A BABE in the Woods

I noticed a week or two ago that Allen St. John in his "By The Numbers" column in the Wall Street Journal introduced a new rate stat to rank pitchers called BABE or bases per batter. This stat is calculated as follows:

BABE = (TB + BB) / BFP

where BFP is batters faced by pitcher. As you can see this stat is attempting to measure how many bases a pitcher gives up per plate appearance something like Total Average (TA) for offensive players developed by Thomas Boswell in the early 1980s. St. John makes the point that some traditional stats such as wins and losses don't accurately measure a pitcher's effectiveness and so a stat like BABE is needed. The example he gives is Randy Johnson's 11-8 record but .345 BABE, currently leading the league. I can't argue that BABE is a better measure than wins and losses but is BABE really needed?

When you boil it down wins and losses are fundamentally about runs. Indeed Bill James calls them the currency with which wins are purchased. That's the reason that when evaluating offensive contributions sabermetricians strive to estimate the number of runs a player has contributed through formulas like Linear Weights, Runs Created, Base Runs, and so on. It seems only natural that for pitchers we would want to do the same thing by estimating the number of runs they are responsible for. But we already have that with Earned Run Average. ERA organically captures the interaction of offensive events that result in runs. BABE merely adds up the individual bases and does not weight offensive events like Linear Weights does or take a multiplicative approach as does Runs Created.

And so as with TA, BABE cannot be as accurate (in terms of predicting how many runs a pitcher would give up) as these other formulas. Since it combines components of on base average and slugging percentage it will be more accurate than either of them taken separately and yet not as accurate as OPS (OBP + SLUG) or BRA (batter run average = OBP * SLUG) as demonstrated by Albert and Bennett in Curveball. If anything St. John would be better off looking at OPS against the pitcher.

But ERA does have its problems, the two biggest of which are:

1) It doesn't include all the runs that the pitcher is responsible for since some percentage of unearned runs will be the pitcher's fault. See this post for an attempt at assigning responsibility

2) It doesn't work very well for relief pitchers since they pitch alot of partial innings and come in with runners on base

And so clearly at least for relief pitchers ERA is not the stat that should be used for comparisons. For starters and long relievers I think ERA is the appropriate measure. However, in order to compare all pitchers, rather than invent a new stat St. John should look no further than the existing sabermetric work of Component ERA (ERC) or Expected Earned Run Average (XERA), both of which were derived just for the purpose of comparing starters and relievers and which get to the heart of the matter - runs.

In summary, while BABE is not without its merits it can be safely discarded in the sabermetric trash can.