By the fourth round somebody has already published a list of the league’s most clinical finishers. A striker has scored three from five shots and now owns a conversion rate of sixty percent. A goalkeeper has faced eleven shots and saved ten. A promoted side sits fourth and the word being used is transformed.
None of those three numbers is wrong. All three are useless, and for the same reason: the denominator is too small to carry the claim being hung on it. A rate calculated from five events is not a rate. It is one event with decoration.
The useful question in September is not which team is best. It is which measurements have already accumulated enough raw material to mean something, and which are still noise wearing a decimal point. That line falls in a consistent place, and knowing where it falls is most of the skill.
The short list and the ignore list
- Usable now. Shot volume for and against, shot locations, touches in the opposition box, corners and final-third free kicks won, passes allowed per defensive action.
- Usable but weak. Chance-quality totals, expressed as a difference rather than as a per-match figure.
- Not usable yet. Conversion rate, save percentage, set-piece goal share, anything ending in percent that came from fewer than a hundred events.
- Not a measurement at all. Points after four rounds. That is a fixture list.
- The single most common error. Comparing a September number with a November number as if the conditions were the same.
The mistake is not the number, it is the denominator
Every statistic is a count divided by something. Counts grow quickly because matches generate them in bulk. In most competitions a team takes something on the order of ten to fifteen shots a match, which means five rounds produce enough shots to describe how much a team shoots with reasonable confidence.
Rates are the opposite. To know how often a team scores from those shots, the useful sample is not the number of matches but the number of goals, and goals arrive two or three to a match across both sides. Five rounds might give a single team six or seven goals. Nobody would estimate a national average from seven observations, yet the same person will happily rank finishers on five shots each.
The underlying idea is old and unglamorous. The law of large numbers says an observed average approaches its true value as the sample grows, and says nothing comforting about what happens before then. Early in a season, the gap between observed and true is the whole story.
Volume settles early. Conversion does not.
The practical division is between things a team does and things that happen to it. A team decides, through structure and personnel, roughly how many shots it takes, where it takes them from, how far up the pitch it defends and how many corners it wins. Those are behaviors, they repeat every match, and they settle within a handful of games.
Whether those shots go in is a much thinner signal. Finishing quality exists, but its influence is smaller than the run of the ball over any short window, and the events are too rare to separate the two. A team shooting well from good positions and scoring nothing is far more likely to be unlucky than to be broken, and the reverse is equally true.
That asymmetry has a practical consequence. Judge a new coach on where the shots are coming from, not on whether they went in. The first is available in September. The second is not available until spring, if at all.
One contamination cuts across both categories, and it is worth naming early. Game state changes behavior: a side that leads after twenty minutes drops deeper, concedes possession and shoots less, and none of that reflects how it intended to play. In a five-round sample a couple of early goals can rewrite a team’s entire volume profile. Where the data allows it, filter to periods of level scoring before comparing anything.
The table is a schedule wearing a ranking’s clothes
After four or five rounds no two clubs have played the same opponents, and roughly half have had the easier half of their fixtures. A side sitting fourth may have played three of the bottom five at home. A side sitting fourteenth may have played the same three clubs away.
Neither position contains information about quality until the fixture list has evened out, which in a single round-robin format means most of the way through the first half of the season. Until then the table is an accurate record of results and a poor estimate of ability, and those are different things.
There is a simple correction available to anyone with a fixture list and ten minutes: write down each club’s opponents so far, and where. It will not produce a rating, but it will stop the worst mistakes, because the outliers at both ends of the table are usually explained by the column you just wrote.
You are measuring a squad that did not exist in June
Domestic squads here turn over heavily every summer. Contracts run short, and a club can replace a third of its outfield group in one window without anything unusual having happened. The eleven that starts the opening round has, in many cases, trained together for a matter of weeks.
Early numbers therefore describe a team in an unusual state: fit, unfamiliar, and still learning where the second and third runners are. Defensive metrics suffer most, because a defensive line is a coordination problem and coordination is exactly what a new group lacks. Attacking volume tends to arrive before attacking coherence.
The reasonable inference is not that the team is bad. It is that the September version and the February version are different objects, and a metric collected on one does not automatically transfer to the other. A club that changes coach and signs heavily in the same window is a particularly poor candidate for early conclusions.
Heat moves the numbers before the team does
Opening rounds are played in the hottest part of the year, and in the southern venues that means conditions in which no team can run the way it will run in December. Total distance falls. High-speed running falls further, because sprints are the first thing a fatigued player rations. Pressing intensity drops, possessions get longer, and the match slows down.
Every one of those changes shows up in the data as a tactical fact. A team that appears passive in an August fixture on the Gulf coast may press aggressively in the same shape a month later in the northwest. The metric did not lie; it measured the weather.
The correction is to compare like with like: southern away fixtures against other southern away fixtures, evening kick-offs against evening kick-offs. That is a smaller sample again, which is the recurring trap. Every honest correction costs sample size.
Set pieces distort a small sample more than anything else
Dead-ball goals arrive in clumps. A team can score two from corners in one match and none for the next eight, and in a five-round window those two goals can be a third of everything it has scored. Any early statistic built on goal origin is therefore close to worthless.
