Thursday, March 28, 2013

Marginals Mayhem 2: Labor Losing The Lot?

 Advance Summary

1. A new JWS Research marginals robopoll has been reported as showing that Labor would lose all its 25 marginal seats, including Dobell.

2. The poll was partly commissioned by ECG Advisory Solutions, a firm with strong Liberal Party ties, but this was not disclosed in initial media reporting.

3. The poll uses demographic scaling to obtain a result from an on-average older voter pool.  However the breakdown by age of this poll is more credible than that of the previous poll.

4. Results of the poll are more Coalition-leaning than current national polling.  This may indicate inaccuracy, but could also be caused by an intensification of post-leadership-spill anti-Labor feeling, or by a very strong post-spill swing in Queensland.

5. The poll finds that 15% of intending Coalition voters (about 8% of the total sample) say they would support Labor if Kevin Rudd were Prime Minister.  However the claimed poll finding that voters want Labor to win is misleading.

6. The poll is consistent with many other polls that have suggested voters indicating a preference for Others in polling are often Rudd supporters and are opposed to both Julia Gillard and Tony Abbott.

7. The reported finding that Labor would lose all 25 marginals is unlikely to be correct as a few seats are at threshholds above or near the state swings and as swings are not uniform from seat to seat. 

8.  Nevertheless in an election "held now" Labor would lose most of its marginal seats and probably several others besides.


Internet searches for "JWS Research" following the new poll out today made my original Mayhem in the Marginals article on the January JWS marginals poll the most visited page on this site today.  The reason for that was that a new JWS Research poll has been reported showing the Liberals poised to win every single Labor marginal seat (25 including Dobell) plus perhaps up to another 15.

I'll start here with a brief description of electoral reality as it stood early this week.  No analyst really believes the 58:42 Newspoll in isolation tells it like it is, and the more benign polls from other pollsters suggested the Coalition's true polling position is really "only" somewhere in the 55s.  Bludgertrack (now with added LOESS) has it at 55.4, Mark the Ballot (assuming zero-sum polling effects) has 55.1 and my own quick and primitive version reckons 55.3.  All of these are affected by a degree of pre-spill ballast, so it's possible that the real picture in the aftermath of the spill shambles was more like high 55s, and perhaps it will get worse.

It's not rocket science that a government that went to an election polling a 2PP in the 44s would lose heavily.  Such a swing, uniformly distributed, would claim something like 20 to 25 seats leaving Labor with at most about a third of the House, and giving the Coalition about 100 seats.  Not every "marginal" (those seats requiring a 6% or less swing to fall) would actually fall on a swing of 6%, but those that did not would be counterbalanced by others on higher margins falling. 

Just as reporting of the previous JWS poll claimed that things were even worse than that in the marginals in both seat loss and 2PP terms, so this one claims that in "marginal" seats (those that were close last time) the Coalition is winning 59.4:40.6, for an average swing of a whopping 9.6%.  On this basis, the poll projected the loss of every one of the 25 marginals won by Labor at the last election.

Murky Origins

It was widely reported at the time of the previous JWS marginals poll that it was commissioned by ECG Advisory Solutions, which as my previous article noted, is a a corporate strategic advice unit boasting "unique access to the highest levels of government". ECG Advisory Solutions is a registered lobbyist in Victoria and, according to The Australian,  its principal operators are former advisers to former Liberal Treasurer Peter Costello (who is in turn associated with the group). One, David Gazzard, contested Eden-Monaro last election.

The initial burst of mainstream reporting did not hint at the poll having a commissioning source with strong links to the Liberal Party.  The Australian Financial Review initially reported the poll as exclusive to the Australian Financial Review, which was fair enough since its first release was, but only a later website edit added that the poll was "commissioned by The Australian Financial Review and ECG Advisory Solutions".  By that time the horse had bolted, with sources including, Herald Sun and others misreporting it as a poll for the AFR, rather than an AFR/ECG poll.  Thus it is not just a poll commissioned by a media outlet, but one commissioned by a media outlet in tandem with a firm with strong Liberal Party links.  This should be a cause for caution, as discussed in my previous article - or at least a cause for clear and immediate disclosure. 

