Do you really need to dig deep into the data to know that?

You absolutely have to dig deep into the data. The only other study I am aware of that correlations anything to pitchers landing on the DL is Woolner’s work back in 2002 with pitch counts In fact, as I pointed out in the article, none of the variables I used were even weakly correlated by themselves. If you feel this is easy then I encourage you do look at whatever variable you want and do the correlation because I am very confident the result you get will be that the variable you choose is not correlated (less than 0.3 correlation).

Fatigue is the only this data will be able to predict any kind of injury. Thanks to individualized mechanics and genetics, pitchers fatigue differently.

If genetics and mechanics aren’t added to the data (and I have zero ideas about making that work), there’s no way this or the future studies he suggested can be applied as any kind of predictive model.

I appreciate the work he put into the study, but it’s about as practical as using pitch limits to prevent injury. If that worked, relievers would never hurt their arms.

The whole point of using this advanced set of PITCHf/x data is to compare pitchers to themselves when they are healthy. This takes into account things like mechanics because if a pitcher’s mechanics have been altered his release point and movement on his pitches will change. This means you can properly evaluate Sabathia to Sabathia and not to other pitchers who tire earlier like Woolner’s pitch count study.

Yes, as it currently stands this study is probably not going to be able to prevent many injuries but I strongly feel this is the right track. It already probably would do a better job than something like PAP to look at pitchers after the game to determine if they were really overworked. Also, something like PAP is completely useless for relievers. This metric, while not tested on relievers yet, has a chance of being useful. It doesn’t care about the pitcher’s pitch count so it doesn’t matter if they have thrown 20 or 120 pitches up to this point. All it cares about is how that pitcher is throwing compared to a healthy version of himself.

Based on the study as presented, the results were pretty mixed. My hypothesis would be that the link between the two is tenuous at best.

This current link is far from tenuous. Depending on where exactly you want to make the cut and define the injury zone the correlation is at least weakly correlated and often strongly correlated. It never will be 100% correlated however.

Such a study would need to:
-find all groups of 10 pitches that fit the fatigue criteria.
-relate these clusters with stays on the DL with consideration potentially given to number of days on the DL
-compare frequency of these clusters for "healthy" and "unhealthy" pitchers

If the frequency is just as high for the healthy group as it is for the unhealthy group, then different is not necessarily bad. It’s a lot of work, but I can’t think of any other way to prove the link.

I think you are overestimating the need for adding something like number of DL days. Yes it certainly might be somewhat helpful but what do you do with pitchers who get hurt and the season ends? How many days should be listed there? I think that issue alone is a good enough reason to not use something like time on the DL as a weighting factor.

The network output already says that the frequency is clearly not the same for the healthy group compared to the injured group. As the pitcher gets closer and closer to one his chances of ending up on the DL rise astronomically. If the frequency was the same then the injured pitchers curve would be exactly the same as the healthy pitcher curve only with much less statistics.

-joshkalk