A result calculated from an odds archive usually applies to the fixtures with the required data. If missing quotes are silently excluded, the reader cannot tell how that sample relates to the original fixture list. A coverage report should therefore come before the headline outcome rate.
Start with the intended fixture population
In this illustrative study, the scope contains 200 fixtures. A valid quote for the chosen market and observation time exists for 140 of them; 60 have no usable quote. Price coverage is 140 / 200 = 70%. That is a measure of data availability, not an outcome hit rate.
Suppose the selected outcome occurs in 84 of the 140 covered fixtures. The observed frequency is 84 / 140 = 60%. Report it as “60% among the 140 fixtures with valid quotes”. Calling it the rate for all 200 fixtures treats unobserved records as though they were part of the calculation.
Distinguish missing prices from missing results
To illustrate the possible effect of unknown outcomes, temporarily suppose the results of the other 60 fixtures are also unavailable. If none produced the selected outcome, the full-population rate would be 84 / 200 = 42%. If all did, it would be 144 / 200 = 72%.
The resulting 42–72% range is not a confidence interval. It is a pair of extreme counting bounds under this particular missing-outcome assumption. It says nothing about which extreme is plausible.
In practice, the results may be known while only the prices are missing. You could then calculate the outcome frequency for the complete fixture list. However, you would still not know whether the missing-price fixtures belonged inside a specified odds band. Knowing the result does not restore the quote or determine membership of a price-based filter.
Account for exclusions explicitly
- No market recorded by the source
- No quote observed inside the required time window
- Unresolved fixture or market identifier
- Invalid record or ambiguous settlement scope
- Postponement, cancellation or another predefined exclusion
Give each excluded fixture a primary reason so that the counts reconcile with the starting total. Additional issues can be recorded separately. Define the rules before inspecting which exclusions improve the result.
A replacement value introduces a new assumption
Using another bookmaker's price to fill a gap changes the meaning of the original series. Substituting a later closing quote for an earlier missing observation may introduce information unavailable at the research cutoff. These are methodological changes, not neutral tidying operations.
Publish the original fixture count, usable observations, exclusion counts and coverage by competition. Read the source's collection notes, too. Limited coverage does not automatically make a study unhelpful, but it limits the population to which its findings can reasonably apply. The coverage table makes that limit visible.
Sources and method
Calculations and scenarios are original teaching examples, not observed match data, product performance or predictions. Sources support the stated concepts and methods. This article was prepared with AI assistance; sources and calculations were checked.
- Football-Data: data coverage and collection notes — accessed 14 September 2026
- scikit-learn: data leakage — accessed 14 September 2026