Has Brandon Lyon’s long awaited regression finally started?

In the preseason I was touting a quick collapse for Brandon Lyon, and to my chagrin I even made a prediction that he would be one of the first two closers to lose their job, along with all-star Brian Wilson. Whoops.  These results go a long way to explaining why we always focus here on the process of decision making and not the actual result.

When we predict a regression to the mean, what we are saying is that, essentially, water seeks it own level and that level can be divined by analysis of a pitcher’s skill set. In Lyon’s case, coming into the season he had a 4.94 xERA last year compared to a 2.68 actual ERA, a K rate below five per nine innings and a K/BB ratio below 2.0.  These are the hallmarks not of a top closer but of a mop-up man. He has never had an expected ERA below 3.88 in any season and this year so far has been no different, with an expected ERA of 4.55.

Sometimes these regressions take a long time; more than one season in many cases. This is doubly true with relief pitchers, since their in-season sample sizes are small. Even a 60 inning season is still a small sample size. Water will seek its own level but that could take 100 innings or more. With relief pitchers we really are in the realm of the uncertain.

This is partly why relievers have such short shelf lives, aside from the creme de la creme like Mariano Rivera. They win and lose jobs based not on skill but on fluctuation. Most relievers have similar skill sets and can burn hot or cold over the course of more than one season.  Like the blackjack dealer who deals an inordinate number of blackjacks, a reliever can shine based on nothing more than the chance fluctuations of statistics.

Lyon may be a good example of this right now. It has taken over a year, which seems like a long time, but in reality he has only 110 IP or so since the start of 2007. He was very fortunate in 2007; he had one of the highest differentials between ERA and xERA in baseball last year. This year gave us more of the same, and all year his ERA has been below his xERA. It still is right now, but the tracks are getting closer as he has a 4.55 xERA and a 3.86 ERA.

It has been his last two outings that have caused his ERA to spike, with 7 ER and 8 H in his last 1.2 innings. His earlier strikeout rate increase now seems a thing of the past, with only 2 strikeouts in his last 7 appearances. He had expected ERAs of 5.06 in June and 6.31 in July so far. Even his superficially excellent May was a mirage; his 0.00 ERA was the result of a 100% strand rate.

There are no guarantees and this analysis may or may not end up as accurate, depending on when you decide to examine his record. A player’s performance vacillates up and down over his expected level of performance, and eventually the actual and expected performance meet somewhere down the road. When that is impossible to determine, but all we can do is make our best decision in the short term, focusing on what we look and and how we process the information. All we see is a sliver of that vacillation, and from that sliver we try to divine the future and analyze the past. After that, the dice are cast and out of our control.

Right now it looks as if his regression may finally be in the process of starting. And it’s about damn time.