If a state election were held today, the Bartlett Labor government would have a fair chance of winning it outright.  This is the conclusion I draw from a combined study of the February and May EMRS polls.

Before I start, a small but important correction to my previous article. In that article I wrote that “Lyons has a very high “undecided” rate (around 35%)” and that “It is difficult to be sure what the undecided rates for all the other electorates are as the claimed sample sizes for Bass and probably Franklin appear to include the undecided voters”.  Both these comments were made on the incorrect assumption that EMRS still samples roughly equally in each electorate.  However, had I noticed the sample sizes in the electorate-by-electorate breakdown for the otherwise usually meaningless Preferred Premier stat, I would have seen that this is actually no longer the case and both my quoted comments were therefore incorrect.  In the February poll, EMRS sampled 242 voters in Bass, 200 in Braddon, 188 in Denison, 222 in Franklin and just 148 in Lyons.  The actual undecided rate in Lyons, assigning those leaning to a party to that party, was just 12%.  The May samples (236, 188, 203, 215, 158) show a similar if slightly less extreme level of skewing.

I have not been able to determine for sure whether or not EMRS scales their percentage of total respondents to account for the large difference in sample sizes between different electorates.  There is no explicit indication either way and it is difficult to tell because the figures are rounded to the nearest whole point, and because the difference created to the total by including extra voters in Bass and fewer in Lyons is not great (typically, the difference between an average across all electorates and a scaled average is not more than about half a point).  But firstly, EMRS should state clearly whether they are using scaling or not, and secondly they should use scaling.  To use a sample of five similarly populated electorates with 50% (or more) extra voters sampled in one than in one other, and then not scale those averages, would be statistically ludicrous.  This applies even if the impact on the results only changes a point here or there in the statewide standings, especially as the least sampled electorate is one in which Labor’s lead over the Liberal Opposition is large, and the most sampled one is one where the major parties are closer and the Green vote relatively high.

The May poll is available as a download at http://www.emrs.com.au/. As usual, Peter Tucker has been first with the latest with an excellent basic write-up and graphs at http://www.tasmanianpolitics.blogspot.com/.  He shows that “None of the movements in the raw numbers [since February] are statistically significant, although the trends are perhaps starting to show a couple of things.”  Peter also correctly singles out this Mercury piece piece as a beat-up on account of it declaring that the poll shows Premier Bartlett has lost support.  Of course a two-point drop from poll to poll does not show that support has been lost, as it is not statistically significant.  It is becoming tedious to haul mainstream backsliders into line about this article after article.  Perhaps such claims by journos should simply be made illegal, and offending newspapers be subjected to the ultimate in inhumane punishment: having to allow me to fill page three with whatever takes my fancy for a month.

An article by Loretta Johnson in The Examiner (probably subscriber-only by the time you read this) is rather different to the usual interpretive swill.  It quotes Tony McCall arguing that the Liberals’ poor standing in Denison and Lyons casts doubt over whether anything apart from Franklin would change if an election was held now – especially, in Denison the Liberals are over a quota behind.  However the article does not come with any warning about drawing conclusions from very small sample sizes, and electorate samples of only 200 votes or so can easily be several percent out.

Matthew Denholm in The Australian argues that the opposition parties “are failing to deliver the Bartlett Labor Government a killer blow” and says that the poll suggests the Liberals would pick up seats from Labor in Franklin and Lyons.  Again, the reservation about small sample sizes applies but is not stated.  Concerning Lyons, it breaks 45-35-15-5 in the sample; if we redistribute the 5 in proportion we get 47.4-36.8-15.8, which in quota terms is 2.84-2.21-0.95.  The Liberals would indeed win a seat on such figures if accurate, but it would only narrowly be at the expense of Labor rather than at of the Greens.  Since the Lyons sample only contains 133 voters who were able to give an opinion, and given that Labor is within striking distance (given the massive error margin for such a small sample) of holding their third seat one way or the other, any conclusion that Labor looks like losing a seat there based on that sample is meaningless.

