FIP and DER (December 30, 2003)
There is no correlation between FIP and DER. None, nada, zip. The R squared between the two is .002. This is good, actually. It means that FIP and DER measure two completely different things.
This says that how a pitcher performs with BB,K,HR is completely unrelated to his hits per BIP.
So it represents the proportion of a pitcher's ERA for which he shares responsibility with his fielders.
Not technically accurate, but the general point is there.
Excellent article all-in-all.
--posted by TangoTiger at 11:07 PM EDT
Posted 2:02 a.m.,
December 31, 2003
(#1) -
Brad Wenban
From the article:
As a pitcher's FIP decreases, defense becomes less relevant to the overall outcome. Not negative, not positive. Just less relevant.
**********
So superior FIP pitchers are more valuable to a team with a worse defense and vice versa.
The moral from a team building point of view is that every increase in defensive quality ($H) makes an increase in pitching (FIP) quality relatively less valuable (since the run value of a BIP falls) but makes an increase in batting quality relatively more valuable (since Runs/Win falls). Similarly, every increase in pitching quality has the same impact, making an increase in defensive quality relatively less valuable but also making an increase in batting quality relatively more valuable.
Are these effects large enough to ever be functionally determinative in optimal player usage? For example, how would '03 Jeter's custom defensive LWts given his pitching staff differ from league average?
Posted 6:50 a.m.,
December 31, 2003
(#2) -
Guy
Nice piece. It seems that the numbers on the relative power of FIP and DER in predicting ERA have implications for the debate over the proper pitching/defense allocation. The R2 for FIP is .54 and for DER is .31. If we used RS instead of ERA as dependent variable, and replaced DER with BIP-Slg ((1B + 2x2B +3x3B/IP)) to better capture impact of balls in play, we should find that BIP accounts for more than 31% of the variation. Let's assume the split would be something like 55/45 FIP/BIP if we eliminate unexplained variance due to chance.
In "Solving DIPS," Tango et. al. argue that the division of responsibility for DER/BIP is about 62/38 pitchers/fielders, leaving aside park effect and random chance. I find this much more believable than the original Voros thesis -- which Studes seems to accept (sorry if I'm misinterpreting, Studes) -- that DER is almost entirely driven by fielding.
Combining these, it would suggest a pitcher/fielder split of around 83/17, giving fielders far less influence (about 9% overall) than they are assigned in WS. Thoughts?
Posted 7:37 a.m.,
December 31, 2003
(#3) -
studes
(homepage)
Thanks for posting the article, Tango. How is the DERA definition not technically accurate?
Guy, every time I've looked at it, I've come to the conclusion that responsibility for DER is around 50/50, maybe as high as 60/40 pitching.
This isn't inconsistent with Voros' theory, by the way. Saying that pitchers don't have a lot of control over BIP is different than assigning responsibility for BIP. IMO, if fielders can't be expected to reach a batted ball, than the pitcher is responsible for the hit.
BTW, I was going to compute XBHDER, using your formula, but that data isn't available historically so I backed off. I should probably conduct the analysis, but with a more recent group of pitchers.
I was thinking that the next step is to construct a pitching/fielding split for specific pitchers and teams, in which the pitcher credit/responsibility increases proportionally as FIP decreases. Not sure what the exact formula might be.
Posted 9:45 a.m.,
December 31, 2003
(#4) -
tangotiger
Studes, it's not a "proportion". FIP does not share the same scale as the other metrics.
Posted 10:33 a.m.,
December 31, 2003
(#5) -
Joel Wertheimer
I emailed Tango about this recently because I found that with Derek Lowe, one can notice that his BB rate is correlated with his balls in play rate.
I wonder if one control's for a subset of pitchers, groundball pitchers for example, the finding might be different. Simply because there is not a high R^2 across all pitchers does not mean that a certain group of pitchers.
Studes, the regression I ran in the Derek Lowe study was BBrate = constant + DERX + DER*krateX + u. If you still have that database, that might be something to play around with.
Posted 10:42 a.m.,
December 31, 2003
(#6) -
studes
(homepage)
I don't mean to bog this down, but why doesn't FIP have the same scale as ERA? I thought that was the point of it.
Posted 11:42 a.m.,
December 31, 2003
(#7) -
tangotiger
To get technical about it, studes said:
As a next step, I computed an intermediate stat called DERA (a combination of DER and ERA -- get it?). It equals ERA minus FIP. So it represents the proportion of a pitcher's ERA for which he shares responsibility with his fielders.
dipsERA = FIP + 3.2
DERA = ERA - FIP
DERA = ERA - dipsERA + 3.2
when the pitcher and fielders are both league average at hits / BIP, ERA = dipsERA
so,
DERA = 3.2
So, there's no real "proportion" being allocated between FIP and ERA.
A proportion would be: "50% of his salary is paid by the Yanks" or some such.
