As previously mentioned (HERE), the August 2009 EMRS poll showed a massive swing away from the Bartlett Labor Government, compared with the May 2009 poll in which it had been in a reasonable position.

The November 2009 poll is at best (for Labor) more of the same of what we saw in August, but also contains hints things may be getting even worse. This is not a great surprise since after a horrible polling cycle in the leadup to the August poll, Labor badly needed to get clear air in this one to improve its position. It has been completely denied it by confusion over its Tasmania Tomorrow education reforms, dissatisfaction with sewage and water rates implementation and land tax rises, incidents involving Infrastructure Minister Graeme Sturges.

After the polling period, there was an intensification of fallout that went straight to the top over the short-lived and at best hamfisted McCreadie reappointment. As with the previous cycle, Labor’s problems are more or less completely self-inflicted and despite the good condition of the state’s economy, the government is failing to create confidence that it could ever extract itself from stuff-up mode. It is no surprise then that the Opposition has a substantial (and now probably to at least some degree real) lead in the polling.

The EMRS figures are here (PDF download, HERE). Firm support for the Liberals is up four points to 37, and the raw Undecided rate (on which see more below) is down three to19. With tentative responses added in (this forms a result comparable to the headline rates in most other pollsters’ polls) Labor is at 29, Liberals are at 40, Greens 19, Independents 2, 11 undecided. Again the only serious change is in Liberal support. There is a rounding issue because the numbers add to 99 in the previous sample and 101 in this one, but what may have happened is that some of those leaning towards the Liberals last time have firmed for them, and that a new batch of voters are now leaning Liberal, many but not all of whom were previously “undecided”. Having said that, the difference between the Liberal votes in the two polls is only bordering on statistically significant. We can’t be certain things have got worse for Labor since August, but we can be very confident that they have not got any better. Since we are now four months from the election instead of seven, that’s another three at-best wasted months to add to the loss of a whole year of previous steady progress that was wiped out in the previous three-month cycle. The net result is at best no gain in eighteen months – and eighteen months ago Labor was in the public-opinion pits of the final days of Lennon. Labor can forget about climbing the mountain mentioned in my previous article; their challenge now is to avoid falling off the cliff.

Or, in more prosaic language, unless EMRS polls have somehow become completely defective, Labor now has no realistic chance of retaining its majority (with the usual proviso that extraordinary events sometimes change things). The question thus is whether it can contain the swing to the loss of a few seats, or whether things will just keep getting worse. This government has yet to descend to the open leadership brawling and major internal policy division displayed by the Lowe/Holgate team in their race to the bottom of the barrel back in the Franklin Dam days, and nor has it trashed the state’s economy (indeed, given the global circumstances, the state is doing remarkably well). Apart from those two things, it has ticked most of the boxes on the how-to-destroy-your-own-government checklist, and has continued to check them regularly in the course of the last six months. It is no wonder that polling is now suggesting a double-digit swing from Labor to the Liberals compared to the results of the (very one-sided) 2006 state election. I will not be at all surprised now if we really see a 10-12 point swing (something like what happened to Groom in 1996) when we finally get to the polls.

ABC reporting on the poll (HERE) on the poll has already included the at this stage premature (or at least ambiguous) prediction that the poll shows the Liberals on track to win the next election (an 11 point lead is typically not quite enough for a party to win a majority, especially not if it is the Liberals). The Examiner is similarly over-hyping the Libs’ position (HERE); they are said to be “with a strong chance of winning government in an immediate election” (actually, that would most likely depend on the Greens). The Mercury’s report (HERE) sticks to the facts, although it gives a lot of prominence to the electorate-by-electorate samples that are too small to be reliable. You can see what the Premier thinks HERE and the Poll Bludger thread is HERE; I suspect Peter Tucker (HERE) will have something up by the time this piece is posted. Antony Green correctly declares the poll “dreadful news” for Labor and has some nice graphs showing how starkly things have gone pearshape for Labor in the last two polling periods (HERE).

It has previously been possible to argue that even when Labor are behind in an EMRS poll, they are perhaps really in front, because the hardcore “undecideds” in EMRS sampling have proved in the past to be soft Labor voters. The problem for Labor in this poll is that even giving them all the non-leaning undecideds only puts them up to level-pegging and a hung parliament. Furthermore, if most of the undecideds jump to the Liberals, then Labor is in very big trouble indeed.

