Standard Deviation Calculator

Enter numbers separated by commas, spaces, or line breaks to calculate standard deviation, variance, mean, confidence intervals, frequency distribution, and detailed statistical metrics.

Enter Your Dataset

DatasetComma, space, or newline separated

Parsed 8 numeric values

It is a

What Is Standard Deviation?

Standard deviation measures how spread out a dataset is around its mean. A small standard deviation means values cluster tightly around the average; a large one means they scatter widely. Mathematically, it is the square root of the variance — the average squared distance of each value from the mean. Because it lives in the same units as the data itself, standard deviation is the most readable single number for describing variability.

This calculator returns both the population (σ) and sample (s) standard deviation, the matching variances, the mean, sum, count, minimum, maximum, range, median, mode, quartiles, percentiles, coefficient of variation, skewness, excess kurtosis, the standard error of the mean, and a full confidence-interval table from 68.3% to 99.9999%. It pairs naturally with the mean median mode range calculator, the z-score calculator, and the probability calculator.

How Standard Deviation Is Calculated

Compute the mean

Add every value, divide by the count. The mean is the balance point of the data — the value around which deviations are measured.

Subtract the mean, square

For each observation x, compute (x − mean)². Squaring removes the sign and weights large deviations much more than small ones, so a single far-away value dominates the sum.

Sum the squared deviations

Add every squared deviation. The result, Σ(x − mean)², captures the total squared spread of the dataset around its centre.

Divide and take the root

Divide by n for a population, or by n − 1 for a sample (Bessel's correction), to get variance. Take the square root of that variance to return the answer to the original units — and you have the standard deviation.

6 Ways to Use This Standard Deviation Calculator

Describe variability

Quote the standard deviation alongside the mean whenever you report an average — together they describe both location and spread, the bare minimum for any honest summary.

Build confidence intervals

Use the standard error of the mean (σ ÷ √n) with the 68.3%, 95%, or 99% critical Z to bracket the true population mean with a known margin of error.

Detect outliers

Flag any value more than 1.5×IQR outside Q1–Q3, or more than three standard deviations from the mean, as a candidate outlier worth investigating before averaging.

Apply the empirical rule

For roughly normal data, 68% of values fall within ±1σ of the mean, 95% within ±2σ, and 99.7% within ±3σ — a fast sanity check on any sample summary.

Standardise features

Subtract the mean and divide by σ to put any column on the standard scale (mean 0, σ 1). Most ML algorithms — linear models, k-NN, neural nets — train faster and generalise better on standardised features.

Compare two datasets

The coefficient of variation (σ ÷ mean) is unit-less and lets you compare variability between datasets with completely different scales — say a temperature time-series with a financial return series.

Best Practices for Working With Standard Deviation

Decide whether your data is the full population or only a sample of one before reporting σ vs. s. Survey responses, lab measurements, and historical returns are almost always a sample, in which case dividing by n − 1 (Bessel's correction) gives an unbiased estimate of the population variance. Census data and the entire historical record of an account are populations, in which case dividing by n is correct.

Always report standard deviation alongside the mean, and never pre-round either one. A mean of 100 with σ = 10 tells a very different story than the same mean with σ = 50. The two together give the reader a sense of both centre and spread; reporting only the mean hides the difference.

Inspect a histogram before treating σ as a percentile gate. Standard deviation is most informative for symmetric, single-peaked distributions; for skewed or bimodal data, the median and IQR (or per-cluster summaries) describe the shape more faithfully. The Charts tab here gives you a histogram, a bell-curve overlay, and a box plot in one click — use them.

Why Standard Deviation Matters

Finance & investing

Volatility, value-at-risk, Sharpe ratio, beta, and option pricing all depend on the standard deviation of returns. A larger σ means a wider distribution of possible outcomes — exactly what risk means in modern portfolio theory.

Science & research

Every lab measurement carries an uncertainty band reported as ±σ or ±2σ. Quality-control charts flag samples beyond ±3σ. Confidence intervals on experimental effects are built directly from the sample standard deviation and the standard error.

Machine learning

Feature scaling, z-score normalisation, BatchNorm, and dropout all reference σ. The variance of weights drives initialisation schemes (He, Xavier). Without σ, none of the standard preprocessing pipelines would work the same way.

Education & testing

Standardised tests (SAT, GRE, IQ) report scores as z-scores against a population σ — that is what a percentile rank actually means. Curving exam grades around the mean and standard deviation lets instructors compare classes that took different forms.

Where Standard Deviation Gets Tricky

Skewed and heavy-tailed data

For income, response times, or insurance claims, the long right tail pulls the mean above the median and inflates σ. The 68-95-99.7 rule no longer applies; report median and IQR alongside (or instead of) mean and σ.

