Polls, pundits and punters yet again

It’s too late to influence anyone’s vote, and I doubt that many of my readers are in much doubt as to which way they will go, so I’ll return to one of my favorite topics, the relative predictive power of polls and betting markets. Most of the time, comparing the two is a tricky exercise, since the odds in betting markets and those implied by polls tend to converge as the election nears. Not this time, however. The polls have remained at or near 50-50 throughout the campaign, with the additional complication that 5-10 seats may be won by independents or minor parties. Yet, as of a couple of days ago, the LNP was paying $1.08 for a dollar bet, implying a winning probability of around 90 per cent.

In classical statistical terms, that means that, if the LNP loses (does not form a government) the null hypothesis that the betting odds are correct can be rejected with 90 per cent confidence, which is commonly considered good enough for social science work. Unfortunately, things don’t work the other way around. If we disregard markets, the chance of an LNP win is probably a bit above 50 per cent, so a win for the LNP wouldn’t prove anything one way or the other. For that, we’d need a case where the polls strongly predicted a winner, while the markets were even money or going the other way (the key concept here is the “power” of the test).

In an important sense, anything less than a clear LNP majority would constitute a rejection of the betting market hypothesis. If the majority is small, then the outcome will inevitably have been decided by a few thousand random votes reflecting unpredictable chance.

Anyway, we’ll see soon enough.

36 thoughts on “Polls, pundits and punters yet again

  1. I suspect the polls are now telephone based, which means they sample from a directory. This skews the sample as generally older peopkle have landlines or mobile numbers in directories.

    Betting markets also do not reflect the community.

    Betting markets are influenced by polls while polls are not influenced by betting markets.

    The only poll that matters is the ballot box.

    Although if you are keen and believe in a Turnbull triumph, you can always stick money and get a quick 8% gain.

  2. Alright John.

    But Its never too late to vote – at least until postals close. I just changed mine 🙂


  3. Yes John Chapman, that’s a possibility. I can’t see any compelling public evidence for who is going to win – mainly a lot of speculation. And it doesn’t seem possible that there is some hidden data which we don’t have access to. A breakdown of who’s doing the betting might enlighten us, but there’s no chance of that.

  4. Brexit is being cited as supporting polls over bookies, and with good reason. Like the Australian election, the polls on Brexit were very close but the bookies strongly favoured one side (remain). Of course the bookies turned out to be wrong. A Labor win would be the equivalent tonight.

    One explanation I saw from a bookie (more credible because it was made before the referendum, not afterwards) was that odds follow the money, and the big money was for remain *but* this was biased by relatively small number of very large bets (maybe from London-based city types?) whereas the greater number of bets (i.e. lots of small bets) favoured ‘leave’, better reflecting punters’ views. That was why this particular bookie said he thought there might be an upset (although most others interviewed were supremely confident). It was also noted that whereas polls try to sample everyone, bettors tend to be a subset of men.

    I don’t know whether the same factors apply here, but in any case tonight we will have one more data point for this debate.

  5. @Dave
    Somebody over on Poll Bludger was saying something similar about the Australian betting market, i.e. that there was a larger number of bets on Labor but that the size of the bets was smaller. Not sure how reliable the source is though.

  6. I really would like to see a breakdown of number and size of bets for each side. Do you think this is possible after the event?

  7. This blog post would be illegal in New Zealand. In New Zealand, the only thing you are allowed to say on election day is you should vote.

    Celebrities tweeting photos of your ballot paper from the polling booth has led to police investigations. The leader of the opposition was threatened with the police investigation for saying something a little bit more than get out and vote.

    There are no how to vote cards or any canvassing outside of polling booths, but that is more to do with different electoral systems.

    All social media are taken down by the parties to avoid prosecution.

    The High Court managed to persuade itself recently that if you put something on your website and it still up on that website in the 2 days before an election when there are more serious laws about corrupt practices and lying, you can be prosecuted despite the fact you said that weeks ago.

    They decided to is a difference between the words publish and first publish.

    Next New Zealand election day, I plan to turn the commenting function on my blog often not blog at all on politics because it is not legal

  8. Looks from his language that John is one of those damned frequentist trying to predict the future when the latter school will tell you simply cant predict the future from statistics.

    Being on the other hand a Bayesian I am less anal but curious about how John got his 90% probability from the $1.08 return. It doesnt look right….they look like a lousy return for a 90% likelihood even after the TAB takes its cut. Regretably John doesnt explain the calculation but hopefully might some day. Anyway its all academic as the votes have now been cast except in WA which has the advantage (?!) of seeing what the smarter states do first before it responds strategically.