There is a relationship in what studes is doing, but it is not a "proportion". You can't have FIP divided by ERA and have it tell you anything. FIP is just a number. To give it meaning, you have to add 3.2 to it.
(FIP + 3.2)/ERA will tell you something useful, though.
Posted 1:42 p.m.,
December 31, 2003
(#8) -
studes
(homepage)
Thanks, Tango. I understand. However, given that you basically use linear weights for your weights in FIP, I think it comes pretty close to a proportion. Average DERA across all pitchers was 3.00, BTW.
I struggled with this article a lot, and I don't think I nailed it. The conclusion bugs me. In theory, shouldn't DERA decrease as FIP decreases? Less BIP, less impact of a hit. Instead, DERA increases (and DER stays flat) as FIP decreases. I think I'm not doing this quite right.
My guess is that I'm using FIP in a way that wasn't intended, so I'm probably misinterpeting or misapplying something. In particular, it would seem to me that applying league-average weights to FIP for extraordinary pitchers is throwing off the conclusion. What do you think?
Posted 3:43 p.m.,
December 31, 2003
(#9) -
Tangotiger
However, given that you basically use linear weights for your weights in FIP, I think it comes pretty close to a proportion
What is Pedro's "proportion", what with his negative FIP? My only problem is with you using the word proportion.
In particular, it would seem to me that applying league-average weights to FIP for extraordinary pitchers is throwing off the conclusion.
Correct. FIP is a shortcut to DIPS, and a damn good one. You lose at the extremes,as you would with any shortcut.
What you REALLY want to do is apply BaseRuns. You plug in the known BB,K,HR, and use league average for hits,2b,3b,nonKouts. That'll give you the dipsRA. Use the player's real numbers for hits,2b,3b,nonKouts to get the RA.
THEN, you can perform your analysis.
Posted 3:49 p.m.,
December 31, 2003
(#10) -
studes
(homepage)
Thanks, Tango. Agreed. I might try and play with this to see if I can develop a proxy for the weights (don't want to re-run your spreadsheet for every pitcher). Maybe I'll use the table you posted a week or two ago, in some way.
Posted 4:48 p.m.,
December 31, 2003
(#11) -
Charles Saeger(e-mail)
One thing I feel the need to add to all this is that the non-FIP portion of RA/9 has changed (diminished, IIRC) over baseball history. (I ran FIP for postwar leagues awhile ago and found this.) No idea what to do about this, but I thought it worth saying.
Posted 1:44 a.m.,
January 1, 2004
(#12) -
Guy
" In theory, shouldn't DERA decrease as FIP decreases? Less BIP, less impact of a hit. Instead, DERA increases (and DER stays flat) as FIP decreases. I think I'm not doing this quite right."
Yes, a low FIP should tend to produce a lower DERA as more Ks = fewer BIP (but shouldn't affect DER in the same way). However, keep in mind that DERA is really a composite of two factors: pitcher-DERA and fielder-DERA. And there are other possible relationships at work. In terms of basic pitcher ability, low FIP is probably associated at least weakly with low pitcher DERA -- it's hard to believe the skills are wholly unrelated. In addition, you'd expect that financial disparities among teams could produce some association between low FIP (good pitching) and low fielders DERA (good fielding).
Against all of those tendencies is one major factor that cuts the other way: if you are a high-FIP pitcher, you must have a reasonably good pitcher-DERA or you won't pitch for long in the major leagues (i.e. if you can't strike batters out, you'd better be able to get them out on BIP). This selection process should contribute a negative correlation between FIP and DERA. (It would be interesting to see if FIP and DER are somewhat correlated in the minors, but not in the majors). Apparently, this last factor slightly outweighs the impact of the others (and maybe I'm missing others). The zero correlation between FIP and DER may just be an interesting coincidence.
Posted 10:05 a.m.,
January 1, 2004
(#13) -
studes
(homepage)
Agree about the high-FIP pitchers. But even looking at pitchers with FIP of 1.5 and lower, DERA actually rises as FIP declines.
Posted 11:07 a.m.,
January 1, 2004
(#14) -
Tangotiger
Studes,
Until you rerun using BaseRuns, what you are finding may simply be a byproduct of the shortcut.
hits per ball in play is fairly static across all quality of pitchers, quality as measured by BB,K,HR.
So, what you are left with is how much run impact a hit per BIP has. And, we know with custom LWTS that the run value of a hit decreases with the run environment. So, I would expect a slight slope downwards with a real DERA and FIP, as the FIP decreases.
Posted 2:19 p.m.,
January 1, 2004
(#15) -
studes
(homepage)
Yes, I agree, Tango. I need to figure out some way to do that as a next step. I'd like to refine this analysis, and then use it to replace the current Win Shares methodology for splitting runs allowed between pitching and fielding.
