Why numbers need questions
Sport and esports produce a constant stream of statistics: win rates, averages, rankings, streaks, ratings. Broadcasters, commentators, fans and news sites all use them to tell stories. A number can be perfectly accurate and still give a poor picture, because the choice of what to count, over what period, and compared with what, shapes the story it tells.
Statistics literacy is the habit of asking a few plain questions before accepting a figure. It does not require advanced maths. It requires knowing where the common traps are.
Ask how many games sit behind the number
The first question is always about the size of the sample. A player who wins four of their first five matches has an 80% win rate, but five matches say very little about ability. Results in games with any element of chance, and that includes almost all sports and esports, wobble a lot over short stretches. The idea behind this is the law of large numbers: averages settle down only as results accumulate. Our article on small sample sizes covers the details, and the related idea of short run results shows how much random variation can look like a real trend.
Ask which average is being used
“Average” can mean the mean (add everything up and divide) or the median (the middle value). When a few extreme performances are present, the two can differ a lot. A single enormous match can drag a mean upward while the typical match looks much more ordinary. When a report says “averages 18”, it is fair to wonder whether the median would say the same thing. The comparison of median versus mean is worth knowing, because many player statistics are lopsided in exactly this way.
Ask about the time window and the split
A claim such as “unbeaten in their last six” or “scores more in away games on Tuesdays” may be true and still tell you little. The start and end points of a window, and the way data is split into categories, are choices someone made. If many windows and splits were tried and only the striking one was reported, the result is likely to be a product of selection rather than a pattern. This is the territory of cherry-picked statistics.
Ask whether the surprise will last
After an exceptional season or a remarkable run, the next one is usually less remarkable. This is not necessarily a slump or a loss of form. Extreme results tend to contain an unusually favourable dose of luck, and luck does not repeat on demand. This pattern, known as regression to the mean, explains many “second season” disappointments and some apparent effects of coaching changes. A related error is the gambler’s fallacy, the belief that a run of results makes the opposite result due. The hot hand debate sits on the other side of that coin: whether streaks reflect a real change in ability or just the clumps that random sequences naturally produce.
Ask what else could explain a link
Two numbers moving together does not mean one drives the other. A team that uses a certain tactic may win more because it is also the stronger team. Correlation versus causation is one of the most common mistakes in sports commentary. Closely related is Simpsons paradox, where a trend seen in overall data reverses once the data is split into groups.
Ask who is missing from the story
Success stories are told by those who succeeded. Advice from a famous player about training habits ignores the many people with the same habits who never reached the top. That is survivorship bias, and it affects how we read biographies, coaching advice and “what the best players do” articles.
Read percentages carefully
A rise from 10% to 12% is a gain of two percentage points, but also a 20% relative increase. Both descriptions are true, and a headline will pick whichever sounds bigger. Knowing the difference between relative and absolute change is essential, and it connects to the broader skill of working with odds and percentages. Where a story involves a rare event, ask also about the base rate: base rate neglect is the habit of ignoring how common something was to begin with.
Treat model-based numbers as estimates
Many modern statistics are not counts but model outputs. Expected goals, win probability graphs and rating systems such as Elo ratings are built from assumptions and past data. They can be informative, but each rests on choices about what to include, and each produces estimates rather than facts. The wider field of sports probability models and esports analytics rests on the same principle: a model is a simplified description, and its output carries uncertainty.
Be cautious with uncertainty itself
Good statistics come with some sense of how uncertain they are. Confidence intervals, the standard deviation of results and the shape of the normal distribution are all ways of describing how much a figure could vary by chance. When a report gives a single figure with no hint of spread, treat it as a rough guide.
A short checklist
- How many games or events are included?
- Is the mean or the median being reported, and would the other tell a different story?
- Who chose the time window, and could another window say something different?
- Is a link between two numbers being treated as a cause?
- Who or what is missing from the sample?
- Is a change described as relative or absolute?
- Is the number a count or the output of a model?
None of these questions guarantees you will reach the right conclusion. They simply slow down the step from “I saw a number” to “I believe a story”, which is where most statistical mistakes enter. Sports and esports are also games of chance and skill mixed together, and numbers rarely separate the two cleanly. Holding conclusions loosely is a reasonable response.