Settings

Language, shortcuts & timezone

Language

Timezone

Time zone changes apply on this device. A new device may start at GMT+8; check match times before reading.

Skip to content
Referrals

Settings

Customize your preferences

Survivorship Bias in Tipster Rankings and the Missing Services

By Opta Tips Team Updated

Check which services entered and left a tipster comparison, preserve the original cohort and distinguish observed returns from scenarios about missing records.

Concept cards distinguish the original service cohort from the services still visible in a current directory.

A ranking of currently active football tipsters describes the services still visible today. It cannot automatically describe everyone who began publishing at the start of the period. If unsuccessful services disappear more often than successful ones, the remaining table can show an encouraging record even when the original population performed poorly.

That is the survivorship problem. It concerns which analysts remain in the comparison. A complete list of selections for each visible analyst does not, on its own, resolve the disappearance of entire analysts from the sample.

Identify the population before calculating its result

Suppose a reader wants to know what happened to services available on the first day of a season. The relevant population is the cohort present on that date. A table containing only services available at the end of the season answers a different question.

Record the original identifiers and the date the cohort was fixed. Keep renamed, suspended and closed profiles attached to that initial list where the evidence supports their identity. Do not infer that similar names represent the same operator, or that a removed profile must have been unsuccessful.

The reason for a departure matters. A service can stop through retirement, a business change or a platform decision. Survivorship bias is a possible consequence of selection into the observed sample, not permission to assign losses to every missing analyst.

Follow an invented cohort through the arithmetic

Consider four fictional services, each with 100 units staked over a specified observation period. A earns 20 units, B earns 10, C loses 15 and D loses 25. Combined net profit is minus 10 units on 400 staked, giving a cohort ROI of minus 2.5%.

Now assume C and D stop publishing and vanish from the current table. A and B together show 30 units profit on 200 staked, an ROI of 15%. The remaining services’ calculation is correct for its stated scope. It becomes misleading if presented as what a reader would have experienced across the original four services.

Nothing about the removal improved the original selections. The reported population changed. This simple example uses equal turnover to make the weighting transparent. With different stakes, aggregate ROI must be calculated from combined net profit divided by combined eligible stakes.

A simulation can illustrate the mechanism without measuring a real platform

Football-Data published a 2017 thought experiment in which simulated tipsters had no forecasting advantage and those falling below a performance threshold could disappear. Its surviving group could display positive aggregate results. The exercise demonstrates how a withdrawal rule can select a flattering remaining sample. It does not measure OptaTips or prove that any current platform applies that rule. Football-Data’s original model description.

Keep that evidential boundary visible. Simulations depend on their assumptions about entry, selection frequency, outcome probabilities and removal. Changing those assumptions changes the numerical results. The mechanism is more portable than any one simulated percentage.

Separate missing services from missing selections

There are two different completeness checks. Within a profile, ask whether its published wins, losses, prices and corrections remain visible. Across profiles, ask whether the table retains services that stopped during the review.

A platform might preserve every selection for active profiles yet omit inactive ones. Another might keep inactive profiles while their old selection history is incomplete. Both limitations deserve disclosure, but they should not be merged into one vague claim that records are verified.

For OptaTips, inspect the actual directory and available archives. If no historical membership list exists, state that the original cohort cannot be reconstructed from current public pages alone. Do not fill that gap using remembered names or retrospective guesses.

Keep entry and exit dates in a cohort ledger

A practical ledger contains the analyst identifier, first observed date, last observed date, current status and evidence source. Add a field distinguishing confirmed closure from a profile that simply could not be accessed at one observation.

Temporary access failure is not proof of permanent disappearance. A changed URL can also break an old link while the same service continues elsewhere. Verify identity through public evidence and retain uncertainty where no reliable mapping is available.

Services joining after the start date belong in an entrants group, not silently in the original cohort. They have had different exposure time and may operate under different conditions. Show their outcomes separately before attempting a combined view.