The volume version of the same measurement survives. Corners won, free kicks conceded in dangerous areas, throw-ins into the box, first-contact wins at the near post: these are counts, they occur many times a match, and they describe intent and set-up rather than outcome.
The same reasoning applies to penalties. One penalty is worth roughly three-quarters of a goal in chance-quality terms, so a single award swings a team’s totals in a way that has nothing to do with how it plays. Where a chance-quality figure is quoted early, ask whether penalties are included before anything else.
Where the data itself is soft
A match data sheet is not a single object. Some fields are close to mechanical: shots, corners, substitutions, cards. Others are coded by a person watching video and applying a convention, and the conventions differ between providers and drift between operators on the same provider.
Duels won, key passes, big chances, pressures and progressive carries all belong to the second category. Two suppliers can watch the same match and disagree by a wide margin on any of them, which is why a number quoted without a source is not a number. Analytics work at club level normally starts by picking one supplier and staying with it, precisely so that internal comparisons remain valid even when external ones do not.
Positional and tracking data, where camera and wearable systems are installed, is more consistent but far from universal across venues. A league in which some grounds are tracked and some are not produces a dataset with a hole in it, and the hole is not random: it correlates with budget.
What early season football statistics can honestly support
Three claims survive the first month. First, a description of how a team wants to play: where it shoots from, how high it defends, how much of the pitch it concedes. Second, a comparison of a team against its own recent past, held at the same venue type and the same time of year. Third, a flag for something to watch, which is not the same as a conclusion.
What early season football statistics cannot support is any ranking of individuals by efficiency, any judgment of a goalkeeper, any claim that a coach has been vindicated, and any statement containing the phrase on course for. The events required to justify those claims have not happened yet.
This is a less exciting position than the one taken in most previews, and it is the one that survives contact with April.
A reference table for the first six rounds
Chance-quality models sit awkwardly in this list because they are a hybrid: a count of shots weighted by a rate estimated elsewhere. That construction makes expected goals more stable than raw conversion and less stable than raw shot volume, which is roughly where it belongs.
| Metric | Type | When it becomes usable | What it says in September |
|---|---|---|---|
| Shots for and against | Count | A handful of matches | How much volume the team creates and allows |
| Shot location profile | Distribution | A handful of matches | Where the attack is built to finish from |
| Touches in the opposition box | Count | A handful of matches | Territorial reach of the attack |
| Passes allowed per defensive action | Ratio of counts | A handful of matches | Pressing intent, not pressing quality |
| Corners and final-third free kicks | Count | A handful of matches | Dead-ball volume the side generates |
| Chance-quality difference | Hybrid | Roughly a third of a season | A weak signal, better than points |
| Conversion rate | Rate | Most of a season, often longer | Nothing reliable |
| Save percentage | Rate | Most of a season, often longer | Nothing reliable about the goalkeeper alone |
| Set-piece share of goals | Rate | Most of a season | Nothing reliable |
| Points per match | Outcome | After the fixture list evens out | The schedule played so far |
Three checks before you quote a number
- Count the events, not the matches. Ask how many shots, saves or duels produced the figure. Under a hundred events, treat any percentage as decoration.
- Name the opponents and the venues. Write out who has been played and where. If the answer is three home fixtures against the bottom third, the number describes a schedule.
- Separate count from rate. Counts describe behavior and settle fast. Rates describe outcomes and settle slowly. Half of all early-season arguments are two people quoting different types at each other.
Run those three and most September claims collapse on their own, which saves the trouble of arguing with them. The competition’s own published regulations and match reports are the primary record; anything derived from them inherits whatever was recorded there.
Frequently Asked Questions
How many matches does a metric need before it means something?
The wrong unit. What a metric needs is events, and different metrics generate events at very different speeds. Shot counts arrive by the dozen every match; goals arrive one or two at a time. A working rule is that anything expressed as a percentage needs at least three figures’ worth of underlying events before it is worth an argument.
Is expected goals reliable in the first month?
More reliable than goals, less reliable than shots. It converts each attempt into a value, so it uses every shot rather than only the ones that scored, which is a genuine improvement in sample efficiency. It still inherits the small number of high-value chances, and one penalty can distort a five-round total.
Why do early tables so often mislead?
Because the fixture list is unbalanced by design at that stage and because two or three results decide the whole ordering. A single deflected goal moves a club four places. That sensitivity does not vanish later, but it shrinks as the number of matches grows.
Can a goalkeeper be judged on save percentage at all?
Not on its own, and not early. Save percentage measures the shots faced as much as the goalkeeper, and shots faced are decided mostly by the defenders in front of him. A shot-quality-adjusted version is better, needs a large sample of its own, and still says little about distribution or command of the area.
What is worth watching in September instead?
Structure. Where the team builds from, which player receives between the lines, how the press is triggered, who takes the second ball at a corner. None of that requires a large sample because it is observed rather than estimated, and it is what the numbers will eventually describe anyway.
Do these limits apply to lower divisions as well?
More so. Coverage is thinner, coding is less consistent, and squads change more between seasons. Everything above is true one level down, with a smaller and noisier dataset to work from.
Numbers earn trust by accumulating, and in September they have not accumulated. The reasonable position for another six weeks is to describe what teams are doing, keep the percentages in a drawer, and let the sample do the work it was always going to have to do.
