Monday, March 16, 2015

New South Wales: March Poll Roundup And Seat Modelling

(For the final week's article go here - note added 26 Mar because this one for whatever reason is still getting more hits!)

Current aggregate (updated March 19): Coalition 44.6 Labor 35.3 Greens 10.2 Other 10.0
2PP Estimate: 53.3% to Coalition (54.6 by last election preferences)
Current median seat projection: Coalition 52 Labor 37 Others 4


1. Recent polling shows the Coalition continuing to maintain a primary vote lead of at least eight points over Labor.

2. Recent polling also shows that while preferencing patterns among minor party voters may change, they are unlikely to change by nearly as much as in Queensland.

3. However, current polling when translated to seat projections suggests only a small Coalition majority.

4. There is more uncertainty than normal in translating the Coalition's current lead to seat results, and for this reason current voting intention levels do not quite assure the Coalition of victory. 

5. Any narrowing of voting intention from this point would make a hung parliament significantly more likely.

There haven't been a lot of polls in the lead-up to the New South Wales election lately, but with Galaxy just out (and a ReachTEL a week and a half ago that I hadn't covered yet) it's time to post an update and some seat modelling discussion. 

Recent Polling

The Galaxy shows a very mainstream result of Coalition 44, ALP 36, Greens and Others each 10, with a 2PP of 54:46 based on last-election preferences.  Mike Baird maintains a strong Preferred Premier lead over Luke Foley (49:26) and voters were as usual not too good at paying attention to state politics, with just 65% able to name Baird as Liberal leader and 38% Foley for Labor.  (The figure for Baird may sound low but Barry O'Farrell was doing worse eighteen months into the job.) Galaxy also showed only 32% believing the Upper House would be "obligated" to pass Baird's partial electricity lease infrastructure if he wins, compared to 48% saying no.  

Results of the ReachTEL taken on March 6 show the following:

* Insignificant shifts in primary support only since the previous ReachTEL in late February.  The primary vote gap between the major parties (9.2 points) was comparable to the average for all polls this year (9.03).  

* A slight popularity improvement for Premier Mike Baird, with the proportion of voters finding his performance Poor or Very Poor down from 21.3% to 18.9%.  While the proportion finding his performance Good or Very Good was stable at 43.8, a higher proportion of these found his performance Very Good.  

* A strong improvement for Opposition Leader Luke Foley as voters become more aware of him.  Foley's Good/Very Good score increased from 16.5 to 23.6 and his Poor/Very Poor score was down from 27.1 to 23.8.  While not comparable to Baird's, these are quite good ratings for an Opposition Leader by ReachTEL standards.

* Voters oppose the government's plan to lease a share of the state's electricity poles and wires (25.8-51.7) with 22.1% of Coalition supporters opposed.

* There has been a substantial weakening in the tendency of voters to say that the Abbott federal government will make them more likely to vote against the Coalition in NSW.  Very strangely, 12.4% of state Labor voters press the button to say they will be more likely to vote Coalition on account of Abbott's performance.  

* Voters for each party tend to cite their party's stronghold areas as reasons for voting for that party.  The economy followed by health come out as the two most cited issues from the list available.


The most useful feature of the ReachTEL - and one I don't recall seeing in previous public polling of any kind - is that it publishes the exact breakdown of how voters say they will distribute their preferences.
This is much better than trying to work out what is going on with respondent-allocated preference claims from released 2PP figures rounded to whole numbers and where the pollster's preferred method for calculating the last-election figure may well differ from the modeller's!  Anyway, on this basis if the above preference pattern was repeated at the state election, the changed preference flow would make a difference of around one point, perhaps slightly more, compared to using 2011 preferences.  I make the 2011 preference flow for ReachTEL's published primary about 54.6 (Adrian Beaumont gets 54.8 by batched last-election preference flows) and I make the respondent-allocated 2PP to one decimal place to be 53.3 (ReachTEL published it rounded to the nearest whole number as 53).  

