Standard Deviation Calculator - Variance, Quartiles & More

Divide by n if your list is the complete population you care about, for example every reading taken during a two-hour shift when your question is how variable that shift was. Divide by n−1 if your list is a sample drawn from a larger population, for example fifty customers surveyed when you want to estimate the spread of your whole customer base. The correction exists because your mean was calculated from the same values you are measuring deviations from, which pulls it slightly toward the centre and makes each squared deviation slightly too small. Dividing by n−1 inflates the variance by n/(n−1) to cancel that bias. The practical effect is that sample standard deviations run about 5 percent above population standard deviations at n = 20, only 1 percent at n = 100, and is irrelevant past a few thousand values. Publishing a standard deviation without stating which form you used is the most common statistical omission there is.

The mean is the sum divided by the count and it weights every value fully, so a single extreme observation pulls it a long way. The median is the middle value after sorting and it barely moves when an extreme value appears, because it depends only on the order of the data, not on how far apart the values are. They coincide on symmetric data and separate on skewed data. For a list such as 2, 4, 4, 4, 5, 5, 7, 9 the mean is exactly 5 while the median is 4.5, because the two largest values pull the average upward. Use the mean when you need it for further arithmetic, such as computing a percentage of the total or a weighted combination, and use the median when you want to describe a typical case in a distribution where a few large or small values dominate. If you report only one number for a skewed set, you will mislead whoever reads it.

Because the variance is measured in squared units and nobody can think in seconds squared. The squaring is a mathematical necessity rather than a choice: deviations from the mean come in opposite signs, and without squaring, a value 5 above cancels a value 5 below exactly, so every data set would report a variance of zero. Squaring makes all contributions positive, and the resulting total scales with the square of the spread. Dividing by n or n−1 gives a variance in units of time squared, which cannot be compared with anything meaningful. Taking the square root restores the original units, so a standard deviation of 0.83 seconds means the data genuinely sits about 0.83 seconds from the mean, a statement that can be checked against a specification, added to a threshold or fed into a subsequent formula. This is why the standard deviation, not the variance, is what people quote when they say how much their data varies.

The interquartile range, or IQR, is Q3 minus Q1 and it measures the width of the middle half of the data, discarding the bottom quarter and the top quarter entirely. That is exactly what makes it robust. The standard deviation uses every value in proportion to how far it sits from the mean, so one corrupted reading or one customer who entered their income as 900000 instead of 90000 inflates it dramatically. The IQR is unaffected. For the list 11.2, 11.5, 11.9, 12.1, 12.4, 13.8 the standard deviation is about 0.91 seconds and the IQR is about 0.73 seconds, a modest difference; add a single stray value of 500 and the standard deviation climbs past 180 while Q1 and Q3 barely move. Reporting the IQR alongside the median is the standard defensive pairing for skewed data such as income, house prices and latency. The common rule of thumb is that values more than 1.5 times the IQR beyond Q1 or Q3 are worth inspecting as possible outliers.

The coefficient of variation divides the standard deviation by the absolute size of the mean and expresses the result as a percentage, which turns an absolute spread into a relative one. A standard deviation of 10 is uninterpretable on its own: it is enormous if the mean is 12 and negligible if the mean is 5000. Expressing it as a ratio removes the units and the scale entirely. In the calculator, the default list has a coefficient of variation of about 35.5 percent while a tighter set of exam scores has about 10.8 percent, which tells you at a glance that the first list is scattered relative to its own size while the second is not. This is what makes it the right statistic for comparing datasets measured in different units, such as comparing the relative variability of daily sales in euros against relative variability of daily sales in dollars, or across populations with very different scales. One caveat: it becomes meaningless when the mean is close to zero, since you are dividing by something tiny.

The standard error of the mean, written SEM, is the sample standard deviation divided by the square root of n. It answers a different question from the standard deviation itself. The standard deviation describes how far your individual data points sit from the mean; the standard error describes how precisely your estimate of the mean is known. Because the mean of many observations settles down around the true value, its uncertainty shrinks as more data arrives, specifically at a rate of one over the square root of n, which means you need four times as much data to halve the standard error and sixteen times as much to quarter it. For the default list of ten values with a sample standard deviation of 10.898, the standard error is about 3.45, so a mean of 29.1 carries an uncertainty on the order of plus or minus three or so. When you report a mean in a paper or a report, reporting the standard error alongside it tells the reader how much to trust it. Do not confuse it with the standard deviation of the raw data, which is usually the larger of the two.

The calculator reports a mode only when at least one value occurs more than once, because a list where every value appears exactly once genuinely has no mode. This is the strict textbook definition and it differs from how some software behaves. Excel, for instance, can return multiple results or the lowest value in ways that surprise people. The tool also reports all tied modes rather than picking one, so the list 2, 4, 4, 4, 5, 5, 7, 9 returns both 4 and 5, since each appears three times. A bimodal distribution is often more informative than either value alone: two peaks usually mean your data contains two distinct populations, such as two customer groups or two operating conditions, and lumping them into one mean and one standard deviation will hide that structure. When you see two modes, split the data and analyse the halves separately.