Statistics Reference
Standard Error vs Standard Deviation
The two statistics share a word, a symbol style, and a formula connection — and answer completely different questions. Standard deviation describes how much individuals vary. Standard error describes how much an estimate would vary if you repeated the study. Mixing them up changes conclusions, chart interpretations, and confidence intervals, so this guide pins down the difference with one dataset carried all the way through.
Two Different Questions
Standard deviation (SD)
“If I pick one individual from this data, how far from the average is it likely to sit?” SD is a description of the data itself — the typical distance of observations from their mean. It has the same units as the data and a fixed true value that more data merely estimates better.
Standard error (SE)
“If I redid this whole study with a fresh sample, how much would my computed mean move?” SE is a description of an estimate's precision — the standard deviation of the sampling distribution of that estimate. It shrinks as the sample grows.
The bridge between them is the square root law: for a mean of n independent observations, SE = SD / √n. Averaging cancels individual randomness, and it cancels it at the rate √n — which is why precision is expensive: halving the SE takes four times the data.
Side by Side
| Standard deviation | Standard error | |
|---|---|---|
| Describes | Spread of individual values | Precision of an estimate (usually the mean) |
| Formula | s = √(SS / (n − 1)) | SE = s / √n |
| As n grows | Settles toward the fixed population value | Shrinks toward zero |
| Used for | Describing data, z-scores, reference ranges | Confidence intervals, test statistics, comparing means |
| Error bars show… | How variable the observations are | How trustworthy the plotted mean is |
One Dataset, Both Numbers
Take the five values 4, 8, 6, 5, 12 (the same set used across this site's calculators, so every step can be checked).
- Mean: (4 + 8 + 6 + 5 + 12) / 5 = 7.
- Sum of squared deviations: 9 + 1 + 1 + 4 + 25 = 40.
- Standard deviation: s = √(40 / 4) = √10 ≈ 3.1623. A typical observation sits about 3.2 units from the mean.
- Standard error: SE = 3.1623 / √5 ≈ 1.4142. A repeat of the whole study would typically move the mean by about 1.4 units.
Same data, two very different numbers, and both are correct answers — to different questions. Reporting “7 ± 3.2” describes people; “7 ± 1.4” describes the estimate. A reader cannot tell which claim is being made unless the ± is labeled.
Why the Distinction Changes Decisions
Reading charts. SE bars are √n times narrower than SD bars. With n = 100, that is a factor of 10 — a plot of barely-overlapping SE bars can come from populations whose individuals overlap almost completely. Neither is wrong; they show different things.
Building intervals. A 95% confidence interval for the mean is roughly mean ± 2 × SE. Using the SD by mistake produces an interval √n times too wide — it stops being a statement about the mean at all.
Combining groups. When several groups are assumed to share one spread, their SDs are combined with the pooled standard deviation — weighted on the variance scale, not averaged — and that pooled SD then feeds the standard errors used by two-sample t-tests.
Scaling studies. Doubling data does not calm individuals (SD stays put); it calms estimates (SE drops by √2). Planning a study to reach a target precision is therefore an SE calculation — which is exactly what a sample size calculator automates.
Try the Standard Error Calculator
Compute the standard error of a mean or proportion from raw data or summary statistics, with the SD-to-SE step shown explicitly.
Frequently Asked Questions
What is the main difference between standard error and standard deviation?
Standard deviation describes the spread of individual data values around their mean; standard error describes how much a sample statistic (usually the mean) would vary across repeated samples. SD is a property of the data, SE is a property of an estimate. They are linked by SE = SD / sqrt(n), so the SE is always smaller for any sample with more than one observation.
Is standard error just standard deviation divided by the square root of n?
For the mean of independent observations, yes: SE = s / sqrt(n). But 'standard error' is a general term - every statistic has one. The standard error of a proportion is sqrt(p(1-p)/n), the standard error of a regression slope has its own formula, and so on. In every case it means the same thing: the standard deviation of that statistic's sampling distribution.
Does the standard deviation get smaller with a larger sample?
No - and this is the key conceptual difference. The population has a fixed spread, and a larger sample only estimates that fixed value more accurately; the estimate settles, it does not shrink. The standard error, by contrast, genuinely shrinks with sample size because averaging more values cancels more random noise. If someone reports spread that keeps falling as n grows, they are reporting a standard error.
Should error bars show standard deviation or standard error?
It depends on the claim. Use SD bars to show how variable individual observations are, and SE bars (or better, confidence intervals) to show how precisely the group mean is estimated. SE bars are always sqrt(n) times narrower, which can make noisy data look deceptively clean - a chart should always label which one it displays.
Which one goes into a confidence interval?
The standard error. A confidence interval for a mean is the estimate plus or minus a critical value times the SE: for large samples, roughly mean +/- 1.96 x SE for 95% confidence. Using the SD there would produce an interval for individual values (a prediction-style range), not for the mean - a common and consequential mix-up.
Can I compute the standard deviation from a reported standard error?
Yes, if the sample size is reported: SD = SE x sqrt(n). This is routinely needed in meta-analysis, where some papers report SD and others SE. For example, a reported SE of 1.2 from n = 64 implies SD = 1.2 x 8 = 9.6. Without n, the conversion is impossible, which is one more reason sample sizes belong in every table.