Two football websites disagree: how to compare their numbers

Before deciding a statistic is wrong, align the competition, time period, event definition and treatment of corrections.

You check a match on two websites and find different totals for an attacking statistic. The disagreement may be a mistake, but it may also come from different definitions or update times. Comparing the numbers without checking their scope can create a controversy that the data does not support.

Make sure the records describe the same thing

Start with the obvious details: team, competition, season and match. Check whether a player page covers league matches only or all competitions. Check whether a cup total includes extra time. A season named by its starting year on one website may be labelled by both years elsewhere.

For live matches, record the retrieval time. A provisional event may later be reassigned or removed. Comparing a screenshot taken during play with a corrected post-match page is not a clean test of consistency.

Read the event definition

Terms that sound familiar are not always identical across providers. An assist may follow competition rules or a provider’s own statistical convention. A chance-creation figure may count a defined type of final pass. Do not silently combine these into a single table unless their meanings align.

If the definition is unavailable, label the limitation. “Provider A reports four; Provider B reports five; their definitions were not confirmed” is more informative than selecting the larger value because it supports your preferred player.

Work through one hypothetical discrepancy

Imagine a player with five assists on a league page and seven on a general profile. First inspect competition coverage. If the profile includes two cup assists, the figures may be perfectly consistent. If both claim to cover the league, inspect dates and match logs. Only after those checks should you investigate event classification or an error.

Avoid averaging incompatible figures

Taking the mean of two disputed totals does not resolve a definition problem. A value halfway between them might correspond to neither provider’s method. Choose a consistent source for the comparison, explain the choice and keep the original attribution visible.

The same caution applies when combining player data. If one player’s numbers come from a different competition or provider, readers need that information before seeing a ranking. Consistent presentation is not the same as comparable evidence.

Keep a compact source note

For a published analysis, record the source URL, date accessed, competition and definition where available. Link to the relevant page rather than only the provider’s homepage. If you later correct a figure that changes the conclusion, add a visible correction instead of quietly rewriting the argument.

Data literacy is often less about advanced mathematics than careful bookkeeping. Matching the scope and meaning of a number can prevent a more serious error than calculating it to another decimal place.

Continue reading: More statistics guides.

Reconcile a disagreement one field at a time

Suppose two pages show different shooting totals. First establish that both concern the same match and the same final update. Then split the disagreement into event categories. Under Opta's published definitions, an effort that hits the frame and stays out is off target; a block by the last defending player that prevents a goal can count as on target. A label that looks obvious may therefore have a precise collection rule.

These rows illustrate possible causes; they do not explain a real discrepancy until you check the actual pages. Keep the URLs, access date and relevant definitions with your notes so another reader can follow the comparison.

When the figures still disagree

Do not average conflicting counts to create an apparently neutral answer. An average of two incompatible definitions has no clear meaning. Choose one provider consistently for the comparison and explain the unresolved difference if it matters. For modelled figures such as expected goals, even matching the shot list may not produce identical values because models can use different information. A provider-specific number should remain attached to its provider.

Expanded on 10 October 2026 with a worked exercise, practical checks and source notes.