Tuesday, August 11, 2015

Poll Roundup: More Trouble For Coalition

2PP Aggregate: 53.4 to Labor (+0.6 in one week, +1.6 in four weeks, highest since start of March)
Labor would win election "held now" comfortably

After four and a half months in which voting intention hardly moved, we've finally got some real action. In the last week we saw expense claims by all sides of politics come under scrutiny following the scandal that brought down former Speaker Bronwyn Bishop.  Although Labor (and especially Tony Burke) looks far from spotless in this, the government seems to be taking all the heat.

Not too much else happened in the issue mix except that Australia's men's cricket team was dismissed for a ludicrous 60 runs and soon after that surrendered The Ashes.  As there is ample evidence that sporting results can affect electoral outcomes, it stands to reason that extremely bad sporting results that are harmful to national pride might slightly impair government polling.  Especially if they undermine this sort of thing:

(More likely, Australia actually started winning at cricket in Nov 2013 mainly because they were playing at home.)

This week's polls

This week saw the surprise release of a second consecutive ReachTEL, this one sponsored by Fairfax so they had something to go with for the return of parliament, with Ipsos scheduled for next week.  We also had Newspoll, Morgan and Essential.

The ReachTEL (53:47) was barely different from last week's but there was a slight shift in the primary votes with the Coalition dropping 0.5 points and Labor picking up 0.3.  Essential also stayed at 53:47.  Newspoll had 54:46.  Morgan had a feral headline rate of 57:43, but that was respondent-allocated.  The last-election preferences came in at a more modest 54.5 to ALP, but that said, off those primaries I'd have expected 55.   The 2.5 point gap between respondent and last-election preferences is the second-highest since the last election.

I aggregated this week's ReachTEL at 53.1 to ALP, Essential at a flat 53, the Newspoll at 54.1, and the Morgan at 53.7 after adjusting for house effect; as a result my aggregate now sits at 53.4 to ALP, its highest level since the start of March.  Here's the smoothed tracking graph:

After three readings, the new Newspoll has on average seemed a touch friendly to Labor, although that could well be just a result of random noise rather than an actual house effect.  My current estimate of the new poll's lean is 0.47 points to ALP, but this will be refined over coming weeks and it could still be that the real lean is zero.  Anyway, this appearance of a possible lean has led to me knocking another 0.1 points off my aggregate calculation method, and backdating that change to the first appearance of the new Newspoll.   I've also considered the tendency of the slightly Labor-leaning polls to appear more often than the slightly Coalition-leaning ones, and whether I should be adjusting for that. My estimate of the difference it makes is only 0.07 points and I want to see more evidence about the new Newspoll first.

The Newspoll was the government's worst poll from anyone except Morgan since a 54:46 ReachTEL in late March. It's also notable that the Greens are getting record-level readings in Morgan though other polls are more muted (and likely more accurate) concerning their support.

Other aggregators are currently at: 53.1 for Mark the Ballot, 52.8 for Phantom Trend, and I'll add in others as I become aware of their readings for this week (update: Bludger Track is 53.7).  Aggregates that use a sample-size term will probably be clustered around the 53-ish result of the two ReachTELs.  Mine doesn't.


Most of this is the usual story.  Both leaders have a Newspoll netsat of -28, which represents a fall of one point for Abbott (33-61) and a rise of four points for Shorten (29-57).  It is looking like the new Newspoll has lower undecided rates for these ratings than the old one.  Abbott and Shorten are tied 38-38 as "better Prime Minister" in Newspoll, and it looks like the new Newspoll has higher undecided rates on this score than the old one (which is not surprising since Galaxy has often had high undecided rates on preferred leader scores).  Over at Essential, Abbott is at -15 (38-53), Shorten at -23 (29-52), both slight increases from last time, and Abbott leads 36-32 (a narrowing of three points).

The most interesting leadership stat was in the Fairfax ReachTEL in which Shorten gained 3.4 points to lead 58.5:41.5 on ReachTEL's "forced answer" scale, in spite of the negligible change in primary voting intention.  Abbott was down 3.8 points among Coalition voters, 2.6 points among Labor voters and 2.7 points among Greeen voters but up 5.5 points among Others voters, the last result probably being caused by small sample size.

Other Polling

There hasn't been that much of it, but Essential found (66:18) that voters think Bronwyn Bishop should now resign from Parliament; even Coalition voters (52:35) agreed with this.  Essential also found that voters didn't think AFL fans booing Adam Goodes are racist (29:45) with Greens voters the only party thinking they were, and Others fans disagreeing even more than Coalition supporters.  Both ReachTEL and Essential predictably (and by more or less identical margins) found voters don't think much  of Productivity Commission recommendations on Sunday penalty rates.

ReachTEL's Bronwyn Bishop poll asked "Has the controversy surrounding Bronwyn Bishop's expenses made you more or less likely to support Tony Abbott?"  These sorts of questions are always hard to unravel because many locked-in voters give silly answers, saying that issues make them more or less likely to support a leader out of blind loyalty or equally blind aversion.  Thus, it's hard to take 27.3% of Coalition supporters saying they were more likely to support Abbott now seriously, or 61.3% of Greens supporters saying less likely.

