Sunday, January 20, 2013

Silly Lilleys: Is Wayne Swan Losing His Seat? - Updated

Note: this is an older article that has been updated a number of times.  Everything below the line is not very relevant to the upcoming election following the change in Prime Minister, and with Swan now on the backbench, the answer for many will be "who cares?"

Update: 31 August: The great poetic irony that was supposed to occur if Kevin Rudd's resurrection as PM saved Swan's seat is again in danger, with the seat jittery based on seat margin polling, reports of internal polling and now a public poll.  A JWS Research robopoll reports a 53.8:46.2 result in this seat in favour of the Coalition, although it should be noted that seat polls have been delivering generally nasty stuff for Labor compared to national polling.  By no means a surefire indicator of defeat for Swan, but it is better not to poll such numbers than to poll them.


Advance summary of original article:

1. An opinion poll has been reported as showing the Treasurer, Wayne Swan, headed for defeat in his seat of Lilley.

2. While this finding may yet prove correct, there are many reasons to treat the finding cautiously.

3. These reasons include the likely closeness of the poll once preferences are considered, the poll's small sample size, the non-naming of the key candidates, and the poll being commissioned by an interest group.

4. With all these reasons considered, the poll provides no clear insight into the likely fate of Swan's seat, which is already known to be marginal and at substantial risk.

5. Polls commissioned by interest groups should be treated with much caution, because even if they are impeccably designed, groups may decide whether or not to release such polls depending on whether they like the results.

March Update: (More details at bottom of article.) Some similar comments apply to a second Lilley poll in early March, except that the key candidates are named, the sample size is larger, and this is the second released poll now claiming this result, so the level of overall doubt about the findings is lower.

The primary figures provided would produce a closer 2PP result than claimed.  The second poll's finding that a change to Rudd would result in an increase of about eight points is unreliable as polls in which voters anticipate how they would respond to hypothetical situations are generally not to be trusted.  Nonetheless a return to Kevin Rudd is capable of making a big difference to the chances of this seat being held. 

June 3 Update: Things are looking dodgier for the Treasurer now, considering the dire state of the national 2PP and the recent Galaxy Queensland poll which showed a state swing large enough (but only just) for the Treasurer to lose if uniformly applied.  An "internal poll" has been reported as showing him on a primary of just 28%, however the usual caveats apply to internal polling that has not been released in full detail and that may be being released selectively or be methodically suspect.

June 11 Update: However we now (see here) see the first publicly released polling result (Fairfax-commissioned ReachTEL) in which the Treasurer has been in front (53:47), although the lead in this sample is within its margin of error based on a sample size of 600.  This does not necessarily mean Swan has turned anything around; it is quite possible that all the published polls are within the neighbourhood of an actual picture in which the fight for the seat is very close.  Full details of the poll are yet to be released.

Original Text from January:

An AAP report today, published by a few newspaper sites, has declared the Treasurer, Wayne Swan, to be "on track" to lose his seat of Lilley at the next election.  The basis of the report is that a "ReachTEL telephone poll" (the report omits to note it is in fact an automated telephone poll, or "robopoll") conducted on Thursday night showed Swan trailing the LNP candidate, Rod McGarvie, by 38% to 45.2%.  The Herald-Sun caption screams "FEDERAL TREASURER WAYNE SWAN WOULD LOSE HIS SEAT IF AN ELECTION WAS HELD NOW, A POLL INDICATES".

That sounds pretty scary for the Treasurer.  But here are some details that should be considered before jumping to conclusions (as many from both sides of politics on social media sites and forums did) that the 2011 Euromoney Finance Minister of the Year is toast:

Firstly, the AAP report gave no indication of minor party votes or preferences.  In 2010 Wayne Swan in fact trailed McGarvie on primary votes, 41.22% to 41.09%, but won 53.18% to 46.82% after preferences.  Thus, 68.3% of the votes to minor candidates flowed to Swan.  A repeat of this outcome, as applied exactly to the ReachTEL figures, would result in a very close contest (50.5 to 49.5 in the LNP's favour).  Of course, without the breakdown of other votes it is hard to say if this is likely.

The  poll as promptly published (with a much more credible and cautious heading) on ReachTEL's site provides the missing primary breakdown, with the Greens on 7.5%, Katter's Australian Party on 6.2% and "Other" on a bearably low 3.1%.  The Green preferences, based on the 2010 election where they split 76:24, would reduce the 7.2 point gap in the poll to 3.3 points.  It's possible Green preferences will be even more pro-Labor in 2013, but probably not by much if so.

