Canning By-Election: Prospects and Polls

Seat: Canning (WA, Lib, 11.8%)
By-Election 19 Sep 2015
2PP Contest: Andrew Hastie (Lib) vs Matt Keogh (ALP)
Current outlook (15 Sep): Liberal retain expected with modest swing against.

Note: Turnbull-era updates appear at the top of the article; the rest is archived below the break.

As of 19 Sep, Coverage has switched to Canning: The Anticlimax Live.

Turnbull PM: What Does This Mean For Canning?

Most of this article was written while Tony Abbott was still Prime Minister, and the surprise removal of Abbott prior to the by-election has made quite a mess of the analysis.  Abbott's Prime Ministership was the major risk factor to an expected Coalition victory because the bad polling endemic to it was likely to cause the swing to be larger than normal for a by-election caused by the death of a government MP.



With Abbott gone, it's too soon to say what the federal polling picture will look like on election day (or even if we'll have one), but an improvement of at least a few points in federal voting intentions seems likely.  According to the graph below, about 60% of the federal improvement should be expected to flow on to Canning.  This makes it much more likely that the Coalition will now retain the seat with a comfortable margin.  Tentatively, I'd now expect the swing against to be somewhere in the 4-9% range, meaning a 53 to 58% 2PP for the Coalition, although it could be worse should recriminations from the booting of Abbott become messy.  One slight dampener on the impact of the leadership switch is that over 15% of the electorate will have already voted and cast their vote in the Abbott era.  This could make for some pretty strange postal and prepoll voting behaviour.

Another thing that could change is preferencing behaviour.  It might not be as different from the last Canning poll as I've expected now (though it will be interesting to see if the flow from the Christian right parties is now weaker.)

Andrew Hastie will probably be grateful for being able to get back to fighting for the seat on local issues rather than being dragged down by the Abbott factor and by constant Canberra leadership questions.  On the other hand, Hastie seemed very much in the Abbott mould in what could be discerned of his policy positions and style, and this may make things slightly awkward for him on the campaign trail.

Over the next few days I'll decide what to do with the mess Turnbull has made of my most popular article this year!

15 Sep: New ReachTEL: A ReachTEL taken last night in the prehistoric Abbott era with a sample size of 1127 had a 52-48 respondent-allocated 2PP with Hastie 45.3% Keogh 36.4% Greens 7.4% PUP 2.6% Other/Ind 5.3% and 3% undecided (an interesting comparison to the 12% undecided in Ipsos!)  The flow of respondent preferences was 63.4% to Labor, much lower than in the previous smaller commissioned ReachTELs and broadly in line with Ipsos and what might be inferred from Galaxy's hybrid preference method.

More importantly, the poll asked hypothetically how respondents would vote if Malcolm Turnbull was PM.  This should be treated with mild caution since we need to see how the dust settles from the spill and whether people actually respond how they say they will, but it's likely a much more reliable guide than anything else going.  In this question the 2PP is 57-43 with Hastie 47.6% led Keogh 32.9% Greens 6.4% PUP 3.3% Other/Ind 7.1% and 2.6% undecided.  The undecideds in this case (all 29 or so of them) are more Liberal-leaning and the flow of preferences to Labor drops to 50.7% (similar to the same seat at the last election).

The higher Other/Ind vote suggests some disgruntled Abbott fans will switch to the Christian right parties as well as PUP, but not all that many. Liberal supporters in the electorate still preferred Tony Abbott (48.8%) over Turnbull (28.2%), Morrison (17.6%) and Bishop (a surprisingly low 5.3% for a fellow Western Australian!) but were split on which of Turnbull or Abbott would be most likely to win the next election.

15 Sep: Betting: Currently seeing 1.10 or 1.15 for Hastie and 5.00 for Keogh, so about an 81% implied winning chance for Hastie.  Would not be too surprising to see Labor drift a bit more.

17 Sep: Labor has said ("Malcolm Turnbull ascendancy crushes ALP hopes of Canning win") that their polling does not show them winning, though they hope a swing of 4-5% could still be achieved. No figures from this polling are released.  Betting on Labor has indeed drifted further, with 1.08 for Hastie and 7.00 for Keogh, an implied 87% chance of winning for Hastie.

