Part one on the debate between traditional statistics and sabermetrics.
If any of you noticed,* Parker Hageman of Over The Baggy wrote a post about FIP** and why we should care about it on Nov. 18th. This led to an argument** over the usefulness of ERA vs. FIP between TT of Granny Baseball***, Parker, and David of Twins Fix. I entered the battle of vegetating baseball nerds in the defense of young, righteous FIP by posting this in the comment section:
* It only took me half a week after he posted this for me to actually read it, despite knowing about it the day it was originally posted.
** These are the two most crucial links that you should click on.
*** If you’ve never visited Granny Baseball, its main premise is to defend the older statistics in baseball prior to sabermetrics.
Do I have enough asterisks here? Maybe this is a preview to what the MLB record book will look like in 30 years.
The context of my reply is that I’m talking directly to TT.
Oh, the college math major in me is taking over. Here’s a problem if you rely entirely on ERA with extreme circumstances.
By the way, search FIP on Wikipedia to find the FIP equation that I use for this. Don’t worry if you’re sent to a DIPS page, just scroll down.
Let a player X be a perfectly average pitcher for a team with an infinitely spacious outfield and a perfect defense. To make it easy, I’m going to say that the average pitcher* threw 200 innings, allowed 20 HR, 133 K (2/3 of 200 IP) 40 BB, and 10 HBP. Using The Hardball Times version of FIP, this means our average pitcher X had a FIP of 4.42. The AL average for ERA was 4.46, so this appears that my half-assed number picking was actually quite accurate, since FIP and ERA are supposed to normalize and be similar over time.
* Edit: Add “in average conditions” here.
For this perfect team, you may think that X will have a perfect 0.00 ERA for the season, but this is not the case. Due to walks and hit batters, X will certainly have a miniscule ERA, but it is not guaranteed to be 0.00 (albeit rare, the chance of walking/HBP-ing enough times to allow a run to score). We can’t calculate X’s ERA, but we can assume it will be insanely low. As for FIP, this infinitely large ballpark and perfect defense eliminates all HR, so X’s new FIP is 2.62. Clearly, VERY good.
Now, let X play for a different team the next season, such that all variables are average. Now X will start giving up hits and home runs. This would cause his ERA to regress to the league average, and by adding HR back into the FIP equation, his FIP will return to 4.42, which very nearly agrees with the average ERA.
When X was with the perfect team, his ERA was likely near 0.00, but his FIP was 2.62. When X joined the average team, his ERA would clearly rise, and his FIP definitely rose as well. The argument for FIP is that it’s a better predictor for a pitcher’s ERA in the future than that same pitcher’s current ERA. In this extreme example, that was true. A 2.62 FIP was better indicative of an average ERA in the future (4.46) than the near-0.00 ERA.
You can argue that since this was an extreme example, it wasn’t similar to a real-life example. My response to that complaint is that this extreme example exaggerated the influence that defense and park factors have on ERA, but when normalized, showed that the extreme example FIP was closer to the average ERA than the extreme example ERA was.
Ugh, I wasted too much time on this.
Yes, I did waste too much time on that. Yes, I did.