Congratulations to forecasters from the past several weeks who won profile badges on SciCast.org during Activity Badge Wednesday. Learn more about the special.
Profile Badge Winners
July 23, 2014
laughinghyena13, jcurtis, kelli.thompson, supportcaring, [email protected], parkerrc, Kaspar Snashall won a Helium badge
Tender won a Neon badge
Apeear, Phoenix won an Argon badge
Recycler, Recurring, hippolyta won a Krypton badge
July 30, 2014
thucthuc, DustLover1104, ronald.redfern, mrmark, Jaydady, jimpencelol, jameswillwin1, Awsomeboy360, sferrrl, Bill Humphrey, iknwnbob, melrrorfd, sciamazon, kmcwilliams, richhuah, bdjzn won a Helium badge
Phoenix won a Neon badge
hippolyta, peaigr won an Argon badge
dvasya won a Krypton badge
August 6, 2014
spambot2113, 123bilbobibo, jkim3, DaBagBoy, Kz556 won a Helium badge
Recycler won an Argon badge
Robert, sflicht won a Radon badge
August 13, 2014
banneradman, ejrowi, ejonahor, maria, journeymenc, brigadier won a Helium badge
pressurization won a Neon badge
Robert won a Xenon badge
Hi. I was reading Tetlock recently and noticed his comments on how mathematical models outperformed humans. And I can see from a quick google that some of your forecasters do use models:
https://scicast.org/#!/questions/37/comments/safe
So I was curious how many of your predictions are coming mostly from models and how much people are just answering off-the-cuff. Thanks.
Based on comments it seems that ~5 of our top forecasters use models at least sometimes, and a couple have even started automating forecasts from their models. It’s wise to include models when you can. But in weather forecasting, chess, and many other areas, the top systems are hybrids where humans adjust the model forecasts for local context or “broken leg” scenarios the model can’t see.
In Expert Political Judgment the humans only explained half the variance of relatively simple trend models. But in Tetlock’s Good Judgment Project, the humans have gotten very good, partly by using models, but partly by finding skilled, well-calibrated forecasters, and giving them enough feedback and support to get even better.
Hopefully some of our forecasters can reply on how they mix models, careful thought, intuition, and outrage to make their forecasts.
I’m one of the forecasters using models. I don’t have exact counts, but I expect I have models (formal, written down things/programs I could give another person so they get the same number I do) for fewer than a quarter of the questions I trade on. The rest are just my opinion (expert or not, depending on the question).
There are quite a few questions dealing with technological development, mathematical breakthroughs, and the business decisions of tech companies. None of those really seem amenable enough to modelling to make it worth the effort to try.