Poll Roundup: 50.5 Is The New 51

2PP Aggregate 50.6 to ALP (-0.1 since end of last week)
Aggregate has changed little since loss of about half a point in mid-July
Election "held now" would probably produce minority Labor government

I haven't done a federal poll roundup for a while and today is a randomly opportune time to do one following the second straight 50-50 Newspoll and mention some general themes in recent 2PP polling.  In the last week we have had:

* Newspoll at 50-50 (ALP 32 L-NP 38 Green 12 ON 7 others 11)

* Redbridge at 50.5-49.5 to ALP (ALP 32 L-NP 38 and the rest not published yet, but I'm expecting Greens either 10 or 11)

* YouGov at 50-50 (ALP 32 L-NP 37 Green 13 ON 8 others 10) (Note: normally the 2PP for these primaries would be 51-49 to Labor, though it is possible to get 50 from these primaries sometimes because of rounding and perhaps also the makeup of others.).  

* Essential at 48-46 to ALP, equivalent to 51.1-48.9 (raw primaries ALP 29 Coalition 33 Greens 13 ON 7 UAP 1 others 11 undecided 6 - meaning the major party primaries are effectively more like 31-35)

* Morgan at 50.5-49.5 to Coalition by respondent preferences (50-50 last election) (ALP 29.5 L-NP 39.5 Green 13 ON 4 IND 9 others 5 - Morgan has a standalone IND option on the ballot everywhere, which is likely to be overstated)

I could have written this article any time in the last two and a half months and it would have looked a lot like that.  I could have written it any time since March and it would have been not very different.  Even before that not a great deal happened since November.  This has been one of the most becalmed periods in federal 2PP polling you would ever see - but still what has happened isn't quite nothing.  Here are my tracking graphs:

From about late March my aggregate sat at around 51 to ALP with remarkably little week to week variation.  In the first week of July it made it up to 51.4 (a three-month high), but then it dropped in the next two weeks.  The next week had only a single poll (Morgan) and it fell to 51.1, then the next week it fell to 50.6.  The most obvious event preceding this slump was the departure of WA Senator Fatima Payman from the party on 4 July, but I am not saying this was necessarily the cause.  

Whatever the cause the half point or so lost at that time has stayed off.  And while it may seem like half a point dropped in a polling aggregate is no big deal (certainly less than the margin of error of the aggregate at any given time), nonetheless the shift from 51-ish readings to 50.5-ish readings was very statistically significant, even after only three weeks of it.  The reason for this is the stability of polling on both sides of the shift.  

The incredible shrinking government act of the first-term incumbents has certainly attracted some very jaded media and social media reviews.  And yet, for everything they've had to deal with (this week including a farcical self-inflicted census stuffup and semi-reversal that I doubt moved any real number of votes), the government still clings (on last-election preferences) to the aggregated last-election preferences lead it has held since the election.  It is now into its 28th month in the lead.  In about four weeks if it stays there it will pass the Howard Government in 2001-4 (essentially, Tampa to Latham). In eight weeks it will pass the Rudd Government (2007-10) and have the longest polling lead that any government has held since there was only one infrequent and unreliable poll to aggregate (Menzies/Holt/Gorton in mid-late 1960s).  If, that is, it can actually stay in front for that long.

I should note again that some aggregates that use pollster-issued preferences have already crossed over.  The reason for this is that Morgan and Essential both use respondent preferences as a headline figure and Essential's have this year run 1.8% lower for Labor than my estimate of last-election preferences for their poll.  (As stated before, I don't think this is likely to be real.)  Morgan's have run 0.8% lower.  Respondent preferences are less reliable than previous-election preferences, but as the size of the third party vote grows, shifts in preferencing patterns are having more impact on the 2PP.  If the respondent-preference polls are half-right, Labor's current situation is even more lackluster than my aggregate suggests.

