Tuesday, October 27, 2020

Post-Budget Federal Polling Roundup

I haven't done a federal polling roundup for a long time, because most of the time at present we are only getting Newspoll.  However, last week saw the quarterly batch release of Essential's new poll results and there is actually enough information out there to make it worth sneaking in a general if slightly dated polling review in the small window of spare time I have between the ACT and Queensland election counts.  My previous comments about federal polling (or mostly, the Australian's lousy coverage thereof) were here.

This year we saw very little of a common polling trope in previous years - fevered speculation about whether the government of the day would get a "Budget bounce".  Actual budget bounces are rare, but the extremely well-received 2019 Budget not only saw an immediate lift of about 0.6% in aggregated polling for the Coalition, but also either coincided with or kickstarted a longer recovery that continued through the campaign.  We now know that all that polling was wrong, but we don't know if it was wrong by the same amount all along.  In any case, 2019 was another example of the strongest evidence (such as it was, since it could be coincidence rather than causation) for the Budgets that most help government polling usually occurring in a Coalition government's election year.  

Newspoll recently recorded a 51% 2PP for the government before the Budget and a 52% after it, which people unfamiliar with the idea of random statistical noise may have taken as evidence of another Budget bounce.  However, the evidence from Essential weakens if not completely negates the evidence for the bounce, depending on how Essential is interpreted.


Essential has released seven new readings over the previous three months (including four consecutive recent weekly readings), of which six have the Coalition ahead, with the Coalition behind in one sample in early September.  The late September sample was extremely strong for the Coalition and this complicates any attempt to assess the impact of the Budget on aggregated polling (or, at least, any attempt that uses that sample.)

As noted in my initial review of Essential's rather unusual "2PP Plus" method of presenting two-party results, Essential now uses a variant of respondent preferencing.  Respondent preferencing generally makes two-party results bouncier, and there's no reason to think it makes them more accurate.  The following graph compares Essential's released readings so far (as converted to standard 2PPs by redistributing "undecided") with last-election estimates from the released primary vote figures:


Using last-election preferences, the Coalition wasn't behind in the 7 September sample, was not as far ahead in the 28 September sample, and isn't ahead in the most recent sample.  On average, Essential's respondent preferences are returning 2PP+ results that are 0.4% friendlier to the Coalition than if they used last-election preferences, and if that is actually true even despite the presumed between-election subsiding of the United Australia Party vote, then Labor has a big problem.  The track record of respondent preferences is that such things are often not true.  

In the previous article about Essential's new methods, I pointed to an apparent house effect of their recent polling compared with Newspoll.  Here's an updated version of the graph comparing each Newspoll's reading with the closest comparable Essential reading (as converted to a standard 2PP):


For the four Newspolls added, the average difference I previously noted hasn't been apparent.  Not all the Essential readings are included here but in any case the average of readings for the two in the last three months has been more or less identical: a Coalition 2PP of  pretty much the same as the 2019 election (51.2%).  Even assuming that these polls are accurate, this means very little looking ahead; the government will have different challenges as the rivers of free money run dry when Jobkeeper ends, and what the government's polling will look like in 12 months is not something I'd care to predict.

I am not currently running a federal polling aggregate because at most times there is only Newspoll to process.  However, I've crunched some numbers for the last few weeks using similar assumptions to those I used in past aggregates, and once Essential is included the evidence for a budget bounce disappears; I get a shift of 0.1% to Coalition (though that is still better than the historic average trend, which is that governments go slightly backwards post-Budget).  

Leaderships

There has been little to see on the leadership front in recent Newspolls, with Scott Morrison continuing to enjoy high net satisfaction (most recently at +34) and a larger lead as Better Prime Minister than the house effect for that indicator (most recently 57-28).  There might be some room in Labor for concern about Anthony Albanese's declining personal rating, which has fallen fifteen points from +11 in late April to -4 (39-43) now.  If it falls too much further, it will start to look a bit Bill Shorteny.  Any time that goes by without Morrison's rating going sharply downwards is probably good news for the Coalition, because it means that as far as voters are concerned he's not seriously stuffing up - but we have also seen that Australian voters are giving leaders a lot of latitude in these difficult times.  

Newspoll Budget Polling

The 2020 Budget (see tables here) was, again, well received, though not so much as its predecessor.  The two major Newspoll indicators of budget reception are what voters think of the Budget's impact on the economy, and what voters think of its impact on them personally.  Here is my updated graph of these, showing where 2020 fits in, and also labelling various other Budgets that are recent or of note:

Red dots are Labor Budgets and blue dots are Coalition.  This was one of only seven Budgets that voters have considered good for them personally - all seven delivered by Coalition governments in the years 2004-7 and 2018-20.  

