The state breakdowns in the Nielsen (in a preliminary version the PUP row was mistakenly labelled "Independent") are especially interesting because they show the Greens on 27% in WA, compared to only 20% for the ALP, despite the two-party-preferred in that state being 52:48 to Coalition. I've modelled the implied 2PP from the Senate poll as 53:47, so 52:48 in WA is about what we'd expect. Of course, the sample size for WA is tiny, so we shouldn't take too much notice of the figure. All the same, might such things as a national Greens vote of 17% really be true?
Whenever we see startling results like these, there are a few general principles to keep in mind. The first is that the further off-trend a result is, the more likely it is to be nonsense. Under normal circumstances, a party does not sit on around 10.5% all year and then suddenly attract 17% out of nowhere. So the reflex response is to suspect that this result is just rogue. But an event has just occurred that could be driving the Greens vote up, specifically the WA Senate by-election, so that obvious conclusion shouldn't be automatically jumped to. It's just hard to believe it could be driving it up by that much, and it probably isn't.
As for the 27% in WA, The Greens polled 15 in the WA Senate by-election, but a few percent of the vote also went to a gaggle of Greens-friendly Senate micro-parties, so one might say the 15 is really 18. Throw in the general tendency for pollsters to slightly overstate the Greens vote, and assume a bit of Bullock factor backlash against Labor, and a bit of random variation, and you're just about there in terms of explaining how 15 might become 27.
Bouncing Vote Or Bouncy Poll?
However, too much commentary on results like the WA 27% involves accepting them at face value as if they were just like any other sample with a given margin of error, and thus assuming that it is highly likely the real result is somewhere between 19% and 35% (and could just as easily be at either end of that scale). In reality, when a result is out of the normal ballpark, not only is the probability of it being rogue higher, but also the probability of the real value being closer to the trend is higher than that of it being further away. Also, when a large number of state results are released simultaneously, as in Nielsen, it's almost inevitable that at least one of those is rogue, and if there's an obvious candidate it's most likely to be that one.
I have the advantage of having lived in Tasmania when local pollster EMRS used to release 200-vote electorate samples. Implausibles like the Greens polling 36% in a seat were common enough here for me to be very much used to very high Green readings in small samples. They're not worth taking seriously while there are only one or two of them.
That said, the Greens are also up in the state breakdowns everywhere else as well in this poll - by four points in NSW and Victoria, two points in Queensland and three points in South Australia.
On that basis, we can't rule out that what we see here is a "Bullock Bounce" in the Greens vote nationally, as a result of a by-election result in which the Greens were seen as campaigning with vigour and commitment and scored a huge vote. Meanwhile Labor preselected the wrong candidate to top its ticket, and did so in the wrong way, and finished up with a disastrous one out of six result, that is already producing internal recriminations.
Unfortunately, the only other test of the Greens' support level this week was Essential. And Essential's lack of dynamism from poll to poll (even accounting for its two-week cycle) is so well established that a lack of action from Essential doesn't prove all that much by itself. A one-week bounce of five points in the Green vote would make a 2.5 point difference to the rolling result, all else being equal, but an under-dispersed poll might not even pick up that. So we have to wait for more polls to come in to see if the 17% is partly real, or a completely rogue result. However, if it is a very short-lived bounce then we may never know the difference.
I actually think several months of the Green vote running very high over the Bullock issue at Labor's expense might be a good thing in terms of forcing Labor to get serious about reviewing its preselection tactics, but I tend to agree with Adrian Beaumont that even assuming this surge is real, it probably won't last very long.
Labor's WA result shocker, caused partly by an uninspired campaign and apparently partly by the preselection of Joe Bullock*, isn't the sort of thing that swings votes from left to right. No-one is going to be scared into switching from Labor to Liberal because Labor picked Joe Bullock over Louise Pratt and ended up losing Pratt's seat. But on the left, this seems to be a moderately big deal for the moment. There's a sense of betrayal that Labor cost itself a seat by putting a right-wing religious socially-conservative unionist on the top of its ticket ahead of a sitting Senator. It suggests the party hasn't learned any of the lessons about internal machinations and the excessive power of union hacks from its 2013 collapse and defeat. There may have been some on the Left who were OK with the Bullock-Pratt ordering so long as both got elected, but are rather less pleased with it now that it was not only ideologically repugnant but didn't even work. Is there enough in all of this to swing 650,000 votes on account of a by-election only a tenth of the electorate voted in, though? Probably not.
(* It is vaguely debatable. The primary vote-based evidence for a Bullock Factor is that the Greens polled eight points higher on the day than on postals, compared to the 2013 poll which displayed less than half that difference. The Greens apparently ran no organised postal vote campaign for the 2014 re-run but it is hard to see that accounting for the whole of such a big difference.)
Nielsen On Rogue Watch
Of the four Nielsen polls since the election, this is now the third that has attracted criticism as a potentially rogue poll in some way or other. Though this poll wasn't that unusual in terms of its base 2PP of 52:48 to Labor, Nielsen has thus far had the least consistent 2PP results as measured by difference from my aggregate at the time, of those pollsters that have issued more than one poll. The most consistent - but also the most skewed - has been Morgan (+1.38 for Labor +/- 0.95) followed by Essential (-0.45 +/-1.07), ReachTEL (+0.4 +/- 1.44), Newspoll (+0.22 +/- 1.57) and Nielsen (+0.13 +/- 2.00). Galaxy has only polled once.
