There were some obviously false ones—predictions of a terrorist attack in the US, eg. (There were attempts of varying levels of entrapment, but I didn’t find any actually successful ones.)
For me, the Times Square attempt falls straight into the ‘self-radicalized incompetents’ category akin to the British airport attack or the JFK plot; it makes perfect sense to me that ‘real’ attacks will be rare and with minimal casaulties (I have argued that Terrorism is not about Terror and Terrorism is not Effective in the past, and now that I think about, supposedly there’s a power-law governing terrorist attacks).
As far as credible attacks go, I guessed from going through Wikipedia’s categories of terrorist attacks by year that a better prior would be more like 1⁄3, not 1⁄2.
But when I say ‘obviously false’ I don’t mean that it’s an insane prediction (if you read through the thread you’ll see I do call out some predictors for making predictions which I consider ‘on crack’ - not naming any names here, but none of his seem any more likely to be true after the passage of a year), I just mean that it’s objectively and clearly judgeable as wrong now that the year is out.
This is a rare and valuable property, as one will learn after reading through a few compilations of predictions for 2011.
I just mean that it’s objectively and clearly judgeable a wrong now that the year is out.
This is a rare and valuable property, as one will learn after reading through a few compilations of predictions for 2011.
Thanks for the clarification; my first reading was that you were holding people to the wrong standard. If you made 10 predictions at 70% each, and 6 of them come true, then you should be lauded rather than criticized. If all 10 of them come true (and appear to be causally independent of each other), then you should be criticized for underconfidence.
I’m guessing that many of your 80% and 90% predictions would be positively correlated with one or two major trends (e.g. there was no global economic meltdown in 2010), so it’s not quite that simple. Looks like pretty good calibration to me overall.
The 70-90% bracket for me is ~70 predictions judged; I don’t think most or even many of them are economic-related (except in a weak sense).
If you wanted to check my intuition, there doesn’t seem to be any easy way to filter my user page for just the judged ones within a probability range short of downloading and processing the HTML, but you could look through a few dozen or score of the recently judged predictions (http://predictionbook.com/predictions/judged) since I have looked at every prediction on the site and registered my own probabilities for essentially every prediction which isn’t either sports or highly personal (and even then I’ve frequently given it a shot anyway).
Wow, really? I’ll happily make the same bet again for 2011 if you’ll offer it again—though I’d like to slightly tighten the rules to say it has to be a conspiracy, because I don’t think something like Fort Hood should count.
Oh, my own odds are lower than mattnewport’s, just not low enough to call his estimate crazy.
Also, my bid-ask spread is pretty high if we’re talking actual bets. I’d take the “yes” side at something like 10:1 odds, and the “no” side at 2:1 odds the other way. And I’d only feel morally comfortable betting the “yes” side in the context of a formal prediction market, where the positive externalities to having accurate odds would assuage my guilt about collecting in that case.
There were some obviously false ones—predictions of a terrorist attack in the US, eg. (There were attempts of varying levels of entrapment, but I didn’t find any actually successful ones.)
In my opinion, mattnewport’s 50% odds weren’t that badly calibrated (especially given the Times Square attempt).
For me, the Times Square attempt falls straight into the ‘self-radicalized incompetents’ category akin to the British airport attack or the JFK plot; it makes perfect sense to me that ‘real’ attacks will be rare and with minimal casaulties (I have argued that Terrorism is not about Terror and Terrorism is not Effective in the past, and now that I think about, supposedly there’s a power-law governing terrorist attacks).
As far as credible attacks go, I guessed from going through Wikipedia’s categories of terrorist attacks by year that a better prior would be more like 1⁄3, not 1⁄2.
But when I say ‘obviously false’ I don’t mean that it’s an insane prediction (if you read through the thread you’ll see I do call out some predictors for making predictions which I consider ‘on crack’ - not naming any names here, but none of his seem any more likely to be true after the passage of a year), I just mean that it’s objectively and clearly judgeable as wrong now that the year is out.
This is a rare and valuable property, as one will learn after reading through a few compilations of predictions for 2011.
Thanks for the clarification; my first reading was that you were holding people to the wrong standard. If you made 10 predictions at 70% each, and 6 of them come true, then you should be lauded rather than criticized. If all 10 of them come true (and appear to be causally independent of each other), then you should be criticized for underconfidence.
/looks at http://predictionbook.com/users/gwern
: (
I’m guessing that many of your 80% and 90% predictions would be positively correlated with one or two major trends (e.g. there was no global economic meltdown in 2010), so it’s not quite that simple. Looks like pretty good calibration to me overall.
The 70-90% bracket for me is ~70 predictions judged; I don’t think most or even many of them are economic-related (except in a weak sense).
If you wanted to check my intuition, there doesn’t seem to be any easy way to filter my user page for just the judged ones within a probability range short of downloading and processing the HTML, but you could look through a few dozen or score of the recently judged predictions (http://predictionbook.com/predictions/judged) since I have looked at every prediction on the site and registered my own probabilities for essentially every prediction which isn’t either sports or highly personal (and even then I’ve frequently given it a shot anyway).
Wow, really? I’ll happily make the same bet again for 2011 if you’ll offer it again—though I’d like to slightly tighten the rules to say it has to be a conspiracy, because I don’t think something like Fort Hood should count.
Oh, my own odds are lower than mattnewport’s, just not low enough to call his estimate crazy.
Also, my bid-ask spread is pretty high if we’re talking actual bets. I’d take the “yes” side at something like 10:1 odds, and the “no” side at 2:1 odds the other way. And I’d only feel morally comfortable betting the “yes” side in the context of a formal prediction market, where the positive externalities to having accurate odds would assuage my guilt about collecting in that case.