Tuesday, November 25, 2014

EMRS: Liberals Fall To Four-Year Low

EMRS: Liberal 42 (-4 since September) Labor 31 (-2) Green 19 (+3) Ind 6 (+3) PUP 2 (+1)
Interpretation: Liberal 43 Labor 34 Green 17 PUP 2 Others 4
Likely outcome based on this poll "if election held now":  Hung parliament (12-9-4) with slim chance of Liberal majority (13-9-3)
New aggregate of all polls: Likely Liberal majority government (13-9-3)

A new quarterly EMRS poll of Tasmanian state voting intentions has been released.  See also the excellent trend tracker on the EMRS site.  This poll continues a downward trend in this pollster's readings noted in my previous EMRS article (Closest Gap In Four Years) and sees the polled gap between the major parties close further and the Liberal vote at its lowest level since November 2010.  It's a rather poor poll for the new Hodgman government that suggests it has significantly lost support since winning office just eight months ago.  If this poll is to be relied on, the government would be in danger of losing its freshly minted majority in an election proverbially held now.  The next election is not until 2018 so there is not too much for opponents to get excited about yet.

As well as the usual EMRS habit of overstating the Greens vote, the vote for "Independent" in this sample (at 6 points) is too high, and this is a common issue with polls that include "Independent" as an option by itself in the readout.  (Voters may intend to vote for indies, but when there are no high-profile independents that suit their politics, most of them then don't.)  Nonetheless this figure for independents does suggest voters are looking for fourth-party alternatives.  It is the highest figure for "Independent" in EMRS polling since May 2012, and there is a history of this figure being higher in the pollster's May samples because of interference from Legislative Council elections held in that month.

The poll shows Will Hodgman still with a large (and relatively meaningless) lead over Bryan Green as Preferred Premier, 50-22, with 28% (up five) choosing neither.  Preferred Premier scores are usually skewed to the incumbent, especially against new Opposition Leaders, so there's not too much to see here.  If it's still like this in, oh, mid-2016, it might be time for Labor to be more concerned.

The poll echoes some curiously bad results for the government in two very small Morgan SMS samples taken since the last EMRS and ReachTEL polls, and suggests that those samples, while very likely skewed against the government, were not as skewed as I at first suspected.

EMRS Poll Translated to Seats
The past record of EMRS is that the pollster tends to overstate the Green vote, understate the Labor vote and get the Liberal vote more or less spot on.  Also the poll has a high undecided rate even after prodding unsure voters, in this case 11%.

With some corrections made for these tendencies, I model the votes based on this poll as Liberal 43 (-8 from election) Labor 34 (+7) Green 17 (+3) PUP 2 (-3) Other 4 (+1). A straight swing of that magnitude would result in the Liberals losing seats to Labor in both Braddon (where the Liberals remarkably won four seats in March) and Franklin.  The Liberals would also probably lose their third seat in Lyons to the Greens.  However if the swing was not quite uniformly distributed, the Liberals might hold their third seat and keep a majority, especially if the distribution between the three sitting Liberal MHAs was as uniform as it was last election.

An election being held with these kinds of primary votes is a very artificial scenario.  We have seen in the past - a good example being the Lennon Labor government in 2006 - that the party that leads in the polls can make the case that only it can win majority government.  So if the Hodgman government went into an election campaign with this sort of voting intention base, it's quite possible voters would come back to it rather than risk causing another minority government.

Morgan Was Here First (Sort-Of)
I didn't cover them in much detail at the time because I wasn't sure what to make of them as they were so different to other published polling, and as the sample sizes were extremely small.   But it is time to note that Tasmania has another state voting intention player now with the return of Roy Morgan Research after an absence of several years.  Alas, it's not a return on the scale of the pollster's old Tasmanian phone polls, but if we can make some kind of sense of it, it will be better than nothing.

Two very small 300-vote SMS samples have been included in the pollster's recent state polling releases.  The first in late September produced results of Liberal 39.5 Labor 33.5 Green 19.5 PUP 3 Ind/Other 4.5.  The second in late October produced results of Liberal 41 Labor 34 Green 19 PUP 2.5 Ind/Other 3.5.  The second is very close to the current EMRS poll.

At the time I treated these polls with vast suspicion because of their very small sample size and a history of apparent pro-Labor skew in Morgan's small Tasmanian federal samples (conducted by a combination of SMS polling and face-to-face polling.)  While Morgan's face-to-face polling is known to skew slightly to Labor (their SMS polling apparently doesn't), I suspected that the skew in Tasmania's case might be greater and that there might be issues with the Tasmanian SMS panel specifically.  This is still probably the case, but with the EMRS poll providing some evidence of the government genuinely losing support over the past few months, the Morgan samples are looking a little bit less outlandish now.

