One Annotation, Many Analyses | Dodgeball Analytics

A detailed match annotation is not merely a way to calculate today's statistics. It is a reusable analytical investment. It can answer the question that motivated the annotation, support new questions that arise later, become more informative as additional matches are collected, and be reanalyzed whenever better methods are developed.

Sports analysis rarely ends with the first answer. More often, the answer produces another question. The value of a match record therefore depends not only on what it can tell us today, but on how many future questions it leaves open.

Practical example

Questions that may arise:

Which player has the highest hit percentage?
Are isolated throws or coordinated multiball attacks more successful?
Does throwing accuracy change as a set progresses?

Two analysts annotate the same match to compare throwing accuracy. Both record the thrower and whether each throw hits or misses. One also records the precise time of every throw.

Both can calculate hit percentage. Later, someone asks whether players are more successful during isolated single throws or coordinated multiball attacks.

The detailed record can group throws from the same team that occur within a short interval and compare the attack types. The simpler record cannot determine which throws occurred together without the match being watched and annotated again.

The same timestamps may later support questions about performance over the course of a set, rapid follow-up throws, or team pace. The extra information did not improve the first answer; it preserved the possibility of asking different questions.

What a useful match record must preserve

Hit percentage is a statistic; the individual throws and outcomes are the data. Points per minute is a statistic; the events and court-time intervals are the data. Contextual event value is a statistic; the ordered events and surrounding game states are the data.

The distinction matters because a statistic reflects one interpretation of a match. If the observations remain available, definitions can change, events can be regrouped, and new methods can be applied. The match record should preserve what happened. The analysis can decide what it means.

Detailed events can always be collapsed into totals, percentages, and box scores. A hit percentage cannot later reveal when the throws occurred, whether they formed one attack, who was targeted, or what the game state was. Those details were discarded before the percentage was calculated.

Missing information may sometimes be recovered by returning to the video, but that means repeating work whenever a new question appears. It also assumes that the recording still exists and shows the relevant detail clearly.

Why annotation is an investment

Detailed data become more useful as the dataset grows. One match may contain too few examples to compare multiball attacks reliably, but after twenty matches those early events contribute to a much stronger estimate. Rare states become analyzable, and player, opponent, and role effects become easier to separate.

Better methods also increase the value of old data. Recording a detailed dataset may later support analytics that do not yet exist at the time of collection. A summary dataset remains limited to the questions built into it; a detailed dataset can become more informative over time.

Collect depth with purpose

More data are not automatically better data. Every field must justify the time, difficulty, and possible error involved in collecting it. Precise timing may unlock attack coordination and pace; target identity may unlock opponent adjustment and target selection; player-count and possession states may unlock contextual event value.

The goal is not the largest possible dataset. It is a reliable and flexible representation of the match: detailed enough to preserve valuable possibilities, but efficient enough to collect consistently. That also means storing the event rather than only its current interpretation. A recorded catch can be rescored later; a stored two-point total preserves only one scoring rule.

The question may change

Questions will change, more matches will be added, and better methods will emerge. The same match should not need to be annotated from the beginning each time they do.

Record enough of the match that the data can outlive the question that created them.