Where the government did especially well was on the question of whether the opposition would have delivered a better Budget.  It had a net lead of 49-33 (No-Yes) on this score, which is the largest for a government since 2009 (when the Rudd government was, for a while, very popular).  This may play into the widespread perception that the federal Labor Opposition is just useless, but I think that's premature.  These are very difficult times for Oppositions generally, and we have seen conservative Oppositions beaten convincingly at elections in the NT and the ACT and thrashed completely senseless in New Zealand, and polling pretty awful leader ratings in Victoria and especially WA.  So maybe we should cut Labor some slack here, though I'm sure a lot of my readers on both the left and the right would prefer not to!

Is Frydenberg More Popular Than Andrews?

And now we turn to something, well, completely ridiculous.  The Australian, yes, the national premium broadsheet, published (October 20) the following claim:

"Political observers have noted he [Frydenberg] is a popular figure in his home state, winning 48,928 first-preference votes in his seat of ­Kooyong at the 2019 federal ­election while [Daniel] Andrews ­received 19,649 votes in his state electorate."

It's really difficult to describe the astonishing cluelessness of this comment without using severely ableist language.  For starters, Andrews' state electorate of Mulgrave only had 40588 enrolled voters in 2018, meaning that Andrews would only have needed 120.55% of all enrolled Mulgravians to vote 1 for him in order to match the Treasurer's tally.  So it should have been obvious that if you are going to use a candidate's primary vote as a measure of personal popularity (which is a bad idea anyway) then you should be looking at percentages, not raw primary votes.  And on this score, Andrews polled 56.7% to Frydenberg's 49.4%.

That's still an obviously silly comparison too, because Andrews' result was achieved in the context of a landslide election win for his party, while Frydenberg's came in a reasonably close federal poll.  But even ignoring that, the idea of using votes received in an election as an indicator of popularity is a foolish one because it is easier to get a high personal primary vote in some seats than in others, and in some electoral circumstances than others.  

Josh Frydenberg's primary vote percentage in 2019 was in fact the Liberal Party's worst in Kooyong since the party's inception, and ditto for his two-party preferred result.  But that's not a reflection on him either - it's a reflection that the issue mix in 2019 was bad for the Liberals in rich inner-city seats.  Sensing this, the Greens, Labor and various third party groups threw the proverbial kitchen sink at at least making Kooyong competitive enough that the Liberals would be forced to divert massive resources into sandbagging it - which they did, but still won the election anyway.  Frydenberg's result compared poorly, for an incumbent, with a demographic model, but that might be down to the profile of his Green opponent, Julian Burnside.  

Ultimately the best way to measure Frydenberg's popularity and compare it to Andrews would be to simply get YouGov to ask a Newspoll approval question for the Treasurer.  It's possible he'd do very well; he's delivered two well-regarded Budgets and has few political enemies compared to some on the right of the party.  Instead, the article as published went with a silly tactic of presenting any evidence, however irrelevant, rather than the writer just admitting he had nothing.

Nasty Seat Polls For Labor - Sort-Of

Another piece of media silliness came from Nine News on 18 October - and it's even earned a place in my hall of shame for ratioed tweets, with a fitting ratio of 6.66.  The report announced "polls revealed he'd [PM Morrison] win in a landslide, if votes were cast tonight".  At least for once the "private research" was described as "given", rather than leaked, but it hardly supported the claim.  Rather, this is two seat polls by Redbridge, who have produced a large volume of interesting research but have not yet attempted any publicly testable election forecasts that I'm aware of, and apparently these polls show 10%-ish swings against Labor in Macquarie and Dobell.  That's two seats, not 151, and you don't use seat polling (with its poor Australian track record) to measure the national mood (not even after the 2019 polling failure).  It's unlikely but not impossible that two seats would show swings of this size while there was more or less no swing nationwide (as Newspoll and Essential have had it) - double-digit 2PP swings did occur in a few seats at the federal election.  Sloppily, these results have been reported by Nine without full primary figures or sample size results.  

The Council Arrives

Finally, it is pleasing to report the formation of the Australian Polling Council, by YouGov (which also administers Newspoll), Essential, uComms, Ipsos, Lonergan, Telereach and JWS.  They've pretty much collected the full set there, except for Morgan.  Hopefully the start of an improvement in Australian poll transparency and reporting, which cannot come soon enough!

I may update this article with notes on the next Newspoll if there is anything of note to see there.

No comments:

Post a Comment