Sunday, March 24, 2013

The Gillard Bounce Barely Existed - What About The Spill Cliff?

Advance Summary

1. Kevin Rudd was in an improving and apparently election-winning polling position when deposed as Prime Minister in 2010.

2. Although it is considered that Labor received a bounce upon Rudd's replacement by Julia Gillard, a published aggregation of polls shows that that bounce was actually small (about a point) and short-lived.

3. The recent Labor leadership spill (called but not contested) was another event expected to produce dramatic changes (this time negative) in the party's polling but the results of polls conducted since the spill have been mixed.

4. Two polls (Essential, Galaxy) have measured no meaningful change in voting intentions, one (Morgan) has returned a significant but not massive change and one (Newspoll) has recorded a very large change.

5. The six-point swing in Newspoll is probably produced by a combination of the swing, the previous Newspoll having been below trend, and this one being above trend.  

6. The current Newspoll is in many ways among the worst Newspolls for the Government and best for the Opposition in this term.

7. Continuing support for Kevin Rudd from voters open to voting for Labor, despite the issue being apparently internally resolved, is likely to present a problem in building enthusiasm for the party's struggling re-election attempt.


In 2010, psephology, of the pseudo- kind, was a big part of the public case for removing Kevin Rudd as Prime Minister.  Rudd's rapid fall from the glory days of 60-40 2PPs and 70% approval ratings to tossup 2PPs and negative netsats was one thing, but since that didn't quite seem to be doing the trick (as Labor was recovering) it was necessary to drag out "internal polling".  So internal poll results were leaked to Andrew Bolt, who never published the figures "on request", and this was used to argue that with Rudd as leader the party was headed for a net loss of about 19 seats and defeat.

It never passed the proverbial smell test in the first place, with claimed results in certain seats being ridiculed by experienced psephs, and the whole idea that Labor would lose with a 52-48ish lead being very far-fetched to begin with.  The fact that the details of the polling were not available for public scrutiny made it all the more suspicious.  Even after the election, had the internal poll results in these seats been robust, then releasing them would have done a great deal to shore up the case for having removed Rudd.  I strongly suspect the reason these poll results were never released in detail is that they would never have stood up to scrutiny.  In a rather hit-and-miss article I published at the time, I believe the hits included both my attack on the "internal polling" in question and the following:

" My own view, in hindsight, was that his autocratic style within the party made him so many enemies and so diminished the payoffs of factional influence, that once the chance of winning without him was seen as not much worse than the chance of winning with him, his troubles started.  A factional figure might rationally see an 85% chance of retaining office with a leader who respected and upheld factional processes, even from a different faction, as a better deal than an 90% chance of retaining office with a leader who sidelined the factions and diminished their influence.  Thus, to assume that Rudd would not have been rolled had he not been rightly seen by the party as headed for a likely defeat is unwarranted.  In the absence of firmer evidence that the SS Rudd was really going down with all hands on deck, I believe that Rudd was rolled not only out of fear that he would lose, but equally as much (by the primary instigators) out of fear that he would win, and further cement his presidential style as superior to the factional system that represents the traditional career path within the ALP. "

The 85% proved to be wildly optimistic for Labor (though they still "won" anyway) and I should have added the words "in public perception" after "superior" there, but this doesn't affect the general idea.  At the time the idea that Rudd was rolled because he would have won was probably seen as grassy-knoll material, but I these days it's tacitly (and sometimes even half-jokingly) conceded as true, hence the shift of the case for his removal to the many ways in which his rule was dysfunctional and annoying to his colleagues. 

In contrast, beyond:

(i) the many polls that claimed (unconvincingly) that Labor would receive a massive 2PP boost if Rudd was brought back
(ii) the various polls that showed Labor supporters split between the two options
(iii) the general backdrop of the government's shabby polling position, especially the 56-44 Nielsen

... there wasn't really a lot of chewy pseph to see in last week's arguments about whether or not to switch Prime Minister.  But one report was amusing.  Simon Benson revealed breathlessly that:

"CRITICAL internal Labor polling in South Australia, taken on the eve of Kevin Rudd's political assassination, revealed a rebound for the PM and suggested Labor could still have won an election in 2010 under his leadership."

