from Amazon
Sunday, March 07, 2010
A new way of showing forecast results
Well, I gotta say: great job. Even though Chone won 52% of the head-to-head matchups with Marcel, Jared is showing Chone with a 93-69 (.576) record because:
How likely is it that Chone was really no better than Marcel this year and simply got lucky? Just as likely as it is that an 93-69 team was really an average team that got lucky. Not at all likely.
This is a really cool way to show it. And even though Steamer did better than PECOTA against Marcel, Jared is showing his own system as performing worse because:
The reason for it’s inferior record is that Steamer’s projections were considerably more similar to Marcel’s and therefore, given that it finished below Marcel, we can be more confident in rejecting the hypothesis that it’s just as good as Marcel.
I love the thinking behind it.