Monday, May 9, 2016

Poll Roundup and Seat Betting Watch: Election Called

2PP: 50.3 to Coalition (+0.6 in two weeks, -0.1 since end of last week)
Coalition would probably narrowly win election "held now" (seat projection Coalition 79 Labor 67 Others 4)
Updated after Essential

On Sunday the Prime Minister, Malcolm Turnbull, visited the Governor-General and asked him to dissolve both houses and call a double-dissolution election for the long-expected date of July 2.  I hope the G-G asked him if he was absolutely sure the High Court would dismiss the Senate reform challenge case first, but in any case the G-G is in no position to be refusing such requests.

In the last week Scott Morrison delivered his first Budget.  So far if there's a budget bounce at all, it's very small.  We've got Ipsos at 51-49 to Coalition (up one on last month) but Newspoll unchanged at 51-49 the other way. We've also had two post-Budget appearances from polls not sighted for a while - a federal Galaxy at 50:50 and a Morgan Phone at 51:49 to Coalition.  In pre-budget days we had a 50.5 to Labor from Morgan multi-mode (by last election preferences, 51-49 respondent) and a 52:48 to Labor from Essential, which has mostly leant strongly to the ALP since mid-November last year.

The post-Budget ReachTEL has a 0.7 point primary vote swing from Labor to Coalition on last month's.  While the 2PP in the ReachTEL was reported as just 50-50, that's a result of firstly ReachTEL shifting to respondent preferences, and secondly ReachTEL respondents giving Labor a much better share of preferences than has usually been the case in other polls (though Ipsos this week has concurred).  I calculate the ReachTEL 2PP by last-election preferences at 51.6 to Coalition, or 51.4 applying ReachTEL's previous adjustment for the Nick Xenophon Team.

I'm aggregating the  ReachTEL as 51.6, Ipsos and Morgan phone at 51, Morgan multi at 49.6, Galaxy at 49.8, Newspoll at 48.8 and Essential at 48.7.  The ReachTEL put the Coalition back in the lead on my aggregate after sixteen days in arrears, and I now have the government at 50.2.  Here's the smoothed tracking graph:

The Galaxy was the subject of some poor poll reporting from someone who I would have hoped knew better. (May be paywalled).  A 50-50 poll does not mean there is a 50-50 chance of either side winning!

Firstly if the 2PP were exactly 50-50 on election day, the Coalition would be not a tossup for a win but highly likely to do so (Labor's target for a 50-50 chance is probably almost a point higher - see What 2PP Vote Does Labor Need To Win?)  Second, even without the deck being in the Coalition's favour, a poll taken this far out is a snapshot of current voting intention, not a prediction of chances.  To convert a sample of public opinion into an estimate of victory chances it's necessary to look at the history of how governments, and how the parties, tend to perform relative to their polling at a given time from the election.  Even then, doing this from a single poll rather than an aggregate would be very inaccurate.  I haven't yet rerun my historical regressions but I expect they'd still give the Coalition about a 75% chance of winning from this point.

Of course it can easily be argued that there are election-specific factors that make Labor's chances better than that, but none of that is proven by polling.  The polls plus history are telling us the government should be returned.  Whether those patterns actually hold up this time, or whether the government runs a bad campaign and collapses, or whether the polls might even just be wrong (like they weren't last time when many people thought they could be) - this all remains to be seen in the weeks ahead.


A brief roundup of leadership figures:

* Newspoll has Turnbull on a net satisfaction rating of -11 (38-49), his first improvement since November.  Shorten is on -19 (33-52), his new least worst since August.  Turnbull still has a large "better PM" lead considering his party's 2PP polling (49-27).

* Ipsos has Turnbull down five points to +8 (48-40) and Shorten up eleven to -11 (38:49). Turnbull has the same beauty contest lead as in Newspoll (51-29).

* Morgan phone has Turnbull at +2 (43-41) - bear in mind this poll once had him at +50 - and Shorten at -15 (34-49).  Turnbull is preferred 57-24.  

* Morgan phone has Turnbull (41) preferred Coalition leader over Julie Bishop (24) and all the rest have single figures.  This is similar to what Turnbull was polling for the post before he was installed in it.  On the Labor side this series always produces wretched reading for Bill Shorten fans, and he (14) again trails Tanya Plibersek (22) and Anthony Albanese (20) as preferred Labor leader.  Supposedly 8% still prefer Wayne Swan!

* Not quite a leadership stat but Galaxy found only 48% could name the Treasurer correctly and only 18% could name "Labor's Treasury spokesman".  (Perhaps had they said "Shadow Treasurer" Chris Bowen would have rated higher there, or perhaps not.)

