Poll Roundup: Turnbull Surge Alone Can't Explain Shorten's Problems

2PP Aggregate: 53.4 to Coalition (-0.2 in a week)
Coalition would easily win election "held now", probably with increased majority

(Updated on Friday to 53.8 to Coalition, then went to 54.0 to Coalition on weekly reset)

There are only two new federal polls and one state poll of federal voting intention to add so far this week, but there is still quite a lot to discuss.

Last week there was a sign of a possible Paris-attacks surge to the government in the 56:44 result from Ipsos, but this wasn't repeated by either Morgan or Essential.  The lack of replication from those two didn't mean a lot because Essential doesn't do dynamism and Morgan's behaviour under Turnbull has been strange, but this week Newspoll didn't play ball with Ipsos either.  I'm still inclined to wait to see what ReachTEL says before completely discounting it, but it looks likely that there wasn't really a Paris attack bounce in 2PP polling, and the Ipsos sample just had a couple of extra points of sample noise for the Coalition.  This week's Newspoll at 53:47 (which I aggregated as exactly that) and Essential at 52:48 (which I counted as 51.7 considering the primaries) have so far knocked two-tenths of a point off last week's result.  The smoothed tracking graph, however, does not yet show the surge as having peaked.

I add that the behaviour of Essential, which I have as having a Turnbull-era skew so far of 0.75 points to ALP (it would be more in other aggregates), is under investigation by the stewards.  A house effect correction may be applied if this continues, though I'm reluctant to do that given that I haven't corrected other Essential excursions in this term.

Australian federal leadership change polling surges, excepting those that happen just before elections, usually take between three and eight months to peak.

The other poll out this week so far has been a Queensland-only Galaxy of federal voting intention with a result of 58:42 to the Coalition.  This result was so bad for Labor (especially given the "Queensland strategy" touted as an antidote to Malcolm Turnbull) that even the Courier-Mail failed to make it sound worse than it actually was.  That said in 2013 the Coalition beat its national average by 3.5 points in Queensland, even against a home-state PM, so the Galaxy result is nothing all that strange.

Shorten's 15% Shocker

The most remarked-upon poll figure of the week was Bill Shorten's 15% better Prime Minister score in Newspoll (he trails Turnbull in the "beauty contest" by a mere 49 points).  This is the worst figure for an Opposition Leader since Turnbull himself held the position, and one point above the all-time worst for a Labor leader, set by Simon Crean in 2003.  Both Turnbull and Crean were booted immediately after their 14% results.

The worst better-PM figures of all time, however, were set by Brendan Nelson, who averaged 12% over 16 Newspolls with a low of 7% and a high of 17%.  Nelson was on a hiding to nothing as a new Opposition Leader facing an extremely popular new PM, so his results were tolerated for a while.  So why is Shorten now polling figures that, Nelson aside, were previously polled only by leaders who were circling the drain, or at state level, almost entirely by failures?

The line being run by the ALP is that Shorten is struggling because Turnbull is supremely popular both as a new PM who replaced the much-disliked Abbott, and in the wake of the Paris attacks.  But this does not hold up.  Kim Beazley went no lower than 25% in the aftermath of the S11 attacks in 2001, although John Howard's netsats then were almost as high as Turnbull's and the Coalition's 2PP lead was much greater.  Even Crean was not doing quite this badly (18%) following the Bali bombings.

The Better PM score of an Opposition Leader correlates strongly with their party's 2PP voting intention and with the Prime Minister's net rating, but only weakly with the Opposition Leader's net rating.  If I try to explain Shorten's scores using only the PM's ratings and the 2PP, I get this linear regression:

Better PM (LO) = 1.159*2PP(Opposition)-0.1351*PM Netsat-26.716  (+/-4.76)

For where Shorten is at the moment, this gives an expected value of 22.6.  His actual value of 15% is 7.6 points below where he "should" be, and he's been about that far below the prediction in every poll since Turnbull became PM.  In contrast, Shorten's preferred-PM scores when Abbott was PM were generally a couple of points higher than expected.

However, if we throw Opposition Leader netsats into the mix, this is the regression:

Better PM (LO) = 0.604*2PP (Opposition)-0.244*PM Netsat+0.1908*Opp Ldr Netsat+1.516 (+/-3.34)

Adding the third term makes the equation a lot more predictive - it now predicts 85% of variation, up from 69%.  And this time the projection for Shorten (14.7) is bang on, as it has been on average in the five Turnbull-PM polls so far.

So all Bill Shorten's lousy Better PM score is really telling us is that he is an unpopular Opposition Leader facing a popular PM, and that Labor are well behind on 2PP.  Shorten's Better PM scores were actually more informative when Abbott was leader, since he was punching above his weight by a few points at that time, even when unpopular.  (I suspect Abbott had more to do with this than Shorten did.)

Turnbull's net satisfaction reached a new high of +38 (60-22) this week while Shorten's was at -31 (27-58); the 69 point Newspoll gap between the two is a new record, breaking the one set four weeks ago.  The Galaxy Queensland-only sample had Turnbull favoured over Shorten 65-14 on the indicator of "having the best plan for Queensland".

