Mean absolute deviation is the average distance between each observation and the centre of the data. It answers the same question as standard deviation — how spread out is this? — but it gets there without squaring the deviations, so the answer stays in the original units and no single distant point dominates the total. This mean absolute deviation calculator reports MAD about either the mean or the median, prints the deviation for every observation so the total can be checked by hand, and shows the standard deviation alongside for comparison.
Arb Digital included the per-point table because MAD is most often met in schools, where the marking is on the method rather than the number. It is also the measure that makes the underlying idea clearest: spread is the average distance from the middle, and every step of that sentence is visible in the table. Squaring, as standard deviation does, hides the arithmetic behind an operation that has to be justified separately.
What This Mean Absolute Deviation Calculator Does
It computes the centre of the data — mean or median, your choice — takes the absolute difference between each observation and that centre, and averages those distances. The headline figure is that average. The table below lists every observation with its signed difference and its absolute difference, so the sum can be reproduced on paper.
The supporting grid gives the mean, the median, the sample standard deviation and the median absolute deviation. Having all four together makes the comparison that matters: MAD and standard deviation measure the same thing on different scales, and the gap between them tells you something about the shape of the data. For a roughly normal sample, standard deviation runs about 25% larger than MAD, and a much larger gap is a sign of outliers or heavy tails.
This is a different statistic from the one on our standard deviation calculator, which squares the deviations, averages them and takes the square root. It is also different from our variance calculator, which stops before the square root and reports in squared units, and from our interquartile range calculator, which measures spread by the width of the middle half rather than by distance from a centre.
How to Use It
- Paste the data in any separator. Commas, spaces and line breaks all work, and a mixture is fine.
- Choose the centre. School exercises almost always mean the mean; the median option is there for the robust version and for the comparison below.
- Read the deviation table before the answer. If a single row dominates the total, the data has an outlier and the spread figure is describing that point more than the sample.
- Sort by deviation size to find outliers quickly. The largest absolute deviations rise to the top, which is faster than scanning the original list.
- Compare MAD against the standard deviation. A ratio far above the usual 1.25 for normal data is worth investigating before you report either number.
The Formula and How It's Calculated
MAD = Σ|xᵢ − x̄| ÷ n. Take the default data: 4, 8, 6, 5, 3, 10, 14, 7. The total is 57 across eight values, so the mean is 7.125. The absolute deviations are 3.125, 0.875, 1.125, 2.125, 4.125, 2.875, 6.875 and 0.125, which sum to 21.25. Dividing by 8 gives a mean absolute deviation of 2.65625.
The absolute value is not optional decoration. Without it the deviations sum to exactly zero for any data set, because that is what the mean is: the balance point where positive and negative departures cancel. This is why every measure of spread has to do something to the signs — take absolute values as MAD does, or square them as variance does. Both work, and they answer slightly different questions.
Switching the centre to the median gives 6.5 here, and the absolute deviations become 2.5, 1.5, 0.5, 1.5, 3.5, 3.5, 7.5 and 0.5, summing to 21 and averaging 2.625. That is smaller than the value about the mean, and it is smaller for every data set — the median is the point that minimises total absolute distance, a fact that turns out to be the defining property of a median. The related measures of scale and how they differ are catalogued in the NIST/SEMATECH e-Handbook section on measures of scale.
Why the Median Minimises Absolute Deviation
Pick any point c and ask how the total Σ|xᵢ − c| changes as c moves slightly to the right. Every observation to the left of c gets further away by that small step; every observation to the right gets closer by the same amount. The total therefore changes by the number of points on the left minus the number on the right, times the step.
The total is at its minimum when moving in either direction makes it worse, which happens precisely when the counts on either side are equal. That is the definition of the median. Move away from it and the imbalance in the counts pushes the total up, no matter which way you go.
The same argument run with squared deviations gives the mean instead, because squaring makes distant points contribute proportionally more and the balance shifts toward them. This one difference explains almost everything about the two statistics: absolute deviation pairs with the median, squared deviation pairs with the mean, and the choice of centre and the choice of penalty are the same decision made twice. The range of centre measures this choice sits between is summarised in the NIST/SEMATECH e-Handbook section on measures of location.
MAD Versus Standard Deviation
Standard deviation squares each deviation before averaging, so an observation twice as far from the centre contributes four times as much rather than twice as much. That makes standard deviation more sensitive to outliers. One badly mistyped value can inflate it substantially while barely moving MAD, which is exactly the behaviour you want if you are trying to describe a typical distance and exactly what you do not want if the extreme values are the point of the analysis.
Standard deviation nonetheless dominates practical statistics, and not from habit. Variances of independent quantities add, which absolute deviations do not; the normal distribution is defined through its standard deviation; least squares, regression, analysis of variance and most of inferential statistics are built on squared error because squares are differentiable everywhere and absolute values are not. MAD has no comparable theory attached to it.