Morgan has been doing a lot of SMS sampling in the lead-up to the Victorian election and its samples have been massively over-estimating the Green vote compared to everyone else's.  At the moment I am conservatively estimating that Morgan's Victorian sampling has had the Green vote five points too high there on average, but we'll find out on the weekend.

As a first attempt at benchmarking these Morgan samples off my corrected readings of the two EMRS polls either side of them, I reckon they have the Liberal vote about four points too low and the Greens vote about four points too high.  Because of the small size of the Morgan samples these estimates may not be very accurate and will need to be fine-tuned with more evidence.  For instance if this EMRS poll is itself affected by sample error in Labor's favour, then it is possible the Morgan samples are more erratic than I've suggested.  But at least there is some basis for roughly using the Morgan data in my state aggregate now rather than throwing it away entirely.

New Aggregate
My Tasmanian aggregate was last revised in mid-September following a very large-sample-size ReachTEL that contradicted the early September EMRS and showed the Liberals still with a hefty lead.

Considering that the two Morgan samples are by a method that has never been tested in Tasmanian elections, are a smaller sample than the EMRS when combined, and are older than the EMRS, I can only give them a very small weighting (I've gone with giving the two combined one-fifth of the weighting of the EMRS, though even this is generous.)  I've weighted the new EMRS at 37.5%, the October Morgan at 4.5%, the September Morgan at 3% and the previous aggregate value at 55%.  There's no new electorate breakdown data so I've applied uniform swings to the previous electorate values, which were taken partly from the ReachTEL and partly from the state election.  Here's the aggregate:

The aggregate gives the fourth Braddon Liberal seat back to Labor (and will probably keep doing so all the way to the election).  It also now gives Labor the Liberals' third Franklin seat.  On these figures the Liberals with three evenly-matched incumbents and a healthy flow of fourth-party preferences should retain their third seat in Lyons, but it's not that far from being lost.  There are two Labor-Green battles: one in Bass and one in Lyons.  Although Labor leads the Greens by 0.1 quotas on party score in the Bass aggregate, and the Greens lead Labor by 0.1 quotas in Lyons, I suspect the result in both cases would more likely be the incumbents (Green in Bass and Labor in Lyons) retaining, because of profile and leakage issues.  Anyway one-all seems a fair division of these two contests so my new aggregate of all state polling is Liberal 13 Labor 9 Green 3.

Why is it so?
Well really, I don't know.  It's easier to explain data you expect to see than data you don't.  We still have to see more polls to see if this one is exaggerated, but I also said that last EMRS (and that one was four points higher.)  They new government has pretty much governed as it intended to do, even if it did make an abysmal mess of anti-protestor laws (finally about to pass in a greatly watered-down form after the initial draft went way too far.)

If anything, these lacklustre polling results (combined with the drubbing for the party's candidates in the Legislative Council elections) suggest the last election was just about chucking the previous mob out and that Tasmanian voters don't actually want the Liberals to keep their promises to get tough on public sector spending.  Perhaps voters just want a harmless government that's vaguely pro-development, competent and not in league with the Greens.  As seen at federal level, it doesn't seem that easy any more for new governments to surf into office and get months and months of 50+% polling and uncritical reviews.  The honeymoon period aint what it used to be.

Update: Another New Morgan Sample

Another new Morgan SMS sample is out with a sample size of a feeble 281 and figures of ALP 41 Liberal 39 Green 13 PUP 2 Ind/Other 4.5.  Not only is the sample size tiny but this poll provides still further evidence that this polling form skews to the left in Tasmania, though the low Green reading this time suggests the skew is coming on average from both Labor and the Greens, and not just the Greens as previous samples suggested.

I've redone my calculations of the aggregate above with the new Morgan included and the one two months ago removed, with appropriate re-weightings and reassigned house effects, but the impact was so small as to not be worth reposting.

I do hope any journalists reporting this Morgan "result" will take note that even if the sample had no house effects the figures would still have a margin of error of 6%.  As it is the strong evidence of house effects in these polls implies they could be even further out than that.  And the pollster's claim that Labor would "win an election easily" based on those unlikely figures is false; it would more or less certainly be a hung parliament. 

1 comment:

  1. And Labor + Greens add up to over 50% for the first time in a long time.