|
Post by uvavolleystats on Apr 17, 2013 9:01:23 GMT -5
|
|
|
Post by meanmug on Apr 17, 2013 10:04:37 GMT -5
Option 1 is (IMO) by far the best method for evaluating passing. The other options are just doing math for the sake of math and aren't actually telling you anything useful.
The only issue is that he should be using historical averages from your league, not just from one season's worth of one team. That is what is creating the noise. Additionally, if you want to create a "value of replacement" type of statistic (which is what he is getting into with options 3 and 4), he should compare to historical averages for passing numbers in the league.
And I would recommend making a conversion from FBSO probability to total SO probability, which will allow you to translate into points so that passing quality can be directly compared to other skills.
|
|
|
Post by alantech on Apr 17, 2013 10:16:47 GMT -5
I already have a website called VolleyMetrics on advanced volleyball statistics: volleymetrics.blogspot.com/ . I've operated the site since 2007. I'm pleased to have additional sports analysts looking at volleyball, the more the merrier. However, can you please change your user name to avoid confusion with my site? Thanks!
|
|
|
Post by uvavolleystats on Apr 17, 2013 10:19:59 GMT -5
Option 1 is (IMO) by far the best method for evaluating passing. The other options are just doing math for the sake of math and aren't actually telling you anything useful. The only issue is that he should be using historical averages from your league, not just from one season's worth of one team. That is what is creating the noise. Additionally, if you want to create a "value of replacement" type of statistic (which is what he is getting into with options 3 and 4), he should compare to historical averages for passing numbers in the league. And I would recommend making a conversion from FBSO probability to total SO probability, which will allow you to translate into points so that passing quality can be directly compared to other skills. I agree that we should be using a larger data set for this study. I am in the process of compiling numbers for the entire league from the 2012 season. Also, these numbers are FBSO probabilities. Sorry that this wasn't clarified in the article).
|
|
|
Post by n00b on Apr 17, 2013 11:46:06 GMT -5
Option 1 is (IMO) by far the best method for evaluating passing. The other options are just doing math for the sake of math and aren't actually telling you anything useful. The only issue is that he should be using historical averages from your league, not just from one season's worth of one team. That is what is creating the noise. Additionally, if you want to create a "value of replacement" type of statistic (which is what he is getting into with options 3 and 4), he should compare to historical averages for passing numbers in the league. And I would recommend making a conversion from FBSO probability to total SO probability, which will allow you to translate into points so that passing quality can be directly compared to other skills. 1) If he's trying to better his team, why go to league averages? For a team that is extremely physical with poor defense, the importance of getting a 3 pass is huge. If my team is undersized and wins by defending and playing long rallies, getting 3 passes isn't as important as just minimizing 0 passes. Now if you want a generic stat for everybody to use, I would agree with you. 2) I agree that it should use the overall odds of winning the rally, not of getting a first ball kill. Finally, what is your reasoning for saying there is a lot of noise? Could that just mean that 1, 2, and 3 passes aren't really that different in terms of importance?
|
|
|
Post by uvavolleystats on Apr 17, 2013 11:58:04 GMT -5
Option 1 is (IMO) by far the best method for evaluating passing. The other options are just doing math for the sake of math and aren't actually telling you anything useful. The only issue is that he should be using historical averages from your league, not just from one season's worth of one team. That is what is creating the noise. Additionally, if you want to create a "value of replacement" type of statistic (which is what he is getting into with options 3 and 4), he should compare to historical averages for passing numbers in the league. And I would recommend making a conversion from FBSO probability to total SO probability, which will allow you to translate into points so that passing quality can be directly compared to other skills. 1) If he's trying to better his team, why go to league averages? For a team that is extremely physical with poor defense, the importance of getting a 3 pass is huge. If my team is undersized and wins by defending and playing long rallies, getting 3 passes isn't as important as just minimizing 0 passes. Now if you want a generic stat for everybody to use, I would agree with you. 2) I agree that it should use the overall odds of winning the rally, not of getting a first ball kill. Finally, what is your reasoning for saying there is a lot of noise? Could that just mean that 1, 2, and 3 passes aren't really that different in terms of importance? 1) Good point-- But this would also mean that all of my data from last season is irrelevant for the upcoming season and even for this Spring since we have a totally different group of athletes on the court. I think you can ascertain general trends on the whole from league totals and team specific efficiencies from your own totals. 2) I think the pass quality is more telling of success in FBSO opportunities than overall opportunities because it directly correlates to the FBSO attack. I am working on doing the same thing with transition attacking based on dig quality. 3) The noise would come from other factors that would be used to properly rate and balance the probability based on the pass quality. Is using kill % alone the best way? I also look at the error % in each situation and calculate that probability. Is that differential more important than scoring probability? Also, what else should be figured in? The quality and speed of the set? The hitters in each rotation? The opponent's blocking and digging efficiency? And to what factor should all of this be figured in, if at all?
|
|
|
Post by n00b on Apr 17, 2013 12:23:21 GMT -5
2) I think the pass quality is more telling of success in FBSO opportunities than overall opportunities because it directly correlates to the FBSO attack. I am working on doing the same thing with transition attacking based on dig quality. While you're probably right that FBSO is more highly correlated, why is that what we care about? Isn't what we're trying to figure out how and why we win rallies? I would think only taking FBSO into consideration underestimates the harm of a 0 pass (since a 0 pass is equivalent to a 3-pass attack that gets dug). I think this is the power of large sample sizes (and you're right, one team for one season might not be enough, I'm not sure). As your sample gets bigger and bigger, these things tend to even out. Over 2500 sets, you should get a realistic proportion of botched sets as well as perfect sets. Same goes for opponents defending.
