Friday, June 24, 2016

Rolling Poll Roundup: The Final Week

2PP Aggregate: 50.6 to Coalition (Ended last week at 50.5)
Seat estimate if this is the final 2PP: 78 Coalition 66 Labor 6 Other
Voting intention may be volatile in final week because of Brexit. (Or not.)

Smoothed 2PP Aggregate. Graph last updated 26 June post Newspoll
Here's another rolling roundup of incoming polls, which will probably run at least until the middle of election week.  For last week's see here, and for my thoughts on the state of the betting markets through last week (and reasons for caution about the current projections of a narrow Coalition win) see here. New polls and thoughts will be added, and the aggregate graph and header updated, through the week as polls arrive.

I've been eagerly awaiting the fresh data in this evening's national ReachTEL, given that as usual in this rather sparsely-polled election, there were no national data that were less than five days old.  Especially it was important to see whether national polls had picked up the shift to Labor implied in that batch of commissioned ReachTELs in NSW earlier this week.  If that shift was real, then Labor's Medicare scare campaign may have bitten hard, and it would not have been surprising to see the Coalition drop a few points off its primary vote tonight.  Tasmanian polling which I will report on tomorrow is also none too flashy for the government.

As it happens though, tonight's ReachTEL (taken last night) shows only modest and not even really "statistically significant" changes*.  There is an important methods change in this poll in that ReachTEL only now polls the Nick Xenophon Team vote in South Australia, causing it to drop from 4.2% to 1.4% through no fault of its own (the NXT SA vote was likely in the ballpark of 19% plus or minus lots, on the low side compared to others).

The NXT handling change (an excellent idea, though likely to be widely misreported) releases 2.8 points of primary vote to other parties.  Others are up 2.4, Greens are up 1.4, Labor are up 0.1 and the Coalition is down 1.2.  So possibly Coalition votes are going to Others while ex-NXT votes go to Others and Greens, but more complex churn could also be at play.  There's no movement on the respondent preferences 2PP (still 51-49) but my estimate of the last-election 2PP from these primaries is down from 52.3 to 51.1%.  And by the way, the lack of difference between the last-election and respondent 2PPs in this ReachTEL is another point against the view that this election will see a large preference shift.

Once again because of the sparseness of recent data the ReachTEL is heavily weighted, taking my aggregate from 50.3 to 50.5 as of Friday night, which will go to 50.8 on weekly reset if no other polls arrive by midnight. Again the lower frequency of polls makes the aggregate more volatile than it would normally be at this stage of the campaign and the pattern so far has been that boosts for the Coalition are pulled down by other polls.  And the caveats mentioned in the article below - about whether our polling is even reliable at all - still apply.

The ReachTEL has a curious leadership result, with Malcolm Turnbull's net rating jumping 9.1 points to +0.2, a three-month high, for no evident reason whatsoever (again, this is pre-Brexit).  Bill Shorten is also up to -7.1, which is again his second-best rating so far.  Turnbull gains slightly as "better PM", out to 58.4 to 41.6.  Bearing in mind that ReachTEL does not have a neutral option (which seems to be the major cause of skew to incumbents on this question), this seems a serious lead, and might say something unflattering about perceptions of Bill Shorten's readiness to lead.

The only slight change in the issues mix is a slight uptick for health issues at the expense of "creating jobs".  It is interesting that the Coalition have stressed "Jobs And Growth" as their mantra and yet generally polling hasn't shown that employment is the issue it was in 2013.

(* There is a fallacy that changes that are not statistically significant mean that nothing has changed, and should be ignored.  Small changes do generate far too much nonsense commentary when they are often caused by random movements.  However, any non-tiny change in data alters the underlying probability distribution of voting intention in an aggregate.)