Not so much is being made of Shorten's unpopularity these days, especially with his ratings on the rise.  Historically, Opposition Leaders who have been unpopular rarely win at federal or state level, but this is mainly because they are often removed and/or leading uncompetitive parties.  I think that Shorten is not really comparable to most other Opposition duds, because part of his poor rating history is that he is disliked by strongly left-wing voters for not being far enough to the left.  This might be seen as part of a global tendency (Sanders, Corbyn) for a large chunk of the left to crave something less safe from their side of politics.

Budget Polling

The most critical Budget poll is the Newspoll because of the very long history of the same two questions being asked by it: whether voters think the Budget will be good for the economy and whether voters think they will personally benefit.  The budget scored +5 (34-29) on the economic front and -21 (18-39) on personal circumstances.

The latter probably sounds bad but in fact it's pretty normal.  Voters are usually negative about their own finances.  It's the economic rating that is weak - it rates in the bottom half for all Budgets, and below all but four past Coalition budgets.  However given the new Newspoll methods, it's not clear how comparable the figures are.  Here's an update to the Newspoll budget poll graph, showing where 2016 fits:

Ipsos found 37% of voters considered the budget fair (43% unfair) and 24% thought they would be better off to 39% worse off.

The first polling on the Budget came from ReachTEL which found 26.2% rating the budget good, 38.9% average and 34.9% poor.  Responses were, as usual, party-polarised.  There is no comparison available for the 2015 Budget since ReachTEL didn't ask exactly the same question.  We can however compare how voters feel about the impact on themselves, with just 7.1% (much lower than other polls) thinking they'd be better off and 33.4% worse this time (compared to 16.4% and 30.3% last year).  Coalition supporters were especially less likely to see their economic prospects improving.

Galaxy found that voters were easily whipped into outrage that "only workers earning more than $80,000 get a tax cut" (28% agree, 62% not) and that almost everyone likes Labor's plan to continue the Deficit Levy on high income earners (63:21).  (Had the question not specified that this was Labor policy, the response may have been even stronger.)

Essential respondents gave the Budget a net -9 response (20-29) compared to +1 last year.  Both the neutral and don't-know responses were much higher than in the last two years, suggesting that the Budget has been overshadowed.  I lack the time for a detailed summary of Essential's budget polling but the approval for company tax cuts and internships for the long-term unemployed were interesting.  Scott Morrison's advantage over Chris Bowen as preferred Treasurer increased to 31:20.

Seat Polling

I have not seen any neutrally-conducted seat-polls in the last two weeks.  There was a Seven News story about internal polling supposed to show the Liberals with the following 2PPs: Eden-Monaro 46.9, Reid 50.3, Banks 50.9, Gilmore 50.2, Bennelong 51.6, Lindsay 51.2, Hughes 58.8, New England 53.1 (vs IND).  This was accompanied by reports of a declining trend in some of these seats, so it sounded like some sort of tracking polling.  These polls were attributed to the Liberals but the government denied knowing anything about them, as did usual Coalition pollsters, so where these figures come from, the methods behind them and even if these polls exist remains a mystery at this stage.

Anyway, seat polling was frequently cack at the previous election, and this sort of dire internal polling report (always supposed to show something you wouldn't see in the national polls) is a polling trope as old as the hills and with a dismal predictive record.  But what's most striking about this lot is that the 2PPs in the first six listed seats actually aren't much worse (on average maybe one point) than should be expected based on current national polling.  Indeed on current national polling Eden-Monaro, Banks and Gilmore are all more or less tossups.  The alleged poll got attention because of the sacred bellwether status of Eden-Monaro, but one of these days someone will finally win that seat and lose the election.  Its past history does not make it immune to local factors.

Betting Watch

The usual disclaimer: the purpose of this section is to monitor the effectiveness of betting odds, and not to encourage either betting nor the belief that betting markets are strongly predictive.

The headline rates as of Friday again had the Coalition between a 69% and 72% implied chance of winning depending on the market.

However there have been massive moves on the Sportsbet seat betting markets. Labor is now favourite in 14 Coalition-held seats.  Labor is not even considered close to losing any seat now (apart from one apparent error) while the market is giving it realistic shots in another 20 Coalition seats.

Loss (Coalition to Labor): Barton*, Paterson*, Solomon
Close Loss (Coalition to Labor): Dobell*Eden-Monaro, Macarthur,  Petrie, Capricornia, Hindmarsh, Lyons, Deakin, Hasluck, Swan, Burt
Close Loss (Labor to Coalition): Lingiari (??)
Loss (PUP to Coalition): Fairfax

Coalition Close Holds: Banks, Reid, Page, Gilmore, Lindsay, Robertson, Macquarie, New England (vs IND), Bonner, Brisbane, Forde, Leichhardt, Herbert, Dickson, Longman, Corangamite, Dunkley, La Trobe, Cowan, Stirling

Assuming Lingiari is a misprint, that makes just 77 Coalition, with 69 Labor and 4 others based on individual-seat favourites, but the number of close holds suggests that the collective wisdom of the seat-betting markets is probably a hung parliament.  In coming weeks I'll have a colour chart showing the movement in these seats over time.

