Saturday, September 19, 2015

Canning: The Anticlimax Live (Plus Post-Count)

CANNING (Lib 11.8%) - called, Hastie (Lib) retain.


On-the-day votes and most prepolls counted.  Current 2PP 55.3% (-6.5% swing) to Liberal. Unlikely to change greatly after remaining postals and other votes.


Thursday 1st Oct: There has been very little movement in the 2PP as remaining votes are added (I believe there are now only a few hundred left).  The major interest now is in the internal polling claims mentioned below.

We finally have more detail of these in a report from The West Australian, that says not only that the Liberal Party had Hastie on 57% the weekend before Tony Abbott was rolled, but also that Liberal tracking throughout never dipped below "about 53-47".  However, if the internal poll series really included both a 53 and a 57 then the most likely explanation is random noise, rather than voting intention actually changing that much.  Usually internal polling is less reliable than public polling, but public seatpolling has been so bad in recent years that it's hard to dismiss competing internal poll reports entirely.

Thursday: There hasn't been a lot to add here, but 755 prepolls split 58.4% to Liberal (a 7.5% swing). Although this is a point larger than the overall swing, it's not especially meaningful as the number of non-PPVC prepolls is much smaller than at the general election.

Tuesday:  Not much extra progress to report in the postcount.  However an article in The Guardian incorrectly reports that postal votes are showing only a 2.3% swing to Labor.  This figure is derived by calculating the swing on postals from the 2013 overall result for the electorate, and ignores the well-known trend that postals favour the conservatives.  The correct baseline for calculating the swing on postals is the result on postals last time, ie 65.1%.

The other thing to bear in mind is that the postals added so far will mostly include the early postals, and while these are more likely to have been cast in the Abbott era, they are also more likely to have been cast by slightly more conservative voters. All the same by whatever measure the impact of the leadership change on the by-election was small: my reading of the polls is that they were pointing to it being only a couple of points overall, and such an impact (if it happened) can easily be lost in the noise of shifting voting method patterns or strengths of postal campaigns in an actual election.

Sunday 8:40 pm: Still more postals added and the total break in postals is now 59.5% to Liberal, meaning that the swing on postals (5.6%) is still slightly less than the swing overall (6.4%).  The margin now won't change much from here at all as there are not that many votes left to add, a few thousand at most.  It's going to be pretty much exactly the average by-election swing under the circumstances (whether measured with or without reference to the state of federal polling).  Meanwhile in the minor party race the Sustainable Population Party looks to have joined the Liberal Democrats in suffering the indignity of being beaten by Teresa van Lieshout.

Sunday 5:40 pm: A big bunch of postals, 4751 of them formal, have been added, breaking 58.8% to Hastie, a 6.3% swing.  This has increased the overall Liberal 2PP to 55.2% (6.6%).  If the remaining postals break that way then the overall swing will come down a little more and the swing in postals will end up virtually the same as the booth swing.  The Liberals did make a strong postal effort but this is more evidence that the impact of the leadership change on the vote was not massive.  Indeed it's another way in which the election results so far don't reflect the change at all.

I've also had a look at what we might draw from polling by way of evidence about the impact of the leadership change.  The difficulty there is that only ReachTEL polled after the change, and that only twice, with one of the questions a hypothetical on the night of the spill.  Based on the Pollbludger national last-election preferences for the various polls, the Abbott-era ReachTELs were on average 2.9 points stronger for the Coalition than the average of all other polls (54.7 vs 51.8).  The change in the Turnbull-era ReachTELs averaged 1.8 2PP points.  The appearance of a 2PP bounce of five points in the ReachTELs was largely because the early commissioned ReachTELs had extremely weak respondent-allocated preference flows, that would have never actually happened.  How much weaker preference flows would have been under Abbott we will probably never know, but the polling suggests the change made at most 3 points difference, and probably less.

2PP Booth Figures Up: The AEC now has 2PP booth figures up and these do provide some succour for the view that the change in federal leadership made little difference, because the swings at the Mandurah and Armadale PPVCs are not greatly different from the swings for each city.  However I do not know how much effort each party was making on prepolls compared to in 2013, and massive increase in PPVC voting means that inferences that the same sorts of people are voting at them are not reliable.