ABC News 24 will be covering the by-election.

It is worth pointing out that we have not as yet seen a public poll taken since Malcolm Turnbull became PM, though we may well get some before the day.  I'll have some live coverage on the night mainly to give a third or fourth opinion to the other attempts to project the margin off booth figures.

19 Sep: Final ReachTEL 57-43.  Liberal internal polling said to show same margin.

(Abbott-Era Content Archived Below This Line:)



======================================================================
Advance Summary (Abbott Era)

1. This article attempts to model the upcoming Canning by-election, taking into account the current state of federal and state polling, the personal vote of Don Randall and any other factors that might be relevant.

2. Historically (since 1950) there has been a rather strong relationship between the state of federal polling at the time of by-elections, and the swing recorded at that by-election.  

3. On the basis of that relationship alone, a swing of about 8.7 points would be expected based on current federal polling.

4. The balance of other relevant issues - none of which can be modelled precisely - suggests the expected swing should be slightly larger than that.

5. However, factors that might counter the expectation of a near-winning swing include the qualities of the candidates and any change in how the Coalition is travelling generally. State factors might push in the opposite direction, but may already be partly factored into WA federal polling.  In the final week of the campaign, leadership speculation is emerging as another factor that could affect the result.

6. While great uncertainties exist in predicting by-elections, this article therefore agrees with the widespread view that a large but not quite sufficient swing to Labor is the median expected outcome.

7.  Because of those uncertainties, there is a realistic chance that Labor will win the seat.

========================================================================

With the date for the Canning by-election announced, it's time for some analysis of this event. You can find more detailed guides at The Tally Room,  ABC Elections and doubtless other places.  This piece is concerned with factors relevant to trying to project the outcome.  It is likely to be updated until the by-election night, when I may or may not do a live or postcount thread.  The by-election is seen as a major test of how the Abbott Government is really travelling, and whether its presently poor polling is entirely real.  A loss would be a disaster and would evoke comparisons with other famous by-election losses that have been portents for changes of government, and would certainly fuel increased leadership speculation.

A comfortable win will obviously be a triumph for the Coalition, and one which might raise questions about how real Labor's national polling lead is.  Of more interest is a narrow win.  A win by any margin will come as some relief, but a win with a large but not quite sufficient swing will be spun by both sides as a victory.  This piece considers what the goalposts for this by-election should be, and will also cover polling and, when I see any, betting.

Here is a brief overview of the major relevant factors, before we get on to candidate quality, campaigning, the possible impact of state issues and so on:

1. By-elections in government seats normally produce swings against incumbent governments.
2. The current federal government is polling very poorly one month out from the by-election
3. The current federal government is polling especially badly in Western Australia.
4. The former sitting member Don Randall had a significant personal vote.

Avoid Double Counting

The major trick for new players in assessing the prospects for this by-election is to avoid double-counting the above factors.  It is easy to allow, say, six points for factor 1, seven points for factor 2, two points for factor 3 and five points for factor 4, and conclude from that that a swing of, say, 20 points is possible and Labor should win this easily.  However, that is unsound reasoning.

It is unsound reasoning because some of these factors overlap.  One reason why by-elections in government seats normally produce swings against the government is that the departing MP takes their personal vote with them.  Another is that, at the time of a by-election, a federal government is normally travelling worse than its result at the previous election.  Indeed, based on these two factors alone, the conventional wisdom about "by-election backlash" may well be exaggerated, and the number of voters voting strategically to "send a message" could be much lower than it appears.

Are By-Elections Caused By Bereavements Different?

It would be naturally expected that by-elections caused by the death of an incumbent would produce a weaker swing than those caused by retirements or other factors.  After all, voters are often annoyed by unnecessary by-elections caused by MPs retiring, while a party can't help it if one of its own MPs dies.  Some voters even think it is unseemly for Oppositions to contest such by-elections.  (I don't quite agree; the voters elected a person not a party and in some cases only that person would have won that seat; if enough voters still wish to be represented by that party they will vote accordingly.)

