Imagine this: Jon Gray has a 400 mile journey ahead of him. He has his playlist, a list of rest stops he hopes get cleaned regularly, and snacks. Most importantly, he has directions. Let’s say there are 15 steps to take. They will get him to his destination in the most efficient way possible. And he does everything right—except he misses two of the steps. Those mistakes constituted just a small portion of all of the directions, but they added 150 miles and two hours to the trip.
Now, imagine saying this: “If you take out those two missed turns Gray didn’t drive any extra miles and actually arrived before he actually arrived.” Pretending the mistakes didn’t happen doesn’t whisk them away from reality.
The baseball version of this is, “if you take out those two starts,” then real reality comes forth and gives a truer version of what we have in the record books. In general, this isn’t a great way to do baseball analysis, especially when thinking forward to what the player in question will do next.
But, there’s a but. Rob Mains of Baseball Prospectus recently analyzed ($) and relayed an article by Mitchel Licthman which suggests that, in certain circumstances, it does make sense to take out some particularly awful appearances when projecting future performance. One of the pitcher seasons that most exemplifies this is Jon Gray’s 2016.
In order to draw those conclusions, the games to take out need to meet certain conditions. In this experiment, a game has to be truly awful to be excluded from a pitcher’s projections: eight or more runs in five or fewer innings. Gray had two such outings in 2016. On May 19, he allowed 9 runs in 31⁄3 innings against the Cardinals, and on August 7, he gave up 8 runs in 32⁄3 innings against the Marlins. These weren’t his only bad starts of the season, but they were his worst.
If we take those two starts out, Gray’s season ERA falls from 4.61 to 3.86. In other words, according to Mains by way of Licthman, projecting Gray’s 2017 using either ERA or simple RA9 will be a bit higher than it should be.
It’s necessary to note that this doesn’t mean one can take out any old start to make the projection look better and better. For instance, the time Gray gave up seven runs in fewer than five innings and the two times he gave up six runs in fewer than five have to stay. Removing those would probably lead to less trustworthy projections, whereas what’s so compelling about the analysis summarized here is that removing the truly terrible starts lead to more trustworthy ones.
Ultimately, this is just a fun nugget of optimism as spring training plods along (thank the stars for the World Baseball Classic). We can safely take out Gray’s two terrible starts from 2016 when thinking about his 2017. Now let’s see what he actually does. Maybe this time around he won’t miss any turns at all.