What the Standard Deviation Actually Tells You
Standard deviation is the typical distance of your values from their mean, measured in the same units as the data. A class whose test marks have a mean of 72 and a standard deviation of 4 is bunched together. A class with the same mean and a standard deviation of 15 is scattered from the very weak to the very strong. The mean alone cannot show that difference.
It is the square root of the variance. Variance averages the squared gaps from the mean, which is mathematically convenient but awkward to read (squared marks?). Taking the root brings the number back to marks, centimetres or dollars. Related ideas: the range (max − min) uses only two values, the mean absolute deviation averages gaps without squaring, and the standard error (s ÷ √n) describes how precisely a sample mean is known, which is a different question.
Sample or Population: Which Standard Deviation Do You Need?
| Your data is… | Use | Divide by | Excel / Sheets | TI-84 |
|---|---|---|---|---|
| Every member of the group (all 30 students in this class, all 12 months this year) | Population σ | n | STDEV.P / STDEVP | σx |
| A subset used to learn about a bigger group (a survey, a trial, 30 of 5,000 parts) | Sample s | n − 1 | STDEV.S / STDEV | Sx |
| Unsure | Sample s | n − 1 | STDEV.S | Sx |
In practice almost every real dataset is a sample, because you rarely measure everything. Textbooks therefore default to s unless the question says “the entire population”. If your teacher or software gives a different answer than you got, this is the first thing to check. Many graphing tools report both and it is easy to copy the wrong one.
How to Calculate Standard Deviation by Hand
- Add the values and divide by n to get the mean x̄.
- Subtract x̄ from every value, then square each difference.
- Add the squares. This total is the sum of squares, SS.
- Divide SS by n − 1 (sample) or n (population). This is the variance.
- Take the square root of the variance.
Take 2, 4, 4, 4, 5, 5, 7, 9. The mean is 5 and the squared gaps are 9, 1, 1, 1, 0, 0, 4, 16, so SS = 32. As a population: variance 32 ÷ 8 = 4 and σ = 2. As a sample: 32 ÷ 7 = 4.571 and s = 2.138. Load the example above and the calculator reproduces every line.
Using the shortcut formula s² = (Σx² − n·x̄²) ÷ (n − 1) is faster on paper but loses digits on a calculator when the mean is large compared with the spread. The deviation method is safer, and it is the one shown here.
Why Divide by n − 1? Our Simulation
A sample’s values sit closer to their own mean than to the true mean, so dividing by n underestimates the spread. Dividing by n − 1 (Bessel’s correction) removes that bias for the variance. We checked it instead of just quoting it: we drew 200,000 samples of size 5 from a distribution whose true variance is exactly 1.
| Method | Average variance | Average standard deviation |
|---|---|---|
| Divide by n | 0.799 | 0.840 |
| Divide by n − 1 | 0.998 | 0.939 |
| True value | 1 | 1 |
Dividing by n lands near 0.8, which is (n−1)/n. Dividing by n − 1 lands on 1 for the variance, as promised. Notice the last column though: even the corrected s averages 0.939, not 1. The correction fixes variance, not the standard deviation itself, because a square root of an unbiased estimate is slightly low. For small samples that is a real, small, unavoidable bias. Few pages say so.
The 68–95–99.7 Rule, and When It Fails
| Range | Share of a normal distribution | Chebyshev minimum (any shape) |
|---|---|---|
| Mean ± 1 SD | 68.27% | — |
| Mean ± 2 SD | 95.45% | at least 75% |
| Mean ± 3 SD | 99.73% | at least 88.9% |
The middle column holds only for bell-shaped data. In a simulation of 100,000 normal values, 4,525 fell beyond two standard deviations (theory predicts 4,550) and 278 beyond three (theory predicts 270). Real data is rarely that tidy. Income, waiting times and response times are skewed, and for them Chebyshev’s inequality is the only guarantee you get. Check the chart under the calculator before quoting “95% fall within two SD”.
What Counts as a High or Low Standard Deviation?
There is no universal threshold, because the number carries the units of the data. Compare it with the mean using the coefficient of variation (SD ÷ mean), which the calculator reports. A CV near 5% means very consistent measurements. Near 50% or above means the spread is as large as half the typical value.
Our opinion: the most over-used habit is calling a standard deviation “high” without a comparison group. 12 minutes is huge for a bus timetable and trivial for a hospital wait. Always compare against a baseline, a previous period or a competitor.