The report actually, when stripped of its sensationalism, makes a molehill out of a mountain.  The withheld polling consisted of a ReachTEL of just a single seat, Kingston in South Australia (a marginal requiring a 4.5% swing to fall going into the election, eventually won by Labor with a 9.5% swing to the government!)  A turnaround in a single seat would have proved very little, and what's more, the point was already nailed to the wall by 2PP polling.  After a bad patch in late May 2010, in which most polls had Labor at 50-50 or somewhat worse, bad polls had become much scarcer through June.

Mark the Ballot has presented this very usefully in a Bayesian aggregate here (also see full post). Rudd is removed just before the end of the graph with the regression at about 52%.

Compare then the one for Labor under Gillard in the leadup to the election.  It starts (a week into her leadership) at 53.1%, fluctuates around the mid-52s to 53 for a while then declines to the eventual inconclusive result.

At that time I referred to an intial 2PP bounce of about two points for Gillard. It could have been argued to be more, given that the polls for which timely before-and-after data were available showed an average initial jump of about four.  But the MtB aggregate shows that in the case of Nielsen, the last Rudd poll was an outlier off the bottom of a trend that had improved by the time he was rolled, while the inaugural Gillard Newspoll (55-45), which I have often referred to, was also well above the multi-poll trend at that time, suggesting that Gillard's first Newspoll was a rather lucky sample.  The actual "benefit" of the leadership switch is shown as maxing out at a little over a point, and dropping back to half a point after only a couple of weeks - chickenfeed, and not necessarily even caused by the leadership change. 

Hype like this and praise for Gillard as a "red-headed Hawke" with great popular appeal was all part of a mountain of pseudo-psephological bulldust designed to justify a decision which had little if anything to do with electoral prospects.  We can only laugh now at mainstream articles that made a big deal of Gillard closing on Rudd as preferred PM in 2010, now that we have for some time seen polling showing leaders of both parties losing out as preferred PM to another from their own party (Abbott to Turnbull, Gillard to Rudd) by margins of around two to one - largely because voters for either major party will usually prefer that the incumbent leader of their opponent be replaced.

It is hard to see how anyone examining the evidence could rationally conclude Rudd was in a losing poll position when rolled, as was so widely argued at the time.  Any ALP figure who tries to claim that this was the case in future should be expelled from the party for stupidity.

Fast forward to 2013

Those who support Gillard over Rudd have probably despaired of this article already and concluded that I must be suffering from Stockholm syndrome as a result of shabby treatment at the hands of a prominent Rudd-booster.  But no, the above isn't intended at arguing any kind of case that Rudd should have been returned to the leadership last week (though for what it's worth I wanted it to happen if only to see what happened next!), and it's mainly thrown in because I think the MtB aggregations provide a nice bookend to an old debate that again became vaguely topical. 

At the end of last week we had another event that was supposed by some to have a massive influence on polling.  Following a week of intense leadership speculation by the press, reports finally emerged that indicated something was actually going on, with real names who were said to have withdrawn support for the PM being named.  It came to a head on the final sitting day of Parliament with former leader Simon Crean publicly calling for a leadership spill, which was at first refused then granted when a petition for a spill started circulating.  At no point did Kevin Rudd publicly state he would run for the leadership, but clearly support was being canvassed on his behalf.  However he was somewhere between marginally short and substantially short of having sufficient numbers to roll the PM (accounts vary).  While Rudd has maintained he would not challenge and would not contest (barring broad cross-factional support) his very late announcement that he was not running is being widely taken as evidence that if he had been assured of victory by even one vote he would have changed his mind and run. Debate between the Rudd camp and Simon Crean about what exactly happened has been going over the weekend.