We do not know how KAP preferences would distribute at a federal election yet.  At the 2012 Queensland state election, KAP polled 5.2% in the matching electorate of Clayfield.  47% of the KAP preferences exhausted under Optional Preferential Voting, and those that did not split 46% LNP, 27% ALP, 28% GRN.  In the other matching electorate, Nudgee, KAP polled 6.8%, with 52% exhausting and the remainder splitting 48% LNP, 29% ALP, 23% GRN.  However, this was against the backdrop of a complete thrashing for Labor at that state election, and most likely in a federal election (if KAP did not preference either party) the split would be closer to even.

The drop in the polled Green vote compared to the election does suggest a worse preference distribution for Swan, so that a reasonable reading of the figures would be something like 52:48, or perhaps a little better, to LNP (a five-point swing). But even that's rather more competitive-sounding than just reporting the primary figures with a 7.2 point lead to the Opposition.

Secondly, the press report (as usual) gives no consideration to the margin of error.  Assuming that Swan was really on 48% 2PP off a sample of 511, the margin of error in that sample (Poll Cruncher is a great gadget for testing margins of error) is 4.3%.  If the sample was a fair one and was repeated at an election, we would strongly expect a swing against Swan, and there would be the possibility of a large one.  But there would also be a fair chance that Swan would hold his seat.  While it is better to be ahead off such a small sample than to trail, trailing slightly doesn't represent a clearly losing position.

Thirdly, the poll design as reported by ReachTEL on their site may slightly disadvantage the Treasurer in a poll of this sort by failing to name him, even though both he and McGarvie (also not named) are clearly known candidates.  The poll only names the options by party.  This contrasts strongly with some other recent ReachTEL polls such as Fisher, Sydney (state), Dobell, Denison and New England where expected candidates were named.  Of course, naming only one expected candidate (because the others are not known), as in the cases of Dobell and Denison, gives the sitting member an unfair advantage in the poll.  But in the case of Lilley, the two major party candidates are known, so this is not a great concern.

The omission of names for the major party candidates may matter because of Wayne Swan's personal vote.  Sitting members usually have positive personal votes, and one attempt by Peter Brent (Mumble) to measure them involves considering how well the sitting member polls in their seat, compared to how well their party polls in the Senate in their seat, adjusted for the party's average across the state.  On Mumble's measure, Swan's personal vote of 3.6% is only a middling one, but in the context of a poll pointing to a candidate polling 48% (2PP), the idea that name recognition might be worth a few points is a big one.  It's easy to overplay this point, though - firstly, only some voters would not think of Swan as their local member when they answered the poll question (so to a fair extent his personal vote would be built into the response), and secondly, Lilley is above the state average in terms of the suppression of the Green vote in the House of Reps compared to the Senate.  The second point suggests that a big part of Swan's personal vote is his ability to convince Green-leaning voters to vote 1 Labor instead of 1 Green 2 Labor, and that his personal vote in terms of competing with the LNP is at present modest.  (It also suggests that, given that Swan was not named in the poll, this is a rather sad poll result for the Greens.)

Finally, and perhaps this should have gone at the start,  the poll is not a neutrally-commissioned poll.  It is, rather, a poll commissioned by a union group called United Voice which has been campaigning to pressure the government on childcare.  According to the AAP article, the group is calling for " a cap on childcare fees and an increase in childcare workers' wages to overcome a predicted shortage of workers in Queensland."

Polls that are not neutrally-commissioned but rather are commissioned by groups with an interest in a certain outcome (or by parties) should always be treated with caution, especially when the result serves the commissioning group's purposes.  In this case the poll not only provides a sensational-sounding result that provides a big media hook for publicising the union's concerns, but it also enables the union to claim that Labor is suffering in the polls because of their pet issue (something that is almost certainly bulldust, but Labor operatives can be gullible creatures and have been known to fall for worse.)  Some interest groups even commission polls not because they actually want to know the results, but because they can use the results to put out press releases that gain publicity for their cause.

