THE BOOK--Playing The Percentages In Baseball

A few years ago I saw some private work done by Sig Mejdal on this very subject.  Basically, a player is most likely to get injured returning from an injury, and the probability goes down following a power law curve.  The longer a player remains healthy, the less likely they are to get injured.

Unfortunately, I think that you will find that there are many more lawyers around here than doctors.  There is a significant proportion of lawyers who hate their jobs, and sabermetrics provides a nice relief from the pain.  Also, it is much easier for lawyers to play around with baseball statistics during “work” than doctors.

- Matthew Bultitude, Esq.

Uh, yeah, good point!  Don’t know why I didn’t realize that doctors and construction workers are under-represented on message boards.  Good call

This also depends on how much truth the teams tell us.  i.e. what the actual injury, and how much detail there is.  Basically, it may be impossible to get useful numbers if we get no useful information.

This is not entirely true.  As long as it’s at least 1% useful, then you will get something that is not random (albeit at a high uncertainty level).

As long as the lies we’re told are random, not systematic, then their lies simply mean that we need to get a larger sample size.

There is so much more that it would be useful to know.  From an actuarial point of view it would be helpful to have flags in the injury database for non-baseball related sickness or injury, batting injury, baserunning injury, position specific fielding or pitching injury.  Also for new injury or reoccurence of a previous injury.  I would also expand the database to include all lost player time and injury information that did not cause lost time, but may have caused a decrease in performance.  It is important to know whether a player who is normally a starter is not in the lineup because he is sick, or injured, or on bereavement or family leave, or is being rested, or is being platooned, or has been suspended by the league or team, or replaced due to lack of performance. 

I mentioned in the other thread that the Gameday XML files have much more additional information that could be useful.  For example, in September of 2009 there are 20 DL instances in the current injury database.  But in the XML files there are 50 instances of players leaving the game due to an injury and several hundred more injury time outs that did not result in a player leaving.  The player ID is given in both instances and the specific injury is described if the player has to leave the game.

A comprehensive database like the one I envision would require a lot of work, but would yield valuable results.

Peter, fantastic stuff.  Agree with you all the way.

I took a look at the predictive value of Will Carroll’s red, yellow and green lights (based on his actuarial table):

http://sites.google.com/site/steamerprojections/quantifying-carroll

I’m hoping you guys will give me some feedback on where to go with this (if anywhere).