Measuring how fast a game moves isn’t just a matter of clock time — it’s about possessions, transitions, and the bursts of activity that decide outcomes. Event data from play-by-play logs gives us the raw materials to estimate game speed, and a well-designed pace model turns those events into a repeatable, comparable metric. This article walks through the theory, the common formulas, practical implementation, pitfalls, and real-world tips for building an event-based pace model.
Why pace matters and what it really measures
Pace is a short-hand for how many meaningful possessions or action cycles occur in a given time. In basketball, pace often means possessions per 48 minutes; in other sports you might use actions per 90 minutes or seconds per play. The value of a clear pace metric is that it separates how often teams get chances from how efficiently they use them.
For coaches, analysts, and sportsbooks, pace helps normalize performance across teams and eras. Two teams can average the same points, but if one plays faster it changes substitution patterns, conditioning needs, and matchup choices. A robust pace model built from events lets you compare teams and games without relying on subjective judgments.
Core event logic: which events indicate a possession
To estimate possessions, you must decide which play-by-play events start and end a possession. In basketball, common possession-ending events include made field goals, turnovers, and defensive rebounds after a missed shot. Free throw sequences require careful handling because multiple free throws can correspond to one possession.
A widely used possessions estimate comes from Dean Oliver’s work and is implemented across many sites: Possessions ≈ FGA – ORB + TO + 0.4*FTA. That 0.4 factor approximates how often free throw attempts represent a separate possession versus part of the same possession. Using this approach on event logs yields a consistent possession count without having to parse every second of clock time.
From possessions to pace: formulas and an example
Once you have possessions, converting to a speed metric is straightforward. Two common expressions are possessions per 48 minutes (common in basketball) and seconds per possession. The basic formula for possessions per 48 is:
pace (per 48) = (team possessions + opponent possessions) * 48 / (2 * game minutes)
For a single team you can also report possessions per 40 or per 90 depending on the sport or level. Seconds per possession is simply total seconds of recorded game time divided by total possessions, which gives a more intuitive feel for how long each possession lasts.
Example table: computing possessions and pace
Here’s a concise example that shows how event counts map to possessions and pace using a 48-minute game.
| Event | Count |
|---|---|
| Field goal attempts (FGA) | 85 |
| Offensive rebounds (ORB) | 10 |
| Turnovers (TO) | 14 |
| Free throw attempts (FTA) | 22 |
Possessions ≈ FGA – ORB + TO + 0.4*FTA = 85 – 10 + 14 + 0.4*22 = 103.8 ≈ 104 possessions. If both teams combined for about 208 possessions in a 48-minute game, pace per 48 would be 208 * 48 / 96 = 104 possessions per 48 for the combined game, or roughly 104 possessions per team when divided appropriately.
Practical parsing: turning play-by-play into a pace model
When you parse event logs, build a clear event taxonomy first: made shots, missed shots with rebound type, offensive rebounds, turnovers (with subtypes), free throw starts/ends, and period boundaries. Label each event with a timestamp and a game clock so you can resolve sequences that span stoppages.
A typical processing flow is: normalize events, aggregate counts, apply the possessions formula, and compute pace. For more accuracy you can explicitly follow possession state in the log: start a possession on a defensive rebound or inbound and end it on a made shot or turnover. That state-machine approach is slightly more complex but reduces reliance on approximations like the 0.4 multiplier.
Pitfalls, edge cases, and sport-specific differences
Free throws, technical fouls, and replay stoppages are frequent sources of error in event-based pace models. Multiple free throws after a shooting foul should usually count as one possession. Technical fouls that do not change possession should not increment the possession count. Overtime requires rescaling — don’t mix regulation and overtime minutes without adjusting the pace formula.
Different sports require different event choices. Soccer is continuous and lacks discrete possessions; modelers often measure tempo with actions per minute (passes, carries, entries into the final third) or by using tracking data to compute meters per second in transition. Hockey, with rapid turnovers and line changes, benefits from tracking “zone entries” and shot attempts per 60 minutes. Always tailor the event set to the sport’s structure.
Validating and refining your model
Validation is critical. Compare your model’s pace estimates to league-provided metrics when available — for basketball, NBA.com and Basketball-Reference publish tempo and possessions statistics that serve as a benchmark. If your event-derived pace consistently differs, inspect how you treat offensive rebounds and free throw sequences; small differences there cause systematic offsets.
Refinements include weighting certain turnovers differently (e.g., shot-clock violations vs. live-ball turnovers), using rolling-window smoothing to capture in-game tempo changes, and splitting pace by lineups or phases of play. When tracking data is available, supplement event-based pace with continuous measures like player movement speed and transition frequency for a richer picture.
Real-world experience: tips from applied work
In my work analyzing college and professional games, starting with the simple possessions formula gave fast, reliable results that matched official pace metrics closely. The most time-consuming part was cleaning play-by-play feeds — inconsistent naming conventions for events (rebounds labeled differently, missing free throw tags) created more noise than the math itself.
A practical tip: build unit tests for your parser. Create a few canonical game logs where you know the true possessions and run them through your pipeline whenever you change parsing rules. That small discipline prevents subtle shifts in your pace estimates when the underlying event feed changes format.
When to use a pace model and how to present results
Use a pace model whenever you need to compare scoring rates or action intensity across teams, seasons, or rule sets. Pace becomes particularly useful in scouting reports, lineup optimization, and betting markets where tempo influences expected totals and line movement.
Present pace in both relative and absolute terms: show possessions per 48 (or per 90) for direct comparison, and include seconds per possession (or actions per minute) for intuitive interpretation. Visual tools — rolling pace charts, heat maps of transition frequency, or possession-length histograms — help stakeholders grasp how pace varies within games as well as between teams.
