Validating the New xBABIP Equation With the Surgers

by
February 13, 2017

A week ago, I introduced the newest version of our ever evolving xBABIP equation, this time incorporating the much-needed shift data. Last Thursday, I identified the 10 fantasy relevant hitters with the greatest BABIP upside in 2017, given the gap between their 2016 BABIP and xBABIP. In the comments, I was asked if I could perform a retrospective analysis to see how the new equation would have done if I ran it heading into the 2016 season.

I’ll start by looking at all hitters that posted an xBABIP of at least 0.030 above their actual BABIP marks in 2016 (e.g., a .300 xBABIP vs a .270 actual BABIP), which totaled 39 batters. I also included 2016 Steamer projections to determine whether xBABIP is truly providing any incremental value. So here we go.

2016 BABIP Surgers
Name 2015 BABIP 2015 xBABIP 2015 BABIP-xBABIP 2016 Steamer Projected BABIP 2016 BABIP 2016 BABIP – 2015 BABIP
Jeff Mathis 0.186 0.280 -0.094 0.266 0.318 0.132
Nick Franklin 0.213 0.299 -0.086 0.291 0.325 0.112
Chris Herrmann 0.203 0.281 -0.078 0.282 0.364 0.161
Jon Jay 0.246 0.317 -0.071 0.312 0.371 0.125
Jed Lowrie 0.233 0.300 -0.067 0.281 0.316 0.083
Chase Utley 0.230 0.291 -0.061 0.267 0.299 0.069
Alexi Amarista 0.232 0.292 -0.060 0.263 0.313 0.081
Cliff Pennington 0.253 0.311 -0.058 0.283 0.289 0.036
Hanley Ramirez 0.257 0.314 -0.057 0.308 0.315 0.058
Jake Smolinski 0.208 0.264 -0.056 0.287 0.257 0.049
Kirk Nieuwenhuis 0.280 0.334 -0.054 0.302 0.302 0.022
Jonathan Lucroy 0.297 0.347 -0.050 0.296 0.322 0.025
Tyler Saladino 0.269 0.316 -0.047 0.285 0.329 0.060
Jose Ramirez 0.232 0.279 -0.047 0.280 0.333 0.101
Aaron Hill 0.253 0.299 -0.046 0.275 0.284 0.031
Jayson Werth 0.253 0.297 -0.044 0.316 0.288 0.035
Justin Smoak 0.254 0.294 -0.040 0.274 0.295 0.041
Carlos Sanchez 0.270 0.310 -0.040 0.303 0.257 -0.013
Jarrod Dyson 0.296 0.336 -0.040 0.303 0.315 0.019
Albert Pujols 0.217 0.257 -0.040 0.253 0.260 0.043
Ryan Zimmerman 0.268 0.307 -0.039 0.301 0.248 -0.020
Wilson Ramos 0.256 0.295 -0.039 0.276 0.327 0.071
Ryan Howard 0.272 0.310 -0.038 0.284 0.205 -0.067
Chris Coghlan 0.284 0.321 -0.037 0.287 0.235 -0.049
Danny Santana 0.290 0.327 -0.037 0.324 0.305 0.015
Luis Valbuena 0.235 0.271 -0.036 0.264 0.315 0.080
Chris Owings 0.305 0.340 -0.035 0.307 0.334 0.029
Carlos Correa 0.296 0.331 -0.035 0.309 0.328 0.032
Paulo Orlando 0.291 0.325 -0.034 0.302 0.380 0.089
J.T. Realmuto 0.285 0.319 -0.034 0.286 0.357 0.072
Tim Beckham 0.279 0.313 -0.034 0.293 0.349 0.070
Steve Pearce 0.232 0.265 -0.033 0.270 0.318 0.086
Gordon Beckham 0.229 0.262 -0.033 0.263 0.245 0.016
Welington Castillo 0.263 0.295 -0.032 0.295 0.337 0.074
Omar Infante 0.255 0.287 -0.032 0.284 0.278 0.023
A.J. Ellis 0.265 0.297 -0.032 0.264 0.252 -0.013
Coco Crisp 0.218 0.250 -0.032 0.264 0.252 0.034
Jay Bruce 0.251 0.282 -0.031 0.276 0.266 0.015
Alexei Ramirez 0.264 0.294 -0.030 0.270 0.265 0.001
Averages 0.254 0.300 -0.046 0.286 0.301 0.047

Obviously, there’s little need to dive into any specific names here. Of the 39 batters who underperformed the most, only five of them posted 2016 BABIP marks below their 2015 BABIP marks. That means that 34 of the 39 (87%) of these underperformers increased their BABIP marks!

The most important row is “Averages”. You can see that the group suffered from some awful BABIP fate in 2015, posting an unweighted (I ignored at-bats for simplicity) BABIP of just .254. However, xBABIP thought these hitters should have enjoyed much greater success, figuring a mark of .300, right near the league average. On average, these batters underperformed their xBABIP marks by a whopping 0.046 points. That’s bad! But perhaps greener pastures were in their future.

Being rather familiar with BABIP, we would typically look at an average at .254 and almost blindly forecast an increase the following season. So gee, xBABIP thinks guys averaging a .254 BABIP will improve in 2016? I could have guessed that myself! That’s why I included the Steamer projected BABIP, and you’ll learn that the best projection system did indeed forecast vast improvement to .286. It’s interesting to note that Steamer doesn’t even project batted ball distribution (FB%/GB%, etc.), so it’s not like the system is observing a high line drive rate relative to BABIP and expecting better results.