One Outlier Changes Everything
Take the eight values 12, 13, 13, 14, 15, 15, 16 and 90. The mean is 23.5, the median is 14.5 and the sample standard deviation is 26.9. Remove the 90 and the mean drops to 14, with s = 1.41. One value multiplied the standard deviation by nineteen. Because deviations are squared, extreme points dominate. If your chart shows an isolated dot, find out why before reporting s. For skewed data also give the median and the IQR (see the mean, median and mode calculator).
Standard Deviation in Excel, Google Sheets, TI-84 and Python
| Tool | Sample | Population |
|---|---|---|
| Excel / Google Sheets | =STDEV.S(A1:A10) (Sheets also accepts STDEV) | =STDEV.P(A1:A10) (Sheets: STDEVP) |
| TI-84 | STAT → CALC → 1-Var Stats → Sx | same screen → σx |
| Python statistics | statistics.stdev(x) | statistics.pstdev(x) |
| NumPy | np.std(x, ddof=1) | np.std(x) (default ddof=0) |
NumPy’s np.std divides by n unless you pass ddof=1. Pandas Series.std() does the opposite and divides by n − 1. Mixing them up is a classic source of “my answers do not match”.
Common Mistakes
- Using σ (÷ n) for a sample. Most homework wants s (÷ n − 1).
- Squaring the mean instead of the deviations, or forgetting the square root at the end.
- Typing the data once as a list and again as value:count, which doubles the weights.
- Reading standard error (s ÷ √n) as if it were the spread of the data.
- Applying the 68–95–99.7 rule to clearly skewed data.
Next steps: standardise a value with the z-score calculator or find the centre with the mean, median and mode calculator.
Why Students Get Stuck on Standard Deviation, and What a Good Calculator Should Do
Searching for help with standard deviation usually starts late at night with one problem that will not work out. Sample and population versions differ, and the long formula has many steps. A good calculator fixes that by showing each step, checking its own answer and explaining the idea in plain words. This section covers what students struggle with, what they want, and how this page responds.
In short: students need to compute it step by step and know what it means, not just see a final number, and a calculator earns trust only when its steps can be checked.
Understand the pain points
What students get stuck on
Most mistakes here are not about arithmetic. They come from one rule applied the wrong way round, which is why seeing every step matters.
- Dividing by n instead of n − 1 for a sampleStandard deviation of 4, 8, 6, 5, 3, 7↗
- Forgetting to square the deviationsSample standard deviation of 2, 4, 4, 4, 5, 5, 7, 9↗
- Forgetting the final square rootPopulation standard deviation of 10, 12, 23, 23, 16, 23, 21, 16↗
- Not knowing what the value meansVariance of 3, 5, 7, 9↗
Understand the frustrations and desires
What they feel, and what they want
Each frustration points to a wish: a clearer step, a built-in check, an explanation in plain words.
- Wants the table of deviations shown
- Wants sample and population side by side
- Wants a plain-language reading
Build the page around their intent
Each intent has a place on this page
The page follows intent, not features, so each need has a clear starting point.
- I need an answer now: type the problem into the calculator at the top of this page.Standard deviation of 4, 8, 6, 5, 3, 7↗
- I need to understand: the sections below explain the idea in plain words and list the common mistakes.
- I need to check my work: every step is shown, so the first line that differs from yours is the one to study.
- I need practice: open the solver, tap + and choose Create Practice Test.
- I need it explained aloud: in the solver, tap + and choose Create Video.
Our View: What Is Overrated, Underrated and Wrong About Standard Deviation
Overrated. Working it out by hand for long lists. Doing the squared-deviation method once on four or five numbers shows what it means; after that it is only arithmetic risk.
Underrated. The range divided by 4 as a quick check. For roughly bell-shaped data the standard deviation is often near a quarter of the range. For 2, 4, 4, 4, 5, 5, 7, 9 the range is 7, so the answer should be near 2, and it is exactly 2. An answer of 20 or 0.2 points to a typo.
Plain wrong. “Always divide by n.” That gives the population standard deviation. When the numbers are only a sample from a larger group, divide by n − 1. The same eight numbers give 2 as a population and about 2.14 as a sample.
Test it: enter 2, 4, 4, 4, 5, 5, 7, 9 and compare the population and sample results with the two values above.