Using FHM game rating (GR / GRO / GRD) for real hockey

mikelupu · 01-08-2017, 11:17 AM

Just a thought. I was wondering if it would be possible to use the calculations behind GR/GRO/GRD in the game to analyze actual player performance in NHL games.

There are a couple of methods being developed for quantifying the individual performance of a player in a game, e.g. Game Score, GVT (Hockey Prospectus | Numbers On Ice: Understanding GVT, Part 1) or THoR.

It would be interesting to see how GR/GRO/GRD compares to other metrics. According to the documentation the game rating takes into consideration the role defined for the player which might require a bit of tuning to the metric itself, but maybe there's still potential for a new measure.

Mike

JeffR · 01-09-2017, 05:17 PM

Wouldn't really be doable with real games, since the GR system uses a lot of stats that aren't tracked in real life (and can't be readily identified from things like game logs) - things like targets of failed passes, dump-in attempt results, the location to which clearing attempts make it to (or if they fail), etc. Since we can identify events like that (and things like player intentions) with perfect consistency and accuracy, we're at a huge advantage over the guys trying to develop real-life measures. But those kind of advanced stats do provide a useful check on our systems and would give the players extra feedback, so you'll probably see some of them in the game in the future in addition to the Corsi/Fenwick, PDO, etc. we already do.

And, like you said, the roles the player is being used in would also need to be identified, which could only be done in a very arbitrary way.

01-08-2017, 11:17 AM	#1
mikelupu Bat Boy Join Date: Nov 2016 Posts: 14	Using FHM game rating (GR / GRO / GRD) for real hockey Just a thought. I was wondering if it would be possible to use the calculations behind GR/GRO/GRD in the game to analyze actual player performance in NHL games. There are a couple of methods being developed for quantifying the individual performance of a player in a game, e.g. Game Score, GVT (Hockey Prospectus \| Numbers On Ice: Understanding GVT, Part 1) or THoR. It would be interesting to see how GR/GRO/GRD compares to other metrics. According to the documentation the game rating takes into consideration the role defined for the player which might require a bit of tuning to the metric itself, but maybe there's still potential for a new measure. Mike

01-09-2017, 05:17 PM	#2
JeffR FHM Producer Join Date: Apr 2002 Location: Kelowna, BC Posts: 16,623	Wouldn't really be doable with real games, since the GR system uses a lot of stats that aren't tracked in real life (and can't be readily identified from things like game logs) - things like targets of failed passes, dump-in attempt results, the location to which clearing attempts make it to (or if they fail), etc. Since we can identify events like that (and things like player intentions) with perfect consistency and accuracy, we're at a huge advantage over the guys trying to develop real-life measures. But those kind of advanced stats do provide a useful check on our systems and would give the players extra feedback, so you'll probably see some of them in the game in the future in addition to the Corsi/Fenwick, PDO, etc. we already do. And, like you said, the roles the player is being used in would also need to be identified, which could only be done in a very arbitrary way.