What is the standard error of the mean?
The standard error of the mean is how far your sample's average is likely to sit from the true average of the population you took the sample from. It describes the precision of one number — the mean — rather than the spread of the data points behind it. Collect a second sample the same size and you would get a slightly different average. The standard error puts a size on "slightly."
Here it is on real numbers. Six measurements: 4, 8, 6, 5, 3, 7. Their mean is 5.5, their sample standard deviation is 1.871, and the standard error is 0.7638. A results section would write that as M = 5.50, SE = 0.76. That last figure is what goes after the ± in a table, and it's what sets the height of an error bar on a chart.
It turns up anywhere somebody reports an average taken from a sample: an assay run ten times, a survey's mean satisfaction score, a week of page-load timings, a class of exam marks. The notation moves around by field — SE, SEM, s.e.m., or nothing at all in the axis label — but the quantity is identical every time, and so is the arithmetic that produces it.
How to use the standard error calculator
There are two ways the question arrives — with the raw numbers, or with somebody else's summary of them — so the standard error calculator takes both. Pick the mode that matches what you actually have in front of you.
- Choose From data if you have the raw numbers, or From SD + n if you're reading someone else's results table. Switching between the two keeps whatever you typed in the other one.
- In From data mode, paste or type your values. Commas, spaces, tabs, semicolons and line breaks all work as separators, so a column copied straight out of a spreadsheet goes in unchanged. Decimals and minus signs are fine.
- In From SD + n mode, type the standard deviation and the sample size. The third box — the mean — is optional; fill it in if you want the mean ± SE range too.
- Read the headline number. That's the standard error, shown to four significant figures, and it appears as soon as there's enough to compute it.
- Check the supporting numbers underneath: the sample standard deviation that was divided by √n, the mean, the count, and the mean ± SE interval. The formula line shows the same division with your real numbers substituted in —
SE = 1.871 / √6 = 0.7638— which is usually the part an assignment wants to see. - Copy the bare number for a spreadsheet, or use Copy report for the APA-style line at the two decimal places journals expect.
There's no Calculate button, no account, and no point where the page asks who you are. That's worth saying out loud, because this whole category usually runs the other way: a licensed statistics package billed per seat per year, a spreadsheet add-on sitting behind an annual contract, a fourteen-day trial that wants a card before it will divide one number by another. A single division does not need a subscription attached to it. This is just the calculator, and it's the whole calculator.
The standard error formula
SE = s / √n
s is the sample standard deviation — the one computed with an n − 1 divisor. n is how many observations you have. That's the entire standard error formula. The square root is the reason bigger samples pay off more slowly than people expect: precision improves with the square root of your effort, not with your effort.
How to calculate standard error by hand
Take 4, 8, 6, 5, 3, 7 again and do it by hand.
- Mean first: (4 + 8 + 6 + 5 + 3 + 7) ÷ 6 = 5.5.
- Deviations from the mean: −1.5, 2.5, 0.5, −0.5, −2.5, 1.5. Square them and add: 2.25 + 6.25 + 0.25 + 0.25 + 6.25 + 2.25 = 17.5.
- Divide by n − 1, so 17.5 ÷ 5 = 3.5. Square root: s = 1.871.
- Divide by √6 = 2.449. Standard error = 0.7638.
The mean ± SE range is 4.736 to 6.264 — the error bar you'd draw at ±1 SE.
When you only have the SD and n
Most of the time you're not holding the raw data. A paper reports s = 12.5 from n = 30 and leaves the rest to you. One division: 12.5 ÷ √30 = 2.282. Step 1 through 3 above were already done by someone else, and the page skips straight to the last one. The only thing worth checking is which standard deviation you were handed — the sample one, with the n − 1 divisor, is the correct input. Nearly everything published uses it. If a source explicitly reports a population SD, the answer comes out slightly too small: on this six-value example, the population divisor understates the standard error by about 8.7%.
How the standard error shrinks as your sample grows
Hold the standard deviation fixed at 10 and watch what sample size alone does to the precision of a mean of 50.
| Sample size (n) | √n | SE = 10 / √n | Mean ± SE (mean = 50) |
|---|---|---|---|
| 2 | 1.414 | 7.071 | 42.93 to 57.07 |
| 5 | 2.236 | 4.472 | 45.53 to 54.47 |
| 10 | 3.162 | 3.162 | 46.84 to 53.16 |
| 25 | 5 | 2 | 48 to 52 |
| 100 | 10 | 1 | 49 to 51 |
| 400 | 20 | 0.5 | 49.5 to 50.5 |
| 1,000 | 31.62 | 0.3162 | 49.68 to 50.32 |
The spread of the data never changed. Every row has the same standard deviation of 10, which is the point: more data doesn't make people less different from each other, it makes you more certain about their average. Going from 2 observations to 25 cuts the standard error from 7.071 to 2. Going from 400 to 1,000 — more than doubling the work — buys you a drop from 0.5 to 0.3162.
The practical rule falls straight out of the square root: to halve the standard error, you need four times as many observations. Take the published example of s = 12.5 at n = 30, which gives 2.282. To get that down to 1.141 you need n = 120. That's the calculation worth running in the standard error calculator before you commit to collecting anything, because it answers "is another month of data collection worth it?" in a number rather than a feeling.
