This standard error calculator computes the standard error of the mean, SE = s ÷ √n — how precisely a sample mean pins down the population mean — and the 95% confidence margin (±1.96 × SE) that turns it into an interval. Standard error is the most misreported statistic in science: endlessly confused with standard deviation, despite answering an entirely different question.

–

Two Questions Wearing Similar Names

Standard deviation asks: how far do INDIVIDUAL observations stray from the mean? Standard error asks: how far would my sample MEAN stray from the truth if I sampled again? The first is a property of the world (doesn’t shrink as you measure more of it); the second is a property of your study (shrinks like 1/√n forever). The defaults: s = 15, n = 36 → SE = 15/6 = 2.50, and the 95% margin ±4.90 turns a sample mean of 72 into the interval [67.1, 76.9]. Confusing the two in print is a documented epidemic — papers report “mean ± SE” with unlabeled error bars that look impressively tight because SE is smaller than SD by the factor √n; honest reporting labels the bar, because a reader deserves to know whether the spread shown is the data’s or the estimate’s.

The √n tyranny deserves its own sentence, because it budgets every empirical science: to halve your uncertainty you quadruple the sample; to gain one more decimal of precision you pay a hundredfold. It’s why election polls cluster at n ≈ 1,000 (SE of a proportion ≈ 1.6 points, margin ±3.1 — the familiar ±3 in every news graphic comes straight off this page’s formula, via the proportion version SE = √(p(1−p)/n)) and why physics experiments run for years chasing the next √n.

Worked Examples (Verified)

  • Defaults: SE = 15 ÷ √36 = 2.50; margin ±1.96 × 2.5 = ±4.90.
  • The poll: p = 0.5, n = 1,000 → SE = √(0.25/1000) = 1.58% → the ±3.1-point margin every survey graphic quotes — same machinery, proportion flavor.
  • The diminishing return: from n = 36 to n = 3,600 (100× the fieldwork) SE improves 2.50 → 0.25 (10×) — √n collects its tax without negotiation.

Standard Error Discipline

  • LABEL every ±: SD, SE or CI half-width — the three differ by √n and by 1.96, and unlabeled bars are how small effects dress up as precise ones.
  • Small samples use t, not 1.96: for n under ~30 the 95% multiplier comes from the t-distribution (df = n − 1) and is larger — the interval here uses the large-sample z as labeled.
  • SE of a proportion uses √(p(1−p)/n) — worst case at p = 0.5, which is why poll margins quote the max.
  • Clustered/weighted designs inflate true SE beyond the simple formula (design effects) — survey statisticians adjust; naive √n understates uncertainty in clustered samples.
  • Finite population correction: sampling a large FRACTION of a small population shrinks SE by √((N−n)/(N−1)) — relevant in audits and quality sampling, invisible in infinite-population formulas.

Related Tools

Frequently Asked Questions

How do you calculate standard error?

SE = s ÷ √n — sample standard deviation over the square root of sample size. Defaults: 15 ÷ 6 = 2.50. Multiply by 1.96 for the 95% confidence margin (±4.90 here). This calculator returns SE, the margin and the interval around your sample mean in one pass.

Standard error vs standard deviation — what’s the difference?

SD describes the spread of individual data points around their mean — a fact about the population that does NOT shrink with more sampling. SE describes the spread of the sample MEAN around the true mean across hypothetical repetitions — a fact about your estimate’s precision that shrinks like 1/√n. Reporting one while meaning the other (especially SE, which is always smaller) makes results look deceptively precise; labels are ethics here.

Why does sample size enter as a square root?

Because averaging independent errors partially cancels them: variances of sums add (variance of the mean = σ²/n for independent draws), and standard deviations — hence SE — go as the square root. The consequence is the diminishing-returns curve all research budgets live on: precision gains cost quadratically more data. Dependence between observations (clusters, time series) breaks the independence the derivation needs, which is why design effects exist.

How does SE build a confidence interval?

Mean ± (multiplier × SE): 1.96 for 95% in large samples (from the normal curve’s central 95%), ~2.58 for 99%, 1.645 for 90%. Small samples swap the normal multiplier for Student’s t with n − 1 degrees of freedom — always wider, honestly reflecting that s itself is estimated. The calculator uses the labeled large-sample 1.96; our confidence-interval tooling in the Statistics hub handles the t variants.

What is the standard error of a proportion?

SE = √(p(1−p)/n) — the binomial flavor. It’s largest at p = 0.5 (maximum uncertainty), giving the canonical poll arithmetic: n = 1,000 → SE 1.58 points → ±3.1 at 95%. Proportions near 0 or 1 carry smaller SEs at the same n, which is why landslide races poll tighter than coin-flips — the formula, not the pollster, decides.

Does a bigger sample make SE zero?

Only in the limit of sampling the entire population (or n → ∞): SE ∝ 1/√n approaches zero asymptotically but never arrives for finite samples — and measurement error, bias and non-response, which √n cannot fix, dominate long before. Precision (small SE) is not accuracy (small bias): a huge biased sample estimates the wrong number very precisely, the most dangerous configuration in statistics.

What does “mean ± 1 SE” in a paper mean?

That the interval shown spans roughly a 68% confidence range for the mean (if normal) — narrower than the conventional 95% CI, and a common source of over-optimistic visual impressions. When comparing two groups by eye, non-overlapping ±1 SE bars suggest (roughly) p < 0.05-ish only in specific cases — proper practice reads the stated CI or test, not the bar geometry.

Is this standard error calculator free?

Yes — free, instant, no sign-up. Enter SD and n for SE and the 95% margin; add your sample mean for the interval itself.