Every Rally Isn't Equal: What a Markov Chain Reveals About Volleyball
TL;DR
A volleyball set can be modeled with just two numbers: each team's side-out percentage.
Small edges compound. In our model, a 6-point side-out edge wins about 79% of best-of-5 matches.
Points aren't equal. A rally at 23–23 swings set win probability about 46%; at 5–5 it's about 13%.
The same model gives coaches a baseline for testing momentum and a common currency for valuing every skill.
Two numbers run the whole set
What actually decides a volleyball set? We modeled it and found the answer is simpler than most coaches expect.
At any moment, a set is fully described by two things: the score and who is serving. Mathematicians call a system like this a Markov chain, the same tool used in genetics, queueing theory and board games. Each rally moves the set to a new state, and the odds of each move depend only on the current state.
That framing collapses the whole game into one stat per team: side-out percentage, how often a team wins the rally when it's receiving. A team's point-scoring rate on its own serve is simply 1 minus the opponent's side-out rate. Feed in those two numbers and you can compute the odds of winning from any score.
Small edges compound
A tiny side-out advantage turns into a big match advantage. We started with two evenly matched teams that each side out 60% of the time, then gave Team A a small edge.
One extra side-out per 100 rallies is almost invisible on the court, yet it lifts Team A's match win probability from 50% to 55%. The edge compounds twice: rallies into sets, then sets into matches. It's the same arithmetic that gives a casino its house edge.
Not every rally is worth the same
A rally at 23–23 is worth about 15 times as much as a rally when you trail 20–24. We measured each rally's leverage: how much winning it instead of losing it changes your chance of taking the set (two evenly matched teams, each siding out 60% of the time).
Score
Swing in set win probability
5–5
about 13%
20–20
about 27%
23–23
about 46%
Down 20–24
about 3%
This gives coaches a rigorous basis for decisions they usually make on feel. Save timeouts, key substitutions and your best server for high-leverage moments, and don't spend them on rallies that barely move the needle.
Momentum becomes testable
Every coach has watched a five-point run and felt the gym tilt. But is momentum real, or does randomness just look like momentum?
A Markov chain is memoryless: it assumes the next rally doesn't care about the last five. That makes it the perfect baseline. If real teams string together runs more often than the model predicts, momentum is real and we can measure how big it is. If they don't, most runs are what chance alone produces.
Published research on the "hot hand" in volleyball is mixed, and many runs turn out to be ordinary streaks. Rally-by-rally data is exactly what's needed to settle the question for a specific team or league.
One currency for every skill
Is a serve error worse than a missed dig? A shanked pass worse than a blocked attack? Coaches argue about these trade-offs constantly, often without a shared yardstick.
The model provides one. Measure how much each skill outcome moves a team's side-out percentage, then convert that into set and match win probability. Serving, passing, attacking and defense all end up in the same unit: wins.
What this means for Loggerhead users
If you log who served and who won each rally in Loggerhead, you already have everything this model needs. That opens the door to:
Live win probability that updates rally by rally.
Leverage flags that highlight the rallies that decided a set.
Momentum checks that compare your team's scoring runs to what chance predicts.
Player value measured by how much each player moves your side-out percentage.
A note on the numbers: the figures in this post come from modeling hypothetical, evenly matched teams, not from Loggerhead match data. The model also simplifies real volleyball by assuming side-out rates stay constant through a set. Real data will show where that holds and where it breaks, and that's the next post.