"Events, dear boy, events"

As fresh as this ReachTEL sample is, there's a case it may as well be from a million years ago.  The elephant I mentioned last time has gone from being in the room to in the house.  The Brexit referendum has been carried (a disaster for most of the already-stricken UK polling industry, but that's another story.) It is rare for such a potentially significant world event to happen during an Australian campaign, and especially while the polls are open, and especially following a campaign in which so little has happened to date.  The biggest such event to happen during a campaign was the declaration of war in 1914 (which probably swung the election result).  Other major external events during campaigns have included the 1954 Petrov Affair and the 1983 bushfires.  Some other campaigns have occurred in the immediate aftermath of major events that have seen major swings in voting intention during the campaign - the 1975 Dismissal, the 2001 S11 attacks and the 2013 reinstatement of Kevin Rudd, for example.

The usual battle-lines are already being drawn.  The Coalition says this is no time for reckless spending and running up the national deficit.  Labor responds that this is no time for cutting revenue through tax cuts either. Labor also says that it steered the nation through the GFC.  The Coalition responds that Labor did this by trashing the hay their man stashed in the barn, and then says that Labor spent too much.  The last part may not actually be true, but Labor struggles to contest it.  Even if Labor debates the issue effectively, the perception is that time spent on the economy rather than on medicare or education is only a good thing for the Coalition.

Maybe it just won't matter.  Perhaps no-one will care.  Maybe voters are so sick of elites telling them what they care about that they will just ignore it all, or vote Labor in droves for no reason just to prove all those insiders wrong. A colleague points out that if the Coalition think that they can win now by just saying "Brexit" all week, they could be in for a rude surprise:

Indie Raids Done Dirt Cheap

Cowper.  Is this really happening? Do I really have to write about this seat?  Rob Oakeshott, a waffly crossbencher who delivered government to Julia Gillard and then ran away from the impending judgement of voters in 2013, pops up in the adjacent seat at the very last minute and runs a shoestring campaign with no money, no signs, and worse still the support of GetUp! ... and within weeks he is polling 50-50.  This to go with the similar if larger-scale resurgence by Tony Windsor against Barnaby Joyce in New England.  If these seat polls are accurate (and it's not just one company either) then the National Party has serious problems.  It simply shouldn't be this easy. It doesn't even seem to be about the candidates - Luke Hartsuyker and Rob Oakeshott rate about the same (both fairly well).  Perhaps these results will not stand up when votes are counted, but in the past indies have tended if anything to build up steam as polling day approaches.  If you know of an indie candidate who's running hard against a National and has more profile in their electorate than David Cameron's cat, drop me a line about them; who knows, they might win.

Galaxy Mass Seatpolling

I've seen reports of a large pack of Galaxy seat polls with mostly grim readings for Labor.  Details when I have them.  Hindmarsh is 50-50 with NXT on 16%.  Boothby is 53-47 to Coalition with NXT on 19% and potentially still competitive (though apparently needing to overhaul Labor to win). Kennedy 58-42 to Katter (oh well), and the following are Coalition vs Labor: Longman, Dunkley and Corangamite all 53, Reid, Banks and Brisbane 52, Lindsay, Gilmore and Dobell 51, Macarthur 50.  Sounds grim for Labor but on tight margins from small samples they would win some of these by chance.  Labor are in front in Petrie 52-48, Capricornia 51-49 and their own seats of McEwen 52 (CFA making it uncomfortably close-ish despite the wretchedness of their opponent), Bruce 52 and Griffith 53.  There's nothing surprising in these seat polls - the average swing to Labor is 1.8%, which is about what we'd expect given the state of polling in each state and given that several of these seats have sophomore effects in the Coalition's favour.

The Galaxy results are strongly contradicting some of the commissioned ReachTELs from this week, though neutral polls have yet to bother with Page and Eden-Monaro.

Updates will be added as new polls arrive.

