Saturday, June 18, 2016

Rolling Poll Roundup: ReachTEL vs The Rest?

2PP Aggregate: 50.3 to Coalition (ended last week at 50.6, week before 50.3)
Seat projection for this 2PP: probable Coalition majority (estimated 77-67-6)
(at time of starting this rolling roundup, aggregate was 50.7)

2PP aggregate tracking graph.  Last updated 21 June
As foreshadowed at the end of the previous session I've decided to put out a rolling poll roundup thread to cover the time from the new polls out on Friday night until at least the expected release of Essential on Tuesday.  Updates will be added at the bottom as new polls arrive.

As I write a difference between one poll, ReachTEL, and the other most recently released polls (Ipsos, Morgan and Essential) has developed.  ReachTEL had the Coalition at 50-50 2PP last week and 51-49 this week, using respondent allocated preferences, but if last election preferences are used both polls come out to about 52-48 to Coalition.  Meanwhile Ipsos, Morgan and Essential have all issued headline rates of 51:49 to Labor.  ReachTEL had the Coalition primary on 42.7 last week and 43.5 this week while Ipsos had it on 39 and Essential on 41.  We're still looking for Morgan's primary results and expect to find them somewhere between the black box of MH-370 and the sign that reads "Beware of the Leopard", but the Coalition was probably on 40 or less with them too.


For those who believe the majority verdict here, it might be worth noting this: ReachTEL polled at the 2013 federal, the 2014 Victorian and the 2015 Queensland and NSW elections.  Essential and Morgan polled all of these and Ipsos polled Victoria and NSW.  In terms of accurate predictions of the primary votes, ReachTEL leads Morgan 4-0, Essential 4-0 and Ipsos 2-0, and a lot of those matchups were not even close.  

I aggregated the Ipsos at 48.9 for Coalition and the ReachTEL at 52.3, and as a result my aggregate ended the week at 50.6 (its highest end-of-week reading for the Coalition for nine weeks) and has since gone to 50.7.  The aggregate is, however, based on last-election preferences and there is a case that it is slightly generous to the government based mainly on uncertainties surrounding the Xenophon Team.  This is discussed further below.

There's an elephant in the room, by the way, and I've not seen much discussion of it.  This week the UK votes on leaving the European Union.  If (as mostly expected) it votes to remain, there will be no impact on the election.  But if it votes to leave, the world will suddenly be a different and more economically unpredictable place, and the Coalition's hand on economic responsibility will be greatly strengthened.  

Leaderships and debate

A little bit more leadership data this week.  Ipsos has Malcolm Turnbull up two points to a net +5 rating (47-42) and Bill Shorten at -4 (43-47, Shorten's best since April 2015).  The Preferred PM gap closes another 4 points to a nondescript 48-34 lead to Turnbull. 

ReachTEL, however, finds Turnbull stretching his lead a couple of net points with no change to speak of in their personal ratings.  One thing to note is a spike in the percentage considering Bill Shorten's performance "Very Good" (to 12%), probably caused by increased familiarity and impressed responses to his campaign-trail performance.

ReachTEL found only a slim lead (52:48) for Labor on trust to manage rollout of the NBN network, and the response was very party-polarised.  I suspect that this is a different question to whether voters prefer an all-fibre network or the Coalition's version, since it becomes one about service delivery as well as technology, and some voters asked if they trust Labor to deliver anything will just think "pink batts" or "budget surplus".  Voters also overwhelmingly think that Medicare is more likely to be privatised under a Coalition Government than a Labor one.  I would have not even bothered comparing the parties, and asked if voters believed the Coalition was likely to privatise Medicare, since that is the subject of a major ALP scare campaign.  

Debate

A Facebook-hosted debate was held at the ridiculous time of 6 pm Friday.  I thought Shorten won the debate (after roughly sharing honours in the first two), and many commentators did so too for various reasons.  Among other things I think Turnbull comes across badly when he uses his "can I just say this to you" repartee on ordinary people as opposed to journalists, and for this reason he is less well suited to people's-forum type debates than press-club types.  Town-hall debates also have an irritating tendency towards a left-wing issues mix.  But the whole novel debate setup was undermined by the gimmicky decision to award the result based on the vote of a panel of just thirty "undecided voters", who voted for Shorten 17-7 with six not even pressing a button.  In polling terms, Shorten's 71% result had an 18 point margin of error. Such a small sample size is as bad as watching a live focus group. If that's the best you can do, bring back the worm.

