Contextual versus Counting Statistics

I was alerted to fellow SABR member Cyril Morong's analysis of game situation or contextual statistics such as Player Win Averages (PWA) and other similar stats spurred by the recent article in Business Week which I blogged about a while back.

I don't disagree with Cyril's conclusion that measures of these sorts correlate very strongly with OPS, Runs Created, or other simpler and non-context related statistics. This is simply because the common sabermetric wisdom is that "clutch hitting" as defined in the popular mind doesn't exist and so the context dependent statistics will ultimately measure the same thing as the simple counting statistics given a large enough sample size. (I should mention that the authors of Curve Ball did find some evidence to suggest that hitting with runners on base may be linked to ability - "Although there is evidence that players differ in their ability to perform in the clutch, the evidence is relatively weak..." p108 of the 2003 revised edition).

However, I do appreciate these kinds of stats since they put a player in the context of his team and give an actual measure of how valuable he was to the team in a particular season given the way he was used (where he batted, how often, when he pinch hit etc.) This is because there is certainly a correlation between batting order position and who hits in front and in back of him and these contextual stats. From that perspective, contextual stats are the best way to pick MVPs of each team and perhaps of the league if you want to normalize the contribution based on a percentage of the team's wins.