Skip to content
OneKitly

Dividing by n Minus One Is an Apology for Not Knowing the Mean

Published 11/17/2026 · 3 min read · Everyday calculators

Lena Hoffmann

Lena Hoffmann — Science & education writer at OneKitly

Mathematics · Physics

Checked against 2 sources

View profile →
In short

There are two standard deviations and the difference is one character in the denominator. Divide the squared deviations by n and you get the population version, 6.2237 for the eight numbers 78, 85, 92, 88, 76, 95, 82 and 89. Divide by n minus one and you get the sample version, 6.6534. The first is correct when the eight numbers are the whole group you care about; the second when they are a sample from something larger. Using the population formula on a sample understates the spread systematically, because the sample's own mean sits closer to its own points than the true mean does.

Advertisement

The same eight numbers give a population standard deviation of 6.2237 and a sample one of 6.6534 — 6.9% apart. Which is right depends on what the data is.

Where the missing degree of freedom went

The sample mean is computed from the same data whose spread is being measured, and that costs one piece of information. Knowing seven of the eight deviations and the mean determines the eighth, so only seven of them are free to vary. Dividing by seven rather than eight corrects for the fact that the mean was fitted rather than known. Every model that estimates parameters from data pays the same tax — a regression with three coefficients divides by n minus three — and the pattern is called degrees of freedom for exactly this reason.

When it stops mattering

Quickly. At ten observations the correction raises the standard deviation by 5.4%, at a hundred by 0.5%, at a thousand by 0.05%. Anyone arguing about the denominator on a dataset of ten thousand rows is arguing about the fourth decimal place. The argument is worth having on small samples, where the correction is large and the sample is exactly the case where every assumption matters most — two measurements corrected upward by 41% is not a rounding detail.

Variance raised by
How much the correction matters
Sample sizeVariance raised byStandard deviation raised by
2100%41.4%
1011.1%5.4%
1001.0%0.5%

Worked with our own calculator

Population standard deviation calculator

Given

Data
78, 85, 92, 88

Result

Population σ
5.117
Variance σ²
26.188
Mean
85.75
Count N
4

These figures are produced by the calculator below, not typed in by hand — they are recomputed whenever the tool changes.

Run it on your own figures →
Advertisement

Frequently asked questions

Which one does a spreadsheet use?
Both, under names that are easy to mix up. The plain function name usually means the sample version with n minus one, and a variant ending in P means the population version. Picking the wrong one is silent — no warning, no error, just a number a few per cent off — and it is one of the most common mistakes in spreadsheets that get shared. Writing the intended denominator into the cell comment costs nothing and saves the next reader a guess.
Is the corrected standard deviation unbiased?
The variance is; the standard deviation is not. Dividing by n minus one makes the variance unbiased, but taking a square root of an unbiased quantity does not give an unbiased result, so the sample standard deviation still runs slightly low. A further correction exists and almost nobody uses it, because the remaining bias is small and the formula is ugly. It is a good example of a convention that is defensible rather than perfect, kept because the alternative buys too little.

Articles you may find interesting

All guides →

Related tools

Sources

Spotted a mistake in this article?