Skip to content

Sample Size Calculator

How many bets before your results actually prove anything? The answer is larger than almost anyone expects.

Break-even here is 52.38% — the rate you are trying to beat.

%

Be honest. Optimism here shortens the answer and lengthens the reality.

Optional — converts the answer into time.

Bets required

5,689

To distinguish a 2.62% edge from break-even at 95% confidence. At 10 bets a week that is 10.9 years. This is why almost nobody proves an edge from results alone.

Break-even rate
52.38%
Your expected rate
55.00%
Edge to detect
+2.62%
Time to get there
10.9 years

At 10 bets a week.

Confidence / power
95% / 80%

Confidence is the chance of a false positive; power is the chance of detecting a real edge that exists.

How the requirement scales with edge size
EdgeWin rateBets neededAt 10/week
+0.5%52.9%156,542301.0 yrs
+1.0%53.4%39,11375.2 yrs
+2.0%54.4%9,76618.8 yrs
+3.0%55.4%4,3348.3 yrs
+5.0%57.4%1,5543.0 yrs

The requirement scales with the square of the edge: halving the edge quadruples the bets. This assumes independent bets at a constant price and a constant edge — none of which holds in practice, so the real figures are larger. Which is the case for tracking closing line value instead: it reads on every bet whether it won or lost, and converges in a fraction of the sample.

Runs entirely in your browser · nothing is sent or saved

How it works

You think you win at 55% against prices that need 52.4%. How many bets before that claim is distinguishable from luck? This is the standard statistical planning calculation, applied to the question every bettor eventually asks.

Two settings control the answer. Confidence is how unlikely you want a false positive to be — declaring an edge when there is none. Power is the chance of detecting a real edge that genuinely exists. Both cost sample: demanding 99% confidence instead of 90% roughly doubles the bets required.

The dominant term is the edge itself, and it enters as a square. Halve the edge and the requirement quadruples. A five-point edge might be provable in a few hundred bets; a one-point edge — which is still a professional-grade edge — takes tens of thousands. That relationship is the single most important thing on this page.

The practical consequence is that most bettors will never prove an edge from their own results. At ten bets a week, a realistic 2.6-point edge takes years to establish, during which the edge itself will have drifted, the markets will have changed, and the sample will no longer be measuring one consistent thing.

Which is why closing line value is the metric serious bettors actually track. It reads on every bet regardless of the outcome, so it converges in a small fraction of this sample and answers the same underlying question: is this process finding something the market has not?

The formula

α  = 1 − confidence
z_α = the two-sided normal quantile at 1 − α/2
z_β = the normal quantile at the power
p̄  = (p₀ + p₁) / 2

n ≈ [ z_α√(2p̄(1−p̄)) + z_β√(p₀(1−p₀) + p₁(1−p₁)) ]²
    ────────────────────────────────────────────────
                     (p₁ − p₀)²

p₀ = break-even rate at your price
p₁ = the rate you believe you have

p₁ = p₀  →  no sample size suffices

The (p₁ − p₀)² in the denominator is why edge size dominates everything else — halving the edge quadruples the requirement.

A worked example

You bet −110 lines and believe you win at 55%. Break-even is 52.38%, so the edge is 2.62 points. At 95% confidence and 80% power, you need 5,689 bets.

At ten bets a week that is nearly eleven years — during which the edge has to hold, the markets have to stay comparable, and you have to keep betting the same way. A 2.62-point edge is a strong one, and a decade of betting is what it costs to prove it from results.

Now suppose the real edge is one point rather than 2.62. The requirement rises to 39,113 bets — seventy-five years at the same rate. A one-point edge is genuinely profitable and, from results alone, unprovable within a lifetime.

This is not an argument against tracking results. It is an argument for tracking closing line value alongside them, because the results will not answer the question within any useful timeframe.

Common questions

Why do I need so many bets?
Because betting edges are tiny relative to the variance. Beating −110 means winning 55% instead of 52.4%, and separating those two rates statistically requires the sample to overwhelm the noise — which takes thousands of trials.
What is the difference between confidence and power?
Confidence is the chance of falsely declaring an edge that is not there. Power is the chance of detecting a real edge that is. Both cost sample size, and 95% confidence with 80% power is the usual convention.
Why does a smaller edge need so many more bets?
Because the edge enters the formula squared. Halving the edge quadruples the sample required, so a one-point edge takes roughly seven times the bets of a 2.6-point one.
Is there a faster way to tell if I am winning?
Closing line value. It gives a reading on every bet whether it won or lost, so it converges in a fraction of this sample. It is the metric professional bettors actually track for exactly this reason.
Does this account for varying odds and stakes?
No — it assumes independent bets at a constant price and a constant edge. Real betting has none of those, so the real requirement is larger than the figure shown. Treat it as an optimistic floor.

The guide behind this calculator

Related calculators

For informational and analytical purposes only. These tools do not predict outcomes and do not recommend wagers.