After the Queensland experience I don't think there is much use in placing any weight on 2011 preference distributions for NSW.  The ReachTEL figures above seem realistic and for now I will be using them.  

In summary, what we have from polling so far is as follows:

* No evidence of the primary vote gap narrowing yet.  It is currently at nine points.
* Evidence (and ReachTEL is about consistent with other pollsters on this) of a likely shift in preference behaviour, but nothing like on the same scale as in Queensland.

As a result, even using ReachTEL's respondent-allocated preferences and ignoring the 2011 preferences entirely, I am still getting the Coalition on 52.8% 2PP, almost four points above what the LNP got in very narrowly losing Queensland.

Well, that sounds like they should win rather easily.  But it's not that simple...

Sophomore Shortage

In most state elections that follow a big seat swing at the previous outing, you will see me going on a lot about sophomore effect and how a new government gets padding in its marginals by winning seats from opposition members then building personal votes for its own new MPs.  In the purest form (when an MP from one party faces the election and loses to the other party, then the new MP recontests) this is called "double sophomore" effect and is often worth about two 2PP points in that electorate.  

But this NSW state election is something of a sophomore desert, especially when compared to Queensland.  Without sophomore effect, Labor's win in Queensland would have actually been much clearer.

The Coalition won 29 seats from Labor at the 2011 NSW election, but in twelve of those cases the defeated Labor MLA was fleeing the sinking ship anyway, so Labor had no personal vote factored into those twelve 2011 results.  Of the remaining seventeen, the Coalition has since lost two in by-elections, and in three cases the sitting member has been ICACed.  This leaves only twelve real double sophomore seats, and most of these have been affected by redistribution, so that the new MLA on average has a personal vote across only 80% of his/her electorate.  Also, most of the double sophomore seats are well below the sacrifice line (it's assumed the Coalition generally won't be keeping seats on 7% and below), with nine on margins of 8.6% or less and three, oddly, at 19% and above.  Sophomore effect does still make a difference - chiefly to the Coalition's chances of retaining the seats of Coogee and Kiama - but on the whole there's not a lot in it. 

Indeed after spending hours putting all the various personal vote effects into a seat model, I found that personal votes aren't that much use for the Baird government in this election.  Allowing for them bolsters the government's position in my model by an average of just 0.9 of a seat - remarkably low given the damage inflicted in 2011, and chickenfeed compared to other uncertainties in this election.

Seat Model

The seat model I am using for this election is very similar to the one I used for Victoria and Queensland.  It attempts to model the chances of each side winning the Coalition vs Labor 2PP in a given seat, in order to draw overall conclusions about whose favour the deck might be stacked in.  It relies on a fairly accurate 2PP and if the 2PP is way out (as was the case in Queensland because of dramatic preference behaviour shifts) then the seat distribution won't work well.

The seat model has two advantages over pendulum-based models.  Firstly it adjusts for the impact of personal votes, and secondly it catches cases in which the pendulum model has one side winning a number of seats narrowly.  Because seat swings are not in practice uniform, if a pendulum model projects one party to win a bunch of seats each by very little, most likely that party won't actually win all of these.

The seat model assumes (pending evidence otherwise) that all four notionally crossbench seats will be won by their owners, so I am just dealing with the remaining 89 seats.  Actually I am far from sure the Greens will both win Newtown and retain Balmain.  I also think there could be losses to independents, perhaps even from both major parties (in Labor's case Wollongong is at risk) elsewhere.  But my basic assumption is that of the 69 seats it notionally owns (this includes Miranda) the Coalition must win the 2PP in 47 seats.  One or two losses to conservative independents within that 47 may be bearable; if the Coalition wins just 46 seats then it's hard to see a bush independent siding with a bunch of Greens and left independents to prop up a Labor minority mess.