Age Groups and Weighting

Full tables for the current JWS poll can be found here.

The current poll is the same as the last - a large robopoll of 4070 voters in 54 "key marginal seats", of which 29 are held by the Coalition, 24 by Labor and one by Craig Thomson.

The tables show that this poll shares a characteristic of the previous noted by Peter Brent.  The poll catches a very small sample of young voters and has to employ scaling to give their views proportional weight.  In the previous poll the sample had only 7.6% of respondents in the 18-34 age bracket, with 67% aged over 55.  The margin of error on the first group was very high, meaning that after scaling to make the results more representative, the poll was not as reliable as a representative sample of 3350 voters.  Brent suggested that the surprisingly high Coalition vote in the younger age bracket off a small sample size, magnified by the scaling, could have explained why the first poll was out of whack with other measures of the national 2PP at the time.

The current poll again has a very small sample of 18-34 year old voters, but it's a bit better this time around with 404 of them in total, or 9.9%.  And it seems that Peter's point about the previous poll is correct since on primary votes, in a poll showing a substantial swing against Labor since the last, the Labor vote this time is up 2.2 points, the Liberals down 0.7 points, the Greens down 2 points and Others up 1 point.

But if that explained some of the poll's departure from the national aggregated 2PP last time, what explains it this time?

There are a number of possibilities:

1. JWS has a moderate house effect in the Coalition's favour, the size of which is unknown because of a lack of comparable and regular benchmarking. 

2. The JWS poll was taken slightly later in the week than the other post-spill polls.  There may have been further drift away from Labor as the event and the post-spill recriminations were digested by the voters.

3.  The over-representation of Queensland seats in the national marginal seat picture (17/52 or 33% out of a total of 30/150 or 20% of all seats), could be coupled with a very strong Queensland reaction to the spill events last week (on account of Queensland being Kevin Rudd's home state.)  This isn't impossible since we have seen no polling from other agencies to contradict it, and what we have seen in Queensland ReachTEL polling lately is a strong resurgence by the Newman government to positions near its election result. 

Indeed the poll shows no further swing in NSW and Victoria, and the increased swing in WA since the previous poll is not by a statistically significant amount.  So it's really Queensland where the action is.

All up while there is always a bit of an X-Files aspect to any marginals poll that claims to be four 2PP points over the national aggregate,  I can see ways in which it may - really - not be that far over the top.

The Routine Findings

Most of the findings in the minor tables are no great surprise in the context of other recent polling.  Tony Abbott is preferred prime minister by eight points, but the design of the preferred PM question differs from others in giving the option of "Neither" as well as "Undecided".  Not surprisingly a quarter of voters for each party think Neither would make an excellent Prime Minister, and 47% of Greens voters and 78% of Others voters agree.  The massive figure for the Others vote is scarcely surprising because, as I've mentioned before, there is a lot of evidence that the Others category in recent polling is packed with Kevin Rudd fans who have deserted the ALP.

Gillard has a fairly benign netsat of -24 and on that basis it's not surprising to find Abbott soaring at -2.  Others voters can't stand either of them, giving Gillard a -34 thumbs down and Abbott -47.  Voters are split on the question of whether the crossbench should let the Gillard government enjoy its remaining days in the sun or dispose of it immediately, but the moderate support (20%) for the intermediate option of forcing Gillard to resign but permitting Labor to continue under someone else, not surprisingly, is most heavily supported by Others voters (33%). 

The Most Interesting Finding

The key finding of this whole poll, whatever might be concluded about a point or two of 2PP in its overall weightings, is to be found in Table 9.  Twenty-one percent of Coalition voters prefer a scenario under which Labor wins.


One would expect a trivial percentage of voters intending to support either party would want the other party to win.  For instance, they might think their local member had done a good job as part of a rotten government or Opposition.  But that doesn't explain the figure being anything like as high as it is. 

