Reports in The Mercury: Poll tips dead heat and The Examiner, based on the most recent EMRS poll suggest the Lennon Government is at the end of its parliamentary noose and would lose its outright majority if an election was held today.
The prediction given is the same as for a previous EMRS poll: ten Labor, ten Liberal, three Green, two unclear.
These results have been gleefully received by those (mostly wishfully-thinking Greens and media sensationalists with a vested interest in change and constitutional crisis) who claim to be certain or confident that the next election will result in a hung parliament.
In fact, such conclusions are completely premature, and are based on a combination of extrapolation from small sample sizes, unfamiliarity with some quirks of the Hare-Clark system, lack of acknowledgement of the concept of margin of error and, surprisingly, an outright mistake by EMRS in calling its seat-by-seat breakdowns.
The only thing that this poll proves is that at anything like present support levels, either Labor will remain in majority or else there will be a hung parliament. It says nothing that stands up to scrutiny about the likelihood of those two results.
I shall start with the mistake. The following are the seat-by-seat percentage breakdowns stated by EMRS, with undecided votes reallocated, given in the order Labor, Liberal, Green:
Denison 45,26,23 Bass 35,46,17 Braddon 45,38,14 Lyons 46,36,12 Franklin 46,34,16.
(A longwinded but necessary note at this point: These totals do not quite add to 100, presumably because some voters specified other parties or independents. In reality, the “other” figure in opinion polls usually overestimates the votes actually gained by other candidates, showing that many of those intending to vote for “other” either vote informally or else vote for one of the three main parties. Those who vote for minor parties, except when right-wing parties like Tasmania First are unusually prominent, actually tend to direct preferences to the Greens (in 2002 the Greens ultimately gained 41.4% of minor party preferences, the Liberals 29.8% and the ALP 25.4%). However it is not clear how many of those saying they will vote Green (for instance) actually end up voting for a minor party (Democrats, Socialist Alliance, Liberals for Forests) and giving their preferences to the Greens. I’ve assumed the minor party issue will have no real effect and the missing percentage points will be split evenly between the three main parties.)
A quota is 16.67% of the vote in each electorate.
Two of the five electorates are relatively straightforward. In Bass, Labor has over two quotas, the Liberals have well over two, and the Greens have just on one, so the seats would split 2/2/1.
In Franklin, Labor has well over two, the Liberals have just over two and the Greens have very nearly one, so again 2/2/1.
In Lyons and Braddon, the Liberals have 2.16 and 2.28 quotae respectively, so they’d only be winning two in each of those. So to win ten seats the Liberals need to also win two in Denison, and on EMRS’s figures, they won’t do it.
Green preferences would need to favour the Liberals
Denison is shown as splitting 45% ALP (2.70 quotas), 26% Liberal (1.56) and 23% Green (1.38). All other things being equal, the Liberals would need to gain 0.14 of a quota from the .38 of a quota left over by the Greens. Not only do about a third of Green votes exhaust from the system when there are no Green candidates left in the count, but those that remain in the count favour Labor over Liberal by a margin of around 60:40, which means the final Labor candidate should get home by around .23 of a quota, which is really rather comfortable.
The Green preferences would need to favour the Liberals over Labor by more than three to one to give the Liberals two seats. So this is my first point of difference with the EMRS story: the most likely allocation of seats if these results were recorded is 11/9/3 with two undecided, not 10/10/3 with two undecided as claimed.
This makes a huge difference, because now if Labor can win three seats in both Braddon and Lyons, Labor retains majority government with 13 seats and only one seat lost (Franklin).
In Braddon, Labor polled 2.70 quotas according to EMRS, the Liberals 2.27, the Greens 0.84. In Lyons, Labor polled 2.75 quotas, the Liberals 2.15, the Greens 0.72. Liberal preferences when there are no remaining Liberal candidates in the count exhaust from the system about 60% of the time, and the remainder favour Labor over the Greens by about 55% to 45%. This means that Labor makes very small gains on Liberal preferences. On the figures above, the Greens would be most likely to win Braddon, and Labor would be most likely to win a third seat in Lyons, but the latter would be very close.
Thus, on the seat-by-seat breakdowns provided by EMRS the most likely single outcome is 12 Labor, 9 Liberal, 4 Green. The next most likely outcome is 11/9/5 and the next after that is a Labor majority government with 13/9/3.
So, assuming the 200-vote EMRS electorate samples are very accurate, the above has already shown that a hung parliament is merely likely, not certain.
The next question is: how accurate are these samples individually? The answer is: not very.
Unfortunately in Australia, unlike in the USA, it is not common to see margins of error cited for opinion polls. (Because opinion polls can influence election outcomes, I believe margin-of-error reporting should be not just common practice in Australia but legally mandatory with heavy penalties for failure to comply.)
For a 200-vote poll, by my calculations the margin of error is just over nine percentage points (an additional small error is introduced by rounding the results for popular consumption). Apply this margin to the seat-by-seat sample results and chalk up just nine seats to Labor, six to the Liberals, one very likely to the Greens, and the other nine are still up in the air. Of course, it’s extremely unlikely that a party would do several percent worse than predicted in several electorates at once, so more complex probability modelling could put a few more seats for each party in the “more or less certain” basket (without telling you where those seats might be), but with the difference between winning and losing a seat being typically only a few percentage points (if that) seat-by-seat projections based on 200 votes are not much better than useless.
