Poll Roundup: Hockey Gaffe Breaks The Monotony

2PP Aggregate: 52.4 to Labor (+0.6 in a week)
Labor would probably win election "held now" with small majority

Finally we have some movement at the station.  After four consecutive weeks of 51.8% to Labor my polling aggregate has finally moved on, with a modest step in Labor's direction.  The move is not large enough to be completely certain it is real, and there is enough going on to be unsure what is causing it if so, but it's most likely Treasurer Hockey's remarks on housing affordability (and some faltering and ill-judged attempts by his colleagues to put that genie back in its bottle) have contributed to a small-scale revival of perception that the government is out of touch.

This week's polls

Four polls have come out so far this week.  Fairfax-Ipsos opened procedings with a 53:47 to ALP (by 2013-election preferences, 54:46 respondent-allocated), which was interesting because Ipsos has usually thus far leaned to the Coalition by about a point.  There was some scepticism about it based on a premature perception that Ipsos is very bouncy (see below).  However the Ipsos result was more or less totally backed in by a 2.5 point move to Labor in Morgan, to 54.5 - bearing in mind that Morgan skews to Labor.  2-0 for the proposition that Hockey had really put his foot in it, but this was contradicted by a 51:49 Newspoll, while Essential did as Essential usually does (ie very little) and stayed at 52:48.



The Ipsos appeared in an environment that was very data-poor (with only three other polls still in the aggregate) and produced an immediate 1.1 point jump, but about half of that was lost on the later polling.  After processing the primaries I treated the Ipsos as 54 to Labor, the Morgan as 53.4, the Newspoll as 51.2 and the Essential as 51.8, and my aggregate now sits at 52.4 to Labor (though it was very close to being 52.5).  That puts the Coalition back to where they were before their moderately well-received second Budget.  Here's the smoothed aggregate graph (note that this week's result is smoothed back into last week, so that's why Labor has gained over two weeks in this graph):

The proportion of voters in my sidebar Not-A-Poll who believe the Coalition will not recover the 2PP polling lead at any stage in this term has come down very slightly but is still at 75%.  It is, of course, possible they could be re-elected without recovering the lead and without even winning the 2PP at the election.

One thing we should expect, if the current move to Labor is caused by Hockey's comments, is that the move against the Coalition should be strongest in younger age demographics that do not own their own home.  It turns out that the Ipsos age breakdowns support this to a spectacular degree while the Morgan ones have the biggest move in the middle age bracket.

I should also note claims that asylum seekers were paid to return to Indonesia as another factor perhaps influencing polling, though it's hard to see this playing out well for Labor for any length of time given claims their government did something similar, and given the Coalition's joy in thumping the stopped-the-boats mantra at the slightest chance.

Ipsos and Volatility: Much Too Soon To Know

The shift from a 50-50 2PP to 53-47 in Ipsos last month produced some comments (not for the first time) about Ipsos being bouncy, volatile, all over the road (and so on).  So far Ipsos has produced readings for Labor of 51, 52, 54, 51, 54, 50, 53 for an average poll-to-poll move of 2.7 points, which seems like a lot.  The first problem with these sorts of comments is that they do not compare like with like: we should expect a monthly poll to be more volatile from reading to reading than a weekly poll, simply because it is more likely there has been a genuine change in voting intention over a longer period.  Comparing Ipsos' poll to poll moves with Newspoll's or worse still Essential's is therefore silly.

So in comparing the volatility of Ipsos with that of other pollsters, it's necessary to discount changes in voting intention between polls, and this sort of thing is what aggregates are made for.  After applying such a correction, it still turns out that Ipsos has been more volatile than the other pollsters so far.  I have the average poll-to-poll change in Ipsos' position relative to my aggregate at just under 2 points, compared to 1.4 points for Newspoll, 1.2 for Morgan with all of Essential, ReachTEL, Galaxy around 0.7.  (Essential uses fortnightly rolling, ReachTEL has large sample sizes and Galaxy ... we're not quite sure how Galaxy does it, but it may involve witches.)

Two points is a large average bounce for a corrected result given that each poll itself influences the aggregate, so if Ipsos continued behaving like this forever we might indeed call it "overdispersed" (ie bouncy).  But those making that call right now and saying they've discovered an undesirable property of the new poll are massively jumping the gun. While the mean poll-to-poll move after accounting for underlying changes is 1.95 points by my reckoning, that estimate of the mean has a margin of error of a massive 1.15 - hardly surprising given that the individual data points range from an effective 0.3 point shift to an effective 3.9 point shift.

It's possible based on the data we have that Ipsos' methods might actually be very stable with this just an unusual run, or that they might be even less stable than they've seemed so far.  I reckon it is most likely they've just started with a randomly bumpy run - or perhaps with some extra bumpiness while getting sorted - and that their average poll to poll change will go down somewhat over time.

It's another case where a little understanding of statistics principles as they relate to polling is a dangerous thing.  If someone can grasp the concept that a poll with a sample size of 1400 is likely to bounce by a certain amount because of random variation, then they should already know far, far better than to draw the conclusion that that poll is bouncy on the basis of a mere six measurements of a poll's volatility. But they don't.  It is far too early to be drawing such conclusions.

Leaderships

Leadership polling this fortnight has produced wildly contradictory results.  Last week Essential had Tony Abbott on a net rating of -11 (39% approve 50% disapprove), his best rating since polling the same last November.  Bill Shorten sank to -13 (32-45), his worst ever from this pollster, and Abbott led Shorten as Better Prime Minister 38-33 (his best lead since October).  This was, however, before Hockey's comments and Abbott's minor involvement in the damage repair attempt.

