Universal Measures of Player Value: GVT

Wouldn't it be nice to have a single "objective" value for each player, whether they're a goalie or a skater? Then we could evaluate trades and contracts "objectively". "Edmonton traded a value of 8 and only got back a value of 2. On paper, this looks like a bad trade." Or "Alex Steen has a value of 36. $6m a year for +36 is a great deal." Over the years there have been several attempts to create this type of universal currency for the value of players. So far, these currencies have come up short. I'm going to look at what goes into these metrics and then why they ultimately fail. These metrics have 7 basic aspects:

Inputs. Goals? Shots on goal? Corsi events?

Outputs. Estimated goals? Wins? Some other value?

Zero-point. What level of production equals 0? It's not always obvious.

Slope. If you have a USGA handicap, you're familiar with slope. If the input changes by 1, how much does the output change?

Symmetry. What's the value of offense versus defense?

Comparability. Is a goalie with value X really equal to a skater with value X?

Complexity. How hard is it to compute?

Tom Awad, one of the founding fathers of Hockey Analytics, developed GVT in 2009. Since it's Goals Versus Threshold, the output is obviously estimated Goals.

Goaltenders

The input for Goaltender GVT is Shots On Goal. GVT uses all shots on goal, but considers ES, PP, and PK separately. I've pointed out previously that goalies do not differ on the PK. Goalies probably do differ on the PP, but don't face enough shots for the difference to matter. In fairness to Tom, 2009 was before we knew much about the performance of goalies.

You might think that the Zero-point for goaltenders is defined in terms of the goalies who come up from the AHL for a game or two on an emergency basis, but in fact it's defined as 4% more goals yielded than the number yielded by an average goalie. If 0.920 ES is an average goalie, GVT sets ES threshold level at 0.917. If you look at the goalies who came up from the AHL for a game or two then went back down, their actual ESS% is about 0.897. That's a big difference.

Since goaltenders give up goals, you would think the slope is 1. For goaltender GVT, the slope is actually 0.75. A goalie who give up 40 goals fewer than threshold gets a GVT of +30.

Goaltender GVT also contains a small subjective factor for puck-handling skill and a small factor for shoot-out performance.

Skaters - OGVT

Skaters have two main parts to GVT, an Offensive part (OGVT) and a Defensive part (DGVT). Like goalies, skaters potentially get a small factor for shoot-out performance.

The inputs for Skater OGVT is Goals per Minute and Assists per Minute.

Like Goaltenders, Skater OGVT is defined in terms of an average player. Threshold level is set at 75% of the average Points per Minute. NHL teams average about 2.4 goals or 4.1 points per 60 minutes of 5v5 play. That sets threshold OGVT at about 3.0 points per 60 minutes of 5v5 play. Once again, if you look at AHL players who come up for a game or two, they actually produce about 0.4 points per 60 minutes of 5v5 play.

Once again the Slope isn't 1. Assuming 1.7 assists per goal, each goal is worth 0.47 towards OGVT. Each assist is worth 0.31 towards OGVT.

Skaters – DGVT

Calculating DGVT is incredibly complicated. Inputs include team shots against, a player's relative plus-minus, and each player's share of the 25% of goaltender performance that wasn't assigned to the goalie. The defensive performance of Defenseman is also given more weight than that of forwards. As a result the slope is not calculable directly. It's also not completely clear what the zero-point is.

Symmetry

The adage has it that "a goal prevented is as good as a goal scored." Of course that's not true when you look at NHL salaries and it's not not true in GVT either. If you look at all skaters in 2010-11 the sum of OGVT is +1347 and the sum of DGVT is +1113. Not as lopsided as I remembered but not symmetric either.

Comparability

The top skater in 2010-11 was Daniel Sedin with a GVT of +19.7. In 67 games he had 34 goals and 47 assists for 81 points. Just behind him was the 7th ranked goalie, Cam Ward, at +19.5. That was Cam's best year, but of the goalies with 20 or more games, he was the 8th best goalie in overall save percentage, and 19th in ESS%. I'm not sure that that equating his performance to Sedin's performance makes sense. It's also not clear that 67 games of Daniel Sedin would result in only 19.7 more goals created than 67 games of a career AHL player.

GVT was the first attempt at a universal currency for the value of players. It has problems with zero-points, slopes, symmetry, and comparability. Goaltender GVT has a subjective component. GVT is very hard to calculate. Three or four years ago GVT was in common discussion. These days it rarely gets used.