Where the standard error gets misread
The single most common error isn't arithmetic — it's a label. Error bars drawn at ±1 SD and error bars drawn at ±1 SE look identical on a chart and mean completely different things. For ten height measurements averaging 67, the sample standard deviation is 2.944 and the standard error is 0.9309. The SD bars are more than three times taller. A figure that doesn't say which one it used isn't readable, and plenty of published figures don't say.
Three more things the standard error can't do for you:
- It doesn't detect bias. The standard error measures sampling noise and nothing else. If your sample was collected badly — self-selected respondents, a broken instrument, a convenience group — a tiny standard error just means you've measured the wrong thing very precisely.
- It isn't a confidence interval. Mean ± 1 SE covers roughly 68% of the sampling distribution, not 95%. For a real interval you need a critical value from the t distribution, and at small n that multiplier is well above 2.
- It inherits every outlier. Add a single stray 100 to those six measurements and the standard error jumps from 0.7638 to 13.52. Check the data before trusting the precision.
At the edges, the standard error calculator declines rather than guesses. One data point gives no spread to estimate, so the standard error comes back as an em-dash and an explanation instead of a zero. A negative standard deviation is treated as a typo, not a number. A non-integer sample size is floored with a note that says so — type 30.7 and you'll see "Using n = 30."
A quiet accuracy note. Many implementations compute the variance with the algebraic shortcut — the mean of the squares minus the square of the mean. It's correct on paper and one pass faster, but on a computer it subtracts two nearly identical large numbers and the meaningful digits cancel. On the data set 1000000.1, 1000000.2, 1000000.3, the shortcut returns 0.05846 where the true answer is 0.05774, about 1% wrong from arithmetic that should be exact. This page does the two passes.
Related calculations
The standard error is one step in a chain, and the neighbouring steps each have their own page. The standard deviation calculator gives you the spread itself — useful when you want to describe your data rather than the precision of its average. The variance calculator gives the same quantity before the square root, which is what most statistical tests actually work with.
Going the other direction, the confidence interval calculator turns a standard error into a real 95% interval with the right multiplier for your sample size, instead of the rough ±2 approximation. That multiplier comes from the t-value calculator, which gives the critical t for your degrees of freedom. And once you have a mean and an SD, the z-score calculator tells you where any single observation sits relative to them. If you just need the average of a list first, the average calculator handles that on its own.
All of them run in your browser, none of them ask for an account, and 10% of every dollar Microapp earns goes to charity, off the top, audited quarterly.
Frequently asked questions
Is the standard error the same as the margin of error?
No — the margin of error is the standard error multiplied by a critical value. For a 95% interval you take the standard error and multiply by the critical t for your degrees of freedom. With the six-value example, SE = 0.7638 and the critical t at 5 degrees of freedom is 2.571, so the margin of error is 1.964 and the interval runs from roughly 3.54 to 7.46. People often use 1.96 as the multiplier, which is only right when the sample is large; at n = 6 it would understate the interval by about a quarter.
How do I calculate standard error in a spreadsheet?
No spreadsheet has a built-in SE function, so you compose one: =STDEV.S(A2:A7)/SQRT(COUNT(A2:A7)). On the values 4, 8, 6, 5, 3, 7 that returns 0.7638, the same figure this page shows. The trap is STDEV.P, which uses the population divisor and hands you a number that's too small — always the .S variant for a standard error. Pasting the column into the standard error calculator is faster than debugging a nested formula, and it shows the intermediate standard deviation too, so you can see exactly where a wrong answer came from.
If two error bars overlap, does that mean the difference isn't significant?
No, and this is one of the most durable misreadings in published charts. Bars drawn at ±1 SE can overlap noticeably while the difference between the two means is still significant at p < 0.05. The reverse trap also exists: bars that just barely fail to touch are not a test result either. Error bars are a description of precision, not a significance test. If you need to know whether two means differ, run the test.
How many more observations do I need to halve my standard error?
Four times as many. The √n in the denominator means precision scales with the square root of sample size, so quadrupling n halves the standard error. Concretely: s = 12.5 at n = 30 gives 2.282; the same standard deviation at n = 120 gives 1.141. This is the single most useful thing the formula tells you before you collect data, and it's usually sobering — the last increment of precision costs far more than the first.
Can the standard error be larger than the mean?
Yes, and it's not a bug. It happens when the data straddles zero or clusters near it, and it happens easily at tiny sample sizes. Two values of −4 and 6 have a mean of 1 and a standard error of 5 — the mean ± SE range runs from −4 to 6, which is exactly as informative as it sounds. A standard error that's large relative to the mean is a real signal: the average you computed isn't pinned down well enough to build anything on.
Does this work for a percentage or a proportion?
Not with this formula. The standard error of a proportion is √(p(1−p)/n), which is a different calculation with different inputs — it needs the proportion itself rather than a standard deviation. This page computes the standard error of the mean, and deliberately doesn't offer the proportion version, because a page that quietly computes the wrong one for you is worse than a page that says which one it does. If your data is a set of measured values, you're in the right place.