Sunday, March 18, 2018

2018 South Australia Election Wrapup And Postcount

SEATS APPARENTLY WON LIB 25 ALP 19 IND 3
(Mawson was in doubt but retained by Labor) 
Expected Legislative Council result 4 Liberal, 4 Labor, 2 SA-BEST, 1 Green

This thread will provide some general comments on the South Australian election and will also follow the post-counting in the few seats in doubt.  The post-counting comments will not be updated all that regularly as I took three days off work to follow the Tasmanian post-count and should probably get back to earning some money.  I'll try to check every day or so to see if there's anything worth noting.

The Liberal Opposition led by Steven Marshall has won the election, and is more or less certain to have an outright majority.  If it did fall short in a seat somewhere because of some freakish postcount result, Troy Bell could be counted on for support.

This is only the seventh case since 1969 of an Opposition winning an election while the same party is in power federally; for the previous six see here.  On the other hand, it confirms two other historic patterns: that governments no longer seem to go on forever (it is now 32 years since any state or federal government older than 16 years was returned) and that unpopular state premiers don't get re-elected.



At least on the figures for the classic (Labor vs Liberal) seats, the Weatherill government appears to have received a small swing back to it from its somehow winning 2014 2PP of 47%.  On raw figures the swing back is 1.7% but after declaration votes I expect it to drop back to around 1.2% (I may attempt a 2PP estimate at that time). However, much of the swing to the government was wasted in safe seats such as Croydon, Elizabeth, Kaurna, Playford and Reynell.  As a result the swing back does not seem to have delivered any net seat gain at all.

Much will be said about the failure of the SA-BEST party, which started off polling incredibly well but continued to tank through the campaign and has won at most one seat.  Following a similar failure by One Nation in Queensland the issue will be whether minor party uprisings are failing because of mistakes by the parties in campaigning, or whether the two-party system is so stacked against them when it comes to single-seat contests that they have no chance.

Poll Accuracy

This was a rather bad election for opinion polling.  The final Newspoll had the Liberals on 34% but the Liberals are currently on 37.4% and likely to finish close to 38% should postcount patterns from 2014 hold.  While the continual sliding in the SA-BEST vote meant that SA-BEST was probably still shedding votes even after the final poll was taken, the Newspoll overestimated third parties generally, also having the combined Greens and Others vote around three points too high.

Whether the final ReachTEL was any better is hard to say.  Perhaps it was, but the innumerate presentation of the poll by clueless Sky News presenters lumped the ReachTEL "undecided" vote (in reality voters with a soft preference for a party who should be allocated to that party) in with Others.  Had they not done this, it's possible the poll would have shown a more accurate picture of the total major party vote, though it is unlikely it would have had the major party gap any more accurate.  Once upon a time ReachTEL used to promptly publish full details of its media polling on its site and it's a pity this is generally no longer done.

As for seat polling, the YouGov-Galaxy series during the campaign did correctly predict the winner in nine out of eleven polls taken.  (I count a 50-50 2PP as correct if the margin lands within two points either way.)  However some of these were easy targets and the average error on the winner's 2PP is currently running at just over four points.  These errors aren't confined to the non-classic contests affected by the SA-BEST nosedive, with 5-point errors also in Lee and Morphett.    The less said about the pre-campaign seatpolls, the better.

As for statewide 2PP polling, I can't comment on it because there wasn't any.  One would have thought this was a fine opportunity for someone to test whether respondent preferencing could obtain an accurate outcome; apparently not.

Postcount 

ECSA is expediting postcounting in Adelaide, Mawson, Newland and Heysen, so we should get progress on them on Monday; for the rest things won't get serious until Tuesday.  The most important thing to know here is that post-counting in SA typically favours the Liberals.  In 2014 their 2PP rose by an average of 0.5 points per seat.  There were only six classic seats where it fell at all, and in four of these the drop was trivial (Light with 0.5% was the largest Liberal fall).  There were some seats where the Liberals gained greatly on the 2014 postcount, such as Giles where they gained three points.

