Learn
Variance and Sample Size
The hardest lesson in betting: short-term results barely tell you anything.
Variance is the natural scatter of outcomes around their expected value. A prop with a true 55% Over rate will still go Under plenty — and can go Under five times in a row without anything being wrong. That randomness is exactly why betting is hard to read in the short term.
Short runs lie
Flip a fair coin ten times and you might see seven heads. That doesn’t make the coin biased. The same is true of bets: a winning week can come from a losing process, and a great model can have an awful month. The smaller the sample, the louder the noise.
How much sample do you need?
Enough that skill outweighs luck — usually hundreds to thousandsof bets, depending on edge size. A 2% edge is real money over a season, but it’s invisible over 20 bets. This is the flip side of expected value: EV only pays out across a large number of wagers.
What to do about it
- Judge your process (are the bets +EV?), not last week’s record.
- Keep bet sizes steady so a downswing can’t bust you — see bankroll management.
- Track results over a meaningful sample before drawing conclusions.
It’s also why we publish hit rates honestly on PropProphet — see the methodology. A model is only as good as its record over the long haul.
Frequently asked
How many bets before I know if I'm winning?
Often hundreds to a few thousand. A small edge is easily buried by variance over a few dozen bets, so a week or a month tells you very little.
Why did I lose money on +EV bets?
Variance. Positive expected value is a long-run average; short runs swing widely in both directions. Losing stretches are normal even when every bet was correctly priced.
Is a hot streak proof of skill?
Not on its own. A great week can be luck. Judge a process by its expected value and a large sample, not by recent results.