HFBoards Calculating GVT Question
 Register FAQ Members List Search Today's Posts Mark Forums Read
 Mobile Hockey's Future Become a Sponsor Site Rules Support Forum vBookie Page 2
 By The Numbers Hockey Analytics... the Final Frontier. Explore strange new worlds, to seek out new algorithms, to boldly go where no one has gone before.

# Calculating GVT Question

 07-06-2013, 12:01 AM #1 Starwood Rookie User   Join Date: Jul 2013 Posts: 1 vCash: 500 Calculating GVT Question Hey guys, I'm new around these parts. Always loved statistics, always loved math, and always loved sports, now I'm trying to piece all of that together with some of my free time. I'm currently messing around with calculating GVT for an NHL 13 league I'm playing in, since it seems to be fairly simple as far as not needed too many in depth statistics to calculate. I'm using the link below: http://hockeyprospectus.com/article.php?articleid=236 I had a question: When I get to calculating the offensive threshold for forwards and for defensemen: TOCf = [sum (Gf x GVf + Af x AVf)] / [sum (MPf)] * OTV TOCd = [sum (Gd x GVd + Ad x AVd)] / [sum (MPd)] * OTV I'm a little bit lost as to what I'm supposed to be using for GVf, AVf, GVd, and AVd. The summation of all of the forwards goal values (assist values) for that team? For the whole league? The same question would fall into the MPf as well. Thanks in advance guys. You have some awesome threads around here, just spent the past hour or so stalking the site
07-15-2013, 11:08 PM
#2
Registered User

Join Date: Jan 2006
Location: bohemia
Country:
Posts: 4,056
vCash: 500
Quote:
 Originally Posted by Starwood I'm a little bit lost as to what I'm supposed to be using for GVf, AVf, GVd, and AVd. The summation of all of the forwards goal values (assist values) for that team? For the whole league? The same question would fall into the MPf as well.
Yes, I believe those values are for the team only.

07-27-2013, 09:13 AM
#3
Cunneen
Registered User

Join Date: May 2013
Posts: 94
vCash: 500
Quote:
 Originally Posted by Starwood Hey guys, I'm new around these parts. Always loved statistics, always loved math, and always loved sports, now I'm trying to piece all of that together with some of my free time. I'm currently messing around with calculating GVT for an NHL 13 league I'm playing in, since it seems to be fairly simple as far as not needed too many in depth statistics to calculate. I'm using the link below: http://hockeyprospectus.com/article.php?articleid=236 I had a question: When I get to calculating the offensive threshold for forwards and for defensemen: TOCf = [sum (Gf x GVf + Af x AVf)] / [sum (MPf)] * OTV TOCd = [sum (Gd x GVd + Ad x AVd)] / [sum (MPd)] * OTV I'm a little bit lost as to what I'm supposed to be using for GVf, AVf, GVd, and AVd. The summation of all of the forwards goal values (assist values) for that team? For the whole league? The same question would fall into the MPf as well. Thanks in advance guys. You have some awesome threads around here, just spent the past hour or so stalking the site
I'd stay away from GVT. it is an awful statistic.

07-29-2013, 04:05 AM
#4
Registered User

Join Date: Oct 2011
Country:
Posts: 36
vCash: 500
Quote:
 Originally Posted by Cunneen I'd stay away from GVT. it is an awful statistic.
Care to elaborate?

07-29-2013, 08:19 PM
#5
Cunneen
Registered User

Join Date: May 2013
Posts: 94
vCash: 500
Quote:
 Originally Posted by Sixbladeknife Care to elaborate?
haha, yea sure. Sorry, I've argued this point so many times with guys like Rob Vollman I get tired of it, but one more time is ok.

I have several problems with GVT.

The first is that it has several seemingly arbitrary values in the calculations. Here are some examples.

First, goals are given the value of 1.5 assists. Why? I don't know, it's a number that was chosen without any empirical backing.

Second, defensemen are assumed to have 1.5 times more defensive responsibility as forwards. Again, where did this number come from? It's completely random/arbitrary.

Third, defensemen are given 25% of the credit for every save. This assumes the existance of team shot quality over the course of the season, which again and again has been proven to have a minimal affect at best. Again, an arbitrary #.

I also have a problem with the fact that GVT incorporates plus minus (adjusted) into the calculations. That is a terrible way to evaluate defensive performance. Also, when evaluating defensive performance, why does GVT use Team shots against rather than on ice shots against. Another example of a terrible method.

