Survivorship Bias in Betting Records: Who Is Missing from a Winning Sample
Why visible or surviving wins cannot replace a complete decision log, and how to restore the denominator for an honest performance review.
A record can keep only the “survivors”
Survivorship bias occurs when an analysis includes people, strategies or bets that remain visible while those that disappeared are left out. In betting records, this is not limited to public posts: a journal may retain winning slips but lose deleted picks; show active strategies but omit withdrawn ones; or include only matches still available in an archive.
The problem is more than missing losses. If inclusion depends on the result, the estimate is biased. For example, a publisher could remove unsuccessful posts or an owner could close a losing profile, while a historical table contains only records still online. Performance research on investment funds explicitly studies the bias created by examining funds that survived rather than the original full group. The same selection logic can affect betting records, although the data and selection mechanisms differ.
An example with the right denominator
Suppose 100 picks were posted before their matches. A later archive contains only 60 records, 36 of which won. The visible sample shows a 60% hit rate. But the other 40 picks do not disappear from history just because their copies were lost: if 48 of the original 100 won, the full-sample hit rate in this illustration is 48%. These figures are invented for the example; they show how changing the denominator changes a percentage.
Keep a list of decisions with unique identifiers and publication-time snapshots, including losses, pushes, voids, missed matches, deleted posts and strategies no longer shown. For every missing record, mark the reason and status: outcome unknown, original post unavailable or bet void under its rules. Do not label “result not preserved” as a win or loss.
Do not confuse it with other distortions
Visibility bias differs from trying many strategies and publishing only the best one. In the first case, whole participants may disappear from the sample; in the second, a promising configuration is selected after many attempts. Both can happen at once. Research on backtest overfitting shows why repeated searches for a historically successful rule need a separate evaluation on data that was not used to select it.
Record more than win or loss: the price available before the event, publication time, stake, market, settlement terms, voids and applicable fees. For a hit rate, keep wins, losses, refunds and unknowns separate; for a money result, state the formula and included costs. The betting journal checklist helps preserve those fields.
What you can and cannot conclude
Ask for the full timestamped list of picks, not a selection of winning screenshots. Reconcile the number of published decisions with the number settled and explain each exclusion. A complete record still does not remove the randomness of short runs, different prices or possible settlement errors.
If the denominator cannot be recovered, the honest conclusion is that the data are insufficient for a reliable evaluation. That proves neither poor nor strong quality; it limits the conclusion. The next article discusses probability forecast scoring, which also requires every forecast to have been saved in advance.