Fielding Independent Batting, A ShH Revolution
We have discovered something marvelous! We only need to know four simple measures to predict what a hitter can do: Their walk rate, strikeout rate, home run rate and BABIP.
For weeks now, we’ve been exploring and playing with my initial discovery, and now we’re to the point where I feel comfortable calling it Fielding Independent Batting (FIB) — an homage to Fielding Independent Pitching (FIP), which predates it and partly inspired it.
I have created a final, consummate Google Doc, aptly titled Complete Should Hit (Advanced Predictor!) — as seen above. The document has several tabs, so let me explain:
- —> Fielding Independent Batting: The previous rendition of ShHAP! focused solely on BABIP fluctuations. FIB allows us to calibrate everything. Above, we see the present state of a one Ben Zobrist, titan among men. With FIB, we can choose to calibrate his present season according to his career rates (which I have done above).
If, say, I think his homer rate in 2011 is more accurate than his career rates, then I can adjust that accordingly. If I think he’s been unlucky on the walks this year and could be walking even more through the end of the year, then I can alter that as well. FIB allows us to give a quantitative prediction to changes we expect.
—> ShHAP! BIP: Want to see what a player should hit with just a normalized BABIP? Well, ShHAP! BIP is your man. Unlike ShHAP! 1.0, “BIP” only requires a present wRC+, a present BABIP, and a future or expected BABIP. The results are the same, but the process is simplified.
—> ShHAP! Walk: The same with BIP, “Walk” allows us to simply answer the question: “What difference does Ben Zobrist’s career and present BB% make on his wRC+?”
—> ShHAP! Punch: Punch helps us answer the question: “What if Zobrist was punching out (K%) 1.5% lower (his career rate)?”
—> ShHAP! Homer: With homer, it gets trickier. The formula itself calculates the HR% using home runs and plate appearances — two stats more commonly available than an already-computed HR%. So, as I have done above, we can just put the player’s present and career numbers in; but we can also say crazy things, like: “I believe the 2009 Ben Zobrist was the real one,” and then substitute those — much higher — homer numbers into the formula.
If you think his 2.9% homer rate is incorrect (and given his crazy-bad first two seasons, it may be) then moving it to an arbitrary 3.2% or something is a bit more difficult. Basically, you’ll need to reverse calculate it using maffematix n’such.
So, for 3.2%, you’d put 3.2 homers in the career section and then 100 plate appearances in the corresponding, neighboring career PA box.
Now go! Play with this new FIB toy and tell me what’s wrong, what can be improved, and what surprises you!
NOTE: In order to edit and play with the Google Doc, you will need to make a copy or download your own version from my original.
“Their walk rate, strikeout rate, home run rate and BABIP.”
…
…
Like, no offense, but…you understand that this is like…uh…already known?
Not necessarily. Cutting out triples and doubles, ignore HBPs, entirely excluding RBOE and line drive data — I don’t think that’s an intuitive approach to evaluating a hitter, but that’s what Should Hit does.
HBP shouldn’t be ignored since there’s some pretty strong evidence that it’s a skill (since some of the same guys regularly appear at the top of HBP leaderboards). It should, at least, be included with walks to some extent.
Chase Utley got hit by a pitch twice while JD was typing that comment.
I’m curious – Is there anyway we can get an AVG/OBP/SLG line out of FIB?
Well you can get a wOBA out of it, for what it’s worth. From there, you can use the walk rate to get OBP in some form (try a multiple regression with wOBA and walk rate to find OBP), and if you do a simple regression with wOBA and OPS (which I’ve done before), you can get the approximate OPS, from which you can subtract the OBP from before to get SLG.
OK, I’ll try that. Thanks for the help.
Yeah, I’m not sure if this was a “marvelous discovery” or not, but if nothing else, it’s a fantastic tool. Anyone can now use this to predict future performance if they feel that a player’s current performance is off in some way. Really cool.
I wonder if it is just a predictive tool, or if it could possibly be descriptive and used in WAR. If we can use this to regress a hitter’s BABIP to what it should be given how well he is playing, then that would seem to be a better option than assuming the batter has complete control over BABIP.
The question is, is a metric like xBABIP descriptive or predictive? Is it saying, this is what a player’s BABIP should be if we remove luck? Or is it saying, this is what we expect the player’s BABIP to be in the future based on his skills, but his previous BABIP was still legitimate?
Sorry to keep commenting on my comments, but on this note, it might be helpful to integrate xBABIP into FIB, so we don’t have to go to the xBABIP calculator and then put that result into FIB. Just cut out the middle man. That way we can just say, “Look, how well should he be hitting with X GB%, Y SBs, and Z HRs (or whatever the variables that go into xBABIP are)?”
Yeah, I think FIB could eventually go into WAR just like FIP goes into it, but first I think we need a universally-accepted and powerfully accurate xBABIP calculator. Right now, we have a handful of calculators, but they each having significant flaws.
Yeah, I’d love to throw an xBABIP in there somewhere — I just don’t think there’s one true and correct calculator out there.
