image

ReachTEL (Liberal-commissioned – beware!) Lib 51.5% Labor 22.7% Green 17.7% Other 8.1%
Outcome if reliable: Liberal majority government (probably 13-15 seats)

Welcome to ReachTEL, a new player in the lonely Tasmanian state opinion poll market! It would be nice to extend a less equivocal welcome, if only out of sheer relief at writing about a poll that isn’t marred by the endless problems with EMRS, but alas, things aren’t that simple. At least, not when the Mercury article reporting it starts with stuff like this:

The Tasmanian Labor Party faces a rout at the next election, with new polling showing the Liberals are assured of forming a majority government.

If this piece appearing under the name Matt Smith is to be believed, all you need to do to be assured of government in this state is to release an opinion poll, which you commissioned yourself, saying that you are winning bigtime, a year and a half from the election, and hey presto, you’ve got it in the bank. May as well save the money, replace the Tasmanian Electoral Office with the Smith Electoral Office and send for the Governor now.

In fact, even if this poll release did not require caution because of it being commissioned by a party, its findings would still not prove the Liberals were certain to win the next election. They would just say that at the moment their position appears very strong.

Daft mainstream coverage aside, the poll’s sponsorship is the main obstacle to taking it too seriously. The polling is essentially internal polling, in that the party had no prior commitment to release it, but has chosen to do so after seeing (and liking) the results. We cannot know how many other polls the Liberals might have done through any number of pollsters, and whether this is consistent with other results that they have not released, or just the best of the bunch. We may also suspect, at least until proven otherwise, that the timing of the poll – reported as conducted on a single night – could have aimed to capitalise on a bad news cycle for the state government, making it less reliable than the results of polls taken at more regular intervals. Despite this, the Liberals are to be commended for at least releasing much more detail of their internal polling than is usual for releases of this kind.

Parties seldom release polling they have commissioned themselves – accurate or otherwise – without a motive for doing so. In the Liberals’ case, there has been a drift away from them in EMRS polls that has caused people to start to mutter (very prematurely) about the possibility of a hung parliament, and even about Will Hodgman’s leadership. The Liberals would rather have the focus on attacking the government and would hence prefer that they be seen to be going well. This poll is consistent with that purpose. (In other cases, parties may even release polls that show them to be doing badly, in order to manage expectations coming up to a nasty by-election, for example.)

You would not know anything about the EMRS poll drift over the last year or so from reading the Mercury coverage, which contains the following error:

The Liberals’ polling is, in part, consistent with EMRS polling in August that showed the Liberals had 55 per cent support, Labor 22 per cent and the Greens, 18.

Actually, the August in which such results were recorded was 2011 (around the Liberals’ EMRS peak), not 2012. The comparable result for August 2012 is 49-27-22-3, meaning that the ReachTEL sample has a Liberal result that is 2.5 points higher, Labor and Green each 4.3 points lower, and the rest 5.1 points higher.

I’ve pointed out routinely that EMRS overstates the Green vote by using unsound assumptions when redistributing its high undecided rate. In this case, that alone explains the difference between the two Green figures, and ReachTEL’s figure is much more likely to be correct. However, EMRS also tends to understate the Labor vote for the same reason. So, in a poll with a much higher Other response than EMRS’s, ReachTEL has the Liberals up and Labor probably, in effect, a long way down on the EMRS August sample.

There are major differences between EMRS and ReachTEL in their polling methods. The polling method used by ReachTEL is not stated in the story, but ReachTEL typically uses recorded voice polling with interactive options (so-called “robopolling”), a method noted for generating low response rates (requiring a lot of scaling) but being relatively cheap to run. While the accuracy of robopolling is often disputed, my impressions of the efforts by JWS and ReachTEL in Australia so far have been that they are not bad at all, and that a neutrally-commissioned robopoll by these companies is many times more useful than the average EMRS. A good example of ReachTEL’s recent performance was the Melbourne state by-election, in which it got the Green vote correct to within 1.6 points and the Labor vote correct to within 3.2 points, off a sample of only 403 voters and for a kind of electoral event that is extremely challenging to poll for. ReachTEL has released recent polling of the Denison federal contest, which can be found here.

ReachTEL asks its questions in a superior manner to EMRS. The undecided rate is kept down by asking, as part of a single question, “if undecided, to which do you have even a slight leaning?” Indeed, I am not sure if there is even a possibility for a voter to record themselves as undecided, short of hanging up the phone. Any lack of an undecided option, however, may manifest itself in unsure voters picking the option “other”.

