THE BOOK--Playing The Percentages In Baseball

I read over both the THT article and PC’s post (and the comments) and decided to see what I could do to do my own sim scores. I used the following categories:

Age
Handedness
Weight
Height
PA
$BB=BB/PA
$SO=SO/(PA-BB)
$HR=HR/(PA-BB-SAO)
$H=(S+D+T)/(PA-BB-SO-HR)
$E=(D+T)/(S+D+T)
$T=T/(D+T)

Components were done as per MGL’s component regression article:

http://www.tangotiger.net/mgl/regression.pdf

Everything was converted into z-scores (minus average and divided by standard deviation) except for handedness, which I sorta faked. Sim scores are computed by summing the difference of sim scores in the different categories. Lower is better.

Here’s the top 100 comps for a few players:

http://www.editgrid.com/user/cwyers/sim_scores_test

I think that I need to weight PAs and BABIP more heavily. Any suggestions?

I think that the answer depends on what question that you’re trying to answer.  If you want to know whether players have similar run values, you should weight things to their run values.  Or just look at their total value/runs created/linear weights/etc.

If you want to look for qualitative differences between players, (passably non-slow power hitters vs. slow power hitters), then don’t weight things.  This may not tell you anything useful (that’s an empirical question), but suppose we know that different “types” of players age differently.  The trick is finding measures that reliably distinguish between types.

Pizza’s method seems much more relevant and applicable if we think about it in terms of projecting pitchers. Imagine how much more successful a projection system could be if we knew the kind of pitcher someone was. A soft-tossing minor leaguer would project differently than a fireballer if they had the same track record. I don’t think Zach’s method works as well for projections as Pizza’s would.

IOW, when scouts say a player “looks like” another player, Pizza’s method would match up with that evaluation better than Zach’s. I think.

Yeah, it all depends on what you’re interest is. I’m looking at sim scores as a means to projections, with some (limited) success. (At some point I need to integrate three-year weighted averages into the process.) I’ve since added in basestealing as:

$SBA = (SB+CS)/(1B+BB+IBB+HBP)
$SB = SB/(SB+CS)

80% of singles, 60% of walks and hit batters, and 0% of IBB should do the trick.  That’s what I use.

Those are the approx rates of how often 2b is open.