I like the approach here, especially relying only on prior years' data so the predictions are true predictions. To some extent, what this suggests is that guys who pitch high in the zone will have lower BABIP (more popups, more Ks). A couple of questions:

1) the (weak) link between BABIP and K-rate has been established. Does using swing-miss rate improve the prediction significantly, vs. using K/PA?

2) Is popup rate a better predictor than using the more general FB%?

Interesting stuff, Dan! Are there any (famous) exceptions that the formula doesn't catch or does it seem to grab all of the guys who came ot mind?

To some extent, what this suggests is that guys who pitch high in the zone will have lower BABIP (more popups, more Ks).

I wonder if there's pitch f/x data available that we could use to test this potential explanation.

bump!

To some extent, what this suggests is that guys who pitch high in the zone will have lower BABIP (more popups, more Ks).

If we had a "slow roller back to the pitcher" rate, and it were not predictive, I'd tend to agree. I think the focus on popups is simply because we don't have stats on the GB equivalent.

IOW, I don't think it's a high-in-the-zone thing as much as a not-where-expected-in-the-zone thing. It's an incomplete indicator of a skill at inducing weak contact.

Still, I think the high-zone thing is worth checking through PITCHf/x, as #2 suggested.

Knuckleballer, lefty, lefty, lefty, Matt Cain.

Is BABIP adjusted for platoon split or is it just regressed to "league average"?

Tom Tango’s FIP assumes that all pitchers have exactly league-average BABIP ability

Guess I should RTFA. This strikes me as a terrible decision. The platoon split is real and LHP should have a different BABIP than RHP. Has anyone done a FIP with platoon split BABIP?

We saw the good outliers. Who are the bad outliers for popups and contact rate? I want to know which guys are unlikely to outperform their FIP too. I looked up Scott Diamond and he looks pretty ugly by this formula.

I will mention that it's very easy to construct a FIP type stat that doesn't assume league average BABIP from every pitcher. It still would just use walks, strikeouts, homers, etc. but to the extent those numbers correlate with BABIP you could adjust the coefficients accordingly.

Indeed if you were to do a linear regression, the regression would take care of that for you (and maybe FIP does take that into account). This would have the added benefit of still being defense independent. Dan's work above is also defense independent a makes use of the newly available (relatively speaking) dataset.

Voros/8: By design, FIP does not take into account any relationship between BABIP and Ks, BBs, or HRs. It's designed to reflect only the actual run values (approx) of those 3 variables.

It's actually not quite right to say FIP "assumes" average BABIP. It simply tells you what a pitcher's expected ERA will be IF he has an average BABIP.

I think the focus on popups is simply because we don't have stats on the GB equivalent.

Perhaps, but I doubt it. BABIP is higher on GBs than on FBs. Pitchers who get hitters to hit the ball in the air should, on average, have slightly lower BABIP. I suppose it's possible there is a subset of GB pitchers who induce especially weak GBs, but it's not even clear that the hit rate on such balls would be especially low (I imagine they would produce a lot of IF singles).

GuyM--Yes, Z-Contact% is a significantly better predictor of BABIP than strikeout rate, and if you put both of them in a regression soup K rate falls out. I suspect that getting batters to swing at non-strikes (O-Swing%) or take called strikes (Z-Swing%), the other components of K rate, don't have much bearing on BABIP, causing K rate to muddy the waters.

And yes, a lot of the signal in GB/FB comes from the IFFB subcategory. Once you have popup rate in there, adding in GB or FB terms doesn't add much.

And you're right that the K rate-BABIP correlation is implicitly included in FIP, so it's not actually presuming league average BABIP for all pitchers as DIPS ERA does. I misspoke.

Arnett--It did seem to get all the big positive outliers, yes--I don't think it missed anyone famous.

DL from MN, it's in the presentation, which you'll see whenever MIT puts it up (or I can email it to you). Zach Duke, John Lackey, Paul Maholm, Jeff Suppan, and Livan Hernandez were the bad-BABIP guys correctly identified by the equation.

And you're right that the K rate-BABIP correlation is implicitly included in FIP, so it's not actually presuming league average BABIP for all pitchers as DIPS ERA does. I misspoke.

Actually, the K:BABIP correlation is NOT implicitly included in FIP. By design, Tango constructed it to reflect only the immediate run value of the K (and HR and BB). (You could select coefficients that took account of the correlation, as I believe Voros' DIPS ERA did, but FIP does not do that.)

Where I would quibble is when you say FIP "presumes" league average BABIP. FIP is not taking a position on how much hit-prevention skill varies, or saying we should expect average BABIP from a pitcher. It's simply telling you how well a pitcher did IF you ignore his BABIP (and also the timing of events).

I really enjoyed the conference. I do think the real value came from the presentations by guys like Dan and not so much from the big names in the ballroom chatting for an hour. I suppose that draws the crowds and was fun to watch, but I found the more specific work being talked about in the other rooms more interesting and valuable in terms of giving me ideas for work in the future.

That's true every year. The big panels with the famous names are always fluffy, because no one is allowed to talk about anything publicly.

Shouldn't it be possible to apply this data to hitters?

I like the approach here, especially relying only on prior years' data so the predictions are true predictions. To some extent, what this suggests is that guys who pitch high in the zone will have lower BABIP (more popups, more Ks).

So, is it possible to assert that guys with low BABIPs who tend to swing more at pitches high in the zone and hit more popups and K more will continue to have lower BABIPs?

Do any of the major projection systems for hitters 'correct' for BABIP? If they do so without regard for the type of hitter and what he swings at, adjustments (regressions) of BABIP to the league average might be in error.

Actually, K are positively associated with BABIP for hitters--the harder you swing, the more likely the balls you do make contact with are likely to become hits.

BABIP for hitters is also such that using BABIP as a means to predict future BABIP is likely to be the strongest signal out there, especially with several full years of data.