There are two alternatives here.  Either Rudd would have broken his earlier commitment to not challenge (given the numbers) in which case his commitment now to never accept the leadership isn't necessarily trustworthy either, or else numbers were being counted for him by people who were making completely false assumptions about his nature.  Either Rudd is just another politician who plays the same leadership-denial game that Gillard played before removing him, or else Rudd's parliamentary support base is as shambolic as the shambles it was seeking to exploit and replace.  (It may even be that both of these things are true.)

With most of the devoted Rudd-backers, as well as the opportunistic Rudd-backer Simon Crean, now on the backbench, the upshot of the failed spill is that Gillard will go to the election (assuming she makes it that far) with a ministry that is almost completely dominated by her side of the party.  This probably doesn't make a huge difference to public perceptions of the core of the ministry, because the ministry was dominated by Gillard-backers anyway - it is not comparable to the problems Bob Hawke faced when Paul Keating went to the backbench and Hawke could not convincingly fill the gaping hole at Treasury. But it's not a good look to have a respected party elder like Crean and other talents on the backbench because he attempted to resolve the party's problems unsuccessfully.

Some believed the outcome of the failed spill - especially the impression of disorder and farce that it projected - would be extremely catastrophic for the party.  For instance, Rodney Cavalier predicted the Labor primary vote would crash to 23.  Others believed the spill would destory Rudd's public popularity and that Labor voters who have supported him would now see him as a coward and a wrecker and decide the party was right to remove him in the first place.

The first poll released after the spill, this Galaxy , shows hardly any reaction. The 2PP is unchanged from two weeks ago, primary vote movements are trivial, and Abbott has a narrow lead as preferred prime minister (a measure which greatly advantages the incumbent).  The PPM reading is odd in that the undecided figure of 30% is staggeringly high, apparently implying that over 60% of Greens/Others voters could not decide which they preferred out of Gillard and Abbott, a result I find very difficult to credit even with what I have said before about the Others column containing lots of Rudd fans.  However I can not find any other remotely recent Galaxy PPM reading between these two leaders to compare (Galaxy far more often polls PPM between prospective leaders of the same party.)

 There is a typo in the Labor leaders table (Coalition voters of course support Rudd over Gillard 56-18 not the other way around), and Labor voters in the sample are a little more friendly to Gillard cf. Rudd than in some other recent polls that showed Rudd in the lead.  But even with the spill finished, Labor voters are divided 46-43 on whether the spill was resolved the right way or the wrong way. And when presented with the obviously false dichotomy:

"Do you think that Julia Gillard has now confirmed that she is the only legitimate leader of the Labor Party or has she become a lame duck leader?"

...a very high 39% of Labor supporters agreed with the latter option (47% for the former.)  It's also notable that Labor fans were more supportive of the idea that Rudd had been honorable and true to his word (61%) compared to the idea that he was acting like a prima donna (29%), as compared to Coalition supporters (54-32).

So far then, it seems, the internal narrative that Rudd is "finished" and the leadership battle is over - mainly because Rudd is too risk-averse to be worth going out on a limb for in a challenge and the Gillard supporters will not let him have it on a plate - is a message that is not stopping many Labor supporters still wishing it were otherwise.  That's although there seems no reason to doubt that it is true.  Should Gillard's polling decline catastrophically to the point that the party thinks it is worth putting someone else in for a few months, the party might be inclined to turn to someone else, but if a heavy defeat was totally inevitable, change would seem pointless anyway.  If Labor-leaning voters continue to be unhappy with Labor's solution to the leadership mess, then even if this does not directly affect voting intention, it may well affect their enthusiasm for attempting to shift the opinions of others.