Oh, and there's a missing poll by the way, and it isn't the Newspoll many misinformed people expected and then said stupid stuff about all over Twitter last week.  Rather, among the list of commissioned questions in the Seven News ReachTEL from the week before last (taken, now, 12 days ago) was one on same-sex marriage - I know this because someone taped it for me.  I suspect the reason it hasn't been released is simply that the media source commissioning it hasn't found it interesting enough to release given the rest of the news mix, but has continued to retain the option to do so.  (If it does happen, expect the result to be several points closer than for live phone polling.)  I would hope it is clear given today's events that this is a much too important issue for a commissioning source to be sitting on and concealing a poll result from the public.

(By the way, in my sidebar Not-A-Poll concerning the Speakership, the surprising early trend continued with Wyatt Roy ultimately the top scorer with 29 votes ahead of Sharman Stone on 18; none of the really serious candidates made any great impression on it.)

Textor's Tickled Polls

We've now seen the Coalition trail in the polls for twenty months straight, and by my estimate it's been in an election-losing position (if an election was held at that time) for thirteen of the last fifteen months.  Of course, polls will change between now and the election, and governments in deeper strife for just as long have sometimes won.  But what if the reason polls might change is not that opinions will change but that the opinions in current polling might not be actually real?

A Mark Textor interview in The Guardian discusses this possibility, suggesting that voters are strategically "gaming" polls to "tickle up" parties.  The voter may intend voting for party A but will say they will vote for party B in order to prod party A into doing something.  Textor doesn't say it, but the glaringly obvious candidate for "something" in the context of the current Liberal Party is leadership change.  It might also be alleged that some left-wing ALP supporters intend all along to vote Labor, but tell the pollsters they'll be voting Green.

The same Textor interview also alludes to the problem of strategic voting, in that close to an election, a voter might say they are thinking of voting A, when in fact A is their ideally preferred party but they will actually vote B for pragmatic reasons.  The suggestion is that proper question design might overcome this.

I think these two suggestions run contrary to each other when it comes to using the UK election as evidence of something that might affect Australia.  The UK strategic voting situation (with a risk of a hung parliament beholden to the Scottish Nationalists) doesn't have a parallel in this country, except for in Tasmanian state elections.  It makes it difficult to use the most recent UK election as any evidence that voters were being tricky all along, since the idea they were naive about their own thought processes at the end might well explain the Poll Fail (re which here are my previous comments) by itself.

Perhaps there is other evidence though, and the question to ponder is: if some voters are lying to the pollsters about basic voting intention, then how can we know?  Especially, can this be tested using public polling data in any way?  (It's easy enough for a pollster to test it by running parallel polls and in one case telling respondents that the results won't be published.) Suggestions are welcome.

(I do think many voters lie in response to some questions, especially ones like "Has the controversy surrounding Bronwyn Bishop's expenses made you more or less likely to support Tony Abbott?" That doesn't establish though that they lie on voting intentions.)

It's a bumpy ride for the Government at the moment, with bad polling, the contentious same-sex marriage debate and renewed concerns about the PM's leadership style.   I expect to be back next week with a brief roundup to follow the Ipsos.


  1. Does the drift away from the Coalition on preferences (in polls where respondents allocate preferences) have any impact on your fed 2PP estimate? Or is the 2013 preference flow still the best yardstick for 2PP?

  2. My aggregate uses only last-election preference flows but they are subdivided into Greens, PUP and Others. A major part of the Coalition's fall in the share of all respondent-allocated preferences is that the Greens' share of all non-major party votes has increased and that part is fully captured in my estimate.

    However there has also been a tendency for voters to say that they will preference Labor to an even greater degree than would be expected based on the breakdown of third-party votes. My aggregate doesn't capture that aspect, mainly because experience in federal elections has been that these differences usually don't hold up on polling day, or are cancelled out by errors in primary vote estimates to the extent that they do. Especially, in the last three months, overall respondent preferences have been trending towards 80:20 in Labor's favour, which suggests that the preferences of all third-party voters would flow about as strongly as Greens preferences did in 2010. I absolutely don't believe that that would happen.

  3. Thanks. The preference flow was decidedly different between the 2012 and 2015 Queensland state elections. But my impression is that was an extreme case.

    1. Yes, it was also an optional preferencing case and it seems, oddly, that overall preferencing behaviour changes more in extreme cases under OPV because of the exhaust option. Voters who would never preference the Coalition may sometimes preference Labor and sometimes just vote 1 depending on their view of the ALP. In compulsory preferencing, these voters will always preference Labor.

      We'll never publicly know what respondent preferencing would have said about Queensland because no-one polled it (perhaps partly because of persistent ridicule from people like me!); it does seem Labor/unions had a good handle on it in their internal polling though. In NSW (another, but slightly less, extreme case), the most conservative estimates of respondent preferences were fairly close to accurate but there were also some that were over the top.

  4. At this stage, what if any, bearing do the opinion polls have on the final result? Are we at a point where we can deduce a likely winner based on time ahead/behind in the polls?

    1. We can infer that the Coalition is at significantly more risk of defeat than the average first-term government but really that's about all. In terms of time behind, the historic chances for governments that have trailed for this long are about even. The state of polling at any given time (especially while a government is going through a rough patch) doesn't prove much unless that polling is much better or worse than now.