Monday, May 20, 2013

EMRS - Greens hit new low

EMRS: Lib 54 (-1) Labor 28 (+5) Green 14 (-4) Ind 4 (+1)
Interpretation: Lib 53 Labor 33 Green 11 Ind 3
Outcome: Comfortable Liberal Majority Win (Approx 14 seats)

The May 2013 EMRS poll has been released and the trend graph for the headline figures is here.  The headline figures have a history of overestimating Green support so there is a possibility that this poll is pointing to an even worse result for the party than indicated.

During the last polling period the Tasmanian forestry peace deal passed parliament and the party split on this major legislation on the floor of the House of Assembly, with four members voting in favour of the version returned by the Legislative Council and Bass Green MHA Kim Booth, long the least Labor-friendly and forestry-friendly of the five, voting against.

The poll shows the Liberal vote largely unaltered from the soaring levels of the last two polls, but what is interesting here is that the Labor vote is up to its highest headline level since November 2010 (not that 28% is any great triumph) while the Green headline rate is the lowest of this term in office, at 14%.  The core Green vote, counting only firmly supportive voters, is shown at a ridiculously low 9%.  I cannot remember it being this low for a very long time and suspect it would be necessary to go back to the earliest EMRS readings from the late 1990s to find a similar figure. 

However the news is not entirely good for the Liberals.  In the February article I noted that the party was making great progress in terms of the percentage of secure support (Table 2), which had hit its highest level in EMRS history at 46%.  In this poll that figure has gone back down to 40%, and in the more useful Table 3 (which includes voters leaning to a party, as do other pollsters) it is down four points from 48 to 44.  Both the "undecided" figure without pressing for a response (30%) and the "undecided" figure after pressing for a party response (19%) are at what I believe to be their highest levels in EMRS history, and certainly their highest levels in this term.  Once again these very high claimed indecision results call the usefulness of EMRS polls into question, since major pollsters do not get "undecided" rates that are nearly so high even when using the comparable (and lower) of the two undecided figures.  This is a pet rant that I have covered many times in previous articles but if EMRS cannot find better ways of getting meaningful responses out of respondents then there is a gap in the polling market waiting to be filled by someone who can.

As I noted last time, converting EMRS polls into likely state-level results is very difficult because of the paucity of electorate-level data; the distribution of votes between the different electorates makes a big difference.  Again, I suspect that Labor's vote will hold up better in Franklin than in other electorates, as outlined in my Uneven Swing to Liberals article.  However as the data used in that article age, with no new electorate-level data available (beyond a ReachTEL for one Legislative Council electorate) it gets harder to have confidence about the possible electorate level picture.

Anyway we should certainly not assume a uniform swing against the Greens from this poll, as that would leave them with only a handful of votes in Braddon.  This is what the uneven-swing model looks like if applied to my "interpretation" figures above (with the 3% for unknown independents reallocated) :  (Note that these "interpretation" figures rely on a history of EMRS understating Labor support and overstating the Greens.)

The projection based solely on this poll suggests 13 Liberal, 9 Labor, 2 Green and the last seat in Bass is anyone's.

In this distribution Labor gets off light with a relatively small swing against it and might conceivably even hold all its seats.  However the Greens lose seats to the Liberals in Braddon, Lyons and possibly Bass (unless Labor loses there instead) and so the Liberals still win government.  In this simulation the Liberals are actually diddled by being close to a fourth quota in two seats and a third quota in two.  On the figures above they would probably not get any of those extra seats and hence end up only winning by one.  However, it's more likely that in reality they would be up on the above in some electorates and down on the above in others, and hence that they would get at least 14 seats, perhaps more.

In the above simulation the Greens might still hold Bass, but with a vote well below a quota Kim Booth would be at the mercy of the dreaded Ginninderra Effect and hence might lose.  I have reservations about projecting Booth's vote given that he would be seen as having "kept the faith" with the hardliners while the other Greens would not.  However we only have to consider the forest peace deal scenario to see how hard and confusing things are going to be for the Greens and their voters at the 2014 election.  Voters will want to know whether given Green candidates are for or against the peace deal, since if the Liberals do fall short of majority government, which Greens get elected to parliament could determine the fate of this major policy legislation.  Booth could be targeted as a potential supporter of Liberal attempts to rip up the deal.  It will be interesting to see if the Greens can even keep all the dissenting forces in-house without the party actually splitting and greener-than-thou candidates running as significant Independents.