I am cautious about assuming that the other polls will follow Galaxy here in showing no net 2PP change, because Galaxy has often seemed a surprisingly static poll in the past.  There may also be delayed reaction to what has unfolded.  A move to the Coalition in the next Newspoll should be expected because the events that have occurred will not influence many voters in Labor's favour, and the previous poll was a three-point shift to a position well off-trend.  But unless that move is very large (to at least 56% 2PP for Coalition) then nothing should be read into it.

After the Newspoll arrives, I will update the 2001 comparison graph and attempt to determine whether or not Labor is still in a position from which there is historical precedent for recovery.

A few things have been said about that pet subject of historical recoveries from bad polling lately.  John Watson ventured forth with another essay on the hazards of trying to predict results from polling.  The crux of his argument is:

"Voters can and do change their minds in a matter of months. The Gillard government's current deficit - somewhere between 44-56 and 48-52, according to the Age/Nielsen poll and Newspoll - is not significantly different from the government being at least eight points behind in September 1995, 12 to 14 points behind in May 2001, six to 10 points behind in April 2004 and 16 points behind in May 2007 (even then, Labor won only eight seats more than a majority in November).

It was not a complete surprise that the Howard government recovered from a four-point deficit in April 1998, albeit with a minority of the vote, but it was in 2001. Nor did anyone in late 1992 predict the Keating government's return."

All this is fine as an argument that declaring defeat certain based on polling remains premature (as it often is right up until election week) but it is not a valid argument against the view that polling shows a Labor recovery to be quite improbable.  To round up the usual suspects above - the governments in 1995-6 and 2007 lost the governments in 1998 and 1992-3 were not so far behind, and at least according to Newspoll, Howard was over the worst by April 2004, though his position was not much better than in 2001.

Secondly John Stirton of Nielsen (I'll add the link when I find it again; may have been paywalled) has pointed out that governments that have lost elections have trailed for over two-thirds of the Nielsen polls in their term, and that this government has trailed in every single Nielsen poll since the 2010 election.  Going back further, the old Morgan Gallup examples of Chifley in 1949 (competitive for most of the term but fell over in the last few months) and Menzies in 1954 (trailing almost the whole term, at times by massive margins) show that the link between the proportion of term a party trails for and the actual result is not that watertight.  This is to be expected because polls taken well out from elections are much less predictive than those close to it, and most of a parliamentary term is well out from an election in that context.

Lastly, a word about betting.  Betting markets went absolutely nutty on the day of the spill, with Rudd's price to be leader at the election first lengthening, then shortening dramatically following the Crean intervention, then blowing out when he announced he wouldn't run.  See the graph here.  At one point it was possible to  get $6 on Gillard winning the afternoon's spill even though no-one knew whether Rudd would stand.  Gillard is now at $1.20 to lead the party to the election and I won't be surprised if this shortens further.

Similar crazy moves were seen in betting on who would form government after the 2010 election.  Whatever might be said about betting as a predictor (or not) of election outcomes it is clearly very unreliable in predicting the outcomes of events that depend upon party insider decisions.

This piece will be updated over the next few days.

UPDATE (26 Mar):  Four polls are in now with two (Essential, Galaxy) showing no 2PP change and one (Morgan multi-mode) moving two points, while Newspoll moved ... six!  Of course, not all of that six point move was real since (as mentioned above) about two points were down to the previous Newspoll being well below trend.  It would not be surprising if at least a point or two of the whopper surge was down to the same thing again in the opposite direction.  The general trend of the polls other than Newspoll was that any votes leaving Labor were going to the Greens or Ind/Others, with Newspoll delivering an increase in the Coalition primary to a whopping 50.

Newspoll's poll-to-poll bounciness has attracted a fair bit of flak in recent months.  For a while it's been possible to argue that it hasn't been too much bouncier than the Newspoll long-term average of a bit under two points, but in the last eleven Newspolls, the average poll-to-poll bounce has been 3.2 points.  At some stage soon I may check whether this is actually statistically suspicious.

