Sample Size And Significance

Betting records are almost always too small to support the claims made about them. Understanding how large a sample an edge actually needs is the fastest way to become immune to marketing — including your own optimism about your own results.

The Numbers

To distinguish a 55% bettor from a break-even 52.4% bettor at conventional statistical confidence takes roughly 1,000 to 1,400 bets.

The smaller the edge, the larger the sample. A 53.5% bettor — still genuinely profitable — needs several thousand.

Read the 100-bet row carefully. A genuine 55% bettor can show 45% over a hundred bets, and a genuine break-even bettor can show 62%. A hundred-bet record cannot distinguish between them, which is why hundred-bet records are the ones most often advertised.

Why Advertised Records Are Meaningless

Two mechanisms make published records uninformative even when the individual numbers are true.

Selection. Given a thousand handicappers with no skill, chance alone produces a handful with excellent hundred-bet records. Those are the ones with websites. The rest are not lying about their results; they simply stopped advertising.

Choice of window. Any record long enough contains a flattering sub-period. "27-11 over our last 38 plays" is a statement about where the window was drawn, not about ability.

The defensible standard is a continuous, dated, fully disclosed record over hundreds of plays, graded at the price subscribers actually received. Anything short of that is a marketing artefact. See how to read a betting record and how to choose a service.

What To Look At Instead

Since results converge too slowly to be useful in real time, evaluate inputs.

Closing line value. Whether plays consistently beat the closing number. This is measurable in tens of bets and is the strongest short-run indicator that exists. See closing line value.

Whether the reasoning is stated and specific. A documented method can be evaluated on its own terms. "Trust the process" cannot.

Whether the claims are calibrated. Anybody advertising 65% or 70% long-run win rates is describing something that does not exist, which is information about them regardless of their record.

Whether losses are disclosed. A record without visible drawdowns is not a record of real betting. See variance and drawdowns.

Applying This To Yourself

The same arithmetic applies to your own results, and it is harder to accept there.

After 100 bets you know essentially nothing about your edge. After 500 you have a rough idea. Any conclusion drawn from twenty bets — that you are good at NBA totals, that a system works, that you should raise your stakes — is a conclusion about noise.

Which is genuinely difficult, because betting produces frequent, vivid feedback that feels informative and mostly is not. Holding a method steady through a sample too small to judge it is the discipline that most of the arithmetic on this site is ultimately about.

Frequently Asked Questions