I also don't believe the 2.17 quotas projected for Labor in Denison in the model above as they will be going to the next election with a fairly weak candidate team. One of their incumbents is Graeme Sturges, who was massively thrashed at the last state election only to return on a recount.  This projected second Labor seat is very ripe for being picked off by an independent.  If Andrew Wilkie loses Denison (which is possible though at present he is favoured to retain it) he would probably win that seat very easily.

As this poll gives extreme results in the context of the last few years for both Labor and the Greens, and given the uncertainties created by the high undecided rate, I've taken a cautious approach in projecting where the parties are at in the sidebar, and considered the results of the previous poll as well. 

I will very probably add more comments in coming days in response to articles on the poll and so on and as I analyse the results further.

Postscript (22 May):  Checking old records I have found that:

* The last result worse than this for the Greens in an EMRS poll was in August 2001 in which the party polled 7% of firmly decided voters, which was extrapolated to 9% with undecided voters redistributed.  The party's vote in December 2001 and May 2002 with undecided voters redistributed was the same as it is now (14%) but off a higher committed vote in both cases.

* I cannot find any previous instance of an EMRS poll with a raw undecided figure as high as 30%.  Indeed during the early 2000s, EMRS raw undecided rates were typically much lower than now.

Some people may wonder why I put EMRS polls through such contortions to try to get something resembling a meaningful result.  It's worth remembering the situation in November 2005 in which, with an election only months away, EMRS released a poll claiming that Labor under Paul Lennon held only a 40:36 lead over the Liberals.  This was widely interpreted as pointing to a 10-10-5 result in 2006 (the result we actually got at the election after).  Paul Lennon retaliated by releasing selective details of internal polling, on which basis he claimed that the party would not lose a seat.  They didn't; in fact they went within a very thin whisker of winning one.

The vagueness of the latest poll makes it a choose-your-own-adventure for just about everyone (Green triumphalists excepted) and apart from the predictable over-focus on the "undecided" rate there hasn't been too much to see in public commentary on this poll and I haven't noticed any spectacular or novel howlers so far.  The Libs have copped a bit of stirring for dropping on the base rate from 46 to 40.  But the smaller drop when the question is asked in the form used by major pollsters (48 to 44) is not even a statistically significant change (unlike the drop in the Greens vote on the same indicator) and against the backdrop of such a high undecided vote, it is hard to know whether to even make anything of it at all.  There is often pressure on analysts to provide reasons for polling shifts and sometimes the best thing to do is just to say you don't know if it's real, and if it is real you don't know what has caused it.

A little while ago it was put to me (by someone who just might have a small stake in the outcome) that the Libs were deliberately gaming the EMRS polls by timing ad releases just before them (something they would not have needed to do this time because the LegCo election campaigns and results did the job for them.) I would think that if this was systematically the case and it was working and having a major effect that there would be internal polls floating around getting very different results to EMRS.  To my knowledge there are not, and I have seen bits and pieces of a few.  But it is something that is worth noting and testing, because Tasmanian elections can have a snowball effect in which a party that appears to be winning majority government gains supporters for that reason.  So it's understandable if a party would try it - overrated as the impact of political advertising often is.

I have also heard that some Greens are even clutching for the same straw being clutched at by Labor supporters federally, the mobile vs landlines issue.  Polling is difficult for pollsters that only call landlines because, proverbially, hardly anyone under the age of 40 has one and answers it.  However, people who think this means that the result is going to be massively skewed because the poll sample will include virtually no young people, and hence the result will be very conservative, are underestimating the pollsters.  Australian pollsters are not completely dumb and employ demographic scaling, so that even if the proportion of young people in their sample is small, they can still scale the youngsters so that their combined opinions carry as much weight as their share of the vote suggests.  This is why federal pollsters that do poll mobiles, federal pollsters that don't poll mobiles, and federal pollsters that don't even use phones at all are able to get very similar results.

