Sunday, May 15, 2016

ReachTEL Points To Tasmanian Status Quo

ReachTEL (federal) Bass 51-49 to Liberal, Braddon 53-47, Lyons 51-49, Franklin 54-46 to Labor, Denison 66-34 to Wilkie.
(See comments on 2PP estimates below)
Poll shows Lambie support sufficient for one Senate seat.

This weekend the Sunday Tasmanian is publishing ReachTEL polling of the five Tasmanian federal seats.  The poll includes several questions and other questions will be released by the Mercury through the week.  At this stage two questions have been released - voting intention and a question on the Budget and economic management.

Tasmania has three marginal Liberal-held seats: Lyons (Eric Hutchinson, 1.2%), Braddon (Brett Whiteley, 2.6%) and Bass (Andrew Nikolic, 4.0%).  All were won from Labor at the last election so the new members should have some buffer against swings because of their new personal votes.  It also has the Labor-held seat of Franklin (Julie Collins, 5.1%) and independent Andrew Wilkie's seat of Denison (15.5% vs ALP), neither of which have been considered really "in play".  Until now there has been no released polling of these seats since Malcolm Turnbull became Prime Minister.  There have been small state samples aggregated by Poll Bludger, which have generally looked OK for the Liberal incumbents except for a wobble after the short-lived "state income tax" proposal.

This is the first large-scale poll we have seen, but it has a number of unusual aspects that make interpreting it challenging.  While the Liberal incumbents could take some heart from it, there is plenty of room for other interpretations.



The Poll Design

The poll uses large samples of around 600 voters (based on recorded-voice polling or "robopolling" of both landline and mobile phones) per seat.  Such large ReachTEL samples were also the only public seat polling available for Tasmania before the 2013 election.  They predicted the winner of every Tasmanian seat correctly, but had the Liberal 2PP vote on average 4-5 points higher than the ultimate outcome.  There was a widespread national tendency of published seat polling to skew to the Coalition compared to both national polling and the end results in 2013.  It is probable that pollsters have taken measures to combat this, but I am not aware of any details of such measures.

The main oddity of the poll design as commissioned is the inclusion of the Jacqui Lambie Network in the readout.  At this stage, JLN appear to be contesting the Senate only in Tasmania.  While including the JLN is very useful in terms of getting some kind of a vague idea of their possible Senate support, this does mean that respondents are being asked what sounds like a House of Reps question ("If a federal election were to be held today, which of the following would receive your first preference vote?") but includes a Senate-only party.

The published preferences are all respondent-allocated.  If last-election preferences were used I get some different results.  See discussion of the 2PPs below.

As with many other recent ReachTEL polls (but not all) questions have been asked in a two-step format - in the first stage voters can press a button for "undecided", but those who pick that option are then asked which party they are leaning to.  While this can provide useful data about the softness of particular party votes, it also means the results of the headline question are not directly comparable to those of other polls like, say, Newspoll.

Primary Vote Results

I have published the results below in a format that more closely resembles that of other polls, eg with the Undecided voters all reallocated.  I am showing the swing for Labor, Liberal, the Greens and Wilkie since the last election for each sample.  It should be kept in mind that the in-theory margin of error for any given result here maxes out at about 4%, but that is ignoring the impact of scaling, which tends to increase it slightly. Furthermore because so many individual figures are being published at once, it is quite likely that at least one or two will be outside the margin of error.


The inclusion of the Lambie Network creates swings against parties that may not actually exist in a lower house vote (though some of these voters may vote for Ind/Others).  However, the Lambie vote in this poll by electorate very closely follows the PUP vote at the last federal election.  Whether the Lambie voters and PUP voters are more or less the same people is not so clear.

Bearing all this in mind, the poll shows an average primary "swing" of 1.1 points to the Liberals, 2.3 points against Labor and 3.5 points to the Greens.  Seat by seat, the Green vote in Lyons at least looks too high and the Labor vote there too low, and I think this is also true of Franklin and probably also Denison.  (The Labor brand can't be too trashed in Denison given that the party just won a Legislative Council seat covering slightly over a third of it with a 10% swing!)  Even allowing for this, the primaries look rather strong for the Liberals.

There are three possible explanations for this, any or all of which could be valid or partly valid:

1. The poll could be supporting the finding of the other state samples that the Liberal vote is holding up rather well in Tasmania.

