/cdn.vox-cdn.com/photo_images/1177543/GYI0061676443.jpg)
Hitters' offensive statistics are substantially influenced by random variation---which we sometimes refer in shorthand as being "lucky" or "unlucky." Batting average on balls in play (BABIP), which affects many standard batting statistics, is the most variable measure. However, measures of overall offensive performance also can vary from season to season for reasons other than the player's skill.
Now that the season is at a midpoint, and some of the peripheral stats have begun to stabilize, we can peek at signs of under- or over-performance by Astros' hitters. "Over-performance," as used here, means that the actual batting results are better than expected based on prediction models. "Under-performance" means the converse. So, this test will tell us if the players' performance has been lucky / unlucky--- as opposed to reflecting the players' true skill level.
Expected BABIP, or x-BABIP, will be compared to hitters' actual BABIP this season. A predicted wOBA, based on Fielding Independent wOBA, will be compared to actual wOBA. As you will see, these two methods are linked together, since wOBA is dependent on BABIP. Why is this useful? It is possible that the players' future offensive performance may regress toward the predicted batting performance.
x-BABIP
As I have done in the past, I rely upon a model developed by Chris Dutton and Peter Bendix and published in the Hardball Times in 2008. (Since Dutton and Bendix were students at Tufts University at the time, I will call it the Tufts Model.) I use the x-BABIP calculator based on this model provided by Hardball Times. The model reflects the historic relationship among HR, SB, LD, GB, FB, and IFFB variables for predicting BABIP.
If the Astros were a team loaded with veteran hitters, we might not require the x-BABIP model, since a player's BABIP averaged over 6 or 7 full time seasons might be a reasonable indicator of the player's normal BABIP. But the young Astros' team doesn't have hitters with sufficient ML experience to reliably identify their normalized BABIP. Therefore, x-BABIP models are particularly useful in evaluating a team like the Astros.
Based on the x-BABIP results, only Chris Johnson, Jose Altuve, Brian Bixler, and Brett Wallace out performed their predicted BABIP. Because Altuve's predicted BABIP was less than 1% below his actual BABIP, he probably shouldn't be in the over performance group. His actual BABIP is pretty much right on target.
The following battters had an x-BABIP above their actual BABIP, and perhaps could be viewed as "unlucky": Buck, Marwin Gonzalez, Lowrie, Bogusevic, Downs, Castro, Martinez, Maxwell, Snyder, and Schafer. Bogusevic, Downs, and Snyder had particularly large deviations from their expected BABIP.
The table below shows the Astros hitters' 2012 BABIP compared to the x-BABIP for that period., based on the Tufts model's version of x-BABIP.
actual |
x-BABIP |
diff. |
Percent |
|
Buck |
0.286 |
0.334 |
0.048 |
17% |
Lowrie |
0.269 |
0.281 |
0.012 |
4% |
Gonzalez |
0.293 |
0.308 |
0.015 |
5% |
Maxwell |
0.313 |
0.333 |
0.02 |
6% |
Castro |
0.306 |
0.337 |
0.031 |
10% |
Bogusevic |
0.264 |
0.329 |
0.065 |
25% |
Schafer |
0.339 |
0.342 |
0.003 |
1% |
J. Martinez |
0.279 |
0.319 |
0.04 |
14% |
Snyder |
0.256 |
0.329 |
0.073 |
29% |
Downs |
0.157 |
0.295 |
0.138 |
88% |
Johnson |
0.36 |
0.336 |
-0.024 |
-7% |
Altuve |
0.338 |
0.335 |
-0.003 |
-1% |
Bixler |
0.359 |
0.332 |
-0.027 |
-8% |
Wallace |
0.5 |
0.341 |
-0.159 |
-32% |
The Tufts model isn't the only regression based model for predicting BABIP. Fangraphs has a different x-BABIP calculator, which uses different variables and purports to reflect reduced league wide BABIP resulting from increased use of shifts. I'm not sure I would accept this method as preferable to the Tufts model. But rather than dissecting the pros and cons of each x-BABIP method, I will describe the major differences pertaining to Astros' hitters. The Fangraphs x-BABIP is calculated on the spreadsheet downloadable from here.