A risk factor in this seat model is that it currently assumes that the swing will not be related to a seat's existing margin.  Under optional preferential voting there is a mild tendency for seats on larger margins to swing by more on a 2PP basis, and we saw this in Queensland.  In the case of New South Wales this is possibly counter-balanced by the demographics of Sydney.  There are massive concentrations of Liberal voters in the North Shore suburbs, and in those super-safe electorates Labor support is never all that great.  These electorates swung by less than the state average in 2011, so if they now also swing back by less then it is likely there will be some very large swings in other seats on large margins.  At the moment I've made the model a bit bouncier by raising the standard deviation for individual seats, after taking personal votes into account, to 3.5 points, but there could be some large individual outliers.  It's possible that with a state swing in the low teens there will be some seats with swings back of around 20 points.

Here's the model's current output for the 69 Coalition-held seats, assuming a state 2PP of 52.8.  I've not bothered showing the Labor-held seats as the model gives Labor a 100% chance of winning the 2PP in all of them, except for a 99% chance in the by-election-disrupted seat of Charleston.  For each Coalition seat, the swing required for Labor to win is shown, followed by the swing as adjusted for personal vote issues, followed by the expected probability of the Coalition holding the seat if the 2PP is 52.8. I welcome advice on which seats readers expect to buck the statewide swing and that I should consider whacking Caution notes on:

(Click for larger version).  Note that for Miranda I have adjusted the swing required downwards by six points for the impact of the by-election.   This adjustment is conservative given the size of the by-election swing.  For Tamworth I have treated the seat as National vs Labor, although there was a much closer National vs Independent contest in 2011.  Sportsbet indeed currently has former independent occupant Peter Draper a $1.50 favourite to recover his seat.  I will note that once independents lose their seats it is sometimes harder for them to recover them than the results of the previous election imply.

I've placed Caution notes, meaning I distrust the model's 2PP win chance estimates, on the following:

* The three far north coast seats of Tweed, Ballina and Lismore.  Ballina especially is receiving attention because of the retirement of 27-year Nationals incumbent Don Page, and because seats in this area have been behaving dramatically differently at federal and state levels. The model's standard adjustment for the loss of a rural sitting member (around four points) probably doesn't do justice to the impact of losing one there for so long, and there is some feeling that this seat could actually fall to either Labor or the Greens.  (If both Labor and the Greens are in the mix then they could, however, suffer from preference loss because of votes exhausting.)

* Drummoyne and Ryde, mentioned by Antony Green as examples of seats with vastly inflated margins from the 2011 debacle; there are probably others.  

* The Entrance from the list above.  Sitting member Chris Spence is another of the many catches of ICAC and so it's been assumed the seat will fall.  However, comparable cases such as Dobell (federal 2013), Frankston (Vic 2014) and Robertson (federal 2010) show that when a troubled sitting member is got rid of, the result is often closer than people think.  In this case, the Coalition still has a personal vote advantage in the seat, because in 2011 the seat had a Labor incumbent and this time it has none.  My model has the seat more likely than not to fall, but only by a whisker.

Overall the median outcome in this model is an underwhelming 51 Coalition seats, four more than needed for the barest majority.  The 51 seats includes Tamworth which the Coalition may very well lose back to Draper.  If the core assumptions of the model hold up, the Coalition will still almost certainly win the election - in Monte Carlo simulations with a 2PP of 52.8, they failed only two out of 2000 times.  However, as noted above, these assumptions are fairly easily broken, for instance if blocks of adjacent seats buck the state swing pattern.

The other thing I have noticed is how insensitive the seat tally is to changing the headline 2PP swing.  To get the Coalition's median seat tally in the model down from 51 to below 46.5 (ie a 50% chance of loss of majority) I have to increase the statewide swing back to Labor from 11.3 points to 14.6, ie a 49.5% 2PP for the Coalition.  On the other hand, if I give the Coalition a 56% 2PP, they only get 56 seats.  This is because of the relative lack of close seats compared to Victoria and especially Queensland.  If I give the Coalition an extra 2PP point by keeping the 2011 preference flow, their position improves by only one or two seats.  

