Revisiting 2011 SP DL Projections
Last off season, I looked at the chances of a SP going on the DL. I have finally had time to go back and look at how my predictions fared.
The predictions used logistic regression to find the percentage chance that a pitcher would end up on the DL. I used age, games started in the previous 3 years and how many of the previous 3 years did the pitcher go on the DL. The equation I ended up with was:
1/(1+e^(-z))
where:
z = (.2209)(Years with Trips to DL)+(-0.0040)(GS in last 3 year)+(0.0509)(Age in previous season)-1.7692
Using the equation, I projected the chance that a starter would go on the DL and here is a list of those projections.
To see how the projections stood up, I examined the 25 most and least likely players to end up on the DL. I look to see if they were on the DL in 2011 and what was their chance of going on the DL. I added up the individual percentage chances and figured out the percentage of pitchers that actually went on the DL. I removed any pitchers that did no pitch in 2011 because of retirement like Andy Pettitte.
The 25 players most likely to end up on the DL values ranged from 43.5% (Roy Halladay) to 55.1% (Daisuke Matsuzaka). Of the 25 players, 12 went on the DL in 2011, or 48%. The average percentage chance predicted that 12.2 players or 49% would make the DL. The model held pretty good. This group of players had an average age of 31 years old, pitched in only 65 games over the past 3 season and went on the DL 1.75 times over that time frame.
The pitchers with less of chance to end up the DL ranged in value from 27.3% (Clayton Kershaw) to 32.7% (Paul Maholm). Of the 25 pitchers, 9 went on the DL in 2011, or 36%. The average percentage chance predicted that 7.75 pitchers or 31% would end up on the DL. The final prediction for these pitchers was not as good as the DL prone pitchers, but close. The pitcher’s average age was 25 years old, pitched in 79 games in the previous 3 season and has never been on the DL.
The key is that some pitchers are historically more likely to go on the DL than others. The percentage chance can be to the tune of almost 30% ( Daisuke Matsuzaka at 55% vs. Clayton Kershaw at 27%). An older, DL prone pitching staff will need to have several more options ready when its starters go on the DL than a staff of young healthier pitchers.
Jeff, one of the authors of the fantasy baseball guide,The Process, writes for RotoGraphs, The Hardball Times, Rotowire, Baseball America, and BaseballHQ. He has been nominated for two SABR Analytics Research Award for Contemporary Analysis and won it in 2013 in tandem with Bill Petti. He has won four FSWA Awards including on for his Mining the News series. He's won Tout Wars three times, LABR twice, and got his first NFBC Main Event win in 2021. Follow him on Twitter @jeffwzimmerman.
In the formula, do DL trips and games started in the minors count?
No on both counts. I could do minor league starts, but didn’t use them in the original regression. I used starts to help find starters that missed starts because of injury, but didn’t do on DL
There is no current MiLB injury database. The MLB is a mess to begin with (I am still trying to get the 2012 one figured out and have worked too much on it).
Brilliant last sentence.
So half od the SP’s you predicted went on the DL and half did not?? Did you just flip a coin??
Did you not read the article or do you just not understand it?
My point is this statistical analysis of who will be hurt and who will not based on if they were hurt or not in the past is kind of a stretch in any sense. Pitchers that undergo Ulnar collateral ligament (UCL) reconstruction surgery have often had long productive careers post procedure. Some players are just injury prone or just plain unlucky.
There’s a 50% chance of rain every day. It either rains or it doesn’t, right?
“The 25 players MOST LIKELY to end up on the DL values ranged from 43.5% (Roy Halladay) to 55.1% (Daisuke Matsuzaka). Of the 25 players, 12 went on the DL in 2011, or 48%. The average percentage chance PREDICTED THAT 12.2 PLAYERS or 49% would make the DL.” (emphasis mine)
The model predicted that 12.2 players would hit the DL. In actuality, 12 players hit the DL. Sounds like a damn good model to me.
I was trying to be humble.
Yes, and the probability inherent in that luck can be measured. Jeff measured it, and found reasonably strong and statistically significant relationship between certain variables and the probability in question. That’s what statistics is.
This really seems like a situation where a nonlinear model would do better.
Also I imagine you could do better at predicting # days on DL for a whole team’s SPs. That should cut down on random variation and also highlight any teams that have to potentially watch out.
When isn’t a nonlinear model better? You need some seriously simple data for that to be the case.
As it stands, Jeff used a logit model, so…
I do like the idea of taking a closer look at “at risk” and “relatively safe” teams heading into next season rather than just focusing on players alone.
What do you mean by a nonlinear model? Can you explain or give an example?
A time series model would work wonders here. Days spent on DL in each of the past 5 (or however many you want) seasons, games started in each of the past 5 seasons, and current age.
Another suggestion: you could add in dummy variables for TJS, rotator cuff surgery, etc. This way you’re allowing for differences between injuries that have long-lasting effects, like a rotator cuff, versus surgeries that pitchers generally rebound from and have long, healthy careers afterwards, like TJS. Obviously, not all days on the DL are created equal – TJS will give you a ton of DL days and severely cut down on your GS, but it doesn’t nearly have the predictive value as to whether a pitcher will hit the DL 3-5 years later as many other injuries (or the aging process) do.
There is a huge study on TJS in the works (the words need to be put around the numbers). I helped the person with it and have actually seen the results. I am actually looking at expanding it myself already. There will be some good additional information from it once it is released.
Have you looked at the data and tested whether adding more years to your data would yield more or less predictive results?
Would a pitcher that went on the DL for Tommy John 4 years ago be more likely to go on the DL? If the answer is yes then it might make sense to creating a larger range of years from which to draw your data. (Clearly adding more years would eliminate some pitchers that don’t have as much experience and as such it might not be useful.)
In other words I’m asking whether injuries from long ago increase the odds of future injuries or does the correlation fade with increased time?
I would assume no because the longer ago the injury happened, the more time it has had to heal. More recent injuries have a better chance of happening again or causing other injuries due to a pitcher’s attempt to compensate it.
This is hilarious.
That is some crazy math right there
Older pitchers tend to get hurt more than younger pitchers and guys who were on the DL in the past tend to go on it again are the conclusions of the study? A lot of time was spent on something that was obvious at the beginning.
Agree, I just wanted to find how much of difference it made.
So, Jeff, you using this against us in the upcoming draft?