The Sportsbet 2PP market is still, however, crediting the Coalition with about 52.5% 2PP.  Perhaps the reason it has not come down is that the rake on this market is outrageous so nobody with that much sense goes near it.  The William Hill exact seats market which is a slightly less (but still) severe scalp of the poor punter has an implied 77 seats for the Coalition now, and also hasn't moved a lot.

I will update the 2PP estimate tomorrow night post-Essential, and may add further comments through the week.  I am on remote fieldwork this week (I expect this to be the last week of it pre-election) and finding time for all this is difficult!

(Update: Essential moved from 52-48 to Labor to 51-49, and this caused the Coalition to gain another 0.1 on my aggregate.)


  1. Why are you convinced Lingiari is a missprint? It has a Labor majority of 0.9%, and the result there is always subject to the changeable winds of Indigenous politics. It's possible that people with local knowledge think Snowdon could lose.

    1. It's also possible the odds were real and were affected by a single rogue punter; such things sometimes happen. I see it's now switched back to a close Labor hold (1.67 vs 2.20). I would have thought the extreme disarray of the CLP government would have made things very difficult at federal level in those parts of Lingiari that are pollable (even though they are only a minority). I have seen some obvious misprints that get swapped back on Sportsbet before.

  2. Hey there, great article as usual!

    One thing I've noticed after immersing myself into the political scene and watching psephologists such as yourself engaging in ... cautious predictions about upcoming elections based off of polling data/seat data etc is what aside from opinion polling on the Friday prior to elections, opinion polling leads psephologists to understate seat swaps rather than overstate them (aside from the SA State Election, but that was weird for other reasons altogether).

    My question is: Is there some subconscious bias towards conservative predictions in seat gain/loss numbers? Or is the data itself naturally more conservative? Or is there some other reason for this phenomenon to occur?

    Thank you!

    1. I wonder if this is a pattern on the whole or not, and how strong it is if so. I slightly overestimated the seat swing in 2013 for instance, and quite a few others did too. Experts overall greatly overestimated the seat swing in 2007 (with some exceptions) but underestimated it in 2010. Estimates of the seat swing in NSW and Vic were generally about accurate. Queensland at the last two elections would be a case in point of overcaution - in 2012 possibly because some were struggling to believe their eyes and in 2015 because of over-conservative assumptions about preferences.

      Usually we tend to assume that certain things (preferencing, personal votes etc) will stay about the same as in the past unless there is very strong evidence otherwise. Some assumptions will be wrong but if the errors are in a range of directions they can cancel out. Blowouts can happen when everything that we make middle-of-the-road assumptions about goes the way of the same party. It's possible also that modellers tend to underestimate the chances of what seem to be "freak" results - things for which there is no historical precedent.

      Concerning the current election while the median projection at the moment is a narrow Coalition hold there is still plenty of time for either side to possibly win convincingly. It might not seem all that likely from here but past examples (eg 2004) show that a lot can happen during a campaign.

  3. Your comments on this welcome

  4. I think the chance of the Greens dropping down to one seat is fairly low. Their vote was already damaged by the issue he refers to in 2013 and they will get preferences from left-microparty voters. I agree with all the rest.

  5. With the calling of an election comes the hilarious attempts at filling the 24 hour news cycle.

    Funniest thing i've seen in a while: Liberals in danger of polling below 15% in South Australia (or maybe 13%, given preferences).

  6. Kevin why is there a discrepancy between the national polls and the pols in 'marginals. I doubt that safe seats are going to Labor as implied

    1. Do you mean the marginal polls or the marginal seat betting? If the former I'm not really aware of any marginal-seat polling - at least not neutrally-conducted marginal polling - showing safe seats being lost. If the latter the main answer is Western Australia where it is believed that swings of 6-9% could claim some normally relatively "safe" seats.

      There was a big discrepancy between national polling and seat polling in 2013 though some pollsters were more extreme in that regard than others. The reason for it was never really adequately explained.

  7. Sorry Kevin

    If say the polls say it is 51-49 to the ALP but a few marginal seat polling has marginals holding firm does this not mean ipsofacto that there must be some big swings in 'safe' seats to the ALP?

    1. From what I can see the average seat swing in 11 more or less marginal seats that have been polled by neutral agencies is 1.8 points to Coalition. However six of those are double-sophomore seats and another two are single-sophomore so this muted swing is about what would be expected off a 2PP of about 50-50 - which is where I actually think it is. The other thing to note here is that none of these "marginals" are in WA and in WA even seats that are not really "marginal" are in the firing line.

      There will usually be a weaker swing in Government first-term marginals than elsewhere - this only adds a few tenths of a point to the swings in safe seats on average.