Something I noticed from Peter’s graphs is that the Greens’ raw support reading of 13 appears to be their worst for over four years, worse even than those taken when the undecided vote dropped in the leadup to the 2006 election.  Although one bad reading by itself means nothing, their last reading (15) was also vaguely sub-par, and I think this is the first time since late 2005 the party has polled two readings of 15 or less in a row.  There may not be anything there, but with economics and employment front and centre for many voters in the current financial environment, the Green vote could be softening. I wonder if the journo who called their previous 15% “crushing” has now fallen silent for lack of a still more exaggerated putdown.

The poll has been described as showing movement away from all parties, with a four-point increase in EMRS’s perennially high “undecided” rate and a one-point increase for “independents”.  Indeed, the “Independent” percentage as an average across the five electorates, including “leaning” voters is up from 2.4 points in February to 4.2 points in May.  This is not statistically significant but it is worth noting a factor (as an alternative to dissatisfaction with all parties or GFC-induced confusion) that could have caused the Independent percentage to rise.  This factor is the Legislative Council elections.  A very small proportion of Tasmanians are probably sufficiently confused that when they see Legislative Council elections with high profile independents, they imagine that they might vote for the same candidates or similar creatures in the Lower House elections, even though such candidates tend not to actually exist.  Notably, there is no Andrew Wilkie blip on the Denison radar yet.  That may be a function of him not being specified as an option in polling, or it more likely indicates that his current appeal is largely confined to leftist intelligentsia, like that of the Democrats in their final years of anything resembling electoral relevance.

Both Tony McCall and EMRS’s Tony Hocking suggest the high undecided rate is a symptom of voter uncertainty caused by the global financial crisis.  I am unsure if Hocking has any inside knowledge to justify this conclusion but in any case it’s probably unmerited.  There is nothing unusually high about an undecided rate of 24 in an EMRS poll.  It was 22-23 for most of 2008 and about that level at the same point of the previous electoral cycle.  The four-point gap between February and May only indicates that the February result (20% supposedly undecided) was unusually low by EMRS standards (about as low as they ever get except in an election leadup).  Furthermore when those “leaning” to a party are allocated to that party, the four-point difference disappears. 

The main use of having two polls in a row with electorate breakdowns is that it is possible to pool the samples to provide a very rough electorate-by-electorate indication of voter sentiment over the past six months.  And this is exactly what I have done in the figure provided.  I have pooled the February and May electorate samples to give a combined sample of 2000 voters with Lyons the least surveyed (306 voters) and Bass the most (478).  (Annoyingly, EMRS have over-sampled the most boring electorate and under-sampled the most critical one.)  These combined samples have margins of error around the range 4-6%, which makes them more useful than samples of only 200 votes or so, although a much larger sample would be better.  It would also be possible to weight the more recent sample more heavily as a stronger indication of current sentiment.  I have not done this as I am unsure whether it decreases or increases uncertainty when dealing with such small samples.

In the second set of figures for each seat, I’ve redistributed the claimed votes for Other or Independent, and the Undecided vote all in proportion to party support.  I’ve done this even with Other/Independent because the reported Other vote is negligible while I believe the reported Independent vote to be almost entirely non-genuine.  Assuming this set of redistributed figures reflected voter sentiment perfectly, Labor would lose a seat in Franklin.  It would probably lose one to the Greens in Braddon, although there would be the possibility of the Green either getting squeezed out by candidate distributions within the parties, or else losing on the preferences of one major party flowing strongly enough to the other (in this latter case, the Liberals would be well placed to get the seat).  It would probably lose one to the Liberals in Denison, depending on leakage issues (probably more a problem for Labor) and the strength of Greens preferences.  It would probably lose one to the Liberals in Lyons, although it might just beat the Greens on Liberal preferences.  So assuming Labor held one of these three possibles, it might break something like 11-9-5 or 11-10-4; even 10-10-5 would be possible – as (remotely) would be 13-9-3.

