rogue respondent preference flow; it is worth noting that even if that was the election result Labor would be very likely to win in minority.  The highest (back-)published pollster 2PP was 59.7 to Labor from Morgan in the second week of January.  

I conduct last-election preference conversions off the primaries of all the published polls.  On my estimates Labor won the last-election 2PP for all 109 polls this year, though that includes one each from Newspoll, Morgan, YouGov and Freshwater that round to 50-50 to one decimal place.  The highest last-election conversion was 61.5 from Resolve in April.  Excluding Resolve which has been far better for Labor than other polls, the highest was 59.4 from Morgan in mid-January.

The average of my aggregate for the year was 54.8 to Labor, just down on their 54.9 in my article last year.  But this average is pulled upwards by some very high readings early in the year.  My aggregate shows a levelling out at about 56-44 from late February through to June, and a clear downhill trend from the middle of the year.  I have linked this to a sharp rise in voter pessimism reflected in a big upswing in voters saying the country was on the "wrong track" or heading in the "wrong direction" as opposed to the right track/direction from June onwards.  Cost of living issues are likely to be the primary factor in this narrowing, which brought an end to the run of lopsided honeymoon polling in early September.  One can speculate that running the Voice referendum did not help but it's hard to isolate a point for that as the damage was spread out across several months.  Here are my annual tracking graphs:

It should be noted that this tracking assumes preferences will flow as they did at the last election.  Morgan (which allocates preferences by respondent preferencing) and Essential (which allocates preferences by respondent preferencing unless the respondent is unsure, in which case last-election preferences based on their primary vote choice is used) both find a pessimistic and worsening picture for the government on that front.  In the first half of the year Essential's 2PPs averaged 0.6% worse for Labor than my last-election estimates implied, but Morgan's averaged 1.0% better.  Between July and the mid-October Voice referendum defeat Essential had 0.85% worse for Labor, Morgan had 0.1% worse.  Since the Voice defeat Essential has had 1.3% worse for Labor, Morgan 1.0% worse.

Last-election preferences have outperformed respondent preferences over many years but they are less reliable now than in their golden age from 1993 to 2010.  As the non-major-party share of the pie gets bigger, slight swings in preferences have more impact on the 2PP overall, with three of the last four elections seeing shifts worth about 1% (to Labor in 2013 and 2022 and to the Coalition in 2019).  Therefore at present it is possible that last-election preferences are overstating Labor's position and their 51.7 might in reality be more like 50.7.

Another point to note here is that Labor tends to perform more weakly compared to its during-term polling than the Coalition does; what I refer to as the "Labor fail factor".  One possible interpretation is that voters simply tend to be more likely to prefer Labor during a term than when they have to actually decide to vote, but another is that it's a systematic issue with mid=term polling.  For this reason while Freshwater is a persistent leaner to the Coalition side, its readings shouldn't be dismissed.  

With the Albanese government losing support through the year, are the minor parties picking up?  A little bit, but not that much.  In the second half of the year the non-major party vote (as an unweighted average of all polls) averaged 31.1% compared to 30.1% in the first half, a gain of only 1 point despite Labor's primary being 3 points lower.  Even in the post-Voice phase with Labor down 4.4% on primary on the first half of the year, the Coalition has taken about two-thirds of it.  The combined non-majors polled 31.7% in 2022 and at this stage aren't doing any better than that.  

Leaderships

Anthony Albanese's average net Newspoll rating for the year was +7.1 but his ratings mirrored the government's downhill trend, with a high of +24 in early February and a low of -13 in late November.  Peter Dutton averaged net -13.5 with a high of -9 in the last poll of the year (mid-December) and a low of -20 in late September.  On the Better Prime Minister front (this indicator skews to incumbents by about 15 points) Albanese had an average lead of 24.1 points with a high of 32 points in late March-early April and a low of 10 points in early November.  

Albanese had a higher netsat than Dutton in 14 of the 15 Newspolls but there was one tie in late November.  Looking at other polls, Albanese recorded exclusively positive net ratings through to late August, a mix of positive, negative and zero scores from then til the Voice defeat, and exclusively negative net ratings since the Voice defeat in mid-October.  Dutton had a higher personal rating than Albanese in five polls towards the end of the year: two Resolves, one Essential, one YouGov and one Freshwater.  (Indeed at year's end Newspoll is the only poll he has the better current net satisfaction in, by a single point.)  However Dutton is yet to poll a positive net rating or a Better Prime Minister lead from any pollster.  

There are strong parallels between what has happened with Albanese's ratings in 2023 and what happened with Scott Morrison's in 2021.  However in 2021 Morrison's government recorded a polling average about six points lower than Albanese's, with a similar pattern of decline through the year.  Whereas Morrison in 2021 outperformed his party's voting intention polling (largely I think off pandemic polling effects), Albanese in 2023 has clearly underperformed his.   

There was plenty of commentariat speculation that winning the Voice would damage or destroy Dutton's leadership.  At this stage there's no evidence for that at all - on the contrary his Better Prime Minister deficit halved in the poll after the result.  

Betting

I mention betting odds in these annual roundups for interest, although betting odds are, as a canonical tweet by Ben Raue put it, "a unique combination of the conventional wisdom (fed by polling and media narrative) and a bunch of rich idiots' crazy bets."  They are not reliably predictive and are probably not even "well calibrated" (a noted issue there is longshot bias, but there may be others.)  Checking several bookmaker sites I found the government's average implied chance of re-election to be about 55.5% with very little variation.  Even this, which seems remarkably low for a first-term government that is so far yet to trail in polling and that a hefty swing is probably needed to dislodge, has followed the Newspoll upwards after dipping to about 52 a few weeks back; I believe some sites even had the Coalition favourite!  

2024 a weightier year?

The latter half of 2023 should have given the Albanese Government a fair hint of polling mortality.   The Rudd government was unstoppable in polls throughout its first two years and a sudden plunge in 2010 with an election only months away contributed to the panic that was a part of Rudd being rolled.  It may be an advantage for the current Government to have been given its wake-up call with still half its term to go and to still have got off to the Christmas break in a reasonably good polling position.  When it returns in 2024 we will see whether this year's slide continues and the government gets into trouble, or whether it can maintain a lead and put some heat on Peter Dutton's leadership.  It would be surprising if there was not some attempt at a strategic reset and probably a reshuffle to move on some poor performers.  The Prime Minister's own position may appear a little fragile but John Howard polled far worse ratings in his first term and still went on to triumph.  The current Government does not have Howard's advantages in terms of first-term new member personal votes, but it does face an Opposition with strategic headaches after losing its former heartland seats to teals, and with only a few really juicy target seats (pending the redistribution).  

2024 is a plausible election year.  First term governments generally go early, but usually have stronger reasons to do so.  The Queensland election due on 26 October and redistributions expected to complete by 17 October present some practical complications there; perhaps an election could be squeezed in despite those (say, 30 November), but 2025 is generally seen as more likely.  

Polling at this stage has almost no predictive value.  Towards the end of 2024, if an election hasn't been called by then, we should have a much clearer view of the backdrop to the next election contest.