Probably Nielsen's figures will stabilise with more polls (since they've only released four) but given the large sample sizes of Morgan and ReachTEL and Essential's fortnightly average-rolling, it may now be Newspoll and Nielsen that are the most prone to bouncing around. To this stage we haven't had any completely ridiculous and obviously rogue 2PP polls from anyone, but this is not the first from Nielsen to raise eyebrows. At least the theory that it was leaning Coalition (a rather speculative concept based on the two polls this year and ignoring the one last year) has been well and truly put to bed.
In that light, I'm not sure quite what to make of the news that the current poll's respondent-allocated preferences have Labor ahead by an implausible 54.5:45.5, except that the previous Nielsen also had a big gap between last-election and respondent preferences. This Nielsen, overall, seems a little bit generous to the left, whatever the true picture between Labor and the Greens.
National 2PP Polling Barely Moving
The 2PP result in the Nielsen, when added to my aggregate, produced a small move to Labor, that was then slightly countered by the 50:50 from Essential. The final result for this week barring more data is 50.8 to Labor, with the party having had a 52% result in Morgan and a 49% in Newspoll last week to go with this week's figures. But overall, Spiky Aggedor isn't showing much movement beyond the ups and downs of the odd noisy poll:
(The bottom axis is weeks since the aggregate started last October.)
As for the smoothed version, it's just about fallen asleep:
"Those who are aware of the severity of some of the budget decisions that have been made, are warning the situation will most likely worsen before it starts to improve. The Coalition is bracing to be “deeply unpopular by December”, said one source."
As for leadership figures, Newspoll last week saw Bill Shorten (-11) record his worst result yet and fall behind Tony Abbott (-7) for the first time since his opening poll in the job of ALP leader. It also had his satisfaction rating at just 31%, below his initial satisfaction on taking the job. Nielsen also had Abbott on -7 but with Shorten on +2. ReachTEL tends to produce ratings that look horrible for both leaders, but I've argued that this is because their "satisfactory" rating includes lukewarm positive sentiment. By my usual ReachTEL conversion method Abbott came out as -7.7 on their latest, Shorten at +3.1. Essential's most recent readings were -6 and -4.
The polls also had divergent readings on which of Abbott and Shorten would make the better (or preferred) Prime Minister: Nielsen had Shorten only one point behind (the equal closest from any mainstream poll so far) while Newspoll had an eight-point gap and Essential had ten. This is probably more evidence that Nielsen's sample was a little bit left of average.
That said, even an eight-point gap is underwhelming for the Newspoll, given that it showed a 51:49 to Coalition lead and that Newspolls usually favour incumbent Prime Ministers by more. A notable trend so far with Newspoll is that every time the Coalition has been ahead, Abbott has severely underperformed on the better PM question compared to the historical average for a given 2PP, but when the Coalition has been behind, the PPM figures have been about as expected.
Finally some attribute polling last week has canvassed some of the Coalition's ideological-fetish culture-war type issues: the surprise reintroduction of knights and dames scored a -15 net rating from both Nielsen and ReachTEL. Nielsen also asked "Should it be lawful to offend, insult or humiliate someone based on race?" (a question that goes to the Coalition's proposed revisions of the so-called "Bolt law", section 18C of the Anti-Discrimination Act). 88% said no and only 10% said yes. Most likely 99% were unaware that it is already lawful under certain conditions, but that those conditions do not apply to people who are as sloppy in their research on a sensitive issue as Mr Bolt. These results have been seized on as evidence the Coalition government is out of touch and has its priorities wrong, but it is unlikely they affect many votes.
This piece will be updated with more evidence concerning the strength (or not) of the Green vote post their WA success, as more data arrive.
17 April: Bludgertrack has given the Greens 5.1 points of credit for their polling this week, jumping them to a national primary of 15.4.
Update 23 April: This week's polls have come in with Morgan showing the Greens on 13 and Essential showing them on 11. While in both cases these are up, and are strong readings in the context of the Greens' post-election form, they are not evidence for any dramatic surge of the sort captured by Nielsen. Noting that the Morgan sample is a two-week rolling sample like Essential (and hence should have captured any surge post the WA Senate re-election), I'm thinking it is most likely now that last week's Nielsen reading was rogue and greatly overstated the post-WA surge to the party.
I thought I should have a rough go at applying my own aggregation weightings to the Green vote given the Bludgertrack figure above and just while there is interest in the level of the national Greens vote and the possibility of a major surge. At present by this method I get a raw figure of 12.8%, but since pollsters all overpredicted the Greens vote at the last election, it could be argued something in the mid to high 11s is more likely.
Meanwhile the 2PP picture has seen no real change this week with Essential returning 51:49 to Labor in line with the aggregate, and Morgan returning 52:48 to Labor, which is also in line once they are docked a point for house effect. Labor sits at 50.9 on my aggregate, which will go to 51.0 at the end of this week if there are no further polls released by then.
April 24: This week's Bludgertrack update (51.1 to Labor) has lots of state-by-state goodies that are well worth a look. Bludgertrack has the Greens vote nationally at 13.6% and I think the higher reading there compared to my estimates above is partly because mine retains old data slightly longer (including the Galaxy sample from three weeks ago which had the Greens on only 10%), and partly because of Bludgertrack's different treatment of Essential.