Fishy Fire Stuff From Vic Liberals

It's refreshing to see some dodgy polling from the right for a change rather from the usual quarters.  Liberal internal polling of their own seats of Corangamite, Dunkley and La Trobe and Labor's seats of Ballarat, Bendigo and McEwen is reported as saying that Labor will be hard pressed to gain any seats there (which I suspect is true) and may be fighting to avoid losses.  (It would be useful to see public polling of some of those - unlike McEwen - where the Liberal candidate isn't actively hopeless.)  The CFA saga is a dead weight on Labor's chances in Victoria but the reporting of the poll goes beyond what is believable in terms of impacts.

It's believable that 65% of respondents in such seats would say they opposed the push to "unionise the CFA" (I'd still like to see the question wording) but it's very hard to credit that 27% of Labor voters would have said they would change their vote over the issue.  It turns out that they didn't!  27% was the net result of a poll in which voters were asked if they were more or less likely to vote Labor - not if they would actually change their vote.

Polls of this kind are almost always rubbish.  Many voters will say "less likely" just to give a party they dislike anyway a kicking, though they would never have voted for it anyway.  Some voters say they are more likely when they are only very slightly more likely.  Effect sizes of 30% can be produced in this kind of polling even using issues that are obviously trivial or irrelevant.  It's a stupid polling genre and the best thing any journalist offered such a poll can do with it is burn it.

Galaxy (26 June)

There is a Galaxy taken 22-23 June with a 2PP of 50-50 off primaries of 42-35 and a rather optimistic-sounding 11 for the Greens.  I've aggregated it at 50.1 to the Coalition, which takes things down to 50.6.  This doesn't tell us anything yet about the impact of Brexit (if any) on voting intention.  Galaxy found that voters would have preferred Labor 53-47 if Tony Abbott was still PM.

Melbourne Ports Lonergan

After their previous poll of Melbourne Ports had Michael Danby's vote crashing but didn't actually show the Greens winning the seat (although they may have thought it did), the Greens have another one out.  This one's a large-sample (1300+) Lonergan said to show primaries of Lib 42 Green 27 ALP 25, based on which they might win the seat on Danby's preferences (debatable - see Mark Mulcair in comments.) The poll would be a 7% swing from Labor to Green.  I don't think our assessment of the chance of that happening, whatever it is, should be affected by the Greens releasing a Lonergan that shows it.


A new Newspoll that straddles the Brexit referendum is out with the Coalition up two points on the primary vote to 43% with the 2PP at 51-49.  My aggregate has gone to 50.8 with that (the Newspoll was entered at 50.9).  There is no change to speak of on the leader ratings with both leaders up a point on netsat and down a point on better PM.


A MediaReach robopoll of just over 500 voters has been reported as showing Natasha Griggs in a hopeless position down 39-61, similar to polling for the incumbent CLP territory government through its term.  Primaries are not available and it is not known to me yet who this was commissioned by.

28 June Daily Reset

Here we are in the last week and we have no polls that are entirely less than four days old!  So with the special weightings taken off both ReachTEL and Newspoll the aggregate comes back a tenth of a point to 50.7 to Coalition.

South Australia is posing problems for my 2PP aggregate.  State polling implies a 2PP swing to the Coalition but the reliability of it is unclear given the NXT issues.  If I enter a full-strength or even watered-down swing to Coalition in SA then, for instance, Hindmarsh becomes an almost certain hold despite a 50:50 seatpoll.  The model also now thinks Labor will lose about two seats to the Coalition - most likely one in Victoria and one in SA, and thinks Wakefield could fall to the Coalition although no evidence for this has been noted and the bookies see NXT as the bigger danger.


A ReachTEL by an anti-WestConnex (tollway) group has Julie Owens leading 52-48.

What Vote Will Others Get?

There's a piece by Antony Green here on the size of the Others vote, in which he suggests that the Others vote is being inflated by voters naming NXT in seats they're not running in, and also by voters picking "other" or "independent" as an ambit claim, including in seats where they won't have anyone to vote for.  He reckons Others could crash to 7-8%.

