Using Batted-Ball Data to Measure Hitter Performance

Imagine a batter hits a long fly ball that’s destined for the right-field seats only for the outfielder on the other team to clear the wall and rob him of his home run. In traditional stat sheets, this is treated the same way as any other out and there’s no real way of distinguishing that from a dribbler down the third-base line. But intuitively we know that these are two very different things, and a batter who does more of the first is going to end up being more valuable than one who does more of the second. Thus, if we wanted to truly measure how well a player has performed, we need to separate the performance from the results. The best way of doing that is to break down a batted ball in the most granular way possible and look at the average performance for similar batted balls, and today I’ll reveal a personal tool to do this. This work was inspired by Tony Blengino’s terrific posts on batted-ball data, and I suggest reading his introductory post as background on the theory and methodology that I employ.

This tool uses information on the type, velocity, direction, and distance of a hitter’s batted balls to calculate an expected AVG, OBP, and SLG for him. It divides batted balls into buckets based on the type (GB, FB, LD, PU) and either the direction and velocity or the direction and the distance and calculates the resulting AVG and SLG for all batted balls that meet that criteria. It then goes through all of a batter’s plate appearances and uses these data to calculate both the observed and expected AVG/OBP/SLG for each PA. The table below shows the top 30 hitters by Expected wOBA (xwOBA) as of 5/26/2015.

Name	AB	PA	Velocity	AVG	OBP	SLG	wOBA	wRAA	xAVG	xOBP	xSLG	xwOBA	xwRAA
Bryce Harper	151	191	89	0.331	0.471	0.722	0.505	29.1	0.298	0.445	0.650	0.467	23.3
Miguel Cabrera	164	195	93	0.341	0.446	0.610	0.453	21.7	0.304	0.415	0.665	0.457	22.3
Prince Fielder	182	199	93	0.363	0.417	0.571	0.425	17.7	0.349	0.404	0.640	0.443	20.5
Mike Trout	168	194	92	0.298	0.392	0.548	0.404	14.0	0.321	0.412	0.615	0.438	19.3
Anthony Rizzo	161	197	88	0.311	0.437	0.565	0.433	18.7	0.304	0.431	0.589	0.438	19.6
Ryan Braun	154	173	94	0.266	0.347	0.532	0.376	8.7	0.298	0.375	0.661	0.436	16.9
Paul Goldschmidt	160	190	93	0.338	0.442	0.631	0.459	22.0	0.290	0.402	0.615	0.433	18.1
Adrian Gonzalez	158	179	89	0.342	0.419	0.620	0.443	18.5	0.322	0.401	0.614	0.432	16.9
Todd Frazier	164	187	92	0.256	0.348	0.549	0.382	10.4	0.304	0.390	0.620	0.429	17.2
Yasmani Grandal	104	124	95	0.288	0.403	0.462	0.379	6.6	0.310	0.421	0.574	0.428	11.3
Brandon Crawford	151	170	93	0.298	0.376	0.510	0.383	9.5	0.316	0.393	0.608	0.426	15.2
Brandon Belt	139	156	93	0.302	0.378	0.496	0.379	8.2	0.316	0.391	0.606	0.424	13.8
Nelson Cruz	170	186	92	0.341	0.398	0.688	0.456	21.2	0.295	0.356	0.654	0.423	16.3
Alex Rodriguez	146	170	94	0.260	0.365	0.541	0.388	10.2	0.283	0.384	0.612	0.423	14.9
Joc Pederson	146	179	95	0.247	0.385	0.548	0.401	12.6	0.257	0.394	0.592	0.421	15.4
Mark Teixeira	147	177	87	0.231	0.362	0.551	0.390	10.9	0.281	0.402	0.560	0.414	14.2
Hanley Ramirez	158	170	94	0.259	0.312	0.468	0.336	3.2	0.318	0.366	0.590	0.406	12.6
Stephen Vogt	131	155	87	0.298	0.406	0.580	0.423	13.5	0.283	0.394	0.544	0.404	11.2
Cameron Maybin	109	126	92	0.248	0.349	0.404	0.332	2.0	0.304	0.398	0.537	0.403	9.0
Jose Bautista	133	165	92	0.211	0.364	0.444	0.353	5.4	0.252	0.397	0.530	0.401	11.5
Josh Reddick	153	170	90	0.314	0.382	0.536	0.395	11.1	0.302	0.372	0.561	0.399	11.6
Brian Dozier	174	196	90	0.247	0.332	0.494	0.355	6.6	0.284	0.365	0.572	0.399	13.4
Adam Jones	167	178	91	0.311	0.354	0.479	0.360	6.8	0.319	0.361	0.571	0.397	11.9
Freddie Freeman	169	188	92	0.302	0.372	0.485	0.372	8.9	0.304	0.375	0.553	0.397	12.6
Giancarlo Stanton	174	198	97	0.230	0.323	0.500	0.353	6.4	0.249	0.340	0.598	0.396	13.1
Matt Carpenter	165	184	91	0.321	0.391	0.582	0.416	15.0	0.293	0.366	0.557	0.394	11.9
Eric Hosmer	171	192	91	0.310	0.385	0.520	0.391	11.9	0.306	0.382	0.534	0.394	12.4
Lucas Duda	161	186	92	0.292	0.387	0.491	0.381	10.2	0.285	0.381	0.536	0.394	12.1
Mark Trumbo	144	152	93	0.264	0.303	0.507	0.345	3.9	0.298	0.335	0.600	0.394	9.8
Corey Dickerson	111	117	90	0.306	0.342	0.523	0.370	5.3	0.317	0.352	0.573	0.393	7.4