    In case anyone is interested in using the statistics properly i.e. by being a Bayesian, here is the latest offering from the venerable Nate Silver and co. http://projects.fivethirtyeight.com/2016-election-forecast/#now

    It looks pretty impressive and needs a bit of time to analyse. But he is certainly being transparent unlike most pundits who wont admit the correlation between the predictions and satisfactory bowel motions. Still I am waiting to here the word Bayes used in Oz political predictions. Maybe its time to start a local company with a name that alludes to the Delphic Oracle or those entrails they used to predict the events of the Ides of March.

  9. @Jim Rose

    How does NZ inforce the laws you have described when a blog is registered in say California but fed by a New Zealander via emails to either a person or a computer with the instruction to feed the emails into the blog?

  10. > Yes, John, but what I want to know is WHY the money’s going on the Coalition?

    On average gamblers lose. To be a gambler you need to be confident you know better than the bookmaker… and, y’know, the bookmaker’s a professional. It’s a fool’s wager, literally, driven by overconfidence and a disdain for expertise.

    A-priori, and I’d suggest pragmatic experience agrees with this, there’s strong reason to believe both that the gambling sample overrepresents people with conservative leanings and that they’re unaware of this. I’ve made this point before, last time election wagering came up, but to no effect.Which is as you’d expect, really.

    [also, of course: markets are weighted by disposable income of the market participants, which means that a market-set wagering price overrepresents the opinions of wealthy people. So that’s another rightward bias that most gamblers don’t correct for.]

  11. @Newtownian
    That’s all very well, but what was Mr Silver’s prediction for the Australian election?

    Anyway, looks to me (with most of the results in) like the coalition may fall over the line, leaving Malcolm with reduced authority and the sound of knives sharpening in the background. Stability anyone?
    And God knows what the senate will look like, but it will be a nightmare to work with.

  12. Whatever the final outcome (it’s 72/66/5 with 7 undecideds at the moment), the money was clearly wrong. ALP should have been much shorter than eights.

  13. @Newtownian

    I’m a Bayesian, or more accurately a post-Bayesian (don’t accept all of the Savage axioms). I’m just trying to translate into classical/frequentist terms because they are better understood.

  14. Yep, polls were spot on. The TPP looks as though it is going to almost exactly 50%.

    But the coalition is almost certainly going to form government (its still possible they’ll need an independent but there’s no way ALL the independents will support Labor). The punters (and I) were right – if you start with a big majority you will have an inbuilt advantage in the form of a locally well-known sitting MP in the marginal seats, so the opposition needs more than 50% of the TPP to knock you off. Both the polls and the betting markets proved right, because they were measuring different things (votes on the one hand, seats on the other).

    Plebiscites such as Brexit are a different matter.

  15. You are calling it too early.

    ABC News shows:

    L/NP – 65 seats
    ALP – 67 seats
    OTH – 5 seats (of who at least 3 lean to ALP)
    In doubt – 13

    As at 10:42 EST with 77.7% counted. Admittedly late votes (postals) could favour L/NP.

    Too early to call but ALP might even have their nose in front.

  16. @derrida derider

    The punters bet heavily against both a “hung” Parliament and the emergence of a Labor government (overlapping possibilities). On the most generous interpretation, the implied probability of anything other than an LNP majority was less than 20 per cent. So, it’s already clear that the markets, and the pundits, got things badly wrong. By contrast, the polls were consistent and spot-on for anyone who cared to read them without invoking dubious phenomena like “seat buffers”, one-term incumbency benefits and so on.

  17. @Ikonoclast
    Yes, it was amazing how much Anthony Green’s projections changed later in the night. From 72/66 to 65/67 now. But it’s clear that neither side can form a majority in their own right, with 13 seats in doubt (ALP currently leading in 6, Coalition in 7).

  18. @Ernestine Gross
    Yes those laws are very much out-of-date to the ravages of social media and smart phones. But anyone in New Zealand takes those laws very seriously they do not want to have a knock on the door from the police

  19. @Lyn Gain

    I was amazed when I first turned on the ABC coverage how far in front they had the LNP – and seemed to always have the LNP well ahead of Labor TPP, whereas 7 and 9’s projections seemed to have the ALP just in front most of the time.

    Right wing bias at the ABC ! 🙂

  20. The only group in our society stupider than punters (who amuse themselves by handing over their money to bookmakers) are the Canberra press gallery. They joined forces and drove down the odds on a LNP victory. The bookies are laughing their heads off.

  21. Lyn Gain
    There are differences between the model Antony Green was running last night, and the model he is running today. Today’s numbers are effectively a feed from the AEC, and as the postal and absent votes come in and are counted, about 8 to 10 of those 13 doubtful seats will end up staying with the Coalition, as the Coalition always do better in absent and postal votes.
    The model Antony Green was running last night was including an estimate of how much the absent and postal votes will affect the final result, so that is the reason it was predicting more for the Coalition. So no right wing bias in Antony Green’s estimates.
    Antony clearly thought his model running last night was declaring seats as wins or losses too early, so that was one of the reasons he pulled that model offline.
    But it may well turn out that his model from last night turns out to be an accurate prediction of the final result.
    We will have to wait and see.

