So far in this series, we've tackled luck, wOBA, and WAR, but now it's time we got into something a little more difficult - pitching. Before we get into the nitty-gritty of how to analyze pitchers, here's something I think all of us can agree on: evaluating pitchers is damn hard. Pitchers are some of the most fickle players in all of baseball, and their performances can vary wildly from year to year. A pitcher may be mediocre one year, but then bust out and become one of the best in all of baseball the next year (ahem, Cliff Lee, Randy Johnson). Some may burst onto the scene as perennial All-Stars, but flame out due to injuries and general ineffectiveness (cough, Scott Kazmir). Some may have one star season, but then struggle to ever recapture that glory and effectiveness (Dontrelle Willis and Oliver Perez, anyone?). Unlike position players, which have been shown to - on average - peak around age 27 and slowly decline afterwards, there's no set "aging curve" for pitchers. Some peak at 22, others peak at 32. It's friggin' insane.
And so, considering that we all know how volatile and unpredictable pitchers are, when evaluating a pitcher we want to find statistics that predict the future the best. How well should this pitcher perform going forward? That's the question we want answered and although no statistic - or even clump of statistics - is going to be even 80% accurate, we'll take whatever we can get.
In the late '90s, this guy called Vorus McCracken came up with a radical new idea: let's evaluate pitchers only on things they have direct control over. Instead of focusing on Wins - which depends upon the pitcher's team to score runs - and ERA - which depends upon the quality of the defense behind the pitcher - let's take a look at the outcomes that only involve the pitcher and batter: strikeouts, walks, and homeruns. How well do these three things predict future success for a pitcher? Surprisingly to all, including himself, they worked quite well and launched a new branch of baseball analysis: defense independent pitching (DIPs)*.
*Brief interlude: for those that are unfamiliar with the concept, you may be asking yourself, "Why do we need to separate pitchers from their fielders? Doesn't ERA account for fielding because it only counts earned runs?" Yes and no. ERA accounts for some fielding, disregarding runs that are the result of an Error in the field, but how many times have you seen Carl Crawford make a fantastic catch out in leftfield that turned a double into an out? If the Rays had any other outfielder out there in his place, those balls would fall in for hits and the Rays' starters would most likely allow more runs to score. Like umpires, defense is invisible to the casual fan unless its really, really good or really, really bad, but it can have a large impact on a pitcher's results.
These days, there are a number of DIP statistics, all of which are attempting to do the same thing: predict to the highest degree of success how a pitcher will perform in the future. Since there's no one perfect statistic out there, let's run through three of the most common ones: FIP, xFIP, and tRA.
The golden standard for DIPs statistics, FIP uses McCracken's three variables - stikeouts, walks, and homeruns - to calculate what a player's ERA should have been over a given time period. It's scaled to look exactly like ERA, so it's easy to tell what's a good value and what isn't, but it's a much better predictor of future success than ERA. Does that mean it's perfect? No, but it's a step in the right direction and one of the best DIPs stats available at the moment.
Whenever you see a tiny "x" before a statistic, that means the stat has been regressed to some degree. Since McCracken came up with his radical new idea, it's been shown that pitcher homerun rates are unstable as well. A pitcher may let up homeruns on 5% of his flyballs one year, but then let up homeruns on 15% of his flyballs the next year. There's no rhyme or reason to it, and all pitchers have their homerun rates fluctuate regardless of if they're high strikeout pitcher, induce lots of groundballs, or are one of the best in the league. For example, Roy Halladay let up homeruns on 10.6% of his flyballs last season, but is at only 8.9% this season; James Shields let up homeruns on 9.8% of his flyballs in 2008, but is at 14.3% this season. Since this statistic is volatile, xFIP says, "Screw this, we want stability!" and regresses every pitcher's homerun rate to league average (10.6%). This has been shown to have slightly more predictive value than FIP by itself.
Oh boy, this is where things get interesting. tERA incorporates the same statistics as FIP - strikeouts, walks, and homeruns - but it goes a step further by also including batted ball statistics like line drive rate (LD%), flyball rate (FB%), and groundball rate (GB%). The theory behind it is simple: if a pitcher lets up lots of flyballs and line drives, they'd also be more likely to let up lots of homeruns and doubles, and that's bad. But if a pitcher makes batters hit groundballs all the time, they'll be more likely to get outs and be effective. Pitchers do have some control over their groundball and flyball rates, so we should include those in our calculations as well. tERA is also based on an ERA scale, making it easy to tell what's a good score and what's not.
Those are all very simplistic descriptions of the above statistics, so if you have any additional comments or questions, please let us know in the comments. Also feel free to check out the pages on FIP, xFIP, and tERA on The Sabermetrics Library.