Thursday, May 17, 2018

EMRS: No Real Change Since Election

EMRS: Liberal 47 (-3.2 from election), Labor 30 (-2.8), Green 14 (+3.7) Ind/Other 8 (+1.2)
Interpretation: Liberal 49.5 (-0.7) Labor 31.5 (-1.3) Green 11 (+0.7) Ind/Other 8
Outcome if election "held now" based on this poll: Liberal majority government (c. 13-9-3)
However it is unlikely in practice Greens would be in a position to regain Bass so quickly.
Poll provides no evidence that any party's support has changed.

The December 2017 EMRS poll, taken three months out from the 2018 state election, proved to be completely unpredictive of the outcome.  It had a 17% swing against the Government (the actual swing in the end was 1%), a 3.2% swing to the Greens (the actual swing was 3.5% against) and an 8% vote for the Jacqui Lambie Network (who in the end got 3.2).  If a poll taken three months out is predictively worse than useless, what can we say of one taken two months into an expected four-year term?

Predictively, nothing at all.  However, the May 2018 poll still does provide some data we can consider.  Just not as much as it used to do, because EMRS have changed the report format.  The new presentation provides pretty graphs in the PDF but no longer includes the old "Table 2" (a breakdown including "undecided" voters), and the number of undecided voters isn't stated in the new report anywhere.  Other Australian pollsters don't generally provide tables including "undecided" either (except for ReachTEL who define the term differently to everyone else) but the percentage of undecided respondents should be published.

What did we learn about EMRS polling at this state election?  The previous four elections EMRS polled at all featured Labor governments and this one featured a Liberal government.  EMRS had been especially prone to underestimate Labor support in the past, but in this case underestimated Liberal support even in its final poll (by 4 points, outside the poll's claimed "margin of error", but this could have been down to voting intention changes over the last week).

What we've seen suggests that if there was a specifically anti-Labor house effect in EMRS polling, it isn't there anymore, but it could be that EMRS polling underestimates the government of the day (whoever it is).  One thing that did follow the script was that EMRS overestimated the Green vote (this time by 1.7 points), as it did in 2006 (by 3.4 points), 2010 (by 6.4 points after redistributing undecided), 2014 (by 3.2 points, albeit a month out).  The pattern here seems to be that the poll is prone to exceed the Green vote by more when the Green vote is high and less when the Green vote is low.  My approach to modelling the Green vote in EMRS this term will be to reduce it by around 20%, and to split the resulting slops in favour of the government.  As concerns Independent/Other, EMRS were accurate at this election, so I will only be adjusting the Ind/Other vote if it starts becoming inexplicably high.  (That could include if it is still this high at the next election but no significant minor parties are running.)

What can we learn from this poll?  Predictively nothing of course, and the changes from the election result are trivial and probably caused by a combination of random error and the poll's pattern of high Green results.  State voting intention having not changed in two months is consistent with what we saw in Prosser, where a seat that would have returned a fairly narrow Liberal two-party win based on state results ended up doing exactly that.  What we don't see in this poll is any sign of the government bouncing to above its election result.  There wasn't such a bounce after the last three elections either, even after taking into account the poll's possible house effect against the Labor governments.  Such bounces are quite often seen in other states and federally, but perhaps they don't happen here because winning a majority in Tasmania frequently requires a party to get just about every vote it can, while minority wins are never popular.

We also have a nothing-to-see result on leaderships, with Will Hodgman leading Rebecca White 47-41 as Preferred Premier, compared to 48-41 pre-election.  This suggests that at this stage Labor's defeat has not damaged White's leadership as losses sometimes do.  Indeed that is quite a good result in the context of the voting intentions

Lastly The least important bit - if an election was held today ... There's hardly any point in talking about this now but my adjusted poll results if accurate would probably result in a 13-9-3 split with the Greens recapturing Bass from Labor.  Bass was extremely close - had 138 Labor voters (0.2% of the vote in the seat) instead voted Liberal, the Greens would have held it, and had twice as many voted Green the Greens would also have retained.  That should give rise to some reflection on why the Greens' ground game in the seat was, from what I've heard, so feeble.  For that reason, I don't take the idea that they would really be in a position to win it back right now seriously, and in any case if we did have an election "now", the fact that we were having an election now would distort the outcome massively compared to any polling.

I intend to restart my polling aggregate when we have at least one more poll and I have a lot more time.  I've been working on a federal Budget polling roundup but it is still some way from finished.

Edit: Oh No I Forgot To Mention Braddon I've been reported as saying an apparent drop in Labor support in this poll could spill over into the expected Braddon by-election, but I actually wasn't that specific (at least not intentionally) - I was referring to the strong state Liberal support shown at the election and at this poll.  I've been doing some work on relationships between state and federal voting, along the lines of a party that does well at one level in a state tending to also do well in that state at the other level at the next opportunity.  (That might sound strange given "federal drag", but there are sometimes issues or organisational factors that prop a party up in a given state at both levels at once.)  There seems to be something in it on average, but it's a weak relationship, with a heck of a lot of outliers, especially in Tasmania.  I don't take this poll as altering my previous assessment that the seat should be retained.

Now With Bonus ReachTEL!

A ReachTEL sample of 645 taken for Residents Opposed To The Cable Car, confined to the electorate of Denison (soon to be Clark) finds the following figures after redistributing undecided voters:

Labor 38.6% (-3.3) Liberal 35.9 (-1.8) Green 18.2 (+0.7) Shooters Fishers and Farmers 1.8 (-) Other 5.6 (+4.5).  The increase in Other is wishful thinking - others polled poorly at the election because Tasmanians4Tasmania was the only choice available. ReachTEL was impressively accurate at the state election but also has a history of overestimating the Greens at Tasmanian elections.  Prior to this election it also had a history of underestimating Labor, but this time (as with EMRS) didn't do so.

Comments about the cable car poll will be updated on the running article about cable car polling.


  1. If I recall correctly, in the lead up to the state election there was the highly unpredictive EMRS poll in December, and then no more publicly available polling until a week or two prior to the election (at which time there was a collective "holy shit" realisation that the Liberals were doing much better than anyone had anticipated). Is such an enormous gap in publicly available polling common in the lead up to an election? It seems like an important factor in the outcome. Had their political opponents known that the Liberals were making such tremendous gains, perhaps campaign priorities could have been adjusted accordingly.

    1. There was plenty of internal Liberal polling being released to the media in Jan-Feb, which may have been a strategy to control the polling narrative and discourage the media from commissioning their own polls. Media are suckers for free polling, no matter how dodgy it is.

      In terms of public polling, what happened is unusual, and unprecedented in my memory. For instance the 2006, 2010 and 2014 elections were later in March meaning that there was a February EMRS (as well as a March one in the first two cases). In 2014 there was also a ReachTEL in February.

      Labor were conducting their own internal polling and declining to release most of the results so I think they knew what was going on. However it's been reported that they had spent a lot of money late last year, for some reason expecting an earlier election. And by this stage they were locked into the anti-pokies policy anyway so I'm not sure if they could have done anything.