Here's the Newspoll rolling average graph updated

 Following the spill that wasn't, the Newspoll rolling average comparison shows Labor as now being in a worse position than the Howard government was not only at an equivalent stage but indeed at any stage in 2001.   Of course the double-weighting of the 58 plays a part here.  But on Newspoll data Labor has fallen off the cliff at exactly the same time as the Howard government made one of its most significant improvements.

It's also no cause for comfort for Labor that the two polls that showed no 2PP change were the two with the smallest post-spill sample size.  (Close to half of Essential's sample is pre-spill).  My own aggregate attempt has only gone to 55.3 but a minority of its input is still pre-spill data, which will wash out over coming weeks.

There are many notable statistical features of the current Newspoll:

* The six-point jump is only the 12th case of a government falling by six points or more in a single Newspoll.  The previous such cases were:

Sep 1986 (Hawke): -6, apparent rogue
May 1989 (Hawke): -6, previous poll was apparent rogue
Nov 1991 (Hawke): -6, Hawke leadership terminal
Aug 1993 (Keating): -9 (record), re-elected Keating government honeymoon period over
May 1994 (Keating): -6, Opposition bounce for Downer replacing Hewson
Sep 1994 (Keating): -8, previous poll was apparent rogue 
Feb 1995 (Keating): -7, Opposition bounce for Howard replacing Downer
Oct-Nov 1998 (Howard): -6, apparent rogue
Mar 2006 (Howard): -6, previous poll was apparent rogue
Oct-Nov 2009 (Rudd): -7, apparent rogue
April 2011 (Gillard): -6, previous poll was apparent rogue against general backdrop of government declining following carbon-pricing decision

So moves of this size against a government from poll to poll usually involve rogues on one side or other, as even major political events do not normally cause six-point swings in two weeks.  In this case we have the cocktail of the previous poll being roguish, a major event, and the probability of overkill in the opposite direction.

* Julia Gillard's approval rating of 26 is her equal second-worst (she polled 23 in Sep 2011 and 26 in Feb 2012).

* Gillard's disapproval rating of 65 is her second-worst (she polled 68 in Sep 2011).

* Gillard's net satisfaction of -39 is her second-worst (she polled -45 in Sep 2011).

* Abbott's approval rating of 39 is his best since polling the same in Sep 2011.  He last polled over 39 in July 2011.

* Abbott's disapproval rating of 50 is his best since July 2011.

* Abbott's net satisfaction of -11 is his best since the -7 in July 2011. 

* Abbott has now recovered 25 net satisfaction points in four and a half months.  This already is tied with John Hewson (1990-1993, who lost) as the equal second-best netsat recovery by an Opposition Leader in a term in polling history. For the full list see this now rather dated piece here

Indeed, Abbott is now only two points from recording a netsat better than -10 before the election, something which I (unwisely underestimating Labor's ability to return to self-destruction) considered to be a less than 20% probability of occurring in that review of Opposition Leader recoveries, written during the ancient geological epoch known as the Gillard Recovery.  

* Abbott's preferred prime-minister lead over Gillard of eight points, on an indicator which heavily favours incumbents, is his highest since Sep 2011 and his second highest ever.

* In previous articles I have mentioned the concept of a "disconnect score".  The disconnect score measures the minimum proportion of voters who are supporting a party while dissatisfied with its leader.  The formula is:

Disconnect = max(0,(Party2PP+Leader disapproval - 100).

Abbott's last two readings for this score have been 7 and 8.  His current rolling average is nine, which is the lowest it has been since August 2011. 

* And lastly, the 58-42 is the widest gap between the parties in one poll since April last year.

Yes, the Newspoll is probably over the top, but a number of these figures reflect trends that have been developing in any case.