Seat Polls (Coalition vs Labor) vs National Polls

Overnight Newspoll released a set of seat polls, which showed the Coalition leading 54-46 in Herbert, 53-47 in Lindsay, 52-48 in Bass and Dunkley, 51-49 in Corangamite, Robertson and Brisbane, with Macarthur and Capricornia both 50-50 and Labor leading 52-48 in Burt.

In common with other seat polls that have been released by Galaxy and ReachTEL, these polls are broadly consistent with what would be expected based on the state swing and personal vote effects.  This can be seen in the overall figures compiled by William Bowe, which show an average swing in seat polls of 2.1 points, slightly less than the overall aggregate national swing.  In many Coalition-held marginals the party is assisted by sophomore effect caused by its winning the seat from a Labor incumbent last time, but this is partly cancelled out by seats like Dunkley, where the Coalition is affected by retirement.  So this is overall what we should expect to see.  (See also Tally Room for a similar conclusion).

Not only is the general seat poll picture therefore consistent with national polling, but even if it wasn't, we should ignore it, because seat polls were so badly skewed in 2013.  Seat polls shouldn't be considered capable of shifting our views about the result all that much, unless they show something that is consistently odd but that is not a national skew to one side.  Probably the most important thing seat polls are doing is casting doubt on the swing shown in national sampling for WA.  If the national WA samples are correct then Cowan, Hasluck, Burt and Swan would be expected to be toast, but seat polling has shown Cowan and Hasluck at least to be closer than expected.

Currently my federal seat model has the Coalition losing only these seats: Barton, Paterson, Dobell (all notionally Labor), Petrie, Capricornia, Solomon, Cowan, Burt and Mayo to NXT.  However it's expecting another two or three Coalition seat losses overall, and perhaps even a Coalition gain somewhere (though as the sum of very small chances in a number of Labor seats.)

Respondent vs Last-Election Preferences

In a rearguard against the Newspoll, someone in Labor has finally provided some systematic data about ALP seat polling, telling Barrie Cassidy that Labor's internals have the party ahead or square in 18 presumably Coalition-held marginals (a breakdown into ahead and square would have been nice here) and trailing by 2 points or less in six others. Specific claimed results have included 57-43 to Labor in Paterson, 53-47 in Robertson, 50-50 Hasluck and Coalition leading just 51-49 in Pearce.

Cassidy explains the issue by reference to respondent preferences, arguing that Labor's vote was down in 2013. If Labor is up by a few 2PP points this time then, he says, you would expect Labor's preference flow to be up "by that amount".  There are many major problems with this argument.

Firstly, if this all held water then you would expect that Labor's preference share in 2013 would have fallen (given that its vote share was down).  This didn't happen; Labor's share of Green preferences in 2013 rose from 78% to 83% and their share of Others preferences rose from 42% to 47%.  The same theory also failed in 1993, 1998 and 2007 (Labor's vote share rose in all these cases but preference share, when adjusted for changes in the support level for established minor parties, was unchanged), 2004 (swing away from Labor, preferences unchanged), 1990 (Labor was saved from a vote share crash by greatly improved preferences) and 2001 (swing against Labor, Labor preference shares improved).  In fact the only times this theory has actually worked recently were 1996 and 2010, and both of those were cases where both Labor's primary vote and its preferences crashed.  The last time Labor significantly gained on both 2PP and preference flows by party at the same election was 1980.

Secondly, respondent versus last-election differences do not cause four-point gaps in a seat like Robertson.  In Robertson at the last election, Greens preferences flowed 86% to Labor and Others preferences flowed 54% to Coalition, a more or less typical national breakdown.  To get from the Newspoll primaries (45-41 to Coalition, 8 Green 6 Other) to the Labor-poll 2PP (53-47) it would have to be assumed that the Coalition's share of all preferences was falling from an expected 31% to about 14%, even though its share of Green preferences can hardly fall much further.  Assuming a 90% Greens flow, Labor would need 80% of Others preferences.  So there must be substantial differences in the primaries for Robertson between Newspoll and the ALP.  (A more fruitful line of enquiry might be how Newspoll gets 51-49 to Coalition off their Robertson primaries, which would normally imply a 2PP of about 50.5% to Labor, but there may be rounding or a drilldown into the nature of these "others" at work here.)

