Friday, November 28, 2014

Victorian Poll Roundup: Endless 52-48 Edition

2PP aggregate: 51.8 (-1.4) to ALP by last-election preferences (52.6 expected preferences)
Current seat projection based on aggregate: 48-40 to Labor

This is probably the second-last of my posts on Victorian polling prior to the main day of polling for the election on Saturday.  I will be out on a field trip in the day on Friday.  Another post is expected either overnight Friday or Saturday morning - once I am satisified we have the results of all the major polls we're going to get.   Then it will be on to the live coverage on Saturday night from 6 pm.  I cannot promise you fancy graphics or dancing swing charts but I hope my live blogging will be of some interest.

(Speaking of things of some interest concerning Victorian elections, a plug for Adam Carr's page on early Victorian elections 1851-1864!)

This week's polling

Lately we've had new state polls from Morgan (twice), Essential and Galaxy.  The first Morgan SMS poll taken from 21-24 Nov and with a sample size of 1152 produced a 52:48 2PP off primaries of 39.5 to the Coalition, 33.5 to Labor, 17.5 to the Greens and 9.5 to others.  This series had favoured the Greens by an average of at least five points compared to other pollsters (indeed 17.5 was its lowest reading for them so far), so it was hard to tell whether this result signified that trend waning slightly, or that it was actually a not so good sample for the party.  The series also clearly skews against the ALP compared to other pollsters.

The Morgan was widely reported as showing a narrowing in the election race but this was mostly fallacious - the drop of three points from the previous Morgan was from a result that was abnormally high compared to other pollsters.  The Morgan showed Denis Napthine recpaturing a slim preferred premier lead from Daniel Andrews, but it is difficult to compare Morgan's readings to other pollsters who allow an undecided option.

The Essential poll represented three weeks of data rolled together (Nov 7-24) with a sample size just below 800 and had a 53:47 2PP (a correction of the 2PP to 52:48 turned out to be incorrect.)  The primaries of 40 L-NP, 39 ALP, 13 Green, 8 Others were identical to an earlier Galaxy that had a 2PP of 52:48  (the difference may be explained by rounding and/or different 2PP calculation methods).

The Galaxy of which details are just emerging also has those same primaries and again has a 52:48 2PP!  Significantly though, Denis Napthine's preferred-premier lead over Daniel Andrews sinks to a mere three points (41-38), a lower than expected lead for that set of voting intentions and a further indication that the Coalition's attempts to brand Andrews as an unsuitable, dangerous or unelectable leader have more or less totally failed. 

Finally, the last Morgan SMS threw a minor curveball with a 50-50 2PP (the first such from anyone for a very long time) off primaries of 44 Coalition, 36 Labor, 13.5 Green and 6.5 Others.  While Morgan says Labor are favoured to win off those results, any model that takes personal votes into account will show that 50-50 would probably not be enough to win.  As usual any unexpected Morgan finding is accompanied by an essay explaining why the poll respresents the absolute truth of the electorate rather than one about it being perhaps an outlier.  Morgan even claim that John Howard's appearance on the campaign trail (in a state that never much liked him) diverted voter attentions from the Abbott government.  Fortunately I am saved from having to express any original ridicule upon this point by this:

Attitude questions in the Essential poll are reported in the full report.  The first thing to notice here is that the inclusion of the Greens in the "trust to handle issues" section disadvantages Labor compared to the Coalition, since many Greens supporters would trust Labor over the Coalition on the listed issues.  The Liberals' leads on energy, public transport and support for regional Victoria are thus illusory and their other leads are closer than the poll shows them to be.  That the Liberals are preferred on the economy, roads and cops and Labor on health and education can all be filed under business and normal, so the stat that really deserves attention is Labor's lead over the Liberals on "unemployment and jobs".  So often in Australian polling lately we've seen the economy and jobs marching together in the service of the Coalition.  Not this time.

The Galaxy showed a thumping 60-29 support level for the proposed East-West link, but a tweet from Antony Green shows you all you ever need to know about the way this issue plays out:

The outer suburbs like the link but the seats where Labor is trying to fend off the Greens cannot stand it.  This probably isn't just NIMBY mode, but also that green-leaning inner city dwellers often aren't mad-keen on large-scale road projects wherever you might build them. If Labor loses, it will suddenly be immediately obvious to everyone that the East-West link was why.

Pre-Election Exit Poll Ahoy!

