Statistics Calculator

What one pass through a data set gives you

A statistics calculator earns its keep the moment a question stops being "what is the average" and becomes "what is the average, how spread out is the list, and is any point strange". This page answers all three at once. Paste the numbers, and the widget returns the count, sum, mean, median, mode, minimum, maximum, range, both quartiles, the interquartile range, the outlier fences, the sample and population versions of the variance and standard deviation, and the standard error of the mean, each labeled with the divisor it used.

That bundle is what statisticians call a descriptive summary, and the NIST Engineering Statistics Handbook treats it as the opening move of any analysis: before you model anything, describe what you have. The summary splits into three natural groups. Centers describe the middle of the list (mean, median, mode). Spread describes how far points stray from that middle (range, variance, standard deviation, IQR). Position describes where individual points sit (quartiles, fences, the minimum and maximum).

Numbers separated by commas, spaces or line breaks all parse the same way, so a column copied out of a spreadsheet goes straight in. The worked example below runs through the whole output for ten quiz scores so each row of the table has a story behind it.

Mean and median: two centers that disagree on purpose

The mean adds everything and divides by the count. The median sorts the list and takes the middle. For the ten quiz scores 78, 85, 92, 88, 76, 95, 89, 84, 91 and 73, the sum is 419 for the first five (78 + 85 + 92 + 88 + 76 = 419) and 432 for the second five (95 + 89 + 84 + 91 + 73 = 432), so the total is 851 and the mean is 851 / 10 = 85.1. Sorting gives 73, 76, 78, 84, 85, 88, 89, 91, 92, 95, and with an even count the median is the average of the two middle scores: (85 + 88) / 2 = 86.5.

The two answers disagree by a point and a half, and the direction of the gap says something. When the mean sits below the median, the low scores are pulling it down, which usually means a mild left tail. The NIST handbook chapter on measures of location makes the broader point: the mean is the best descriptor for well-behaved, roughly symmetric data, but a handful of extreme values can drag it far from where most of the data lives, while the median, built on ranks, ignores how extreme the extremes are.

A familiar illustration is household income. One billionaire in a room of teachers moves the mean by millions and the median by nothing. Quiz scores rarely hide a billionaire, but a single 12 among 90s would, and that is exactly when the median becomes the number to quote. Neither center is the right one in general; the gap between them is itself a quick symmetry check.

Quartiles, IQR and the outlier fences

Quartiles extend the median's logic. This calculator uses the median-of-halves method: sort the list, take the median of the lower half for Q1 and the median of the upper half for Q3. For the sorted quiz scores the lower half is 73, 76, 78, 84, 85 and its median is 78, so Q1 = 78. The upper half is 88, 89, 91, 92, 95, its median is 91, so Q3 = 91. The interquartile range is the width of the middle half of the data: IQR = 91 - 78 = 13.

The IQR also powers the most common outlier test. Any point more than 1.5 IQR below Q1 or above Q3 is flagged. Here the buffer is 1.5 x 13 = 19.5, so the fences land at 78 - 19.5 = 58.5 on the low side and 91 + 19.5 = 110.5 on the high side. Every score in the list sits inside those fences, so none of the ten quizzes counts as an outlier. A missing test with a score of 40 would breach the lower fence and earn a conversation rather than a silent average.

One honest caveat: quartile conventions are not universal. Some software interpolates between orderings and can move Q1 or Q3 by a point or so on small samples, which shifts the IQR and the fences with it. The median-of-halves method used here is the one taught in most introductory courses, matching the approach in the Penn State STAT 200 lessons on describing quantitative data. With large data sets the conventions converge and the choice stops mattering.

Sample or population: pick the divisor before you divide

Variance and standard deviation both start from the same place: measure how far each point sits from the mean, square those distances, and add them up. For the quiz scores the squared deviations come to 484.9 in total. What happens next depends on a question only you can answer: is this list the entire group you care about, or a sample standing in for something bigger?

A sample divides by n - 1 to correct for the fact that a sample mean is itself estimated from the data, which makes raw spreads look slightly too small. That gives the sample variance 484.9 / 9 = 53.8778 and the sample standard deviation, its square root, 7.3401. Treating the same ten scores as a complete population divides by 10 instead: the population variance is 484.9 / 10 = 48.49 and the population standard deviation is 6.9635. The sample numbers are always the larger pair, and the gap between them shrinks as n grows; at n = 10 it is already under 6 percent, at n = 100 it is half a percent.

Class test scores are usually a sample of what the students could in principle score, so the sample column is the default in practice. The population column fits when the ten scores are the entire story, such as the ten machines on one production line.

Standard error and what it says about the mean

The standard error of the mean answers a different question from the standard deviation. The standard deviation of 7.3401 describes the spread of the individual scores. The standard error describes how steady the mean itself would be if you repeated the whole collection many times, and it shrinks as the sample grows: SEM = s / sqrt(n). For the quiz scores, the square root of 10 is about 3.1623, so the standard error is 7.3401 / 3.1623 = 2.3211. Ten observations pin the mean down roughly three times as tightly as one observation would.

Standard errors feed confidence intervals. A 95 percent interval for the true mean uses the multiplier 1.96: 1.96 x 2.3211 = 4.55, so the interval runs from 85.1 - 4.55 = 80.55 up to 85.1 + 4.55 = 89.65. Read that as: if the class is typical, the long-run average sits somewhere in that band. The NIST handbook section on confidence limits for the mean walks through the same construction and warns when the small-sample version should swap 1.96 for a t value with n - 1 degrees of freedom.

Keep the two "standard" quantities straight by their units of spread: standard deviation spreads scores, standard error spreads means. Reporting "SD 7.34, SEM 2.32" tells a reader both how ragged the class is and how much to trust the 85.1.

Worked example: ten quiz scores end to end

Here is the full output for the running example, exactly as the calculator reports it.

A few rows reward a second look. The mode row saying "None" is a legitimate result, not a failure: with ten distinct values there is no most-common score. The midrange, the midpoint of the extremes, works out to (95 + 73) / 2 = 84, close to but not the same as the mean. And the mean of 85.1 sitting 1.4 below the median of 86.5 matches the visual: the two lowest scores, 73 and 76, trail farther from the pack than the top scores do.

When a full workbench beats a single-purpose calculator

Leifi already ships focused tools for several of these rows, and each is the faster choice when you want exactly one number. The table below matches the job to the tool.

The case for the workbench version is strongest when the rows interact. Deciding whether a point is an outlier needs the median, the quartiles and the IQR together. Quoting a survey result responsibly needs the mean and the SEM together. Rebuilding that from four separate single-purpose tools means re-pasting the same data four times and hoping the quartile conventions agree. Paste once here, read the whole picture, then jump to the focused tool if one row needs a deeper look.