The “other” vote in this poll is quite curious. Statewide it is 8.1%, but with Denison way out in front with 21.9%, Lyons 8.4, Franklin 6.7, Bass 6.1, Braddon 5.3. (These figures add up to almost 10%, so I assume the samples are “raw” and some kind of scaling is employed to get the state result.) This resembles a pattern seen in EMRS polling except that the figures for “other” are much higher than for EMRS’s “independent”, even at times when the EMRS rate is boosted by Legislative Council election contamination. It is very doubtful that these “other” voters are all just some of those tagged as “undecided” by EMRS by another name, since if this was the case Denison would have shown much higher undecided rates than other electorates in past polling. Although EMRS removed their electorate sample sizes from their reports after May 2010, making this difficult to confirm, it did not appear up to that point that that was happening. So my take on this is that there is a statewide background ReachTEL “others” rate of about 5.5%, varying little by electorate, and a Denison-specific rate of about 13.5% that is connected to the Wilkie phenomenon. (See note above on possible scaling to perhaps explain why these add up to 19 for Denison and not 21.7.)

A Denison voter asked how they will vote in the 2014 state election, after all, cannot know at this stage whether or not Andrew Wilkie will be a candidate. Furthermore a voter who voted for Wilkie, if asked to choose between Labor, Liberal, Green and Other, is quite likely to select Other, even if they do not know that any particular Others might be running. However, this apparent ReachTEL figure is higher than rates for “Independent” that EMRS have been getting in Denison. There are many possible explanations for this, but I’m more inclined to believe the implied ReachTEL result, given the problems with EMRS understating votes for non-party candidates in the past. So it seems like there are almost a quota of voters in Denison who are waiting to either give Wilkie a consolation prize should he lose his federal seat, or else vote for a high-profile Wilkie-like independent, should one be found.

It’s hard to say whether the remainder of the “others” voters are mostly glorified don’t-knows or whether there is also an undercurrent of “none of the above” sentiment in all electorates that is higher than shown by EMRS. In any case, in electorates other than Denison there is probably not enough of it yet to elect anybody, so at this stage it only matters if these voters will lean strongly to a particular party come election time, and even then, it may not matter much.

Given that the ReachTEL Liberal figure is not far above the recent EMRS Liberal figures, it could be argued that there is really little difference between the polls at all, except that the softest Labor voters are hiding in the ReachTEL “others” column. I doubt this, because if it is true then those who are “undecided” according to EMRS (even when asked to express a leaning), but who do have a view according to ReachTEL (let’s call these voters “extra-soft”) skew strongly Liberal. To have the mildly soft voters breaking evenly, the extra-soft voters skewing Liberal and the ultra-soft voters skewing Labor would be quite odd.

So my interpretation of the ReachTEL poll is that it does show a genuine difference to the August EMRS poll, of a level equivalent to a five (or so) point Labor-to-Liberal swing. But as the Liberals are both the commissioners of the polling and its political beneficiaries, we can’t take for granted that such a swing has actually happened (on any lasting basis) since then.

The ReachTEL electorate samples (published in the paper version) need to be treated with extreme care, not only as the poll is Liberal-commissioned but also because they are quite small (230-241 votes each), and they do not add up to the state figure. All the same, even making allowances for these things, they are consistent with the picture shown in every EMRS poll for some time, that the Liberals are on track for majority government. If the figures in ReachTEL’s electorate samples were reproduced exactly at an election – unlikely as that is – Lyons, Bass and Braddon would probably all go 3-1-1 (Bass is almost 4-1-0 but the Liberals would probably miss the fourth because of leakage). Franklin could go either 3-1-1 or 2-2-1 (Liberals would suffer from leakage from the massive Hodgman surplus, and Green preferences might help Labor a bit more than in 2010) and Denison would go 2-1-1-1 if there was a Wilkie-style independent available (if not, the fifth seat could go anywhere.)

What is surprising is that even in this very favourable sample, which the Liberals liked so much that they have released it publicly, the electorate breakdowns would not assure the Opposition of winning more than 13 seats if reproduced at an election. And that is even with the breakdowns taken as if perfect, without even discussing the large margin of error in such small samples or the vagueness introduced by the high “others” rate.

Lastly, March 2014 is a long time away, and should the state coalition go the distance then the current double-whammy of Green-supported Labor governments at both state and federal level may well be a thing of the past. The broad left should not, however, get more than the faintest flicker of hope up on any account. A Liberal outright victory is not “assured”, least of all by polling paid for by themselves, but it is still overwhelmingly likely.

First published: 2012-10-08 04:00 AM