Does the 2PP lead really matter?  Probably not.  I estimate the government needs low 51s for a 50-50 chance of a majority, and 50 point not much or just below are likely to be different shades of probably bearable minority.  Things could get much uglier if the government loses the 2PP 49.0-51.0, especially if that comes with high Green and Coalition primaries.  I have looked at this in modelling some of the polls that have come out with that 2PP, especially the two recent Freshwaters.  I concluded that the AFR's reading of those polls was actually optimistic for the Labor side.  49-51 is a loss of only seven seats by the pendulum, but that is treating Aston at its by-election margin whereas something between its last-election and by-election margins works better for a "disrupted seat".  Also, in the draft redistribution pendulum, the 3-4% swing zone has four seats on 3.3% or 3.4% that would be about even chances to fall all else being equal, so my probabilty-based seat model puts Labor on only 68.5 seats if the 2PP is 49.  Throw in the couple of losses to the Greens that would be likely to occur off a low Labor primary, add in some Coalition recaptures from the teals and 49-51 looks on average like a serious mess, and possibly even loss of office.  This is even without mentioning the Muslim independent push on in western Sydney and north-west Melbourne, though I think it will be difficult for the Coalition to justify recommending preferences to anyone remotely concerned about Gaza while trying to tell Labor that it shouldn't preference the Greens.

49-51 seems like quite a different landscape to the current polling.  On present numbers (50.4) my model has Labor on about 73 seats, and while it might drop one or two more to the Greens that would still be a reasonably manageable minority government.  Ideally if Labor is to lose its majority, it would like to have paths to passing legislation through the Reps that do not go through the Greens, even if the Greens can still block in the Senate.  

Leaderships

There is not too much to say about leaderships at the moment.  Both leaders are modestly but not very unpopular (both at net -13 in this weeks Newspoll, an equal lowest so far for Anthony Albanese.)  Albanese continues to lead in most preferred leader polling, but by margins that are below historic averages as preferred leader polling skews to incumbents.  (For instance his current eight-point lead in Newspoll is below the long-term average lead of 14.7 points for an incumbent PM).  

Simon Benson wrote that Albanese's and Dutton's "combined poor standing among voters is virtually unrivalled at this level" but that is hype and nonsense.  In the history of Newspoll nearly one-sixth of federal Newspolls have had both leaders at net -13 or worse.  This recently included nearly every Turnbull vs Shorten poll between the 2016 election and May 2018, and also the last Newspoll before Turnbull was removed.  Five leader pairs have at some stage both together been worse than -13, and all of those were much worse; the following was as bad as it got for those combos:

-30 Keating/Hewson

-27 Keating/Downer

-31 Gillard/Abbott

-28 Abbott/Shorten

-27 Turnbull/Shorten

Heck, I was writing about this sort of stuff in the Gillard days before I even had this site!

Your Unlucky Word Today Is ... Underdispersal!

Recently the last-election preferences I have calculated for Essential have been favourable to Labor compared to the average of all polls, though their mostly respondent-preference "2PP+" has been anything but.  The difference is worth about a point, and I'm still not completely sure it's a thing rather than a random streak (it's at around the 1/100 chance level). But in aggregation you don't need to be completely sure to make changes - especially as this is something Essential has done before - and I've made a small adjustement for it for the time being. 

As polling has moved away from the last election, Essential's primaries are looking more like the last election than other polls are.  This effect was very obvious in the previous term, when it was attributed to Essential's use of party identification in weighting.  They since switched to using 2022 election vote, but I can see why that would be prone to the same thing - at least if it is asked freshly of the respondent (which it appears Essential does) rather than held on file from just after the election as YouGov do.  At times when a party is on the nose, voters who voted for that party without much conviction last time are more likely to not only switch their vote but also misremember how they voted in the first place.  Hence, those who identify themselves as voters for that party are more likely to be committed to it and less likely to switch.  Likewise when a party is doing fantastically, voters are more likely to over-remember voting for it.  

