1. We have added a Gift Upgrades feature that allows you to gift an account upgrade to another member, just in time for the holiday season. You can see the gift option when going to the Account Upgrades screen, or on any user profile screen.
    Dismiss Notice

I'm so sorry: Australia is having an election and it's going to be very dumb

Discussion in 'Off-Topic' started by Arwon, Apr 10, 2019.

  1. onejayhawk

    onejayhawk Afflicted with reason

    Joined:
    Jul 6, 2002
    Messages:
    13,357
    Location:
    next to George Bush's parents
    You would have to define reasonable climate change policy. IMO wait-and-see is a reasonable policy at this time, which is what they seem to have settled upon.

    Is there any truth to the bookmaker story? I have heard the polling was even further off than in US and GB. The story is that the bookies paid early, before the results were in, and wound up paying some losers then having to pay the winners too.

    J
     
  2. Arwon

    Arwon

    Joined:
    Oct 5, 2006
    Messages:
    17,473
    Location:
    Canberra
    Yeah Sportsbet paid out early as a promotional gimmick, but the political markets are relatively small and shallow (arbitrage opportunities on seat winners between bookmakers are pretty common) so I doubt they lost much money.

    The polling was differently wrong to the US and UK. Here, the national vote was predicted as about 51.5-48.5 and came out about 48.5-51.5 instead, about a 3 percentage point error.

    In both Trump and Brexit cases there were late movements that clearly gave a decent chance of the correct outcome happening by the day of the vote. Here, there's evidence from early voting that there was no late movement at all. In the Trump case the national polls were basically accurate on national vote share, but the distribution in some crucial states in a malapportioned system led to the electoral college outcome being the less favoured one. In Brexit, the aggregate vote error was similar to here (2 or 3 percentage points), but there were still polls predicting a no win due to the margin of error on individual polls.

    In Australia there were two evident problems. One is that there was widespread herding in the last few weeks. Polls came in more similar to each other than should have been possible regardless of what the real result was. With a 3% margin of error on each poll, they should have been more scattered rather than all being within 1 percentage point of each other. At least a couple of polls should have gotten close to the actual result just by random chance.

    Second and more crucially, seemingly a long term structural issue has arisen with weighting/stratification missing a crucial variable. The Coalition basically didn't lead in any poll for three years - that points to something systemic that was missed in the poll methodology that had worked in the past. The most plausible explanation is that something like education level, or level of political engagement and awareness, became a crucial variable and wasn't weighted for in converting data to estimates. If you only weight by age, sex and location, for instance, you will miss if a big cleavage by education level suddenly appears, or if disengaged voters all go one way instead of going both ways like normal (remember with compulsory voting, predicting what totally disengaged voters will do is a challenge at least comparable to modelling turnout in voluntary voting systems).

    There were signs in some state elections recently. Polls in the Victorian election predicted a very narrow result, but there was a big error (over 3%) in the pro-Coalition direction in those polls, and so Labor ended up having a pretty big win there which nobody predicted. Incidentally, the fact that this previous big error was in favour of the Coalition, in a campaign which was loudly racist and pretty shameful, is also evidence that there was no "shy tory effect" here.

    Trouble is, Australian polling is a small field and none of them are very transparent about what they do, compared to in the US or UK, so unless they start being more open about methods and inputs in order to regain trust, we won't really know why their estimates were off. One problem is they've broadly gotten every federal election right since about 1980, so we've been pretty accustomed to polling in Australia being quite easy and exact compared to other countries.
     
    Last edited: May 21, 2019

Share This Page