But it does not take much variation in Labor’s favour from this second set of figures before they hold all of Braddon, Denison and Lyons.  And indeed, this is underlined by my third set of figures, which are based on the observation (both in Tasmania and elsewhere) that voters who call themselves undecided in opinion polls – EMRS polls especially - most commonly break to the government.

What I’ve done with my third and in my view more credible set of numbers is redistribute “undecided” according to the formula suggested by Peter Tucker: 50% Labor, 25% Liberal, 10% Green, 15% Ind/Other.  I think this is a good middle-of-the-road, perhaps even cautious idea of how EMRS-undecided votes may break; for reasons why I don’t go even higher in Labor’s favour see my article from six months ago.  But then I’ve re-redistributed the 15% Ind/Other according to the same formula, since I do not know what Ind/Others will contest apart from Andrew Wilkie and possibly Ben Quin, and I believe that if they are Green-friendly, they will mostly just be taking votes off the Greens and giving them back as preferences.  The result is that I am redistributing the combined Undecided/Ind/Others results in the ratio 59-29-12.

The results of the combined sample still show Labor losing one in Franklin, though holding it is not outside the margin of error of the sample.  In Denison, Labor is on virtually a quota lead over the Liberals and wins on Green preferences, though 2-2-1 is within the margin of error.  Bass goes 2-2-1 with nothing else likely.  Lyons goes 3-2-0 with both majors just over a quota, though the Greens hanging on is well within the margin and can occur with equal likelihood at the expense of either major party.  In Braddon, Labor is neck-and-neck with the Greens for the final seat on primaries, but most likely wins it on Liberal preferences, though 2-2-1 and much less probably 2-3-0 are also within the margin. 

The sum of the individual most likely outcomes according to this modified sample is a Labor majority (13-9-3).  However, allowing for a few points of variation in particular seats a wide range of outcomes are possible, and even if Labor are considered individually likely to hold each of Braddon, Lyons and Denison, there is a high stochastic risk of them still losing one of them.

Thus, considering the last two polls electorate-by-electorate and applying a plausible allocation of the undecided vote produces a scenario in which Labor could win outright.  However, there are sample size uncertainties as well as (large) uncertainties in the assumptions for reallocating the undecideds.  What I can say is as follows:

• The widespread view that Labor will probably not hold their third seat in Franklin remains consistent with the polling as interpreted.
• Although Labor is at risk of losing a seat somewhere in at least one of Lyons, Denison and Braddon, there is no strong evidence that they are likely to lose any specific one of those three.
• Although the Greens squeaked home in Bass in 2009, it is probably Tim Morris rather than Kim Booth who will be most at risk of defeat in 2010.
• The poll, taken after the Legislative Council elections, supports the view that there is no major anti-Government backlash.

Much could change in the next ten months before the election is scheduled, assuming that the parliament runs full term.  These are potentially turbulent times and exceptionally good or poor handling of the state’s economy in them could easily and quickly produce swings of at least a few percent.  But much could also stay the same. 

As I have noted in previous articles, claims that a hung parliament would almost certainly occur in 2006 were proved completely premature, and such claims are equally so now, whether one actually happens or not.  But more than that, I do not now even consider a hung parliament to be the clearly most likely outcome for the 2010 election.  With two polls in a row showing Labor substantially ahead of the Liberals, with signs that Green support could be declining, and with the known issue of undecided voters breaking very unevenly at election time, I think we are in 50-50 territory.  Perhaps next poll I will crawl back into my burrow, declare I was getting excited about nothing and there will be a hung parliament after all.  Or perhaps not …

Dr KEVIN BONHAM

If a state election were held today, the Bartlett Labor government would have a fair chance of winning it outright.  This is the conclusion I draw from a combined study of the February and May EMRS polls.