Predicting Injuries based on Past Injury Reports (1 Viewer)

Did a search with google and also in the Shark Pool--I was a little surprised to see little discussion on predicting injuries.

I've been bantering with a league-mate for a little bit about this. Most of our conversation has been about RBs. His new strategy is to draft players with clean bills of health. He doesn't want to struggle, as he puts it, with "injury prone" players and the weekly injury report and practice logs. And he thinks the past is a good indicator of future performance as far as health is concerned.

Personally, I think it's bollocks. All RBs get hurt and dinged over time. To the extent someone is actually injury prone, we're dealing with such small sample sizes that accurate predictions are nearly impossible. Have Mathews and McFadden suffered more injuries than league average RBs? Sure. Are they more likely to continue suffering more injuries than average? I'm not so sure, and I've never really read anything on the matter that's anything other than speculative and conclusory.

Which lead me to the interwebs, and I found this site: 4for4. It's a pay site, so much of the guy's information is hidden. But in that introductory article, he claims that all players that don't appear as questionable or doubtful on an injury report one year are much less likely to appear on the injury report the following year than the average NFL player.

Of course, you can't see his data without paying so who knows if he's comparing RBs to RBs. Or RBs to entire NFL rosters. Etc. I'd imagine that if you're not controlling for position, then kickers and punters could really skew the data since they get listed on the injury report and miss far fewer games than league average.

But if he is controlling for positions, then that's an interesting find: healthy starting RBs one year are more likely to be a healthy starting RB the next AND looking to the previous year's injury report is a good predictor of the upcoming season's injury report. I'm not sold without seeing more.

Do any of you use information like this? Successfully?

I think there are different forces at work. One is, is there a predictive methodology in predicting injury. As we all know, any player can get hurt at any time. Today is an excellent example. Jason Witten hasn't missed a game in 8 years, yet his status for Week 1 is now an issue.

There may be a higher probability of players having a second injury because the first one wasn't fully healed, ie a guy comes back too soon so he pulls his groin again. Or he had an ankle injury and was not as balanced running so he tweaked a knee. That may be more predictable, but not having access to the data and not being a doctor I can't speak to that one.

The best argument (meaning that we can look at the data and then conclude a trend) is to look at a specific player and then decipher how likely he is or is not likely to play in 16 games. For example, Ben Roethlisberger has only played in1 6 games once in his 8 year career. I think it would be reasonable to project him to only appear in 14 or 15 games, even though I have nothing to hang my hat on that he will get hurt or miss time. I can see the argument both ways on this one. He already said he has a shoulder problem but claims it will not effect his play. I guess we'll have to wait and see.

Thanks, David.

The closest analysis I've seen with football is with a rather crude measurement: games played. And prognosticators typically ding a player like Roethlesberger a game or two in their projections.

It makes sense. But I've never really seen anyone say WHY. And I've never seen any studies on it. The closest is the above linked website, which claims that labels on previous injury reports (such as probable, questionable, doubtful, and out) is an accurate predictor of future designation on injury reports.

This may be true. I think we can agree that the more trauma a body suffers, the less resilient it becomes. This is why athletes typically decline in performance after multiple injuries (and age).

But how accurate have your forecasts been on future games played? Have you ever ran analysis on that portion of your projections and compared players with injury histories against those with lesser injury histories?

I once did a study on whether or not you could predict injuries based on previous season information. What I found was, 1) past injury reports are just untrustworthy. Coaches play too many games to make a certain label mean anything. 2) For QB's I actually found a decent predictor of games played in season X by regressing age and # of pass attempts, and games played in the previous season. This made sense (to me) in that an older QB who played a bunch last year was subject to more hits than a younger qb who didn't play as much last year. Sort of a cumulative "punishment" would lead to less games played.

Was it perfect? No, but it was, imo, pretty decent. I don't see the same success with the other positions. Since then I've lost the study due to a failed harddrive, but while I think it was a decent predictor, I didn't see the worth in trying to recreate the study.

