Quoting: CD282
xGF% doesn't single out the player any more than GF% does, that's silly. The common issue most take with these metrics are sample size and repeatability. I don't think either are an issue when looking at 10 years of data.
When looking at a player who has been on several different teams over the course of 10 years and you see a pattern of results, that says something about the player, not the team. In any case I used the Rel stat, which compares Barrie to his teammates. It's NOT a team stat in any way, shape or form.
And when a player outperforms his xGF% year after year after year, it's imprudent to assert that xGF% is a better representation of the player than GF%. That can't possibly be true.
I would read up a little on these stats you are using because I think we got little mixed up here. I will try to explain a little here.
xGF% does not isolate the individuals performance against his teammates. But it does smooth out things like puck luck, bad goaltending, linemates who cannot shoot etc. (and it actually works better with smaller samples).
But understand why each stat is used and computed. GF% is not much different from +/- (it’s just a ratio). But when models are made to predict the future, synthetic variables are often made to smooth out “noise” or “lumpy data”. The xGF% stat, for instance, is a combination of a number of this figures.
For example, player X has been playing on a line with McDavid and Draisaitl and Arizona wants to determine if he could preform similarly playing with Lawson Crouse and Travis Boyd. One would need variables to balance the items such as the shooting ability of each player.
When the smoothing variable is mixed with the original measurement, one gets the synthetic variable, which would provide a more accurate picture of how well Player X would play alongside the Arizona duo.
When single regressions are done to back test data, the new synthetic variables have a much higher r2 than the original figures (if running multiple regression analysis, best to use all the original inputs). In other words, the expected data-points actually have much more predictive value as a singular figure. Again, if one is running multiple regression models, there would be a ton of predictive inputs that could, cumulatively, give a better prediction. And they wouldn’t be equal parts in the model. But often the goal is to take large sets of complex information and reduce it all to the most recognizable form, which is why is easier to use the single datapoint approach.
So back to the team vs player discussion. If one wants to see how well a team has been playing, goal differential is a good way to do so. But if one wants to predict how well a player will perform in the future, the xGF figures are more useful. Since we are taking about trade values here, we would want to practice the latter. After all, a player’s value is based on how they will likely play going forward.
I hope that helps.
Devil et NoVaSpartan a aimé ceci.
Répondre
CapFriendly
CapFriendly