tRA update

If you're unfamiliar with my work on tRA, here's a decent explanation.

A correction

Since I came out with the 2007 spreadsheet, one thing has really bothered me. ROA (runs above average) doesn't sum to zero. At first, when I noticed it, I did a lot of mental gymnastics to excuse the fact, but eventually I realised why this was happening: I was using innings pitched as the measure of how much the pitcher threw, rather than the entirely more sensible option of total batters faced. Why was that a problem? Because pitchers face a different number of batters per inning. Once I readjusted the formula for ROA to work with TBF rather than IP, leaguewide runs above average summed to zero. Problem solved.

An improvement

tRA+ is like ERA+, except with tRA. It's useful for comparing pitchers on a per-inning basis, as I doubt many of you have a good grasp on league averages for tRA. It also lets me start working on projection systems because around tRA, because it gives a performance level which scales linearly - if you were predicting a tRA+ of 100 with a 2% improvement, you'd expect a tRA+ of 102 (ok, ok, I'd use the regressed versions for projection). If you're just given a tRA of 5.00, how do you know what a 2% improvement would be? You don't. This is the first stepping stone into turning this from an interesting analytical tool to what would probably be one of the better pitching projection systems around.

Thoughts on defense

xRuns-R (expected runs-actual runs) for a team should be a pretty good measure for team defense over a year. As it turns out, there's a problem: Summing for the AL (using proper park and HR factors) yields a value approaching -200. The NL? Roughly +200.

Now we could say that the NL is ridiculously good at defense and the AL is terrible. What I think is happening, however, is that the run environments differ to the point that it skews expected run values for certain events. In the extreme case, a HR in the NL is going to be worth less runs on average than a HR in the AL, because the good league will tend to have more baserunners. When you apply that to every event, you'd expect NL pitchers to give up less runs than their AL counterparts (which they do). My current work does not take this into account, although it's easily solved - I plan on including a run environment modifier in xRuns (it won't affect ROA or anything like that), and when I do team defense should be pretty easily calculable from xRuns-R.

Future Plans

Minor league tRAs are in development (i.e. being thought about). A projection system incorporating tRA is also... in development. Hopefully both will see the light of day in 2008. Also on the cards is an automated data gathering system. That would make me very happy.

Any questions, or if you just want to poke the newest incarnation of the spreadsheet, feel free to get in touch.