Small samples

With n < 30, the sample σ is itself a noisy estimate. For confidence intervals you should usually switch to Student's t-distribution and a t-table instead of using the standard normal critical values.

Different units

You cannot compare two σ values measured in different units — a σ of 5 inches and a σ of 5 grams are unrelated. Use the coefficient of variation (σ ÷ |mean| × 100%) to compare relative variability across units.

Outliers dominate

Because deviations are squared, a single extreme value can dominate the sum. If you suspect a typo or contamination, identify it on the Charts tab before computing further statistics.

The Core Standard Deviation Formulas

Population mean

μ = Σx ÷ N

Centre of the full population.

Sample mean

x̄ = Σx ÷ n

Centre of a sample drawn from a population.

Population variance

σ² = Σ(x − μ)² ÷ N

Mean squared deviation, divides by N.

Sample variance

s² = Σ(x − x̄)² ÷ (n − 1)

Bessel's correction — unbiased estimator of σ².

Population StdDev

σ = √[Σ(x − μ)² ÷ N]

Square root of population variance.

Sample StdDev

s = √[Σ(x − x̄)² ÷ (n − 1)]

Square root of sample variance.

Standard error of mean

SEM = σ ÷ √n or s ÷ √n

Standard deviation of the sampling distribution of the mean.

Margin of error

ME = z* · SEM

Half-width of a confidence interval.

Confidence interval

CI = x̄ ± z* · SEM

Range that contains the true mean with the chosen confidence.

Coefficient of variation

CV = (s ÷ |x̄|) × 100%

Unit-less relative spread — compares across units.

Z-score

z = (x − μ) ÷ σ

Standardise a value into σ-units around the mean.

Empirical rule

P(|Z| ≤ 1, 2, 3) ≈ 68.3, 95.5, 99.7%

Quick sanity check for normally distributed data.

Common Standard Deviation Mistakes

1
Mixing up σ and s
Dividing by n when your data is a sample under-estimates the true population variance — and the bias becomes large for small samples. If the data is a sample of a larger population, divide by n − 1.
2
Reporting variance instead of σ
Variance is in squared units (dollars-squared, kilograms-squared) and is rarely meaningful on its own. Take the square root and report σ — it speaks the same language as the data.
3
Treating σ like a percentile gate
The 68-95-99.7 rule only applies when the data is approximately normal. For skewed or bimodal data, the actual percentage within ±1σ can be quite different.
4
Quoting σ without n
A standard deviation computed from three observations is far less reliable than one from three hundred. Always state the sample size alongside σ, especially for confidence-interval claims.
5
Forgetting units
If x is in dollars, σ is also in dollars; variance is in dollars-squared. Carrying units through the math catches mistakes that the calculator alone cannot.
6
Comparing σ across different scales
A σ of 5 cm and a σ of 5 kg are not comparable. Use the coefficient of variation (σ ÷ |mean|) for any cross-unit comparison of variability.

Real-Life Uses of Standard Deviation

Risk and volatility

Investment volatility is reported as the annualised standard deviation of returns. A fund with a higher σ has a wider distribution of possible outcomes — the foundation of Sharpe ratio, beta, and modern portfolio theory.

Process control

Six Sigma processes target ±6σ around the specification mean — equivalent to roughly 3.4 defects per million opportunities. Statistical process control charts flag any sample beyond ±3σ as out-of-control.

Curved grading

Many instructors curve grades by computing each student's z-score against the class mean and σ, then mapping ranges of z to letter grades. The procedure standardises across exams of different difficulty.

Feature scaling in ML

StandardScaler subtracts the mean and divides by σ of each column so the training data has mean 0 and unit variance. Most gradient-based learners — including neural nets — converge faster on standardised inputs.

Methodology you can verify

Every statistic is computed live from your dataset using the canonical NIST formulas — Σ(x − mean)² for the deviation sum, divided by n (population) or n − 1 (sample) for the variance, with linear-interpolated quartiles matching Excel's PERCENTILE.INC and NumPy's numpy.percentile. Confidence-interval critical values come from the standard normal table. Read more on the methodology and editorial policy pages.

Frequently Asked Questions

Standard deviation is a measure of how spread out the values in a dataset are around the mean. It is the square root of the variance — the average squared distance of each value from the mean. A small standard deviation means most observations cluster tightly around the average, while a large one means they are widely scattered. Because standard deviation is in the same units as the original data (dollars, kilograms, percentage points), it is the most readable single summary of variability. The calculator reports both the population (σ) and the sample (s) versions, along with the matching variances, on every result.

Population standard deviation (σ) is used when your data represents the entire group you care about — for example every product manufactured in a single batch or every score on a finished census. It divides the sum of squared deviations by n. Sample standard deviation (s) is used when your data is a sample drawn from a larger population — for example survey responses, lab measurements, or historical returns. It divides by n − 1 instead of n (Bessel's correction) to produce an unbiased estimate of the population variance. The smaller the sample, the bigger the practical difference between the two; for n = 5 the two differ by about 12%, but for n = 100 the gap is below 1%.

