Yin and Yang

I thought I'd finish off January with a couple of links...

Lovin' on Bannister. MLB Trade Rumors did a great interview with Royals pitcher Brian Bannister in three parts. Part 3 is where it gets really good as Brian reveals that he does his own statistical analysis (we already knew he was a BP reader) and gives us his take on DIPs theory. Several well-renowned analysts have already started the discussion into his insights and I'm sure we'll be seeing more in the future. What really encourages me about this is recognizing the value that thoughtful players like Bannister of Jeff Francis can provide and it makes me wonder how teams are utilizing those resources in parterning with the analytical resources they have.

Not so Much Insight. Kind of the anti-Bannister kind of observations were offerred by MLB.com reporter Marty Noble back in early January. In a previous article Noble used RBIs per 100 at bats to make a comparison between newly acquired catcher Brian Schneider and departed backstop Paul Lo Duca. In the response I've linked he tries to explain himself and although his main point that Schneider and Lo Duca are no longer as different offensively as some people claim is valid, there's simply no way he can get out of the hole he's dug. He has the right idea, namely that opportunities are important and rate statistics rather than counting stats are key, but of course he fails to select the right kind of opportunities to make the kinds of comparisons he's going for.

That said, he dropped two little gems that I couldn't pass up:

Computers have contributed to a current glut of statistics that, to a degree, distort the picture. We have so many now that we lose focus on what is most important. The objective of the game is to win, and to win a team must outscore its opponent. Nothing, therefore, is more important than runs -- both producing and preventing them.

To what degree and to which statistics is he referring? Actually, I would argue that by translating traditional statistics into the currency of runs assuming an accurrate weighting, the vast majority of the supposed "glut" of statistics (VORP, BaseRuns, Linear Weights, defensive metrics, base running, etc.) have served to paint a more accurrate picture of "what is most important" - creating run differential that leads to winning games.

That Lo Duca might have had a higher on-base percentage or slugging percentage means less to me than the number of runs he produced. The next time a team wins a game because it produced a higher on-base mark and scored fewer runs than its opponent, please alert me.

Here I think there are two points of confusion.

First, it turns out that the very combination of metrics he mentions, on-base percentage and slugging percentage (OPS), is a very strong predictor of runs produced since it accounts for the key ingredients (getting on base, moving runners, and avoiding outs) that are so problematic in looking at things like RBIs per 100 at bats which only measure one part of the equation. Additionally, by not accounting for context nor understanding how other metrics predict offensive output Noble ends up inverting the relationship between offensive production between the statistics he discusses.

Second, in his last sentence he stumbles across the problem of scale. It is tautological to say that run differential is a perfect predictor of wins and losses at the level of an individual game. Therefore RBIs and run scored (at least for the offense) take on primary significance in that context and at that scale while OBP and SLUG are less predictive. However, once you raise the aggregation level, those counting stats take on less significance in player evaluation because a particular player's role in generating offense is about more than the tallying of the end result (an RBI or run scored) to the point where it quickly becomes the case (and well before the level of seasons) that OBP+SLUG and other derivative metrics are more indicative of offensive contribution and therefore wins and losses.

This confusion of effects at various scales reminds me (not coincidentally because I'm now reading this book) of one of the primary themes in the writing of the late Stephen Jay Gould. He often railed against the position of ultra selectionists or adaptationists who insisted that natural selection was the exclusive driver and shaper of the pattern of life on earth. Gould contended that evolution operated differently at different levels through various mechanisms and that what worked at one level did not necessarily have power at another. For example, he argues that while natural selection works through differential reproductive success to build adaptations at the level of individual organisms (coloring, wings, claws, size, etc.) those adaptations may have little or nothing to do with survival at the higher level of species. In one of his favorite examples he liked to point out that the small size and adaptability of mammals during the age of the dinosaurs was likely the result of the domination by dinosaurs in the niches available to larger animals. However, when the meteor struck it was those "negative" traits that allowed the mammals to survive but doomed the dinosaurs.