2. The poll could be overmeasuring Liberal support, as the equivalent series did in 2013.

3. While the JLN votes and 2013 PUP votes are distributed very similarly, they might not be the same voters.  The Lambie Network might be "taking" more votes from Labor in this poll, and fewer from the Liberals, than PUP did.

A more House of Representatives specific survey design would have assisted in ruling out 3.

Two-Party Preferred

Here I will deal with Denison first.  The 2PP has been published as between Andrew Wilkie and the Liberals.  However, in 2013 the Greens' preferences in Denison split 55.8% to Wilkie, 40.7% to Labor and 3.5% to the Liberals.  With the split from "others" also slightly favouring Labor, my own estimate is that Labor would overtake the Liberals and make the final two in this sample, albeit by only 0.1 of a point.  It's very likely that the Labor vote is understated compared to the Liberal vote in this sample anyway, so the poll's Wilkie-vs-Liberal 2PP is likely to be irrelevant.  However, the 2PP for Wilkie against Labor would also be about 66:34 based on this sample.  Although the sample shows some primary vote swing away from Wilkie (probably down to declining influence and profile in a lopsided parliament, if real) he would still be returned easily on these figures.

For the other seats, the table above includes the pollster's published 2PPs for Labor and Liberal, but also my own estimate of the 2PP based on last-election preferences (2PP (LE)).  Here I have assumed the Lambie voters would split fairly evenly between the majors, as the PUP voters did in 2013, but this might not actually be the case.  Note that Green preferences in Tasmania flow very strongly to Labor.  Anyway, on that basis I find my "last-election" 2PPs to be about a point more favourable to the Coalition on average (52:48 Bass, 53:47 Braddon, 54:46 Lyons, 47:53 Franklin) than those published in this poll.  The one that really deserves comment here is Lyons.  There is just no way that a Liberal lead of 16.8 points in Lyons would really come down to only a 51:49 lead after preferences.  This would require Labor to get about 80% of all preferences!

There are some risks involved in using respondent-allocated preferences in small samples.  A sample of 600 has a non-major party vote of about 100-150 voters per electorate.  These subsamples then have a margin of error of 8-10 points on a good day, which means that the use of respondent preferences can easily cause up to two points of 2PP error by itself.  This then increases the effective error margin on the 2PP estimate.

However, it's possible (especially see the table for question 2 below) that the Lambie voters lean fairly strongly to Labor.  If this is the case then the published 2PPs all seem very accurate off the primaries, except in Lyons.  In the case of Lyons it may just be that the sampled minor party voters happened by random chance to be unusually pro-Labor.

So What Does It Mean?

Firstly, while the Liberals lead in their three seats in these samples, their leads are narrow.  If a properly random sample of 600 voters shows a 51:49 lead in an electorate (as is the published 2PP for Bass and Lyons) then there is actually about a 30% chance the party polling the 49% would win the seat.  Even if the 2PPs aren't really that close there would still be a realistic chance based on this sample that Labor would gain one of the three seats (could be any one).  The sample also doesn't prove Franklin is safe for Labor by itself; it is just consistent with the widespread belief that this is so.

 Also if the overestimation of the Coalition vote in seat-polling was seen to the same degree as in 2013, it's plausible these results could be re-spun as four ALP wins.  (I think this is unlikely).  If we take the poll as close to face value as we can, then it's not by itself saying any of the "Three Amigos" are rock-solid, but it is still better for their chances to be polling narrow leads than to have no information.

Lambie

While the poll gives us some insight into the size of the Lambie vote and hence her Senate prospects, this should be treated with caution.  Some voters who intend to vote for a major party in the lower house but Lambie in the Senate may have interpreted the question as being about the House of Reps only, so the poll may underestimate Lambie's appeal.  On the other hand, Lambie is named in the readout while the major party candidates are not.

Anyway with the undecided redistributed, the poll has the JLN on 5.1% statewide, plus what little it might get from Andrew Wilkie's voters.  Although the Senate quota is 7.7%, in practice because of the new preferencing system it is very likely a primary of 5.1% would be enough for Lambie to retain her seat - and this could well be an underestimate anyway.  The poll is therefore consistent with the idea that Lambie herself will retain but will not have enough vote for a second seat.  The poll provides no support whatsoever for the idea that the Greens will win more than two seats, while the strong Coalition primary points towards them gaining a fifth at Labor's expense (much as I have doubted this outcome recently).  Extrapolating such a poll to the Senate is always risky especially given the new parties that may well emerge and become factors.