Thirdly, let's have a general look at how respondent preferences are faring in national polls.  In Ipsos, the last four polls have on average shown Labor half a point better by respondent preferences than last-election preference flow, with the last two showing no difference.  However these four polls have attributed 57% of the non-major vote to the Greens, compared to 40% at the last election, and Greens voters preference Labor more heavily.  Ipsos uses a batched preference flow method that ignores such changes.  So even if there had been no change in preferencing breakdowns for either Greens or others individually, we should expect Ipsos to show Labor doing about 1.4 points better on respondent preferences, 0.9 points higher than is actually happening, meaning that Ipsos implies Labor's preference flow is worse.   Ipsos has thus provided no evidence whatsoever of an improved Labor preference share - if anything the contrary.  (It is possible that Green voters who preference the Coalition in Ipsos are actually often Liberal voters, which would also explain Ipsos' huge Green vote, but that is not good news for Labor either.)

If we move to ReachTEL, the picture is slightly different.  In the last six ReachTELs (since the start of May) the average difference between respondent and last-election preferences has been one point in Labor's favour.  And in the last three fully published Morgans the difference was negligible (an average of 0.2 points in Labor's favour).  So the evidence on respondent-vs-last-election preferences from public pollsters is actually all over the shop.  Overall I suspect there will be a slight move in Labor's favour because of the uncertainties created by the Nick Xenophon Team, but the case that there will be more than half a point in it is weak.  It should be weak too.  Why should Greens voters be more likely to preference a Shorten-led ALP against a Turnbull-led Coalition than they were to preference Labor against Tony Abbott last time?  Why should Christian voters disappointed with the Coalition's half-hearted moves towards same-sex marriage preference a Labor Party that is phasing out even letting its MPs vote against it?  Xenophon aside, if many Others voters are really swinging to Labor then who are these people?

Yet another point here is that it is not just the last-election Newspoll seat polls that have Labor struggling in the marginals about as much as they should be expected to struggle.  ReachTEL - which is using respondent-allocated preferences - is getting the same thing.

Of course, anyone who wants to play mix and match with the polls can do so and get the result that they like. Combine, say, Essential's primaries and ReachTEL's finding on respondent preferences and you get 23 Labor seat wins and majority government (just!) with a 52-48 2PP.  Combine ReachTEL's primaries with Ipsos' finding on respondent preferences, however, and you get Labor hardly winning a seat.  Throw in the chance of an industry wide Poll Fail (which is higher than it has been for some time) and possible shifts in the last two weeks and almost anything can happen.  However, the balance of probabilities is that the key questions about polling land somewhere in the middle, and if so that favours the government winning the 2PP in the majority of seats.

Non-Classic Seat Polls

Newspoll has three new seat polls of seats that will or might not come out as classic Coalition-Labor contests.  In Batman, where the Greens' seemingly crushing position against David Feeney has been mangled by Malcolm Turnbull's call on preferences, Newspoll still has a seven-point swing against Feeney for a 53-47 Labor-vs-Green result.  That seems about right to me - he's had a truly terrible campaign but maybe not enough to blow all ten points.  Things might still get rocky here should the Coalition be clearly headed for the win nationwide come election day, so I don't think the Greens' chances in Batman can yet be entirely ignored.

Newspoll produced yet another Mayo poll that shows Jamie Briggs losing (indeed on the primaries of 42-38 he might not even make their 48-52; that said the primary for Rebekha Sharkie of NXT is much higher than other pollsters have had so far).  Newspoll also produced a very rosy outlook for Christopher Pyne in Sturt, with 50% of the primary vote.  No poll has yet convincingly showed Pyne losing and I'd rather see more data on Barker, Boothby and Kingston, for example.  Odds in Mayo have crashed to $1.85 apiece on Sportsbet and $1.80 for Briggs vs $1.90 on Luxbet.

There is still the possibility that the NXT vote will surge in SA in the final week and lead to multiple seat wins and perhaps cause a hung parliament.  The latest data point was a 26% vote share for Others in SA/NT in Ipsos, bearing in mind that NXT scores nothing in NT, so that probably puts him on high teens at least in South Australia without even being in the readout. Ipsos samples are small though.

NXT are not the only minor party that may be surging.  Which brings me reluctantly to ...

