Filed under:

# Demuddling some common statistical mixups

So it seems like a lot of people are missing a crucial point in player evaluation:

The stats used to evaluate pitchers are not necessarily the ones we should use to look at hitters.

Analysing run prevention is a different beast to measuring run scoring, and that's because we're trying to solve seperate problems with different constraints. The key concept here is that runs win baseball games, and we should consider the twin goals of calculating present run value and future run value as paramount.

When attempting to measure a player's output in terms of runs scored, we need to figure out the individual contribution to what is a team effort.  Since each player is only 1/9th of a lineup, we can't blindly look at a given player's runs scored/runs batted in. We must determine the components of players that best describe their offensive worth, which is where OBP, SLG, and more advanced metrics come in. These turn out to be reasonably stable year to year.

We run into a different problem with pitching/defence (if everyone will permit me to ignore the difficulties in isolating a pitcher from his fielders for a few minutes). We know how to model runs scored. We take... runs scored against. The correlation here is 1.00, so there is no need to look at components to measure present value of the pitching/defence units. The problem lies in the stability of our run prevention statistics - namely that there barely is any. The components used to evaluate a pitcher should be those that decrease the volatility of our future run value estimate, and therefore they need to be defence independent. This is the province of FIP and tRA.

So when you feel like applying tRA to hitters or OPS against to pitchers, remember the context in which the stats were developed. It's very easy to think that they'll be accurate when applied to the other side of the game, but doing so loses sight of the constraints they were designed to overcome.