However, when it comes to by-elections involving government seats, the evidence doesn't suggest much difference.  As noted in previous comments on Canning, for cases for which a 2PP swing can be computed, the average swing has been much the same for 26 such government seats since Federation as for the average of all government vacancies (6.3 +/- 4.7 points in either case).

I did however omit one unusual case from the previous assessment.  In the 1980 general election for the Liberal-held seat of McPherson, enmity between the Liberals and National Country Party (now Nationals) in Queensland at the time resulted in an unofficial NCP candidate running as an independent and directing preferences to Labor.  This caused a 12.7% swing to Labor.  The Liberal incumbent died in 1981 and a sitting Nationals Senator quit the Senate to try to wrest the seat from the Liberals in the by-election.  The result of this was that Labor finished third, and the ALP's 2PP result in that by-election is unknown; there was probably a substantial swing against Labor on a 2PP basis.

Also, by-elections involving the voiding of results are different from resignations or bereavements: they tend to occur early in a government's term while it is probably still polling well, and voters sometimes blame officialdom for causing them.  All this taken into account, there is a slight difference in the expected result for a government death vs a government resignation.

More importantly, it's possible the behaviour of voters in by-elections caused by the death of government incumbents has changed as this has become a less common event.  Ten government incumbents died in the 1910s and between three and six in every decade from the twenties to the sixties.  However since the 1970 Chisholm by-election there have been just two more such cases: the McPherson example mentioned above and the 2001 Aston by-election which saw just a 3.7% swing.  The last five such by-elections, going back to 1964, have all shown reduced (if any) swing to Oppositions.

I wouldn't read too much into that, based on the small sample size and the unusual nature of pretty much every by-election when you look closely enough.  (For instance in Higgins 1968 one Prime Minister replaced another, while in Aston 2001 a 15-candidate ballot caused a three-point increase in the informal rate.)  But maybe we should give some concession to the possibility that voters are more sympathetic when a government MP dies these days, and expect that at present the difference in expected swing between a by-election caused by resignation and death might be something above one point.

Added Aug 21: Bill Shorten has been trying to manage expectations by pointing out that the average 2PP swing against governments at by-elections caused by deaths since 1949 is only 2.5%.  But this is disingenuous since that figure includes both Government and Opposition deaths.  In the case of an Opposition death it is the Opposition that is losing a personal vote, which reduces the swing against the Government.  Indeed since 1949, 55% of by-elections caused by death have involved Opposition seats.

The Federal Government Is Polling Poorly

One would expect some link between by-election swings for government vacancies and how a government is polling at the time.

To investigate this I've dug into my ancient scrubbled list of Morgan results as well as the good old Newspoll rolling average.  For each government by-election held since 1946 for which a 2PP swing can be found, excepting technical vacancies such as Lindsay, I estimate the 2PP swing in national polling and compare it to the swing at the by-election.

There are 34 usable by-elections and the scatter-plot looks like this (Labor government by-elections red, Coalition government by-elections blue):

With a pretty decent 29% of variation explained (especially given some of the ropey old estimates off Morgans taken months apart for 1950-1970), I get the following regression:

By-election swing = .587*Polling swing + 5.46

...which on present national polling would imply a 9.5% swing is to be expected before we consider Don Randall's personal vote or state issues.

However, note that the red dots are mostly above the line and frequently nowhere near it.   Labor has a historic tendency at general elections to underperform its polling compared to the Coalition - so the former should be expected - and the Labor dots include both one retirement of a recent leader (Bill Hayden) and one case where the candidate was a popular ex-Premier (Carmen Lawrence).

If we just use Coalition incumbents, which I do believe we should, the equation becomes:

By-election swing = .615*Polling swing + 4.46

... with 41% of variation now explained, which is rather good.  On present national polling that would imply a more modest 8.7% starting point.  However, that starting point will move based on national polling (which the Coalition would hope would get better rather than worse).

(By the way, the average national polling swing against a government at the time of a government-seat by-election is 1.6%, from an average previous-election base of 52.2%.  There's a whiff of a suggestion there that governments that are polling badly tend to discourage retirements.)

A different historical approach can be seen at Poll Bludger.  A regression of state and national by-election swings against polling swings since 1990 finds 46% of variation explained and projects a lineball outcome for a national polling swing of 7%.  That should be a rather alarming graph for the Coalition given it does not even consider whether the departing members in those by-elections were of the government or not.

