standarddeviationcalculator.net

Updated

Spread, shape and outlier calculators

Calculators that measure how spread out a dataset is, how lopsided its distribution is, and which values look like outliers. They also cover relative spread for comparing datasets on different scales, and the capability indices used in quality control.

Which calculator do I need?

You have or wantUse
Every measure of spread at once, to compare themDispersion calculator
The simplest spread: largest minus smallestRange calculator
The average distance from the mean, without squaringMean absolute deviation
A spread measure that outliers cannot distortMedian absolute deviation
Compare spread across datasets with different units or meansCoefficient of variation
Whether the distribution leans left or rightSkewness calculator
Which values are outliersOutlier calculator
Whether a process fits inside its specification limitsProcess capability index

Full summaries

One paste gives the whole picture: centre, spread and shape in a single table.

Range and deviation measures

Absolute measures of spread, in the same units as the data, from the simplest to the most outlier-resistant.

Relative spread and categorical variation

Spread as a ratio or percentage of the mean, and variation for data that are categories rather than numbers.

Shape and outliers

How lopsided the distribution is and which values sit unusually far from the rest.

Quality control

Spread measured against a specification or a peer group, as used in manufacturing and clinical labs.

Measures of spread compared

Two datasets can share a mean and still look nothing alike. Both of these have a mean of 50:

MeasureA: 48, 49, 50, 51, 52B: 30, 40, 50, 60, 70
Range440
Mean absolute deviation1.212
Sample standard deviation1.58115.81
Coefficient of variation3.2%31.6%

Every measure agrees that B is ten times as spread out; they differ in how they get there and how they react to awkward data.

  • The range uses only the two extremes, so a single outlier changes it completely.
  • The mean absolute deviation averages distances without squaring them, so it is easy to explain and less swayed by one large value than the SD.
  • The standard deviation squares the distances, which gives large deviations more weight. It is the measure that connects to the normal distribution, confidence intervals and tests.
  • The median absolute deviation is built from medians, so up to half the data can be wild before it breaks. Multiplied by 1.4826 it estimates σ for normal data.
  • The coefficient of variation divides by the mean, which lets you compare the spread of weights in kilograms with heights in centimetres. Use it only for data on a ratio scale with a true zero.

Shape and outliers

Skewness near zero means a roughly symmetric distribution; positive skew means a long right tail (a few large values), negative skew a long left tail. The outlier calculator runs Tukey fences, z-scores, modified z-scores and Grubbs’ test together, because they do not always agree: fences built on quartiles hold up well with skewed data, while z-score rules assume rough normality and can be masked by the very outliers they are looking for.

An outlier is a prompt to check the value, not a licence to delete it. Look for typing or measurement errors first, and report any values you exclude.

Guides to read alongside

Common questions

Which measure of spread should I report?
Report the standard deviation for roughly symmetric data, since most readers and methods expect it. For skewed data or data with outliers, report the interquartile range or median absolute deviation alongside the median.
What is the difference between mean absolute deviation and median absolute deviation?
Mean absolute deviation is the average distance of each value from the mean. Median absolute deviation is the median distance of each value from the median, which makes it far more resistant to outliers. Both are sometimes abbreviated MAD, so say which you mean.
How much skewness is too much?
A common rule of thumb treats skewness between −0.5 and 0.5 as roughly symmetric, 0.5 to 1 in either direction as moderate, and beyond ±1 as highly skewed. With small samples the estimate itself is noisy, so check the standard error the skewness calculator reports.