In this post I will talk about a concrete experimental design that I'm considering implementing with our team to help us make better approach decisions at the service line. I’ll also discuss how I’ll talk to athletes about what we’re trying to do. When I look at a lot of the public statistical analysis out there, people are relying almost exclusively on match data. This makes some amount of sense, that’s the data that is public, outsiders don’t get access to a team’s practice data. But, match data lacks the counterfactual of “what would have happened in that match if they had served differently?” By collecting data in practice, we can use the counterfactual to help make approach decisions.
Since I started coaching, I’ve always felt silly calling serving zones for athletes. When I was an athlete none of my coaches ever did that so it’s not something I was socialized into. Also, I think that there isn’t just one correct serving zone for any given situation. There are definitely some wrong decisions, but on any given serve someone could be justified in serving to probably at least three spots. I also don’t think a coach has a significant informational advantage over athletes in making those decisions. Picking a serving zone isn’t that strategically complicated. If a person has a moderate amount of preparation, scouting, and volleyball acumen; and is intentional about using them, they will make good decisions on zones. I also think it helps confidence and execution to see the game and make a plan yourself, and that the gains from that confidence are much greater than any marginal gains from making a slightly better zone decision. All of this combines to make me not super excited to be giving athletes serving zones. This is probably one of my closer-held unpopular takes, I think giving serving zones is pretty common. I’d be interested to hear from yall who do it about how it helps you win.
I would like to replace helping athletes make zone decisions with helping athletes make approach decisions. There is a lot of information that goes into making a good approach decision and I think that the band of good approach decisions is much narrower and more difficult to access than the band of good zone decisions. It seems like a tall task to keep all the information (what rotation the opponent is in, how well our opponent sides out in that rotation, the respective shapes we expect an athlete’s various approaches to generate, how the side out situation and the approach combine to generate an expected win probability added, how the athlete is feeling about a given approach on the day, and others) that goes into an approach decision in your head while you’re also playing a match. I mentioned that I don’t think coaches have a significant informational advantage over athletes in picking zones. I think coaches can have a significant information advantage over athletes in picking an approach for a given serve, especially because coaches can reference notes and athletes cannot.
So, if signaling in approach decisions to servers is where we want to get, how do we generate the data that we need to get there? Naturally occurring data is of limited effectiveness in creating prescriptions for approach changes. I can look at data about an athlete’s past performance and quantify how productive they have been. Even if we assume that past performance is a good indicator of future performance, which is a big “if,” that doesn’t help me much with making prescriptions for how to change. Prescriptions about how to change require a counterfactual, “what if I had done X instead of Y?” We can’t get information on counterfactuals from match data with no treatment on approach. Furthermore, match data is noisy. There’s variance in opponent quality, gym environment, how athletes were feeling physically and emotionally on the day, approach changes that we don’t know about, and more that generate noise in the data. With enough reps, a lot of that could come out in the wash. But, if you remember from the blog about statistical significance, a person needs 100+ serves in a given approach for us to get good data on it. I imagine that number goes up when we introduce all those other variables into the mix. So, if I’ve got someone who started last year and is returning, I’ve probably got enough match data to make some kind of prescription. But, for people who didn’t start or are new to the team, I have no data, and I want to be able to help them make good decisions before the season is over. So, I think it makes sense to generate data in practice.
Here are a couple of ways I could see going about it (I’ll just be using the technology we have in our gym so no VolleyMetrics):
Experimental Design 1:
Put a serve-receive formation on one side of the court.
Put a set of servers on the other side of the court
Put someone with a radar gun on the side with the servers. Have them record who the servers are
Put someone with a way to take notes on the side with the receivers. Have them record who the receivers and servers are.
Instruct each server to imagine three different levels of serving aggression, a 3/5, a 4/5, and a 5/5. Whatever this means to them is fine
Instruct all the servers to always serve with a level of aggression of your choosing
Each server gets a turn serving. They serve a ball, the person with the radar gun records the speed, the person with the receivers records the results of the pass (who passed it, pass rating). If the server makes an error or surrenders a 3 pass, their turn is over. If the server surrenders a 2 pass they go again if it’s their first two pass of the turn. If it’s the second, their turn ends. If the server surrenders a 1 pass or makes an ace, they go again. Servers should receive a downball to pass any time their serve stays in play.
Cycle through servers, each taking turns, until you’ve gotten as much data as you want for the day. Servers in line should do something that raises their heartrate before they start their turn. You can also change what rotation the receivers are in, making sure to record that change as well.
You should end up with a series of data points on discrete serves, server, passer, result, speed, approach, how many serves they had in the turn before that serve. Since you’ll have multiple people recording data on serves, make sure you have a system for getting those to match up.
Once you have a chunk of data for each serving approach, you can analyze that data as you choose to generate a representation of the shape of production of each server’s various approaches.
Some explanation of that experimental design:
I want to have the radar gun to have an objective measure of some kind. I have no idea what it’ll look like but it’d be interesting to see if there’s a noticeable shift in speeds between approach.
I want the cycling through servers and the downball passing to disrupt some of the muscle memory and mimic a game-like serving scenario. It’s much easier to serve aggressive consistently when it’s your fifth serve in a row and your heartrate is low. It’s harder to serve aggressive consistently when it’s your first serve and your heartrate is high. I want to both practice doing so and be collecting data in game-like situations.
I don’t love that this drill isn’t totally game-like, it’s not full 6 on 6 play. This means it isn’t perfect at mimicing a game scenario and you have to take practice time away from playing to get your data. I do like that this gets you more data in a given chunk of time than playing 6v6 would.
Experimental Design 2:
Set up a 6v6 drill as you would in practice. Either just playing a game straight up or whatever drills you would normally use.
Set up people gathering data as before. Add a column for side-out or not.
Instruct servers to serve at a set aggressiveness level.
Play the drill and record the results. Make sure that your servers are serving at a variety of serve receive situations. One server shouldn’t serve at row 1 every time.
Analyze the data how you like
Some explanation of that experimental design:
I like that playing a 6v6 drill mimics game situations better. I like that this is something you’d normally do at practice so the data collection isn’t taking up practice time. I like that you can get points scored data to generate win probability added numbers. I don’t like that we’re collecting not very much serving data at a time. It could be difficult to get to that 100 serves per approach mark in preseason or between contests.
Once you’ve collected data, either through these methods or other ones, you can start to make decisions based on it. You can fit approaches to situations and help athletes make good decisions.
Next post: Serving Series 9: Implementing an Approach
Craiyon prompt for today: "Degas painting of a volleyball practice"
Comments