The Federal Government Is On The Nose In WA

While the national swing against the Coalition in polling is currently running at about seven points, it's nearly ten points in WA according to the Bludgertrack sidebar.  It's hard to know how much use to make of this as there is not the same level of historic data on state-by-state federal polling at the time of recent government by-elections.  Indeed there have been only two Coalition government vacancies in the Newspoll era until now.

State-by-state variation in government standing would be one of the causes of variation in by-election results, but any measure of state polling is less robust than one of federal polling, so it's challenging to decide how to weight this factor. Arguably, up to three points could be added to the expected swing, but I'd not be inclined to add that much.

Don Randall's Personal Vote

This has been a big deal in a lot of analysis of this by-election already, but it's a point that is easy to overcount.  As the ABC elections chart shows, Canning has generally been a slightly Labor-friendly seat over the past few decades, when compared to other WA federal seats.  However, when Don Randall has been the defending incumbent, it has leant Coalition compared to the state average three times out of four.  The exception was when Labor ran a super-strong candidate, Alannah MacTiernan, against Randall in 2010.  After adjusting for redistributions, the average outcome since 1980 has been that the Coalition have done 5.6 points better relative to the rest of WA when Randall was the incumbent than when he wasn't.

That's quite large, but that doesn't mean that is Randall's personal vote.  It might be the difference between his personal vote and the personal vote of other incumbents, who in the period under scrutiny were mostly Labor ones, though some of them were Liberal, further confusing the picture.  We also don't know whether demographic change in the electorate might have played a role.

Another way of trying to get a handle on personal votes is by comparing the House and Senate vote for the given parties and Peter Brent has done this in handy tables like this.  On his adjusted measures (see the bottom for explanation) Don Randall takes with him the sixth-highest personal vote in the country (5.9 points), one which turns out to be 4.3 points higher than the average for a recontesting MHR.

So then, do we throw Labor 4.3 points?  No, because the candidates whose deaths or resignations cause by-elections tend to have been in parliament for longer than the average MHR.  For by-elections in the time period covered by my sample above, the average by-election period was 16 years.  Those who had been in parliament in their current seat for between 12 and 20 years at the time of the 2013 by-election had higher average personal votes than the average MP (2.3 points, 3.6 below the estimate for Randall).

But there's more, because MPs whose departures cause by-elections tend to be more prominent than even the average MP who has been there for 16 years.  They are disproportionately former PMs or other major ministers, who might well be expected to take higher personal votes with them.  So while Randall appears to have been very popular, the loss of a high personal vote at a by-election is a rather common event in by-election history.

For this reason the adjustment made for Randall's personal vote should be at most two points, maybe less.  It's also worth noting that some of Randall's personal vote history derives from elections where Labor have either stuffed up their preselection or not really tried that hard to beat him.  It might also be that Randall's 2013 vote was so high because of the issues mix ideally suiting his electorate (see AFR link in Other Reading).  That said, all this doesn't really matter, assuming Labor will be trying hard in this event and that that issues mix no longer applies.

One way in which an issues-mix component in Randall's apparent personal vote could matter is if FIFO mine workers remain loyal to the Coalition over the abolition of the mining tax.

Adding It All Up

Here then is my subjective tallying of the expected swing factors:

* Existing margin: 11.8 points
* Starting point based on Coalition history in by-elections based on current polling: -8.7 points
* Adjustment for slight historical difference between by-elections caused by retirement and by death, with some expectation that that difference might have increased: +0.7 points
* Adjustment for bad Coalition polling in Western Australia: -1.2 points
* Adjustment for Randall personal vote compared to personal votes at other elections: -1.8 points

Revised expected margin 0.8 points.  (Plus or minus at least four!) 

(As of 3 Sep, the expected margin based on these issues had increased slightly to 1.2 points).

So assuming the Coalition is still polling as badly federally in four weeks time as now, assuming that the candidates and campaigns are of equal quality, and assuming that no sleeper issue like one arising out of state politics comes to dominate the by-election, my estimate is that Labor will more likely than not fall a bit short.  But granting all those assumptions, they would have a realistic chance.