I Was Part Wrong About Hanson

Recently I published an article making the case that the new Senate voting system does not advantage Pauline Hanson's One Nation.  That article's fine and I stand by every word of it.  If Hanson is elected to the Senate, it will very likely be because she has polled well enough on primaries that she would probably also have won under the old system, especially on ticket preferences from right-wing (and even left-wing) micro-parties.  

What is not bearing up so well is my initial estimate that Hanson would struggle to poll a competitive primary, following her party's very weak performance at the 2013 election.  I now think that the speculation about Hanson having a good chance at that early stage was right, but was right for the wrong reasons.  Firstly, there is some polling evidence that Hanson may be doing very well indeed on primary votes.  For instance (and it's not the sole example) the current Newspoll has One Nation on 11% in Herbert, which in 1998 polled the average One Nation vote for Queensland as a whole.  Hanson could very well become Senator Hanson, not just because of the collapse of Palmer United, not just because of it being a double dissolution, and not at all because of the change of electoral system.   There are other things going on here.

The 2016 election campaign has pitted a moderate, tolerant Liberal Prime Minister against a socially progressive urban ALP leader whose campaign pitch is rather left-wing.  It's been a very upper/middle-class, technocratic, rationalised civilised election campaign with barely a dog-whistle heard. It's been a PM who is excited by innovation and prosperity vs an ALP leader with "100 positive policies". Old-style, irrational, populist protectionist right-wingers could relate to Tony Abbott, since he was one of them.  But he got rolled, and there's really no major-party voice for those feeling "left behind" by it all, or who just feel that both leaders are too left, too politically-correct and not paranoid enough about Islam.  It's the ideal vacuum for the Hansons and the Lambies to be filling - and as with Donald Trump, Hanson gets free votes because the media give her publicity.  Every article about her chances in the Courier-Mail gives her still more appeal as a way to give major parties the finger.  More One Nation seat polling data will be watched for with interest.

I will update this piece following Newspoll tomorrow night, if not before, and around mid-week I will update the seat betting picture.

19 June: AEU Seat Polls

There are reports of some Australian Education Union commissioned ReachTELs said to show Labor ahead 55-45 in Eden-Monaro, 52-48 in Page and trailing 49-51 in Lindsay.  Neither Eden-Monaro nor Page have yet attracted a neutral seat-poll despite the interest in the former famous "litmus seat" (which is widely thought to be at risk of losing its status this election); the result in Lindsay is better than in the public seat polls.  Presumably these results are based on respondent preferences and the primaries are yet to be seen.  (Update: Yes - William Bowe relates that the Eden-Monaro result is blown out by an unlikely flow of respondent preferences.)

19 June: Recency Update

It being Sunday, the Ipsos poll taken from Tuesday to Thursday no longer qualifies for the special weighting for polls taken in the last four days, while the ReachTEL still does.  This drives my aggregate to a very bullish 51.0 for Coalition. but the two most recent ReachTELs are between them carrying 43% of the aggregate's weight - because of the age, unreliability and in Morgan's case data inadequacies of the others.  So we have to see if this figure survives Newspoll this evening.  The heavy weightings for the final three weeks are designed to stop the aggregate from being dominated by old rope, but polling data at this stage is much sparser than in 2013, so maybe those weightings are too strong in this case.  We'll see!

Through the election campaign my aggregate's seat projection and BludgerTrack have seldom differed by more than a seat but there is presently a six seat difference off a 2PP difference of only 0.6 points.  This becomes seven seats considering that I am giving Mayo to NXT.  This gave me a lot of pause for thought so I tried rerunning my model using the exact same seat weightings as BludgerTrack, the same 2PP and stripping out the seat polls.  Even then there was still a three-seat difference.  I looked at this by state and found the difference to be over a seat in NSW but no more than about half a seat in any other state.  Only 75 Coalition seats off a 2PP of 50.4 is lower than I would expect.  A possible cause is rounding by state (my aggregate just calculates a national total).  It should be kept in mind that Bludger Track was stunningly accurate in its 2013 final seat projections.

19 June: Newspoll

It really is ReachTEL vs the rest now with Newspoll coming out at 50-50 off primaries of Coalition 41 Labor 36 Green 10 Other 13.  I've aggregated that at 49.7 for the Coalition, bearing in mind that Newspoll has consistently seemed to lean Labor by several tenths of a point since Galaxy took over.  That puts my aggregate back to 50.7 2PP.

