Friday, April 07, 2006

Casual Conversations

Over at Black Betsy, each game so far this season has been the subject of what's called an "ERV box score". Essentially, what SuperNoVa is doing is computing the net change in expected run value between two situations and crediting the batter with it. I'm sure he's published which matrix he's using. I had wondered about doing a similar thing every since I picked up a copy of The Hidden Game of Baseball by John Thorne and Pete Palmer two decades ago. Palmer developed the matrix as a basis for his linear weights method (LWTS). SuperNoVa actually did it.

(Aside: I didn't actually like Hidden Game, by the way; I didn't like the "I've got a secret and I'm willing to share it with you" tone of the book. Hidden Game reads like a college professor teaching a lecture hall of 500 students with a big VU-graph projector, absolutely certain that his information is infallible. The rival Bill James Baseball Abstracts of the time were like a seminar taught in a bar; you were with friends, and this guy James was always careful to explain how his formulas were created, and to assert that they were quite imperfect, but they were the best ones he had. The Palmer/Thorne book was off-putting where the James books were inviting. I have no idea who was "more accurate".)

I don't think the method is perfect, by the way. It has three obvious weaknesses, all related to lack of context, and all only pratically "corrected for" by assuming "all else is equal":
  • The batter is not the only factor in the outcome. Not only the pitcher is important; the fielders, ballpark, and umpires all can change the "expected" outcome of an at-bat. A two-out bases-loaded grounder in the hole that is incorrectly called "safe" on the throw to second would score as a +1.00 instead of the earned value, which is negative. (I'd say how negative but I can't locate my copy of Hidden Game.)
  • All situations don't have equal best-case or worst-case scenarios. A way to improve the method, I think, would assess both the change in expected value and the best and worst case scenarios possible for each situtation, and grade the outcome as a percentage, perhaps called "Net ERV efficiency". I don't think this is a substitute for the raw numbers, but as a separate metric it would be meaningful.
  • All situations don't have equal leverage. This is best solved with win probabilities, which other people have done.
So it's not perfect, but it is probably very useful. I think the Net ERV per plate apperance, the average Net ERV efficiency, and the Win Probability Score would make a very interesting (and inter-related) trio of numbers to go with the New Holy Trinity (AVG,OBP, and SLG).

As I wrote, I insist I did think of this idea twenty years ago; but I discarded it because I assumed somebody had already done it and discovered that it wasn't worth the extra work as it closely tracked the New Holy Trinity. Why? Because, over the course of a season, "All Else Is Equal". Now I'm wise enough to believe it's never been done (much) because it's been too much work, and twenty years ago, before STATS and Project Scoresheet/Retrosheet and the cornucopia on the Web, getting play-by-play data for most major league baseball games was impossible unless you got a scoresheet (or you kept it yourself).

What would be most interesting is if there were sustainable season-to-season correlations between (ERV/PA,NERV,WPS) that didn't track the New Holy Trinity, because it would mean that the idea of "productive outs" truly does have more merit than the majority "All Else Is Equal" crowd insists!

1 comment:

SuperNoVa said...

The thing is, Doug, is that it can show that someone like Timo Perez wasn't quite as bad as his stat line suggested in 2004. I went through all of his at bats that year (link) and found that his actual ERV was about 9 runs better than his stat line - Runs Created Above Average - would have otherwise suggested. Now, he was still -7 ERV (vs. -16 RCAA), but he was measurably better, which matched the anecdotal (Harrelsonian?) evidence about Timo.

P.S., I use the 2005 BP Expected Runs Matrix (subscriber only, if you can believe it). But I can plug in any Matrix to the spreadsheet and pop out the results.

P.P.S. DM over at the Nats Blog does a win value - based scoring. His system is more complex, but he did the entire 2005 season for the Nationals. The WV scoring was disappointing inasmuch as it placed a disproportionate value on closers.