What is the standard deviation of percentages?
The standard deviation of percentages tells you how much a rate like “60% hit rate” can wobble by luck alone. A percentage from a small sample is not a fact. It’s an estimate with an error bar around it, and that error bar is often far wider than bettors assume.
Here’s the version you actually need. If you observe a rate p over n tries, the standard deviation of that observed rate is:
SD = √( p × (1 − p) ÷ n )
That’s the standard error of a proportion. Same formula, different name depending on who’s teaching it. Plug in a 60% hit rate over 10 games and you get √(0.6 × 0.4 ÷ 10) = √0.024 ≈ 0.155, or about 15.5 percentage points.
Read that again. The wobble on a 10-game 60% is roughly 15 points. One standard deviation covers roughly 45% to 75%. Two covers about 29% to 91%. That is not a signal. That is noise wearing a signal’s clothes.
Why prop bettors need this formula
Every prop tool on earth, including ours, shows you hit rates. A hit rate is a percentage. So every hit rate carries an error bar, and knowing its size is the difference between reading data and reading tea leaves.
Take a real example from our own board. A StatsBench cheatsheet snapshot from August 23, 2026 listed Paige Bueckers rebounds over 3.5 in a Dallas vs. Seattle game at a 60% hit rate (6 of 10) with a best price of -161. Six for ten looks tidy. Run the formula and the honest read is “somewhere in the mid-40s to mid-70s, probably.”
That doesn’t make the prop bad. It makes the 60% a weak input on its own, which is exactly why price and matchup matter more than the headline number.
The one-line rule of thumb
You can skip the square root most of the time. Near 50%, the standard deviation of a rate is close to 50 ÷ √n percentage points.
- 10 games: about 16 points of wobble
- 25 games: about 10 points
- 100 games: about 5 points
- 400 games: about 2.5 points
To cut the error bar in half, you need four times the sample. That’s the brutal math behind why small samples never settle anything.
How to build an error bar around a hit rate
Three steps. No spreadsheet needed.
- Write the rate as a decimal. 60% becomes 0.60.
- Compute √(p(1−p)/n). For 0.60 over 20 games: √(0.24 ÷ 20) = √0.012 ≈ 0.110, so 11 points.
- Go two SDs each way. 60% ± 22 points, so roughly 38% to 82%.
That two-SD span is your rough 95% confidence interval. If the break-even rate for your price sits inside that span, you have not proven an edge. You’ve proven you need more data.
Here’s a table of the wobble on a 60% observed rate at different sample sizes:
| Games (n) | SD of the rate | Rough 95% range |
|---|---|---|
| 10 | 15.5 pts | 29%, 91% |
| 20 | 11.0 pts | 38%, 82% |
| 50 | 6.9 pts | 46%, 74% |
| 100 | 4.9 pts | 50%, 70% |
| 200 | 3.5 pts | 53%, 67% |
Even at 100 games, a 60% could honestly be a 50%. Which is break-even at -100 and a loser at anything juicier.
Comparing the error bar to your break-even rate
This is where the math pays rent. Every price has a break-even percentage, the rate you must hit just to stay flat. At -110 that’s 52.4%. At -161 it’s about 61.7%. At +160 it’s about 38.5%.
Now line the two up. That same August 2026 snapshot had Alec Bohm’s hits + runs + RBIs over 2.5 at 60% (6 of 10) with a best price of +160. Break-even at +160 is roughly 38.5%. The observed 60% sits about 21 points above it, and the 10-game standard deviation is around 15.5 points. So the gap is bigger than one SD but smaller than two.
Translation: interesting, not proven. That’s an honest read, and it’s the read a sharp bettor makes before clicking.
Compare that to the Bueckers rebounds line at -161. Break-even is about 61.7%, the observed rate is 60%, and the error bar swallows the difference whole. Same 6-of-10 record, completely different meaning, because price sets the bar the percentage has to clear. Run any price through our implied probability calculator to get that break-even number in a second.
Where percentages fool people
A few traps show up constantly in prop research.
Splitting the sample until it sings
“He’s 5 of 6 at home against left-handers on Sundays.” Six tries has an SD near 20 points. Slice a season thin enough and you will always find a 100% split. That’s not a trend, that’s a search.
Treating rates as independent when they aren’t
The formula assumes each try is a coin flip with a fixed probability. Real props aren’t. Minutes change, roles change, pitchers change. Correlated outcomes make the true wobble bigger than the formula says, not smaller. Our guide to correlation for bettors covers why that matters.