Really bad preferred-leader scores often ignite leadership talk, which can make them self-fulfilling prophecies as predictors of leadership change.  But a party willing to ignore the talk could well argue that Opposition Leader personal ratings have very little to do with 2PP, so if both Turnbull's high personal ratings and the government's high 2PP are just results of the bounce for the change to a new PM, then this might all wear off next year and the next election can still be competitive.  Yes, even if Shorten keeps polling -30 netsats.

From that perspective it doesn't matter that Shorten is unpopular and that any alternative leader would be much more popular - what matters is whether any other leader would use the leadership to do a better job of convincing voters not to vote for Turnbull.  This is actually a very hard thing to measure using polling.

As for why Shorten is unpopular, in my view voters don't intensely dislike him, but rather find him boring and lacklustre and too much the party/union hack.  As mentioned before, left-wing voters also wish he was more like them.  There is a lot of Simon Crean in Shorten's problems, and in polling terms just check this out:

Bill Shorten Is Simon Crean

This graph shows the netsat trajectories of Simon Crean and Bill Shorten as the career of each progressed through the Newspoll series.  Note that there is variation in how often Newspolls were taken, and gaps for holidays and so on.  The similarity in career netsat paths is remarkably strong.  In polling terms, assuming Bill Shorten is in fact a slightly less bad version of Simon Crean explains 77% of variation in Bill Shorten!

Other Polling

There is quite a lot of national security type polling about, unsurprisingly, and not a great deal of anything else.  Yesterday I had a go at using the special Newspoll data to reverse engineer the votes in Newspoll; this attempt failed, but it does appear that Others voters in this Newspoll tended to have right-wing views on every question canvassed (in many cases probably to the right of the Coalition).  The Newspoll, not surprisingly, shows that voters are easily scared by terror attacks, with three-quarters considering a large-scale attack within Australia inevitable, likely or very likely.  Expectations would have been similar after September 11 but fourteen years later Al Qaeda has not inspired major attacks on Australian soil.

Pretty much any terror-related question has a predictable partisan skew with Coalition supporters more afraid of attacks and more hawkish and Greens at the other end.  Essential also shows that more left-wing voters are more likely to blame Western interventions in the Middle East for the current situation.

I will update this piece should a ReachTEL arrive at the end of this week.  Otherwise I expect the next volume to follow in two weeks' time.

Friday: ReachTEL Update

Well that's rather interesting, as ReachTEL has partly agreed with Ipsos, coming out with a 55-45 to Coalition result, which I'm aggregating at 55.3 after considering the primaries.  That takes the Coalition out to 53.8% at the end of the week, another term high.  The weekly reset takes them to 54.0 as of midnight, and some others will be higher (Andrew Catsaras has the same).  If renewed scrutiny of Special Minister of State Mal Brough over the now ancient Ashby/Slipper affair is having any impact on voting intentions, then this poll shows no sign of it.  It's not quite the size of jump seen in Ipsos, but it's not far short of it.

In results released from the poll so far, Turnbull's netsat is up 1.9 points to +33.4, and in the five weeks since the last poll the proportion finding his performance poor or very poor is down from 16.1 to 14.8%, so this is not just undecideds moving into the positive column.  Shorten is down 3.5 points to -26.9, his second worst so far, and his "very good" rating has dropped from 8.7 to a pitiful 6.4  The ReachTEL preferred-PM method gives Shorten more leeway than others but even then he's still thumped 71.3-28.7.

In an amusing matchup, voters overwhelmingly feel "safer" on national security under Turnbull than Abbott (74:26); this applies across all parties but slightly more strongly on the left.  I'm not entirely sure that Abbott being stronger on national security than Turnbull was really a thing (Abbott being stronger on it than Shorten most certainly was).  If it was then there are many different ways of asking the question, and perhaps "stronger" on national security would have been an interesting comparison.  (A perception was that Abbott would stand up for the country but might go overboard.)

Another poll out today was a Morgan SMS poll on sending ground troops into Syria and Iraq, which got a much more negative result (39% approve 61% opposed) than other such polls.  This might be a reflection of PM Turnbull's dismissal of the idea sinking in, but it could also be a reflection of the issues with this polling method.  Again there was no 2PP released to benchmark it so it should probably just be ignored.

There were some comments on the Shorten ratings issues in a generally good piece by Peter Brent (Mumble: Bill Shorten Newspoll not the worst by any stretch) that I should debate.  Peter mentions the case of John Howard in 1987 as one in which a hapless Opposition Leader polling dire better PM scores and in a terrible 2PP position recovered to a pretty close defeat.  However, the 1987 election was called early by Hawke to try to exploit opposition infighting - effectively an attempt to "bounce" an election, which often doesn't work too well.  (Alan Carpenter in WA tried the same trick and even lost!) As such its value as an example is debatable.

I'm a moderate sceptic of the Mumble theory that for a specific set of voting intentions, it is better to have poor leader ratings than sky-high ones.  I think it is very difficult to distinguish evidence for this claim from evidence for the more straightforward view that large 2PP leads just tend to contract close to polling day whatever the leader ratings (and large 2PP leads tend to go hand in hand with high leader ratings).  It's also important to determine whether cases like 1987, 1990, 2004 and 2007 are really just caused by the mysterious Labor Fail Factor (the tendency of the Coalition to outperform its pre-election polling more often than Labor does.)

Anyway, if we go into the campaign with the Coalition leading 54:46 and leader ratings the way they are now, Labor might pick up three or four points and make a close thing of it, but it's extremely unlikely that they could pick up six and win.