from Amazon
Saturday, January 30, 2010
Idea for a study…
I am thinking of making this a permanent thread, although I have not discussed this with Tango yet. The idea is to have a section where anyone can post a suggestion for a study that they can’t do themselves for whatever reasons. Then maybe someone else can take these ideas and run with them, perhaps writing an article on one of the various sabermetric web sites. (Of course, if there is already some research in that area, someone can point that out too.)
What do you guys think?
Anyway, the idea for a study that I presently had was this:
Take all the FA signings in the last X years, and split them into 2 (or more) buckets: The ones that were overpaid according to some projection system and the ones that were underpaid.
Then look at their performance (either rate or counting) in the subsequent year or years and see if teams/GM’s have any skills that enable them to project performance better than the projection system. After all, any GM or baseball insider you speak to will say something like, “Sabermetric projections are nice, but....” implying that they know some things that the projections don’t, like work habits, injury status, etc. Well, this kind of a study should shed some serious light onto that hypothesis!
You can also break it down by team/GM, although you would be getting into some sample size issues.
Has this ever been done?
Anyone up to it? It shouldn’t be that difficult to do. Basically look at projected WARP (or whatever metric you want to use that reflects value) from some projection system versus actual WARP and compare that to the salary paid to a FA minus their “fair” salary given their projected WARP.
Personally, I would use a rate stat for this kind of study to remove some of the fluctuations associated with playing time and injury, but in doing so, you are removing any possible skill at projecting playing time/injury by the team/GM. On the other hand, a GM/team, especially for their own players, probably SHOULD have better knowledge regarding playing time than a projection system would, so maybe using a counting stat (for example, WARP) would not be so interesting.
Probably both ways (doing it with a rate stats AND a counting stat) is the best way to go.
You can use the “buckets” method (what I like to call a “poor man’s regression"), or, since you have lots of data points, you can do a linear regression (X is dollars over or under “fair dollars” and Y is performance over or under projection in total runs/wins for each season or runs/wins per PA or some number of “games,” also for each season). Presenting both would probably be a good idea.
Recent comments
Older comments
Page 1 of 344 pages 1 2 3 > Last »Complete Archive – By Category
Complete Archive – By Date