Sunday: Internal Polling Rumours: There are various reports (most prominently by Andrew Bolt) that internal polling showed the Liberals winning by more under Tony Abbott than it seems they have actually done.  Bolt has published an unsourced claim (attributed to "I'm told authoritatively") that internal polling under Abbott was showing a 57:43 win.

These claims first began appearing on Twitter shortly after Abbott was challenged, but the initial claim was that internal polling with Abbott showed a 55:45 win. This was the most commonly claimed margin on Twitter after the spill with some tweets claiming 54 and some 56.  None of these sources were linked to the press or cited primary figures, they were all just of the form "I have heard".  What is interesting is that if internal polling showed a comfortable hold under Abbott, no-one referred to this before Abbott was challenged.

Bolt's 57:43 under Abbott was not one of the Twitter-rumoured figures until Bolt posted it on election morning.  A 57:43 figure was referred to by journalist Joe Spagnolo, but with no sign it was under Abbott.

Bolt also misrepresented William Bowe (who he called a "pollster") as "tipping" a 55:45 margin when in fact Bowe was saying that if the primaries reported by ReachTEL in one of their early polls were correct, 55:45 would be about the result, because the preference flows in some of the early commissioned ReachTELs were not believable.  An interpretation of what a poll shows and a prediction are not the same thing.

Andrew Bolt is not a cautious or reliable source; the entire 18C furore being sparked because of Bolt's spectacular sloppiness in the handling of facts that got in the way of his political sensationalism.  With so many people with agendas in defending the old or the new leader, I do not believe we will ever reliably get to the bottom of what internal Liberal polling was saying about the likely result under Abbott - and nor that it matters all that much, because internal polling is quite unreliable anyway.

The relationships between federal and by-election polling suggest that making Turnbull leader should have only boosted the result by about 2.4 points.  If the shift did have this sort of impact, then it could easily have been lost in the noise or house effects of individual small-sample seat polls.

End-Of-Night Wrapup: An Unremarkable Result

We've pretty much finished for the night with a projected 55.1% 2PP to the Liberal candidate Andrew Hastie, a 6.7% 2PP swing.  So what have we seen?  Basically, nothing surprising.

The 2PP swing is currently very slightly larger than the average swing (6%) for by-elections caused by the death of a government MP.  Also, taking the government to be at about 50.5% 2PP federally based on the two Turnbull-era polls so far, it's very slightly larger than the expected swing (6.3%) for a government with such a national poll standing.  Whatever happens in the postcount, there's basically nothing to explain in the 2PP swing, and if we did need to explain it we could add in Don Randall's personal vote being slightly larger than for most departing MPs.  It's a result close enough to expected that the only thing to learn from the size of the 2PP swing is that there is nothing to learn.  No shifting of the blowtorch to Bill Shorten, no wave of euphoria for a new Prime Minister beyond the obvious bounce in national polling, no voter condemnation of Canberra chaos.

The primary vote pattern shows that the swing to Labor is amplified, and the swing against the Liberals reduced, by the predictable implosion of the Palmer United Party vote (PUP saved from greater embarrassment by drawing the donkey vote).  The most interesting thing in the minor party votes is that the Christian right parties gained nothing from the replacement of Tony Abbott with the more socially liberal Malcolm Turnbull.  Having an openly Christian candidate may have actually helped the Liberals in that regard.

The result for the Greens was a poor one, and while competition from left-micros like the Animal Justice Party and the Pirate Party didn't help their cause, they failed to pick up any voter disillusionment with the direction of federal Labor or Bill Shorten in this election, and have been seen as largely irrelevant.  And, with the switch to Turnbull doubtless a contributing factor, preference flow change looks like it was a non-issue.

The sole truly Turnbull-era poll by ReachTEL seems to have scrubbed up fairly well (especially by seat-polling standards), but to have overestimated the Coalition slightly.

Both major parties would I think be relieved by this result after a tumultuous week.  Labor should be heartened by the strong improvement in their primary vote, and if they do intend to run Matt Keogh for the new seat of Burt, then he's certainly passed the audition.  For the Liberals, Andrew Hastie was an unconventional candidate, who had the odd rocky moment on the campaign trail but succeeded in maintaining focus on local and candidate issues.

