Julien's Faster than Light Blog
Jump to top
Saturday, June 21, 2003
mailbag
paxton from chicago:
ok about your numbers:
wal = (bb + hbp) / (ab + bb + hbp)
con = (ab - k) / (ab)
pow = (tb - h) / (ab - k)
maybe you could discuss why you chose these exactly, because my impulse would be to use these formulae:
wal = (bb + hbp) / (ab + bb + hbp) // unchanged
con = (ab - k) / (ab + bb + hbp) // changed denominator
now con is more directly comparable to wal, also k-rate is 1 - wal - con. i think that's cool. also, philosophically, i think it better because when you are thinking about outcome of a partcular at bat you can just use the con number directly as the chance of the ball being put in play, rather than being forced to first multiply it by the chance of not walking. does that make sense?
pow = tb / (ab - k) // changed numerator
i think this number is just more elegant. and look at what you can do with it:
pow * con + wal = offense in bases per plate appearance
doesn't that seem like a great single number to compare players with? and it is so easily derived from the detailed stats pow, wal, and con. how good is player x? look at this number. why is he that good? look at wal con pow.
the reason for the stats i'm using is they represent the three fundamental skills of hitting: first, whether or not the hitter walks. (2) if he doesn't walk, then does he put the ball in play? (3) if he puts the ball in play, how hard does he hit it?
those are some cool features you describe, but the system as you construct it is just another "how good is this guy?" system. There are many, such as equivalent average, runs created per 27 outs, and ops. all have special features, but all basically say the same thing.
my system, i contend, takes into account the nature of the game, thus is a constructive way of thinking that increases understanding.
the reason hits are not included in pow is we want to be able to get meaning from small sample sizes, and hits are the most variable thing out there. it takes 1000 at-bats for hits to normalize. plus, the more extreme ratio between how hits are counted (hr/3b/2b 3/2/1) makes differences show up faster. ie it takes less plate appearances to know that corey patterson hits harder than troy o'leary than it used to.
the thing to bear in mind with pow is that home runs in some parks are doubles in others, so look to make sure that the player's hr and 2b are in line with career averages. actually, a refinement of pow would substitue for hr and 2b the following:
(hr + 2b)(career hr percentage) and
(hr + 2b)(career 2b percentage)
hr % = (hr) / (hr + 2b)
2b % = (2b) / (hr + 2b)
or instead of career you could use last three years.
Comments:
Post a Comment