Describe partial histories honestly

When a departing service’s history is unavailable, the original cohort result may remain unknown. Report the number of services and the amount of turnover whose outcomes can be inspected. That coverage statement is more defensible than treating the visible subset as the whole.

A sensitivity analysis can ask what different missing outcomes would imply, provided those are explicit hypothetical assumptions. It cannot recover the actual missing record. Keep an observed subtotal separate from any scenario-based range.

A current shortlist and a historical evaluation serve different purposes

One useful distinction is between a fixed cohort and a rolling directory. A fixed cohort keeps the original services in the denominator throughout the review, including those that stop. A rolling directory changes membership as services enter and exit. Neither design is automatically wrong. They answer different questions and should carry different labels.

For a rolling directory, show membership changes beside the performance summary. If a profitable service joins halfway through a month, its earlier off-platform record should not automatically be inserted into the directory’s observed history. Similarly, removing an inactive service should not silently remove its already observed losses from a claimed historical platform total.

Do not repair a selected sample with an arbitrary penalty

Knowing that information is missing does not reveal its value. Assigning every disappeared analyst a large loss can create the opposite distortion. A closure might follow poor results, but that possibility is not an observed settlement.

Use distinct statuses for a known result, a missing result and a hypothetical scenario. Suppose the visible services have 200 units of known turnover and 30 units of profit, while two former services have unknown turnover and returns. The observed subtotal is 15% ROI for the visible subset. A full-cohort ROI cannot be computed until its missing denominator and numerator are established.

If a scenario assumes each unavailable service staked 100 units and lost 20, label all three assumptions. The scenario then gives minus 10 units on 400 staked, or minus 2.5%. That number belongs to the stated scenario, not to the actual population. Changing the invented missing outcomes changes the result.

Check what a ranking claim promises

A statement about the top ten current profiles is narrower than a statement that following the platform’s analysts would historically have produced a particular return. The second claim needs a rule for which analysts were followed, when membership changed and how stakes and costs were allocated.

A retrospective top-ten list also selects on the outcome being celebrated. Even if no profile disappeared, choosing the best performers after the period is over does not establish that a reader could have identified them at its beginning. Survival filtering and retrospective ranking can operate together, so identify both instead of treating a retained archive as the end of the investigation.

An observation plan chosen beforehand can clarify this. Write the eligible population, observation dates, availability rules and calculation convention before examining the final ranking. Later changes should remain visible as changes. This does not eliminate uncertainty or guarantee a useful forecast, but it prevents the evaluation rule from shifting unnoticed with the results.

A reader looking for services available now reasonably begins with current profiles. Availability is part of that practical task. Historical evaluation still needs the selection process behind the shortlist explained, especially when past returns are used as evidence of widespread forecasting ability.

Do not interpret an adjusted cohort result as a verdict on every surviving analyst either. A service can have an independently inspectable record even when the population around it is selected. Individual evidence and population inference require separate conclusions.

A useful final note states which cohort was intended, how many services remain observable, what happened to entrants and exits, and which financial results could actually be reproduced. The uncertainty should remain visible where historical membership or returns are missing.

Connect the cohort check to an individual record review

Once the population boundary is clear, inspect the records that can actually be accessed. Keep the selection unit, settlement period, original price and stake basis consistent. A complete cohort with incompatible return calculations still cannot support a useful aggregate.

The record-comparison guide explains returns and costs, while the market-mix comparison shows how different selection populations can reverse a ranking within the surviving services themselves. Cohort membership adds a further question about who entered that comparison.

These checks can end with a limited finding. The visible records may be reproducible while the original population remains unknown. State both facts. That gives the reader a clear boundary for the evidence without turning missing information into either an endorsement or an accusation.

For adults aged 18 and over. All cohort figures above are fictional teaching examples. Historical results do not guarantee future outcomes or profit.

Reader discussion

Comments 0

No comments yet.

Leave a comment

Your email address will not be published. Required fields are marked *