SoxProspects News
|
|
|
|
Legal
Forum Ground Rules
The views expressed by the members of this Forum do not necessarily reflect the views of SoxProspects, LLC.
© 2003-2024 SoxProspects, LLC
|
|
|
|
|
Forum Home | Search | My Profile | Messages | Members | Help |
Welcome Guest. Please Login or Register.
Regressing Volleyball statistics to the mean
|
|
Post by mgoetze on Feb 3, 2015 14:29:52 GMT -5
Well, Football is over and Baseball hasn't started yet, so I went to the logical conclusion and started watching Volleyball. Apparently there is no such thing as advanced stats in volleyball, so I decided to try and create some, and need a bit of help.
What I'm doing right now is watching games, recording which player served, which player received the serve, what the result of the serve/reception was (Ace, Overpass, Out-of-system attack, In-system attack or Service Error) and what the result of the rally was. By doing that for a lot of games I can calculate an average point value for each serve/reception result.
To get from there to a serving player's added value, I want to calculate the expected ratio of outcomes by averaging all the receptions from the team they are actually playing against and x receptions by an average team.
For a receiving player, I want to calculate the expected ratio from all the serves of the player actually serving, plus y serves by an average player.
How do I choose appropriate values for x and y?
|
|
|
|
Post by Chris Hatfield on Feb 3, 2015 15:20:42 GMT -5
Huh, so I'm actually probably your guy on this one - ran vball stats for a college women's team for two years.
What're the scoring goals in the games you're analyzing (for lack of a better term)? For example, the women's matches I was doing were best of five games, running scoring (not where you have to serve to score), first to 25 in games 1-4, first to 15 in game 5, win by 2.
Frankly, not sure that the nature of volleyball scoring allows for "per X" statistics, in any meaningful sense, other than per point played. It's like if baseball were best of nine innings, and innings 8 and 9 only had two outs.
That's if I'm getting your question, which I may not be. You basically want something like "per nine innings", right?
|
|
|
|
Post by pedroelgrande on Feb 3, 2015 15:53:40 GMT -5
The Dominican Women's Volleyball Team is great and we have the best libero in the world, Brenda Castillo. That's all I know about Volleyball.  |
|
|
|
Post by mgoetze on Feb 3, 2015 16:08:35 GMT -5
Yeah I think pretty much all games are best of 5 sets, first four sets to 25, last to 15, win by 2. I'm looking at international women's volleyball in particular. I don't believe that winning sets is predicted by anything other than winning rallies (i.e. points), unless someone can prove the opposite.  So if I did a rate stat I would do it per rally (or perhaps figure out how many points are scored in an average set, probably around 45). But to start I just want a simple counting stat, like Expected Points Added in Football. The Dominican Women are ranked 6th in the world currently so, yeah, they're not bad. |
|
|
|
Post by Chris Hatfield on Feb 3, 2015 18:54:22 GMT -5
Ah, I'm out of my league there then, I think.
I'll destroy all of you on inputting on volleyball StatCrew though. I barely needed a spotter by the time I left my job in sports info.
|
|
|
|
Post by mgoetze on Feb 4, 2015 5:24:58 GMT -5
Some preliminary results after a whopping sample size of 1 game. This is from the server's perspective: | Result | Points | Occurred | Avg | EPA | | Ace | 11 | 11 | 1.00 | +0.61 | | Overpass | 3 | 5 | 0.60 | +0.21 | | Out-of-system | 31 | 52 | 0.60 | +0.21 | | In-system | 28 | 111 | 0.25 | -0.13 | | Error | 0 | 10 | 0 | -0.39 | | Total | 73 | 189 | 0.39 | |
Observations: - A lot of clearcut cases and a significant difference in result, but there are more borderline cases than I thought where it takes quite some judgement whether the attack was in-system or out-of-system. - I wonder whether I should special-case the flukey serves where the ball rolls over the net. It's obviously much tougher for back-row receivers, and in some cases might be handled by the setter, who by definition cannot pass to an in-system attack, which is somewhat unfair to the setter. - There was one really flukey instance where what should have been a service error turned into an ace by glancing off the receiver's foot. Oh well, sample size should weed that out in the long run but it does show the importance of regressing to the mean eventually... |
|
|
|
Post by plantierforever on Feb 23, 2015 21:26:42 GMT -5
mgoetze, former collegiate volleyball player here. Not sure where you're going with your first pass, but I applaud the effort. If I can make a suggestion that I am too lazy to do any work on, I'd suggest trying to calculate the average value added for each particular skill on the team. For example, compare % of points won off the serve of player X versus the % of points won by the team as a whole (ex player X) when serving. Run enough games, and you'll see pretty handily which players are the more effective servers. Of course, winning points when you are serving is also a function of who is front row during that rotation, so I can see how filtering out the effect of a single player will become difficult. In VB, as with many other team sports, you don't have a mano-a-mano confrontation like pitcher v batter where you can measure individual impact more directly.
Another stat I'd like to see is % of points won when player X is front row versus % of points won by the team generally. I think you are onto something with filtering out flukey stuff, like services errors, aces, and otherwise goofy points that aren't decided by anyone's skill. Good luck.
|
|
|
So if I did a rate stat I would do it per rally (or perhaps figure out how many points are scored in an average set, probably around 45). But to start I just want a simple counting stat, like Expected Points Added in Football.