from Amazon
Friday, May 09, 2008
By The Numbers
Two excellent issues from editor Phil Birnbaum (Nov 07, Feb 08). Here are my thoughts:
1. Eric does an excellent study about “low reward/risk” pitchers, which is those free agent pitchers signed for 1 year at a low cost. He finds that from 2002-07, they end up costing about 2MM per year per win. The free agent cost of players between that time period has been a bit over 3MM per win. However, if you combine all players (free agents, arb-eligible, slave), the average cost per win is right around 2MM per win.
So, what we have here is an EXTREMELY efficient market, whereby a team will pay a premium for “real” free agents (those that are actually being bid upon by many teams), and will pay the going MRP (marginal revenue product) rate for “happen to be” free agents (those who really have no sought-after value that would cause a bidding war).
Great job, Eric.
***
2. Victor expands on his study of the draft. I love all these types of articles, as they get into the risk/reward, but not with the immediacy of free agents, since it takes years to evaluate what you end up having. That Victor can conclude that the average hitter in the 51-75th best hitter class can be equivalent to any subgroup of pitchers (1-10, 11-25, etc) tells you all you need to know about the incredible risk that pitchers provide. Basically, picking a pitcher in the first round, by ANY team, is foolish. You’d have to be such a much better prospect than your typical first round pitcher to be even considered being pick in the top 30.
Now, it’s possible, as Victor pointed out, that he happens to have selected draft years where the pitching talent was fairly low. So, that’s a huge provision there, as he notes the pitchers who came after that didn’t make the study. And that makes the conclusions in the previous paragraph moot.
This is an area that deserves alot of research, and Victor has added to the already great research of Philly and Rany. And I look forward to alot more of it.
3. The conversion of R+RBI, outs into W/L is a good idea, but the implementation requires changes.
First of all, it’s not R+RBI you want, but R+RBI-HR. I’ve detailed the reasons already.
Secondly, W/L of pitchers make alot of sense, in that if you do Wins minus Losses, all divided by 2, you get a pitcher’s Wins above average (WAA). This works best at the career level.
So, if you want to create a similar metric, you need to ensure that a player’s hitting W/L adheres to this property.
Since we have R+RBI-HR, and you have outs, it’s fairly straightforward to get runs (and therefore wins) above average. RAA = R+RBI-HR - outs*x, where x is some league or team constant. If a team is +50 runs above the league average, then you set the “x” such that RAA = 50. Easy stuff.
RAA to WAA is fairly straightforward as well, using Pythag. Or, just divide by 10 if you want a shortcut.
Finally, since you have 162 games, you can get the “game share” for the hitter as his percentage of his team’s PA times 162. So, a guy who gets 11.1% of his team’s PA, he gets 18 “games”.
Now, we do have one problem. We are giving 162 “games” for pitchers (which includes fielders) and 162 “games” for hitters. But, that’s not the correct balance (because of the fielders), which is what the author came up against. The “shares” you want to give out is roughly 30% higher for hitters than pitchers.
The author comes out with his own divisors which are 30% higher than the original baseline we would have started with. Kudos to the author in making an arbitrary intuitive decision that in fact is justifiable.
Anyway, I don’t know how much what I’ve said here changes. It’ll certainly change alot for the Raines and Gwynns and Boggs.
So, a good idea, with an implementation that needs work.
And, as I noted in my original Win Shares article, all of these things are simply Linear Weights, but trying to come up with a second number to accompany it, to give it more context.