Standard Error vs Standard Deviation — What Each One Measures
Standard deviation and standard error look almost the same on paper, are calculated from the same numbers, and are printed after the same ± sign — yet they answer two completely different questions. Quoting the wrong one is one of the most common faults in school practical files and in undergraduate lab reports. This article works through a single set of titration readings and computes both, so the difference stops being a definition to memorise and becomes something you can see. (If you need the basics of mean, median and mode first, that ground is covered in the separate article linked at the end.)
The one-line difference
Standard error answers: how uncertain is the average I calculated from them?
Standard deviation is a property of the data. Standard error is a property of the estimate. Take more readings and the standard error shrinks, because you know the average better. The standard deviation does not systematically shrink — the readings are as scattered as the method makes them, no matter how many you take.
The formulas
Sample standard deviation: s = √[ Σ(x − x̄)² ÷ (n − 1) ]
Standard error of the mean: SE = s ÷ √n
Relative standard deviation: %RSD = (s ÷ x̄) × 100
What each symbol means
| Symbol | Meaning |
|---|---|
| x | An individual reading |
| x̄ | The mean of the sample you actually measured |
| μ | The true mean of the whole population — usually unknown |
| n | Number of readings in your sample |
| n − 1 | Degrees of freedom (Bessel's correction) — see below |
| s | Sample standard deviation, the everyday one |
| SE | Standard error of the mean, sometimes written SEM |
Why divide by n − 1 and not n?
Because you used the same data twice. The mean x̄ was itself calculated from those readings, so the deviations are measured from a centre that has already been nudged towards them. That makes the squared deviations slightly too small, and dividing by n − 1 instead of n corrects the bias. When the true mean μ is known independently — which is rare — you divide by N and the correction is not needed. In exams: if the data is a sample, use n − 1.
Worked example — five titration readings
A burette gives these titre values (mL): 24.8, 25.0, 25.2, 24.9, 25.1.
Step 1 — the mean.
Sum = 24.8 + 25.0 + 25.2 + 24.9 + 25.1 = 125.0
x̄ = 125.0 ÷ 5 = 25.00 mL
Step 2 — the deviations and their squares.
| Reading x (mL) | x − x̄ | (x − x̄)² |
|---|---|---|
| 24.8 | −0.2 | 0.04 |
| 25.0 | 0.0 | 0.00 |
| 25.2 | +0.2 | 0.04 |
| 24.9 | −0.1 | 0.01 |
| 25.1 | +0.1 | 0.01 |
| Σ(x − x̄)² | 0.10 | |
A useful check before going further: the deviations themselves must add to zero. −0.2 + 0.0 + 0.2 − 0.1 + 0.1 = 0 ✔ If they do not, the mean is wrong.
Step 3 — sample standard deviation.
s = √(0.10 ÷ 4) = √0.025 = 0.158 mL
Step 4 — standard error of the mean.
SE = s ÷ √n = 0.158114 ÷ √5 = 0.158114 ÷ 2.2361 = 0.0707 mL
So the readings scatter by about 0.16 mL, but the average of five of them is pinned down to about 0.07 mL. Report the result as 25.00 ± 0.07 mL (standard error, n = 5) — and always state which quantity the ± refers to, because the reader cannot tell otherwise.
The same data, treated as a population
If those five values were the entire population rather than a sample:
σ = √(0.10 ÷ 5) = √0.02 = 0.141 mL, against s = 0.158 mL.
The population figure is smaller because it divides by the larger number. With only five readings the gap is about 12%; with fifty readings it becomes negligible. This is exactly why the two buttons marked σ and s on a scientific calculator give different answers from the same list — neither is broken.
What happens when you take more readings
Suppose the experiment is repeated until there are 20 readings, and the scatter stays the same, so s is still about 0.158 mL.
SE = 0.158114 ÷ √20 = 0.158114 ÷ 4.4721 = 0.0354 mL
Quadrupling the number of readings has exactly halved the standard error, because SE goes as 1/√n. That is the honest arithmetic of improving a measurement: to halve the uncertainty you need four times the work, and to reduce it tenfold you need a hundred times the work. The standard deviation, meanwhile, has not improved at all — a better method, not more repeats, is what reduces scatter.
Turning standard error into a confidence interval
A standard error becomes an interval once it is multiplied by a factor from the t-distribution for your degrees of freedom. For n = 5 (4 degrees of freedom) the two-tailed 95% value is 2.776:
With a large sample the factor approaches 1.96 and the interval would be 25.00 ± 0.14 mL. Use the t-value for small samples — using 1.96 with five readings understates the uncertainty. The exact t-values come from a statistical table, so check the one supplied with your paper rather than remembering a number.
Relative standard deviation — comparing across different quantities
An absolute spread of 0.158 mL means nothing until you know whether the titre was 25 mL or 2.5 mL. The relative form fixes that:
Under 1% is respectable schoolwork for a burette read to the nearest 0.05 mL. Because it is dimensionless, %RSD lets you compare the precision of a titration with the precision of a weighing.
The caution that matters most
Neither figure says anything about whether your answer is right. Both measure precision — how closely repeats agree — and are completely blind to accuracy. A burette with a mis-set zero, or a solution standardised against a wrong concentration, will give five beautifully consistent readings and a tiny standard error around a value that is simply wrong. Small scatter is evidence of a careful technique, never proof of a correct result. Systematic error is found by calibration and by a second independent method, not by more repeats.
Other common mistakes
- Printing ± without saying what it is. "25.00 ± 0.07" is ambiguous. State standard deviation, standard error or confidence interval, and give n.
- Dividing by n for sample data. It quietly understates the spread — use n − 1 unless you genuinely have the whole population.
- Expecting more repeats to shrink the standard deviation. They shrink the standard error. The scatter is a property of the method.
- Quoting standard error to make results look precise. SE is always smaller than s, so choosing it for that reason is misleading rather than clever. Report SE for the reliability of a mean, s for the variability of the data.
- Keeping too many decimals. If the burette reads to 0.05 mL, an uncertainty quoted as 0.070711 mL claims a precision the instrument never had. One or two significant figures in the uncertainty is enough.
- Forgetting the square root. The variance is the mean square deviation; the standard deviation is its square root and carries the same unit as the data.
Where this appears in exams and lab work
| Setting | Which one to report |
|---|---|
| Class 11–12 practical file (repeat titrations) | Mean with standard deviation, and comment on the scatter |
| BSc / MSc analytical chemistry | Mean ± standard error, plus %RSD for method precision |
| Research papers and project reports | Mean ± SE or a 95% confidence interval, with n always stated |
| Physics and chemistry error-analysis questions | Both, plus the explicit distinction between random and systematic error |
Check the whole calculation in one step. The Statistics calculator takes a list of readings and returns the mean and standard deviation, so you can confirm the values worked out above before dividing by √n to get the standard error.
Open the Statistics Calculator →Working through Class 11–12 practicals and finding the error analysis harder than the experiment? ABC Chemistry runs Class 11–12 chemistry coaching at the Gurugram centre plus online classes across India — details at abcchemistry.in.