Predicting the track of polling is a mug's game but the Coalition are currently going through an obvious rough spell and on that basis it would be risky to predict things would get worse for them rather than regress towards their average indifferent 47.5%-ish 2PP result for this term.  So on that basis alone I see some doubt that it is quite as close as the above estimate suggests.  (It did occur to me to float the idea that anti-Abbott forces within the Coalition might seek to cause the by-election to be lost as a way of getting rid of Abbott, but it would be hard to do something that dastardly without the plot being obvious.)

There is also a widespread view that the candidate picked by the Liberals, 32-year old former SAS soldier Andrew Hastie, is a particularly good choice because he can make a very strong case about how he can personally contribute to one of the government's few apparent strengths (national security).  I'd suggest some caution about assuming that the by-election will just neatly follow what sounds like an Abbott Government script (they frequently don't), but possibly Hastie will demonstrate other strengths as well.  At any rate, choosing a very different candidate who is decades younger than the inimitable larrikin Don Randall seems like a smart move.

The current Labor frontrunner (edit: now candidate), 33-year old Law Society of WA President (edit: he's just stepped down from that) Matt Keogh, is not attracting these sorts of rave reviews so far, but opposition candidates at by-elections do not need to. They are most likely to win if they are solid and uncontroversial and something is annoying the voters enough to switch the seat.

Curiously, neither major party candidate lived in the electorate at the time of the by-election being called.  Hastie has since moved in, having no particular historic connection to Canning to my knowledge, while Keogh's family live in the electorate and he has a background there but currently lives in North Perth. Hastie's particularly weak connections to the electorate (indeed he has only lived in WA for five years) make me suspect that the case for Hastie's candidacy making a big positive difference is being overblown.

On 22 August the Fairfax presses revealed that Hastie was commander of a troop under a long-running investigation for chopping the hands off dead Taliban fighters for identification purposes. It appears that Hastie himself did not command that this action occur, and the candidate says that he referred it up the chain of command.  If his explanation as given in his speech the same day checks out then my judgement is that the issue is unlikely to harm his candidacy and could even help.

There is no evidence any other party or independent will be remotely competitive at this stage.  Other candidates are the Greens' Vanessa Rauland (announcement), Palmer United Party's Vimal Sharma, Jamie van Burgel of Australian Christians, Michelle Allan (Pirate Party), Greg Smith (Australian Defence Veterans Party), Connor Whittle (Liberal Democrats), Katrina Love (Animal Justice), Jim McCourt (Family First), Angela Smith (Sustainable Population) and independent serial candidate Teresa van Lieshout.  A large field does not assist Labor because they are more likely to lose votes to the informal pile, while the ballot order also gives the Liberals the very small advantage of the donkey vote.  If the ADVP poll anything of substance then their flow to a career soldier will be interesting.

Preference assignments by the minor parties between the major parties on how-to-vote cards (which are not binding, the voter must decide whether to follow them or not) are:

Greens, Pirate, Animal Justice: Labor
Australian Christians, Family First: Liberal
Liberal Democrats: Preference flows to Labor at position 10 with Liberals at 11 and Greens at 12 - see statement of reasons.  Hastie has alleged a preference deal between Labor and the LDP, a claim which as far as I'm aware is bollocks.
PUP: Donkey vote (preferencing Liberal) given as example, but voters encouraged to choose own preferences.
ADVP: Believed to be open ticket
SPP: Open ticket
van Lieshout: Unknown.

One piece of advice I would give to those campaigning against the Coalition (that includes you, Mr Shorten) at this election is to stay well away from the argument that voters should vote Labor in the hope of causing Abbott to be rolled by his party.  We saw how well that worked in New South Wales.  Campaigners should encourage voters to vote for the best person to represent the electorate, and make the case that that candidate is on their side.

Although my model above points to a relatively close outcome I would be not surprised if it was a few points less close, provided a lid can be put on federal destabilisation.  Overall I see any result between 51 and 54 to the Liberals as ho-hum and not really spinnable either way, anything below 51 as bad (losing being very, very bad) and anything above 54 as good.  Laura Tingle reports that 52 is widely being seen as the line in the sand, a view with some merit because anything below 52 carries a strong signal that the government's bad national polling is very likely to be real. That line might soften if there is relief that the seat has been retained.