|
|
|
Post by uvavolleystats on Apr 17, 2013 13:31:00 GMT -5
2) I think the pass quality is more telling of success in FBSO opportunities than overall opportunities because it directly correlates to the FBSO attack. I am working on doing the same thing with transition attacking based on dig quality. While you're probably right that FBSO is more highly correlated, why is that what we care about? Isn't what we're trying to figure out how and why we win rallies? I would think only taking FBSO into consideration underestimates the harm of a 0 pass (since a 0 pass is equivalent to a 3-pass attack that gets dug). This is a good point. It all depends on what your goal is. In this instance, I am simply trying to figure out which of my passers gives me the best opportunity to win a rally. Figuring out how and why we win rallies is the end game of this study but I am no where near the point of even thinking about that yet. Based on this formula, the negative impact of a 0-pass is included in the calculations. The formula is pass score probability= (3pass%*.452)+(2pass%*.339)+(1pass%*.210)+ (0pass%*0)
|
|
|
Post by meanmug on Apr 17, 2013 13:31:40 GMT -5
(1) You want to use league averages BECAUSE your team might be super-physical or a smaller defensive team. If you only base off what your team does you will be measuring factors other than passing. By factoring in comparison to a more "league average" baseline, you can measure whether you are siding out well due to quality passing or quality hitting.
(2) There's a lot of noise when you base off 1 team (or 1 season from 1 team, etc) because the sample size is too small. Or you pick a team (or group of 5 matches or whatever) that passed really well or really poorly or etc etc. I mean, even just picking 10 matches from any team in any given season will still get you in the ballpark, but obviously if you can get more like 100 matches from various teams then you can be most accurate.
(3) I have gone back and forth between weighting the passing off FBSO or off SO, and I have determined that the most accurate way to is weight the passing by comparing to historical FBSO numbers (or more specifically, FBSO efficiency) and then apply that as a conversion to overall SO... unless you have a ton of data to work with which is often not the case. SO numbers take longer to stabilize than FBSO.
Real-life example: I once pulled about 20 matches (basically imagine a 7-team round robin) and found that FBSO was about 10% higher off 4-passes than 3-passes but total SO was only 2% higher. This is the noise that I am talking about and can lead to faulty conclusions. When I expanded the sample size, the total SO numbers matched up more with the difference in FBSO.
I would love to hear different opinions on this though as I am still working toward solutions that stabilize faster and are more accurate.
(4) In terms of "what other factors" should be factored in, my feeling right now is that you are best with measuring hitting efficiency by each quality of pass without factoring in some of the other things. I think that incorporating more variables can sometimes kill your sample size, which can increase your noise, even though you intended the opposite. Again though, I think this is something to play around with and see.
(5) n00b, you make a good point about FBSO underestimating the harm of a 0-pass (which is also one of the major flaws of 3/2/1 rating), which is why you need to adjust separately for passing errors and overpasses as well. And again, if you have enough data you can just base them straight off overall sideout, but you'd be surprised at how noisy that can be.
|
|
|
Post by meanmug on Apr 17, 2013 13:36:31 GMT -5
The formula is pass score probability= (3pass%*.452)+(2pass%*.339)+(1pass%*.210)+ (0pass%*0)I think you will find a bit higher correlation to SO% (and indeed: winning) if you go something like: 3pass% * 0.65 2pass% * 0.59 1pass% * 0.50 0pass% * 0.00 That is a conversion to expected overall S0% based on those FBSO #s.
|
|
|
Post by uvavolleystats on Apr 17, 2013 13:43:49 GMT -5
The formula is pass score probability= (3pass%*.452)+(2pass%*.339)+(1pass%*.210)+ (0pass%*0)I think you will find a bit higher correlation to SO% (and indeed: winning) if you go something like: 3pass% * 0.65 2pass% * 0.59 1pass% * 0.50 0pass% * 0.00 That is a conversion to expected overall S0% based on those FBSO #s. To calculate the SO% probability, wouldn't you have to also factor in what your opponent's hitting % is in transition as well as what your team hits in transition?
|
|
|
Post by meanmug on Apr 17, 2013 14:18:09 GMT -5
Just use averages. Once you factor your opponent trans hitting (ie, your defense) and your hitting in transition, then you are again measuring things other than passing.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Apr 18, 2013 11:16:09 GMT -5
With recent discussion here on Volleytalk of stats/metrics, is there any chance of the creation of something like www.baseball-reference.com for stats? I know there'd be cost involved, but something like that would make all forms of inquiry easier, yes?
|
|
|
Post by meanmug on Apr 18, 2013 11:37:55 GMT -5
The cost is prohibitive and statistics (and by that I mean DataVolley scout files) are not accurate and standardized enough from team to team.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Apr 19, 2013 9:52:53 GMT -5
I've heard 'costs are prohibitive' before--but no one has actual numbers.
What would it cost? With volunteers to input stats into databases--that would speed things. Would it simply be in hosting, etc that the costs are significant?
|
|