Saturday, November 2, 2013

ReachTEL: Giddings' Northern Problem

ReachTEL (State): Lib 49.2 ALP 23.2 Grn 14.8 Other 6.2 Undecided 6.6
Interpretation: Lib 48.2 ALP 29.4 Green 15.9 Other 6.6
Outcome based on this poll: Liberal Majority Win (13-8-4 or 14-7-4)
Caution: PUP not named in poll; PUP support level unclear.

New aggregate: 13 Liberal 8 Labor 4 Green

Another Mercury/ReachTEL poll is out (see David Killick's report here), with the Saturday release canvassing voting intention and preferred premier scores.  There will also be some questions rating the state government's performance on various issues.  You probably don't need me to tell you what the results of those look like. 

The poll comes as the Labor-Green government, which has to go to an election by June next year, is plagued by rumours of infighting, although these result mainly from the actions of renegade backbench lifer Brenton Best.  It might have been thought this situation would send the Labor vote to new lows, but in fact it has made very little difference.  Compared to the last ReachTEL in August the Liberals are down 1.9 points, Labor down 1.2, the Greens steady, and the winners are Other (+2) and undecided. The impression is that the latest internal tensions are not driving Labor's vote much lower because it was probably scraping along the bottom anyway.

This is the breakdown of the raw results of the poll (click to expand):

Again, a large sample size.  As noted after the federal election, there is an impression that ReachTEL is getting the Liberal vote in the state about right in its baseline level, but underestimating the Labor vote by several points.  At least, that is what the federal election results suggested.  Thus, it may be that both Tasmania's two regular pollsters are overstating the margin between the two major parties.  For years I've been applying corrections to EMRS polls on this basis (my analysis of the September EMRS is here) and in this article I'm assuming that the same thing holds for ReachTEL.

Thus in the following table I present the primary votes of the poll with Undecided redistributed, and then two different result projections.  The first is the outcome of an election "held now" assuming the poll is completely accurate and that the undecided voters split in proportion to the primary votes.  The second is the outcome if it is assumed that the poll is out by similar amounts to the federal ReachTEL polling prior to the 2013 federal election.  For the latter purpose I've moved four points from the Liberals to Labor.

The Liberals win an outright majority in either case as they are clear of three quotas in each of Bass, Braddon and Lyons.  However, the modified result makes a difference in two seats.  Firstly in Braddon, on the results as polled, it is not clear which of Labor, the Liberals or the Greens would win the final seat.  However, with the Labor vote increased as per the modified results, Labor probably retains two seats.  Secondly, on the results as polled, the Liberals easily win three in Franklin, eliminating either David O'Byrne or Lara Giddings.  But on the modified results, Labor becomes competitive in Franklin (and indeed, on those figures would more likely hold, thus retaining eight seats.)  So while the poll appears to show the Liberals with 14 seats and a possible 15th, I believe 13-14 seats is a more accurate reading.  Even considering the modified results, a lot has to shift for the Liberals to only get two in Lyons, especially considering that Labor will only be fielding a single incumbent there.

The poll is a little better for the Greens than some recent offerings.  Especially, their vote in Lyons is higher in this poll than has usually been the case lately.

The big "what if?" about any of this polling at the moment is the difficulty of assessing the PUP vote.  Unfortunately PUP are only canvassed under generic "other", and this certainly underestimates what they would poll if running a campaign as well resourced as their last.  That said, nowhere near all the polled "others" vote is PUP.   We know from the state Senate results that Braddon is PUP's strongest seat and Denison their weakest.  On this basis it's likely that the Others vote in Denison is mostly non-PUP (for instance, Andrew Wilkie voters hoping for a Wilkie clone to vote for) but possible that the Others vote in other electorates has a large PUP component.  Even assuming the Others vote in all electorates other than Denison is all PUP, the polled vote wouldn't put them near winning anywhere - but their real vote might be much higher.  I don't think this poll provides any evidence that PUP are on track to win any specific seats, but they can't be ruled out either.

I should clarify a quote in the Mercury piece.  I'm quoted as saying "That other figure probably does under-poll PUP in Braddon, for instance. It's plausible they might get 10 per cent, but they won't win a seat".  Whether or not I said absolutely exactly that, the intention was to flag a possibility of a 10 per cent PUP vote in Braddon still not necessarily translating to a seat.  It wasn't to call in advance that PUP won't win in Braddon.  I think they do have some chance there, but that more evidence is needed, including polls that specify PUP as an option.  Even these could underestimate the true level of PUP support until close to an election.

