Which Football Stats Actually Predict Results
Which football numbers repeat and connect to results, and which only describe the past: a stability table, the arithmetic behind a six-to-ten-match window, what blending home and away costs on one priced bet, and the inputs each market actually needs.
Most football statistics do not predict anything. They describe what happened, which is a different job. A number is useful for betting only if it repeats from one run of matches to the next and connects to results. Shots on target, chances created and conceded, and venue-split goal difference clear both tests. Possession, conversion rate and "form" over three matches do not.
This page sorts the two groups, sets the window they need, and shows what blending home and away costs on one priced bet. It expands points 5 and 6 of the pre-match checklist.
In short.
- A stat earns its place by repeating: if a side's first ten matches do not predict the next ten, it is describing luck.
- Six to ten matches is the working window: noise on a scoring rate is ±0.50 goals a match at six and ±0.39 at ten, and twenty only reaches ±0.27 on stale data.
- Split every figure by venue: blending a 2.00 home rate with a 1.10 away rate put the Over 2.5 estimate below 11.5 points too high.
- That single error turns a bet you thought was +9.0% into one that is −15.2%, or −₦75.80 on a ₦500 stake.
- A three-match surge is normal: a side averaging 1.3 goals scores six or more in three matches one time in five.
Repeatable or random: the test that matters
Take any statistic and split a club's season in half. If the first half's figure lands close to the second half's, the number measures something the team does; if it jumps around, it measured circumstances. Shot volume passes, because a side's way of attacking does not change week to week. Conversion rate fails: a club scoring 18% of its shots over ten matches has no tendency to keep doing it.
The second test is connection to results. Possession passes the first test easily — a possession side stays a possession side — and still fails as a predictor, because having the ball is a style, not an advantage. A team with 62% of the ball and no shots inside the box has told you nothing about the outcome.
Which metrics carry over, and which do not
| Metric | Repeats? | Predicts results? | Verdict |
|---|---|---|---|
| Non-penalty expected goals, for and against | Strong | Strong | The core number, split by venue |
| Shots on target, for and against | Strong | Good | The best substitute where no xG data exists |
| Shots inside the box | Strong | Good | Removes the long-range attempts that inflate shot counts |
| Goal difference at this venue | Moderate | Strong | Better than points; it keeps the margins |
| Set-piece chances created | Moderate | Moderate | Personnel-dependent; check who is available |
| Possession share | Strong | Weak | A style marker, not an edge |
| Total shots, blocked efforts included | Moderate | Weak | Volume without quality |
| Corners won | Moderate | Weak | Useful for the corner market only |
| Conversion rate | Weak | Weak | Almost pure noise over a short run |
| Save percentage | Weak | Moderate | Needs a season or more to mean anything |
| Points from the last five matches | Weak | Weak | Mixes venues, opponents and luck into one figure |
| Head-to-head record | Weak | Weak | Different squads, different managers, tiny sample |
The bottom half of the table is where most previews live. Form guides, head-to-heads and possession averages are easy to publish and read, which is why they are already in the price.
Why the window is six to ten matches
Goal counts scatter predictably: the spread around an expected number is roughly its square root, which makes the choice of window arithmetic rather than taste. Take a side that genuinely scores 1.5 goals a match.
| Window | Goals expected | Typical spread | Noise per match |
|---|---|---|---|
| 3 matches | 4.5 | ±2.12 | ±0.71 |
| 6 matches | 9.0 | ±3.00 | ±0.50 |
| 10 matches | 15.0 | ±3.87 | ±0.39 |
| 20 matches | 30.0 | ±5.48 | ±0.27 |
Three matches carry ±0.71 of noise on a 1.5 average — the error is half the size of the measurement. Doubling from ten to twenty buys 0.12 goals of precision and costs you data from four months ago: a different manager, a different first eleven. Use ten matches when nothing has changed at the club, six when something has.
One club, two teams: the home and away split
Every figure in the top half of the table has to be split by venue: the two versions of a club are far enough apart to break a price.
| Figure, last 10 at each venue | At home | Away | Blended |
|---|---|---|---|
| Expected goals for, per match | 2.00 | 1.10 | 1.55 |
| Expected goals against, per match | 0.95 | 1.60 | 1.28 |
| Points per match | 2.10 | 0.90 | 1.50 |
Example. This club travels to a side that concedes 1.20 expected goals at home. On the blended figure the match prices at 1.55 + 1.20 = 2.75 goals, putting Over 2.5 at 51.9%. On the away figure it is 1.10 + 1.20 = 2.30 goals and Over 2.5 is 40.4%. One shortcut, 11.5 percentage points.
