Not doubting your data, but having trouble understanding it.
I would have thought that it would be the total hits that would correlate better? If Buff plays 30 minutes, and is involved in 10 hits (1 every 3 minutes), and Petan gets on for 8 minutes and is involved in 4 hits (one every 2 minutes), I would expect, all other things being equal, that the player with the most total hits, Buff, would be more susceptible, not the one with the higher hits per minute but lower total hits.
Also, I would have thought that previous man games lost to injury and also age would also be predictors of future man games lost. Any truth to this?
Sorry, age and man games lost are also.
I wasn't thinking of those, as I was foolishly thinking of that as a given, but it should not be.
As to the ice time thing... I don't know why per ice works better than aggregate, but they both work.
It may be due to the fact that ice time is not always consistent from one year to the next? Aggregate still works though, as the two are interlinked.
Interestingly enough, those same players tend to also have faster WAR/82GP falls after their peek. Adjusting for 82GP removes the impact of man games lost, but it could be that those players perform worse in the games they do play due to playing through injuries.
In the end I will say this:
The extent of the research here is very small and not very specialized.
There's a lot of confounding variables with injuries and recovery.
I could definitely see the extremes (small and large) showing different results with better research, especially when accounting for other variables.