As I have noted previously, EMRS’ electorate breakdowns are generally fairly useless for modelling purposes because the sample size is too high. However, they can be pooled with previous sampling to create an average of two polls. In the case of the August and November polls, this creates merged samples ranging from 312 votes (Lyons) to 472 votes (Bass) with margins of error of around 3.5-5.5%. However, since the November poll is much closer to the election than the August one, and since there is some suggestion of change in the favour of the Liberals between the polls, I have decided to weight the average 60:40 in favour of the November poll. This slightly increases the effective margin of error, but it should still not exceed about 6% for any figure. Incidentally, the spread of voters between the five electorates in the November poll was much more even than in the previous two; one hopes that this continues.

I will give the figures from the merged weighted sample in the form Labor-Liberal-Green-Other-Undecided. I assume based on past experience that Undecided voters will most likely mostly vote for one major party or other, but I do not necessarily assume they will vote for Labor as they have in the past.

In Bass (29-37-16-4-14) the most likely result is the existing 2-2-1. However, the Greens are not likely to have a big (or necessarily any) surplus, so if the undecideds swing very strongly to the Liberals then 1-3-1 is possible, as (distantly) is 2-3-0 if the undecideds swing strongly Liberal but the Green vote drops.

In Braddon (32-43-12-1-11) Labor almost certainly loses a seat. If the undecideds are mostly Labor voters and/or the Green vote lifts or gets a cut of the undecideds then 2-2-1 is possible, but the more likely outcome on these figures is 2-3-0.

In Denison (26-37-23-3-11), the most likely outcome is 2-2-1. If nearly all the undecideds go to the Liberals then 1-3-1 becomes possible, but I don’t think anyone really believes that that is presently on the cards with the Libs with just a single incumbent vs Labor’s three (including the sitting Premier). The November sample with the Liberals a whole quota ahead and the Greens and Labor level-pegging for second spot is a little difficult to take completely seriously at this stage. Still no sign whatsoever of action for independent Andrew Wilkie, though this is exactly the sort of scenario (with stray bits of quotas all over the place) in which a sufficiently high primary vote (I have mentioned before that this must be at least 8%) would make him competitive. It would be interesting to see whether including him explicitly in the sample would give a higher reading, as it did for Ben Quin in Lyons in federal polling.

Franklin (31-35-22-2-10) is a straightforward 2-2-1. I have previously mentioned that 1-3-1 is not entirely off the cards if the swing is really on (given that the seat is notionally 2-2-1 as a result of the redistribution anyway) but the lead for the Liberals is not enough to take that seriously even if all the undecideds break their way.

Lyons (31-35-18-2-12) must be a real worry for Labor; it seems to have thrown close to a whole quota of support there to the opposition parties in just nine months. Lyons would also be a straightforward 2-2-1 on these figures.

So, this weighted two-poll sample suggests a parliament of something like 10-11-4, but a couple of Labor’s ten seats there are not rock-solid. Although an outright Liberal majority remains a remote chance, there is a greatly increased chance of the Liberals becoming the largest party. While David Bartlett has indicated he will go back to the Parliament and defy it to vote him out unless the Libs have a majority, it would be hard to see a Labor government with, say, nine MHAs surviving any length of time – unless, of course, Labor went back on its “no deals” line and offered the Greens a formal coalition. One can only imagine how much the Greens might expect to be happy with such a deal in such a circumstance.

For many years Labor has been able to use the line that only they can win majority government to keep potentially straying major party pro-majority voters in the fold. The problem for Labor now is that on these figures it is hard to see how they could do so, and if either party has a realistic (however slim) chance of winning outright it’s the Liberals. Now that this poll has confirmed (and possibly then some) the result of the August poll, I expect to see the Liberals now using the line that they, and not Labor, are now in a position to push for a majority. Possibly Labor will (as they did in 2006) point to the very poor predictive record of past EMRS polls, and drag out mysterious “private polling” that alters the picture somehow.

In the August poll I analysed what I referred to as the “Bartlett factor” (the extent to which the gap between Bartlett and Hodgman’s preferred premier ratings compares to the gap between the two main parties). This factor, which was then in decline, did not decline further in this period on a statewide basis (in fact, in some electorates it may have improved). But there is something going on that is worth noting. Over the course of the year, Nick McKim’s preferred premier score has been increasing at a faster rate than his party’s support level, to the point that his PP score as a proportion of his party’s support is about the same as Hodgman’s and actually a bit better than Bartlett’s. The Premier’s previously noted attempts to marginalise McKim and the Greens are clearly failing.

Finally, in articles concerning EMRS polls, too many psephologists to mention have commented on the very high undecided rates these polls generate compared with those created by major national pollsters (even at state level). In his comments on the previous EMRS, Antony Green (HERE) offered some comparative thoughts on this issue, pointing out that the comparisons are often overblown:

“What EMRS does with its tables is include data that other companies roll into headings and footnotes. EMRS publishes its findings with the data in a rawer form than Newspoll, but comparable results are available if you dig into the report.”