Ipsos had Tony Abbott down six points to -14 (40-54) but Shorten down only two to a seemingly benign -6 (41-47), though this was still his worst from this pollster.  Ipsos also had Shorten ahead as Preferred Prime Minister 42-41.  This was a rather amusing result after AFR editor Michael Stutchbury had speculated on Insiders that if Abbott again led in the Ipsos it might be the beginning of trouble for Shorten's leadership.  The pilloring Stutchbury received in comments on Poll Bludger was out of proportion to what he actually said, but it could also be argued that no-one should be commenting on preferred leader scores at all without noting that they are skewed to incumbents.

(Incidentally, while PM's historically break even on preferred leader scores when their party has only 47% two-party-preferred, Abbott on average has needed 47.6%.  That's not so unusual - in fact Keating needed about 48%, as did Howard in his first two terms.  Julia Gillard on average could lead as preferred prime minister if her party's 2PP was above a mere 45 - because Abbott was her opponent).

The last ever fortnightly Newspoll as we know it was the big talking point, delivering Abbott a seven-point drop to -22 (34-56) but giving Bill Shorten a horrible net rating of -26 (28-54), his worst by eight points from a series which had already been harsher on him than others.  By way of comparison, Shorten's 28% Newspoll satisfaction rating was only one point better than Tony Abbott's worst in the same position.  (The all-time low was 20%, for Alexander Downer.)  Unsurprisingly, Newspoll still had Abbott in the "lead" as better PM, 41-38.

While all three polls thus gave Shorten his worst rating so far, the sheer magnitude of the Newspoll raspberry does seem suspicious.  It's true that Shorten has come under some heat over his union past, but this has been largely overshadowed by the Coalition's issues, and wouldn't have seemed likely to damage him so much so quickly.  With Newspoll in the process of winding up its current operations prior to the changeover to Galaxy, it would not be surprising if there were staffing issues that had some effect on the reliability of the final poll, and for this reason I'm treating this result with some degree of caution.  (I do not have any specific evidence of such issues, though discontent among staff at the manner of the Newspoll windup was mooted.)

This caution also applies to the relatively strong result for the Coalition on voting intention (largely off non-Coalition voters choosing Greens and Others rather than Labor, a common issue in Newspoll recently), and I did notice Coalition fans more muted in their celebrations of the poll outcome than for a normal 49:51 result.

Greens Through The Roof

One thing everyone except Essential agrees on is that the Greens are on some kind of major surge following an apparently well-received leadership transition to Senator Richard Di Natale, and apparently also on the back of generic disgust with most majors.  While Essential has the party on 10%, those pollsters that use a live interview component have it on 14 (Newspoll and Ipsos) and 13.5 (Morgan).  See Mark The Ballot's aggregate with Essential excluded, in which the Greens have Dirichlet-walked (whatever that is) up to 13.6%.  Most likely all these pollsters are overestimating the Green vote and it would actually be more like 12 in an election "held right now" but the party is certainly doing well at the moment.

Same-Sex Marriage: Nonsense Detected

We have some more same-sex marriage polling this week with Ipsos showing 68-25 support and Newspoll a relatively low (for a phone poll) 58-34.  Caveats apply for possibly slightly left and right leaning samples respectively.

Despite this there was some really strange analysis of supposed Senate seat risks to the Coalition if it allowed a conscience vote, premised on the idea that religious-right micros would not direct preferences, the margin of victory for the last Coalition candidate in NSW, WA and SA having been smaller than the very spuriously so-called "family party" vote, where "family parties" = Family First, Christian Democrats, Aus Christians, Rise Up Australia and DLP.

Let's look at this case by case.

In New South Wales the Liberals won the final seat by about 3.2%.  It's true that the so-called "family parties" polled slightly more than this but the analysis omits to note that the DLP preferences went to the Liberal Democrats and never reached the Liberals.  Moreover the Liberals' opponent for the final seat was the Greens, who these parties would hardly preference ahead of them.

In South Australia the Liberals beat the Xenophon Group for the final seat by about 4%.  However the "family parties" preferences all pooled with one of their number who was elected (Bob Day) and then flowed on to the Liberal-Xenophon contest at vastly reduced value, adding only 0.36% to the Liberal cause.  Moreover, the DLP preferenced the NXG over the Liberals in this contest.  So it is just not true that the "family parties" decided the seat for the Liberals - at all.

In the Western Australian rerun the Liberals narrowly defeated Labor for the final seat.  Here if the "family parties" preferenced Labor, Labor would have won, but they would hardly switch their preferences to Labor over same-sex marriage when Labor also allowed a conscience vote and far more ALP representatives were exercising it.  (Even had they switched their preference to Joe Bullock specifically, he would have been elected before they were excluded).  Moreover Family First preferenced PUP ahead of the Liberals meaning that their votes reached the latter only at vastly reduced value.

Quite aside from the basic premise of the analysis being multiply false (since the Coalition is generally beating left-leaning parties for the final seats and since the micro-right-parties already tend to put each other above the Liberals) it's rather silly to go crystal-balling the next Senate election when the system it will be held under is unknown.

Other Issues Polls

There are a lot of issues questions out there, and I just pick out the ones I find of most interest.

Ipsos respondents said first homes were unaffordable in their region (40-57) and Essential respondents said housing was unaffordable in their region, fullstop (33-60).

Essential last week found that women are more likely than men to report having experienced or witnessed not just sexism but also pretty much any form of "intolerance" (not the ideal word to use for "prejudice", but I digress).  Excluding sexism, are women more likely to either experience or witness prejudice, or is there a gender gap in what is remembered, noticed or construed as prejudicial?  I think this is a very interesting finding.

Essential last week also had some findings of very strong support for the government's plan to remove citizenship from dual nationals found guilty of terrorism, and from sole nationals likewise if eligible to join another nation. It was less surprising that even Coalition voters think this is a matter for the courts.