The tool uses the velocity and direction, rather than the distance and direction, of a batted ball to calculate the expected values with a few exceptions. If the velocity is not available for a fly ball or a line drive, it uses the distance and the direction of the batted ball to calculate the expected values. If the velocity of the batted ball is not available for a ground ball, the tool assumes it was of average velocity and only considers the direction it was hit when calculating the expected values. It does not consider distance for ground balls, as the distances are calculated using where the ball was fielded, so using distance would be describing what actually happened rather than what we expected to happen. For all line drives and fly balls hit over 375 feet it uses distance and direction rather than velocity and direction. The reason for this is that I do not have information on the hang time of batted balls, and in going through the data I found that fly balls and line drives that traveled over 375 feet but weren’t hit very hard were being severely underrated by the tool. As an example of the underlying data, the table below shows the reference data for fly balls hit to center field.

TYPE	Velocity Range (MPH)		Direction Range (90=CF)		AVG	OBP	SLG
FB	105	150	85	95	0.732	0.732	2.511
FB	100	105	85	95	0.314	0.314	0.931
FB	97.5	100	85	95	0.082	0.082	0.247
FB	95	97.5	85	95	0.023	0.023	0.047
FB	92.5	95	85	95	0.000	0.000	0.000
FB	90	92.5	85	95	0.010	0.010	0.038
FB	87.5	90	85	95	0.025	0.025	0.063
FB	85	87.5	85	95	0.000	0.000	0.000
FB	80	85	85	95	0.020	0.020	0.050
FB	75	80	85	95	0.056	0.056	0.070
FB	70	75	85	95	0.220	0.220	0.231
FB	65	70	85	95	0.583	0.583	0.590
FB	60	65	85	95	0.145	0.145	0.145
FB	55	60	85	95	0.073	0.073	0.073
FB	0	55	85	95	0.073	0.073	0.073

I’m providing a link to a Google Sheets document with a leaderboard for all qualified batters, along with leaderboards broken down by each batted ball type. The document also contains a reference page that contains all the information for how batted balls performed in each bucket based on 2015 StatCast data for velocity references and 2014-2015 MLBAM data for distance references. The numbers in the reference page will continue to be updated as more data becomes available from StatCast. Feel free to look through this section and point out any inconsistencies you may see, and note that all data comes from BaseballSavant.

I’ve also provided a Methodology Example in the document so you can dig through what the behind the scenes data looks like as it’s being processed. Note that you may see some discrepancies in a player’s actual AVG seen here and his AVG seen elsewhere, as I treat sac flies as regular outs. The “Notes” tab gives a general outline of the procedure, and also contains a link to an Excel sheet that you can download to perform these calculations on your own.

https://docs.google.com/spreadsheets/d/1-XohbJlWIceDS2Rc8_7-rOxv9avU3IwMCecPkUNxlYU/edit?usp=sharing

Before I wrap up, I should also mention the limitations. It’s been noted elsewhere on FanGraphs that the StatCast data isn’t always completely accurate. Also, the tool currently doesn’t incorporate a player’s speed in any way, so guys like Dee Gordon are going to be fairly underrated in terms of their ground ball performance. I’ve been brainstorming ways to incorporate this and am open to any input you may have. Furthermore, I’ve noticed the tool can be pretty stingy with labeling balls as pop-ups and occasionally pretty generous with labeling them as line drives. I’ve noticed some fly balls with velocities over 95 MPH that only traveled 300 feet, indicating they were hit almost straight up in the air. Unfortunately, without data on the vertical angle of the ball off the bat or on the hang time of the ball in play, it will be difficult to fix this issue.

Even with these limitations, the tool works extremely well at determining how well guys have been hitting the ball and identifying who has been helped or hurt by factors beyond their control. Take the time to dig through the data and the code and point out areas for improvement, and I’ll incorporate them in future versions.