Standard Error & Confidence Interval Calculator

Standard Error Guide

What's the difference between Standard Error and Standard Deviation?

Standard deviation (SD or s): measures spread of individual data points around the mean. How variable are the measurements? Standard error (SE or SEM): measures precision of the sample mean as an estimate of the population mean. How reliable is our estimate? SE = s / √n. Key insight: as sample size increases, the standard error decreases (we get a better estimate of the true mean). For n=4: SE = s/2. For n=100: SE = s/10. For n=10,000: SE = s/100. Example: measuring plant heights. s = 5cm, the natural variation between individual plants, stays roughly constant regardless of sample size, while the standard error of the mean shrinks as more plants are measured.

What's the key thing to understand about Confidence Intervals?

A 95% confidence interval: if we repeated the experiment many times, 95% of the resulting confidence intervals would contain the true population mean. Formula: CI = x̄ ± z × SE. z values: 90% CI: z=1.645. 95% CI: z=1.960. 99% CI: z=2.576. Example: mean=14.5, SE=1.2, 95% CI. Lower = 14.5 - 1.960×1.2 = 12.15. Upper = 14.5 + 1.960×1.2 = 16.85. Interpretation: we are 95% confident the true population mean lies between 12.15 and 16.85. Common misconception: 95% CI does NOT mean 'there is a 95% probability the true mean falls in this specific interval' — the true mean either is or isn't in the interval; the 95% refers to how often this method captures the true mean across many repeated samples.

What's the key thing to understand about Sample Size and Precision?

To achieve a desired margin of error (E) at 95% confidence: n = (z × s / E)². Example: survey with known SD of 10. Want margin of error ±2 at 95% confidence. n = (1.96 × 10 / 2)² = 9.8² = 96.04 → round up to 97 participants. To halve the margin of error: must quadruple the sample size. This inverse square root relationship means large precision improvements require dramatically more data. Practical implication: doubling sample size from 100 to 200 reduces SE by 29% (from s/10 to s/14). Doubling sample size again, from 200 to 400, only reduces SE by a further 29% — diminishing returns that explain why very large sample sizes are needed for small additional gains in precision.

What do I need to know about t-Distribution for Small Samples?

For small samples (n < 30), the z-distribution is replaced by the t-distribution, which has heavier tails. t-distribution accounts for uncertainty about the population standard deviation. As n increases, the t-distribution approaches the z (normal) distribution. For n=10, 95% CI: use t=2.262 instead of z=1.960. For n=5, 95% CI: use t=2.776. The t-distribution is why small clinical trials have wider confidence intervals — more uncertainty from less data. In practice: always report both the confidence interval and the sample size together, since the interval alone doesn't convey how much uncertainty came from a small sample.

What's the difference between Standard Error of the Mean vs Standard Error of a Proportion?

The standard error of the mean (SEM) measures how much a sample mean is expected to vary from the true population mean, calculated as SEM = s/√n, where s is the sample standard deviation and n is the sample size. The standard error of a proportion (SEP) does the analogous job for percentages — like the share of voters favoring a candidate — calculated as SEP = √[p(1-p)/n], where p is the sample proportion. Both shrink as n grows, following the same 1/√n pattern, which is why quadrupling your sample size only halves your standard error, not quarters it. Pollsters typically report a "margin of error" built directly from the SEP (usually ±1.96×SEP for 95% confidence), while scientists reporting experimental measurements usually use the SEM to show how precisely they've estimated an average value.

What should I know about a Worked Example: Calculating Standard Error?

Suppose a sample of 50 students has a mean test score of 72 with a sample standard deviation of 14. The standard error is SEM = s/√n = 14/√50 = 14/7.07 ≈ 1.98. This means that if you repeated the sampling process many times, the sample means would typically vary by about ±1.98 points from the true population mean. To build an approximate 95% confidence interval from this, use mean ± 1.96 × SEM = 72 ± 1.96(1.98) = 72 ± 3.88, or roughly (68.1, 75.9). Doubling the sample size to 100 would reduce the standard error to 14/√100 = 1.4, narrowing the confidence interval — illustrating why larger samples produce more precise (not necessarily more accurate) estimates of a population mean.

What should I know about the Central Limit Theorem and Why Standard Error Works?

Standard error measures how much a sample statistic, usually the sample mean, is expected to vary from sample to sample — but the reason the formula SE = s/√n works, and the reason confidence intervals built from it are trustworthy, rests on the central limit theorem (CLT). The CLT states that if you repeatedly draw random samples of size n from any population with a finite variance, the distribution of the sample means across those repeated samples approaches a normal distribution as n grows large, regardless of the shape of the original population's distribution — even if that original distribution is strongly skewed, bimodal, or otherwise far from normal. This is remarkable because it means statisticians can apply normal-distribution-based methods (like the familiar 68-95-99.7 rule, or z-based confidence intervals) to the sample mean even when the underlying data clearly is not normally distributed, as long as the sample size is reasonably large (a common informal guideline is n ≥ 30). The CLT also explains why standard error shrinks as √n grows: the theorem shows the standard deviation of the sampling distribution of the mean equals the population standard deviation divided by √n, which is exactly the standard error formula. Without the central limit theorem, there would be no general justification for treating the sample mean as approximately normally distributed.

Table: Standard Error Formulas for Different Statistics

The term "standard error" isn't limited to sample means — nearly every statistic you compute from a sample has its own corresponding standard error that measures how much that statistic would bounce around if you repeated the sampling process. The table below collects the most commonly used formulas side by side so you can see how the same basic logic (dividing a measure of spread by a function of sample size) carries across different situations.

Worked Example: Comparing Two Sample Means with a Pooled Standard Error

Suppose a researcher measures reaction times for two groups: Group 1 has n₁ = 25 people with a sample standard deviation of s₁ = 8 milliseconds, and Group 2 has n₂ = 30 people with s₂ = 10 milliseconds. To test whether the group means genuinely differ, the first step is computing the standard error of the difference between the two means: SE = √(s₁²/n₁ + s₂²/n₂) = √(8²/25 + 10²/30) = √(64/25 + 100/30) = √(2.56 + 3.333) = √5.893 ≈ 2.43 milliseconds. If the observed difference between the two group means was, say, 6 milliseconds, you'd divide that difference by the standard error to get a test statistic: 6 / 2.43 ≈ 2.47. A value this large (well beyond ±2) suggests the difference is unlikely to be due to sampling noise alone, which is exactly why the standard error — not just the raw difference — is the number that determines statistical significance. A smaller SE (from larger samples or less variable data) would make the same 6-millisecond gap look even more convincing, while a larger SE could make it look like random noise.

Standard Error in Regression and Hypothesis Testing

Standard error shows up constantly in regression output and hypothesis tests because it's the denominator of nearly every test statistic. In linear regression, each estimated coefficient (like a slope) comes with its own standard error, and dividing the coefficient by its standard error produces the t-statistic used to judge whether that predictor is meaningfully different from zero. A coefficient of 3.2 with a standard error of 0.5 gives a t-statistic of 6.4 — strong evidence the relationship is real. The same coefficient with a standard error of 2.9 gives a t-statistic of about 1.1, which wouldn't clear the usual bar for significance. This is also why standard error is central to margin of error calculations in polling: a poll reporting 52% support with a standard error of 1.5% typically reports a margin of error around 3% (roughly two standard errors), meaning the true population value is plausibly anywhere in that range. Smaller standard errors — from larger, less variable samples — produce tighter, more trustworthy intervals and more decisive hypothesis tests.