@Bradley: I’ve been waiting for years for someone on Fangraphs to say this w/r/t using xBABIP in WAR.
Maybe the SABR community might have backed Miguel Cabrera over Josh Hamilton last year, for example, if that existed.
You can’t use xBABIP for WAR. WAR is meant to be descriptive.
Of course this is why I think FIP stinks for WAR as well, but Fangraphs insists FIP is descriptive. Sure, it describes events that happened, but only a small fraction of events that happened. FIP is like looking at a house flattened by a tornado and saying it has great storm windows.
@ Bradley
woah now you’re confusing me. My comment is the same as noseeum: isn’t WAR descriptive? Thus, isn’t wOBA the perfect batting metric for WAR? It perfectly describes what the batter did, and what run/win value of it was. FIB does not do this, if I understand all this correctly. It’s value is rather to be a simple stat that has better predictive power than others of it’s kind.
[attempts to stop head spinning]
“Thus, isn’t wOBA the perfect batting metric for WAR?”
Yes, except when a BABIP drives it crazy. A high BABIP (as in a lucky one, not an Ichiro one) isn’t usually at the fault of the batter — just like a shoddy defense isn’t at the fault of a pitcher.
And to say FIP and FIB aren’t descriptive is merely to re-define the description. FIP describes very well what the pitcher had control over. FIB does the same thing — assuming it has a perfect xBABIP calculator included. They don’t describe the run-scoring events perfectly, but they do describe the several of the most important, directly-controlled performance indicators.
@Bradley, it doesn’t matter if BABIP drives wOBA crazy, and it doesn’t matter what caused a high BABIP. The hitter still got the hits, and thus a descriptive stat like WAR should credit him for those hits.
I’d be the first to tell someone not to look at past WARs to predict future WARs. That’s not what they’re for.
Matt Joyce was clearly due for a correction later this year, and it happened. But he still did a heck of a lot to help his team win while he was hot. He didn’t get all those hits because the fielders he was playing against sucked.
So any FIB that throws what actually happened out entirely by going with xBABIP is not FIB.
Seriously, I never understand this. “Getting rid of noise” to help determine underlying skills is a very useful endeavor. But the “noise” is not irrelevant. It wins and loses baseball games.
Take away all the “noise” and next thing you’ll be arguing we should take the Giants World Series rings away because their combined FIBs and FIPS were 10th best in baseball last year! [No I don’t know if that’s actually true, but I’m damn sure they weren’t the best team in baseball skill-wise.]
I can’t imagine a statistic more specifically designed to suck all joy out of baseball.
cERA
The horror!
NERD?
net Earned Runs Deserved?
Republican!
You must pronounce homage in the “French-way” with a silent h, because you used the article ‘an’ rather than ‘a’. I prefer to go with “hom-age” rather than “o-mage”.
Also, what’s the deal with the P in OBP? We never express the number as a percentage, so why call it one? Why not “On-Base Average”? Then OBA and wOBA could be buddies.
Why do still call them “the stands” when everyone nowadays has seat?! Why?!
Why do flies fly… but elephants don’t elephant????
Time flies like an arrow. Fruit flies like a banana.
Awesomesauce.
Cool stuff Mr. Woodrum, cool stuff.
Though I do have what I consider a pretty major problem with the statistic. It requires 5 inputs, not 4. The statistic requires you to input a Present wRC+ which it then modifies based on many of the same components as regular wRC+ (HRs, BB) to create the adjusted wRC+.
Now, if you were to find a way to calculate someone’s wRC+ using only BB%, K%, HR%, and BABIP, this would be a lot more groundbreaking.
That being said, as a ‘lab’ for predicting future performance it appears to be pretty cool.
you need to input wRC+ so that it can tell what level of offence is league-average production. With only BB%, K%, HR%, and BABIP, the predictor would have no way of knowing what league-average rates for those should be.
If that’s true, it is fine.
The problem is that I’d bet the input wRC+ is actually covering for all the other stuff that isn’t covered in the spreadsheet such as stolen bases.
I’m having trouble with the idea of calling it fielding independent when one of inputs is BABIP.
A fair complaint. What makes it independent of fielding is the introduction of career BABIP rates. In a single season, a batter’s BABIP could easily be influenced by a string of dandy or terrible fielders.
By using a career rate (and assuming the hitter has a decent amount of PAs), we are assuming the batter has faced a normal distribution of defensive quality, thereby nullifying the impact of defense on the wRC+ and allowing only speed (and such BABIP “skills”) to remain.
With unbalanced schedules and players staying on teams for many seasons, you can not assume a player faces a normal distribution of fielders. If a guy plays for a terrible defensive team then he will face above average defenders for the season under most situations. If a guy plays for the best defensive team then he will hit against below average defenders more often.
I vote for a wOBA predictor as well, one more cell, just makes it more usable.
