Tuesday, November 24, 2015

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 56% 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.

Monday, November 23, 2015

Wonk Central: Reverse Engineering Special Newspolls

Welcome back to Wonk Central, the occasional series of excursions into psephological arcanity that ... well, if you got past those big words and have some sort of head for maths, you'll probably be right at home here.  This one's not as hard as some, probably just a Wonk Factor 4/5.

Today saw the release of extensive results of a special Newspoll on various national security issues. These included:

* support for ground troops to fight the so-called Islamic so-called State
* how many Syrian refugees Australia should be taking
* whether priority should be given to Christian refugees over others
* the chance the so-called Islamic so-called State will carry out a large scale terror attack in Australia
* whether the Muslim community in Australia is doing enough to condemn attacks like the Paris attacks
* whether Muslims living in Australia are doing enough to integrate into something Newspoll calls "the Australian community"

I will comment on the results in the next Poll Roundup.  What I want to discuss now is whether we can use the results to guess at the voting intentions in this Newspoll.  My working will likely assist anyone seeking to do so in similar cases in the future.

Tuesday, November 17, 2015

Poll Roundup: Terror Bounce Kicking Labor When They're Down?

2PP Aggregate: 53.6 to Coalition (+0.8 in a week, +7.2 since Abbott was PM)
Coalition would win election "held now" with increased majority

Another three federal polls are out this week and the Coalition's aggregated polling lead just keeps on growing.  The relentless rise of the Turnbull regime has to peak sometime, but for another week the answer to the question "when?" is "not just yet."  There's now no sign on the smoothed tracking graph that it is even slowing down:

How Bad Is Bryan Green's Rating?

Last week I reported on the EMRS poll which has shown a probably Turnbull-led resurgence for Tasmania's Hodgman Liberal state government.  One figure has dominated discussion of the poll in the subsequent week:


19% is the preferred premier rating of Labor Opposition Leader Bryan Green (compared with Will Hodgman's 56).  I often rail here against the media overuse of preferred-leader stats to spin a yarn, and against polls that poll preferred-leader scores but don't also poll and release approval ratings.  I even have a piece here declaring such scores to be rubbish.  They're not totally meaningless, but they're messy indicators that are often biased to incumbents, they lag behind changes in approval rating, and they don't have a very good predictive record.

Part of the problem is that a preferred-leader score is a comparative indicator, so it's impossible to discuss what it says about one leader without thinking about what it says about another.  Does a big lead for an incumbent Premier say that voters really like the Premier and don't mind the Opposition Leader, or does it say that voters mildly like the Premier and can't stand his opponent?  EMRS have been polling the answers to these questions, but unfortunately they haven't been releasing the results.

As Matt Smith observes (in a notable piece that suggests Labor are disheartened and just going through the motions) numbers like this can spell a lot of trouble for a leader, and can ignite leadership speculation.  We shouldn't overstate the "trouble" angle; some state leaders have dragged on for years with miserable polling, but there have also been cases interstate where just one poll like this has been game over.

Sunday, November 15, 2015

Another Unsound Attack On Proposed Senate Reforms

Advance Summary

1. A recent article by a former NSW politician argues that proposed Senate reforms will create an effectively first-past-the-post system that advantages the Coalition and eliminates minor party candidates and independents.

2. The article exaggerates the impact of the proposed system on minor party candidates, since minor party candidates would have won at least three seats in 2013 under the proposed system.

3. While parties polling very low vote shares would not win without group ticket preferencing, this is not specifically because many votes would exhaust.  Rather, it is because strong preference flows between obscure parties would not exist even if all voters assigned their own preferences.

4. The article's claims about the impact of exhausting votes on preference transfers from Green to Labor and vice versa are undermined by those transfers being much less often important in Senate than in House elections.

5. The article's assumption that it would always be an advantage to run joint tickets rather than split tickets (eg for the Liberals and Nationals, or for Labor and the Greens) is incorrect.  Whether it would be better to run joint or split tickets would vary depending on party vote levels.

6. There is simply no reliable evidence that proposed reforms disadvantage any of the Coalition, Labor or the Greens, or any other force with serious support in any given state.  They disadvantage who they are designed to disadvantage: preference-harvesters.


Tuesday, November 10, 2015

Poll Roundup: Coalition Equals Term High

2PP Aggregate: 52.8 to Coalition (+0.1 in a week, +0.4 in two weeks)
Coalition would easily win election "held now"
(Article also includes comments on the voting age)

Eight weeks into the Turnbull Prime Ministership, we're still yet to see any lasting reversal in the polling trend towards the Coalition.  This week Newspoll, which in its new Galaxy-run incarnation has displayed a slight Labor lean, came out with a headline 53:47 in the Coalition's favour, the Coalition's best Newspoll 2PP since November 2013.  We also had a 53 last week and a 52 this week from Essential and a 55 last week from Morgan (56.5 by respondent preferences).

After considering the primary votes and Morgan's current house effect, I counted the Newspoll as 52.9 to Coalition, last week's Essential at 52.8, this week's Essential at 51.9 and last week's Morgan at 53.1.  This took my aggregate from 52.4 ("Newspoll Smells The Coffee") to 52.7 by the end of last week, and then to 52.8 now. On the assumption that no more polls are released this week, that will be the equal high for this term.  (Early in the Abbott era I did release a figure of 53.1 at one stage, but it was later revised after more accurate preferences were available).  The figure after Newspoll, 52.9, was also the equal highest point the aggregate has reached partway through a week in this term.  Here's the smoothed tracking graph:

Monday, November 9, 2015

EMRS: Liberals Rebound With Abbott Dead Weight Gone

EMRS: Liberal 48 Labor 25 Green 20 Ind 7
Interpretation: Liberal 49 Labor 29 Green 17 Other 5
Result of poll if election held now: Liberal Majority Government (probably 14-7-4)
Aggregate of all state polling: Lib 13 ALP 8 Green 4 (next most likely is 13-9-3)

A new EMRS poll of Tasmanian state voting intentions has been released and this one shows a massive eight-point shift to the Hodgman Liberal government since the previous poll in August. (Also see the trend tracker.) Really not a great deal has happened at state level since the August poll and I would interpret as much of that shift as isn't just random sample noise as being mainly down to federal factors.  The removal of Tony Abbott as Prime Minister has led to a general improvement in what little state Coalition polling has been seen around the country since the switch.  In the case of Tasmania, this is the Hodgman Government's best result from this pollster since a very similar result just after its election, and suggests it would retain its majority with one or maybe two seat losses.

EMRS polling tends to skew to the Greens and "Independent" and against Labor, so I interpret the poll as a little less dire to Labor than it seems.  All the same, the state swing is on paper only enough for Labor to pick up one seat in Braddon and one in Franklin.  In the case of the one in Franklin, a 1.3% swing is required, and while this poll points to about a 2% statewide swing, the Liberals would go into the election with three Franklin incumbents to Labor's one (compared with two each last time).  Thus I don't think Labor would regain that seat based on this poll.