In a few seats - but fewer than I feared - the ECSA has the wrong pair of candidates for the two-party count and will have to realign the two-party count (when this will happen I don't know).  This is the case in Gibson and Morialta, and possibly in Hartley where Nick Xenophon could be overtaken by Labor on Greens preferences.  Looking at the 2016 federal election, I've found that Labor candidates gained on Xenophon candidates using Green preferences in most but not all cases, with the maximum gain rate being about 0.4 votes/preference.  In some cases, Xenophon candidates gained (Greens voters aren't big card-followers), so Xenophon isn't certain to be overtaken.  In Heysen it's unlikely on that basis that Labor would get into second.

Seats that I - at some stage -  consider to be in any significant level of doubt (or just interest) will be listed individually with comments below.  Because of the large size of the postcount and the potential for the redistribution to interfere with post-count patterns it is possible new seats will be added to this list.

Adelaide

This seat was barely mentioned all night but finished with Rachel Sanderson (Lib) leading Jo Chapley (ALP) by just 67 votes (50.22% 2PP).  In 2014 the Liberals gained by 0.77 points in the postcount so if anything like that happens again they will win easily.

Sunday 4:45: Antony Green has tweeted that Labor are now 125 ahead, following corrections during checking including a significantly incorrect figure at one booth. That lead would be unlikely to survive the postcount based on past patterns.

Monday night: Sanderson has gone to a 539 vote lead (51.5%) and while that doesn't include absents yet so while there is some room for recovery that is pretty much the end of that one.

Mawson

This is a greatly redistributed seat where Leon Bignell (Labor) is trying to hang on despite it having a notional Liberal margin of 4.2%.  As widely expected he's given it a very good shake and at the end of election night leads by 387 votes (51.19% 2PP).  That might sound enough but the difficulty for Bignell is that his seat includes redistributed parts of the seat of Finniss, where there was a 1.4% shift in post-counting in the Liberals' favour in 2014.  I suspect that one particular ex-Finniss component that is now in Mawson, Kangaroo Island, might have had a lot to do with that.  This one is worth keeping an eye on to see what the postcount shift is in 2018 with the possibility of it being enough to turn around Bignell's lead.

It is easy to project this as an easy win to the Liberals' Andy Gilfillan if one assumes that a larger post-count means a larger swing to the Liberals, especially given the areas now included in the electorate, but generally in my experience this assumption is unsound.  Postals are the main cause of Liberals gaining in post-counting, but it is actually prepolls that continue to increase.

Monday night: The gap closed to 189 votes (50.5%) for Bignell on postals and prepolls that have been counted - that should be the worst of it but doesn't necessarily mean that he is safe yet.

Tuesday: Bignell lead cut to 90 votes today.

Wednesday: Counting of absents improved Bignell's lead to 198 (50.5% again) which may well be enough as there shouldn't be that much left to throw.  However I don't know how many votes are left.  Informed poster Wakefield on Poll Bludger suggests there could be 1750 or so to throw, which would require about a 56-44 split to Gilfillan.

Thursday: We're up to 89% counted now and with Bignell's lead now at 184 he looks safe.  Earlier today it was reported there are about 600 votes to go.

Saturday: 92% counted (I expect that's the end or very nearly so) and Bignell's lead shrank to 114 but enough to hold on.

Newland

This is another redistributed seat where religious conservative Labor MP Tom Kenyon has had his seat turned into a notional Liberal marginal.  At the end of election night Kenyon trails Richard Harvey (Lib) by 298 votes (50.93% to Liberal) and the ABC coverage reported he had conceded.  However, conceding doesn't affect the count.  In 2014 the Liberals made a 0.3% gain on the 2PP during the postcount; one expects something similar this time as well and that the Liberals will win the seat easily, but just in terms of the reasonably close margin I'll keep an eye on it in case Labor makes any progress.

Update Sunday evening: addition of "ticket votes" (these are votes that would otherwise have been informal but are saved by the savings provisions) has boosted Harvey's lead by 90 votes, eliminating any remaining chances here.

Heysen

There was briefly a glitch in the ECSA computer display that had SA-BEST flipping from 52-48 behind to 52-48 ahead on Saturday but it has now been reverted.

Legislative Council

Figures in the Legislative Council race currently stand at: Liberal 3.779 quotas Labor 3.562 SA-BEST 2.268 Green 0.720 Conservatives 0.423 Lib Dems 0.291 Animal Justice 0.256 Dignity 0.239 (etc).  Eight of the eleven seats will be decided by full quotas.  Then the remainder will be decided by partial quotas unless somebody can catch somebody else on preferences.  But there's no reason to believe the Conservatives could catch up to Labor on preferences on the current numbers.