Normally in the lead-up to elections I talk down the possibility that what seems to be a clear lead for one party could be scuppered by an uneven distribution of seat swings.  In New South Wales it seems the chance of a seriously uneven distribution is higher than normal, and that it indeed is now the major reservation about an election the Coalition should otherwise win.  On current voting intention levels they still should win anyway, but it is by not yet a done deal and a hung parliament is a significant chance.  Should voting intention narrow significantly before polling day, that will make things really shaky. 

Betting Markets

Betting markets made the same mistake us psephologists did re Queensland, panicked in the last days of the 2013 federal election and gave Labor virtually zero chance in South Australia, so you shouldn't pay too much attention to them.  I continue to monitor them to see if they're doing any better at election forecasting than we are.

William Hill (what used to be known as Sportingbet) has the Coalition at $1.05 to Labor's $9.  They have betting on the Coalition's seat total both in bands of five seats and as an exact seat result.  Both have a median of 53 Coalition seats and both have a slightly lopsided distribution that skews towards more Coalition seats rather than fewer from that point (eg 58 seats is slightly shorter odds than 48).  

Sportsbet has a headline of $1.08 vs $7.  They have a 2PP market which expects the Coalition to land in the 52s, with 51s slightly more likely than 53s.  They have an expected ALP primary of about 35.4 and an expected Greens primary of about 10.5.  They have a line bet which gives the Coalition even chances to win 18 more seats than Labor (William Hill's line bet is more conservative).  

Sportsbet individual seat odds have Labor recovering a measly 15.5 seats from the Coalition - everything below 8% plus The Entrance (11.5%) and a tossup in Gosford (11.9%).  However they do have seven seats where they expect the Coalition to hold with Labor at $3 or less: currently Coogee, Holsworthy, Mulgoa, Port Stephens, Miranda, Heathcote (19%!) and Ballina. Seats where Labor is favoured to gain but the Coalition is at $3 or less are Monaro, Oatley, Campbelltown and The Entrance.  As Sportsbet is currently giving Balmain to Labor (but not very convincingly) and Tamworth to Draper, this all adds up to 52.5 Coalition favourites, 36.5 ALP, and four others. 

Overall I think the Sportsbet 2PP market is a bit inconsistent with the rest.  If the Coalition win 53 seats off a 2PP of only 52.4 it will be they, rather than Labor, who has benefited from an uneven distribution of swing.  I cannot see why this should be the case.  

I am yet to find time to deal with the Legislative Council (see Antony Green's comments) or the curious case of the No Land Tax Party (which has of course drawn pole position on the monster upper house ballot paper).   I will be covering the NSW election live on election night. Saturday March 28, and expect to have extensive postcount coverage.

Other Models

The Alizarin Indigo model at Daily Kos has a rather more positive assessment of the Coalition's situation than mine: it is expecting 55 Coalition wins with four tossups, and it has a very different assessment of seats like Monaro and Oatley.

BludgerTrack has also now opened for business with primaries and a last-election 2PP more or less identical to mine but a much lower respondent-allocated 2PP (51.8) probably because it is using multiple respondent-allocated polls whereas I've just taken the distribution from the one with most detail (to this point).  BludgerTrack projects 54 Coalition seats from a last-election 2PP of 54.2 (my model gives 53 for that 2PP) or 51 off 51.8 respondent-allocated (my model gives 49 or 50).

I may fine-tune my model's personal-vote adjustments, which for the moment are copied from the historically-based adjustments I used for Victoria. Perhaps the behaviour of personal votes in NSW is different in some way.

Morgan Update (March 17)

A Morgan SMS poll has been released with some rather bold figures for the Baird Government: a primary lead of 13 points (46.5-33.5 with 11.5 for the Greens and 8.5 for Others).  This comes out to a big 2PP lead of 55.5-44.5 by 2011 preferences.

