Baseballogy

just keep it clean and non-commercial

Baseballogy

Postby DavidTate on Saturday, 20 September 2008, 12:48 am

Eric, in another thread you spotted the "baseballogist" in my .sig file, and asked whether we'd ever explored that common interest. I think we have noted it in passing, but not in any detail. Since we last corresponded, I have been very impressed by your analysis of the "link" between steroids and performance.

My own credentials are much more modest than yours. I was an active researcher for a time in the '90s, mostly through email correspondence and collaboration with the denizens of rec.sport.baseball on USENET, and a few SABR members in the Pittsburgh and (later) DC areas. My biggest claim to fame is that I invented Marginal Lineup Value (MLV), which Keith Woolner later extended and incorporated into his VORP metric for Baseball Prospectus. (BP was founded by several of my rec.sport.baseball e-friends.) Keith is now with the Cleveland Indians front office, and I was tickled to death to see him respond in an interview once to say that, no, Bill James wasn't really a big influence on him -- he thinks of himself more as a David Tate disciple. Now has the student surpassed the master, needless to say.

I haven't done any original work for several years. The last thing I was working on was some unpublished work using linear programming to explore the range of offensive values that players with the same OPS can have. It was interesting, at least to me, to learn that Barry Bonds had pretty much the optimal mix of batting outcomes -- it would be almost impossible for a player with the same OPS to be more valuable offensively. The least valuable mix for a given OPS tends to be a high triples-heavy SLG and low OBP, which is reasonably intuitive. In the last couple of decades, Cristian Guzman and Lance Johnson have come closest to such a mix, so that OPS over-valued their contributions by quite a bit.
David Tate
Professor of Story Problems, emeritus
Rationalist with sombrero
Baseballogist
DavidTate
 
Posts: 187
Joined: Tuesday, 16 September 2008, 9:38 pm

Re: Baseballogy

Postby owlcroft on Saturday, 20 September 2008, 3:13 am

I have always been amused and perplexed by OPS: it is additive when it ought to be multiplicative. Ultimately, the probability that a batter will score a run is the joint probability--and thus the product--of the probability that he will get on base and the probability that he will then be driven in.

Of course, the OBP needs to be adjusted by a factor for the chances he will be thrown out on the bases after having safely reached; but the big variation in methodology is the exact nature of what we may call the "RBI factor". In an OPS-style measure, it is simply the slugging percentage, a measure with all the defects of the batting average, chiefly that it ignores walks. Even so, we'd have a fair proportional (not absolute) measure were OBP multiplied by SA.

(A more sophisticated scheme gives different weights to base hits than the simple number of bases, because each hit, over the long term, participates in both the quality of an on-base and an RBI: the greater the number of bases the easier to later score the man, and the greater the number of bases, the more likely an RBI, so the value ios not linear per base.)

The other major adjustment that many methods overlook is the "compound interest" effect of OBP: not only does a higher OBP increase the chances of a given runner getting on, it raises the overall number of men who will have the chance to get on (or drive in a run).

Bu there, I'm running away at the mouth (or keyboard) again.
Cordially,
Eric Walker, webmaster
User avatar
owlcroft
 
Posts: 213
Joined: Sunday, 14 September 2008, 6:36 pm

Re: Baseballogy

Postby DavidTate on Saturday, 20 September 2008, 10:14 am

owlcroft wrote:I have always been amused and perplexed by OPS: it is additive when it ought to be multiplicative. Ultimately, the probability that a batter will score a run is the joint probability--and thus the product--of the probability that he will get on base and the probability that he will then be driven in.

Certainly any explanatory model of scoring needs to take that into account. I have no problem with OPS as an indicator -- a meaningless dimensionless value that happens to be reasonably well-correlated with something we care about. It doesn't explain scoring at all, but it does predict it modestly well -- certainly better than any other measure that can be calculated by eye from data published every day in USA Today. It's a blunt tool, but sometimes a cheap hammer is appropriate to the task at hand.

That said, you are perfectly correct that the underlying relationship is multiplicative at heart. Bill James got that right with his basic Runs Created formula, which still works pretty well at the team level. (MLV, as it happens, was conceived as a way to make RC meaningful for individual players, by replacing the player's own OBP and SLG in the RC formula with the resulting team OBP and SLG that would result if you substituted that player into an average lineup.)

Cheers,
David
David Tate
Professor of Story Problems, emeritus
Rationalist with sombrero
Baseballogist
DavidTate
 
Posts: 187
Joined: Tuesday, 16 September 2008, 9:38 pm

Re: Baseballogy

Postby owlcroft on Wednesday, 24 September 2008, 6:02 am

. . . replacing the player's own OBP and SLG in the RC formula with the resulting team OBP and SLG . . .

That's one way to go about it, but my sense of it is that it was devised to get around the problem that most of the sundry formulae then running around were felt to over-estimate power, and that the "nine clones" approach would thus over-state the value of home-run hitters.

The first part was, in general, true" most or all of The Usual Suspects did tend to over-value power; but the defect is easily identified, which is that they did not--few if any do even now--take into account the "compound interest" effect of on-base percentage. By that, I mean that a high OBP certainly increases the probability that a given player will get on base, and thus end up becoming a run scored; but it also acts to increase the total number of men who will come to the plate in a game, thus also increasing the number of men who have a chance to get on base and later score. Because The Usual Suspects did not, so far as I ever saw, take that compounding effect into acccount, they did not so much over-estimate power as under-estimate (in a relative sense) OBP.

The formula I have long used, and its efectiveness, can be seen at the here-linked page of The High Boskage Baseball Site. What pleases me about the graph of predicted vs. actual is that the most extreme points, high and low, are almost perfectly on, demonstrating that there is no systematic bias toward high- or low-scoring teams.
Cordially,
Eric Walker, webmaster
User avatar
owlcroft
 
Posts: 213
Joined: Sunday, 14 September 2008, 6:36 pm

Re: Baseballogy

Postby DavidTate on Wednesday, 24 September 2008, 2:27 pm

MLV explicitly takes account of the "compount interest" effect you cite, by computing the derived new team RC on the basis of the derived number of team plate appearances that the new team OBP will result in. It's a counting stat -- marginal team runs added -- not a rate stat. (Though there is a derived rate stat, MLVr, which is simply MLV per PA). As you say, the compounding is the key; ~1.5 additional plate appearances for your teammates for every time you fail to make an out is a very big deal over the course of a season.
David Tate
Professor of Story Problems, emeritus
Rationalist with sombrero
Baseballogist
DavidTate
 
Posts: 187
Joined: Tuesday, 16 September 2008, 9:38 pm


Return to Everything Else

Who is online

Users browsing this forum: No registered users and 0 guests

cron