Posted 7:55 p.m.,
January 1, 2004
(#16) -
Charles Saeger(e-mail)
Yes, I agree, Tango. I need to figure out some way to do that as a next step. I'd like to refine this analysis, and then use it to replace the current Win Shares methodology for splitting runs allowed between pitching and fielding.
Studes: the stuff I sent you, aside from the strikeout adjustment (which is highly malleable, BTW), is based on run values for the pitching/fielding split. I've done a TON of work here.
Posted 7:57 a.m.,
January 2, 2004
(#17) -
studes
(homepage)
Thanks, Charlie. I'll send you an e-mail over the weekend about this, but I'm trying to develop an approach that would adjust the split, based on the underlying run environment established by the pitcher.
Your split formula (and James's) has an implicit assumption about a "baseline" split, if you know what I mean. I'm trying to see if there's a way to establish a "baseline" split based on the fundamental team and pitcher environment.
If I'm able to somehow make it work (a big if), then I hope to add your adjustments, including ADER, to refine the splits.
Posted 10:26 a.m.,
January 2, 2004
(#18) -
Charles Saeger(e-mail)
Yeah, the strikeout adjustment. If you use a run-value based strikeout adjustment, you'll have tons of folks astonished at the differences in defensive Win Shares between one team and the next. The primary reason I wrote it like I did was that no one would accept it otherwise.
In relation to the other elements of the formula, use a run-value of 3/8 for a strikeout if you're going that route. It's the value of the actual strikeout, the value of not giving up a hit and the value of keeping another batter from coming to the plate.
Posted 10:54 a.m.,
January 2, 2004
(#19) -
tangotiger
Pitcher A: 40 PAs, 1 HR, 3BB, 6 K, 21 non-K outs, 5 RPG
Pitcher B: 40 PAs, 1 HR, 3BB, 10 K, 17 non-K outs, 5 RPG
So, pitcher A has 30 BIP where the fielder might be able to do something with, while pitcher B has 26 BIP. (Let's assume that the hits/bip talent rates are different for the 2 pitchers, such that overall, they are the same at 5RPG).
So, if fielders contributed "50 win shares" for a season of pitcher A, he'd contribute 43 win shares for a season of pitcher B.
Charlie, are you saying there's more to it than that? Are you talking about the cases where the hits/BIP talent rates of the pitchers are the same?
Posted 12:15 p.m.,
January 2, 2004
(#20) -
Guy
Why should fielders get more WS under either scenario? If your premise is that pitcher A compensates for his fewer Ks with more outs on BIP due to his ability, shouldn't he receive those WS?
Posted 12:34 p.m.,
January 2, 2004
(#21) -
tangotiger
I didn't say he compensated for it completely.
In any case, the fielders have the same effect on each BIP, more or less. So, I take that as a given. (I know it's different between a GB/FB pitcher, etc.)
Whatever is left over to keep the RGP equals, assume that the different talent levels of the pitchers on BIP is responsible.
It's like the park effect. Say both pitchers are GB pitchers, and they play on turf. But, if one pitcher gives up lots more GB because he strikes out alot less, the "turf effect", per GB, is still the same, but over the whole game, it'll affect the lowK guy more.
Posted 1:01 p.m.,
January 2, 2004
(#22) -
studes
(homepage)
I'm sorry that I'm not following your example, Tango, and I shouldn't be answering a question you posted for Charlie.
But from my point of view, the issue is whether you give a DIPs/FIP weight to K's, or a more typical out-weight to K's, in the Win Shares framework. My interpretation may be off, but I think that James gives them more of a typical out weight, and I think the DIPs/FIP weight is the better approach. I think Charlie agrees with this, too.
Posted 2:42 p.m.,
January 3, 2004
(#23) -
Charles Saeger(e-mail)
What studes said, though there's more numbers floating about here than anyone else here has seen. This is actually why I started with the expected Win Shares approach, since in terms of actual wins being parceled out, I couldn't see how fielders on high-strikeout teams would not earn less than fielders on low-strikeout teams, since technically the pitcher did more of the work and more to earn the win.
This isn't an issue of ability, it's an issue of the system, and that is what I was addressing.
Posted 10:25 p.m.,
January 4, 2004
(#24) -
Guy
Let's say you have a good, well-balanced .600 team, with above-average hitting, pitching, and fielding. Where should you put resources to improve? Studes' article points out that such a team will get diminishing returns on defense: with good pitching, the impact of good fielding is reduced. The reverse must also be true -- adding an ace pitcher will reduce RA a bit less on a good fielding team than on an avg fielding team. Offense is just the opposite: adding a good hitter to a high OBP/high SLG team will increase RS more than for an average team. Since salaries are presumably based on players' impact on an average team, this team should get more bang for their buck by upgrading their hitting.
But, on the other side, you have Tango's point that the win multiplier is higher for marginal reductions in RA than for marginal increases in RS. That would argue for a pitching upgrade strategy. So, which factor outweighs the other? Or maybe it's a wash, and either strategy would deliver an equal W/$ return?