Malcolm Turnbull records a new personal worst netsat of -15 (36-51), albeit only by a point.  Paul Keating won with -25, but you really don't want to be carrying a PM with this sort of rating in a campaign, especially not when you have his face with your candidate on every second billboard.  Bill Shorten records an almost identical -16 (35-51) and the better PM reading is a nondescript 46-31 to Turnbull (again, the standard lead given the 2PP).

The Xenophon Team chalks up 29% in South Australia and I will have to look carefully at their data across various pollsters over coming days to see whether it is now worth projecting them a second seat or even more.  If they actually polled 29% all hell would break loose as they would probably win most of the state.  However this is just one data point that would be from a small sample.

In seat poll news, Barnaby Joyce has a precarious 51-49 lead over Tony Windsor in a New England seat poll.  This is off primaries of 48-36, but with most of the rest Labor and Greens preferences, Joyce doesn't want to fall short of a 50 primary by too much here; the implied 81% flow to Windsor is quite believable.  I've been sceptical along that Windsor would actually win, but even that he is this competitive at this stage would be a worry, unless the Nationals have their own polling showing different.

Meanwhile, the Liberals are attracting lots of flak from people who wouldn't vote for them anyway for this apparently terrible ad.  I'm not likely to be a guest on Gruen Transfer anytime soon, but does stuff like this really work?  And, as a novel break from the endless candidate malfunctions at this election, Labor has suffered an ex-candidate malfunction just where it didn't need it, Cowan.

20 June: Kennedy

It would be interesting to see some seat polling of Kennedy, held by Bob Katter.  Katter is at $1.01 at Sportsbet but he is 71, crazier than ever and almost lost last time.  Today excerpts have been circulated of an eccentric 12-page document Katter sent to his constituents, and he also appears to have lost some influential supporters following the screening of an ad in which he shoots rival candidates.

20 June: Morgan Greens samples and Batman

There is more data on those Morgan Green samples up now.  The sample size is apparently about 310 per electorate.  On the published figures, the Greens would win Batman by a whisker (I get 50.5% two-candidate preferred) based on last-election preferences, but I'd take at least a few points off that for Morgan's usual skew, and then not pay too much notice anyway based on the sample size.  They would also only just hold Melbourne (I get 51-49).

Thanks to Matt Lennon for noticing that the link also includes seat results for 20 seats, though some of the results are obviously silly.  The problem with Morgan (aside from sample size and house effects) is that its SMS polling is a panel-poll and it is therefore extremely prone to local pool selection effects.

Meanwhile there is an internal poll fight between Labor and the Greens with the Greens claiming to be leading on primaries 41-28 in Batman (55-45 respondent allocated, which is an especially dubious method in such a seat) while Labor claim to be leading the primary count.  Labor are also alleging push-polling (by which they probably mean skew-polling, but who knows since full details of neither the poll nor the accusations are available).  It is alleged that the poll included negative claims about Feeney.  Whether these claims were made before or after the voting intention and approval questions is the crucial test of whether or not skew-polling (at least) occurred, but this information isn't available.

21 June: Recency update

The ReachTEL and the Newspoll no longer qualify for special weighting, which means the Newspoll as the more recent is now weighted higher, so that brings us down to 50.4 to Coalition, and costs the Coalition two seats on the projection.  It's still close, and it's still very loseable, and the Coalition have seemed rather jittery in the last 24 hours.

21 June: Adelaide

The seat of Adelaide has from time to time got some mentions as a Coalition target.  There's a report of an ALP internal ReachTEL - but with a very small sample size of below 350 - said to show Kate Ellis trailing 49:51.  The argument is that inner city voters are still responding positively to the replacement of Tony Abbott with Malcolm Turnbull.  Anyway such a close margin in such a small sample can only show the seat is competitive.

21 June: Barker

Barker was on my radar early as a seat at possible risk to NXT.  A CFMEU-sponsored ReachTEL has NXT ahead 52-48 on a large sample ReachTEL.  Again, commissioned poll, but I would not be surprised if this seat is in play.

21 June: Menzies

Stephen Mayne is running a rather colourful campaign against Kevin Andrews, premised on Andrews being an Abbott man who is disloyal to Malcolm Turnbull and describing Andrews as a "1950s dinosaur and social extremist".  It should be kept in mind that while all of this is absolutely true, the Abbott government's scrapped six-month dole-wait policy was an Abbott office brainchild (if anything involving brain is the right word) that was resisted by Andrews and Eric Abetz because it was too extreme and nasty even for them.  Anyway a ReachTEL commissioned by the Mayne camp has Andrews on 51% and Mayne on 7.3%, which seems a rather large gap to bridge to get Mayne into a competitive position.  Andrews' 2PP (presumably vs Labor) is not much affected.

