The central limit theorem calculator above describes the sampling distribution of the sample mean. Give it a population mean, a population standard deviation and a sample size, and it returns the standard error, the z-scores of any two bounds you choose, and the probability that a randomly drawn sample mean lands between them. It also plots how quickly the standard error falls as the sample grows.
Arb Digital publishes this because the central limit theorem is the result that makes most of applied statistics work, and it is routinely stated in a form that is not quite right. It does not say that data become normal with a large enough sample. It says that the distribution of the sample mean approaches a normal shape as the sample size grows, regardless of the shape of the population it was drawn from. Those are very different claims, and the difference decides which questions this page can answer.
What This Central Limit Theorem Calculator Does
It builds the sampling distribution of the mean. That distribution has the same centre as the population, a spread equal to the population standard deviation divided by the square root of the sample size, and — for a sufficiently large sample — an approximately normal shape. The tool computes that spread, converts your bounds into z-scores against it, and integrates the normal curve between them.
This is deliberately not the same job as our normal distribution calculator, which works on individual values drawn from a normal population. That page answers "what proportion of people are taller than 190 cm". This page answers "what is the chance the average height of 25 randomly chosen people exceeds 190 cm". The second probability is far smaller, because averaging suppresses variation, and the standard error is exactly the number that quantifies how much.
The other important difference is the assumption each page makes. The normal distribution calculator needs the population itself to be normal. This one does not: the central limit theorem does the work, so the population can be skewed, bimodal or discrete, provided it has a finite variance and the sample is large enough.
How to Use It
- Enter the population mean and standard deviation. These describe individual values, not averages.
- Set the sample size. This is the number of observations in one sample, not the number of samples you plan to take.
- Set the two bounds for the sample mean. Symmetric bounds around the population mean give you the familiar coverage probabilities.
- Read the standard error first. It is the single most useful number on the page and it drives everything else.
- Check the bars to see how the standard error behaves as the sample grows, before deciding whether a larger sample is worth the cost.
The Formula and How It Is Calculated
The sampling distribution of the mean has mean μ and standard deviation σ ÷ √n, which is called the standard error. A bound is converted to a z-score by z = (bound − μ) ÷ standard error, and the probability between two bounds is the area under the standard normal curve between the two z-scores.
Work the defaults through. The population mean is 100 with a standard deviation of 15, and the sample size is 25. The standard error is 15 ÷ √25 = 15 ÷ 5 = 3. The lower bound of 97 is one standard error below the mean, so z = −1, and the upper bound of 103 is one above, so z = +1. The area between −1 and +1 under the normal curve is 0.6827, so there is about a 68.3 percent chance that the mean of 25 randomly drawn observations falls between 97 and 103.
Note what changed and what did not. The population standard deviation is 15, and an individual value between 97 and 103 would be well within one fifth of a standard deviation of the mean — a much weaker statement. Averaging 25 observations cut the spread by a factor of five, which is the square root of 25. That square root is the whole mechanism, and it is why sample size has diminishing returns: going from 25 to 100 observations only halves the standard error again. Wolfram MathWorld's entry on the central limit theorem states the convergence formally, including the variance-divided-by-n result the standard error comes from.
Why the Theorem Holds: Averaging Cancels Extremes
The mechanism is easier to see than the proof. When you average n independent observations, an unusually high value is only unusual for the average if the other n − 1 values fail to pull it back. For a single observation, one extreme value is the whole result. For an average of 25, it has to survive being diluted by 24 others. The larger the sample, the more coordinated the extremes would have to be to move the mean, and independent draws do not coordinate.
The normal shape emerges from the same argument. The number of ways to land near the middle of the range vastly exceeds the number of ways to land at an extreme, because there are many combinations of individual values that average out to something typical and very few that all point the same way. That combinatorial imbalance produces a peaked, symmetric, thin-tailed distribution regardless of what the individual values looked like. Roll a single die and the distribution is flat; average five dice and it is already visibly bell-shaped, even though the die itself never changed.
How Large Does the Sample Need to Be?
The rule of thumb that a sample of 30 is enough is repeated far more often than it is checked. It is a reasonable default for populations that are roughly symmetric and free of heavy tails, and it is badly optimistic otherwise.