Economic Management

Question 2 is rather interesting.


Economic management is a Coalition strength area so it would be expected that the Coalition would have a healthy edge here, in the context of a poll showing at most a narrow 2PP lead to Labor statewide.  However I think the "and the core issues facing Tasmania" add-on has muted the normal response to economic questions and created a lot more scope for a classically partisan response.  Thus, the lukewarm result for the Liberals may not be as disappointing as it looks.

I will update this article with more as more findings are released through the week.

PS I've had a request for the original primaries so here they are.  Click for larger versions:


Issue Importance (May 16):

A question on issue importance has been released by The Mercury (sorry, the text here is hard to read):


This is at least the third ReachTEL in the past year in which health has topped the list in Tasmania, albeit against various competition.  In the leadup to the 2013 election, "jobs" topped the list in Bass, Braddon and Lyons and was a close second to economy in Franklin.  It is possible that the more restrictive wording "job creation packages" has contributed to a lower finding in this survey, but that would not alone explain the massive differences.   There are no big surprises here - the preference for health among female voters and the economy among male voters are common and partly reflect the standard tendency for male voters to lean more to the right.

MP Ratings (Tuesday)

The Mercury has now also released a very important question on approval ratings for the five incumbent local members:

We see here that Andrew Wilkie is top of the pops with a massive +51 netsat, which highlights why attempts to unseat him at this election are unlikely to get close even though his primary vote in this poll wasn't all that high.  Voters often see Wilkie as having integrity and being something different from run-of-the-mill Canberra politics, and a close election campaign will do no harm to his claims of being able to make a difference.  Andrew Nikolic (Bass) scores an impressive +24.7 showing that his tendency to get into social-media scrapes has not harmed his standing outside the political bubble.

Eric Hutchinson (Lyons) has a reasonable net +9.8, but a very high 15.3% score for "never heard of them before", which is consistent with some doubts about I see about whether he has done enough profile-building.  He also has the highest "average" score of those listed.   Julie Collins (Franklin) also has a high non-awareness rate (perhaps a result of the obscurity of Opposition) but a strong net rating of +22.7.

Brett Whiteley has the weakest net rating in the survey (with almost 20% rating his performance "very poor"), but a net +2.9 is still not especially bad.  Recently Whiteley proposed drug-testing long-term welfare recipients and it's possible this culture-war sop to the irate talkback right may have gone down like a lead balloon with the broader community.  However there are probably other explanations.

Senatorial Obscurity (Saturday)

The Mercury on Saturday has run a two-pager on Tasmania's senators and their obscure nature.  At this stage it's in the print edition only (edit: up online now).  The poll asked voters how many Senators they thought they could name, though it didn't put their ability to do so to the test.  20% said they could name no Senators, 34.6% "Three and under", 30.5% "Between three and six", 12.5% "Between six and eleven", 2.4% "All twelve".  It's not clear how those who thought they could name exactly three or six would have voted, but it really doesn't matter to the point.

Female voters and young voters were least confident of their Senator-spotting abilities (though it would be interesting to see whether females really performed worse than males when put to the test) and there was not a vast amount of variation by voting intention (except undecided).

11 comments:

  1. I was actually polled in this particular poll, if you have any queries about the questions. One thing to note is that there was respondent-allocated preference questions asked, it was simply a first preference question.

    ReplyDelete
  2. Thanks Jamie, I have the text and results of all the questions; I'm just not allowed to release most of them yet. Just to check on the preference bit (and I'm happy not to publish your answer if you don't want any info about your party choice revealed!) am I right that you gave a first preference that wasn't Labor, Liberal or Wilkie and then weren't asked for a further preference? If they were using respondent preferences then a Labor or Liberal voter wouldn't have been asked that question.

    ReplyDelete
  3. ReachTEL have advised me that preferences were all respondent-allocated. That is, voters picking Green, JLN or Ind/Other would have been asked for a preference between Labor and Liberal (or for Denison, Wilkie and Liberal).

    ReplyDelete
  4. > There are some risks involved in using respondent-allocated preferences in small
    > samples. A sample of 600 has a non-major party vote of about 100-150 voters per
    > electorate. These subsamples then have a margin of error of 8-10 points on a
    > good day, which means that the use of respondent preferences can easily cause up
    > to two points of 2PP error by itself. This then increases the effective error
    > margin on the 2PP estimate.

