July 17, 2012
Playoff Odds Updated
Due to reader feedback, we’ve made some changes to the Playoff Odds report:
1) Rather than rerunning sims for past days of the season, we’re locking in sims that have already been run. Originally, we were rerunning sims from previous days to make sure that only updated information was affecting the deltas, but the practice seems to have generated too much confusion as to what the deltas were supposed to mean.
2) In order to smooth out the deltas a bit more, we’ve roughly quadrupled the number of simulations that we’re running each day. That should reduce the amount of random variation we see in the playoff odds.
There are also some things that we have not changed but thought we should take a moment to explain in some detail anyway. The Playoff Odds are governed by expected rest-of-reason results, and as such they depend on our best estimate of what those results will be.
Using the depth charts and our rest-of-season projections, we come up with one estimate of a team’s probable winning percentage going forward. Using a combination of our estimated team runs scored and allowed based on components, the linear weights underlying True Average, and the Pythagenpat formula, we come up with another. We take a weighted average of the two, based on our estimate of the reliability of each.
Now, there is quite a bit of agreement between these two measures, with a correlation of .75. But they don’t always agree. Teams upgrade personnel—the expected wins of a team like the Phillies are much different with Utley and Howard on board than without those two players, for instance. Teams also play worse or better than their personnel would indicate. The more games by which a team manages to exceed its projections, the more weight is placed on observed rather than projected performance. (And thanks to the rest-of-season forecasts, that over-performance also gets incorporated into the projections, to a lesser extent.) We feel that we can provide a better expectation for rest-of-season wins by combining these two estimates based upon our estimated uncertainty than by using either method alone. We’ve published the exact method we use to estimate reliability of one relative to the other, so anyone should be able to look at our third-order wins and the depth charts and see where the estimates of expected rest-of-season win percentage are coming from.