The most spectacular demonstration of this was in Bass
Another problem with seat-by-seat projections in the Hare-Clark system is that even when you know the likely distribution of preferences of other parties, raw quota figures by party are still not always a reliable guide to who will win a close seat.
The most spectacular demonstration of this was in Bass in 2002 when the Liberals started 648 votes (0.067 of a quota or about one percentage point) behind Labor in the race for the final seat, yet won it by 577 votes. This happened largely because the critical votes for the ALP were over twice as prone to leak to another party or exhaust than were the critical votes for the Liberals.
Leakage can also be an issue if one party has far more votes that need to be distributed during the count (eg this stopped Labor having any chance of beating Peg Putt in 1998).
Another quirk of the system is that a party can win from an “inferior” quota position if enough of its candidates stay in the cut-up long enough. Suppose in a given seat the Greens have 0.70 quotas and the major parties 2.65 each (something that most journalists would call as an expected seat for the Greens), but each major party has one candidate who has reached a quota and two others left over, each of whom have more than 0.70 of a quota. The Green candidate is excluded and one of the major parties wins three seats despite starting 0.05 of a quota “behind”. (Something like this would have stopped Kim Booth from winning in Bass in 1998 even if he had not been passed by the other parties on Tasmania First preferences.)
Finally, while it’s all very well to say that a hung parliament would be the most likely outcome if an election was held today, any good modeller is aware of a thing called sensitivity analysis, which involves saying “suppose there is a slight shift in what I am modelling here, how does that affect the outcome?”
The Lennon government hasn’t been travelling well in recent months, but a statewide lift of just a few percent compared to the EMRS figures would result in it holding all its fourteen seats.
The Liberals, however, are nowhere near winning any more than eleven. If the opinion poll figures are similar a few months out from the next election, it will be easy for Labor to point out that only it is within striking distance of majority government -– a claim that would give it a massive advantage in the minds of undecided voters.
Only if the Liberals can bridge the gap in percentage support to a position where they are more or less even with the ALP (ie they can either boast a realistic chance of outright victory or claim that the ALP has none) will a hung parliament become overwhelmingly likely.
For the time being, if those predicting a hung parliament would like to put their money on it at odds short enough to match the confidence they display, I would be more than happy to take them up on the bet!
Kevin Bonham does lots of other things beside being a very experienced election scrutineer and psephelogical hack, but has wasted too much time on the above to bother to list them all right now.
Dr Kevin Bonham
May 28, 2005 at 12:28
POSTCRIPT: Wayne Crawford in the Saturday Mercury (extracted on TT: http://oldtt.pixelkey.biz/index.php/weblog/comments/what-absolute-nonsense/) writes “Ironically, it was the cut in the size of the parliament – an ill-considered attempt to get rid of the Greens by Tony Rundle’s minority Government in league with the Jim Bacon led opposition – that has made a hung parliament more likely by reducing the critical mass. If the results of the latest opinion poll were applied to the old 35-member House, Labor might win anything up to 20 seats”.
I’ll start with the claim that “If the results of the latest opinion poll were applied to the old 35-member House, Labor might win anything up to 20 seats.” This claim by Crawford is simply nonsense. To win 20 seats without getting more than 50% in any electorate Labor would need to win four seats in each – and four seats in Bass with 35% of the vote (not even three quotas since a quota was 12.5% in the 35-seat House) is impossible.
Indeed the only electorate where Labor would be likely to win four seats if the published results were applied to the 35-seat House would be Franklin, and that would be extremely close as they would start half a percent behind the Liberals and need to overhaul them on Green preferences. In Denison the Greens would be much closer to a second quota than Labor would be to a fourth, in Braddon the Liberals would have three quotas and the Greens one (rendering Labor’s seven-point lead over the Liberals useless), and in Lyons the Liberals would be much closer to their third quota and the Greens to their first than Labor would be to its fourth.
So the likely result in a 35-member house would be 16/13/6 or 15/14/6. To win an outright majority Labor would need to both win one very close contest (albeit in which it was precariously favoured) and win two more in which it was somewhat behind. Comparing this with the 25-seat scenario, in which Labor would need to win one contest in which it was narrowly favoured and only one in which it was somewhat behind, and we can see that not only would Labor be nowhere near Crawford’s 20 seats, but also it would be slightly less, not more, likely to win majority government than under the present system.
As for the general issue of whether the changes to the House have caused hung parliaments to be more or less likely, I previously discussed this issue (http://www.oldtt.pixelkey.biz/jurassic/bonhamlester.html) and what I said is just as true for Crawford trying to claim that the change makes hung parliaments more likely: “There is no “yes” or “no” answer to the question of whether the system reduces a minor party’s ability to win seats. Rather, it depends on that minor party’s precise support level.” A support level of around 17% is enough for the Greens to start winning seats under the 35-seat system that they would not win under the 25-seat system.