I have a few other issues, but the above reasons are enough to show that GVT is a terrible statistic which should never be used in player analysis until it's many issues are solved.

07-31-2013, 04:23 AM
#6
Blue Blooded
Registered User

Join Date: Oct 2010
Location: Stockholm
Country:
Posts: 2,810
vCash: 500
Quote:
 Originally Posted by Cunneen haha, yea sure. Sorry, I've argued this point so many times with guys like Rob Vollman I get tired of it, but one more time is ok. I have several problems with GVT. The first is that it has several seemingly arbitrary values in the calculations. Here are some examples. First, goals are given the value of 1.5 assists. Why? I don't know, it's a number that was chosen without any empirical backing. Second, defensemen are assumed to have 1.5 times more defensive responsibility as forwards. Again, where did this number come from? It's completely random/arbitrary. Third, defensemen are given 25% of the credit for every save. This assumes the existance of team shot quality over the course of the season, which again and again has been proven to have a minimal affect at best. Again, an arbitrary #. I also have a problem with the fact that GVT incorporates plus minus (adjusted) into the calculations. That is a terrible way to evaluate defensive performance. Also, when evaluating defensive performance, why does GVT use Team shots against rather than on ice shots against. Another example of a terrible method. I have a few other issues, but the above reasons are enough to show that GVT is a terrible statistic which should never be used in player analysis until it's many issues are solved.
What is your opinion on HA's HARO/HARD/HART ratings?

My summer brain doesn't have the energy to properly analyze the methodology, but at a glance it looks decent. It may try to do too much though.

In practice the results seem quite decent, although it seems like playing on a good team still seems to help your value quite a bit - which is something I think the metric tries to compensate for.

07-31-2013, 07:49 PM
#7
Cunneen
Registered User

Join Date: May 2013
Posts: 94
vCash: 500
Quote:
 Originally Posted by Blue Blooded What is your opinion on HA's HARO/HARD/HART ratings? My summer brain doesn't have the energy to properly analyze the methodology, but at a glance it looks decent. It may try to do too much though. In practice the results seem quite decent, although it seems like playing on a good team still seems to help your value quite a bit - which is something I think the metric tries to compensate for.
I'll be perfectly honest, I have not taken the time to read and understand Johnson's HARO stats. There's only so much time in the day haha.

08-10-2013, 06:20 PM
#8
Blue Blooded
Registered User

Join Date: Oct 2010
Location: Stockholm
Country:
Posts: 2,810
vCash: 500
Quote:
 Originally Posted by Cunneen I'll be perfectly honest, I have not taken the time to read and understand Johnson's HARO stats. There's only so much time in the day haha.
I finally went through it and this is my synopsis:

It is basically based on WOWYs and AYNAYs comparing the players actual GF/GA, SF/SA, FF/FA, CF/CA, to his expected one based on his teammates and matchups.

It is carefully avoiding a player influencing his QoT and QoC and then calculates percentages in several iterations where first the player's ratings are done and then his QoT and QoC. The player's ratings are then updated, and then the QoT and QoC in another iteration and so forth.

Problem #1: The number of iterations seems arbitrary and is not listed. I'm sure the number is the same for all players and likely sufficient, but it shouldn't be very hard to calculate the number of iterations required to reach the convergence point.

After the player's HARO and HARD (difference to expectations measured in %) are calculated, the HART rating is calculated to act as a measure of the player's impact.

HART isn't a relative percentage however, like GF% or CF% but a mean between the HARO and the HARD.

At first I thought this would give misleading results, but it actually gives the correct numbers in terms of absolute impact. (i.e. a player with HART 100 will have 10% higher +/-60 than a player with HART 90).

Problem #2: It does churn out some seemingly inexplicable results at times though.

Joe Thornton has a Corsi20 differential of +4.526 over the past two years. His teammates has a -0.282 over the same period. His opposition +0.377. His HART comes out at +9.4.

T.J. Oshie had +2.657. His teammates +2.57. His opposition +0.317. His HART rating 10.3.

There is NOTHING in the underlying numbers that suggest that Oshie should be in Thornton's stratosphere, yet alone ahead of him.

There are several others that seem weirdly out of place, e.g. Patrice Bergeron being behind four LA players despite putting up similar numbers on a worse team.

Something seems to go wrong somewhere in the iteration process.

Conclusion: Despite the process seeming correct, the end product does not seem to match the data it is based on.