I will post some postcount updates above over coming days.

Live Election Night Comments

Live comments will be posted between the lines below once counting starts (6pm WA/8 pm in the east), scrolling from the bottom up.  Refresh now and then for the most recent comments.  Note: PPVC = pre-poll voting centre.  And I should have mentioned times were in AEST!


11:35 pm: Still waiting for that last PPVC and then I'll post a wrapup.

11:08 pm: A bit of late increase in the swing with Armadale PPVC still to reach the 2PP count.

10:56 pm: The Abbott-era aspect in the PPVCs doesn't seem to be doing vast damage (nor should it according to my model) and it does look like this should end somewhere in the 55s, perhaps even still the 56s, after postals.

10:48 pm: We're close to final for the night now as Armadale PPVC is in with only one significant booth remaining.  Armadale PPVC did not make a massive difference to the primaries.  Still waiting for the big PPVCs to flow through to the 2PP vote and then that will be it for the night.

10:46 pm: A nice line from Hastie in his speech pointing out that Canning and Canberra are further apart than London and Moscow.  Mandurah PPVC is in now but we are still waiting for Armadale to be added.

10:35 pm: Having a look at the minor party votes, it seems the Greens are suffering from the competition from Animal Justice and the Pirate Party.  The Liberal Democrats are still running last behind even van Lieshout.  There is also no evidence of a swing from the Liberals to the Christian parties.

10:23 pm:  Another possible cause of the swing increasing is that there are some big Armadale booths to come, and they are swinging more than the Mandurah booths.

10:16 pm: The projected swing is now down to 6% on the 2PP booths, 6.5% projecting off the primary booths.  But we still need to see what happens when the big PPVCs hit the fan as it should blow out a bit when they are counted.

10:07 pm: An apparent error on the ABC site in the booth of Boddington.  The major party primaries appear transposed. A small booth so if the errors are reflected in the AEC figures then the Coalition could be 0.5% too low and Labor 0.5% too high on current numbers (which will roughly halve by the time we reach the full count.)

10 pm: Probably a reasonable time to wonder whether the seat polls for Canning actually had a pro-Coalition house effect and the seat might even have been lost under Abbott.  But really not enough Turnbull-era data to say.

9:50 pm: The bit below said I have looked at some more preference flows in booths and it seems that strong Liberal booths are showing increasing preference flows, on which basis the overall change in preference flow is actually going to be negligible.  So that puts my projection back on course for about 55% 2PP though the big PPVCs and post-count could knock that down towards 54.

9:35 pm: I currently have the 2PP swing projected at about 8% and this should increase as prepolls and postals are added.  We're not, at this stage, seeing a euphoric by-election bounce for a change of leadership and it's looking like Labor is heading for a pretty solid result.  The model I gave in my previous piece actually expected an 8.6% swing based on current federal polling, and I'm not entirely sure why my expectations crept up a few points over that! (It's looking rather good at the moment.)

9:32 pm: Swing just went up quite a bit as Keogh had a sensational booth somewhere; I think it was Jarrahdale where he has a 16% primary vote swing (8% against Hastie).

9:30 pm: My zillionth reminder for this campaign that the average by-election swing for the death of a member is an inappropriate baseline and the correct baseline is the average swing for the death of a government member (6%).  Which is about what it looks like we are getting, maybe a little more.

9:25 pm: OK, I've looked at overall preference flows in five booths: Carcoola, Dudley, Dwellingup, North Dandalup, Preston Beach.  On average (unweighted by vote total) Hastie is getting 38% of preferences in these booths, which compares to 42% for Randall in 2013.  So if that holds up, with the Others vote around 18%, we can assume the overall swing to be about 0.7% larger than primary swings apply.

9:21 pm: I'm doing some preference flow comparisons for a few booths and will report on this soon.

9:11 pm: Not really a big live effort from News 24, just a trickle of reports here and there.  Coolup and Lake Clifton in, PV swings now -4.6/+8.5.  6.5% of all votes counted.  The 2PP swing will bob about a lot, including in the post-count, but is currently running about 5.8% (which would give a 56:44 result).