  22. @John Goss
    Yes, I knew it had to be something to do with the modelling – but it still doesn’t seem to affect my analysis that neither side can make it to 76, despite postal votes (given very large percentage of them, more than normal which must affect benefit to Coalition.

  23. Lyn Gain. If 11 of the 13 doubtfuls go to LNP, then they will get 76 seats. And although that is possible it is unlikely, as for 3 of the doubtful seats, Capricornia, Cowan and Herbert, the ALP is sitting on 50.7%, and normally 0.7% is enough of a buffer, so that postals and absent votes will not erode the lead. But we can’t be sure.
    I think 75 seats for the Coalition most likely.

  24. @Jim Rose
    Essentially the same thing in Canada in terms of law. I don’t think the enforcement is as strict. On the other hand, handing out election material at all, let alone near a poll would likely mean a nice visit to the friendly police station and some nasty legal stuff after the election.

    I remember being in the USA years ago on an election day and as I walked by a poll several party workers jumped out to hand me material. Weird.

  25. @John Quiggin

    Suggest waiting until the final results before asserting the markets got it wrong. And what is ‘wrong’ anyway? Take a sample of 20 elections and compare the implied probabilities with the results. Something can then be said.

    BTW I haven’t done one formally but have followed the markets very closely and I would “bet’ the proverbial house they are within a 95% confidence range. Punters are mean minded and greedy and quite often stupid but they are not biased politically in terms of their bets.

  26. @Historyintime

    A lot of small-time or one-off bettors are “heart bettors” or “whim bettors”. They bet on their team. They bet on the horse’s name, the jockey’s colours or a numerological scheme. Events like elections , football test matches (national games) and the Melbourne Cup can draw out droves of these “heart bettors” or “whim bettors” to skew the sample. These are not form analysts, these are idle one-off bettors.

    For a “whim bettor”, who risks little, bets seldom and has done no research on form, betting on the horse’s name, the jockey’s colours or a numerological scheme is rational. It’s a simple pseudo-randomisation strategy to make an easy, fun pick and “have a flutter”.

  27. Barry Cassidy says that it’s increasingly likely that we could be heading back to the polls. Being a constituent in the Melbourne seat of Chisholm, I actually feel a touch empowered for the 1st time in an election process 🙂

  28. @Troy Prideaux

    Malcolm may well want this, because the trade union campaign will not be as effective in the scenario of a short media-driven campaign.

    The current result was achieved on the basis of months of ACTU mobilisation and community engagement in particular seats. This cannot be replicated in a short campaign.

  29. @Ikonoclast

    There are small irrational betters as in any market. It’s the aggregate that counts (the price/odds) and needs to be objective.

    The ALP will win 68-70 seats ie small chance of hung parliament if all the cards fall the right way.

    And what are my qualifications for such pontification. Professional economist, lifelong psephologist (short list for Antony Greens job way back in 1988 but withdrew for a Treasury job, bad call!) and lifelong punter. The comments on the election sites are usually ill informed and those about election betting, well ……..I don’t know why I bother really. They seem to apply some theory which is way beyond my education to something which they really don’t know how it works in practice.

  30. Having said all that Sportsbet has the LNP at $1.22 to form government. Correct price $1.02. Run run rabbit.

  31. Turning to the larger ‘betting markets’, the ASX and the FEX, nothing much is happening there as at noon, 4/7/16.

  32. I’m surprised at how woolly some of the above analysis is. My ‘gut feel’ is that the markets maybe had it somewhat shorter priced for Coalition than ‘fair’, but…

    The fact that a 10% chance happens does not mean its odds were underestimated or mispriced. This should be obvious. Roll a (true) 10-sided then ask what were the odds the revealed number was going to come up.

    Another ‘fact’: Pretty much everyone seems to think that the result on the night was close to as bad as it could have been for Coalition, and pretty much as good as Labor could have hoped for. Yet it’s still odds on that Coalition will form Government. Antony Green is predicting minimum 73 and maybe 76 seats for Coalition. Let’s face it, if they get 74 or 75 they will almost certainly form Government (Katter + 0/1).

    Put another way, if we imagine it’s last Friday and running some kind of Monte Carlo simulation of the result, we are seemingly well down in the tail of bad results for the Coalition – and the straight ‘win/lose’ market is still about a 75% chance of paying out for Liberal.

    None of this is to say the outcome will be stable or good.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s