What standard deviation is trying to capture
Two data sets can have the exact same average and still look completely different — one tightly clustered around that average, the other scattered widely. The mean alone can't distinguish between them. Standard deviation is the measure built specifically to answer that missing question: on average, how far do individual values sit from the mean? A small standard deviation means most values sit close to the mean; a large one means values are spread out widely.
Working through the calculation
Take a small data set: 2, 4, 4, 4, 5, 5, 7, 9. The calculation happens in a specific order, and each step exists for a reason.
Step 1: Find the mean. Add the values and divide by how many there are: (2+4+4+4+5+5+7+9) / 8 = 5.
Step 2: Find each value's distance from the mean. Subtract the mean from each value: -3, -1, -1, -1, 0, 0, 2, 4.
Step 3: Square each distance. 9, 1, 1, 1, 0, 0, 4, 16. Squaring is the step that most needs explaining — see below.
Step 4: Average the squared distances. This average is called the variance: (9+1+1+1+0+0+4+16) / 8 = 4.
Step 5: Take the square root of the variance. √4 = 2. This final number, 2, is the standard deviation — it's back in the original units (unlike variance, which is in squared units), which is why standard deviation, not variance, is the number usually reported and interpreted.
Why square the distances instead of just averaging them
Simply averaging the raw distances from the mean (step 2's values, without squaring) always produces zero — the positive and negative distances cancel out perfectly by the mathematical definition of the mean. Squaring solves this by making every distance positive before averaging, so they can't cancel out. Taking the square root at the end (step 5) undoes the distortion squaring introduces to the scale, bringing the final answer back into the same units as the original data instead of leaving it in squared units.
An alternative — averaging the absolute value of each distance instead of squaring — would also avoid the cancellation problem, and this measure exists too (it's called mean absolute deviation). Standard deviation is used far more often because squaring has useful mathematical properties for further statistical work — particularly that it connects directly to the normal distribution and to variance-based statistical tests — that absolute-value-based measures don't share.
Reading a standard deviation once it's calculated
For data that's roughly normally distributed (the familiar bell curve), standard deviation has a well-known practical interpretation, sometimes called the 68-95-99.7 rule: about 68% of values fall within one standard deviation of the mean, about 95% fall within two, and about 99.7% fall within three. This is what makes a number like "standard deviation = 2" actionable rather than abstract — for the earlier example (mean 5, standard deviation 2), roughly two-thirds of similarly-distributed values would be expected to fall between 3 and 7.
Population versus sample standard deviation
One detail matters when calculating standard deviation from a sample (a subset of a larger population) rather than an entire population: step 4's averaging divides by n - 1 instead of n, a correction called Bessel's correction. Dividing by the full count n when working from a sample systematically underestimates the true population standard deviation, because a sample's own mean is calculated to fit that specific sample as closely as possible, which artificially shrinks the apparent spread. Dividing by n - 1 instead corrects for this bias. Most statistical software defaults to the sample version (n - 1) unless told explicitly that the data represents an entire population.
Why this measure shows up everywhere
Standard deviation's usefulness comes from converting "spread" — a vague, qualitative idea — into a single comparable number attached to real units, applicable to test scores, measurement error, financial returns, or any other numeric data. Once it's calculated, questions like "is this particular value unusually high?" or "how consistent is this process?" get concrete, comparable answers instead of subjective ones.