9:10 pm: Really not seeing a lot of reason to doubt that Hastie will win.

9:06 pm: Liberal Democrats currently running last on about half a percent, confirming and then some my view that their decision to preference Labor was a non-issue.

9:04 pm: Nothing near a losing swing on primaries in any of the eight non-PPVC booths in so far.  Remember we should be a bit cautious about the relatively low 2PP swings early so far because the big PPVCs will show large swings (possibly around 10%) since their votes are mainly Abbott era.

8:57 pm: Bedfordale and Mandurah added. Also Dudley Park and Forrestdale, and the primary vote swings are now running at -4.8/+8.3.  So the ALP primary is doing really well but a lot of that is coming off Palmer United.

8:52 pm: Dwellingup in too and the primary vote swings are down slightly.  I'll give them in the form primary swing against Liberal/primary swing to Labor and it's currently running at -6.7/+7.5.  ABC booth results page clearly broken.

8:50 pm: A big lot of votes in from Carcoola and North Dandalup, with the primary vote swings now running at about 8%.  Some small PPVCs are shown as having been counted, and it's likely that swings in them would be higher because of their votes having been cast in the Abbott era.  On this basis the swing could well go down.

8:40 pm: ABC has individual booth results at this tab but this does not include the small PPVCs that have been included.

8:36 pm:  ABC TV computer doing something queer getting an 11% 2PP swing off this booth.  (Edit: Most likely it is partly taking into account the small dribble of PPVC booths that have just been added.)

8:28 pm: First booth is in; it's Preston Beach and there we have on primary votes a 5.2% swing against the Liberals and a 7.3% swing to Labor, so likely a 6.something% swing after preferences (though we don't have those yet) - all completely expected.  The 7.7% swing on the ABC site is a projection only.  I'm not yet clear on whether individual booth totals are accessible to the public.  Indie nutter Teresa van Lieshout is off to a strong start with 3 votes, that's equal fourth!  The LDP are not yet off the mark, which should set the scene for the rest of their night.

8:22 pm:  Note that the AEC will post projected two-candidate preferred scores but for the first 15% of the count they will be based on assumptions from the previous election.  We'll see to what extent they are corrected for booth differences as the count proceeds.

8:04 pm: Reports of a lowish turnout (common for by-elections) with 87,000 voting.

7:50 pm Just about ready to go with polls closing in 10 minutes.  Official results will be going up on the AEC VTR site here.  Probably take half an hour at least for any substantial figures to come through.



Welcome to my election day comments, live counting comments and postcount comments on the Canning by-election.  There will be live coverage of the election on ABC News 24, on Pollbludger, Tally Room and other places, so consider my comments here as being a third or fourth opinion in case the modelling on TV is falling over (or for those who only have or want text feed!)

Only a week ago this shaped as the most important federal by-election for many decades, since arguably not since 1968 (and before that 1941) has the potential impact of a by-election on the office of Prime Minister been more obvious.  As it turned out, the main act has been booted off the stage five days before the vote, with dire polling in the Canning electorate a significant causal factor.  That's even though, as my preview argued, that polling was telling us nothing we did not already know.  With Tony Abbott as Prime Minister, a swing of about ten points was entirely predictable based on the history of by-elections, Don Randall's personal vote and the national and Western Australian state of federal polling.

With Abbott gone, this by-election is still popcorn for political junkies, but except for those voting to choose their new local member, its meaning is not that great.  We've never had a by-election days after an unpopular PM was booted by a vastly more popular replacement, so if the result is weird it will probably just mean that we didn't have a handle on what happens in such cases because we had no other data.  I'm expecting an easy Liberal win with a 2PP in the range 54-59 (a 3-8% swing) - and probably around the middle of that range - but if it's something outside that massive range of margins, then so what?  It's always possible that voters' reaction to a leadership change when voting at a by-election might be much worse or much better than their reaction when answering a poll about voting at a general election.   Voters might want to use a by-election to express disgust or more likely delight in a way voting intention polling wouldn't predict.  Assuming that the Liberals do indeed retain, then none of the debate about the causes of the margin (whatever it is) will be worth a hill of beans at next year's election.