Update 11 Sep: Federal factors are threatening to further impact on the by-election following the publication of claims that the Coalition was intending a major frontbench reshuffle including the removal of underperforming Abbott loyalists Kevin Andrews and Eric Abetz, as well as Trade Minister Andrew Robb and others.  It is disputed whether the plan came from the PM's office but has been hamfistedly executed, or whether the plan is made-up by someone attempting to cause the loss of the by-election in the hope of damaging Abbott's leadership.  Either way there is the potential for the federal Coalition to be focused on internal matters during Canning week, which could cause the loss of the seat by itself.  I've placed my assessment of the seat on hold temporarily til we see more polling or whether this blows over.

Enrolment Saga

It's been reported that neither Keogh nor Hastie will vote in the by-election.  Hastie's enrolment has come under scrutiny based on him having not lived in the electorate for a month prior to transferring from another division.  A month is required under S99 (1) of the Electoral Act.  Hastie's transfer was accepted, the question being why.  Hastie's explanation appears factually consistent with the form he states he filled out.  It appears Hastie's declaration that he was eligible to enrol at his current address is false.  It is only an offence to make a false declaration that is found to be deliberately or recklessly false; a conviction on deliberate falsehood would result in loss of seat.  However, at this stage there is no evidence that this is anything but a stuffup.

Polling

The first poll for this electorate is a Newspoll with a sample size of 508.  As those following my general federal polling would be aware, Newspoll is now done by Galaxy and the phone side of Newspoll's polling involves robo-polling of landlines, with scaling to overcome the issues with such polls.  Galaxy's national Newspolling includes an online component that reaches people who don't use landlines, but I do not know if it was used to any degree in this case or not.

What we do have is a 2PP of 51:49 to Liberal (with no candidates named in the poll) off primaries of 41 Liberal, 36 Labor, 11 Green, 2 PUP, and 10 for the rest.  It is worth noting that in Canning last election, Others voters were mainly Australian Christians voters, and as a result Others preferences flowed 67.6% to the Coalition.  On that basis the expected 2PP to one decimal place off these primaries is about 51.3 to Liberal.  For the umpteenth time I point out that seat polling in Australia is a troubled enterprise and all a poll like this really does is fails to debunk the natural expectation of a close Liberal retain.

As of 21 August there are also reports of a union-commissioned ReachTEL for the AWU on July 21 having shown a similar result.  Sample size was 734 and 2PP 50.8 to Liberal.  Old rope now and full details not available, but broadly consistent with what we should expect.

26 Aug: Reports of a new union-commissioned ReachTEL poll claiming to show a knife-edge result here.  The reported results are a 2PP of 50.1 to ALP "after undecided voters are allocated according to their leaning" off raw primaries (prior to the allocation of leaning voters) of Coalition 44.4 Labor 30.2 Green 8.6.

This is a strange way to present primary vote data and has caused a lot of confusion.  As that only sums to 83.2 (plus probably several points for Others) it can be assumed undecided voters were at least several percent of the sample - the published primaries are not comparable to normal polling primaries, which are figures after the prodding of initially undecided voters.  (Edit: Yes, the missing figures were 2.3% PUP, 5.9% Other, 8.6% undecided as noted by Poll Bludger, who has also noted the primaries after including the leaning votes were 47.3% Liberal, 33% Labor, 9.6% Green, 2.7% PUP, 7.5% Other.

The split of respondent preferences from those answering Green, PUP or Other (including undecideds who preferred one of these) was 85% to Labor, a figure that would not happen at a real election and is best explained by small sample size ... with a strong hint that the preference flow from non-Green third parties is likely to be less friendly than last election.  By 2013 Canning preferences, the Liberals would get 56% 2PP on these primaries!

A side-result of the poll is that "connection with Canning" was found to be somewhat or very important by virtually all respondents, but the wording of this question has not been published.