    It seems to me this should not be the case so long as the voter is in full control of their own preferences. That is, the MoE that applies to the question of which major party they would rank higher is no different to the MoE that applies to the question of which party they'd rank first. Of course if they're not in full control of their own preferences, then it's scarcely worth asking the question.

    > The Labor brand can't be too trashed in Denison given that the party just won a
    > Legislative Council seat covering almost half of it with a 10% swing!

    Given that there are 15 LegCo seats to 5 divisions, it's surely like a third.

    Finally, I wonder if Lambie has missed a trick by not standing lower house candidates.

    ReplyDelete
  5. Elwick is 35.3% of Denison by enrolment so I've corrected that bit, ta. The overlap between Denison and Nelson is greater than I remembered it to be.

    The MoE on a respondent-directed 2PP is actually higher than the MoE for a primary vote for a given major party, generally. The less roundabout reason for this is that MoE is a function not only of sample size but also of percentage - the MoE on a result of 50% is higher than that for a result of 35%, for instance. The publication of single "margins of error" for a polling sample is misleading; different results in a poll have different MoEs.

    ReplyDelete
  6. I wouldn't think that Denison would end up being a Wilkie-Lib 2CP count. I understand that Wilkie would take the majority of his voters from Labor, although i'd suggest he'd pull votes from all 3 of the established parties.
    In the end I think their is enough "naturally" Labor voters in this electorate to pull them into second place.

    ReplyDelete
    Replies
    1. Another thing here is that Denison was Wilkie vs Labor last time (Labor beat Liberal in the cutup by five points). What possible reason could there be for the Liberals to gain five points on Labor compared to 2013? I can't see one so there is probably some sample error.

      Delete
  7. G'day Kevin

    Thanks for the analysis

    I'd imagine, hope, that a number of Labor voters could take advantage of the voting changes in the Senate to elevate Lisa Singh above her spot on the ticket. She may even get some Green support. Thoughts on how that may play out?

    ReplyDelete
    Replies
    1. I'm sceptical about Green preferences being much use to Singh because that requires the Greens to get substantially over two quotas (15.4%) and I don't think they will. If the Greens do get their two quotas more or less on primaries then that will free up left-wing micro votes to perhaps flow to her. It's tough though. She needs to whip up something like 3% in her own right, and have Labor poll well enough that the fifth seat is there for her, but not so well that she can't overtake the fifth candidate. I don't think she has that much of a following outside Denison and even there she wasn't able to save her state seat, so it won't be easy. I definitely don't dismiss it as a possibility though. It's the kind of thing where if it would work anywhere it would be here.

      It would be rather amusing if both Colbeck and Singh got decent BTL votes and the final seats turned into a Hare-Clark style race between candidates rather than parties. Not sure though if voters will be alert enough to the new system the first time around.

      Delete
  8. I am happy to say that my response was Wilkie, so that would explain why i didn't get pushed on further preferences.

    It should be noted that this poll was based on a person being able to press what number they wanted whenever they wanted (i.e. not hearing all the options first). This gives a slight advantage to someone at the top of the readout and slight disadvantage to who was last on the list. Probably doesn't affect the "house" candidate selection, but may have a bearing on what is the most important issue to you.

    I must admit that i was confused with the Lambie being read out and assumed that she might be running a candidate in the house. In my case it wouldn't have mattered, but generally speaking I think people are more likely to pick someone in the top 2 or 3 for the house, but the senate people are more likely to go with their first pick. Would have been nice in this instances if people had been asked about the senate and house separately, if that was the information they were after.

    ReplyDelete
  9. I think it's quite likely that Lisa Singh will poll the best out of Labor below the line and may get 3% as there are a lot of people that were unhappy with the way the preselection went. However with optional above the line voting, i think this may encourage more voting above the line. This leads to party preselection being a bit more powerful than before.

    That said, Tasmanian's are surprisingly good at voting below the line, so anything could happen and generally speaking below the line voters are probably more politically aware and hence may be more inclined to put Lisa Singh first in their labor preferences (If i don't know the candidates i tend to vote bottom up when it comes to BTL voting, just because i like the concept of messing with party pre-selection, however this tends to happen only for people past about preference 20).

    ReplyDelete