The Fangraphs' x-BABIP essentially is the same as the x-BABIP above for Bogusevic, Altuve, Martinez, and Schafer. The x-BABIP above is higher than the Fangraphs' version for Buck, Gonzalez, Maxwell, Snyder, Downs, Bixler and Johnson. The x-BABIP above is lower than the Fangraphs' x-BABIP for Castro and Lowrie. Fangraphs didn't make a calculation for Brett Wallace. The Fangraphs' version doesn't change the conclusion of over or under performance, as reflected above, except in the case of Maxwell. The Fangraphs' x-BABIP predicts a very low .280 BABIP for Maxwell, which diverges significantly from the x-BABIP of .333, above, and changes his BABIP status from under performance to over performance.
Fielding Independent wOBA
As most of you know, wOBA is a composite offensive production statistic similar to OPS, except that it is more accurate. Fangraphs has developed a new offense evaluation approach called Fielding Independent Offense, which is the offensive equivalent of Fielding Independent Pitching (FIP). Fielding Independent wOBA, based on BB%, K%, HR%, SB%, and x-BABIP, can be used to predict a hitters' wOBA. With the inclusion of the x-BABIP formula, the Fielding Independent wOBA is similar to a previous formula, Predicted OPS (PrOPS), which has been used to predict hitters' future offensive performance.
What does Fielding Independent wOBA do for us? It can tell us how much the player's actual wOBA deviated from the predicted wOBA, based on the player's peripheral stats and expected BABIP. Certainly all deviations may not be a matter of luck, but it does suggest the potential for upward or downward regression for a player's future wOBA. I used the Fangraphs spreadsheet for wOBA downloaded here.
As discussed previously, the Fangraphs version of x-BABIP can produce different results than the Tufts Model I used. Therefore, I will present two different calculations of Fielding Independent wOBA based on the different x-BABIP formulas. No. 1 corresponds to the x-BABIP table presented above. No. 2 is based on the Fangraphs' x-BABIP. The predicted wOBA is best represented by a range, since these are estimates subject to error margins. So you can think of the two wOBA predictions as a range of possibilities.
Some conclusions to be drawn from the table below: (1) Jose Altuve may incur some regression in wOBA, but it should be relatively small in magnitude; (2) Most of the Astros' batters should be getting better offensive results, based on their peripheral stats; (3) If you are worried that Jed Lowrie has reached the peak of his performance, the predicted wOBA says you are wrong, and his regression is likely to be upward; (4) There may be some hope for Schafer's offense, given that he has under performed despite what appears on the surface to be an above average BABIP; (5) Chris Johnson may regress downward; and (6) the small sample offense provided by Wallace and Bixler is fueled by unsustainable BABIP.
Actual |
Predicted wOBA |
Predicted - Actual % |
||||
wOBA |
f-wOBA 1 |
f-wOBA 2 |
1 |
2 |
||
Matt Downs |
.227 |
.338 |
.279 |
49% |
23% |
|
Brian Bogusevic |
.288 |
.331 |
.329 |
15% |
14% |
|
Chris Snyder |
.279 |
.327 |
.308 |
17% |
10% |
|
Jason Castro |
. offe308 |
.322 |
.333 |
5% |
8% |
|
Jordan Schafer |
.289 |
.305 |
.305 |
5% |
5% |
|
Jed Lowrie |
.350 |
.359 |
.365 |
2% |
4% |
|
J.D. Martinez |
.314 |
.347 |
.323 |
10% |
3% |
|
Marwin Gonzalez |
.281 |
.299 |
.290 |
7% |
3% |
|
Travis Buck |
.266 |
.281 |
.260 |
6% |
-2% |
|
Jose Altuve |
.343 |
.332 |
.336 |
-3% |
-2% |
|
Justin Maxwell |
.337 |
.363 |
.329 |
8% |
-2% |
|
Brian Bixler |
.321 |
.307 |
.296 |
-4% |
-8% |
|
Chris Johnson |
.319 |
.304 |
.292 |
-5% |
-8% |
|
Brett Wallace |
.434 |
.366 |
N/A |
-16% |
N/A |