22 June: Essential

Essential's result today was another 51-49 to Labor, but the Coalition's primary has slipped a point and the 2PP was probably close to rounding to 52-48.  That makes this the first poll from anyone other than Morgan for six weeks to have Labor in a probably winning (rather than tossup) territory, and knocks another tenth of a point off my polling aggregate.  The Coalition also sheds another seat there, but not to Labor, rather to NXT.  Poll Bludger has reported an NXT aggregate reading of 24% and has access to more reliable state drilldown data than me.  I think that some, at least, of the polls are overestimating NXT based on a general tendency to overestimate Others votes, but even putting it at, say, 21% it is hard to deny NXT a couple of seats.  This is also consistent with what little neutral seat polling exists.

Essential however finds that perceptions of which groups would be better off under a Coalition government have generally moved in the Government's favour.  They also find that voters only slightly more trust the Government to secure jobs than Labor, although jobs are a major part of the Coalition's re-election campaign.  A 5% result for "Pauline Hanson" on this question suggests a lot of right-wingers in the Others category for this sample.

22 June: Melbourne Ports

Crikey today reported a Greens ReachTEL of Melbourne Ports with the Liberals on 44.7 (+3.7 compared to 2013), Labor incumbent Michael Danby on 23.7 (-8) and the Greens on 20.2 (level).  It is not stated how undecided voters were treated.  I greatly doubt that these results are accurate, but if they were, then the Liberals would probably win.  A handy comparison is the state seat of Prahran in which the Liberals with 44.8% were beaten by a whisker by the Greens (24.8%) with a similar result had Labor (25.9%) managed to remain second.

22 June: More Morgan Wibble ... Stop It Now This Stuff Is Painful

Morgan has released some issues findings from Indi, Higgins, Mayo and Menzies.  These are based on SMS samples of 308 voters across the 4 seats, so the individual samples are on average 77 voters per seat.  If Morgan actually used random sampling it would be have a maximum margin of error of a mere 11% on these samples, but it's a panel pollster with dangerous undertones of opt-in, so it is probably not that good.  The seat-by-seat results are useless but I will mention the Menzies one because they claim the candidates are roughly evenly split.  Even Stephen Mayne's own polling - from a much bigger sample - had him trailing 51% to 7% as discussed above!

22 June: More Commissioned ReachTELs

Commissioned ReachTELs for the NSW Teachers Federation have Labor up 55-45 in Eden-Monaro, 54-46 in Macquarie, Page and Lindsay and 53-47 in Dobell and Gilmore.  These are all based on respondent preferences, but only in Eden-Monaro and Page does the flow look unusually strong; in most cases the issue is that Labor are ahead on primaries.  The Eden-Monaro poll has the same 2PP as the other commissioned poll of the seat and Macquarie and Page have not had neutral polls yet, but the figures are much better than the neutral polling of Lindsay, Dobell and Gilmore, and overall show a swing of 7% since the last election.  The suggestion is a shift in voting intention in these seats following the ALP launch and Medicare attack, although the sponsor of course blames their pet issue, Gonski funding.  Neutral evidence is required to back these results.

While I have praised ReachTEL's state and national performance above, I should point out that seat polls by anyone were pretty ropey in 2013.  I am extra-wary of these results but we will find out soon enough if there is something startling going on!

24 comments:

  1. Thoughtful and interesting analysis - enjoyed reading it.

    ReplyDelete
  2. Hi Kevin, question. This was in the Oz on Friday regarding their marginal seat polls. 'While the Coalition has dropped 3.6 per cent in Lindsay, Labor is also down 3 per cent, with the NXT snaring 6 per cent of the primary vote. In two-party-­preferred terms, there has been no gain for the ALP in the must-win electorate, with the vote unchanged at 53-47 in favour of the Coalition.'How accurate can NXT preference allocation be, based on 2013 figures, given NXT didn't run there and it appears there are a lot of question marks about which way NXT voters will split (assuming 6% of people actually decide to go NXT when they have the pencil in their hand)? thanks @warrickjordan

    ReplyDelete
    Replies
    1. I am not sure what Galaxy (who administer the Newspoll seat polls) are doing about NXT last-election preferences. As usual there is not any detail published on this. On national last-election preferences Lindsay comes out to 53.1 on the published primaries, so the NXT flow would have to drop to 40% to Coalition to push the 2PP below 53.

