Pitcher Volatility Part I
For much of this year I have been publishing my ongoing research into player volatility. Those of you that have been following that work will remember that Volatility is defined as the degree to which a player will vary on a given day from their seasonal average. For hitters, seasonal Volatility was measured in terms of Weighted On-base Average.
We found that, generally speaking, while hitters vary to the extent they are volatile in a given season there exists minimal correlation between Volatility in year one and Volatility in year two. That's not to say that some hitters don't exhibit higher or lower levels of Volatility than others over multiple seasons, just that the relationship is prone to wide fluctuation year over year (similar to, say, BABIP).
The next logical question was, of course, what about pitchers? Do they exhibit volatility in the same way as hitters do? If so, was there a better or worse relationship year to year?
To answer head questions I used a similar methodology, but instead chose Fielding Independent Pitching instead of wOBA. I also restricted the analysis only to starting pitchers since 2005, and to those that accumulated at least 20 starts in both the first and second season (n=484).
Here are the initial findings:
The year-to-year correlation between Pitcher Volatility in year one to year two is a mere .04. No, not .40, but .04. That is to say, there is basically no correlation year-to-year. I've looked at the data a few times and I can't find any obvious errors that might be artificially creating the low correlation.
This means that, as an indicator of skill or skill change year-to-year, we simply cannot read much into the statistic.
That being said, there is clearly variation by pitchers over multiple seasons.
If we take the combined Volatility of starting pitchers with >= 80 starts since 2005 a clear hierarchy emerges:
Over multiple seasons we see that certain pitchers exhibit less average variation from start to start than others. The degree of variation among pitchers is also quite large. Not only was Greg Maddux a phenomenal pitcher, but he was the most consistent over this data set. Roy Halladay also shows up as one of the most consistent, dominant starters over the past 6+ seasons.
I should note again that Volatility is not the same as streakiness. Volatility simply shows the degree to which pitchers are consistent, start to start, not whether they tend to string together a series of good starts followed by a string of bad starts.
Additionally, being consistent doesn't necessarily mean you are good. Mike Pelfrey has been one of the most consistent pitchers over the past 5+ years, but all this means is that Met fans (and opponents) can count on getting relatively the same FIP start after start. If he was a low 3-FIP guy that would be tremendous. As it turns out, he's not.
Another interesting note: starting pitcher Volatility has declined in-line with other pitching metrics. There is a .31 correlation between seasonal FIP and Volatility, but there is still a lot left to be accounted for. However, as FIP has decreased over time so to as Volatility.
The average Volatility for starters with >= 20 starts since 2005 is as follows:
2005 - 2.03
2006 - 2.03
2007 - 2.00
2008 - 2.02
2009 - 1.93
2010 - 1.85
2011 - 1.72
This lines up pretty well with what I found when looking for the reason for the decreasing run environment. If that analysis was right, part of the reason is the influx of really good starting pitching beginning in 2010.
As with hitters, I think this is a useful metric to look at when thinking about roster construction. Ideally, you want your best pitchers to be highly consistent, start to start. As David Gasko has suggested, however, you may actually want your back-end starts to have a little more Volatility.
Certainly, when evaluating who was the "better" pitcher I think Volatility could provide an interesting perspective.
I'll tackle that topic next in terms of starters in 2011 and the Cy Young award.