This is absolutely correct – whereas EMRS asks first who you are voting for and then if undecided who you are leaning to, other pollsters typically form their headline rate from a question that aims to capture the “leaning-tos” inside a party figure. Thus, the undecided rate in the November EMRS poll for comparison with a state Newspoll isn’t the 19 in table 1, but the 11 in table 2. In the last three polls, this figure was 15, 15 and 12.

Antony continues:

“What can we say about the relative size of the undecided vote in opinion polls? First, we know from experience that the undecided figure shrinks as you approach polling day. Second, on almost all data I have ever seen, you get a lower undecided rating in Federal polls compared to State polls.

To illustrate these points, look no further than Newspoll surveys before the Queensland election in March this year. Newpoll’s first poll in February 2009, before the election was announced, had 8 per cent of voters uncommitted and a 2% refusal rate, a 10% figure that is nearer to the EMRS result. In the final week of the campaign one month later, only 2% were uncommitted, 1% informal (whatever that is) and 1% refused.

In summary, the EMRS poll has a 12% undecided vote, higher but not outrageously different from similar mid-term state polls on the mainland.”

In August I thought I’d have a look at this in detail and came up with the following observations:

While Newspoll reports a refusal rate, there is no mention of a refusal rate in EMRS’s samples. It is unclear whether EMRS incorporates “refused” responses in its uncommitted category or simply excludes them from its results altogether.

An undecided rate of 8 is actually on the high side for a state Newspoll. I downloaded 103 state Newspolls from five states going back to the start of 2005 (excluding Tas and NT where Newspoll only polled very close to an election). In these polls, 8 was the maximum undecided (as distinct from “refused” or “informal”) figure, and occurred only six times.  Four of those (like Antony’s example) were in Queensland.  The mean undecided rates were: NSW 5.1 +/- 1.0, Vic 5 +/- 1.0, Qld 5.7 +/- 1.7, SA 5.6 +/- 1.1, WA 4.5 +/- 1.5.  These do include a few polls taken just before elections in which the undecided rate is lower, but nonetheless the Newspoll state undecided figure for mainland states is usually 6 or below. The average “refused” rate is about 2 +/- 0.7 with a similar level of variation between states. (As for “informal”, that figure generally appeared only very close to state elections, and I suspect it means exactly that – voters who explicitly tell Newspoll that they will not case a formal vote.)

Although Newspoll state undecided rates fall sharply very close to an election, there is not a clear and consistent pattern of decline through the electoral cycle and as the campaign approaches. Indeed, in Queensland from Nov 06 up til the start of the 2008 campaign, the pattern, if anything, pointed in the opposite direction: 4, 5, 4, 4, 4, 5, 6, 6, 8 (the 8 being the example above).

Thus, while comparing undecided rates in the 20s for EMRS with undecided rates from other pollsters is indeed failing to compare apples with apples, EMRS’s true undecided rates are still substantially higher than Newspoll’s are for mainland state polling, perhaps close to double when the comparison is truly like-with-like.

Of course, this could be genuine voter indecision, as a product of the far greater tendency for Tasmanians to jump from one party to another. That tendency is in turn a result of Tasmanian voters’ tendency to circumvent proportionality by jumping between major parties to avert hung parliaments, an aspect which is very well covered by Antony in that article. So it might seem on surface that undecided rates in the mid-teens in Tasmania could be genuine, and that the Tasmanian system could be encouraging voters to hedge their bets until they see who is in the best position to win.

However, EMRS are not the only pollster to have ever run regular Tasmanian opinion polling. Until early 2007, when it abandoned state polling, Roy Morgan Research conducted regular face-to-face polling in Tasmania that regularly returned undecided rates of less than 5%, similar to its rates in polling in other states. It is very likely that the different polling method contributed to the difference, or that asking the question in different ways (for instance, Morgan often used a long list of parties) made a difference, but the point is clear. Tasmania does not have a genuinely large mass of voters from whom it is impossible to wring a peep concerning even weak voting intention by any method whatsoever. State voting in Tasmania is indeed very volatile because of the hung parliament issue (and perhaps also because of instabilities created by having a small Parliament), but that does not mean that voters are more vague in their intentions at any given time compared to other states; it probably means those intentions change very readily. My view remains, as has been expressed before, that the “undecided” rates are rather high in EMRS polls not primarily because of the way Tasmanian voters respond to their system, but primarily because the other pollsters are doing something better.

It is now a little over four months until the Bartlett government reaches its use-by date and in all likelihood is moved to the minority-government discount bin (if it is not discarded entirely). Polls will come thick and fast from multiple sources once the campaign starts in earnest. It will be interesting to see if we get anything else in the three months before the next scheduled offering from EMRS.