How to interpret sports statistics without fooling yourself

You had 62% possession and lost 1–3.

What does that tell you? Almost nothing, and that is the problem with match statistics. They arrive looking like answers. A number on a screen has an authority that a coach's impression does not, and it is frequently less reliable.

This guide is written for coaches and analysts reading their own team's data — not for betting markets. Six rules, in the order they matter.

Rule 1: A number means nothing without a comparison

Every statistic is a comparison waiting for its other half. Fourteen turnovers is not a lot or a little. It is fourteen.

Three baselines, in descending order of usefulness:

Your own recent average. The best one, by a distance. Same team, same level, same person doing the counting, which controls for the three biggest sources of noise at once. Your last five matches is a good working window.

Your league or age-group average. Useful if you can get it, which at grassroots level you usually cannot.

Elite benchmarks. Almost always misleading for amateur and youth teams, and worth saying plainly: comparing your U14s to professional averages will tell you they are bad at everything, which is both true and useless. Elite figures describe a different sport played by selected adults.

If you take one thing from this article: compare your team to your team.

Rule 2: Rates beat totals

Totals conflate performance with opportunity. This is the most common error in amateur stat sheets and the easiest to fix.

Four players, tackles won across a season:

By totals, the ranking is A, B, C, D. By rate, it is almost exactly reversed: C, B, A, D. Player C, who looks fourth-best, is the most active tackler in the squad by a wide margin — they have simply played a third of the minutes.

Convert to per-90, per-set, per-possession or per-attempt before you rank anything. It takes one column.

Rule 3: Respect the sample size

One match is a sample of one.

A player goes 1-for-3 from three-point range. Is their shooting a problem? You have three data points. If their true rate were 33%, the most likely single-game outcome would be exactly what you observed — and if their true rate were 50%, going 1-for-3 would still be entirely unremarkable. The observation does not distinguish between the two hypotheses at all.

The useful principle rather than a table of thresholds:

• High-frequency events stabilise quickly. Passes, receptions, duels — dozens per match, so a few matches gives you a usable rate.
• Low-frequency events need a season, or longer. Goals, aces, penalties, red cards. A striker's conversion rate over ten matches is mostly noise.
• The rarer the event, the longer you must wait, and the more confidently people will over-interpret it in the meantime.

Specific stabilisation points do exist in the published literature, but they are sport-specific and metric-specific. If you need one, find the study for your sport rather than borrowing a number from football analytics.

Practical version: never change a player's role on one match of data. Never not mention something for a season either.

Rule 4: Separate process from outcome

Outcome statistics (goals, points, results) are what everyone looks at, and they are the noisiest things on the sheet. They sit at the end of a long chain of luck: a post, a deflection, a referee, an opposition keeper having a good day.

Process statistics (entries into the final third, shot locations, first-contact quality, defensive recoveries) are more stable, more repeatable, and much more coachable. A player can act on "get your first touch out of your feet". Nobody can act on "score more".

The rule: coach the process metrics, judge the season on outcomes. A team improving its process numbers while losing is a team about to start winning. A team winning on poor process numbers is borrowing against next month.

Rule 5: The five traps

The possession fallacy. Possession is a style descriptor, not a quality metric. Plenty of very good teams deliberately have less of it. 62% possession in a defeat is not evidence of anything except that you had the ball.

Ratio distortion. 100% pass accuracy sounds excellent until you notice it was two passes. Always show the denominator. A rate without its sample size is a rumour.

Selection bias. You tagged the moments that confirmed what you already believed. This is the reason to watch the footage once through before you start coding — see the video workflow.

Averaging away the tail. Your average defensive line height is fine. The four times it was catastrophic are the reason you lost. Averages hide exactly the events you most want to fix, so look at the worst cases separately.

Correlation read as cause. Your team ran further in the matches you lost. Running more did not cause the defeats — being behind caused the running. Ask what else changed before you draw an arrow.

Rule 6: Pair every number with the evidence

A statistic tells you that something happened. It cannot tell you why, and it cannot change a player's behaviour on its own. Nobody has ever improved because they were shown a spreadsheet.

So every number that matters needs to be traceable back to the moments that produced it. A tagging tool does that by making each data point a clip, which is genuinely the main argument for computerised over hand notation.

There is a second half to this, and it is the one usually missing. The footage shows what happened. It does not record what you understood at the time. You saw the shape break down in the 23rd minute and you said something specific about it, out loud, on the touchline — and by Monday that sentence is gone. That was the half that actually explained the number.

This is the narrow thing Sidetalk does: it records what you said while coaching, works out which player each remark was about, and puts it on your video's clock, so the 23rd minute has both the picture and your reasoning attached. To be clear about the limits: it does not compute statistics, count events or navigate from a stat to a clip. It handles the commentary layer, not the numbers.

A stat sheet a small club can actually sustain

Six to eight metrics, no more. The constraint is not ambition, it is week five.

Any team sport

1. Availability. Sessions and matches attended per player. Dull, unglamorous, and the strongest single predictor of development you will ever track.
2. Minutes played per player. Also the denominator for everything in rule 2.
3. One team process metric, tied to your current coaching priority. Change it when the priority changes, not sooner.
4. One individual process metric per position group.
5. Outcome. Results and goals for and against. For the season review, not for Tuesday.

Football variant

6. Build-up exit rate — build-ups that reach the final third, as a share of build-ups started.
7. Turnovers by third and channel.
8. Shots by location, split inside and outside the box.

Court-sport variant

6. Success rate by phase — serve, reception, attack, transition.
7. Errors as a share of total attempts, per phase.
8. Points won from the opposition's first mistake.

Everything on that list is countable by one person with a phone and the free tally templates. None of it needs a subscription.

Performance analysis, statistics, analytics — which is which?

These three terms are used interchangeably and describe different jobs.

If you coach a team, you are doing performance analysis and you need statistics to do it. Analytics is a different discipline with different tooling, and most content written about it does not apply to you. The full explanation.

Frequently asked questions

How do you interpret sports statistics?

Compare every number to a baseline — ideally your own recent average. Convert totals to rates. Check the sample is big enough to mean anything. Prioritise process metrics over outcome metrics. And trace each number back to the footage that produced it before acting on it.

Which sports statistics matter most for coaching?

Availability and minutes played, plus one or two process metrics tied to what you are currently coaching. Outcome statistics matter for reviewing a season and mislead badly when used to review a match.

How many matches of data do I need before trusting a stat?

It depends entirely on how often the event occurs. High-frequency events like passes and duels stabilise within a handful of matches. Low-frequency events like goals and aces need a season or more. A useful rule: never change a player's role on one match of data.

Is possession a good statistic?

No, not as a quality measure. It describes how a team plays rather than how well. Read possession alongside what was done with it — entries into the final third, shots from good locations — rather than on its own.

How does sports analysis improve performance?

By closing a loop. It replaces recall with a record, turns that record into one specific thing a player can do differently, and then re-measures to check whether it happened. The re-measurement is the step that produces improvement and the step almost everyone skips.

Where to go next

The workflow that produces these numbers — how to analyse sports video. The coding method behind them — notational analysis, with free templates. And if you are choosing tools, the buyer's guide.