
Two accumulator tickets can contain different combinations and still rely on the same football result. If both include the same home win, one draw can defeat both tickets. The recorded accumulator win rate remains a valid description of those settled tickets, but the tickets do not provide two independent tests of the selection process.
This matters when a long record contains many combinations built around a small number of repeated selections. A hundred tickets need not represent a hundred separate sets of football evidence. To assess an accumulator win rate with shared selections, retain the ticket count and add a map of the fixtures, markets and outcomes that those tickets share.
The worked example below is entirely fictional. Its probabilities are assumptions chosen to show the arithmetic, not estimates for real fixtures or OptaTips performance. Every selection has a binary win-or-lose outcome; there are no voids, cash-outs, partial settlements or changes after publication.
The shared selection creates a second layer of dependence
There are two different questions. Within one accumulator, do its selections depend on each other? Across a group of accumulators, do the tickets share selections? Even if every underlying fixture result is independent in a model, tickets that reuse one of those results are dependent.
For example, ticket AB contains selections A and B, while ticket AC contains A and C. B and C might concern unrelated fixtures. That does not remove A from either ticket. If A loses, AB and AC both lose regardless of what happens in the other matches.
The general probability rule uses conditional probabilities; multiplying the separate probabilities is the special case for independent events. OpenStax explains the distinction between the general multiplication rule and its independence case. Applied here, the relevant events must be identified carefully: the independent objects assumed in the example are the three underlying selections, not the two tickets created from them.
Our separate explanation of correlation between selections within an accumulator addresses the first question. This article concerns the second: how repeated selections affect the reading of a collection of ticket results.
Two tickets, three underlying results
Assume A has an 80% chance of winning, B a 60% chance and C a 50% chance. Assume the three underlying outcomes are mutually independent. Ticket AB needs A and B to win; ticket AC needs A and C to win.
AB therefore has an assumed success probability of 0.80 × 0.60 = 0.48, or 48%. AC has an assumed success probability of 0.80 × 0.50 = 0.40, or 40%. Those are separate ticket probabilities. They do not establish how often both tickets succeed together.
For both tickets to win, A, B and C must all win. Count A once: 0.80 × 0.60 × 0.50 = 0.24, or 24%. Multiplying the two ticket probabilities would instead give 0.48 × 0.40 = 0.192, or 19.2%. That calculation incorrectly treats the shared A outcome as two independent hurdles.
The complete set of ticket outcomes can be calculated without inventing any additional matches:
| Ticket outcome | Assumed probability |
|---|---|
| Both win | 24% |
| AB wins, AC loses | 24% |
| AC wins, AB loses | 16% |
| Neither wins | 36% |
The second row requires A and B to win while C loses: 0.80 × 0.60 × 0.50 = 24%. The third requires A and C to win while B loses: 0.80 × 0.50 × 0.40 = 16%. Together with the two other rows, the probabilities sum to 100%.
Why both winning and both losing become more likely
There are two ways neither ticket can win. A can lose, which has assumed probability 20%. Alternatively, A can win while both B and C lose, which has probability 0.80 × 0.40 × 0.50 = 16%. These cases cannot occur together, so their probabilities add to 36%.
For comparison, imagine two independent ticket outcomes with the same individual success probabilities, 48% and 40%. Their chance of both losing would be 0.52 × 0.60 = 31.2%. Their chance of both winning would be 19.2%. The overlapping tickets have more probability at both extremes: 36% for no wins and 24% for two wins.
This comparison holds the individual ticket probabilities fixed. It isolates the effect of their relationship. It is not a claim that replacing one real selection with another will preserve those probabilities, prices or expected returns. Real choices can change all three.
The chance of exactly one winning ticket is 40% in the shared-selection example. In the independent comparison it is 49.6%. Sharing A shifts probability away from the middle and towards joint outcomes. That is why a record can produce clusters of wins and losses even though its individual ticket success probabilities have not changed.
The average can stay the same while the evidence changes
The expected number of winning tickets is 0.48 + 0.40 = 0.88 in both examples. Dividing by two gives an expected ticket win proportion of 44%. Dependence changes the distribution around that average; it does not automatically change the average itself.
In any one run of these two tickets, the observed win rate can only be 0%, 50% or 100%. The 44% figure describes the model average across repetitions of this hypothetical setup, not a result that one pair of tickets can literally record. Nor does it predict the next real pair.
A statement such as “our average win rate is unchanged, so overlap does not matter” therefore leaves out the question a reader is trying to answer. Overlap matters when judging concentration, losing clusters and how much independent evidence supports a rate. The numerical description of settled tickets and the strength of evidence behind a future-performance claim are different issues.
For the basic distinction between ticket wins and individual leg wins, start with the accumulator win-rate counting guide. An overlap audit adds information to that count; it does not replace it with the number of unique fixtures.
Build a ledger that can reveal the overlap
Keep one row for each original ticket, including its publication time, selections, initial fold count, stake, accepted odds and final settlement. Give each underlying selection an identity based on the fixture, market, line, chosen outcome and relevant settlement period. A team name alone is not enough: a full-time home win and a first-half home win are different selections.
Next, map each selection identity to the tickets that contain it. In the example, A maps to AB and AC, B maps only to AB, and C maps only to AC. This small map explains a joint loss immediately without erasing either losing ticket.
Useful descriptive figures include the number of tickets, the number of unique selections, the most frequently repeated selection and the share of tickets containing it. Ten tickets with one common selection have a 100% common-selection share for that selection. That fact is useful even when no reliable probability estimate is available.
Stake concentration needs a separate calculation. If all ten tickets carry equal stakes, the common selection affects every unit staked across them. With unequal stakes, count the actual amounts committed to tickets containing that selection. Do not add the full stake once per leg and present the resulting sum as fresh money wagered.
Do not turn a fixture count into an effective sample size
Counting three unique selections is not a licence to relabel the two-ticket record as a sample of three independent accumulator trials. The objects differ. Likewise, ten tickets sharing one selection cannot automatically be reduced to one statistical observation: their other legs still create different possible outcomes.
A defensible uncertainty calculation needs a model or a method that preserves the relevant dependence. Depending on the actual question, this might require grouping observations by shared fixtures or evaluating complete publication cohorts together. A generic confidence interval that assumes independent, identically distributed ticket outcomes may not describe a record containing different fold counts, changing prices and overlapping selections.
If the available record lacks selection identities, acknowledge the gap. You can report its observed ticket rate, but you cannot verify how much overlap is hidden behind it. An apparently precise interval calculated from the ticket total alone does not repair missing evidence.
Compare the same publication scope
A fair audit retains all tickets from the declared publication window. Excluding tickets because the common selection lost removes exactly the outcome that reveals the concentration. Showing only the surviving combinations after the first fixture is settled changes the population being described.
Also distinguish published recommendations from personally placed tickets. A reader may combine several separately published selections, place the same combination twice or omit a selection after an odds move. Those actions create a personal ticket record, which cannot be attributed to the original service without showing how the tickets were constructed.
When examining the accumulator tips page or any other source, first establish what the displayed item represents. Is it a complete ticket, a selection within a suggested combination, or a historical result? Retain that unit consistently before making a win-rate comparison.
The useful final record has two connected views: every ticket and its actual result, alongside the underlying selections that connect those tickets. A run of several winning accumulators is still several winning tickets. Showing their common dependence lets the reader judge what that run demonstrates, without treating repeated exposure to one football result as repeated independent confirmation.
评论 0
暂无评论。