Tuesday, October 21, 2014

Poll Roundup: Three Polls Have Small Swing Back To Labor

2PP Aggregate: 51.6 to ALP (+0.6 in a week, reversing chances two weeks ago)
ALP would probably just win election "held now"

Somewhat against the recent run of play, this week's new polls have all seen modest improvements in federal Labor's position.  It did seem as results crept ever closer to 50:50 that soon we might see a crossover into slim Coalition leads, but it hasn't happened this week - another reminder that in polling, "momentum" is an elusive quantity.

This fortnight's polls

This week's new crop started with Morgan which turned in almost identical figures to last fortnight. A half-point primary shift from the Coalition to Labor was mirrored in the last-election 2PP result, with Labor's position improving from 51.5 to 52.  (The respondent-allocated 2PP, which I largely ignore, came down from 53 to 52 causing the company to describe the gap as the "narrowest in six months").  As Morgan's current methods have skewed to Labor by about 1.5 points during this term on average, this was equivalent to about a 50.5 from anyone else.

The first Newspoll in four weeks (thanks to a public holiday weekend in several states and a desire to release at the start of a new parliamentary sitting week) surprised with a headline result of 53:47 to Labor.  The primaries (Coalition 38 Labor 36 Green 14 Others 14) would normally imply a 2PP of about 52.3 to Labor so it's likely the actual 2PP was something not much above 52.5 and has been rounded up for Labor (yet again!).  As per the standards announced in the smash hit post Wonk Central: What Do We Do With The Poll Rounding Problem? I've treated this result as 52.6 to Labor for aggregation purposes.

That will probably go down as a slightly friendly sample for Labor, so it was no surprise to see PM Tony Abbott's netsat slip back from -11 to -15, or his "better Prime Minister" lead drop from four points to one (39:38).  What was therefore slightly odd was to see Bill Shorten's netsat down six points to -11 (35:46), his equal worst netsat (the third time he's had that score) and his highest dissatisfaction score so far.

After three weeks of 52:48s, Essential moved back to 53:47 to Labor.  This, again, is likely to have been a case of friendly rounding since the published primaries imply a result of about 52.6; I've aggregated it as 52.8.

Last week's Essential showed Abbott with an improved netsat of -8, his best since a -6 in April. Essential had Abbott with a six-point lead (38:32) as "better Prime Minister", also his best since April when he led by 10. Essential had an unchanged netsat of -1 for Shorten, though this hid a two point shift from "strongly disapprove" to "disapprove".  The current "strongly disapprove" figure for Shorten (12) is the lowest since the same reading in January.  Again, Essential's netsats are milder than Newspoll's.

Here's the current smoothed tracking graph:

Newspoll's High Non-Major Figures

Newspoll is normally a reliable poll, although far too much is made by commentators of often random bouncing from poll to poll in its output.  However, recently I've been curious about a string of very high minor party readings in Newspoll's primary figures.  The non-reading-out of PUP in Newspoll is perhaps no longer such a concern with the party only polling 3-4% in other polls, but prior to the May budget every Newspoll since the last election had a combined Green/Ind/Others figure in the range 22-24, and since it every time this reading has been 25-28 (and only one of those was 25).

The following table gives pre and post Budget averages for combined Green/Ind/Others figures for the four polls that have polled regularly since the Coalition's victory in September last year.  At that election the combined Green/Ind/Others vote was 21.1 points:

The standard deviations (SD) are for individual readings.  All four pollsters show a statistically significant average increase in the combined Green/Ind/Others vote since the Budget.  This isn't surprising as there was a big surge in the PUP vote following the Budget, which has only recently returned to sender (the Coalition mainly).  However while ReachTEL and Essential show a change of 1.2 and 1.5 points respectively, and Morgan's 2-point change has shrunk to more like a point in readings over the last few months, Newspoll results have a four-point increase in third-party voting since the budget, to a level nearly six points above the last election.  Also in Newspoll's case it's far from clear the third-party vote is going down.

The five Galaxy readings so far (20 and 24 before the Budget, 24, 24 and 22 after it) are also consistent with the increase in third-party voting being only modest (and with it being not that far, if at all, above the election result to begin with).

This has also been seen at state level, at which the combined Green/Ind/Others votes at polls taken largely or entirely after the federal Budget exceed those taken before in every mainland state, and by at least three points per state.  Again, the flood of third-party figures around the high 20s (even one 30 in South Australia) isn't generally being repeated by other pollsters. What is going on here?

Enter Ipsos

It is pleasing with so few active polls, and clouds of some sort over most of them, that a new player will soon enter the market with the arrival of Fairfax Ipsos.  Ipsos has run a series of issue polls based on online panel polling, and did a few voting intention polls by this method, but will now be running polls similar to the old Fairfax Nielsen - phone polling in most months (including calling mobiles) with a sample size of 1400.  It will also be running numerous state polls.  Ipsos has a strong international reputation - for instance its US phone polls have an A- rating on fivethirtyeight, ranking them well inside the top 10% of US pollsters.  Because we have already seen from the online polls that Ipsos have some idea what they are doing in the Australian environment, I will be including their phone polls in my aggregate right away, though they will be slightly downweighted for the first four while I get a feel for their behaviour and house effects (if any).  Based on what we know from overseas there is a good chance these polls will be at least as good as the departed Nielsen, if not better.

