.png)
The Identification of Valid Volleyball Performance Indicators in The Women’s Super League 2019-20
Prologue
It all started when I wanted to know which statistics are the most important in volleyball. I looked into this subject for my Masters dissertation using data I collected in the Men’s Super 8s 2012-13 season. The dissertation results showed the Points Efficiency Ratio (PER) and Hitting Percentage was highly related to success in volleyball. I found some support for reception performance indicators but no support for the other performance indicators as related to success in volleyball. My analysis since has been based on these results.
I recently created four graphs on how the Points Efficiency Ratio is related to success in volleyball using match data in the Women’s Super League 2019-20, where I statted nearly every match that season (what else you going to do in Covid-19 lockdown?) For fun, I produced the same graphs for the other performance indicators with the intention of showing there’s a lack of correlation, which should highlight how good the PER is. To my suprise, I found the other performance indicators had a similar correlation, more or less, compared to the Points Efficiency Ratio. Because it looked like there was a correlation, was there an correlation? I wanted a definitive answer so I felt the best to do this was to perform actual statistical testing on the data. Essentially, I was going to run back my masters dissertation, with much less scientific rigour though, but using the Women’s Super League data (again, what else you going to do in the Covid-19 lockdown?)
Purpose of the Study
The purpose of the study is to identify valid volleyball performance indicators of all technical skills currently measured (see link) in the Women’s Super League (2019-20) using the methods of final league points, set outcome and close set outcome.
Introduction
Performance indicators (PIs) normalises data into percentages, ratios or per set/ match averages. This is so it creates a level playing field so we understand and compare the levels of performance. Performance indicators are when they are associated with a successful performance or outcome. Success in volleyball can be defined as final league points, winning or losing a set (set outcome) and winning or losing a close set (close set outcome, i.e. a set score with a 2 point difference, e.g. 25-23).
The literature review into valid performance indicator studies that I wrote for my masters dissertation showed overwhelmingly strong support for attack performance indicators as critical to success. They showed mixed support for serve and block performance indicators, weak support for reception performance indicators and almost no support digs performance indicators as critical to success. There was no studies testing point scoring performance indicators. All the research articles I read was based on elite level volleyball, e.g. Olympics, World League, Italian A1 league etc. There was no research in non-elite volleyball however. Valid performance indicators at elite level may not tbe the same at non-elite level. This is where my masters dissertation study came in where I tried to identify valid performance indicators in non-elite volleyball, using match data in the men’s volleyball Super 8s League (2012-13 season). The dissertation found that point scoring performance indicators and attack performance indicators are the best discriminators for volleyball success in the Men’s Super 8s league. This would make sense because to win a match, you need to win three sets and to win one set, you need to score 25 points before your opponents does. So the ability to concurrently score points and not lose points in errors should be critical to success. Attack would make sense too as most points won and lost are done via the attack. I also found reception is related to final league points.
Both the literature review I wrote and my dissertation results is based on men’s volleyball only. I purposely excluded valid performance indicators studies in women’s volleyball because valid performance indicators for one gender may not be valid for the other. The differences, if any, could relate to the anthropometric and physiological differences between the genders or the 19cm difference in net height.
At this point, it would be best to write up a literature review on valid performance indicators in women’s volleyball. However, this is where the “less scientific rigour” part comes in. I don’t have access of journal articles required and it would take weeks to write up. In any case, I can partly justify not doing this by saying that journal articles studying performance indicators are very likely based on elite women’s volleyball, e.g. Olympics etc. As I said earlier, valid performance indicators for elite women’s volleyball may not be the same for non-elite women’s volleyball.
Method
The sample consisted of 175 sets (out 189 sets played) taken from the Women’s Super League 2019-20 season. The sets were video-recorded and performance data statted with the DataVolley program. The type of peformance data collected are related to serve, reception, attack, block and free balls. Free ball/ overpass dig errors and setting errors were also collected. Performance indicators being tested and its calculations can be seen below (click to enlarge picture). Since I don’t collect stats on defence, I used opposition attack zeros as a proxy measure for digs and opposition attack kills as a proxy measure for missed digs.
Pearson’s correlation was conducted to identify correlations between the performance indicators and final league points. I used this method in Excel to calcualte the r-score and p-value. It would have been ideal if we had a regular league structure (each team plays every other team twice) to run this correlation. With the league structure as it is, with the split into the Top Five and Bottom Four after the first half of the season, I decided to do run the Pearson’s correlation four times to a specific condition. Each condition are (with it’s pros and cons): 1. First half of the season (high validity but a less dataset is used) 2. Full league assuming there was no split (lower validity but incorporates the entire dataset) 3. Top Five league (higher validity but only five data points used to identify a correlation) 4. Bottom Four league (higher validity but only four data points used to identify a correlation)
Wilcoxon signed ranks test was conducted to identify differences between winning and losing performances within a set and a close set. I define a close set outcome as as any set score that finishes with a two point difference, e.g. 25-23, 26-24 etc. Fifth sets are excluded as they only finish to 15 points. I used this website to calculate the z-score and p-value.
Results and Discussion
I will try and simplify the statistical test results. I will ignore the statistical test values (r-score for Pearson’s Correlation and z-score for Wilcoxon sign ranks test) and just show the p-values from those tests. What we are looking for is a p-value of less than 0.05 (typically written as p < 0.05). It basically means there’s a 95+% probability that there is an actual, or , correlation or difference. The smaller than p-value, the higher the probability that there is a correlation or difference and the more confidence we can say that. Below are the p-values of the various statistical tests I ran (click to enlarge picture):
General Points
The results show seven out of nine performance indicators was correlated to league points in both first half of the season and the full season, assuming no split in the league. This means I did see correlations when I was producing the graphs before undertaking this project. This is not practically helpful as it says we better performing teams will accrue more league points. It doesn’t pinpoint what skills to focus on. Something else that is not practically helpful is winning teams perform much better than losing teams in all sets played with p < 0.00001 on all performance indicators measured. Close set should be better at identifying more critical performance indicators. The last thing of note is there is a lack of significant results for any correlations in the Top Five and Bottom Four leagues. I think this is because the correlations must be near perfect for results to be significant since we are trying to correlate five or four data points (i.e. teams).