      The other interesting question re Lindsay is who the hell the 12% for Others think they're voting for. Others in Lindsay in 2013 were 11.3% but that included PUP 5% and One Nation 2%. Neither are running this time around.

      Delete
    2. A chunk of that 12% might be for local councillor and ex-Liberal Marcus Cornish, running as an independent to the right of Scott.

      Delete
    3. Thanks Geoffrey. Yes that would explain the high Others vote. He said he would preference Labor (which would create some issues) but whether his supporters actually do so is another story.

      Delete
    4. If you're right about a significant parcel of voters ditching the Liberals because Turnbull isn't conservative enough, that should show up in Lindsay, especially given the rumours that Fiona Scott voted for Turnbull. Those votes could be flowing to Cornish, Liberty Alliance or Australia First, but should go back to the Liberals on 2PP.

      Delete
  3. Do you know if the federal Newspoll is from the same sample as the marginal Newspolls? Or a completely different poll?

    ReplyDelete
    Replies
    1. I would expect completely different as the sample size for the federal Newspolls only allows for about a dozen voters per electorate, and also given the sample dates.

      Delete
  4. Kevin, are you saying that ReachTel's respondent allocated preference method did better than Essential, Ipsos and Newspoll at the last four elections. If so, shouldn't everyone be adopting that?

    ReplyDelete
  5. No. I am making no comment about the relative accuracy of preference allocation methods there, or the accuracy of 2PP estimates, only about the accuracy of primary votes. I was very explicit on this: "In terms of accurate predictions of the primary votes [..]". In fact, ReachTEL used last-election preferences for its 2PPs at all those elections except NSW. Morgan's 2PP for Queensland was closer than ReachTEL's, because its primaries were inaccurate, but this was cancelled out by the preference shift.

    Overall, for the federal election and Victoria, last-election and respondent preferences were about equally accurate, for Queensland it is suspected respondent preferences would have been more accurate but no public pollster did them, and for NSW the more conservative estimates of respondent preferences were most accurate while the wilder ones and last-election preferences were both inaccurate. However, Queensland used and NSW still uses different preferencing methods to federal elections, and the long-term record in federal elections is in favour of last-election preferences.

    ReplyDelete
    Replies
    1. Many thanks Kevin. Most enlightening. Looks like the split between last-election and respondent-allocated preferences for this election is going to be very important and we'll see how history stands up.

      Delete
  6. I don't know if it's partly in jest, but it's not really "reachtel versus the rest". All polls apart from maybe ipsos are within margin of error

    ReplyDelete
    Replies
    1. Margin of error relates to a single polling result. For a case of two polls in a row displaying similar differences the effective margin of error is substantially smaller. In any case, even when two polls are within MoE of each other, if their results differ substantially they are still each providing different information about the possible spread of results. When we are dealing with an election which has not been determined with 95% probability, I am not going to get concerned about whether two poll results differ from each other at that probability level or not. Even if two results are only near each others' MoE that can still shift an estimate of what is going on quite a lot.

      I am going to write a lengthy piece about all the confusions a basic understanding of margin of error can cause in analysing polling sometime, though it worries me that I have seen so many I will probably never finish it.

      Delete
  7. Hi Kevin,

    A few commentators have remarked that the Greens haven't picked up much of the Primary vote considering the high percentage now attributed to Others in many polls. These commentators seem not to consider the idea that a large portion of that Others vote may flow to the Greens in preferences before it flows to the ALP or Coalition. Their argument seems to be that for many the Greens are now just another major party and are prone to the same discontent with the political mainstream as the ALP and LNP. It seems to me that if that's really the case, and people are set on voting for anyone but the majors then there's a good chance a large chunk of that vote would then flow to the Greens before the other two. That is, perhaps the Greens might be considered less major than the other majors. This may matter in seats like Gryndler, where the last poll I saw has the Greens in 3rd, 0.5 % behind the Liberal, but with 12 % for others! For this reason I'd be interested in what a 3PP vote would look like. Do you have any insight in this area?

