Pitcher Win Values Explained: Part Seven
In talking about how we calculate pitcher win values, we’ve covered FIP, differences in replacement level for each league and role, run environments, the dynamic runs-to-wins conversion, and park factors. What we haven’t done is walked through an example, from scratch, of how the pitcher win values are calculated. That’s what we’re going to do today.
We’ll use Felix Hernandez as our guinea pig. In 2008, he threw 200 2/3 innings with a 3.80 FIP as a starting pitcher in the American League. Remember, we noted earlier that the league average runs per game in the AL was 4.78 last year, so we rescaled Felix’s FIP to make 4.78 equal league average. Adding in a park adjustment for a half season in Safeco Field (with a park factor of .96), we get a 4.28 neutral park FIP scaled to RA for Felix’s 2008 season.
Now, we have to figure out the runs to wins conversion based on Felix affecting the run environment he pitches in. To do so, find his innings pitched per start (6.5), subtract that from 18, and multiply that by the league average runs per game. Then, we add to that those 6.5 innings multiplied by his park adjusted FIP, and divide that by 18, and then use Tango’s +2*1.5 runs to wins converter. So, the formula for Felix would be ((11.5 * 4.78 + 6.5 * 4.28)/18)+2)*1.5, which would give us a run environment of 9.90 runs per win. So, for every 9.9 runs he saves, he gets credit for one win.
His 4.28 FIP is 0.50 runs per nine innings better than league average. What does Felix’s 4.28 FIP translate into in terms of win%? 0.50 divided by 9.90 equals .050. Add that to .500 and we get .550, making Felix a .550 win% pitcher. Remember, an average pitcher would post a .500 win%, and a replacement level starting pitcher would post a .380 win%. So now we subtract .380 from .550, and we get Felix as .17 wins better than a replacement level starter every nine innings.
Factoring in his actual innings pitched, we get .170 * 200.67 / 9, which comes out to 3.8 wins. That’s his wins above a replacement level starting pitcher, or what we call his win value for 2008. Remember, though, these are context neutral win values. Actual wins contributed to a team’s ledger will also be affected by how each pitcher performed with runners on base, as well as the performance of the defenders behind the pitcher. There are going to be cases where a pitcher has a much better (or worse) context neutral win value than you might expect if you’re used to looking at his W-L record or his ERA.
That does not mean these win values are “wrong”. We’ve removed the situational context of the pitcher’s performance, just as we do for hitters. Pitchers can either underperform or outperform their win values with extreme performances in “clutch” situations. We can measure the differences in these situational performances by looking at a pitcher’s WPA or WPA/LI and comparing it to his Win Value. For too long, we’ve lacked a resource for context neutral win values for pitchers, having to settle for situational win values that include a lot of variables. These pitcher win values offer us a great opportunity to explore more of what is in a pitcher’s control and what is not.
Dave is the Managing Editor of FanGraphs.
I remember reading in an earlier piece that after a certain number of innings (150 maybe?) a starting pitcher needed to be compared to bullpen numbers because a replacement level pitcher wouldn’t go deep into games.
The earlier method was essentially a shortcut. It’s not a horrible stand-in (usually within half a win or so), but this one’s better.
What new process accounts for this? Is it the run environment?
The more accurate replacement level.
The thing that confuses me about win values for pitchers is using FIP to determine their value. Theoretically, if a pitcher is untouchable all year, but gives up 100 runs in three games (bear with me), would his year end FIP make any sense to use as a win value?
There’s no evidence that distribution of run prevention is anything but random, so there won’t be any systematic bias in using an overall average.
I’ve been thinking about how you just use a single dollar value to calculate what a pitchers (or hitters for that matter) wins are worth. It seems to me like these wins could be scaled differently. ex.) one win over replacement (Kenny Rogers 4mil) shouldn’t have the same value as Lincecums 5th and 6th win over replacement value. Elite players are a more scarce commodity and the extra wins they accumulate should be worth more, no? My thinking might be flawed but this was something running through my little brain.
