Originally Posted by blandman
Do you have a basis for this or are you just using two outlying sox seasons as your basis for their body of work?
Over the last 8 seasons (including every one of them--not just 2005 and 2008) Baseball Prospectus underestimates the Sox by an average
of 7 wins. 8 seasons is a sizable chunk to be that wrong about, and 7 wins is, coincidentally, the difference between the 2012 projection of 78-84 and the actual record of 85-77. It took the Sox taking everyone by surprise last year and leading the division for most of the summer to even reach the average
amount that BP is wrong by. The system does not take intangibles into account at all--as evidenced by the 82-80 projection the year after a 99-win championship season
. Like clockwork, the Sox outpaced the 2006 projection by 8 games. You didn't have to look far to find other publications picking the Sox to win at least 90 games in 2006--that's because most other projections take such intangibles into account--and guess what? Intangibles apparently matter, because the majority was right that year, and BP was as wrong as ever.
Originally Posted by doublem23
Well first off, anyone who understands Stats 101 knows that even good models turn out bad results from time to time.
6 years out of 8 is not "from time to time"-- it's 75% of the time.
My final argument is this: when numbers are blindly used as the only basis for prediction, the prediction will very often fail. Here's a simplistic analogy: If you roll a die 1000 times and take the average of all the values that come up, you will get something close to 3.5. Yet only an idiot would actually believe that this means that 3.5 was the most likely number to come up on the next roll, because he would be failing to account for other factors, such as the fact that 3.5 is not a number that appears on any side of a die. Similarly, someone who believes a 162-game baseball season can be reasonably predicted just by looking at numbers--and actually makes important decisions that are based on those projections--is just as big a fool.