I ran some rough numbers on this last weekend looking at some DMB 2007 preseason projections, Nate Silver’s PECOTA projections, Nate Silver’s predictions after he tried to adjust things manually, Joe Sheehan’s projections, and the preseason predictions from a poll of espn.com baseball writers. I don’t have the numbers handy right now, but the DMB and PECOTA projections did the best, coming in with a reduced chi square statistic of about one. Nate Silver made his predictions worse by deviating from PECOTA, but not a whole lot worse. The espn.com writers were a fair bit worse (reduced chi square of around 1.5 I think) and Joe Sheehan was last in the group with a reduced chi square of around 1.7.

Basically, the objective preseason predictions did remarkably well if you’re comfortable with the idea that a team’s win total is the sum of 162 random binomial events. The standard deviation on such random variable is going to be about 6 wins over 162 games, which gives a sort of limit on just how well you can predict the season (like there’s a limit as to how well you can predict a series of coin tosses.) If that’s the case no one is going to get much closer without being really lucky (or if games are somehow more pre-determined with match-fixing or something like that.) And for all of the “adjusting for the human element” that analysts can do by hand, they did a fair bit worse than the objective systems in 2007.

2007 could turn out to be special, but I doubt it.

This might actually be a good year to subscribe to Extra Innings to watch some of the other teams.

BTW, why does BP make it so hard to figure out how they project playing time on their depth charts?

Be that as it may, the difference in win values between Ichiro’s PECOTA projection and his real values wont be significant enough to think that the projections here are absurdly low. One or two wins added when your projection is 73 wins wont make a ants’ nuts worth of difference.