That doesn't necessarily mean there is anything really wrong with the poll.  It's just that had it shown the Treasurer coasting to victory 55:45, a group trying to put pressure on the Treasurer would never have let it see the light of day.  Even if this might be the only such poll this particular group has commissioned, we have no idea how many other such polls by various pollsters, commissioned by various issue groups, have been staged yet never been released.  As such, we have no idea whether this is the only such poll anyone has commissioned for the electorate of Lilley, or whether there have been others with different results that are not available.  It may be representative and groundbreaking, or rogue; we cannot tell.

(I have to say the poll's result on the commissioning group's core issue seems underwhelming.  A question like "Do you believe the Federal Government should do more to improve access to quality childcare?" is the kind of question that I'd normally expect to get a big thumbs up.  For starters it implies that the government is not doing enough at present and hence encourages a "yes" vote, secondly it is framed as a positive with no mention of the cost to taxpayers or other programs, and thirdly it's the kind of issue that voters might be expected to be either positive or apathetic about having provided.  A result of just 34.3% Yes 26.9% No really suggests to me that there are plenty of respondents who are accustomed to being cautious about the cost of proposals of this kind.)

What does this poll really tell us about Swan's prospects?

Not a great deal that was not already known.  We know Lilley is an uncomfortably marginal seat.  We know, from examples like John Howard losing Bennelong, that a long-term very high-profile government minister is not necessarily protected against swings, because their status is already factored into their result from the last election.  We have a fair idea that Queensland Labor voters tend to like Kevin Rudd and may not be all that impressed with some of the things that Mr Swan has said about him in putting Rudd's return ambitions back in their box.  We know that seat swings are variable and that a certain proportion of seats on 3%-ish margins can fall even if there is no overall swing.  We know that Labor is not yet getting a clear swing to it in Queensland, even if Campbell Newman is doing his best to create one (Bludgertrack modelled it at 1.4% against at the end of last year, albeit off a national 2PP that's about half a point worse than my current assessment).  Swan won't be out of the woods until his party is clearly set to gain substantially in his home state, while if we start to get robust evidence that he is going down, it may well be because the national picture gets even uglier for his party.  But at present, the evidence that he will lose, or will probably lose, as opposed to merely may lose, is not that strong.

I hope this article is useful as a guide to the level of caution that is often required in interpreting polls of this kind that are commissioned by interest groups.  If Swan loses, this poll will not be the reason why, and nor will the arguments advanced by United Voice.  Unless, perhaps, they campaign against him ...

Update 21/1/13: An article in The Australian today says "The key to the seat is the large number of undecided -- 19.1 per cent -- while 1.8 per cent of those polled said they had never heard of Mr Swan."  I believe this undecided figure is actually taken from the question about approval/disapproval of Mr Swan, which returned exactly that figure.  If so, (i) it does not reflect voters who are "undecided" about voting intention (ii) it is not an especially large figure, and indeed is only slightly higher than the undecided rates the same poll obtained for the two party leaders.  On this basis if, say, Newspoll conducted an approval survey of Mr Swan in his own electorate, the undecided rate would be much lower.


Update: Second Lilley Poll, 3 March

A second Lilley ReachTEL has been conducted, this time intriguingly commissioned by The Bolt Report (in other words, the usually but not always right-wing commentator Andrew Bolt, whose write-up of the poll appears here).  The primary results are remarkably close to the previous Lilley poll, with Labor down 0.6 points, the Coalition level, The Greens up one point, KAP down from 6.2 to 3.0 and Others up from 3.1 to 5.9. Julia Gillard's netsat (approval minus disapproval) is very slightly improved from -24.5 to -22.4, but Tony Abbott's is up substantially from -22.3 to -11.6.

This is claimed to result in a 54:46 2PP by 2010 election preferences, but I don't think that would happen on the primary figures stated.  In the same seat in 2010, Greens primary voters split 79:21 to ALP (the 74:26 I gave above includes votes received by the Greens from minor parties), Family First voters split 59:41 to LNP and voters for the lone independent split about evenly.  The KAP three-party preference split in Lilley at the state election was very similar to that for Family First in  .  There just doesn't seem to be any basis for assuming a greater than 60:40 split from KAP+others to LNP in this seat. Even assuming 60:40 (which is possibly excessive anyway), I get 52.3:47.7 to LNP.