The current lead for Labor's fourth seat over the Conservatives is 1.16%.  While this is likely to come down on post-counting it is highly unlikely to be reduced to zero.  On that basis it is very likely the result will be 4-4-2-1, with the Conservatives losing the seat they won under the name Family First.

Saturday 24th: With 88% counted Labor's lead has dropped back to .06 of a quota (0.5%).  They should still win but one would want to have a look at the scale of below-the-line voting to see if there's any potential for leakage issues (the ECSA site doesn't seem to have this information.)

General Update (25 March)

The postcount effect in the Liberals' favour has been stronger this election than in the past and is currently running at about 0.9 points 2PP.  This is one reason behind Labor falling back further in the Legislative Council count than at first expected.  It does suggest that, firstly, the swing back to Labor at the election overall was in fact below a point (the 2PP might now come out around 52.2 for Liberal, though treat this as just an estimate for now), and secondly, that there was not a late swing to the Liberals (if anything the opposite) so the polls do not have that excuse.

13 comments:

  1. If SA-BEST Preferences go to Dignity (as was promoted by SA-BEST) that might get Dignity into a possible last upper house seat?

    ReplyDelete
    Replies
    1. It's touch and go whether SA-BEST's third candidate or Dignity will be excluded first (usually, larger parties do better at attracting preferences than little ones). However even if this is the case, experience from the Senate election showed that the proportion of voters following upper house how-to-vote cards under this sort of system is low. The completely optional nature of above-the-line preferencing is likely to make it even lower than the follow rates seen in the Senate race. As in NSW upper house contests, I doubt that anyone will be overtaking anyone on preferences except where the gaps are very very close.

      Delete
  2. One interesting point I found from this election were how accurate the moves in seat betting proved to be - basically all the major changes to the odds were in line with the final results (Lee and Taylor the only real exceptions), and the moves came before any polls in these seats were made public. Basically, if you wanted to know what would happen in this election, the best place to look was the Sportsbet markets - more accurate than any media predictions or other online commentary, and this information was there earlier too! The trend continued on election night with Liberals keeping on shortening to 1.04 or so well before Antony Green and others were calling the election.

    This all makes me suspect that there was a LOT of internal polling (backed up by the number of times I kept getting polled in Hartley) and this polling was quite accurate. Kevin, do you ever get to see internal polls post-election or is this something that parties don't like sharing even after the event? Or are they simply never asked? I imagine that sharing these results with someone such as yourself and/or the public would be beneficial for them too, as you can give them feedback on the accuracy of different polling companies, and whether they are asking the right questions etc.

    ReplyDelete
    Replies
    1. It's rare that I get to see internal polling and I don't ask for it (I doubt they'd give it to me anyway.) Generally parties don't share their external polling even after the election; if they want to review its accuracy they do that via within-party reviews.

      Delete
  3. Is something out of the ordinary happening in King? ABC giving to Libs by 1.7% as of last night (68.9% counted), but current (73.1% counted) ECSA TCP has Labor 64 votes in front.

    ReplyDelete
  4. Currently showing 51.3% to Lib for me with the same count. I am guessing there was a temporary error as there was in Heysen on Sunday.

    ReplyDelete
  5. In Heysen and Finniss, the Greens preferenced Liberals over SA Best; the Liberals won these seats by margins smaller than the Greens vote. Did the how to vote cards end up giving Liberals the seats?

    ReplyDelete
    Replies
    1. The proportion of Greens voters who actually follow Greens how to vote cards is very small, typically around 10%. So on current numbers it's unlikely the Greens card caused the result, but had it been a bit closer then it would have.

      Delete
  6. I'm sure this is not an original thought.

    Polls are seemingly universally judged on how closely they predict the election result. Why?

    Polls are disconnected from prediction by time, place, social context and actions.

    Time - a poll that accurately reflects voter intention on a particular day/s will be deemed inaccurate if it doesn't measure the change in voter intention between the time of polling and the time electors cast their ballots.

