The regression predicting 2012 HR needs 2011 HR as an independent variable ... although I'd have thought it rather obvious you need HR/PA not raw HR to adjust for playing time changes. All of that before you get into minor-league stats, etc. Anyway, the baseline for any such analysis has to be "does this indicator add anything over the previous X seasons' performance?" Think of y_t regressed on y_(t-1) as the Marcel equivalent and then see if you can add anything that substantially improves on Marcel.

As to HR max distance ... guys who hit more HR by definition are "sampling" more from the HR distance distribution. That is a guy who hits 40 HR is more likely have hit one of 425 feet than a guy who hits 3 even if they have the same "power".

Isn't home run distance one of the least accurate stats, or are people actually calculating them correctly now? For instance, if a ball hits the foot of the wall, or the top of the wall I thought they say the ball went the same distance even though obviously if the wall was not there, one of them would have gone farther. If the stat tracked where the ball would have landed if the wall or stands were not there, it may have more predictive power.

I don't believe that's correct Scott.

@dr. scott - This is how the home run distances that were used in the study are calculated.

@Walt Davis - Thanks a lot for the feedback, I appreciate it. I did realize that HR/PA would be a better metric to get around injuries/playing time differences (and mentioned this in the comments afterward). When I follow up on this I plan to look at HR/PA. As far as whether it can be useful or not, I will look at a multiple regression of Year N HR/PA and Year N Max HR Distance against Year N+1 HR/PA and see if the Max HR Distance is a significant factor or not. I may still look at raw HR numbers as well out of curiosity. Your sampling point is interesting and well made also. I'll see how this ends up looking in a follow-up study.