
A tipster can have a higher overall win rate while performing worse within every market category you compare. The reason can be the mixture of selections. A record dominated by short-priced favourites faces a different scoring task from one dominated by longer-priced outcomes.
Useful tipster analysis therefore separates two questions. What happened across the service’s actual portfolio? And what happened within comparable groups? Both matter when reading OptaTips profiles, but they need different calculations. Combining them into one league table can reward the choice of denominator before it tells you anything about forecasting.
Start with the decision the comparison must support
Suppose you want to compare two analysts’ goal-total selections. An all-market percentage that includes handicaps, match winners and accumulators cannot answer that question directly. First identify the selections that belong to the stated task, while retaining the full record as the context around that subset.
Use the same publication period and a clear selection unit. A single selection, a completed Series and an accumulator ticket are different units. A Series may close successfully after earlier legs lose. Calling that one successful observation and comparing it with individual match selections changes the meaning of the percentage.
The Series results explanation covers that counting distinction. Establish it before grouping by league, price or market. Otherwise even a carefully formatted comparison can be built on incompatible observations.
Follow a ranking reversal through the numbers
The following is a completely hypothetical teaching example. The categories are simplified price groups, and every entry is a settled binary selection. There are no pushes, partial settlements or unsettled fixtures.
| Group | Analyst A | Analyst B |
|---|---|---|
| Short-price selections | 72 wins from 90, or 80% | 9 wins from 10, or 90% |
| Long-price selections | 3 wins from 10, or 30% | 36 wins from 90, or 40% |
| All selections | 75 wins from 100, or 75% | 45 wins from 100, or 45% |
B has the higher observed win rate within both groups. A has the higher combined rate because A publishes far more selections in the higher-hit-rate category. Neither total is an arithmetic mistake. The figures answer different questions about differently composed records.
This illustrates a reversal often discussed under Simpson’s paradox. Look at the group counts and calculate the total from the actual wins and observations. A’s 75% is heavily influenced by 90 short-price selections. B’s 45% is heavily influenced by 90 long-price selections. Duke University’s statistics notes discuss how aggregation can reverse a comparison.
The example establishes neither analyst’s skill. Ten observations in a subgroup are particularly fragile, and the grouping ignores many differences within each category. Its purpose is to show why a headline percentage can reverse the apparent comparison underneath it.
Show actual weighting before trying an adjusted comparison
The actual combined rate is a weighted average. Each group’s weight is its share of the record. For A, the calculation is 0.90 × 80% + 0.10 × 30% = 75%. For B it is 0.10 × 90% + 0.90 × 40% = 45%.
An equal-weight comparison would instead give each group 50% influence. Under that invented mix, A scores 55% and B scores 65%. This can help isolate the effect of the chosen mix, but it describes a standardized comparison rather than either analyst’s actual published portfolio.
Keep the actual and standardized columns separate. Name the weights and explain why they serve your question. Equal weighting can be arbitrary when one category is rarely used by the service or unavailable to the reader. Choosing weights after seeing the results can simply manufacture the preferred winner.
Do not turn the standardized percentage into an imaginary profit record. It does not supply missing odds, stakes or selections. It is a transparent sensitivity check on one descriptive ranking.
Market labels alone do not make selections comparable
Two records labelled match winner can still contain very different price ranges. A home favourite at 1.25 and an away outsider at 6.00 have different break-even thresholds. A higher hit rate in the first set need not imply a stronger assessment of value.
Even exact price bands leave differences. Competitions, publication times, market liquidity and the quality of available information may vary. A broad band can hide concentration near its boundaries. Keep the original rows available so a reader can inspect those limitations.
Define bands before ranking the analysts. If you repeatedly move the cut-off until the result changes, the table becomes a search for a flattering story. Use a small number of meaningful groups and disclose the counts rather than subdividing a modest record into dozens of impressive-looking percentages.
For football markets, distinguish the settlement period and line. Over 2.5 and over 3.0 do not share the same refund possibilities. A 90-minute winner and a team to qualify also answer different questions. The group name should preserve those differences.
Add returns without erasing the selection mix
A strike rate counts successful outcomes. It does not measure the financial result. For a settled binary selection at decimal odds d, a one-unit win produces d − 1 units of net profit, while a loss costs one unit. Mixed prices require row-by-row accounting.
In the hypothetical short-price group, assume solely for illustration that every selection was priced at 1.20. A’s 72 wins return 86.4 units from 90 units staked, a net loss of 3.6 units. B’s nine wins return 10.8 units from ten staked, a net gain of 0.8 units. The 80% strike rate is compatible with a loss.
Those invented prices are separate assumptions, not facts inferred from the win table. Real records need their original prices. The price-audit guide explains why a quoted price and an attainable reader price should also remain separate.
When combining group ROI figures, weight by the relevant stakes, not automatically by selection counts. Ten large-stake entries can contribute more to turnover than 90 small-stake entries. Report the total stakes and net result alongside the ratio so the weighting is visible.
Keep omissions and uncertain entries in view
Check whether the market split changed during the review period. A service may start with mostly favourites and finish with more outsiders. A monthly total can then move because its composition changed, even without a corresponding change inside either group. Preserve the dated rows and compare both the group results and their shares over time. This makes a change of remit visible without pretending that the older and newer portfolios posed the same forecasting task.
A comparison based on the entries you can verify should say so. Missing historical prices prevent a complete return calculation. They should not be filled with an average price or silently removed while the headline is described as the full record.
List how many publications were observed, how many settled under each rule and how many have usable odds. Report pending, void and partial settlements separately. An entry excluded from a binary win-rate calculation can still belong in the publication ledger.
Original timing matters too. UK advertising guidance discusses advance proofing and odds availability when tipsters make past-performance claims. That provides a concrete reason to inspect the underlying record rather than rely on a retrospective selection of winners. It does not certify an individual OptaTips profile. ASA/CAP tipster guidance.
Separate description from an explanation of skill
A stronger result within an observed subgroup can reflect skill, luck, opponent mix, changing conditions or other differences. Standardizing one variable does not remove every source of bias. The table should narrow the question, not inflate the confidence of the answer.
Repeated fixtures create another limitation. Several selections tied to the same match need not provide independent evidence. A single unexpected result can affect multiple rows together. The overlap guide shows why counting more analyst names does not automatically diversify the underlying evidence.
Use a fixed follow-up period to see whether the descriptive pattern persists. Preserve the original grouping and note genuine changes in remit. A switch from top-flight match winners to lower-league totals may make an old aggregate less relevant to the service now being offered.
Leave a comparison another reader can challenge
A useful review can fit into a short note supported by a clear table. State the dates, selection unit, settlement convention, group definitions, counts, prices available and weighting method. Then distinguish the actual portfolio result from any standardized sensitivity check.
For the fictional example, the defensible conclusion is that A’s higher total win rate is driven by a much larger short-price share, while B has higher observed rates inside both simplified groups. That is already a more informative finding than declaring a winner from 75% versus 45%.
Before comparing services, use the OptaTips guides to clarify the record’s unit and calculation. If the required split is unavailable, describe the limitation. An unresolved comparison is preferable to a precise ranking built from incompatible evidence.
For adults aged 18 and over. All analysts, counts and prices in this article’s worked examples are fictional. Historical comparisons do not establish future profit.
コメント 0
まだコメントはありません。