This poll has the highest Coalition vote and the lowest Labor vote of any of the 12 polls by five different pollsters now released this year.  It naturally also has the highest Coalition to Labor gap.  It should be treated with caution for this reason, and also given that Morgan-SMS polls in the Victorian campaign were very bouncy.

Nonetheless it is still data, something of which we don't have that much recently, and its addition to my aggregate gives the Coalition another 0.4 points on the 2PP vote and another seat on the seat projection (albeit only just; it was giving them 51.6).

Labor-Green Preference Deal (March 17)

The SMH has announced a list of 23 seats, all bar one Coalition-held, in which the Greens will preference Labor.  In pendulum order they are:

Kogarah (ALP 5.4), East Hills (0.2), Swansea (0.3), Prospect (1.1), Macquarie Fields (1.8), Granville (3.8), Oatley (3.8), Wyong (4.6), Maitland (4.9), Londonderry (5.3), Blue Mountains (5.4), Strathfield (6.4), Campbelltown (6.8), Kiama (8.6), Holsworthy (10.7), The Entrance (11.5), Gosford (11.9), Port Stephens (14.7), Penrith (16.1), Heathcote (19), Tweed (21.7), Lismore (24.3) and Ballina (24.6).

Coalition-held seats on margins below 20 points and not included in the deal are: Monaro (2), Rockdale (3.6), Coogee (8.3), Seven Hills (8.8), Mulgoa (12.4), Parramatta (12.5), Drummoyne (17.1) and Bega (18.5).  The disrupted seat of Miranda (23 last election, since lost in by-election and now vacant) also isn't on the list.

However, an online Greens HTV card checker (thanks to Nick Casmirri for pointing that out) shows that, at least online, the Greens are actually preferencing Labor in all those seats except Drummoyne. (Ben Raue also mentioned that the Greens are preferencing Labor in more seats than are in the deal.) Online at least, the Greens are preferencing Labor ahead of the Coalition in 83 of 93 seats.  The exceptions are Camden, Cessnock, Coffs Harbour, Drummoyne, Fairfield, Liverpool, Manly, Maroubra, Shellharbour and Wollongong.

The SMH speculates that this is "potentially worth up to four points" in given seats, but it's not clear what this is based on.  They do refer to the February Ipsos, which had a three-point gap between respondent-allocated and batched last-election preferences.  However that result (a higher gap than in the recent ReachTEL) tells us how voters basic intentions shift irrespective of preferencing decisions, and nothing about how how-to-vote cards impact on those decisions.

Antony Green conducted a thorough study in 2007 (in which the Greens preferenced Labor in 43 of the 73 seats that ended up ALP-vs-Coalition) and in that case Green preferences flowed 46.2% to Labor in seats where Labor were preferenced on how-to-vote cards, but 33.2% where they were not.  Overall the 2007 gain rate for Labor was .350 votes per Green vote in seats where preferences were allocated, and .197 votes per vote where they were not.  In 2PP terms this difference of .153 votes per vote magnifies to a difference in 2PP outcomes of about .17 points per Green vote percentage.  That means that for a Green vote of 10% in a seat, the preferencing decision could on average help Labor by about 1.7% in that seat.   I've seen no such study for 2011, in which the Greens preferenced Labor in very few seats.

It's not clear whether the how-to-vote card itself caused such differences in 2007.  For instance if branches had input into the 2007 decisions, then branches where Green voters were more likely to preference Labor anyway, would also be more likely to allocate preferences to Labor.

I did have some discussion at this point of the chilling impact on the overall preference flow of Greens preferences exhausting in seats outside the deal, but it appears that they will only do so in some largely irrelevant seats.   However, even if the decision had only had the impact of strengthening the flow of Green preferences in the targeted seats compared to remaining Coalition seats (even without changing the overall preference flow), that would itself have been significant.  It would have knocked about one seat off the Coalition's projection in my model.

