16 thoughts on “Go to PredictWise for forecast probabilities of events in the news

  1. Thanks for pointing out this website.

    Of course, you are right about the spurious precision. I think the excessive precision just comes from the arithmetic. For the NHL forecast they say “Step 3: normalized to equal 100% for any mutually exclusive set of outcomes.”

    Note that 3-digit fractions are necessary to report with some clarity their estimate of the probability that the Edmonton Oilers will win the Stanley Cup. The estimated probability is 0.1% for that event (LAST UPDATED: 04-19-2015 10:02AM).

    Given (1) that the 16 teams making the playoffs have already been determined, (2) that Edmonton did not make the playoffs, and (3) the winner of the playoffs gets the Stanley Cup, it seems that this estimated probability overstates the real probability by a large factor.

    P.S. Wouldn’t the Stan Cup be a good name for a prize to be given to an outstanding Stan developer each year?

  2. Sig figs. Back when I was TA my students were surprised I took them so seriously, i.e., that I’d dock them points on homework assignments for egregious violations. Never mind student errors – you’re bound to make errors while you’re still learning a topic – I see sig fig errors from engineers with decades of work experience. The other day someone sent me a table of geographic coordinates with locations reported to the nearest 1/10000th of a meter, i.e., 0.1 mm. Seriously? He knew perfectly well that the uncertainties associated with the reported values was multiple meters. One extra significant figure is forgivable but four extra? No.

    • Sometimes though, the extra significant figures don’t signify stupidity on part of the Engineer just haste.

      e.g. A SCADA / DCS system might dump unnecessary digits & it could be a hassle to clean the data everytime you send it for future consumption.

      My point is, so long as nobody is taking a *decision* that is contingent on the irrelevant decimal places things aren’t a disaster, right?

      • > the extra significant figures don’t signify stupidity on part of the Engineer just haste.

        Yeah, signifies carelessness more than anything else I think.

        > My point is, so long as nobody is taking a *decision* that is contingent on the irrelevant decimal places things aren’t a disaster, right?

        True enough. It’s the carelessness that sticks in my craw. If someone isn’t taking the extra 4.3 seconds necessary to identify an appropriate number of significant figures, what else are they not thinking through?

        Actually, another thing is that sometimes the audience for the numbers isn’t particularly math literate. If you present them numbers with n extra significant figures they may develop the belief that all are relevant and at some future date you may need to correct their expectation. (Correcting expectations is a p.i.a.)

  3. Looking at the latest (April 20th) version of their homepage, I’m struggling to think of any sort of prior distribution that could produce a 9.2% predictive probability that Australia will win the 2015 Eurovision song contest!

    • Iyh:

      I read the linked article. Freese’s argument is provocative and amusing, the thing that could make an excellent blog post. It’s a contrarian point that’s worth making, but I don’t really buy it. Freese says that researchers should report extra significant figures. Much better would be for researchers to make their data and code available. Or appropriate summary data if there are confidentiality restrictions. But not to clutter up articles so that they become hard to read.

  4. Re: Hyper-precision

    The hyperprecise figures obviously have to be taken with a grain of salt, but they do serve an important purpose, particularly near 0% and 100%.

    I used to bet on Intrade/Tradesports. On that platform, they used percentages with one decimal point. I recently began trading on Predictit, which only allows trades in whole percentages. I am not happy about this: For instance, I am currently heavily short on Bernie Sanders for the democratic nomination. This contract is trading at 11%, which is extremely mispriced: The democrats are not going to throw the general election with 11% probability. The market will almost certainly adjust soon. If this was Intrade, I would have been able to get out of the position by selling my positions when the price reaches 0.1% (giving me 99.9 cents per share). Now, I will either have to wait until the contracts expire at the nominating convention in August 2016, or sell my shares at 1% (giving me 99 cents per share, which is less than they are worth).

    In situations like this, I am convinced that allowing an extra decimal increases the liquidity in the market and therefore contributes to more accurate predictions

    • Interesting point!

      By the way, I too think it’s unlikely that Sanders will win the nomination. But if he is nominated, I think he has a chance. His ideological extremism will cost him some votes, but if it’s a good economic year, just about any Democrat could win. Recall that Ronald Reagan was pretty extreme but he still won in 1980. The best estimates I’ve seen are that a more moderate Republican would’ve done better, but Reagan did well enough. Maybe Sanders’s extremism would cost him 3 percentage points of the vote.

      Everyone remembers McGovern and Goldwater as extreme candidates who lost, but they were running against incumbents in good economic years.

    • Anders,

      You probably noticed that in those markets there may be a third alternative available to you for “buying back” your Bernie Sanders, and often at prices significantly better than 1%. But again, you’d have to wait until August 2016 to realize your profit.

Leave a Reply

Your email address will not be published.