Scaling does have risks attached to it (and it also means that a lot of stuff said by pollsters about their polls' margins of error is a bit dubious).  One risk is that because so few young people use landlines, those young people who do use them may be politically unrepresentative.  But scaling greatly reduces the risks associated with landline polling, and means that while those polling landlines only might get burnt to the level of a point or two, they're not going to miss the result by an amount so large as to render their readings totally invalid. 

Lastly for now, there have been a few vibes around in response to the effect that Labor is fine in Denison and not at risk of dropping to one there (at least, assuming they don't get gouged by an indie).  On that basis, if the Libs are not doing as well there as my uneven swing model suggests, this raises the rather scary prospect (for their opponents) that they might be doing better somewhere else.


  1. EMRS has become a sideshow. When Leon Compton put the 30% undecided figure to Bernard Keane this morning, he had to catch himself from laughing.
    Anecdotally I do think that there would be a larger undecided number at present than we would normally expect, however. Large numbers of people are telling me the same basic story; they're sick of Labor, (I read that as the old "its time for a change" mentality combined with perceptions of impropriety over a long period), but the Liberals offer nothing but old politics and negativity. The Greens don't appear to get much of a mention, love nor hate, outside of people associated with mining. I'm not sure what that implies for them, I can see the potential damage to their core voter base as you said, but I can see people considering them as a progressive vote without union attachment.

    But as I said, all anecdotal.

    1. Bernard's response - just saying that the claimed level of undecided voters is just not credible - was good. Of course, 30% is not the "real" undecided figure (since someone who offers a party when prodded is actually intending to vote for that party; they're just not "sure" enough to cough up that view immediately) but 19% as a figure for those not even leaning to a party or to other is still just too high.

      Wish I knew what was causing it but I don't. My guesses would be it's not about question methods, but rather about either staff skills in coaxing an answer or else pressure on staff to complete interviews at a certain rate.

  2. Agreed, its irritating to see polling used as a self-fulfilling prophecy by commercial media as well. The only direct spotlight on the credibility of the poll that I heard all day came from the ABC, while the rest of the narrative was "Liberals winning, Greens smashed!"

  3. If the variation across the state has increased, Bass and Braddon are the likely recipients of the increased vote (although an increase in Lyons cannot be discounted).

    For that to have any chance of an effect on the seat outcome, it would have to see the Liberal 3PP vote increase to somewhere in the order of 65%, which I would have thought unlikely even in Braddon.

    Then again, with the growing unrest up that way regarding the so called 'Tarkine', it might be possible.

  4. Four seats is in theory possible with a 3PP vote of around 62. For instance Liberal 62 Labor 27 Green 11 (which is actually almost identical to my "uneven swing" model projection for Bass). Preferences flow weakly across parties when the last candidate from either Labor or the Greens is excluded. But to get four with those numbers the Liberals would have to get very lucky on the distribution of votes among their candidates (so as to minimise leakage losses).

    There is one precedent of sorts; in 1992 the Liberals polled 65.7% to Labor's 20.8% and the Greens 8.8% in Braddon, which would have easily won the Liberals four out of five. But that was off a base of 57.5% from the previous election, whereas this time the base is only 45.2% in Braddon and 42.6% in Bass. Also that was with a team of four incumbents (they won five seats out of seven) whereas this time they only have two in each seat.

    It's something that polling has been pointing to as possible on and off over the last eighteen months - it would still be rather surprising if it happened, and less surprising if the Libs just overkilled all of Bass, Braddon and Lyons with only one extra seat to show for their troubles in each.

    After the last two days the Best factor could be an extra complication in Braddon.

  5. Braddon may well yield only a single Labor seat, so there is going to be a lot on infighting for that vote. The limited response I've seen on social media from the working Labor side is approval for the act, but not yet for the actor. If he backs it up consistently an with action over the next year Best may well win himself his quota, preselected or not.

    I agree, the fourth spot in any seat is still very long odds. The odds get a lot shorter if the Libs pick up four in the Federal senate this time around.