There are still some interesting things to see here.  To what extent will the Turnbull ascension provoke a drift from the Liberals to Christian right parties?  Do Liberal Democrat preferences go to the ALP when they try to direct them thus?  Will the Green result be as poor as polling suggests, which seems odd given their national surge?  And so on.

We've had only one and a half public polls since the Turnbull ascension - a ReachTEL on the night of the spill showing a hypothetical 57:43 under Turnbull, and a final ReachTEL taken on Thursday night showing the same.  Liberal internal polling is also said to be showing this margin.  Seat polls are not very reliable at the best of times so let's see how these pan out - that said, large sample size plus the experience of repeat sampling improve their prospects of being correct.  A fair deal is being made of the share of votes being cast before Abbott was rolled, but the late polling should be capturing these votes too, although it's possible voters who now wish they had voted Liberal will self-misreport.

Labor has had to ditch signage attacking Tony Abbott and instead has signage attacking Julie Bishop for alleged disloyalty to Tony Abbott, which given that Bishop is Western Australian and popular (and that the charge of disloyalty is contestable anyway) will probably go down like a ton of bricks.  There's been a slightly more original attempt by GetUp! but nobody much outside left-wing social media circles will have a clue what it's about.

Some Details

A few things to point out for those following the entrails of the election.

As this is a by-election there are no absent votes, and very few out-of-electorate prepolls.  As the great majority of prepolls are counted as booth votes, the post-count will consist overwhelmingly of postals.  As of Thursday, 11621 postals had been issued, of which 506 were duplicates or disallowed. Liberal Party-issued postal vote applications (5351) exceed Labor applications (1255), suggesting a greater effort by the incumbents on getting postal votes in the seat.  (Note that party-postals do not necessarily mean the voter will vote for that party.)  The lack of absent votes at by-elections can mean that conservative candidates surge massively in post-counting, but this time this trend should be tempered by most of the postal votes being counted before Tony Abbott was removed.

As of Thursday there had been 225 out-of-electorate prepolls and 10026 within electorate.  The within electorate prepolls are overwhelmingly at the Armadale and Mandurah PPVCs. There wasn't a PPVC called Mandurah in 2013 but the Greenfields (Canning) PPVC was close enough to be equivalent.

There are relatively minor changes in the booths from 2013.  A second Byford booth is added - the original one receiving over 4000 votes last time - and the Greenfields (Canning) booth which received only 550 votes is off the menu.  A few other booths have moved slightly.


  1. What's the baseline swing for a by-election in the middle of a Prime Ministerial honeymoon?

    1. There's not enough data to calculate a comparable one! By-elections when a new PM had just been elected aren't comparable because they usually involved a boost in national 2PP voting to above the previous election. In this case the "honeymoon" still sees the Coalition 3 points nationally below the 2013 election (at least based on the two polls so far).

  2. Yes, so the "Turnbull bounce" is a pretty modest one. Looking at Canning, the real story is not the 6% swing, but the 10 and 11% swings in some of the suburban booths. This spells trouble for the Libs in the Perth marginals Cowan, Burt, Hasluck, Swan. And this happened in a by-election which Labor actually didn't want to win (for fear of getting Abbott rolled), and accordingly spent almost nothing, while the Libs spent $1 million. Feel free to make these points :)

    1. Thats interesting about the Libs outspending Labor, my impression from reading Bolt and the Oz, was that the ALP ( and the Unions) were the big spenders! It looks like the Libs were outspending the ALP by at least 2 to 1, according to the West Australian.
      I suspect that Labor or more accurately the Unions, were more targeted in their campaign using American style marketing tactics. Was Canning a proving ground for these techniques?

  3. The big swing in Armadale is probably due more to that fact that it's a Labor voting area: the sort of people who might vote for a "popular larrakin" like Don Randall personally, but not really supporters of the Liberals. With Randall gone, the area has simply reverted to type. Mandurah and the coastal communities are more naturally Liberal, so the loss of Randall wouldn't affect their vote as much.

    Not sure you can extrapolate to other WA seats like that....