Three properties of the population decide how large is large enough. Skewness matters most: strongly right-skewed data such as income, session duration or order values may need several hundred observations before the sample mean is convincingly normal. Heavy tails matter next, because rare extreme values dominate the average and slow the convergence sharply. Discreteness matters least but is not nothing — averaging a small number of yes-or-no outcomes produces a lumpy distribution that no amount of theory smooths out. And a population with infinite variance, such as a Cauchy distribution, never converges at all, because the theorem's precondition fails. Penn State's STAT 500 applied statistics course devotes a lesson to sampling distributions and the conditions behind them.
Standard Deviation Versus Standard Error
These two are confused constantly, including in published work, and the difference is not subtle. The standard deviation describes how much individual observations vary. The standard error describes how much a sample statistic varies from sample to sample. One is a property of the population; the other is a property of your estimation procedure and shrinks as you collect more data.
The practical test is what an error bar on a chart is claiming. A bar showing plus or minus one standard deviation says "individual values typically spread this far". A bar showing plus or minus one standard error says "if I repeated this study, the average would typically land this far away". The second is always narrower for any sample above one, and quoting it as if it described individual variation makes data look far more consistent than it is. Our standard deviation calculator computes the first from raw data, and this page converts it into the second.
Where This Leads: Confidence Intervals and Sample Size
Everything downstream of the standard error is built on this page. A confidence interval for a mean is the sample mean plus or minus a critical value times the standard error, which is why the central 95 percent range in the results grid looks so familiar: it is the same arithmetic pointed the other way. Our confidence interval calculator builds that interval from your data, and the margin of error calculator reports its half-width.
Sample size planning inverts it again. If you need a margin of error no wider than some target, set the standard error accordingly and solve for n, which our sample size calculator does directly. The square root is what makes this planning painful: halving your margin of error costs four times the data, and reducing it by a factor of ten costs a hundred times. Understanding that trade before committing to a study is worth more than any amount of analysis afterwards. To convert a specific bound into a z-score by hand, the z-score calculator does the same standardisation on individual values.
Arb Digital plans marketing experiments around the precision they actually need, so results arrive conclusive rather than merely finished.
Browse All Free Tools Talk To Our TeamCommon Mistakes to Avoid
- Believing the data become normal — the theorem is about the distribution of the sample mean, never about the individual observations.
- Dividing by n instead of the square root of n — the standard error uses the square root, and using n makes the spread look far smaller than it is.
- Trusting the rule of 30 on skewed data — heavily skewed or heavy-tailed populations need far larger samples before the approximation holds.
- Plotting standard error bars and calling them variation — error bars on a mean describe the estimate, not the spread of individual values.
- Applying it to dependent observations — the theorem assumes independent draws, and clustered or repeated-measures data break that assumption.
Related Free Tools From Arb Digital
Work with individual values using the normal distribution calculator, standardise a single figure with the z-score calculator, measure spread with the standard deviation calculator, build an interval with the confidence interval calculator, or plan a study with the sample size calculator. The full free online tools hub lists every statistics tool we publish.
Frequently Asked Questions
That the distribution of the sample mean approaches a normal shape as the sample size grows, whatever the shape of the population, provided the population has a finite variance and the observations are independent.
It is the population standard deviation divided by the square root of the sample size. It measures how much the sample mean would vary between repeated samples of the same size.
The standard deviation describes the spread of individual values in the population. The standard error describes the spread of an estimate across repeated samples, and it shrinks as the sample grows.
No. Thirty is a reasonable default for roughly symmetric populations. Strongly skewed or heavy-tailed data can need several hundred observations before the approximation is safe.
No, and that is the point of the theorem. The population can be any shape with a finite variance. Only the distribution of the sample mean is claimed to become normal.
Because the standard error divides by the square root of the sample size rather than by the sample size itself. Four times the data gives twice the precision, not four times.
The result becomes an approximation. When the population standard deviation is estimated from the same small sample, a t-distribution is the more accurate basis for the interval.
This page explains a statistical result for educational purposes. It assumes independent observations and a finite population variance, and the normal approximation is only as good as those assumptions.
