1681

Facebooktwittergoogle_plusredditpinterestlinkedinmail

this strip's number is my birth date, if you use day-month-year format

I think it doesn’t quite work, because to determine the accuracy of each individual 90%-chance prediction, you’d have to re-run that specific situation 10 times with all known variables the same, and see if it got the correct answer 9 times out of 10… which of course would be an impossible experiment. Percent-chance predictions really are unfalsifiable.

TEXT OF COMIC:
Predictions with percentages are such a cop-out.
Huh?
If I say there’s a 90% chance of rain tomorrow, or my candidate has a 90% chance of winning the election– no one can say I’m wrong.
If it doesn’t rain, or if my candidate doesn’t win, that doesn’t mean I was wrong. It just means something happened that had a 10% chance of happening.
I guess people would have to judge your predictions on a larger scale, instead of individually.
Yeah. If I’ve given a 90% chance of a certain outcome, like, ten times in a row… and my preferred outcome has happened nine times, and failed to happen once… then I can say my predictions are 100% accurate.
I’m… not sure that’s how it works?

MOUSEOVER TEXT: this strip’s number is my birth date, if you use day-month-year format

Facebooktwittergoogle_plusredditpinterestlinkedinmail
Bookmark the permalink.

3 Responses to 1681

  1. Susanne Nieß says:

    The correctness of your predictions cannot be proven but falsified: If they are wrong in much more than 10% of the cases, this means that you are not very good at predicting something with a 90% chance.

  2. Antonymous says:

    The concept Abby is after is calibration, not accuracy. If her 60%-confidence predictions are right 60% of the time, then she’s perfectly calibrated: she’s inaccurate, but she knows exactly how much.

Leave a Reply

Your email address will not be published. Required fields are marked *