.png)
Why I Don’t Like the Points Won-Lost Differential
Introduction
Performance indicators should normalise data into percentages, ratios or per set/ match averages. This is so it creates a level playing field so we can understand and compare the differing levels of performance. Performance indicators should normalise data into percentages, ratios or per set/ match averages. This is so it creates a level playing field so we can understand and compare the differing levels of performance.
The Points Won-Lost Differential (PWLD) is a commonly used performance indicator. It’s widespread usage and acceptance is perhaps to no surprise because it is easy to understand and simple to calculate (PWLD = Points Scored – Points Lost). Anybody could read a DataVolley match report and understand that a positive PWLD is good (you scored more points than you lost in errors) and the bigger the positive difference, the better the point scoring performance (you scored a lot more points than you lost in errors).
However, the PWLD is not a great indicator of performance. This is because it does not normalise the data effectively, which means it does not create a level playing field so we can understand and compare the differing levels of performance. I will outline below why this is the case.
Why the Points Won-Lost Differential Is Not A Good Performance Indicator
A Bigger Points Won-Lost Differential Does Not Necessarily Mean A Better Point Scoring Performance
You might think the bigger the positive Points Won-Lost Differential (PWLD), the better. However, this may not necassarily be the case. A player with a high PWLD may only be high simply because they had the most point scoring opportunities. Similarly, a player cannot achieve a high PWLD if they had limited point scoring opportuniteis. Total point scoring opportunities can vary greatly depending on player position (opposites and outside hitters will have much more attack attempts than middle players since they can only attack in optimal setting conditions), time on court (players can’t score from the bench), number of sets played (it would be easier to score 50+ points in a 5 set match than a 3 set match), starting rotation (players will have more attack opportunities if player start in the front row/ zone 4) and if they are the go-to hitters (since they will get the most sets to attack).
So when you see a player’s PWLD, you start take into account all the above factors to work out if the level of performance is slightly better or worse than the actual number suggests. For example, a middle player with a +10 PWLD will be a lot more impressive than an outside hitter with the same +10 PWLD. When you do this however, you are essentially trying to normalise the data further to create a better level playing field to compare performances when an effective performance indicator should have done that for you.
You Can’t Scale the Data
What you think of what a good Points Won-Lost Differential (PWLD) should be, it is usually based on values of a single match. I have already described above how there are issues with that alone. However, if you start combining more match data to find a more accurate performance/ skill level, you start to get bigger differences and it’ll get to a point when those massive differences starts to become meaningless. This is because the Points Won-Lost Differential (PWLD) measures the absolute difference between the points won and points lost and not the relative difference.
For example, Polonia SideOut London had a PWLD of +613 in the 2019-20 season. It’s a staggering large number but what does it mean in relation to point scoring performance level? We can say it is good but we just don’t know how good it is. It could be the team was only a decent point scoring team but because I statted so many matches that season, a small but positive PWLD can rack up to extraordinary large numbers. We could try and normalise the data further by calculating a PWLD per set average, so we know what the PWLD was on a per set basis. This will make the PWLD easier to understand since we understand more on what values we get on a per set basis. However, I have already written before why you shouldn’t use per set averages and as I said above, an effective performance indicator should have normalised the data effectively already so you don’t have to.
You Don’t Really Understand Differing Levels of Performance
A performance indicator should create a level playing field so you can understanding differing levels of performance. For example, I would say a hitting percentage of 30% is considered a great level for attacking. Only 12 players hit 30+% in the Women’s Super League (2019-20) and only 2 City University London players in recent history have hit that high. More men can hit at this level as the Top 10 Men hit 35+% in the Men’s Super 8s (2013-14). However, what is that number for the PWLD? As an exercise, try to fill out the table below of what the PWLD should be for differing levels of performance. I did it for hitting percentage and I got you started with hurting the team and minimum level.
You will probably realise that what the PWLD should be at each performance level will depend what it is based on. The PWLD for a great level of performance is a vastly different number for an individual player than it is for a team as a whole. The PWLD for a great level of performance is a vastly different number for a single match than it is for a season as a whole. The root problem is again down to the fact that the PWLD measures absolute differences. This means the more data you base the performance level on, the bigger that difference can become. Since the PWLD doesn’t normalise the data effectively, it doesn’t create a level playing field to understand and compare differing levels of performance.
So What Other Performance Indicators Can We Use?
I created a performance indicator called the Points Efficiency Ratio, calculated as PER = Points Won/ Points Lost. It tells you the number of points scored for every error committed. It is a better performance indicator than the Ponts Won-Lost Differential (PWLD) because it measures the relative difference between points won and lost rather than the absolute difference. This will normalises the data more effectively. So a player who scored 100 points with 50 errors has the same output to a player who scored 20 points with 10 errors, i.e. a 2.0 PER (=100/50; =20/10). I have written previously why I like the PER so much in this post and this post. I’ve also written up studies that shows that the PER is closely associated to success in volleyball (click here, here and here). It also has the advantage that the PER uses same stats used to calculate the PWLD, i.e. points won and points lost. So as part of your calculations, you just need to change the minus (-) sign to a divide (/) sign for a much superior performance indicator.
I recently created another performance indicator called the Points Efficiency Percentage that should be even better than the PER. It is essentially the same calculation as a hitting percentage but for point scoring, i.e. PE% = (Points Won – Points Lost)/ Total Point Scoring Attempts. The PE% tells us the percentage of point scoring attempts that results in a point won whilst also taking into account errors. The reason why the PE% is a better performance indicator compared to the PER is because it takes into account the total point scoring attempts. We know already how many points was scored and lost via errors with the PER but we want to know how many point scoring attempts it took to get there. For example, Player A and Player B scored 20 points and lost 10 points but Player A had 50 point scoring attempts and Player B had 100 attempts. Both players have a 2.0 PER (=20/10) but Player A has a PE% of 20% [=(20-10)/50] while Player B has a PE% of 10% [=(20-10)/100]. This means Player A did twice as good as Player B in relation to point scoring performance. A disadvantage of using the PE% is you do need to have stats on all the total attempts in the attack, service and block to calculate the PE%. Not all teams will stat this, especially the total block attempts (some teams don’t stat block outs, rebounded balls or touch touches).
So Why I Don’t Like the Points Won-Lost Differential
So the reason why I don’t like the Points Won-Lost Differential (PWLD) is because it does not normalise the data effectively so we don’t really know if a player or team had a good point scoring performance. I would instead use a Points Efficiency Ratio (PER) as it measures the relative differences between points won and lost. It has the advantage that it uses the same stats, i.e. points won and lost, and it can be easily explained as the number of points won for every point lost in error. However, a Points Efficiency Percentage (PE%) is better because it takes into account total point scoring attempts. It can easily be explained as the hitting percentage for point scoring, i.e. the percentage of point scoring attempts that results in a point won while taking into account points lost via errors.
Comments