Three Questions:
1) Previous objections pointed out that FIB will underestimate the wRC+ of players who have low HR rates but collect doubles and triples at a high rate. How have you addressed that? I don’t think saying that no stat is complete is sufficient here because the ability to get doubles and triples is an obvious FI skill with clear and accepted explanations. This is a different case than that of Matt Cain regularly well out-performing his FIP, which has no clear and accepted explanation (besides the empty “there must be a skill there” gesture)
2) How is it FI when BABIP is included? Is it that we can predict what the batter contributes in value given a combination of luck and fielding talent he faces?
3) Is the initial discovery surprising given the premises of Sabermetrics regarding the value of being on base at all? FIB looks to me like an analytic truth (or necessary consequence) of the premise that a batter’s value increases as his ability to reach base increases. (I don’t mean to question the truth of the premise. I only mean to point out that it is unclear that FIB is much of a new discovery.)
I think point 3) from LTG is the best summary as to why I don’t find this all that revolutionary.
Hits plus walks are the most significant ways of getting on base. Using BB%, HR%, and BABIP as inputs just means that one of the inputs is essentially OBP (except for players who have an extreme aversion or extreme attraction to HBP, which is a source of error in FIB). HR% also of course functions as a rough proxy for SLG. So it seems like you’re just using rough approximations of OBP and SLG (both somewhat regressed by using xBABIP) to approximate player value… which seems like a pretty standard approach.
I’m basically left wondering if including K% in the calculation helps it all. If it does, I guess that’s the key insight?
1) I looked into including doubles or triples into the model in one of my previous posts:
http://www.fangraphs.com/blogs/index.php/this-shh-just-got-real/
The problem is that including doubles and triples would be double-counting their contribution. So, the alternative then is to use a funky BABIP or using steals as a proxy for doubles and triples. The first option undoes the beautiful simplicity of the regression and formula, while the second option was not statistically worthwhile (the relationship was broken at best).
Moreover, the effect appears to be more like the different between xFIP and SIERA than Cain and FIP. Even the double hitters, like Lou Brock in the above article, stick pretty close to their Should Hit estimators.
2) See my response to Ralph above. In some respects, yeah, it’s not fielding independent. In others, indeed it is.
3) And yes one more time. This is nothing crazy new. I demonstrated in the article I mentioned in (1) that others have explored this arena for years and years. At the same time, though, Should Hit’s predecessors never employed such a simple or effective estimation.
The fact that the ShHAP! formula is intuitive and obvious is a credit to its simplicity, not a detraction from its discovery.
My other problem is that you can’t simply magically lower a player’s K% and assume everything else will stay the same, even though guys like Ortiz and Fielder are doing their best to contest that sort of thought.
Can’t wait until we see FIF, Fielding Independent Fielding. It will take over UZR. It will be partly based on shoe size and hair color.
and what team they play for. Of coarse, Boston players will be heavily favored in this stat, so it won’t be much different than UZR.
everyone will score 0 except david eckstein
I just need to say that I can’t fail but fall behind a statistic that has a predictor basically called “Shit Happens”.
I haven’t followed this series closely since the first article, but that did get me thinking…
Couldn’t you predict dragon-independent hitting more simply by just regressing singles, doubles, and triples down to expected BABIP levels? That is, just multiply each of those outcomes by xBABIP/BABIP (or career BABIP / BABIP, or even just .300/BABIP).
Then you can calculate wOBA based on a sensible BABIP instead of a silly one and find out what the hitter “should hit” when the luck dragon falls to earth.
i.e. xwOBA = (.41*IBB + .7*NIBB + .73*HBP + (xBABIP/BABIP)*(.89*1B + 1.27*2B + 1.61*3b) + 2.07*HR + .25*SB + .5*CS) / PA
…which is just wOBA with 1B, 2B, and 3B scaled to a success rate of xBABIP instead of BABIP.
Reed Johnson, for example, has a wOBA of .404, but if you assumed that his success in hitting singles, doubles, and triples happened at his career BABIP rate of .334 instead of his 2011 BABIP of .438, the xwOBA comes out to around .330, which is about what we’d expect when assessing his true talent.
Of course, this assumes that players’ BABIP luck is roughly equal for 1b, 2b, and 3b, but beyond that it shouldn’t reward certain types of hitters over others. It’s also a lot simpler and more transparent than the numbers that emerged from the regression.
IIUC, this is exactly what FIP does for pitchers, except that for pitchers the xBABIP is (almost) always .300. So FIP just uses the linear weights relative to that expected BABIP, giving a number of what the pitcher “should pitch” when his BABIP returns to normal.
Since this new stat is supposed to be a “predictor,” shouldn’t we see if it actually predicts something before we anoint it as a “revolution”?
Bradley:
Have you seen this article posted by Bill Petti at BTB today?
http://www.beyondtheboxscore.com/2011/8/29/2391327/in-season-batter-regression-tooled-updated#storyjump
Any comparison/contrast thoughts on your tool and his?
Basically, are regressing BB rate and K rate in addition to BABIP and HR/FB adding much to our understanding of the situation?