And it's a potentially big deal in seats on inflated margins and with big Green votes, such as Heathcote, Tweed, Lismore and Ballina (though in the latter two cases Labor needs to come second to benefit!)  In those extreme cases, it really might be worth four points.

What we definitely shouldn't assume is that Green preferences going to Labor ahead of the Coalition in 83 seats will suddenly give Labor, say, 1.5 points extra statewide to what we should have expected before this announcement.  It may well be 1.5 points extra compared to if the Greens had run an exhausting HTV in every single seat, but that would certainly have had a chilling effect on preference flows compared to what respondent-allocated polling is suggesting.  That is, after all, the standard way in which respondent-allocated preferences have got it wrong in the past.  Whether the Greens' decision could even cause Labor to outperform what respondent preferences suggest at all is in my view rather dubious.  Similar decisions haven't done so before.

An important thing here is that in optional preferencing the impact of a Greens preference decision may be - and probably is - greater than in compulsory preference elections, where Greens tend to preference Labor to very similar degrees no matter what the party does.  That explains why the difference between a recommendation to exhaust and a directed preference could have been so large in 2007 - larger than the difference between an open ticket and a directed preference in federal elections.

Lonergan and Tamworth ReachTEL (19 March)

A Lonergan "phone" poll (robopoll) with sample size 1565 was reported by the Guardian yesterday, although bizarrely the Guardian reported everything but the most important bit, the primary votes, which were Coalition 46 Labor 34 Greens 10 Other 10.  That was calculated as 55:45 by 2010 election preferences.

Of what the Guardian did report, nothing is too surprising: Baird leading Foley 53:23, voters don't believe the corruption problem has been solved but think the Liberals are more likely to do so, nobody much likes fracking or coal seam gas extraction, and 29% say they intend to use their vote as a protest against the federal government. 43% say they are less likely to vote Coalition because of Tony Abbott, but as usual there would be some who would be more likely, and relatively few on either side would actually switch their vote over it.

A commissioned ReachTEL on attitudes to coal mining and CSG in the Tamworth electorate has been seen.  The usual caution applies to results of polls commissioned by a group that stands to benefit from the results, but I don't have any issues with the question wording, save that questions 2 and 3 would likely lead to a more negative response to question 4 than if those questions had been about perceived positives of mining.  Anyway. for psephologists the interest here is the Draper vote, but there is no question released on voting intention and it is not possible to accurately derive the Draper vote from the data available.  It does however look like Draper is at least competitive.

Barwon (20 March)

Independent candidate for the immense Nationals stronghold of Barwon, Rohan Boehm, claims to have a ReachTEL showing "that [incumbent Kevin] Humphries' primary vote has halved and that on preferences he would face a clear defeat."  As Humphries polled a staggering 79.1% in 2011, I'd be interested to see the figures (which are at this stage unreleased), since under OPV opposition would need to be strongly concentrated in one candidate for Humphries to lose with even half that figure.  Sportsbet has Boehm at $11 and Humphries at $1.01.  There is a lengthy piece about Barwon here and several solid media grabs for the Boehm campaign can be found online.  I make the observation that moving voter intention in such a massive electorate is a more challenging task than in smaller rural electorates, but this seat is worth keeping an eye on. In 2007 when its previous incumbent retired, independent Tim Horan came within 6% of poaching it.

As with other recent elections there is a massive volume of internal polling being conducted that is mostly not publicly released.  Claims about such polling should always be treated with caution (especially when not all figures are seen) but I think it is worth mentioning them.


  1. Dr. Bonham,
    Many thanks for this thorough and cogent analysis. You might not be aware that the Liberal Party has recommended its preference in the seat of Wollongong go to Independent Arthur Rorris, the well known head of South Coast Labour Council. As he has also been given the Greens preference, Labor incumbent Noreen Hay could be in trouble. All the same, she's a slippery campaigner who's survived strong challenges before, despite incessant exceptionally negative reportage in The Mercury.

  2. Ta; I meant Wollongong where I wrote Newcastle; fixed now. Very interesting contest.


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