Standard Deviation
The variance
captures spread perfectly — but in the wrong units. Square centimetres, square marks, square
seconds: quantities nobody can picture. Imagine a weather report announcing "today's temperatures
varied by 9 degrees squared." You'd have no idea whether to grab a coat.
The fix is delightfully simple: take the square root, and you are back in the data's own units.
That square root is the standard deviation — and it reads as one plain, useful
sentence: the typical distance a data point sits from the average.
s = \sqrt{\sigma^2} = \sqrt{\frac{1}{n}\sum_{i=1}^{n}\left(x_i - \bar{x}\right)^2}.
If the data are heights in centimetres, the standard deviation is in centimetres too — a number
you can picture and lay right alongside the values themselves.
What it means: the typical distance from the mean
Read the standard deviation as the typical distance of a data point from the
mean. A small s says the values huddle close to the
centre; a large s says they wander far from it. It is the single
most quoted measure of spread precisely because it is interpretable: "scores averaged
70, give or take about 8" is a standard
deviation talking.
Worked example: from variance to standard deviation
Earlier we found a set of five test scores 2,4,4,5,5 had variance
\sigma^2 = 1.2. The standard deviation is just its square root:
s = \sqrt{1.2} \approx 1.10.
So the scores sat, on average, about 1.1 marks from their mean of
4 — a number in marks, the same units as the data, that you
can hold up against the scores directly. That is the whole trick: variance does the heavy lifting,
the square root makes it human-readable.
Worked example: reading a real result
A class sits an exam. The mean is 70 and the standard deviation is
8. What does that tell a student?
"Typical distance from the mean is 8" means most students landed
roughly one standard deviation either side of 70 — that is, between
about 62 and 78. A score of
86 is a full two standard deviations above the mean:
genuinely unusual, near the top of the class. A score of 54 is two
below: unusually low. The standard deviation turns a bare number into a ruler
for judging how remarkable any single result is.
Worked example: which class is more consistent?
Two classes both average 70 on the same test. Class A has
s = 4; Class B has s = 12. Same mean —
so which teaching was steadier?
Class A's smaller standard deviation says its scores clustered tightly around
70: almost everyone did roughly the same. Class B's larger
s says its scores were flung wide — some soared, some struggled. When
two data sets share a mean, the one with the smaller standard deviation is the more
consistent, more reliable one. This single comparison is why s
is quoted everywhere from clinical trials to factory quality control.
In finance, the standard deviation of an investment's returns has its own nickname:
volatility. A savings account whose return barely wobbles has a tiny standard
deviation — low volatility, low risk. A speculative stock whose price lurches up and down month
to month has a large one — high volatility, high risk. Two funds might boast the exact same
average return, yet one lets you sleep at night and the other doesn't, and it's the
standard deviation that tells them apart. Whole theories of building safe portfolios are, at
heart, arguments about how to keep this one number small.
See it: one step either side of the mean
The dots are a fixed data set with mean 5 (the dashed line). The band
stretches from \bar{x} - s to \bar{x} + s
— one standard deviation in each direction. Slide s until the band
comfortably covers the bulk of the points: for this set the true standard deviation is
2, and a band of that width swallows most of the data, just as a
"typical distance" should.
Standard deviation versus variance
They are two views of the same spread. The variance is the natural quantity in the algebra (it
adds cleanly, and it is what later theory is written in); the standard deviation is the natural
quantity for a human (right units, "typical distance"). Always:
s = \sqrt{\sigma^2} and \sigma^2 = s^2.
- The standard deviation is the square root of the variance: s = \sqrt{\sigma^2}.
- It is measured in the same units as the data, unlike the variance.
- It reads as the typical distance of a value from the mean.
- Squaring it returns the variance: \sigma^2 = s^2.
Standard deviation is not resistant to outliers. Because it is built from
squared deviations, one freakish value pulls it up dramatically — the squaring blows a
single far-flung point far out of proportion. Nine people earning around
\$30\text{k} and one billionaire in the room will produce a standard
deviation so huge it describes nobody in the room. For skewed data with outliers,
the standard deviation gives a misleading picture of "typical" spread, and a more robust measure
like the interquartile
range — which ignores the extreme tails entirely — is the more honest choice. Use
s for roughly symmetric data; reach for the IQR when a few wild values
threaten to hijack it.
Standard deviation isn't just a spread measure — it is the natural
ruler of the normal
distribution, the bell-shaped curve that data pile into again and again in nature.
On that curve the numbers are astonishingly tidy: about 68% of everything lands
within one standard deviation of the mean, about 95% within two, and about
99.7% within three — the famous
empirical
rule. Mark the mean, step out in units of s, and you can
read off what fraction of the world lies where. That deep link to the bell curve is why
standard deviation, out of all the possible spread measures, became the dominant one across
science, finance, and quality control.
See it explained