BP's 2007
postseason odds report is out.
though it confirms good bets on most of my picks, that's not the point. there are a couple of important things to take away from the report, both dealing with randomness.
first, the postseason chances for some of the teams might surprise you, but BP (correctly) adds in a random variable to account for the naturally imperfect ability to project a team's true level of talent during the season.
if you look at the
RLYW projections, you notice that despite using a fine simulation software in diamond mind, for most teams they come up with intervals of 12 to 14 wins between (-1 SD) to (+1 SD), so a 6 to 7 win standard deviation. this compares favorably to the standard deviation of
6.3 wins you get with absolutely perfect information. (i highly recommend that anyone who bets futures reads this article, which is sort of like a "fooled by randomness" for projecting team standings.)
second, you may also find it shocking that although no individual NL team projection is over 88 wins, the three division champs average 99 wins apiece, and the wild card averages 95. again, this should not be too surprising, though i do think these numbers look too high by a game or two. remember that these projections are inexact, so there's a reasonable chance one or two teams actually start the year with 90-win talent. add in the natural variance in baseball, and you're going to see a lot of dispersion from "expected" results.
imagine sending 16 breakeven poker players to play 162 hands apiece, or 10000 for that matter. do you expect their results to stay close to even, ignoring rake? i would suggest you're more likely to see a few who run well and a few who run poorly, just like in baseball.
not to trivialize the 2006 seasons of the tigers and cubs, but they fit this model well. you can look all day without finding a single viable source who had the tigers down for over 90 wins or the cubs for fewer than 70.
at any rate, projecting the standings with a SD of 6.3 games is impossible, and we should expect a true SD of around 9 games to account for injuries, playing time adjustments, trades, midseason callups, etc. IN ADDITION to the inherent coin-flipping variance.