ReachTEL - State Labor Back To The Lifeboats

ReachTEL (Tas State): Lib 52.5 ALP 19.7 Green 15.6 Other 6 Undecided 6.3
Interpretation: Lib 57 ALP 22 Green 17 Other 4 (Other perhaps higher)
Outcome based on this poll "if election was held now": Liberal Victory (c. 15 seats)
Projection based on all polling: 14-8-3

We have reached a useful milestone in Tasmanian polling since, for the first time since the last state election, EMRS has competition from another poll that is commissioned by the media and that is not merely an internal poll released by the party.  Historically it has been a long time since we have seen public polling by someone other than EMRS, apart from during an election campaign.  Indeed, the Mercury-commissioned ReachTEL is to my knowledge the largest poll of state voting intention ever conducted in Tasmania.

These are the headline figures, which I'll discuss after a small but important diversion:

Diversion: The Dog Ate My Undecided Rate

What I would really like people to notice there is not the yawning disparity between the top two figures but rather the Undecided figure of 6.3%.  This is for a way of asking the question for which the equivalent undecided rate in the last EMRS poll was 19%.  It is still further evidence (not that any is actually necessary as the point has already been proven using old Morgan polling) that the high Undecided rates recorded by EMRS are not an innate feature of Tasmanian politics, and indeed are just not real!  Every time a media source draws attention to there being supposedly 25% or 30% of Tasmanian voters "undecided" on the basis of an EMRS poll, what they are saying is nonsense.   The high undecided rates that occur in EMRS polls are not because voters are genuinely so unsure of their vote, but because EMRS are not as effective in getting a response out of voters as other polling methods.  Indeed, their interviewers are three times more likely to fail than an automated recording.  Scary!

The Overall State Picture

This poll is yet another state poll that shows the Liberals headed for majority government and leading by a ridiculously large margin - the result recorded by every poll whether public or private that I have seen since early 2011.  Indeed, the Labor vote following simple redistribution of the undecided, 21%, is lower (by one point) than any poll in the EMRS series, while the Liberal vote (56%) is one point higher than any EMRS result.  This could indicate that this ReachTEL is an unnaturally extreme result, but it doesn't have to.  Another possibility is that the assumptions that are used in distributing the high EMRS undecided rate are actually a bit generous to Labor and that the EMRS-undecided voters are actually more than proportionally Liberal-leaning.

By comparison with the Federal poll, Labor is down 6.6 points, and the Liberals are up 7.  The Greens are up 5.1 points, and Others (mainly because of the absence of Andrew Wilkie from the mix) are down 5. These differences are practically identical to the gaps between the parties' state and federal performances in 2010 (Labor -7.2. Liberals +5.3, Greens -4.8).

The February and May EMRS polls offered a sharp contrast in their readings of the fates of the two coalition parties.  The February EMRS poll suggested that if an election was held then, Labor would have copped the brunt of the damage while the Greens would have escaped with the loss of one seat.  The May EMRS poll suggested Labor could retain most or even potentially all of its seats, but would still lose because the Liberals would win three seats from the Greens.  Now this ReachTEL poll looks like the February scenario all over again (and then some). It is notable that the last EMRS was taken almost immediately after the passage of the forestry "peace deal", and it's possible that it picked up a backlash against the Greens that has since faded.  Or it could just have been a nasty sample for them.

Anyway this again raises the prospect of Labor incumbent pairs like Michelle O'Byrne and Brian Wightman, Brenton Best and Bryan Green and even Lara Giddings and David O'Byrne, scrambling for the lifeboats and fighting for single seats, a scenario which might well increase interest in the on-again-off-again concept of increasing the size of the House of Assembly.  Some see Brenton Best's recent behaviour in exactly this sort of light - he thinks the party won't win two in Braddon, and he wants to be the sole survivor.

