Transparency

FootyStats predictions: turning a stat into a probability you can check

By FootInsights · Published · 5 min read

People who search for FootyStats predictions usually want one of two things: the site’s own prediction pages, or a way to turn the statistics it is known for into a decision about tonight’s match. This article is about the second, because it is the part a reader controls — and because the same reasoning applies to every stats site, and to the numbers published here.

To be explicit: nothing below claims that FootyStats’ predictions are poor. It describes what kind of number a team statistic is, what kind of number a prediction is, and the work that sits between the two.

What FootyStats publishes

FootyStats describes itself as a football stats and analysis site, and that is the right way to think about it. At the time of writing it advertises coverage of well over a thousand leagues and organises its pages around the statistics bettors ask for most: both teams to score, over 2.5 goals, corners, cards, expected goals and first- and second-half goals. Its predictions section sits on top of that, mixing tips posted by users with predictions the site says are calculated from historical statistics, across the match result, goal lines, BTTS, corners and correct score.

That breadth is a real service. A table showing every team in a league ranked by how often its matches finish over 2.5 goals is hard to build yourself and useful to have. The question is what happens next, when a reader looks at one row of that table and treats it as a forecast.

A stat describes the past; a prediction prices one match

“Both teams scored in 70% of this side’s matches” is a statement about matches that have been played. A prediction is a statement about one match that has not: this opponent, this venue, this week. Getting from the first to the second takes three adjustments, and skipping any of them is where a percentage starts to mislead.

The sample. Seven matches out of ten is 70%. But a team whose true rate is 50% — a coin flip — produces seven or more out of ten about 17% of the time, purely by chance. In a twenty-team league of identical coin-flip sides, about three would be sitting on a “70% BTTS” figure after ten games without being any different from the rest. The arithmetic of how badly small samples deceive is in sample size in betting records; it applies to team stats exactly as it applies to tipsters.

The opposition. A season-to-date figure is an average over whoever the team happened to play. Ten matches can easily contain six against the bottom half. The same side’s rate against tonight’s opponent may be nowhere near its headline figure, and the opponent has a figure of its own that points somewhere else. Averaging the two teams’ percentages feels reasonable and has no statistical meaning: neither number was measured against the other team.

The base rate. A percentage only means something next to what is normal for the competition. Some leagues see both teams score in well over half of all matches and some sit far below, as the tables in BTTS rates by league and over 2.5 goals rates by league show. A team at 60% in a league that averages 58% is ordinary. The same 60% in a league that averages 45% is a real signal.

Four checks before a stat becomes a bet

None of this needs a model. It needs five minutes and the discipline to do it every time.

  1. Count the matches behind the percentage. If the figure rests on fewer than about fifteen games, treat it as a hint. Look for last season’s figure for the same team and see whether the two agree.
  2. Split it by venue. Home and away rates for goals markets are often far apart, and tonight’s match is only one of them. Most stats sites publish the split — use that row, not the overall one.
  3. Set it against the league. Subtract the competition’s average. What is left is the part that belongs to the team.
  4. Convert the price and compare. Odds of 1.60 imply 62.5% before the bookmaker’s margin is removed. If your adjusted estimate is 63%, the stat has told you nothing the price did not already contain. Our odds converter does the conversion, and how to remove the vig covers the margin.

The fourth check is the one that matters most. A statistic that everybody can see is a statistic the bookmaker has already seen. It is only worth money when your reading of it differs from the price, and you are the one who is right.

How to judge the predictions themselves

For the prediction pages — FootyStats’, a tipster’s, ours — the test is the same one we set out in how to compare football prediction sites, and it has nothing to do with how good the underlying statistics are:

  • Is there a probability, or only a pick? A pick can be right or wrong. A probability can be checked for honesty over hundreds of matches with a proper scoring rule.
  • Was it fixed before kick-off? A prediction that can be edited afterwards is not a record.
  • Is every prediction counted? A leaderboard of the best performers, on any site, is by construction a list of whoever has run well lately. What matters is the full history behind each name, losses included.
  • Can you check it yourself? If not, the 30-day audit in how accurate is Forebet? builds the record independently, with a spreadsheet and five minutes a day.

One more caution that is specific to stats sites: two of them can show different numbers for the same team, for reasons that have nothing to do with error. Why football stats differ between sites explains the collection and definition differences, and they are worth knowing before you compare a figure from one site with a price built on another’s data.

Apply the same test to us

A stats site and a prediction site answer different questions, and using both is sensible. FootyStats will tell you what has happened across far more competitions than we cover. What we publish is narrower: tips with a probability behind each one, for every match in the competitions we follow, frozen before kick-off, with every result — misses included — settled on a public track record and a calibration page that shows whether our 70% calls happen about 70% of the time.

We quote no performance figures in articles, ours or anyone else’s. The live record is the claim, and the checks above are the instrument for testing it. Whatever you conclude, stake only what you can afford to lose: a stat is a starting point, a probability is an estimate, and neither one knows how tonight ends.