Ignoring that a line moves
A 70% hit rate on a 3.5 rebound line means nothing if the line is now 5.5. The percentage describes a threshold that no longer exists.
Averaging percentages of different sizes
A 60% over 5 games and a 40% over 50 games do not average to 50%. Weight by sample. Always.
Percentages, distributions and the normal curve
Once your sample gets reasonably large (roughly n×p and n×(1−p) both above 10), the spread of possible observed rates looks like a bell curve. That’s why the two-SD rule works: about 95% of the normal curve sits within two standard deviations of the center.
That’s also the link between “standard deviation” and the normal curve percentages people look up: 68% inside one SD, 95% inside two, 99.7% inside three. The standard deviation is just the ruler; the curve tells you what each tick mark is worth.
If you want the deeper version of that, we wrote about normal distribution in player props and reading z-scores on prop charts. Both build on the same ruler.
One caveat worth keeping: raw stat lines (points, kills, strikeouts) are often skewed, not perfectly bell-shaped. The rate converges to normal faster than the underlying stat does. So use the curve for hit rates, and be careful using it for the stat itself.
How StatsBench data handles the small-sample problem
We show hit rates because they’re useful. We also show what sits around them, because a rate alone is thin.
- Consistency Grades score how steady a player’s output is, 0 to 100. A 60% built on results hugging the line is different from a 60% built on wild swings.
- Edge % compares the price to a fair estimate. In the August 23, 2026 snapshot, several 6-of-10 props carried edges under 1% either way, which is the tool telling you the percentage isn’t buying much.
- Best price across books lowers your break-even bar. Cheaper juice means your true rate needs to clear less.
- SB Score blends the inputs so you’re not staring at one number in isolation.
The StatsBench Prop Finder lets you filter on all of it at once, including hit rate, consistency, trend and matchup. That’s the practical answer to error bars: stop leaning on one statistic.
A quick sanity checklist before you bet a percentage
- How many games is this rate built on? Under 20 means treat it as a hint.
- What’s the break-even rate at the price I’m getting?
- Is the gap between the two bigger than the error bar?
- Has the line moved since those games happened?
- Is the role stable (minutes, batting order, map pool)?
If four of five check out, you have something worth a small stake. If two do, you have a story.
The long-run point
Standard deviation isn’t only about samples of past games. It’s also about your own results. A bettor with a genuine 3% edge can sit underwater for hundreds of bets, because the wobble on your own win rate follows the exact same formula. We covered that honestly in why good bets lose.
That’s the whole discipline. Bet prices you can defend, size sensibly, and judge yourself over samples big enough to mean something. Positive expected value is a long-run edge, never a guarantee on any single night.
Betting should stay fun and stay within your means. 21+ where legal, and if it stops being entertainment, call 1-800-GAMBLER.
Put the math to work
Knowing the error bar is step one. Step two is finding props where the price gap is wide enough to survive it. Open the WNBA props board or the MLB strikeout research tool and sort by edge, not just by hit rate. Then run your price through the EV calculator and see whether the number actually holds up.
Frequently Asked Questions
How do you calculate the standard deviation of a percentage?
Use SD = √(p × (1 − p) ÷ n), where p is the rate as a decimal and n is the number of tries. For a 60% hit rate over 10 games, that’s √(0.6 × 0.4 ÷ 10) ≈ 0.155, or about 15.5 percentage points.
Is a 60% hit rate on 10 games meaningful?
Barely. The standard deviation on 10 tries is roughly 15 points, so a true rate anywhere from about 45% to 75% would produce that result routinely. Treat it as a starting hint, not evidence.
How many games do I need before a hit rate means something?
It depends on how big the gap is between your rate and the break-even price. As a guide, 25 games gives roughly a 10-point error bar and 100 games roughly 5 points. Cutting the error bar in half needs four times the sample.
What’s the difference between standard deviation and standard error?
Standard deviation describes spread in a set of values. Standard error describes how much a computed statistic, like an observed percentage, would bounce around across repeated samples. For proportions the formula √(p(1−p)/n) is the standard error.
Does a bigger sample make a bet safer?
It makes your estimate more reliable, not the bet safer. A single wager still swings on one game. Larger samples reduce the chance you’re fooled by noise when deciding whether an edge exists.