That's one thing - another is the general lack of difference between results from different polls at the moment.  A month or two ago it was common to see people complaining that the polls were all over the place, how to know what to believe when one poll has Labor at 52-48 and another has them at 49.5-50.5, and so on.  In fact, the truth at the time was that the polls were not very variable and the variation just looks greater to unwary observers when it's close to 50-50 than it does when one poll says 55 and another says 58 (for example).  

Even if there are not big differences in what the different polls are getting on average - as has been the case since Resolve suddenly ceased being way better for Labor than other polls at the start of the year - one normally expects to see a few outliers; if Labor is on about 50 you'd expect to every now and then see a random 53 or 47.  Most of the polls have claimed margins of error of around 3%; that implies a standard deviation of about 1.5%.  So if absolutely nothing is happening with underlying voting intention and there are absolutely no house effects between polls (the latter being not quite true but close) then the average poll will vary from the mean by about that standard deviation.

The 29 polls with pollster-released 2PPs since mid-June have been remarkably clustered with an average for Labor of 50.06 and a standard deviation of just 0.88.  There hasn't been a single reading in this time below 48 or above 51.5.  Even if I go back to mid-March the standard deviation is just 1.19, and all that in a sample which includes a lot of Morgan, which is often quite a bouncy poll.   If I use not the pollster-released rounded 2PPs but my own estimates from their rounded primaries, the standard deviation is 1.01 since mid-June and 1.08 since mid-March,  What's going on here?

There are actually some reasons why polls can vary less than their published margins of error imply, especially where the 2PP is concerned.  Firstly a last-election 2PP is a derived calculation; if there is a random undersampling of a major party and over-sampling of Others, then that has less impact on the 2PP because the Others are still counted about half to that major party on the 2PP.  So headline previous-election preference polls (which is at least Newspoll, YouGov and Redbridge) should all else being equal be less volatile than respondent preference polls (and historically are).

Secondly Newspoll and YouGov in particular use targeting to try to make their sample representative from the start instead of relying on heavy reweighting. If you're constraining your sample to have, hypothetically, about the right number of 65+ year old blokes from rural Queensland and about the right number of uni graduates aged under 35, then the former are going to heavily lean to Coalition and the latter to the left parties, and that's going to make your overall results less variable than if you were just randomly sampling a bunch of voters who have a 50% chance of preferring Labor and the same of preferring the Coalition.  

While I'm aware of all that, I am surprised by the low level of variation in polling results lately (not mainly the nearly flat trend since late March but more the scarcity of outliers).  And I think I should mention the less attractive alternatives while stressing that I have absolutely no evidence that any particular pollster is doing it. Some pollsters could in theory be herding (making methods decisions that are influenced by other polls or aggregates of polling), self-herding (using unpublished adjustments that reduce poll-to-poll volatility), or what I call zeitgeist-herding (being influenced by an expectation of what the polls should be saying that isn't necessarily based on other polls).  However, there's no sane reason for any pollster to be doing the first or the third this far from an election.  

The Australian polling industry had a disaster in 2019 when multiple pollsters produced ridiculously similar polls in the final weeks that then all turned out to be wrong in the same direction.  Although the formation of the Australian Polling Council has gone some way to trying to discourage a repeat performance and to at least ensure we know more about how polls work, only four of the seven regular pollsters are members, and some of those four are more enthusiastic about disclosure in a detailed and organised fashion than others.  Even in those cases the disclosure is still much less than often seen overseas.  We still do not know enough about Australian polling to be sure that everything is under control when something looks odd.   

(I am massively tempting fate here of course.  By writing these words I am bound to cause tonight's Morgan to be a monster rogue).  

Update Monday night: not a rogue as such but a fairly good poll from Morgan for Labor: 50.5 by respondent preferences, 52 by last-election.  ALP 30.5 L-NP 36 Greens 13 ON 6 IND 9.5 other 5.  My aggregate just made it to 50.6 off that.