Inventing new pitching stats: Runs per inning and runs per appearance

Guest Post AvatarAndrew Perpetua, Fan Post:

Last year I decided to keep track of all the Citi Field home runs. I made a Tumblr that was featured on MetsBlog a few times over the course of last season. This season, I decided to make a spreadsheet to keep track of all of the pitching stats.

I’m treating every ball put into play equally. An error is a ground ball, fly ball, etc., a sac bunt is a ground ball, a sac fly is a fly ball. A lot of the other sites, from what I understand, do not necessarily treat numbers this way. Instead they have adopted scorecard rules such as plate appearances, sacrifices, and things like that.

It would be boring to run down each of them individually, so I’m going to talk about the two that you probably haven’t considered, and why I like them…

RPI, or runs per inning (which would probably better be named ERIP — ERIP sounds strange and ERPI looks weird so I prefer RPI): RPI is, quite simply, earned runs divided by innings pitched. There are zero pitchers who pitch nine innings on a daily basis anymore, and haven’t been for decades. I see no reason — beyond momentum, I guess — for talking about ERA. What is so different with RPI? Well, a) it isn’t multiplied by an arbitrary constant, and b) when combined with RPA — which I’ll get to in a moment — it gives you more information about a pitcher.

Jon Niese 1 polaroidNext, RPA: Runs per appearance. This also would probably better be named ERPA. Or maybe ERA (cough). However, RPA is what I have always called it (if you know what an ‘official’ name is, please tell me). RPA is earned runs divided by appearance. The stat does well when a pitcher is either a strict reliever or a strict starter, and does really poorly when a pitcher yo-yos between the two roles. Average, Median, and Total RPA give you a rough idea, when used together, how many runs you expect a pitcher to give up in any given appearance. Average and Median RPA are calculated using the RPA of each start whereas Total RPA is found by dividing total ER by total appearances.

For instance, our friend Jon Niese has an average RPA of 2.8, Median RPA of 1.5, and a total RPA of 2.9. From this you can gather that he doesn’t tend to pitch a consistent number of innings per game, and he has had a number of starts where he has been very bad, and roughly an equal number where he has been very good.

Now, comparing this to his RPI of .53, you can see he tends to pitch about 5.5 innings per start because:

RPA / RPI = (ER / Appearance) / (ER / IP) = (ER * IP) / (ER * Appearance) = IP / Appearance.

This is why I like these two stats. You can figure out roughly how well he has pitched, roughly how consistent he has been, and roughly how many innings per start he has pitched. Pretty good stuff, if you ask me.

So I hope I sold you at least a little bit on the value of those two stats. I don’t think sites like Fangraphs or Baseball-Reference keep them (or, at least I don’t know what those sites call them). Everyone else seems to prefer ERA, which quite frankly doesn’t give you the same amount of information, at least not on its own.


Andrew is a consistent commenter on MetsBlog. This post was born out of a comment he made on a previous post, and he was asked to contribute more in a full post this weekend.