On to the very useful crosstab for ratings of the Labor/Green government's performance:

The government has a net satisfaction rating of -44.6 and a frightfully low 3.9% of respondents think it is performing very well.  This table is handy because it gives us a fix on the leanings of Other and Undecided voters.  Unlike the Others voters in the federal poll, who were fairly evenly split in their views of the federal government's performance, these Others voters rate the government's performance poorly, as do the Undecided voters.  The difference is that many Andrew Wilkie voters would be supporting the Greens at state level.  In the absence of significant 4th-party candidates (meaning that the Others vote is probably a bit too high), or assuming they are excluded, this poll points to the Liberals picking up still more support.  Hence my interpretation figure of 57%.  Again, this is an extreme result, but only a touch more so than a number from EMRS.

The Seat Breakdowns

The size of the seat breakdowns (just over 500 votes each) is such that they are actually useful.  As they are the first seat data to come out since a Liberal-commissioned ReachTEL last October, and given the sample sizes, the value of earlier seat data is much reduced.

The following is the breakdown of the poll with Undecided redistributed proportionally:

A few comments at this point:

1. When you have 20 individual readings by party and electorate off a sample size of 500, some of them are going to be significantly wrong by chance. You would expect on average to get one result that would normally be considered "rogue". One that just doesn't seem believable here is that Labor would be below a quota in Bass.  The Liberal score in Franklin also seems a little difficult to credit.

2. In comparison to the Liberal-released ReachTEL last October, this one has a modest Others rate for Denison.  The asking of a federal voting intention question first has probably primed respondents with the information that Wilkie is a federal member and prevented them from getting confused and trying to express support for him in the state poll.

3. The Others vote is highest in Braddon (and generally pretty high across the board, despite the note above.)  There doesn't seem to be a lot of logic to this unless there are a lot of Brenton Best fans who believe he will run as an independent.

4. There's no real suggestion that any Others would win a seat.  In Denison, if a single candidate could tap just a little bit more than the 7.1% Others vote, they would have a good chance.  But there's no evidence yet of such a candidate running.

In past polls I have produced a three-party-preferred breakdown and projection with Others votes distributed equally between the three parties.  That's a rather dubious modelling practice in this case because it appears the Others voters are Lib-leaning, but I think the Libs are having quite enough fun here as it is so let's give it a whirl (click for larger version):

If this poll is correct then Labor loses seats everywhere except perhaps Denison, while the Greens lose only Braddon (and that narrowly) and possibly pick up two in Denison.  A very different picture to the most recent EMRS.

Now that we have two polls from different companies, the natural thing to do is aggregate them!

ReachTEL/EMRS May-June Three Party Aggregate


I've decided to aggregate the ReachTEL in the three-party form above with the uneven swing model for the last EMRS (which represented my reading of the last EMRS allowing for the pollster's known house effects and for evidence on regional variation culled from two EMRS and one ReachTEL polls during 2012.)

In so doing I am weighting the ReachTEL at double in view of its much larger sample size, freshness, not having been taken very soon after a major state event, and in my view better polling design.  (Although ReachTEL robopolls and weights, I'll take that over anything that has an undecided rate of 19% even after prodding the respondent, any day).  There is a case for weighting the ReachTEL much more than double, but I always prefer to be cautious rather than to have an aggregate too captive to a single pollster or method just because one pollster has a large sample size.

Here's the aggregate (click for larger version):

I think this has knocked a few of the rough edges off the results in both of the polls.  The Liberal vote in Denison is still a bit too high (according to what everyone tells me!) but it is what the public data say, and it makes little difference anyway. 

The projection continues to show the Liberals easily winning three in Bass, Braddon and Lyons with the possibility of a third in Franklin.  Also they are not far shy of four in Braddon, though in this aggregate it's more likely Labor would retain.  (Some may wonder why I give the Greens Franklin when they are short of quota and the others are not; the answer is that the others would be much more exposed to leakage.)  The projection allows for results ranging from 13-9-3 to 15-7-3, although Tim Morris in this projection is not that far shy of holding his seat.  While Labor would have slightly better chances than the Liberals in each of the individual close seat contests, I've assumed one each to be the most likely outcome, and on that basis left my state projection at 14-8-3.

And for the record, with 35 seats, the ReachTEL alone comes out at 21-8-6 (five Libs in both Bass and Braddon), the last EMRS would have been about 18-13-4, and the aggregate returns either 20-10-5 or 21-9-5.  Switching to 35 seats, it seems, would save the existing furniture ... and give all the new stuff to the Liberals.