Standard Error Calculator
Compute the standard error of the mean (SEM) from the sample standard deviation and sample size, SEM = s/√n. Quantifies how precisely a sample mean estimates the underlying population mean — the foundation of confidence intervals, t-tests, and any inferential statistic.
Last updated: May 2026
Compare with similar
About this calculator
The standard error of the mean (SEM) is the standard deviation of the sampling distribution of the sample mean: SEM = s/√n, where s is the sample standard deviation and n is the sample size. It tells you how much the sample mean would vary if you repeated the sampling procedure many times. Unlike the standard deviation, which describes spread of individual observations, the SEM describes spread of sample means around the true population mean. The √n in the denominator means that doubling sample size reduces the SEM by a factor of √2 ≈ 1.41; quadrupling sample size halves it. This is the diminishing-returns law of statistics: to halve your estimation uncertainty you need to quadruple your sample size. SEM is the standard scale for inferential statistics: confidence intervals are constructed as x̄ ± z·SEM (or t·SEM); test statistics are (x̄ − μ₀)/SEM; meta-analyses weight studies by 1/SEM². Variables: s is the sample standard deviation (use Bessel’s-corrected n − 1), n is the sample size. Edge cases: n = 1 makes SEM undefined; n = 0 makes everything undefined. SEM assumes simple random sampling — for cluster, stratified, or complex sample designs, the design-effect must be applied. SEM applies to the mean specifically; for medians, proportions, regression coefficients, and other estimators, separate standard error formulas apply. For very small samples the SEM itself has substantial uncertainty (the chi-square distribution governs the SD estimate), which is why t-distributions are used for inference at small n.
How to use
Example 1 — Survey of 100 employees. You measure job-satisfaction scores in n = 100 employees and find x̄ = 6.5 (on a 1–10 scale) with s = 1.8. SEM = 1.8 / √100 = 1.8/10 = 0.18. ✓ A 95% confidence interval is x̄ ± 1.96·SEM = 6.5 ± 0.353, giving [6.15, 6.85]. Interpret: if the survey were repeated many times under identical conditions, 95% of resulting confidence intervals would contain the true population mean. Example 2 — Comparing precision at two sample sizes. Same data as above (x̄ = 6.5, s = 1.8), but you had only n = 25 observations. SEM = 1.8 / √25 = 1.8/5 = 0.36. ✓ The SEM doubled compared to n = 100 — exactly the √n=2 versus √n=10 ratio (10/5 = 2). The 95% confidence interval widens accordingly to about 6.5 ± 0.706, or [5.79, 7.21]. To match the precision of the n = 100 survey, you would need to multiply this n by 4 (to 100) — a striking illustration of why sample size matters so much for inference.
Frequently asked questions
What is the difference between standard deviation and standard error?
Standard deviation (SD) measures the spread of individual observations in a single sample — how much typical data points deviate from the sample mean. Standard error of the mean (SEM) measures the spread of sample means across repeated samples — how much the estimate of the population mean would vary if you took different samples. SD describes the data; SEM describes the estimate. Numerically, SEM = SD/√n, so SEM is always smaller than SD (often much smaller for large samples). Confusing the two is one of the most common errors in scientific reporting: papers that report ‘mean ± SD’ versus ‘mean ± SEM’ tell very different stories. SD bars in a graph show data variability; SEM bars show estimate precision. For visualisation, SD is appropriate when you want readers to understand the spread of individual measurements; SEM (or confidence intervals) when you want them to see how precisely the mean is estimated.
How does sample size affect the standard error?
SEM scales as 1/√n: doubling sample size shrinks SEM by a factor of √2 ≈ 1.41, quadrupling sample size halves SEM, increasing by 100× shrinks SEM by 10×. The square-root law means that improving precision is increasingly expensive: going from n = 100 to n = 200 halves the SEM only by a factor of √2, not by half. For experimental design, this dictates the trade-off between precision and cost — to halve your error bars, you need 4× more data, which is often the binding constraint on what is achievable. Power analysis uses this relationship to compute the minimum sample size needed to detect an effect of given magnitude with given probability. Critical insight: SD does NOT depend on sample size in expectation (the underlying spread of data is a property of the population, not the sample), but SEM very much does.
What is the relationship between standard error and confidence intervals?
A confidence interval for the mean is constructed as x̄ ± (critical value) × SEM. For a 95% CI with a large sample (n ≥ 30), use 1.96 (the standard-normal quantile): x̄ ± 1.96·SEM. For smaller samples, use the t-distribution with df = n − 1: at n = 10 the critical value is t = 2.262, giving a wider interval than the normal. The interpretation: in repeated sampling from the same population, 95% of resulting intervals would contain the true population mean. The interval does NOT mean ‘there is a 95% probability that the true mean is in this interval’ — that is a Bayesian credible interval, which requires a prior distribution. The frequentist confidence interval is a property of the procedure, not a probability statement about any single interval. Wider intervals indicate more uncertainty; narrower intervals indicate more precise estimates — and the relationship to sample size flows through SEM.
What are the most common mistakes people make with standard error?
The first is confusing SD with SEM in plots and tables; this can make the same dataset look 5–20× more or less variable than it actually is. The second is using SEM-based error bars to assess overlap between groups; SEM error bars look smaller than CIs and can suggest false significance — for visual comparison use 95% CIs. The third is using SEM with non-random or clustered samples without applying a design effect; the SEM formula assumes simple random sampling. The fourth is applying SEM to proportions or other non-mean statistics; each estimator has its own SE formula. The fifth is treating SEM as an absolute precision measure across studies of different sizes — papers always need both n and SD/SEM together to be interpretable. The sixth is computing SEM with population SD (σ) instead of sample SD (s) when σ is unknown; in practice you almost always use s with Bessel’s correction. And the seventh is forgetting that SEM is for the mean specifically — for differences between means, the SE of the difference is √(SE₁² + SE₂²) under independence, not just any single SEM.
When should I not use this calculator?
Skip it for non-mean statistics — standard error of a proportion, median, regression coefficient, ratio, or odds ratio each have their own formulas distinct from s/√n. Avoid it for cluster, stratified, or complex survey designs without applying the appropriate design effect; the simple SEM formula assumes simple random sampling and understates uncertainty for clustered data. It is the wrong tool when sample size is extremely small (n < 10) and you need exact inference — use t-distribution methods with appropriate degrees of freedom rather than SEM-based normal approximations. Do not use it for time-series with autocorrelation, where consecutive observations are not independent and the effective sample size is smaller than n. Skip it for inferring uncertainty on transformed quantities (log of mean, ratio of means) where you need the delta method or bootstrap. And for very precise inferential needs (regulatory submissions, medical trials), use the appropriate confidence-interval and hypothesis-test framework rather than a single SEM number.