    Place - the polling place is a distinctly different environment than that in which polling is done. A poll question seems more like like how do you feel about..., or what comes first into your head, as opposed to the process of casting a ballot in which many voters may be more inclined to reflect and question themselves as to what they should do. For example perhaps SA Best and the Australian Conservatives performed worse than expected from polling because on election day many people may have been thinking about stable majority government rather than who they liked the most.

    Social Context - polls are a one-to-one activity. Voting for many people is a social activity. They gather the family around and head to the polling booth, likely talking about their intentions. At the booth they may interact with party workers. They may know some of them. Social interations close to and even during casting a ballot may change one's intention. The casting of a ballot may reflect thinking about others, while polling reflects thinking about oneself.

    Actions - Polls in Australia assume that everyone will vote, but they don't. There is no way to accurately measure which people were polled and then didn't actually vote. What of purposely or unintentional informal voters? More troublesome variables.

    My conclusion is that a poll that is judged to be highly accurate then may not have been. It can only be deemed to have resulted in figures similar to those of election day. Perhaps it actually performed poorly in measuring intention at a particular time and place prior to the election.

    ReplyDelete
  7. Some pollsters have included deliberately informal votes in final polling - eg Newspoll for federal polls has found that in the last few weeks at least you can get some voters who will say that they will vote informally. Beyond that it is indeed difficult to predict who the non-voters will be, but there isn't any evidence that they tend to skew in any given direction, and those who deliberately don't vote are also probably more likely to refuse to participate in a poll.

    The time factor is the reason why final polls are usually taken as close to the day as possible. While there can be change in opinions in the final 24 hours, especially where there is a sort of bandwagon effect going on, it is less likely than when polls are taken further out. I think there are some elections (and this could arguably include both Tasmania and SA this year) where this sort of thing happens anyway and any poll that gets it dead right was wrong at the time and got lucky, but that in most elections there is not much change in the final days. We don't have very detailed exit polling in Australia, making it difficult to test.

    Pollsters who produce inaccurate seat polls often offer the excuse that their poll was taken a week or two weeks out and that in that time campaigns would have changed their strategy in the seat based on what the polls were showing. But I don't think that explains why we have had so many recent cases (both federally and in the SA state election) of seat polls showing close contests (50-50 or 51-49) that then turn into 55-45 or even 60-40 blowouts.

    Social context - yes, but the booth voter still votes alone in the privacy of the ballot box. The poll subject also may well talk about politics with others in the leadup to the election rather than just on the day. And some polling is done online, in which case respondents may well be talking to close family about the issues as they fill out the poll.

    Normally, something like 5% of voters say they decide who to vote for on election day. But of these there would be some whose choice was actually more or less inevitable if you knew enough about that voter, and who only became aware of that choice on the day. There would be others who genuinely could have decided either way and whose choice was swayed by random conversations, what they saw on their way to the booth, perhaps even the weather or what they had to eat and so on. But I suspect these are only a few percent and that it would be rare for them to break overwhelmingly the same way.

    It is difficult to say whether minor parties so frequently underperform because voters tend to turn away from them at the final moment. The main alternative is that polling companies over-capture voters who are more politically engaged and that those voters are more likely to support minor parties.

    ReplyDelete
  8. Re: Legislative Council. Its not just below the line voting is it? What about above the line preferences. Since neither the Greens nor Libs have their final quota, their preferences wont be distributed and it could come down to the micros..Child Protection (who are right next to the Conservatives on the ballot paper and might appeal to some conservative minded voters), Animal Justice, Liberal Democrats, Dignity, Stop Population Growth + the SAB surplus. Since the Lib-Dems (for example) have .29 quotas and the conservatives only need to make up .04, surely there is some potential for them to overtake the ALP even if only a small minority preference?

    ReplyDelete
    Replies
    1. Sorry, missed this comment as I didn't get a notification email. Experience from the Senate election was that big parties tended to do better on preferences than obscure parties. Also the rate of preferences being distributed at all is quite low. Can't say for sure the Conservatives won't get up but it would be pretty surprising.

      Delete
  9. The full 2PP count Liberal Labor is now up on ECSA website for March election. End result 51.9 Lib 48.1 Labor so about 1.1% swing to Labor. Also results where SAB or other independent candidate was in last 2 https://ecsa.sa.gov.au/elections/notional-two-party-preferred-results

    ReplyDelete