In an exercise I've never seen before, Galaxy have been commissioned by Trades Hall Council for an exit poll of early voters (of whom there are a hell of a lot this time around - I'll probably cover that on Saturday).  The poll of the four sandbelt seats of Bentleigh, Carrum, Frankston, Mordialloc shows a 2PP of 52-48 to Labor from a sample size of 602.  In 2010, early voting notably favoured the Coalition in both Bentleigh and Mordialloc, and might have done so by a shade more had there not been a late swing to the Liberals.  All up, the early votes in these seats in 2010 came in at 53:47 to the Liberals.  So that's an average five-point swing, if the poll is reliable. (It's a commissioned poll, hence caution, but the previous THC-commissioned Essential polls all looked OK to me.)

The pattern of early voting favouring conservatives is nothing unusual and given that this poll has been taken before the early voting period ends, it might be expected that this underestimates what Labor should get on the day.  On the other hand it might be that since more people are voting early, those doing so are more representative than usual.

In any case, there's no warrant on this poll alone for assuming Labor will win all four seats.  The sample size would only be 150 for each and it may well be there is variation between them.  Also, given the four-point MOE for the sample overall, it's still possible on this sample that the Liberals might be holding most of the sandbelt.  But, as always, it is better not to poll such adverse results than to poll them.

The other thing highlighted here is that so many votes will be in the can by polling day.  With record early voting (looking like it will settle at thirty-something percent) the impact of any very late swing will be muted.  Unless the polls are almost all getting it wrong, a quarter of the vote is already in the can for Labor and the Liberals need to get enough more on the day to counter it.

Are The Polls Herding?

In the recent US mid-terms polls displayed a strong skew to the Democrats, shown by Nate Silver to be potentially causable by herding. (Silver's article is very well worth reading in full.)  After seeing so many 52:48s in this campaign (seven of the last 14 polls!) I've been running some fake-poll Monte Carlos to see if the polls so far do look too tightly clustered.

Fortunately, I've been unable to find clear evidence of the polls being too tightly bunched together, though at times the percentage of simulations that are more closely bunched than the actual polls has been small enough for it to be close to statistically meaningful if I was looking at a random election (rather than one I was already suspicious about.)  Morgan has been volatile while Galaxy and Essential have been their usually under-dispersed selves, and really that's been about all there's been to it.  We also know from other elections that while Essential does tend to move slowly, it also goes off on tangents compared to other pollsters.

The spread of poll readings has been less than expected by random chance, but not enough to be really suspicious.


Throwing this week's polls into the mix and dropping old polls I'm now looking at the Essential 53, the Morgans 52 and 50, the Ipsos 54 and the two Galaxy 52s.  My aggregate weights Galaxy very heavily and the Ipsos and Essential are downweighted for age, with the Morgan 50 weighted higher than the 52.  (I've used my usual formula, except that I've declared it to be Saturday already.) It shouldn't thus be greatly surprising that the 2PP by 2010 election preferences comes out at 51.8, especially given my corrections to the Green vote.  This is off primaries of Coalition 41, Labor 38.8, Green 12.3 Other 7.9.  The difference between the majors looks a little high to me and will probably go down if the Newspoll and expected Ipsos pour cold water on the Morgan, which I predict that at least one of them will.  But for now that is where  have it.

The handling of the Green vote is extremely difficult.  Established pollsters like Essential, Galaxy, ReachTEL slightly overstated it in the federal election and have consistently had it about 13 leading up to this one.  Ipsos have had it much higher in two polls so far but are new, and Morgan has had it extremely high until the final poll which is conveniently close to the others.

The shift in the Morgans has been so strong that I wonder if they have herded any assumptions to correct the feral overstatement of the Green vote in their early polls, and for this reason I am reluctant to ping the 13.5 in their last poll as much as I have been pinging the others.  I would be less cynical about all this were it not for Morgan's habit of switching from Labor-skewed methods to non-ALP-skewed methods in the shadows of the post at federal elections.  Still, I guess no-one could accuse their SMS series of a pro-Labor skew!  Indeed if the change in their Green vote is due to any methods tweak then this may have skewed the 2PP of their latest poll, since their last five SMS efforts have had the lowest ALP primaries of any pollster for a very long time. 

Preference Strength Again

The other big issue is the strength of preferences.  While I have noted the 51.8 aggregated 2PP based on preferences from the 2010 state election, the general view among psephologists is that the strength of preferences to Labor will be stronger than in 2010.  I covered this in the Late October roundup, but a further development since has been preference-dealing for this election.