29 Aug: ReachTEL are doing rather well out of this one since their third poll of the campaign, this time commissioned by Australians for Marriage Equality, is now out.  Again we've got a very strong flow of respondent preferences with the Coalition getting only 51-49 off primaries of 44.6% Coalition, 31.6% Labor, 7.5% Green (plus, I assume, others and the initially undecided). The electorate marginally favours same-sex marriage according to this poll; the figures are fairly typical for robopolling in a slightly but not very conservative seat.

3 Sep: We have more commissioned polling with a GetUp!-commissioned Essential showing 51-49 using last-election preferences off primaries of 43% Liberal 36% Labor 11% Green, while another ReachTEL by Solar Citizens gives 51-49 respondent-allocated off primaries of 46.7% Liberal 35.6% Labor 8% Green (basic details from Poll Bludger here, I might find more details of the figures later.)  There is something funny going on in that everyone is releasing the same 2PP but some are getting there from high Liberal primaries with weak preference flows, and some from modest Liberal primaries using last-election preference flows.

If we actually see high Liberal primaries (as implied by ReachTEL) with similar-to-last-election preference flows then the 51-ish 2PPs will all be wrong and Labor will bolt in.  If we see medium Liberal primaries (as per Newspoll and Essential) with shockingly weak preference flows (as per or even close to ReachTEL) then the 51-ish 2PPs will all be wrong and Labor will win the seat.

I have not much doubt that the preference flow to the Coalition will be weaker than in Canning last time, and probably weaker than in the 2013 federal election generally (bearing in mind the mix of micro-parties contesting), but I'll be surprised if it is quite as weak as respondent preferences have implied.

I'm treating a number of issues findings reported by The Guardian with a lot of caution until I see full details of the question wordings.

13 Sep: We have a new Galaxy poll - the first non-commissioned poll for some time - with a 52-48 margin off primaries of Liberal 44 Labor 36 Green 9 PUP and Aus Christians 3 apiece and all others 5. That would be a 40% preference flow to the Liberals.  Samantha Maiden notes "The preference flow is based on past election with a minor adjustment based on respondent-allocated preferences".  The sample size is 557.  Tony Abbott is preferred PM 41:38 (Coalition voters 75-12 for Abbott, Labor voters 68-7 for Shorten and I make the minor party voters about 41-27 for Shorten, which given that 45% of those voters are Greens is actually not that bad for Abbott).

In what could be the biggest poll of the campaign, an Ipsos with a sample size of 1003 (massive for a seat-poll) has a 2PP of 52-48 by last-election "overall" (batched) preferences (clearly not electorate-specific) and 53-47 respondent-allocated (primaries 45-36, with 9 for Greens, 2 for PUP, 7 for Others).  The respondent preference flow of an implied 42% is much higher than in the commissioned ReachTELs.  We should keep in mind here that respondent preference flows are volatile because of the small sample size involved.  If the preference flow is really, say, 35%, then the margin of error for the estimate of the preference flow even in a poll of this size is 6.5 points before rounding, and probably at least 10 points after it.  If the preference distribution was based on preference flows in Canning in 2013, then the 2PP would be 54:46).

Tony Abbott is preferred PM in the Ipsos by 42:36.  His personal ratings are 39-54 compared with 34-50 for Bill Shorten.  A gap in polling so far is that nobody seems to have polled any personal ratings for Keogh and Hastie.  Full Ipsos results are here, including that there was a rather large 12% undecided rate.

What Preference Flow Should We Expect?

This section was added on 5 September.  The massive differences between the flows in respondent-allocated preferencing and those from the last election raise the question of what preference flows we should expect in Canning.

At the 2013 election the overall preference flow to the Coalition in Canning was 48.1%.  This included flows of 22.6% from the Greens, 53.2% from PUP, 86.3% from Aus Christians, 67.5% from the Nationals, 47% from Family First, 53.7% from Rise Up Australia and 45.1% from KAP.  The overall preference flow from Others including PUP was 60.9% and excluding PUP it was 67.6%.

This compares to a national preference flow to the Coalition of about 38%.  The major factor here is that a high proportion of the non-PUP Others vote in Canning is Australian Christians, who very strongly preference the Coalition.