09-22-2013, 07:30 AM
#9
mibanov
Registered User

Join Date: Sep 2013
Posts: 1
vCash: 500
Quote:
 Originally Posted by Starwood Hey guys, I'm new around these parts. Always loved statistics, always loved math, and always loved sports, now I'm trying to piece all of that together with some of my free time. I'm currently messing around with calculating GVT for an NHL 13 league I'm playing in, since it seems to be fairly simple as far as not needed too many in depth statistics to calculate. I'm using the link below: http://hockeyprospectus.com/article.php?articleid=236 I had a question: When I get to calculating the offensive threshold for forwards and for defensemen: TOCf = [sum (Gf x GVf + Af x AVf)] / [sum (MPf)] * OTV TOCd = [sum (Gd x GVd + Ad x AVd)] / [sum (MPd)] * OTV I'm a little bit lost as to what I'm supposed to be using for GVf, AVf, GVd, and AVd. The summation of all of the forwards goal values (assist values) for that team? For the whole league? The same question would fall into the MPf as well. Thanks in advance guys. You have some awesome threads around here, just spent the past hour or so stalking the site
I'm also a bit confused with the formulas in the link above. There even seems to be some typos etc. I was wondering whether anyone happens to know if there is a corrected or more recent guide into calculating GVT. Would, for example, subscribing to Hockey Prospectus give access to a more recent version?

09-25-2013, 07:36 AM
#10
wgknestrick
Registered User

Join Date: Aug 2012
Posts: 2,278
vCash: 500
Quote:
 Originally Posted by Cunneen haha, yea sure. Sorry, I've argued this point so many times with guys like Rob Vollman I get tired of it, but one more time is ok. I have several problems with GVT. The first is that it has several seemingly arbitrary values in the calculations. Here are some examples. First, goals are given the value of 1.5 assists. Why? I don't know, it's a number that was chosen without any empirical backing. Second, defensemen are assumed to have 1.5 times more defensive responsibility as forwards. Again, where did this number come from? It's completely random/arbitrary. Third, defensemen are given 25% of the credit for every save. This assumes the existance of team shot quality over the course of the season, which again and again has been proven to have a minimal affect at best. Again, an arbitrary #. I also have a problem with the fact that GVT incorporates plus minus (adjusted) into the calculations. That is a terrible way to evaluate defensive performance. Also, when evaluating defensive performance, why does GVT use Team shots against rather than on ice shots against. Another example of a terrible method. I have a few other issues, but the above reasons are enough to show that GVT is a terrible statistic which should never be used in player analysis until it's many issues are solved.

The 1.5:1 ratio of value stems from 1.5 assists for every NHL goal scored.

My opinion is this:

There has been no better attempt to unite all the areas in hockey into a universal, single metric. Is it perfect? far from it IMO. Should it never be used? I don't buy that for one second.

Forwards GVT is just about perfect. It was the easiest to get right with the current data recorded in the NHL. I don't think defensive forwards get enough credit, but that will come when the refinements below are made.

Defender and Goalie GVT is where things get murky. If you ever graph the GVT populations of defenders vs forwards vs goalies you will see what I mean. If GVT was "perfect" (maybe good is a better word), I would expect GVTs to follow a similar bell shaped curve for each position.

That is certainly not what happens. The goaltenders have HUGE variance (max of around 38GVT and min of around -15GVT) between them when you look at the tails of the population (extreme highs and lows). Defenders, however have very little that separate them in terms of GVT (16GVT to about -1 GVT). If you look at the GVT leader each year on defense, he only "equals" about a 2 line forward.

If I were to assume GVT was close to accurate evaluation of "worth", then I would expect the market to mirror this in salaries. We all know that is not the case. Elite goaltenders don't make far and away the most money in the league, while elite defenders don't make 2nd forward money.

Elite players all make around 7-8mil/year so that tells me that GVT should be close between positions at each extreme.

What needs to happen is to refine the "allocation" of defensive value between the defenders and goaltenders so that populations are as similar as possible. Elite talent should have near the same GVT (position independent). Do you guys really think that BOS has had far and away the best crop of goaltenders in the league (over the last 5 years) or that they are propped up by elite defensive corps (ie Chara and crew)?

I just don't like people discrediting something entirely just because it isn't perfect. Show me your 1 number system that is better, or do something constructive to right the wrong. It is very easy to judge someone else's work.

Forum Jump