Now price it. The coupon offers Over 2.5 at 2.10. On the blended estimate the bet reads 0.519 × 2.10 − 1 = +9.0%, or ₦45 in your favour on a ₦500 stake. On the honest estimate it is 0.404 × 2.10 − 1 = −15.2%, or −₦75.80 per ₦500 staked.
Nothing else in the analysis was wrong. One averaging shortcut moved a bet across ₦120 of expected value, and would do the same on the next twenty fixtures. Venue-split tables are on the tournaments pages.
Three matches is a story, not a sample
"They have scored six in three games" is the most repeated sentence in football previews, and it is usually coincidence dressed as a trend. Take a side whose true rate is 1.3 goals a match. Across three matches it should score 3.9. The chance of it scoring six or more anyway, with nothing having changed, is about 19.9% — one three-match block in five. In a twenty-club division, roughly four teams are in the middle of a scoring "surge" at any moment, and each will have an explanation attached to it in the preview. The same arithmetic runs in reverse for a drought. Three rules follow.
- Never let a three-match run override a ten-match figure. If it does, you are betting on the story.
- Before adjusting for a run, name the cause: a new signing, an injury, a formation change. No cause, no adjustment.
- Check whether the run is in the chances or only in the goals. Six goals from 5.8 expected is a real improvement; six from 2.9 is finishing that will not hold. That distinction is the subject of the guide to expected goals.
Important. Fitting a rule to three matches is the cheapest way to lose money slowly. Any pattern can be found in a small enough sample; the test is whether it was still there in the fifty matches you did not look at.
Which numbers matter for which market
No statistic is useful in general. Each market settles on a different event, so each has its own inputs and its own list to ignore.
| Market | What to read | What to ignore |
|---|---|---|
| Match result (1X2) | Venue-split expected goal difference; points per match there | Last five results across all venues; head-to-head |
| Over/under goals | Both sides' xG for and against at this venue; box shots | Total shot counts; the league average alone |
| GG/NG | Each side's xG against here; how often each failed to score | "Both scored in four of the last five" |
| Handicap | Goal difference per match here; margin of recent wins | League position, which hides very different margins |
| Cards | The referee's cards per match; fouls committed by both sides | A club's reputation for aggression |
| Corners | Crosses and attacking-third entries; who will be chasing | Corner counts from previous meetings |
Reading the wrong row is expensive: a well-built totals estimate says almost nothing about who wins, and a strong result read says nothing about bookings. The full map of markets is in the guide to football betting markets.
Keeping the numbers honest
- Roll the window forward weekly rather than recalculating from the season's start.
- Use non-penalty figures for attacking quality; penalties belong in the total, not the judgement.
- Note the opposition: ten matches against the bottom half is not the evidence that ten against the top half would be.
- Write your figure down before opening the coupon. A number you can compare with the result later is worth ten adjusted in hindsight.
Where to apply it. Venue splits, chance figures and line-ups sit on our match pages, and the fixtures our analysts have worked through are in today's predictions.
FAQ
Is possession really useless for betting?
Not useless, but not predictive on its own. It tells you how a match will look rather than how it will end. Possession becomes informative only alongside what a side does with the ball — box entries, shots on target — and at that point you are reading those numbers instead.
What can I use if my league has no expected goals data?
Shots on target for and against, split by venue, over the last eight to ten matches. It is a coarser measure of the same idea and it repeats well. Add shots inside the box if the source has them, since that removes most of the volume that never threatened a goalkeeper.
Do these stats work the same way in the NPFL as in the Premier League?
The logic does; the data availability does not. Where detailed chance data is thin, fall back to goals and goal difference split by venue over ten matches, widen the edge you demand before betting, and treat divisions you cannot measure properly as a pass rather than a guess.
This article is for information only and is not an inducement to gamble. Betting involves the risk of losing money — never stake more than you can afford to lose. 18+. If gambling stops being entertainment, read our responsible gambling guide and seek help.