Two major differences thus far are the collapse in national support for PUP (who polled 6.9% in Canning in 2013) and the change in the composition of non-PUP minor parties.  As a result of the former factor the Greens' share of all third-party votes should increase.  Current polling has the Greens at an average of 9.6 points and Others at an average of 10.5 (most of this polling is commissioned but the ratio in the one non-commissioned poll is the same).  Thus the Greens are polling about 48% of the third-party vote compared to just 33% at the last election.  Polling tends to overestimate Green votes so it may be their actual share is in the low 40s.

The one data point for the PUP vote we have is 2.3 points, which I'll increase to 2.7 for them having the donkey vote.  To carve up the remaining 7.8 points a good starting point is the WA Senate by-election.  All of the micro-parties contested it except Defence Veterans.  Teresa van Lieshout contested but was a below-the-line only independent, and the Liberal Democrats are more likely to have benefited from ballot paper confusion issues in a Senate ballot.  A starting point is that the Australian Christians will poll about 2.5% and that none of the others bar possibly the Lib Dems will poll much more than 1%.

Based on national preference flows from the last election (or the Griffith by-election for the Pirate Party) and guesses for ADVP and van Lieshout, I estimate the preference flow from the non-PUP Others as being about 58% to Coalition.  I'm assuming the very strong Aus Christians flow was partly driven by Don Randall's personal vote last time and won't be quite so high this time.  PUP are giving a donkey vote as an example of preferencing on their card, while telling voters to preference as they wish.  PUP preferences that follow the card example will flow to Hastie, meaning the flow of PUP preferences to the Coalition might weaken slightly but will probably not change very much.  I expect Greens preferences to be less friendly than at the last election, perhaps down to say 15% to Coalition.

All up I'm thinking the Coalition should, on paper, get about the same preference flow from all third parties combined as they did in the last federal election generally, somewhere around the mid to high 30s, and not the flow seen in Canning last time, nor the very low flows seen in the respondent ReachTEL preferencing.  If that's correct, then the ReachTELs come out in a Coalition 2PP range of 53.4 to 54.7, the Essential at 51 and the Newspoll at 49.8.

Betting

The predictiveness of betting is widely overrated but it's interesting to keep an eye on it.  As of 21 Aug, William Hill has Coalition 1.57 Labor 2.35 Other 21.  Ignoring the "other" odds which I definitely don't recommend taking, that's an implied 60% chance of a Coalition victory.

As of 31 Aug, Hastie (Lib) has firmed to 1.42, Keogh (Labor) is at 2.75 and Other is 34.  An implied 66% chance of a Coalition win.

As of 5 Sep, the punters seem to think the fix is in.  Hastie (Lib) has firmed to 1.22, Keogh (Labor) is at 4.15 and Other is at 51. An implied 77% chance of a Coalition win.

As of 8 Sep as noted by Ross Irwin in comments, a minor easing with Hastie (Lib) at 1.28, Keogh at 3.50 and Other at 51.   An implied 73% chance of a Coalition win.  As of 13 Sep, the same, after some ups and downs in the last 5 days.

As of 14 Sep Sportsbet also has a market which offers separate odds for the minors with the Greens at $34 and PUP, Liberal Democrats and Pirate Party all at $101 (don't touch any of these!) Sportsbet's current picture between the majors is much the same.

Other Reading

Psephos division map
The Guardian's unofficial candidate guide
Poll Bludger thread
Demographic profile.  By WA standards, Canning has the highest average age and unemployment rate, but nationally it is not remarkable on either score.
Antony Green on preferences and donkey votes.  I reckon the impact of the donkey vote will be lower than suggested, but it won't be possible to unpick from people following the PUP how-to-vote card example.
Report on traffic issues and local council lobbying.
AFR article on expectations for the seat, including revelations that some Liberals think Canning contains a bogan vote.
Hastie's religious views and connections attract scrutiny
Hastie fails to substantiate charge that Labor didn't support military
Business Spectator on Solar Council campaign against Abbott government at seat.
Report on speculation that Labor isn't really trying.  (I don't buy this, but I do wonder if they were a bit less organised at the early stage this speculation surfaced.  For instance, people were already voting, so why did the Liberal site at that stage have a how-to-vote card but not the ALP one?)