Australian Federal Election 18th of May 2019

[Wisconsin+17]

Happy man today. I called all the Queensland flips and the Tasmania Flips. Really hope the Coalition can take Cowan too.

All I basically did was assume the actual polling was the exact opposite of abc's coverage on election day. They said that WA was trending against the coalition (it didn't), and that Queensland was dire for the coalition (it had a surge for the coalition), and that Tasmania was a lost cause.

Seriously, abc when someone who's on the opposite side of the world can call your elections better than you can, it's time to retire!

Voting total results indicate that while folks are not terribly happy with the coalition, they like labor even less than before. Labor votes are down 1%!

[Wisconsin+17]

Still TCTC:

Snowdon in NT. Half in, Labor leads by three. currently a 6 point swing to the Coalition.

Solomon in NT. 60 percent in, Labor leads by three here too.

Blair in Queensland. 60 percent in, Labor leads by less than a percent.


Griffith in Queensland. 70 percent in Labor leads by four.


Lilley in Queensland, 70 percent in, Laboer leads by half a percent.


Moretown in Queensland, 60 percent in, Labor leads by two percent.

[Wisconsin+17]

Flint takes the lead again in Boothby! Yay!

67 in, Liberals up half a percent.

Bass 80 percent in, still TCTC, Liberals up by 1.5 percent.

Cowan, Labor still up by half a percent. Half in.

[Wisconsin+17]

Chisholm in Victoria, Liberals still up but very close.

Indi is still TCTC, currently the independents lead, but by two percent.

Dobell in NSW is TCTC, currently Labor leads by a percent.

Eden is NSW is TCTC, currently Labor leads by a percent.

Hunter is still TCTC, currently Labor leads by 2 over the Nationalists. This should have been a completely safe Labor seat.

Macquerie, Liberals still lead here.

Wentworth, Sharma is still down by half a percent with half in.

[Wisconsin+17]

67 in. Labor lead in Cowan down to .2 percent.

And it flips! Liberals have the lead now!

[Wisconsin+17]

But hey, at least we won't get any scary brown people.

[Wisconsin+17]

While a polling error for sure it seems roughly the same size as Brexit.

Consistent error on one side of the line suggests systematic bias. If the error were sometimes on the side of conservatives, it wouldn't be systematic error.

Australian polls are quite simple. Liberals win on election day, and every single pollster between elections says they lose. Every single one.

[Wisconsin+17]

Good job on not responding to the climate change part.

What about it? Apparently only white people cause climate change.

[Wisconsin+17]

75 now, with Cowan lead back to 0.3 for Labor.

Lilley also with a Labor lead of just 0.3.

Blair at 0.8 for Labor


Chisholm at 0.4 for the Liberals, and Boothby also very close at 0.3

Maquerie at 0.2 for the Liberals. Going to come right down to the line here. Still a bunch of TCTCs, though some are going to be called pretty soon here.

[Wisconsin+17]

Any reason for that?

As you've noted, there's a few elections that were the other way. Canada's election in 2015 was one, strangely. Polls showed a conservative minority but that's not what we saw. Something like 10-15 percent for a Liberal victory. Pretty much every 'tossup' broke their way, which is very unusual, I've never seen that before or since.

I got that one wrong due to estimating the usual liberal bias.

I think polling is done pretty poorly. There are all kinds of ways to distort a poll, and most polls I've seen tend to be done by organizations with an axe to grind. The problem is that polling is seen as a way to shape the narrative rather than reflecting on the actual views of the electorate.

Until polling returns to being a reflector rather than a tool to manipulate people, we will see bad polling done.

Also often when a polling error pollsters try to figure out what went wrong and overcorrect, UK perfect example where Tories overperformed in 2015, Labour in 2017.

It shouldn't be a matter of 'overcorrecting'. Again, polling should be a reflection not a narrative shaper. I've done ok every election since 2015, which was a bit of an aberration. Only one I truly missed on was the House election, which Republicans lost due to the mortgage deduction. That's why they lost their blue state high cost of living seats.

[Wisconsin+17]

The Coalition will do f**k all on climate change and you know it.

Which is different from the posturing of Labor how? The only 'action' on climate change that I see is raising taxes on people trying to make a living.

It's a scam. But, I guess it's good if you get your lifestyle choices subsidized while taxing the crap out of people you don't like.

[Wisconsin+17]

Can you two children please take your bickering to the intl discussion forum or else stick to the election returns?

Finally! Thank you!


Returns seem to be slowing down now. Guess the counters are taking a break. 

[Wisconsin+17]

Martin gaining in Indi on Helen Haines. Now just 1.9 percent down. Still TCTC

[Wisconsin+17]

Adma, there was a late break in the polling. Ekos as late as October 10th, showed a conservative majority government. I've never seen a ten point break in a little more than a week, all of it going one way.