We’ve had this discussion quite a bit lately. Here are some good places to start reading.
http://www.insidethebook.com/ee/index.php/site/comments/war_to_dollar_valuation_but_roster_space_management/
http://www.baseballthinkfactory.org/files/newsstand/discussion/si_red_sox_made_play_for_ss_ramirez/P200/
http://www.insidethebook.com/ee/index.php/site/comments/supervorp/
Thanks
I am having difficulty with the pitcher win% part… Does your .550 win% imply that felix *should* win 55% of the time against an average team assuming average offense and defense on both sides?
i have some lame questions about re-engineering these values…I assume scaling FIP to RA is the same as scaling to ERA, in terms of leagueRA-leagueFIP, is that right? Also, I keep getting different RA values…is this figured like ERA, in terms of League R/Inn * 9? Thanks again for making all of this mind-blowingly analysis FREE!!!!
One More Day Til the Big Day…..I’m mad I’m a guinea pig though lol
Jake the guinea pig was apparently under the impression that I’d abandoned him forever. He is clinging and crying.
One More Day Til the Big Day…..I’m mad I’m a guinea pig though lol
You guys aren’t the guinea pig, he said Felix Hernandez was the guinea pig. What’s the big idea here, fellas?
I’m more of a noob, and appreciated the nice explanations and wonderful work here. But did have a couple questions/comments.
Please pardon their noobeness:
1) For WAR for Pitchers (WARp), why no impact for defense? While P is probably the least impactful defensive spot (LF/RF, 1B), it certainly can have an impact, especially on batted balls up the middle. And it IS another factor a pitcher can control. You include defense for WAR position players (it isn’t just about batting), why not pitchers?
Is there a separate stat for defensive impact for all positions one can look at to see a pitchers defensive impact translated to wins?
2) I’ve noticed at BBREF.com that WAR for position players goes back all the way to 1871, but pitcher data starts in n1876. Why? This seems to make is hard to measure some HOF level pichers (Mathews, Spalding, Cummings, Bond) against others in this metric with these 5 years missing. Are there plans to do the final 5 years? There don’t seem to be any stats missing or unavailable that would account for not calculating it?
3) On the whole FIP thing, I must say, as a noob, I’m unconvinced that HR/BB/K’s are the best measures of a pitchers effectiveness. I don’t think BABIP is random or equal, and while randomness can certainly play a factor in over a short time frame, over a long one (162 games/multiple years) I feel the type of contact a pitcher allows, etc is an essential tool in evaluating his effectiveness. Not saying there is a better way, just saying that any metric that takes that out of the equation is limited (like any single stat will be).
Love the work and keep it up.
What accounts for Kershaw’s WAR being so much better
than Matt Cain’s when their stats are very similar. Cain may
have the slight edge in most stats, and they both pitch in
pitchers ballparks. Cain and Kershaw have the same ERA.
Same innings, Kershaw has 4 more K’s Cain has given up
less hits and walks and has a better WHIP. Not sure of the
flyball/groundball ratio. How is the calculation for Kershaw’s
WAR at 3.1, and Cain’s at only 2.2. To complicate matters,
Vogelsong’s WAR is also 3.1. How have these numbers been
specifically calculated?
Hi, I’m trying to work through this example, and I don’t understand these two points:
“Remember, we noted earlier that the league average runs per game in the AL was 4.78 last year, so we rescaled Felix’s FIP to make 4.78 equal league average. Adding in a park adjustment for a half season in Safeco Field (with a park factor of .96), we get a 4.28 neutral park FIP scaled to RA for Felix’s 2008 season.”
What exactly do you do you Felix’s 3.80 FIP to “rescale” it? And then what do we do with the park adjustment?
My current best guess is that the process is: 3.80 FIP * _____ (rescaling?) = 4.458 (intermediate step) * 0.96 (park factor) = 4.28 park neutral FIP.
Thanks in advance!
Chandler, remember that we calculate the FIP constant by subtracting LeagueFIP (-constant) from LeagueERA, so I’m pretty sure that what’s happening here (I’m trying to back this out myself too), is that we’re just adjusting the components of the FIP constant to account for league differences.
I agree with Chandler – this is not complete example because all of the constants are not given to derive the end numbers.
Mostly, just how the park neutral FIP number is achieved.
It is not shown how you “rescale” to league average or how you end up with you park neutral FIP number of 4.28 from the original 3.8 FIP.
A resource to park factors would be helpful as well.
However, most everything else is straightforward. Thanks
Park neutral FIP number scaled to RA that is (excuse me)