By David Roher
I wrote a guest post for the LoHud Yankees Blog that appears this morning. I’m very grateful to Chad Jennings and Sam Borden for giving me the opportunity to write a piece for the best beat blog in the business. The post is about Joe Girardi’s decision to have Jose Molina catch A.J. Burnett in the playoffs, instead of having Jorge Posada do it.
But it’s really an introduction to an idea of mine, Result-Change Probability. I suggest you read the article for more info on it. I’ll be developing it more later as we approach Opening Day. I’ll take the space here to clarify/expand some points I made over at LoHud.
- The analysis could have gone a lot deeper into the specific situation. For instance, it based the quality of the Yankee offense, and the probability that they would win given a certain run total, on regular-season results. Because it was a playoff game, the level of competition brings the expected run totals down, as well as the win probabilities. As a result, Posada’s offensive value increase a bit: the runs he adds to the offense are more likely the 4th, 5th, and 6th instead of the 7th, 8th, and 9th, when the game is probably already won.
- The idea of splitting the win into two separate concepts is somewhat superficial in this scenario. The difference between Posada and Molina was always 1.75%, regardless of how I ran the simulation. But the 14% figure for Molina doesn’t mean that Molina is causing the win, nor does the 16% figure for Posada mean that Posada is causing the win. In the final version of Result-Change Probability, I would like these two numbers to have more significance in and of themselves, to mean that the players are actually causing the win/loss over the other.
My idea is that some players have skills that cause them to be better at preserving wins, while others are better at converting losses (Great closers vs. great pinch hitters, as an example). The win as a single unit takes the difference between the two numbers, when it’s often interesting to look at both. In a contextual environment, this is kind of like Win Probability Added and Win Probability Lost. But I’m interested in a context-free version.