Yet, Steamer’s projected improvement actually proved not to be optimistic enough! Move one column over and you see the group actually posted a .301 BABIP, 15 points higher than Steamer projected and nearly dead on with their 2015 xBABIP.

So the early return suggest that the new xBABIP does provide additional value over and above Steamer. And guess what? The BABIP forecasts in my Pod Projections reflect the new xBABIP equation, which should help boost their accuracy.





 

Mike Podhorzer is the 2015 Fantasy Sports Writers Association Baseball Writer of the Year. He produces player projections using his own forecasting system and is the author of the eBook Projecting X 2.0: How to Forecast Baseball Player Performance, which teaches you how to project players yourself. His projections helped him win the inaugural 2013 Tout Wars mixed draft league. Follow Mike on Twitter @MikePodhorzer and contact him via email.

10 Comments
Oldest
Newest Most Voted
Inline Feedbacks
View all comments
alemungermember
7 years ago

This is awesome stuff, and the issue of xBABIP is something that has frustrated me for years. I’ve played with various regressions of possible indicators myself, so I feel how exciting progress such as this is.

There’s a couple things here that have always confused me regarding this stuff, though. First, it was obviously very astute of you to separate FB% and IFFB% in your regression given the considerable difference in their coefficients, something I hadn’t thought of before. There is something weird about FB% and BABIP for me though. Given that HRs are subtracted out of the BABIP equation, there is an incongruence in the events actually being compared between FB% and BABIP. Especially for power hitters with high HR/FB rates, there could exist a significant number of batted balls that are explicitly subtracted in the BABIP calculation, and there would be no reason I can see for this “white noise” to improve the correlation; it could only hurt it. I wonder if subtracting home run balls from FB% could further assist your improving R-squared values.

There is of course the possibility that the correlation between True FB% and HR/FB% is so high that this extra step is negligible, I’m not sure. There is also the issue of LD HRs, which, through a quick Baseball Savant query, account for about 1/3 of all HRs. Given, separating HRs into their batted ball types seems a bit pointlessly tedious, but the initial FB adjustment would not be too difficult.

One other question/suggestion. Incorporating speed into xBABIP is clearly necessary given the massive boost IFHs can give to a player’s BABIP, so is there a particular reason you used Spd instead of IFH%? To me, IFH% would more directly reflect the impact of speed on BABIP, given that Spd is a conglomeration of metrics that don’t directly have to do with getting base hits (the closest would be the 3Bs component of the metric). Again, it is highly possible that Spd correlates so highly with IFH% that the difference in their respective relationships to BABIP is negligible, but any possible accuracy boost seems valuable at this point. The one reason I can think of that this would not make sense is if the YoY correlation for Spd is drastically higher than IFH%. I assume it would be somewhat higher given the composite nature of the metric, but I haven’t done the math myself.

All in all, this is great, valuable work. Just trying to wrap my head around all of it and do what I can to make it better.

2
Mike Podhorzermember
7 years ago
Reply to  alemunger

It sounds like you’re thinking HR/FB rate is negatively correlated with BABIP, is that correct? I understand this idea, I’m not sure if the data supports it though. I’d have to check.

I’m fairly certain that the batted ball data on FG classifies the vast majority of homers as fly balls, regardless of launch angle. Statcast is completely objective, while the batted ball data here is subjective, so it’s rare a homer is considered a line drive.

Good question and I have the answer — IFH% is a result, whereas Spd is the skill. You can’t include a metric that represents a type of hit into how often the player gets a hit when the ball is in play!

2
alemungermember
7 years ago
Reply to  Mike Podhorzer

Of course, that makes sense about IFHs. And you are right, the Fangraphs batted ball classification only classifies like 3% of HRs as line drives.

I quickly ran BABIP vs. a recalculated True FB% without HRs on the same dataset you supplied in the previous article. Your correlation was -0.335 (which I confirmed myself to make sure our data was consistent). The correlation I got using “in-play” True FB% was -0.362, a marginally stronger relationship. I’m not sure what effect, if any, that modification would have on the final R^2 value of the regression.

Actually, my hypothesis was that HR/FB would have no relationship to BABIP, considering that HRs are explicitly subtracted out of the BABIP calculation. I believe my hypothesis was confirmed; I got a correlation of 0.036 between BABIP and HR/FB, which is likely the reason for a slightly stronger relationship with in-play True FB% than with the variety that includes HRs. This finding actually perfectly analogues with my question about IFHs. If you were to subtract IFHs from BABIP calculation (in the same manner as you do with HRs), I’d bet you’d not find any meaningful relationship between IFH/GB and this new “OFH” BABIP. I would have to check that though.

0
alemungermember
7 years ago
Reply to  Mike Podhorzer

I just ran your identical regression with the revised “in-play” True FB% statistic. My equation looks like this:

xBABIP = 0.1956 + (0.3900 * LD%) – (0.1566 * IP True FB%) – (0.4328 * True IFFB%) + (0.2280 * Hard%) + (0.0049 * Spd) – (0.1572 * Pull GB While Shifted%)

Adjusted R^2 = 0.5407

So, a tiny bit of progress.

The formula for the modified stat:
IP True FB% = (True FB% * (AB-SO+SF) – HR) / (AB-SO+SF-HR)

1
Mike Podhorzermember
7 years ago
Reply to  alemunger

Cool, small gain, but seems like it’s still meaningful.

0