The rapid improvement in Abbott's standing is not just because Labor's infighting has taken the spotlight and the pressure off him and memories of why people disliked him so much last year have to some degree faded.  Nor is it just because Abbott has been more disciplined, and at times reserved, during this year so far (the polling of late last-year having been an obvious wake-up call).  In my view the main thing going on here is that the more disorderly the government gets, the more voters are starting to think that Abbott's relentless negativity towards it was actually to some degree justified.  

It's very well known that in politics, "disunity is death", and this three-word saying will be used after the election to say that it should have been completely obvious to everyone what was coming all along, as with the falls in the distant past of the Bruce, Scullin and first Menzies governments. But this particular episode of disunity was opinion-poll-driven, through bad polling convincing many in the Caucus that Labor was lost under Gillard, and the rapid fall back into a polling position bad enough to drive this was avoidable. 

It's hard to see any reason why Labor should now have a realistic chance of winning the election, since Abbott's unpopularity and the government's generally reasonable economic record are trivial compared to the many reasons why Labor is so unpopular.  In a lot of poll analysis for the rest of this year the question may therefore be not the result but just how big a thumping it will be.  On this, just as I urged caution against premature writing off of the government's prospects based on polling alone (while it was still premature!) I now suggest that it would be foolish to try putting realistic limits on the amount by which the government might lose.

Further Update: (2 April): It's got still worse for Labor with a two-point drop in this week's Essential (apparently, the previous week's sample was also bad but was muted by the rolling average) and a 1.5 point drop in this week's Morgan multi-mode.  My aggregate has gone to 56.0 and Mark the Ballot's Bayesian model is at 56.1.  I am going to review my treatment of house effects shortly but the new model I may implement will only make a very small difference compared with what I've been doing.

This week's Essential also contains some attribute polling that is dire for Gillard and Labor.  Bizarre results, such as the PM being perceived by the electorate as more aggressive than Tony Abbott, suggest that voters are attributing negatives to Julia Gillard whether they are justified or not simply because they dislike her and think she should be replaced.

Labor has had a (partly self-inflicted) problem all term with obtaining credit for its positives from the electorate.  It seems that now the party's position is so awful that not only does the party not get full value for its positives but that the electorate are willing to concoct negatives unprompted irrespective of whether they exist.

A Week Later: (9 April):  Not much better for Labor this week with Essential and Morgan at 56 and Newspoll at 55.  The bounce back in Newspoll was utterly unremarkable since the last Newspoll at 58 was over the top.  My aggregate, which now uses a global house effect correction rather than mucking about with house effects for specific polls, has come down but only slightly, to 55.6.  The rolling Newspoll average is 55 but I haven't bothered updating the 2001 chart yet.  Actually there is some reader interest in a 2007 comparison chart and I might have a go at that next time.

Most notably this week's Newspoll saw Tony Abbott lose all eight points of net satisfaction that he gained last time, putting him back to -19 compared to Gillard's -34.  The last six Newspolls have been 51 (under trend), 56 (over trend), 55, 52 (under trend), 58 (over trend) and 55.  Comparing this 55 to the last 55, which was also the only one not to cop flak for bouncing, Gillard's netsat is down six points, Abbott's is up three, and Gillard's PPM is up one point.  So against the same set of Newspoll voting intentions before and after the abortive Labor spill, the impact on leadership impressions has not been that great.  This may well be because negative impressions of Gillard were already heavily embedded in both the bad pre-spill 2PP results and Gillard's pre-spill approval ratings. 

I note with no particular comment the passing of Margaret Thatcher, except to note that in 1980 she recorded an approval rating of 23% ...  the same as Gillard's all time low. 

No comments:

Post a Comment

The comment system is unreliable. If you cannot submit comments you can email me a comment (via email link in profile) - email must be entitled: Comment for publication, followed by the name of the article you wish to comment on. Comments are accepted in full or not at all. Comments will be published under the name the email is sent from unless an alias is clearly requested and stated. If you submit a comment which is not accepted within a few days you can also email me and I will check if it has been received.