Issue Polling

Some new findings on Iraq come from another new pollster for Australia, Factuality.  This is another online panel poll and, similarly to other pollsters, it finds that voters support air strikes against ISIS (54-25) but are equivocal about ground troops (39-38).  Factuality also finds that voters support these actions although believing (56-25) that they will make Australians less safe from domestic terror.  The full report is well worth a look for those with an interest in these issues, as there is much more detailed drilling down than in most such polls.  A slight concern about the poll is that a massive 20% of the sample were supposedly "unsure" of their voting intention.  The only voter intentions mentioned are Coalition, Labor, Green and Unsure, so whether any "other"-type choice was offered isn't clear.  If others were included under Unsure they should not have been; if they were not then the Unsure figure is too high and more prodding is needed.

Newspoll finds that Australians want Tony Abbott to "confront" Vladimir Putin over the apparent shooting down of flight MH-17 over Ukraine (63:27, including a 51:40 majority of Labor voters).

Essential finds as usual that voters think everything in the economy except company profits is getting worse.  It also finds that Australians believe the government is doing enough to fight Ebola (58:21) based on a preamble that advises that the government has committed $18 million, but has said risks are too high to send personnel.  Last week, Essential had various highly predictable findings about voter attitudes on wealth vs poverty, including the unsurprising results that voters would like the Government to fund its Iraq involvement through higher corporate taxes (68:22) and dropping its paid paternal leave scheme (56:31) rather than by cutting tax concessions or services or raising income tax.  Sometimes the results of Essential issue questions make me wonder that their 2PP isn't permanently jammed at 65:35 to Labor!

We're about due for another ReachTEL soon (if one appears later this week comments will be edited into this post) and the first Fairfax Ipsos isn't far away.  Their opening poll will be a Victorian state poll expected early next week, followed soon after by a federal poll.

Update (25 Oct): Make it 4-0 for the small swing back with ReachTEL reporting a rounded 2PP of 52-48.  Primaries and other details are not yet available but I've provisionally aggregated it and this plus the weekly reset puts Labor up to 51.8.  For now that's Labor's best position in six weeks.  

Update (27 Oct): The full ReachTEL primaries are out and I make the 2PP off these 52.3 to Labor, which makes one-tenth of a point difference to my aggregate, pushing it to 51.9. 

The improvement in Tony Abbott's ratings noted in the September poll is mostly gone, with his Good/Very Good rating now down 0.5 of a point since August, with Bad/Very Bad down 1.2.  Overall Abbott's raw net rating on the ReachTEL scale (which is harsher than that of other pollsters) comes to -19.5, with Bill Shorten much the same at -19.3.  As an experiment I am going to try changing my ReachTEL conversion method for leader ratings to counting a third (instead of a half) of the Satisfactory rating as positive; if this is done Abbott scores -14  and Shorten -7.  Issue questions still to come.


  1. Hi Kevin
    It only takes on poll to win the election , and what happens if morgan poll 2 weeks before the election has labor 54-46 and on the eve of the election 54-56 to labor. On election day if the election is 54-46% to labor.

    Does that mean you will take 1.5% off the result , due to morgan poll being biased to labor ?

    1. Morgan have a history of switching their polling methods in the last few weeks before an election, for instance by dropping the use of face-to-face polling. Last election their final poll had the 2PP dead right, but it was conducted by different methods to the ones they are doing now. I assess the house effect not just by pollster, but for each different kind of poll that a pollster like Morgan might run. So my comments about the 1.5 point skew relate only to the current Morgan polling method (SMS/F2F).

      Supposing that Morgan were still using the same methods at the next election, and their last two polls were 54-46 to ALP and the result was the same, but up to that point Morgan had appeared to be skewing to Labor by 1.5 points. In that case I would look at what the other polls were doing.

      If the other polls were also getting around 54-46 then I would assume that either Morgan had made a very late change to their methods, or else just by chance they'd had two samples that got unusually neutral results in a row. I'd assume the latter pending further testing. If further Morgan polls then showed no skew I'd stop adjusting them.

      If, however, the other polls were all getting around 52.5-47.5, meaning that Morgan had been right all along and the other polls (and my aggregate) were wrong, then I'd give them credit for getting it right, examine how this had happened and adjust my estimates of house effects accordingly for the future. That's assuming there was no obvious last-minute issue that could have caused it.

      But I'd probably eat my hat first. Past elections where Morgan has gone out on a Labor limb all the way to the polls (eg 2001) ended badly for them.