The margin is not spread evenly, and everyone pretends it is
A bookmaker quotes 1.25 / 6.00 / 13.00. Those prices imply 80.0% + 16.7% + 7.7% = 104.36%, so there is 4.36% of margin sitting on top. To find what the book really thinks, you have to take that 4.36% back out.
Almost everybody does it the same way: divide each implied probability by 104.36%. It is one line of arithmetic, every spreadsheet does it, and it is what the overwhelming majority of devig calculators do silently without naming the method.
It also assumes something that is not true. Dividing by the booksum removes the margin in proportion to each outcome's probability — the favourite carries most of the load because it has most of the probability. Bookmakers do not price that way. The favourite-longshot bias is one of the oldest documented regularities in betting markets: margin is loaded disproportionately onto long prices, because that is where the recreational money goes and where a customer is least able to tell 13.00 from 15.00. If you remove margin evenly from a book that was built unevenly, you get the favourite roughly right and the outsider badly wrong.
The margin calculator tells you how much vig is in a book. This is about where it sits.
Four ways to take it off
There is no single correct answer, because removing margin means guessing how the bookmaker added it. Four models are in common use.
Proportional (also called multiplicative, or normalising): divide every implied probability by the booksum. Assumes margin scales with probability. The default almost everywhere.
Equal margin (additive): subtract the same slice from each outcome — the total margin divided by the number of outcomes. Assumes margin is a flat charge per outcome. Simple, and it falls apart on long prices, where the flat slice can be most of the outcome's probability.
Power: find the exponent k that makes the probabilities sum to 1 when each is raised to it. Because raising a small number to a power above 1 shrinks it proportionally more, this takes relatively more margin off longshots than off favourites — the direction the favourite-longshot bias says it should go.
Shin: model the margin as the bookmaker's protection against a fraction of informed money, and solve for that fraction. It was derived from an actual theory of why the margin exists rather than fitted to the shape of the answer, and in practice it lands between proportional and power.
All four produce probabilities that sum to exactly 100%. They disagree about which outcome gives up what.
The same book, four answers
Here is what each method says about the same four markets. The margin is identical in every row of a block — only its distribution changes.
Balanced, 2.10 / 3.40 / 3.60, margin 4.81%
• Proportional: 2.20 · 3.56 · 3.77
• Shin: 2.18 · 3.59 · 3.81
• Power: 2.17 · 3.60 · 3.82
• Spread on the outsider: 1.25%
Strong favourite, 1.25 / 6.00 / 13.00, margin 4.36%
• Proportional: 1.30 · 6.26 · 13.57
• Shin: 1.28 · 6.50 · 15.27
• Power: 1.27 · 6.78 · 15.48
• Spread on the outsider: 18.14%
Mismatch, 1.05 / 15.00 / 34.00, margin 4.85%
• Proportional: 1.10 · 15.73 · 35.65
• Shin: 1.08 · 18.88 · 55.76
• Power: 1.06 · 24.61 · 64.78
• Spread on the outsider: 111.58%
Look at the favourite column: 2.20 to 2.17, 1.30 to 1.27, 1.10 to 1.06. The method barely matters. Now the outsider: 3.77 to 3.82, then 13.57 to 15.48, then 35.65 to 64.78. The entire disagreement piles up on exactly the outcome a value bettor is most likely to be pricing. Every number above comes straight out of the Draw No Bet and Double Chance calculator, which reports that spread as a result of its own.
The bet that changes sign
Take the strong-favourite book: 1.25 / 6.00 / 13.00. You are interested in the outsider, and another bookmaker is offering 14.50 on it.
Devig proportionally and the fair price is 13.57, so a true probability of 7.37%. At 14.50 the expected value is 14.50 × 0.0737 − 1 = +6.9%. That is a clear bet by anyone's standard, big enough to stake confidently.
Devig by Shin and the fair price is 15.27, a true probability of 6.55%. At 14.50 the expected value is 14.50 × 0.0655 − 1 = −5.0%. Not a marginal bet. A losing one.
Same bookmaker, same offer, same arithmetic afterwards. The only difference is an assumption nobody wrote down, and it moved the answer by nearly twelve points of expected value. A 2% edge, which is a perfectly respectable real edge, is noise next to that.
This is why expected value is only ever as trustworthy as the fair price you fed it. The EV calculation is not where bettors go wrong — it is two steps earlier.
So which one should you use?
Honestly: nobody can hand you the right answer, because the right answer depends on how the specific bookmaker built that specific book, and they do not publish it. Anyone who tells you one method is simply correct is selling something.
What the research supports is narrower and more useful. Shin and the power method both beat proportional at predicting outcomes when tested against large samples of settled markets, and they beat it by more on exactly the long prices where proportional is weakest. Shin has the advantage of coming from a model of why a margin exists. Equal-margin is the weakest of the four and is included here mainly so you can watch it break.
A sharper way to think about it: do not pick a method and trust it. Run the book through more than one and look at how far apart they land. If the spread on the outcome you care about is 1%, take any of them and move on. If it is 18%, the method is now a bigger input to your decision than your opinion of the match is, and no amount of care downstream will fix that. If it is 100%, you do not have a fair price at all — you have a range, and you should want a price comfortably outside the whole range before you bet.
The existing no-vig calculator uses the proportional method, which is the right call for the two-way markets it serves: with two outcomes near even money there is almost nothing for the methods to disagree about.
What to do with this
Three practical consequences.
First, stop quoting fair odds without naming the method. A fair price is not a fact about a market, it is the output of a model, and leaving the model unstated is how a 12-point swing in expected value hides in plain sight.
Second, be suspicious of your own longshot edges. If your value bets cluster on prices above 10.00 and you devig proportionally, a meaningful share of that edge is an artefact of the method rather than a property of the market. Re-price a few of them by Shin and see how many survive.
Third, let reality arbitrate. Methods argue; closing lines settle it. If your devigged numbers are genuinely better than the market's, your bets will beat the closing price more often than not, and if they are an artefact they will not. Closing line value will tell you which within a few hundred bets, long before the profit curve does — and unlike your fair-odds model, the closing line is not something you can accidentally tune to agree with you.