Five steps: (1) compute the mean by summing every value and dividing by the count; (2) subtract the mean from each value to get the deviations; (3) square each deviation; (4) sum the squared deviations and divide by n (population) or n − 1 (sample) — this gives the variance; (5) take the square root of the variance — that is the standard deviation. For the example dataset {10, 12, 23, 23, 16, 23, 21, 16} the mean is 18, the sum of squared deviations is 192, the population variance is 192 ÷ 8 = 24, and σ = √24 ≈ 4.8989794855664. The Steps tab on this page works the entire calculation out with your actual numbers.

Variance is the average squared distance of each value from the mean: σ² = Σ(x − μ)² ÷ N for a population, or s² = Σ(x − x̄)² ÷ (n − 1) for a sample. Standard deviation is the square root of variance. The two carry exactly the same information about spread, but variance lives in squared units (dollars-squared, kilograms-squared), which makes it awkward to interpret. Most analyses therefore quote standard deviation. Variance still appears in formulas — analysis of variance (ANOVA), covariance, and propagation-of-uncertainty equations — because additive properties work cleanly on squared quantities.

There is no universal good value — it depends entirely on the units, the mean, and the field of study. A standard deviation of 5 mm is huge for the diameter of a precision-machined bearing and trivial for the height of an adult. The coefficient of variation (CV = σ ÷ |mean| × 100%) gives a unit-less alternative: a CV below 15% is usually considered low variability, 15–35% moderate, and above 35% high. The calculator reports CV automatically on the Summary tab whenever the mean is non-zero.

A high standard deviation means the values in your dataset are far apart from one another — the data is widely scattered around the mean. For financial returns, a high σ means high volatility. For exam scores, it means a wide range of student performance. For a manufacturing process, it means the output is inconsistent. In every case, the empirical rule still applies for roughly normal data: about 68% of values fall within ±1σ of the mean, 95% within ±2σ, and 99.7% within ±3σ — a wider σ simply widens those bands.

A low standard deviation means the values cluster tightly around the mean, so the average is a good summary of any individual observation. For investments this implies low volatility; for process control it implies high precision; for survey results it implies broad agreement. A standard deviation of exactly 0 means every value in the dataset is identical to the mean — there is no spread at all.

First compute the standard error of the mean: SEM = σ ÷ √n (or s ÷ √n for a sample). Then multiply by the critical Z value for your chosen confidence level (1.000 for 68.3%, 1.645 for 90%, 1.960 for 95%, 2.576 for 99%, 3.291 for 99.9%, and so on) to get the margin of error. The confidence interval is mean ± margin. For example, with mean = 18, σ = 4.8989794855664, and n = 8, SEM = 1.732, and the 95% margin of error is 1.960 × 1.732 ≈ 3.395 — giving the 95% CI 18 ± 3.395, or [14.605, 21.395]. The Confidence tab on this page returns the full table from 68.3% to 99.9999%.

SEM is the standard deviation of the sampling distribution of the mean — how much the sample mean would vary if you took many different samples of the same size from the same population. It is computed as SEM = σ ÷ √n. SEM shrinks as the sample size grows (by a factor of √n), which is why larger samples produce tighter confidence intervals. SEM is the building block of every interval estimate and every Z- or t-based hypothesis test on the mean.

No. Standard deviation is the square root of variance, and variance is itself a sum of squared deviations divided by a positive count — so both are always greater than or equal to zero. A reported negative σ is always a calculation or sign-flip bug. The smallest possible σ is exactly 0, which occurs only when every value in the dataset is identical to the mean (and therefore identical to one another).

Squaring achieves three things at once: it makes every deviation positive so they do not cancel out when summed, it weights large deviations much more than small ones (a deviation of 4 contributes 16 to the sum, while a deviation of 2 contributes only 4), and it produces clean mathematical properties — the variance is the unique quadratic measure of spread that is additive for independent variables, which is why variance underpins ANOVA, covariance, and uncertainty propagation. The square root at the very end returns the answer to the original units.

For data that is approximately normally distributed, about 68% of values fall within one standard deviation of the mean, about 95% within two standard deviations, and about 99.7% within three. The exact percentages are 68.27%, 95.45%, and 99.73%. The rule lets you sanity-check any normal distribution at a glance: with mean = 100 and σ = 10, you expect roughly two-thirds of observations between 90 and 110, almost all between 80 and 120, and essentially everything between 70 and 130. For skewed or heavy-tailed data the rule overstates how concentrated the data really is — check a histogram first.

Related Calculators

More statistics, probability, and math tools to pair with your dataset.