Furthermore, the response to the question about how voters would vote if Kevin Rudd was leader, with half of the Others voters saying they would then vote Labor, suggests that the increase in voters picking Others isn't just direct flow from KAP to Other as a result of KAP being a rabble.  Rather, the Others column probably includes voters who are basically Labor-friendly but just don't like Julia Gillard and are therefore parking their vote in the Others column (although they would probably vote Labor or perhaps informally at an actual election.) There are, therefore, reasons to suspect that even 52.3:47.7 is a bit harsh on Labor, disregarding the possibility of KAP issuing a pro-LNP how to vote card, and that the correct reading of these figures is inside 52:48 to LNP.  So it is really a similar result to last time: Swan struggling but within the margin of error and hence not in a clearly hopeless position.  The conclusion drawn by Bolt as fact, that Swan would lose if an election was held now, is not so clear at all.  (I would argue that the margin of error can be effectively reduced if not for the issue with the previous poll not naming the candidates.)

An apparent difference between the two polls is that in this one the major candidates are named.  I suggested previously that Swan was slightly disadvantaged in the previous Lilley ReachTEL by the non-naming of the major candidates.  Given that there has been a general national move to the Coalition since the previous Lilley poll was taken, it is possible that the national trend and the slight change in polling method have more or less cancelled each other out.  Alternatively, neither of these factors makes a huge difference so it could be that any slight change in support has been drowned by random sampling noise. 

The poll not surprisingly asks the Rudd question and not surprisingly a move to Rudd is supposed to increase the Labor primary by 10.5 points, with the Coalition losing 5.1, the Greens 2, KAP 0.6 and Other 2.8.  This then is meant to result in an 8-point increase in Labor's 2PP to 54:46.  Again I think the Labor 2PP is too low (I get 55.2:44.8 again assuming 40:60 for KAP+Others, which in this case is probably a more realistic assumption). The finding of a supposed several-point increase in the case of a change to Rudd is very similar to that found by Galaxy for Queensland generally. 

As I have pointed out many times, poll questions that ask voters how they would behave in a hypothetical situation are not reliable.  Voters may answer dishonestly to try to stir up trouble, or they may simply fail to anticipate the impact of a change on the general political environment and how that would in turn impact on their own vote.  At best, these polls seem to do a good job of capturing the bounce effect resulting from a leadership change, but I do not believe they reliably predict long-term voting intention changes or election results.  Nonetheless, the basic finding Bolt is reporting from the poll - that Wayne Swan's chances would be boosted by having Rudd back as leader - may well have truth in it.  That is, assuming Swan would even wish to recontest if Rudd was restored.  More likely in that case he would retire at the election or resign soon after it. 

The commissioning source question is also a bugbear with this one.  Normally I trust mainstream media-commissioned polls that are conducted by regular polling agencies except where there are specific design flaws apparent.  But Bolt is at least as much agitator as reporter/commentator and hence occupies a twilight-zone position in this regard. That said, I don't see anything untrustworthy or suspicious about the primary vote readings in this poll.

Overall, my conclusion regarding this poll is similar.  We know that Lilley is a line-ball proposition barring clear evidence that Labor's vote is improving in Queensland.  We know that Labor's position has worsened nationally since the previous Lilley poll was taken.  It is fair to say that if Labor went to an election with a national 2PP as ghastly as the current level, they would probably narrowly lose Swan's seat, while if the 2PP improved to end-of-2012 levels, it would go back in the who-knows category.  It is to be expected, on that basis, that a poll would show Swan trailing slightly at present, and that is exactly what this poll, based on a realistic 2PP calculation, does. 

While Gillard remains Prime Minister, Lilley will probably be one of a number of marginal seats in which Labor's chances rise and fall with the state of the national 2PP.  For those interested, the Coalition are now slight favourites to win this seat on Sportsbet (1.70 vs 2.05), with Labor having been very slightly favoured when that market opened.

Addendum: the snarky marsupial has stated that the same broad picture is shown, unanimously, in other (unpublished) polls that he's aware of: "I just went back over Lilley. There's a few ReachTELs, some union polling and a lobby polling over the last 12 months."

However, as the snarky marsupial is employed by a union allied to the one whose commissioned poll I dissed in the original article, and as his comments were accompanied by some patently clueless personal attacks, I'm not sure I should trust that evidence! (Look who's snarky now!)

1 comment:

  1. Rod McGarvie has a good political pedigree and sounds like a nice bloke. He is the nephew of late Victorian Governor and one-time ALP member Dick McGarvie. Rod is of a different party to Dick, but that is not unusual in politics. The son of a New South Wales Labor Premier became an upper house member representing the Liberal Party.