What these figures do show is that if PUP do win, it will most likely be at the expense of the second Labor seat, not the Liberals' third.  The Liberals have a lot to spare in this electorate.  I missed an opportunity to do a radio interview about the whole PUP-causing-minority-government thing, but in my view at this stage it's an overrated prospect.

Unusual Distribution Of Leader Support

I think this is the first state ReachTEL to canvass a three-way preferred Premier question.  EMRS asks this question routinely, and it usually shows ratings that are quite similar to the party ratings.  There was some interest in the September EMRS in that Lara Giddings' preferred premier score suffered a seven point drop to 18%.  I argued that the drop was real.

ReachTEL employs an unusual forcing method in Better Premier polling - the respondent has no "undecided" option but must pick one of the three options to continue the survey.  Federal experience is that this method more or less eliminates the massive advantage that incumbent Prime Ministers have over their opponents (all else being equal).  The suggestion is that allowing a don't-know option results in voters who mildly prefer the Opposition Leader hiding their intentions.  Given that EMRS allows a don't-know option and their results roughly track the party scores, it might be expected that the ReachTEL method would show the Opposition Leader, Will Hodgman, ahead of his party's standing.  And that's exactly what we get; nearly 60% (a staggeringly high figure for an Opposition Leader by any measure) of the survey thought that Will Hodgman would be a better premier than Lara Giddings or Nick McKim.

To some extent these figures track the parties' fortunes in each electorate, but that's not the full story.  In the raw primary figures, 12.8% of respondents picked Other or Undecided, but in this question all those respondents expressed a preference.  All else being equal and assuming the Others and Undecided voters had similar preferences to the rest of the voters, it would be expected that all three leaders would poll better premier ratings somewhat larger than their party's support score.  Statewide this is indeed the case but it is more so for Hodgman than expected, and less so for Giddings and McKim.

The following table gives each leader a Better Premier rating in each electorate that is a net rating with their party's support deducted:

What is significant here is the comparison between Bass and Braddon on the one hand and Denison on the other, especially as concerns Lara Giddings' ratings.  In Bass and Braddon, Will Hodgman's support level is way ahead of his party's, but Lara Giddings' is below her party's.  There are many Labor voters in these seats who think Will Hodgman would make a better Premier than Giddings.
In Denison, however, Giddings is actually way ahead of her party's rating, and the leader who has the problem relative to their party is McKim.

I suggest that what is going on here is that progressive Denison voters including Greens voters are impressed by Giddings, and that she therefore rates well even among left-independent types who no longer prefer Labor or the Greens.  But the more conservative Labor voter in Bass and Braddon has a very different reaction to what they see as a radical left Labor-Green administration.  Even those conventional Labor voters in these seats who have not already switched parties are not, in general, happy with this Government or its leader.  In particular, Giddings' willingness to enter into another minority government with Green support (given the chance) is poison in these electorates.

New aggregate

I've been a bit slack on aggregating state results lately.  The following aggregate uses these inputs:

1. My interpretation of the current ReachTEL, scaled at 40% of total,
2. My interpretation of the current EMRS, scaled at 30%
3. The federal-election based ReachTEL model, scaled at 15%
4. The old aggregate, with 3 points moved from Liberal to Labor (based on underestimation of ReachTEL house effects when it was compiled), with a notional 7% for Others, and scaled at 15%

I've also slightly scaled down the Green vote in those components of 1 and 4 that included ReachTEL data based on the findings of 3.  I could have also done this for the present poll but no projected outcome would change, though Booth and Morris are both quite shaky for the Greens.

And here it is:

Only Franklin and Lyons require any comment here.  In Franklin on the current aggregate figures the Liberals would lose votes from leakage from Will Hodgman's surplus and the high profiles of Giddings and O'Byrne would see both retain.  Nonetheless a 3-seat win for the Liberals is well within the aggregate's likely margin of error.  In Lyons there is a fight between Labor and the Greens for the final seat.  On these figures Tim Morris would probably very narrowly retain because of his high profile and the likelihood of a low-profile Labor team with only one incumbent leaking votes.

My new state aggregate is therefore 13-8-4, though it is very close to being 13-9-3.  The Liberals are clearly winning but still have work to do to snare the 14th seat and win by more than one.

I should note that not only are the Greens only just over the line in a couple of seats on these figures, but the position of Greens leader Nick McKim is not entirely safe.  The aggregate has him at an even quota, but a small swing away from that might put him far enough behind the other parties to be at risk.  Denison is the only seat where the Greens appear safe on current polling.