I suspect the recent Others votes (eg ReachTEL 13.3, Ipsos 14, Newspoll 12) are indeed on average exaggerated, but not by as much as Antony suggests.  NXT will indeed poll only 2-3% but in 2013, Others excluding Palmer United managed 6.9%.  At this election we have NXT, plus the return of Windsor and Oakeshott, so those should add easily 3% to the Others mix.  Moreover, NXT in South Australia takes its votes mainly from major party voters, and will do some damage to the SA Others vote but probably not very much (and SA is less than 10% of the election anyway).  There is also loose PUP vote to be snaffled by different Others at this election.  ReachTEL and apparently Newspoll are already adjusting for NXT only running in certain seats, and BludgerTrack has the remaining Others vote (ignoring NXT) at 11.3%.

I can easily see non-Green Others (including NXT) getting 10% on that basis, and perhaps reaching the Newspoll figure of 12%.  It would take some deliberate voter move away from micro-party voting  at this election to hold the total below the Newspoll figure, but it's possible that will actually happen.

Higgins again - a poll-shaped object

Porcupine Fish Award for Ultra-Fishy Polling (image credit)

There's a poll of Higgins that is easily the worst poll I have seen released in this campaign so far.  The poll is by Community Engagement, of whom I have never heard and regarding whom I can find no information.   Six pages of polling results provide nothing on the pollster's identity, web presence (if any) or contact details for the pollster, and most critically do not even state what the polling method was.

The results (Liberal 44 Green 27 Labor 21 after excluding the undecided) seem doubtful enough given that this would represent a 10-point primary vote swing from the Liberals to the Greens with Labor losing three points as well.  But what's worse is the issue questions, which start with the basic idealistic pony-poll question about whether voters think Australia should do more to end poverty, and then go on to use selective presentation of claimed facts in a blatantly biasing preamble about foreign aid to prime respondents into agreeing with increasing aid.

Then there is the usual question about whether the issue would influence your vote - voters when confronted with practically any issue in isolation will often say it will influence their vote when it actually won't - and various other blatantly primed questions.  From this comes the sudden conclusion that increasing foreign aid could be a "deciding election factor", a conclusion that could have been wrongly drawn from any of a thousand issues polled in isolation using the same dodgy techniques to produce a deceptive result.

I have been quite restrained on the awards for bad polling lately but this Higgins poll is so fishy you could smell it from the electorate of Durack.  Why left-wing groups that would presumably claim to stand for some kind of political morality persist in issuing such dodgy polls is beyond me.


Essential has not moved off last week's 51-49 to Labor and I have aggregated it at 48.6 for the Coalition.  Essential has the Coalition primary vote at only 39% (down one), and appears in the last few weeks to have resumed its heavy pro-Labor skew relative to other pollsters. We already know Essential doesn't follow trends so this tells us not much about whether voting intention has changed. Malcolm Turnbull has a netsat of zero (40-40) and Bill Shorten comes in at -2 (37-39); Turnbull leads the beauty contest 40-29 (unchanged).  I do not know if this is the final Essential or if there will be one from this week only later in the week.  If you believe the current one, Labor is probably winning (or probably was on average over the last two weeks).  Other pollsters with much better track records have a different view.

Essential also samples respondent-allocated preferences, finding a 73-12 flow of preferences from Green to Labor (equals 86% if assuming the undecideds maintain this pattern, or 80.5% if assuming they split evenly) and a 33-30 flow of preferences from Others to Coalition (52% or 51.5%).  If all undecided preferences are assumed to follow the flow of decided preferences, Labor would do just under half a 2PP point better on overall preference flows than in 2016.  If they split 50-50, the overall preference flow would be the same.  The truth is probably somewhere in between.

An interesting finding is contained in the "Best campaign" section - 26% of Greens supporters think Bill Shorten and Labor have run a better campaign than their own party.

NXT Lonergan

Tonight on ABC 7:30, an NXT poll showing SA primaries of Liberal 36 Labor 26 ALP 24 Green 6 Family First 5 was reported.  ABC did not report that the pollster was Lonergan, which has yet to establish a reliable reputation following poor polling at the 2013 federal poll (though its NSW state polling wasn't so bad).