For a sample drawn from a normal distribution the two are related by a constant: standard deviation is roughly 1.25 times MAD, because the expected absolute deviation of a normal variable is √(2/π) times its standard deviation. That ratio is a quick diagnostic. A ratio near 1.25 is consistent with normality; a much larger one indicates heavy tails or contamination, and is worth checking before running any test that assumes normality, such as those on our t-test calculator.
The Two Statistics Called MAD
The abbreviation is used for two different things and the confusion is genuine. Mean absolute deviation is the average of the absolute deviations, usually about the mean. Median absolute deviation is the median of the absolute deviations about the median. This tool reports both — the first as the headline figure, the second in the results grid.
The distinction matters because their behaviour with outliers is completely different. Mean absolute deviation still averages, so an extreme point pulls it up, just less violently than it pulls standard deviation. Median absolute deviation takes the middle of the distances, so it ignores extremes almost entirely and has what statisticians call a 50% breakdown point: half the data can be arbitrarily corrupted before the statistic moves.
Median absolute deviation is therefore the standard robust scale estimate, and it is usually multiplied by 1.4826 to make it comparable to a standard deviation under normality. Outlier detection rules often flag any point more than three scaled median absolute deviations from the median, which is a more reliable rule than the equivalent based on standard deviations, precisely because the outliers cannot inflate the yardstick used to measure them.
Where Mean Absolute Deviation Is the Right Choice
Forecasting is the clearest case. Forecast accuracy is routinely reported as mean absolute error, which is a mean absolute deviation of the errors about zero, because it is directly interpretable: an average error of 40 units means what it says, while a root mean squared error of 40 units means something more abstract. Any forecast evaluation that will be read by a non-specialist is better served by the absolute version.
Teaching is the second case. MAD is introduced before standard deviation in most curricula because it needs no justification beyond the plain-language definition of spread. A learner who has computed a MAD table by hand understands what a deviation is, which makes the later jump to squaring a much smaller step.
The third case is any setting where the cost of an error really is proportional to its size. Squared error implicitly says that one error of ten is worse than ten errors of one. That is true for some problems — structural loads, financial risk — and false for others, such as delivery time slippage where the inconvenience scales roughly with the delay. Choosing the measure that matches the actual cost function is the substantive decision, and the arithmetic follows from it. Our mean squared error calculator covers the squared-cost case.
Arb Digital builds dashboards where the accuracy measure matches the decision being made, rather than whichever statistic the tool happened to output.
Browse All Free Tools Talk To Our TeamCommon Mistakes to Avoid
- Forgetting the absolute values — the signed deviations about the mean sum to zero every time, so dividing that total by n gives zero rather than a spread.
- Dividing by n − 1 — the Bessel correction belongs to sample variance, not to mean absolute deviation, which divides by n.
- Mixing up the two MADs — mean absolute deviation averages the distances, median absolute deviation takes their median, and they can differ substantially.
- Comparing MAD against standard deviation as though they were the same scale — for normal data the standard deviation is about 25% larger by construction.
- Reporting spread without checking the deviation table — if one row supplies most of the total, the statistic is describing an outlier rather than the sample.
Related Free Tools From Arb Digital
Compare against the squared-deviation measure on the standard deviation calculator, work in squared units with the variance calculator, locate the centre with the mean, median and mode calculator, measure spread by the middle half with the interquartile range calculator, or standardise a single observation with the z-score calculator. The full free online tools hub lists every statistics tool we publish.
Frequently Asked Questions
Add up the absolute difference between each value and the mean, then divide by the number of values. Written out it is the sum of the absolute deviations divided by n.
Because the signed deviations about the mean always add to zero, which is what makes the mean the balance point. Taking absolute values is what turns them into distances that can be averaged.
By n. The n minus one correction applies to sample variance and standard deviation, where it compensates for estimating the mean from the same data. Mean absolute deviation uses n.
For real data it almost always is, and for a normal distribution the standard deviation is about 1.25 times larger. Squaring gives distant points more weight, which pushes the standard deviation up.
Mean absolute deviation averages the distances from the centre. Median absolute deviation takes the median of the distances from the median, which makes it far more resistant to extreme values.
Because moving the centre one step to the right adds distance for every point on the left and removes it for every point on the right. The total stops improving only when the counts on both sides are equal, which defines the median.
When the audience needs an interpretable average error in the original units, when the data has outliers you do not want magnified, or when the real cost of an error is proportional to its size rather than its square.
This page explains a statistical method for study and for checking your own working. It is not a substitute for advice from a qualified statistician, and it is not medical, legal, or financial advice.