Issue rating questions (Nov 3):

Issue rating questions were released today covering the government's performance on the issues of economic development, job creation, education, health, law and order, and the environment.

The government is shown as having a net negative rating on every question in every electorate.  However given the apparent anti-Labor lean of the poll, it is very likely that these ratings are exaggerated by several points and that the government does actually have a mildly positive image on a single issue (law and order) in a single electorate (Denison).

I've prepared a table showing an average of the government's net ratings on each issue in each electorate and overall:

The best (or least worst) rating on each issue is shown in green and the worst in red.  Overall Denison delivers the government's least bad reception on every issue, followed in all cases by Franklin.  There's not much between Bass, Braddon and Lyons, which take turns in delivering the government's worst ratings on specific issues.  The government's ratings are the worst on job creation, health and the economy (three issues that persistently rate highly in issue importance polling) with more benign ratings on the environment and law and order, which are generally down the list.

This picture of Bass, Braddon and Lyons being much the same as each other in terms of the government's reception is the same picture we saw in the leadup to the federal election in both federal and state polling.  There were those who believed Lyons must be different in the federal election because it was on a big margin and Dick Adams had been there so long, but in the end it wasn't really.  Voters in these three electorates dislike the current government very strongly and I think it's very likely all three electorates will deliver Liberal primary votes exceeding or at least very close to 50%, with PUP gains from here coming mainly at the expense of the Greens and Labor.

Some may be surprised that the government rates badly on the environment when it has been the greenest government in Tasmanian history.  There are two likely reasons for this.  Firstly some voters would rate its performance on the environment badly for exactly that reason.  Secondly there would be some voters who so much want the government gone that they would rate its performance badly on any issue put to them.  If the government's ratings on this issue inverted the normal pattern then it would make sense to say voters thought the government had not been green enough, but the pattern on the environment is much the same as any other issue that was canvassed.


TT Comment: (Nov 13) A comment by Basil Fitch (#73 here) has been published on Tasmanian Times which implicitly criticises my recent use of ReachTEL polling.

Firstly, Fitch's reporting of the results of the final Bass Federal ReachTEL is not quite accurate, and his reporting of the Bass election results is wrong too.  Fitch claims that " Alison Andrews(Examiner Sept 5) stated exclusive Reach-Tel poll showed Geoff Lyon [sic] Labor 26% and Andrew Nikolic 54% 1st preference" and then claims the actual results were 38% and 48% (significantly wrong in the first case).   Andrews' article can be seen here and it reported the raw poll results correctly, with 51.8% for Nikolic and 26.6% for Geoff Lyons.  These figures are without redistribution of undecided voters.  While these were indeed well off the correct figures of 47.85% and 34.65%, the actual differences were 3.95% (compare Fitch's "6% out") and 8.05% (compare Fitch's "12% out").  So Fitch has exaggerated the difference between ReachTEL's poll and electoral reality by a factor of 50% just by getting all the numbers wrong.

Secondly, Fitch expresses surprise that I rely on the ReachTEL figures in my comments.  But in fact in analysing this poll I adjusted for what we know of ReachTEL's apparent skew based on federal polling, and that is why I told the Mercury that I assessed this poll as projecting seven or eight ALP seats when the raw figures support only six. 

Fitch claims that EMRS is "pretty reliable", ignoring that the pollster projected a hung parliament in every poll leading up to the 2006 election at which Labor obtained a comfortable majority victory, and has a long history of exaggerating the Green vote.  He also lauds a Morgan poll showing a federal result of 50-50, but in the same week Newspoll and Essential (for what the latter is worth) showed 53-47, and the Morgan in question featured a combination of face-to-face and SMS polling without Morgan's usual online component.  Morgan face-to-face has been shown to skew to Labor so this new combination is untrustworthy. And there's the usual stuff about ReachTEL mainly calling oldies - if you are going to even discuss this point, you should at least inform yourself about scaling and reflect some understanding of how pollsters actually work in your comments.

As for Basil's projection of a Green vote of 24%, it aint happening.  Not this election anyhow!

No comments:

Post a Comment

The comment system is unreliable. If you cannot submit comments you can email me a comment (via email link in profile) - email must be entitled: Comment for publication, followed by the name of the article you wish to comment on. Comments are accepted in full or not at all. Comments will be published under the name the email is sent from unless an alias is clearly requested and stated. If you submit a comment which is not accepted within a few days you can also email me and I will check if it has been received.