"I have a VPIP of 30 over 100 hands, am I playing too loose?" Maybe, maybe not. With few hands, poker stats swing a lot, and changing your game because of a number that doesn't mean anything yet is as bad as not looking at the numbers at all.
The problem: the sample
Every stat is a proportion: of the times you could do something, how many times you did. If you've had 100 chances to enter the pot and you entered 30 times, your measured VPIP is 30%. But your real VPIP (the one you'd have in the long run playing the same way) can be quite different: over 100 hands, the luck of the cards you were dealt weighs a lot.
The confidence interval
The honest way to give a number is with its confidence interval: the range where your real value almost certainly (95%) lies. With the Wilson method, which works well with few hands:
| What you measure | Measured value | 95% interval |
|---|---|---|
| 30 of 100 opportunities | 30% | 21.9% to 39.6% |
| 300 of 1,000 opportunities | 30% | 27.2% to 32.9% |
| 9 of 30 opportunities | 30% | 16.7% to 47.9% |
Over 100 hands, a measured VPIP of 30% fits a tight player (22%) just as well as a very loose one (40%). Over 1,000, you already know fairly well where you stand.
Opportunities aren't hands
Careful: what counts isn't the hands you played but each stat's opportunities. For VPIP almost every hand is an opportunity, but for fold to 3-bet only the times you opened and got re-raised count, which might be one hand in 30. That's why preflop stats stabilize quickly and stats for specific spots need many more hands.
When is it really a leak?
A simple, solid rule: a stat is truly out of range when its whole confidence interval falls outside the healthy range. If only the measured value is outside but the interval still touches the healthy range, it's a trend: keep an eye on it, but don't rebuild your game around it.
Your winrate needs far more hands
With your winrate (bb/100) it's even worse, because every hand can win or lose hundreds of blinds. With a typical standard deviation of about 90 bb/100, the 95% margin of error is:
| Hands | Winrate margin of error |
|---|---|
| 10,000 | ±17.6 bb/100 |
| 50,000 | ±7.9 bb/100 |
| 100,000 | ±5.6 bb/100 |
In other words: over 10,000 hands, someone who "wins 5 bb/100" could be a losing player. To improve, look at your decisions (the stats above) before your results.
Flopmortem does it for you
Flopmortem calculates the confidence interval of every stat and shows it next to the number: it separates confirmed leaks from trends and doesn't flag anything as a leak until there's a minimum sample. That way you know what to work on and what not to yet.