In a review on the Crikey site (paywalled) William Bowe noted that Labor has done preference deals for Upper House preferences with the likes of the Country Party and (in a sure sign that hell has frozen over) the DLP.  Labor is however struggling for preference flow in Frankston and a vengeful ex-member wreaking havoc in Ivanhoe. 

Minor party voters aren't strong HTV card followers generally (the follow rates are around 40-45% for the major parties, but significantly lower for micros because Greens voters tend to make up their own minds on preferences and micro-parties tend to struggle for full booth coverage).  So maybe the card games won't matter too much, but I'm expecting changes in preference flow anyhow.  To be cautious on this point, for both Greens and Others I'm assuming as my core model that we will see a preference flow change that is about 60% of that seen between the 2010 and 2013 federal elections.  That brings us to a 2PP of 52.6% (and it may well go up if the Morgan is an outlier).  If I assume preferences exactly as per the federal election (excluding preferences from voters for Independents because of the McGowan tactical voting factor) this goes up to 53.2%.

The other big preferencing story is that the Greens lodged open HTV cards in every electorate and the Liberals in the four inner-city seats eyed off by the Greens.  Lodging of a card doesn't mean that card will be handed out, but means it can be.  If these cards are handed out, the flow of Liberal preferences to Labor in the inner city seats could decrease sharply, and this could well cause the Greens to at least win Melbourne.  On the other hand, so few Greens supporters are influenced by how-to-vote cards that the damage if the Green open cards are handed out will be limited.  What exactly the Liberals are up to with their open card is a big mystery, but we will find out on the day whether this is a sneaky plan to gift Labor an unwanted crossbench, or whether it was just all mischief-making and mind-games.

Seat Model

There's not been much seat polling so my model is pretty much starved on that front.  While I am not directly including data from commissioned polls I do feel the THC exit poll justifies a slight watering down of my loading of the Galaxy seat poll of Bentleigh.

This is the current output for a 2PP of 52.6%:

It's not quite time to switch this over from "nowcast" to "forecast" mode just yet but I'm not expecting the late polls will change it greatly.  I won't be surprised if they are ALP-friendly enough to kick it up from 48 seats to 49, in which case my suspicion is that rather than Labor holding Ripon, Labor will more likely pull off some random seat from the 5-8% layer that the Liberals are apparently now trying to sandbag.  Indeed it's quite possible they could get a couple of those and miss one or two further down the tree.  Almost certainly, the underdog according to the seat model will get up in a handful of seats. 

We still don't have reliable polling information on which to judge the chances of the Greens picking up a seat, and none to judge any independent chances in Morwell (Tracie Lund) or elsewhere.  I can only stay with my previous impression: Green chances in Melbourne should be taken very seriously, but other electorates are not easy for them. 

(If we stick to last-election preferences strictly, then the model likes 47-41).

Quick Betting Watch

I don't have time to do a thorough betting roundup tonight but Labor remain heavily favoured, with the $5.50 or so on the Coalition indicating some caution about the relatively close state of the polls.  The contest between seat betting and polling models is a fizzer this election as we're all getting about the same thing.  The Sportsbet exact seat market has 47-48 Labor seats as a narrow favourite over 49-50 (bear in mind this factors in chances of losses to the Greens) and projects an average of 48.7.  The Greens are still at a mere $1.55 to get nothing, with the market still expecting their primary vote to be below 12%.  The projected 2PP has come up to 52-53.  The ALP are favourites in the same 48 seats my model has them favourites in other than Ripon which as noted will likely be an override anyway.  Tracie Lund in Morwell is at $8.  My guess is the bookies have had people working for them setting initial odds using very similar models to this one, and no-one is betting enough otherwise to push them off those. (Oh, and sadly the Mordialloc arbitrage eventually disappeared, with Labor favourite on both markets.)

Everything (external polls, reported internal polls, pseph models, historical models, betting, behaviour of the parties and commentary) is all pointing to a highly likely moderate ALP win.  Are these expectations all herded together and is the chance of something unusual happening higher than it appears?  It doesn't look like it and it's hard to be surprised at reports that some Liberals just want it all to go away.  But much will be revealed on Saturday night, and most of the rest when the early votes are counted around Monday.  Back with more Saturday morning.

(Oh, and in possibly the biggest surprise of the campaign, The Age has, albeit backhandedly, endorsed the Coalition!)

No comments:

Post a Comment