A sample mean is an estimate, and standard error measures how good the estimate is. Take many samples from the same population, compute each sample’s mean, and those means vary — their standard deviation is the standard error of the mean. It is computed from one sample alone as the sample standard deviation divided by the square root of the sample size, and it shrinks as samples grow, but only as the square root: quadrupling the sample halves the error. This guide explains the concept, the formula, the construction of confidence intervals, the difference from standard deviation that causes most of the confusion, and the mistakes that confidence intervals invite.

What standard error is

Imagine measuring the heights of 25 adults and finding a mean of 170 cm with a standard deviation of 10 cm. The standard deviation describes the people: individual heights scatter about 10 cm around the mean. The standard error describes the mean itself: if you repeated the whole exercise with different samples of 25, the sample means would scatter around the true population mean with a standard deviation of 10 ÷ √25 = 2 cm. The standard error is 2 cm. It answers a different question from the standard deviation — not “how much do people vary?” but “how far might my sample mean be from the truth?”

The sampling-distribution picture is the intuition to keep. Every sample gives a slightly different mean; the distribution of those means centres on the population mean (for unbiased sampling) and narrows as samples grow. The standard error is that distribution’s standard deviation, estimated from the one sample you actually took. It is an estimate of an estimate’s precision, and like all estimates it carries uncertainty of its own — larger when the sample is small, which is why small-sample intervals use the t-distribution’s wider multipliers rather than the normal 1.96.

The formula, and why the square root is there

SE = s ÷ √n, where s is the sample standard deviation and n is the sample size. The square root comes from the variance of a mean: the variance of the average of n independent values is the population variance divided by n, and the standard deviation — the square root of variance — is therefore the population standard deviation divided by √n. Averaging n independent measurements does not divide the error by n, because the errors partly cancel rather than adding: the square root is the mathematics of partial cancellation.

Worked examples: SD 15, n 25 → SE = 15 ÷ 5 = 3. SD 15, n 100 → SE = 15 ÷ 10 = 1.5. SD 20, n 16 → SE = 20 ÷ 4 = 5. SD 20, n 400 → SE = 20 ÷ 20 = 1. The pattern to internalise is the cost of precision: each halving of the standard error requires four times the data. A study designed for SE 3 that is redesigned for SE 1.5 needs four times the participants, with four times the cost — a decision that should be made deliberately, with the budget, not discovered afterwards.

Confidence intervals: building them, and reading them correctly

A confidence interval is the sample mean plus and minus a multiple of the standard error: mean ± multiplier × SE. For large samples the multiplier for a 95% interval is 1.96 (from the normal distribution); for small samples the t-distribution gives larger multipliers that depend on degrees of freedom (n − 1): at n = 10 the 95% multiplier is 2.262; at n = 25 it is 2.064; at n = 100 it is 1.984. Example: mean 100, SD 15, n 25 → SE 3, interval 100 ± 1.96 × 3 = 94.12 to 105.88 (normal approximation) or 100 ± 2.064 × 3 = 93.81 to 106.19 (t, more correct at this size). The interval’s width is 2 × multiplier × SE; halving the width needs four times the sample.

The reading that most often goes wrong is the probability statement. A 95% confidence interval does not mean there is a 95% probability that the true mean lies inside this interval. The true mean is a fixed number; it is either inside or outside. The 95% belongs to the procedure: if you repeated the sampling and interval construction many times, about 95% of the intervals constructed would contain the true mean. Saying “we are 95% confident” is the conventional shorthand for that long-run property. It is not a licence to treat any single interval as 95% likely correct — and it says nothing at all about the probability of any particular value within the interval, which the procedure treats as equally plausible.

Standard error is not standard deviation

The confusion between the two is the most consequential error in elementary statistics, because the numbers can be identical in form and opposite in meaning. Standard deviation describes variability in the data: how far individual observations sit from their mean. Standard error describes uncertainty in an estimate: how far a sample mean might sit from the population mean. Reporting a standard error where a standard deviation belongs understates the variability of the data by a factor of √n — for n = 100, tenfold — and makes a scattered dataset look tight. Reporting a standard deviation where a standard error belongs overstates the uncertainty of the mean by the same factor.

Error bars in charts are where the confusion becomes visible. Bars showing ±1 SD describe the data’s spread; bars showing ±1 SE describe the mean’s precision; bars showing a 95% confidence interval are about twice the SE bars. A chart that does not label its bars is uninterpretable, and a reader who assumes the wrong one will misjudge both the data and the conclusion. The companion article on standard deviation, linked at the end, covers the data side; this guide covers the estimate side. Keep the two questions separate — “how much do values vary?” and “how precisely do I know the mean?” — and the right figure follows.

Five mistakes with standard error

First, using the standard error to describe the data — see above; use the standard deviation for that. Second, building intervals with 1.96 for tiny samples: below about n = 30, use the t multiplier, or the interval will be too narrow. Third, treating the standard error as if it described individual observations: a 95% interval for the mean is far narrower than the range containing 95% of individuals, which is roughly mean ± 1.96 × SD — the empirical rule applied to the data, not to the estimate. Fourth, comparing two groups by whether their intervals overlap: overlapping intervals do not prove the difference is non-significant, and non-overlapping intervals are a stricter test than the proper two-sample test. Compare the difference directly, with its own standard error. Fifth, forgetting the assumptions: the formula assumes a random sample and roughly independent observations. Clustered, stratified, or time-series data need adjusted standard errors, and no formula rescues a biased sample.

Frequently asked questions

What is standard error in simple words?

How far a sample mean is likely to be from the true population mean. It is the standard deviation divided by the square root of the sample size: SE = SD ÷ √n. A smaller standard error means a more precise estimate of the mean.

How do you calculate standard error?

Divide the sample standard deviation by the square root of the sample size. With SD 15 and n = 25: SE = 15 ÷ 5 = 3. With n = 100: SE = 15 ÷ 10 = 1.5.

What is the difference between standard deviation and standard error?

Standard deviation describes how spread out the individual data values are. Standard error describes how precisely the sample mean estimates the population mean. Bigger samples shrink the standard error; they do not shrink the standard deviation.

How do you turn a standard error into a confidence interval?

Mean ± multiplier × SE. For a 95% interval with a large sample, the multiplier is 1.96: mean 100, SE 3 gives 94.1 to 105.9. For small samples use the larger t-distribution multiplier for the sample size.

Is a 95% confidence interval a 95% chance the true mean is inside it?

Not exactly. The 95% is a property of the procedure: about 95% of intervals built this way, over many repeated samples, would contain the true mean. The true mean is fixed; this particular interval either contains it or does not. ‘95% confident’ is shorthand for that long-run property.

Does a bigger sample always help?

For precision, yes — the standard error falls with √n — but with diminishing returns: quadrupling the sample only halves the error. And a bigger sample does not fix bias: a large biased sample estimates the wrong value precisely. Fix the sampling first, then enlarge it.