[Spraguer]

Hmm...

Sorry I'm so late to the party here, guys. It's been a very hectic weekend.

I think that, in order to answer the question, we have to really figure out what it is that we are looking for. I think, in order to prevent this from becoming just a semantic issue, I'll respond as though the question were "which factors and/or statistics best predict a pitcher's contribution to his team's run prevention?"; if that makes any sense. This way, we can ignore things outside of the pitcher's control, such as defence, bullpen support and park factor. Ignoring these things can be tricky.

A pitcher controls: the walks he allows, the strikeouts that he gets, the homeruns that he allows, and batted ball type. The measures that we want to use are k/bb, k/9 and hr/9. Batted ball type is incorporated into this mostly through the hr/9 measure (groundballers don't allow as many, obviously).

How you want to weight those things is up to you. Any approach that focusses on the "three true outcomes" k, bb and hr, will work; you can find measures such as DIPS (defence independant pitching stats), FIP (fielding independant pitching), or BPV (base performance value).

The trick is to ignore ERA. Don't even look at it. It's very misleading.

Peripherals - BB, K and HR - are a better predictor of future ERA than is past ERA.

After you've determined how good the pitcher is, it's time to add in the contextual factors, like defence, bullpen support and park. Take somebody like Mike Maroth - he is a well-below average pitcher, but his defence, bullpen and park are so good for pitchers that a game started by Maroth is likely to be right at or even below league average in terms of runs allowed.

K, BB and HR rates (park-factored, of course) are remarkably consistent from year to year for each pitcher - but contexts change all the time.

What I'm trying to say is that you have to remove context first and boil it down to K, BB and HR, and then re-contextualize it based on the circumstances of the game that you are capping.

Figure out how good the pitcher is and then add in the context. Looking at past ERA just gives you hints about past context, which is no longer relevant to you.

Now, for re-incorporating the current contextual factors after you've weeded out the past ones, you have to decide how you want to handicap and weigh bullpens, defence and ballparks. That is a whole other topic.