'Modog814 said:

I once did a study on whether or not you could predict injuries based on previous season information. What I found was, 1) past injury reports are just untrustworthy. Coaches play too many games to make a certain label mean anything. 2) For QB's I actually found a decent predictor of games played in season X by regressing age and # of pass attempts, and games played in the previous season. This made sense (to me) in that an older QB who played a bunch last year was subject to more hits than a younger qb who didn't play as much last year. Sort of a cumulative "punishment" would lead to less games played. Was it perfect? No, but it was, imo, pretty decent. I don't see the same success with the other positions. Since then I've lost the study due to a failed harddrive, but while I think it was a decent predictor, I didn't see the worth in trying to recreate the study.

Click to expand...

I was thinking the same thing about the injury reports. Seems a coach's motivation for labeling a player may have multiple factors. As for your QB findings, though: Did you find that older quarterbacks who threw a lot and played in all the games the previous season were the highest injury risks? If I'm reading you right, it's interesting that you'd find games played in previous seasons to have a negative correlation with games played in future seasons. This would lead to the bizarre conclusion that you're better off selecting players who missed games the previous year since they got their injuries out of the way...

I happen to be the league-mate he is referring to.

Let me add some clarity to the argument between mr strong and I. This entire discussion started when he emailed me a list of restricted RB free agents who will be available in our league starting next year (dynasty league). Instead of sharing his excitement, I said that besides Foster, I wasn't impressed. On this list were McFadden and Matthews, two players that have consistently suffered PROLONGED injuries. No one can argue that these two players have spent a good amount of time either out injured or with a Q/P next to their name. My point all along was that due to this history, their values were not as high as mr strong suggested.

I'm not going to predict injuries, but I will argue that it is obvious that certain players may not be as structurally sound (physically) as other players, leaving them vulnerable to "dinky" injuries, longer recoup times, or even a significant long-term injury. We see this in every sport, players that sustain injury after injury regardless of their conditioning. I believe McFadden is undoubtedly one of these players. He's a stud with enormous skills, but due to past injury history his value goes way down. Of course this is a small sample size and many will roll their eyes at the "injury prone" label, which is understandable considering how much it's overused. But I personally will avoid players that make deciding a weekly lineup more difficult. Matthews and McFadden have too many Qs, Ps, and Os next to their names for my taste. Does this mean that they will follow this trend this year....maybe.

If anyone's interested, I found a couple links.

Here a poster ran numbers from a study he did. Long and short of it: missing games the previous year was barely a factor for missing games in the next year. The poster's conclusion was that the minor difference (between players missing 0 or 1 games and players missing 8 or more games) was likely due to a player getting hurt near the end of a season and missing a game or two at the start of the next season.

The general consensus among the posters is that "injury prone" is generally something that is exaggerated by FFers.

http://forums.footballguys.com/forum/index.php?showtopic=323671&st=0

Also, in that thread, they link to a study by Doug Drinen in which he found that past injuries were such a minor factor in predicting future injuries that it was something he'd rather ignore in projecting players.

http://www.footballguys.com/drineniron.htm

'strong said:

'Modog814 said:

I once did a study on whether or not you could predict injuries based on previous season information. What I found was, 1) past injury reports are just untrustworthy. Coaches play too many games to make a certain label mean anything. 2) For QB's I actually found a decent predictor of games played in season X by regressing age and # of pass attempts, and games played in the previous season. This made sense (to me) in that an older QB who played a bunch last year was subject to more hits than a younger qb who didn't play as much last year. Sort of a cumulative "punishment" would lead to less games played. Was it perfect? No, but it was, imo, pretty decent. I don't see the same success with the other positions. Since then I've lost the study due to a failed harddrive, but while I think it was a decent predictor, I didn't see the worth in trying to recreate the study.

Click to expand...

I was thinking the same thing about the injury reports. Seems a coach's motivation for labeling a player may have multiple factors. As for your QB findings, though: Did you find that older quarterbacks who threw a lot and played in all the games the previous season were the highest injury risks? If I'm reading you right, it's interesting that you'd find games played in previous seasons to have a negative correlation with games played in future seasons. This would lead to the bizarre conclusion that you're better off selecting players who missed games the previous year since they got their injuries out of the way...

Click to expand...

Honestly, it's been a couple years since I even thought of the study, so forgive me if the details are a bit foggy in my head. But I think it took into account the 3 factors where advanced age was a negative thing, throwing a lot of passes the previous year was a negative, and missing games was a negative. So it essentally started a QB at 16 games and subtracted X amount for each year over age A, subtracted Y amount for each pass last year, and subtracted Z amount for each game misses last year because of injury.