What does explain the figure is that there are many Coalition voters (15%) who say that their preferred outcome of the election is a Labor victory under Kevin Rudd.  It's tempting to suspect that some of these 15% are thinking tactically and hoping to use their answer to stir up further leadership tensions, while others among the 15% might change their mind if dealing with a Rudd Prime Ministership rather than a rose-tinted reflection of one.  But at least, the poll is not showing that all Coalition voters are holders of Coalition values (indeed 10% say they have Labor values) or convinced by the Coalition's argument.  The suggestion is that a substantial share of the Coalition vote comes from voters who would actually prefer to vote for Labor if only Labor would stop being such an internal disaster zone, or change its direction.

Laura Tingle puts this in an especially damning manner:

"The message is clear: even with these terrible polling numbers, the election might still be winnable if Labor had made a decision about its leadership based on the interests of the party rather than the self-interest of the small cabal that now runs it."

Note the implied use of past tense.  Of course, any consideration of whether a return to Rudd really would have produced a winnable position needs to also take into account the possibility of the government collapsing shortly after he was reintroduced. But if we continue to see polling as bad as the recent run, the question may end up being asked: would even that have been any worse?

I'm a little disappointed that although there is now no pressure for a change in Opposition leadership, the poll did not also include "Malcolm Turnbull lead Liberal-National Coalition to victory" in its list of options.  I believe that some of Labor's remaining vote share now is purely and simply an expression of utter opposition to Tony Abbott (such that some voters will cling to a Labor vote no matter what new calamity befalls the government simply because they view anything as better than Abbott) and it would have been nice to have Turnbull in the mix to get a real idea of how many such voters there are.  (My guess would be just a few points.) 

Overall the interpretation that 53% of voters say they want Labor to win is a furphy, since the voters who say they want Labor to win under Gillard and those who say they want Labor to win under Rudd cannot all have their wishes granted.  Moreover at least many of the latter (and anecdotally, I can add, some of the former) prefer a Coalition victory to Labor winning under their less-preferred Labor leader.

Would Labor Really Lose the Lot?

Reporting of the previous JWS poll was very much marred by the far from credible way in which results were presented, as discussed in the previous article.  Different ways of reading the results pointed to wildly different seat loss totals.  Some of them were presented prominently and taken seriously by the media although the poll results showed that they were not appropriate ways to project the results.  What about this one?

This time almost all the different means of projecting the swings recorded in the poll point in the same direction: every Labor marginal goes.  The one that doesn't is the state seat breakdown, which "only" gives away 23 of the 25 seats.  Not given away are Chisholm (Vic) which is over the state swing, and Lingiari (NT) which is not among the state breakdowns but would presumably fall.  However the evidence for giving away the WA seats of Perth (5.9%) and Fremantle (5.7%) on a batched sample with a swing of 7.8%, a sample which would only include about 60 or so voters in each seat and which has been affected by heavy demographic weighting, is so threadbare that to assume those seats would be "lost" in an election now is ridiculous.  The poll would be best seen as just showing that they are in play.

Given that distributions are never uniform (rather, seat swings are distributed about a uniform mean, so some swing much more than others, but overall it tends to cancel out), even on these figures it would therefore be realistically expected that Labor would save a few of its 25 marginals.  But not that many.  To hold any more than five on such figures would be very lucky.

And, of course, other seats would fall too.  The AFR list of seats at risk includes six seats in NSW , two in Tasmania and Griffith.   At the moment I suspect Kevin Rudd could run on a joint Citizens Electoral Council/Socialist Alliance ticket and still hold his seat, so I'm inclined to discount that one, but most of the others are seats that are already on the semi-usual-suspects list.

If this poll's finding about a massive swing in Queensland seats is correct, then I think it would be reasonable to extrapolate the poll as pointing to slightly greater likely seat losses than the uniform pendulum model - say, to the picture for Labor being four or five seats worse than uniform swing would predict.  But that's a very tenuous conclusion.  In either case, to assume that every single one of 25 Labor marginals would go based on such polling is not realistic.

The bigger picture is that if the party is losing dozens of seats then whether or not a few marginals are saved is not very relevant.  And you don't need a marginals poll to tell you that the loss of dozens of seats would be inevitable if Labor faced an election with current polling.

More reading

Also see Mumble's post on this one: Aggregate marginal polls: a bit silly

No comments:

Post a Comment