Current xERA?

[boness]

It appears that the calculation for YTD xERA may be incorrect. Looking at NL pitchers
that have over 100 innings, 49 out of 54 have a lower YTD xERA than YTD ERA. The
situation is similar in the AL. Either every pitcher has been unlucky or there's some type
of glitch.

[RobR]

The values in the sheet don't match what I get using the formula in the glossary with the default intercept of 2.77. And the error's not constant so it's not coming from the intercept.

[ajc730]

I haven't looked closely at this but has the YTD xERA been adjusted to reflect changes in the level of offense between 2006 and prior seasons? The formula for xERA that is published in HQ is based on simulations using data from prior seasons. 2006 has been an offense-heavy year.

[RobR]

Don't think so. That would mean the expected scoring contribution of H/9, BB/9, HR/9, and/or K/9 have changed significantly. Anyway, it's not too useful as a predictor if the coefficients are that sensitive to 3/4 of a year's data.

[ajc730]

Yep. The expected contributions of each stat are definitely going to be senisitve to changes in the level of offense. I'm pretty sure this is what explains the fact that 49 out of 54 pitchers with 100+ IP have a lower YTD xERA than YTD ERA.

In my opinion, xERA, as published, is still very useful though re-running the simulation halfway through the year would enhance its precision. If you're worried about accounting for the level of offense, you can calculate a "quick and dirty" xERA by running a regression of ERA on k/9, bb/9 and gb% with a dummy for league. League offense levels equal, it won't be quite as accurate as Burnson's xERA which is the gold standard but it will account for the run-scoring environment in the league in 2006.

Here's a quick formula to try:

xERA ~ 5.83 - .296*k/9 + .352*bb/9 - .0142*gb% + .150*league

where league = 1 if AL, 0 if NL.
gb% expressed as a number between 0 and 100; e.g. 45 not 0.45.