Based on the R2 above, GAR has the ability to predict roughly 72% of how a team will end up in the standings (retroactively).
Retroactive prediction, huh?
Those are some pretty fucking important outliers on the left side of that graph (and to a lesser extent on the far right) - they could be driving all of the results since without them the slope of the line appears much flatter (to my eye).
If Keith, Sharp, Seabrook and Hossa are all overpaid (and Kane is adequately compensated) if appears to me that perhaps there may be somewhat of a problem. Similar to Burtch's problem with dCorsi weighing too negatively on D with big minutes.
As an aside—why bother putting different colours and symbols if you're going to put the players' names on the graph?
If you combine the result that nearly all of the players on a bad team are overpaid and generating negative value while most of the players on a good team are overpaid, I'm left wondering what the fuck good is this tool.
However, what this result tells me is that a big part of effectively managing a roster will come down to simply not overpaying players. It is extremely hard to find a player that can be signed for less than he is worth
Or you know, that perhaps you forgot to control for contract status in determining whether a guy is overpaid or not (always be thinking of the counterfactual). The real implication is that unless a guy is elite I should comprise my roster entirely of ELCs and never give RFAs more than the required raise (if they want more then I trade 'em).
The point about ensuring that your highest paid guys are "worth their money" is a fair one and certainly one that's been brought up by less quanty people in more accessible language/format. Is this GAR the tool to help determine what guys are worth? Nothing I've seen so far leads me to believe that it's there yet (especially for defensemen and goaltenders).
"If you torture the data long enough, it will confess to anything." - Ronald Coase
"[...]all models are wrong, some are useful." - George E. P. Box