One might expect from my previous comments about NXT that a 24% primary scenario means lots of seat wins.  But if it is accompanied by even a small shift from Labor to Liberal among voters not switching to NXT (or if NXT voters come even a little more heavily from Labor compared to Liberal than my previous model expects) then this does not necessarily follow.  An increase in the Liberal primary vote at the expense of Labor in my NXT model (re which see here) makes it harder for NXT to get over the Liberals in Kingston and Port Adelaide, and steepens the preference mountain in Boothby, Sturt, Mayo, Barker, Grey.  The NXT model I previously constructed gives the win to NXT in just two seats - Boothby and Port Adelaide (!) but there are obviously special factors in Mayo (Briggs bar incident) and Grey (Arrium plus known strong polling).  So, this poll probably doesn't represent the "mayhem scenario" even if it is accurate.  But even very small movements on these numbers could be the difference between winning nothing and winning most of South Australia.


  1. Hi Kevin,
    Thanks for all of the insightful commentary.
    I have some doubts about the sampling methodology of the seat polling. My partner and I live in Gilmore and I suspect because we work from home (so here the great majority of the time) that means we get way over-represented in polls. We were polled 3 times in 27 hours from the middle of last week: the Galaxy one published yesterday, one from Q&A marketing (don't know who for) and Reachtel on behalf of the ALP. We also got called a while back by Reachtel, I think for the Teachers' Federation poll published recently. I wouldn't be surprised if we have been included in every Gilmore poll conducted in recent months.
    Given the number of voters in Gilmore and the typical size of the polls, I doubt the methodology behind choosing those to poll delivers anything like a random representative polling group. The true margin of error is much greater than that based solely on the number of respondents. That may be a good part of why seat polling has such a poor record.

    1. Gilmore has been very heavily polled this election but this comment highlights some of the increasing problems with seat polling and polling more generally. Response rates for the landline portions of polling are so low that a person who routinely answers their landline has a high chance of being polled in a seat poll. There is a bias in favour of people who are interested in politics that is really hard to get rid of with all of the scaling that some companies do to adjust for income, age, gender and so on. And yes, the true margins of error are indeed much greater because of these sorts of issues.

  2. Kevin thanks for all of your insightful commentary and analytical work. One query on this post, though: you say 'worse still, the support of GetUp' for Oakeshott's campaign. What do you have against GetUp?

    1. That flippant comment probably had a few aspects to it. Firstly - and this is something that applies to Windsor too - if you're an indie running for a conservative rural seat then whether it's really in your best interests to have your campaign too publicly associated with an obviously green-left movement with an inner-city youth vibe is questionable.

      Secondly, GetUp!s style of campaigning fairly often bugs me. When they have campaigned in Tasmanian elections, I've sometimes found their summaries of where candidates stand on issues are inaccurate and seem designed to favour who they wanted to win anyway. For the present election, I rather like their strategy of specifically targeting the pro-Abbott right, but I also find their methods rather trivialising. I'd like to think the election for Bass for instance is about bigger things than a politician blocking the other side's concerns on Facebook. (Which is not to say that that is all that Nikolic has blocked.)

      On a deeper level, though I doubt it was anything to do with that particular comment, I disapprove of the commodification of dissent in the form of mass movements that encourage culture-war politics. I see such movements as strengthening the generic/conformist tendencies of both the right and left and training more people into being unable to think beyond their side's position.

  3. Assuming the Melbourne Ports poll is true (huge "IF"), I expect the Liberals would win on those primaries. Enough of Danby's vote comes from conservative Jews for preferences to leak to Liberals, IMHO.

    1. It's a tricky one. They needed 83% off Neil Pharoah and micros in Prahran and got 84%, but the factor you mention does raise doubts about whether they could get 74% off preferences mostly from Danby.