Chase Utley: One of the Best Ever

If you are reading this, there is a remarkably decent chance that you are a somewhat invested baseball fan. As a baseball fan, you probably ...

Attempting to fix xwOBA

xwOBA is an interesting but flawed stat. It takes the exit veto and launch angle of each batted ball and assigns an estimated wOBA to the specific outcome. It then takes the xwOBA on contact, or xwOBACON, and fully factors K% and BB% to create the actual xwOBA. It is pretty good, but there is a general flaw in creating such a broad stat based purely on two inputs. Other factors, like sprint speed and launch direction, are not included in the calculation but will effect the actual outcomes on the baseball field. Unfortunately, Baseball Savant makes it impossible to find launch direction but that's not the point of this post anyways. 

Home runs are a true outcome, just like K% and BB%. The correlation to future success on home runs are just as stable as BB%, but it is not treated as a true outcome in xwOBA. I decided to change that. I took xwOBACON on balls that were not home runs, normalized it to fit the actual league average so things would not be inflated, and then factored in BB%, K%, and HR% using the basic wOBA values. This makes more sense in terms of a descriptive purpose, and I haven't tested things out but it is likely a lot more predictive as well. It still probably lags behind the incredible predictive powers of DRC+, but it is likely better than xwOBA. 

Nolan Arenado went from a 0.344 xwOBA to a 0.385 xwOBA. Alex Bregman went from 0.378 to 0.418. Matt Chapman fell from 0.360 to 0.354. Cody Bellinger went from 0.429 to a very impressive 0.460, just a tick behind Mike Trout, who also heavily improved to 0.461. I do want to acknowledge that it this stat definitely overrates the value of good players by a smidgen. The league average is still 0.314, but there is more variation between players than what there should be.

To normalize the xwOBA I just plugged in the new xwOBA into an equation which puts it on the same scale as the original xwOBA. The rankings and percentiles did not change, but the values did. Then I went and used a different equation factoring sprint speed into the xwOBAcon that I had. For example, Mike Trout dropped from 0.461 to 0.430, but the sprint speed adjustment brought up up to 0.433, and he still ranked first in the league. Jeff Mathis jumped from 0.191 to 0.216, still the worst in the league, but a higher value. Bregman and Arenado, who I mentioned previously, fell down to .397 and .369 respectively, but Bregman remained in the 98th percentile and Arenado in the 93rd percentile.  You can find the full results for 2019 here-> here <- . 

No comments:

Post a Comment