Strong Support
The results show that the points efficiency ratio is the best performance indicator measured as it had significant results across the board (p < 0.01 or lower). It was particularly strong when correlated with league points accrued in the first half of the season and the full league, assuming no split in the league (p < 0.0001). A p-value of less than 0.0001 means there’s a 99.99+% probability there is a correlation. Hitting pct. for all attempts is also a highly significant performance indicator, particularly in close sets where p < 0.0001. I suspect this the points efficiency ratio and hitting pct. to be valid performance indicators across many studies based on this simple premise: to win a match you need to win three sets and to win one set, you need to score 25 points before the opponent does. So the ability to concurrently score points and not lose points in errors should be critical to success. This is what the points efficiency ratio measures. And since most points are won and lost via the attack (see below for a breakdown of points won and lost in the Women’s Super League 2019-20 – click to enlarge the photo), the hitting pct. should be critical to success too.
Block-defence pct, and defence pct. are are shown to be valid performance indicators too. This is a bit of a surprise because there has not much evidence for these skills as valid performance indicators in the scientific literature. There could be several reasons for this. The first is the literature I have read and studied were based on elite men’s volleyball rather than non-elite women’s volleyball. As I said previously, a valid performance for one gender and playing level may not be the same for the other gender and playing level. So these performance indicators could be valid in non-elite women’s volleyball only. The second reason could be the way these performance indicators are calculated. Usually, digs and missed digs are used to calculate defence performance indicators. However, since I don’t stat defence stats, I used oppositional attack zeros as a proxy measure for digs and oppositional attack kills as a proxy for missed digs. In essence, it could be argued we are calculating a slightly different version of the opposition hitting pct. and we have shown hitting pct. to be a valid performance indicator.
Limited Support
There is limited support for hitting pct. after reception and hitting pct. in transition as a valid performance indicator. It is not as strong at hitting pct. for all attacks however. I suspect it could be that it doesn’t matter in which situation a team gets its kills (and limits errors) but as long as they get it somewhere and with good efficiency overall. There is also limited support for serving rating but is particularly strong when correlated against league points after the first half of the season (p < 0.001).
No Support
There is no support for reception rating in the results as a valid performance indicator. I am surprised because there was support for this performance indicator in my dissertation study. This made sense to me because it is the first opportunity to attack and it is easiest situation to put the team in system (as compared to defending an opposition attack attempt). There is also no support for block efficiency pct. as a valid performance indicator. This is in contrast to block-defence pct. as a highly valid performance indicator. However, with defence pct. as a valid performance indicator also, I suspect the defence is the main driver for the validity in both performance indicators. The blocking makes the block-defence pct. slightly better than the defence pct. but not by itself.
Practical Implications
This could have huge ramifications if I was to incorporate these results into my performance analysis practices. I should primarily focus on the points efficieny ratio and the hitting pct. in my analyses (as I have always done) but also focus on defence pct. and block-defence pct. However, since I don’t actually collect defence stats, I don’t have any defence stats for individual players. This means I cannot make any informed decisions on who we can use or sub in if we needed better defence. To get individual player defence stats, I would need to stat digs and missed digs. I don’t have the skills to stat these in real time so I would need to go back and stat these after the match. This requires a lot of work. It is the same with the Block-Defence Pct. I would need to stat additional block stats of block zero (block attempt but does not touch the attacked ball) and block touches (block attempt that touches the ball and play continues on blocker’s side). This will require so much more work (especially statting block zero attempts).
If I was to think about statting digs and the extra block attempts (and the amount of work it involves), then I will need more evidence that the defence pct. and block-defence pct are associated to success in volleyball. This is because there’s quite a few limitations to this study. The first is no literature review done on valid performance indicators studies in women’s volleyball so there is no basis to compare results. Perhaps the results does reflect the current literature. The second is the data itself. Although it is the most complete dataset I have ever compiled with 50 out of 53 WSL matches statted (probability the most ever in top flight volleyball in England), we don’t actually have the final league points. This is because the league was suspended due to the Covid-19 outbreaks. The league is also not a true or regular league, since it splits after the first half of the season. In addition, we only had 26 close sets to find any differences between winning and losing teams in those close sets. We can always do with more close sets to get a bigger sample size for more accurate results. I would have liked to used SPSS software program, the gold standard for statistical testing, to test this data for accurate statistical results. However, the one big advantage this study has is the data is based on the very league I work. So if I was to follow any study, it would be this one.
Valid Performance Indicator Studies Series
This is a part of the valid performance indicator studies series, including an overview of all five studies that I did. Links are below.
6 | Overview of the five valid performance indicator studies 5 | City University London + Opponents (BUCS South Eastern Division 1), 2017-20, Set outcome 4 | IBB Polonia Men + Opponents (Men’s Super League), 2018-20, Set outcome 3 | Women’s Super 8s + League, 2015-20, League correlation (2017-18) + Set outcome 2 | Women’s Super League, 2019-20, League correlation + set outcome 1 | Men’s Super 8s, 2012-13, League correlation + set outcome (Masters Dissertation) 0 | Academic Literature Review, elite men’s volleyball (2004 – 2012)
Comments