Still not sure what to think about it.

[Wisconsin+17]

Imagining a 'systematic' left bias is pretty naïve

What else would you call it? Just looking at the polling aggregates, every poll since about 2014 has said Labor victory. The only two exceptions are both elections. That's not one poll but about 50.

At that point we can estimate a (at least in the Australian context, of about a 1-2 percent systematic error in favor of Labor lean.

Given a sample without systematic bias, we'd expect 25 +/- 8, so there would be no more than 37 polls. The chance of all 50 being Labor without systematic bias is, what, 3.5 standard deviations there about? A tenth of a percent or so.

[Wisconsin+17]

AB 2015 was a bonkers election. I don't think anybody, including Albertans had a bloody clue what was going on.

By Oct 10 that EKOS poll was a clear outlier. The shift happened two weeks earlier following the conclusion of the debates, which are a pretty clear demarcation point for voting intentions. Also, 10-point swings inside a week have happened in UK 2017, Alberta 2015, BC 2013, Spain 2004, and many others besides.

2013 was funny. I changed my prediction very little. Polls screamed Liberals are toast, etc, etc. My perception of that 'shift' is that the electorate barely moved, but the pollsters certainly did!

[Wisconsin+17]

Regardless, your argument implied a global systematic bias against right wing parties. Miles rightly pointed out that doesn't hold water.

Shrug, I predicted a Liberal majority government here. All I did was study the actual history of the polls. Told all my Australian friends that it was business as usual, and to GOTV because it was damned close.

The only time I've had my lunch eaten was in the 2015 Canadian election. That's a lot of elections over the years.

[Wisconsin+17]

Hey, you want to blame Murdoch for Labor's asswhooping here, go right ahead. But the election is basically the same as it was before. No change.

That would seem to indicate pretty settled preferences. Too early to dig into the substance of the returns to make better predictions. Wouldn't it be more simple to say that than blame 'Rupert Murdoch?'

[Wisconsin+17]

3. A brutal observation but this is a brutal result so: the politics of the environment are too important to be Political and there's an urgent need to find a way out of what has already been a very costly (mostly for the environment!) trap; this is not an Australia-specific comment, it applies everywhere. It is possible to persuade Country People and people in resource-dependent regions and localities to support measures that will improve the environment (or at least reduce damage), but only when approached on their terms. This is going to be particularly difficult in Australia for various reasons, but things are too serious to not at least try.

As a country person, you aren't going to sell rural people on climate change by passing bills that screw them especially hard. Are you willing to let go of climate taxes on carbon and preferential treatment of alternative energy? Until then, you're never going to see rural people vote for Labor.

[Wisconsin+17]

If you think that it is a simple case of pollsters always over/underpolling one side, deliberately or not, then it is because you have no idea of how polling actually works (in part because it is hardly in the pollsters own interests to always get it wrong - their own financial viability relies on actually being vaguely credible).

It's not just one poll. It's *every* poll for the last five years. There's a difference. Systematic bias on the part of pollsters is the best explanation given the amount of error and the consistent error of +1, +2 for Labor.

[Wisconsin+17]

My hypothesis is that the bias against One Nation is the source of the error. The polling error is equivalent to their improvement in the polls.

[Wisconsin+17]

That was not your hypothesis.

That's an explanation as to how and where the systematic error arose.

[Wisconsin+17]

It is not, however, a hypothesis that there is a global underestimation of right wing parties (because ... reasons?), which was your original contention.

I would argue there is, but the error on the part of Australia in particular seems to come from the underestimation of One Nation support. Now, this could be explained in other ways too. It's too early to show for sure. Both explanations given here could also be partially correct. We shall see.

[Wisconsin+17]

the poster who called Alabama Senate for Roy Moore at 50% reporting, despite the clear counting bias and the NYT needle being against him. The justification? A incredibly weak model that assumes there never is counting bias ever and all precincts are reasonablly similar.

I'm not sure how it's an incredibly weak model when I beat out 538 on Trump calls. Missed two. 48-2 for calls on the night. I called PA for Hillary and VA for Trump. It sometimes misses on incredibly close elections, like Roy Moore's.

The benefit is that I don't rely on exits or precincts at all. It works because of math, not because of election data.


In Moore's case, he was up by 5 percent with 50 percent in. The only time that's been a loser is... Moore.

VA was a miss as was PA because the opening precincts were stacked so heavily for Hillary in PA, and in VA for Trump.

[Wisconsin+17]

I'm not sure why they would have thought what works in Victoria would work in the rest of Australia...