    ReplyDelete
  8. If that Grayndler poll is the Morgan sample (the only public poll I've seen for Grayndler) then I'd pretty much ignore it - Morgan samples are too small, skewed to non-majors and prone to local pooling effects. Plus the data is anything from recent to six months old.

    I don't believe the Others vote is going up as much as Morgan and the latest Ipsos say. PUP is gone and is replaced as a rough proportion of Others voters by NXT, and there might be an increase for xenophobic right micros like One Nation. I don't believe there is a serious increase in voting for left micros. I suspect that working out whether the Greens can get into second from say 2-3 points behind in some of those inner city seats (plus maybe Richmond) will be a seat-by-seat matter that varies according to the mix of parties running in each seat. As for those who are picking Others even when there are no prominent Others to vote for in their electorates, a lot of those will vote for major parties.

    ReplyDelete
  9. the Wentworth electorate poll is intersting suggests 12 to 15% swing and Turnbull in Danger....... Barker ... 52/48 for nxt

    ReplyDelete
  10. I'd take even the poll commissioned by Turnbull's opponent more seriously than that Morgan sample.

    ReplyDelete
  11. Kevin - On the other hand, Bob Katter didn't empty his gun. That must mean something up North.

    Wentworth is an artsy fartsy high-wealth progressive electorate which was marginal in 2004 - 2007. Huge numbers of voters would feel betrayed by his righ-turn. I will not be a bit surprised if there is a massive swing against Turnbull. How big - who knows.

    Just out of curiosity: has the reliability of previous election preferences as a predictor depended basically on the stability of green voters. Are we now seeing a new component of disenchanted anti-incumbent "others" coming through? Is there any way of assessing that.

    ReplyDelete
  12. There is a good chance that One Nation voters will strongly favour the Labor party this time. Labor has muscled up on border protection, has a male leader and is running a brilliant "scare" campaign on medicare, NBN, etc etc. I bet you that, after border protection, most One Nation supporters worry about medicare.

    ReplyDelete
  13. It's not just Green preferences; the preferences of generic "Others" have also been very stable from election to election at many elections in a row. There was a 5% shift to Labor in the flow of Others preferences in 2013 and a 7% shift in 1990 - the latter being caused by the rise of green candidates in the Others category. There was a 5% shift to the Coalition in 2004. A 5% shift now would be worth about 0.65 2PP points. Apart from that, even with such ructions as the rise of One Nation and other cases where the composition of "Others" changes, shifts have been less than 5% at every election since 1980. At the 1980 election Others preferences shifted from 30% for Labor to 47% for Labor but this had little impact since the Others vote was only 2%. I think if there is a major shift in the Others department it will come from NXT supporters.

    One Nation preferences went 55% to Coalition in 2013. They could well go more to Labor this time but I don't think hugely so as I suspect most of their vote will fall in conservative electorates. I also don't think Labor has really "muscled up" on borders to any greater extent than its usual ritual of doing so before each election. One Nation are only polling 1% for the Reps nationwide although that might be an underestimate given their strength in Queensland.

    Wentworth does have a slightly less lopsided history than, say, Warringah, but the major reason it was marginal in 2004 at least was the dumping of the sitting member Peter King in favour of Turnbull.

    ReplyDelete
  14. Kevin - Many thanks. That is most interesting. I do still wonder if there is an anti-incumbency "other" group brewing out there after so much political instability (maybe even identifying themselves as NXT supporters) but there is no way of registering it (if it exists). Wouldn't be surprised either if there are a lot of "shy" One Nation supporters out there as well. Highly unlikely Turnbull will lose Wentworth, but two (tiny sample) polls now suggest he's going to get a big kicking. People in Wentworth are very different to the surf-billies in Warringah.

    ReplyDelete
  15. It would be very useful to see respondent preferencing data on NXT voters broken down into SA and outside SA. NXT voters will be receiving an open preference card so it might be that for them respondent preferences are actually accurate. The only NXT preference estimate I saw was the early 52:48 to Labor from ReachTEL but that was a small sample and their vote has increased since.

    ReplyDelete
  16. Don't forget the reachtel Grey poll. See qn 1c.

    https://www.reachtel.com.au/blog/7-news-grey-poll-9june16

    ReplyDelete
  17. It's also possible that the NXT preferencing data was embedded in the internal polling that the ALP leaked to Barrie Cassidy. That explains the higher polling for Labor in Lindsay etc etc. A tantalising thought. But we just don't know.

    ReplyDelete