2063 - Marcel Projections: 2063 Pitchers Edition *REVISED*

[Bob Breum]

Here are the Marcel projections for pitchers. It uses a weighted average of the last three seasons with some regression to the mean and an adjustment for age. *** Revised ***

See my previous post with the batter projections for more on the methodology:
http://montybrewster.net/forums/viewto ... 23&t=55082

Alphabetic sort:

Marcel Pitching.csv

(94.13 KiB) Not downloaded yet

Team sort:

Marcel Pitching by Team.pdf

(289.18 KiB) Downloaded 6 times

.csv format:

Marcel Pitching.pdf

(277.73 KiB) Downloaded 2 times

[cheekimonk]

Not liking the +4.00 mERA across the board

[Bob Breum]

Not liking the +4.00 mERA across the board

There are two issues with this release. First and foremost is that the projections are too flat. Let me explain my diagnosis.

Tango uses a technique to regress the projections to the league mean. With batters, his formulae are well documented. It involves comparing a weighted average of plate appearances over the last three seasons to the sum of 1200 (two seasons of 600 PA each) plus the weighted average and adjusting the projections based on this. He calls this ratio "Reliability." In other words, the more data that we have for a player the less need there is to regress toward the mean.

For the pitching projections, he just sort of waved his hand and said it's the same as the batters, with a couple of tweaks. He did not mention any revision to the reliability calculation or the 1200 factor. I looked at the Python code that was available and it appeared to use 1200, so I went with it. You can see the results.

After consideration, I have decided to change the 1200 to 400, based on two seasons of 200 IP. This results in projections that are much closer to what all of us were expecting.

The other, lesser, issue is that a number of young pitchers with no BBA experience snuck into the dataset. Once I get this resolved, I'll post an updated version.

[Bob Breum]

The revised release is now available.