Mean Absolute Deviation Calculator – Free MAD Stats Tool | StatsUnlock

Mean absolute deviation tells you how far a typical value sits from the mean, on average, ignoring whether it is above or below. Add up the absolute gaps between each value and the mean, then divide by how many values you have. The answer is in the same units as your data, dollars stay dollars, kilograms stay kilograms.

Rule of thumb: MAD = 0 means every value is identical. For roughly normal data MAD ≈ 0.798 × SD, so a MAD/SD ratio near 0.80 suggests normal-shaped data, while a ratio below 0.70 flags heavy tails or outliers.

The mean absolute deviation (MAD, also called average absolute deviation or mean deviation) is a measure of spread. It answers a simple question: on average, how far is a value from the centre of the data? Because it uses absolute values, distances above and below the mean count equally and never cancel out. The result is easy to explain to a non-statistician: "on average, readings sit 4.2 units away from the mean."

What it measures

• MAD, average distance from the mean, in the original data units.
• Median absolute deviation, median distance from the median, in the same units; robust to outliers.
• Relative MAD (CV of MAD), MAD divided by the mean, expressed as a percentage; unit-free.
• MAD/SD ratio, a quick shape diagnostic; near 0.798 for a normal distribution.
• Total absolute deviation, the numerator Σ|xᵢ − x̄|, useful for hand-checking.

A worked one-liner

If five readings are 52, 48, 55, 61, 47 the mean is 52.6. The absolute gaps are 0.6, 4.6, 2.4, 8.4, 5.6, which sum to 21.6. Divide by 5 and MAD = 4.32, a typical reading sits about 4.3 units from the mean.

Who uses it

Middle-school and high-school curricula (Common Core 6.SP.B.5c) introduce spread with MAD before standard deviation because it needs no squaring. Forecasters use the closely related Mean Absolute Error to score predictions. Quality engineers, ecologists reporting count variability, and finance analysts describing dispersion in returns all use MAD when they want a figure that is easy to explain in plain units.

How it differs from the nearest alternatives

4.6 Decision guide: what to do with your MAD

4.7 Common mistakes this calculator prevents

• Forgetting the absolute-value bars. Signed deviations from the mean always sum to exactly zero, so the "average deviation" without absolute values is always 0. The calculator shows the signed sum so you can verify this.
• Dividing by n − 1. Bessel's correction exists to make the variance unbiased. It has no justification for MAD. This tool always divides by n and states so on screen.
• Confusing the two MADs. "MAD" means mean absolute deviation here; in robust statistics it often means median absolute deviation. Both are computed and labelled separately so you never quote the wrong one.
• Comparing MADs across different units. A MAD of 4 kg and a MAD of 4 grams are not comparable. Use the relative MAD (percentage of the mean) for cross-scale comparison, which is shown in the results table.
• Reporting MAD to more decimals than the raw data. If your readings have one decimal place, MAD to six decimals is false precision. The decimal-places selector lets you match your instrument.
• Treating MAD as an inference statistic. MAD describes the sample you have. It is not a test, has no p-value, and by itself says nothing about a population.

When reporting mean absolute deviation calculator output in a paper, thesis or report, always pair the MAD with the mean and the sample size, state the units, and say explicitly which centre you used. Five ready-to-paste templates follow; press Copy on the one that matches your outlet.

5.6 Table and figure conventions

• In a descriptive table, give one row per group with columns: n, M, SD, MAD, Mdn, IQR. Align decimals and use the same precision throughout.
• Define MAD in the table note the first time it appears: "MAD = mean absolute deviation, Σ|x − M|/n."
• If you plot MAD as an error bar, say so in the caption, readers default to assuming SD or SEM.
• Never mix MAD error bars in one panel with SD error bars in another.

5.7 Discussion-section wording

In the Discussion, translate the number: "a mean absolute deviation of 4.32 mg/L means a typical sample sat roughly 4 mg/L from the batch average, which is within the ±5 mg/L tolerance stated in the protocol." Anchor the value to a threshold your reader already cares about rather than describing it as merely "low" or "high".

5.8 Final reporting checklist

• ☐ Sample size n stated for every group
• ☐ Centre (mean or median) named explicitly
• ☐ Divisor stated as n, not n − 1
• ☐ Units given, matching the raw data precision
• ☐ Mean reported alongside MAD
• ☐ Median absolute deviation added if outliers are present
• ☐ Software and tool cited (see Section 17)
• ☐ Raw data or a summary table available in the supplement

Technical notes

• All computation runs in your browser in double precision. Deviations are summed in input order; for very large datasets (n > 10⁶) accumulated floating-point error is possible but negligible at typical research sample sizes.
• The median uses the standard convention: for even n it is the average of the two central order statistics.
• Quartiles use the linear-interpolation method (equivalent to R's type = 7 and NumPy's default), so IQR may differ by a small amount from SPSS, which uses a different quartile definition.
• Choosing "Median" as the centre changes formula ③ to formula ④; the on-screen label updates accordingly so you never mis-report which centre was used.

1. Enter your data. The default tab takes comma-separated numbers, for example 52, 48, 55, 61, 47. Newlines, tabs, semicolons and spaces work too, so a column pasted straight out of Excel is fine.
2. Name your group. Click the group-name box and type something meaningful such as "Site A, dissolved oxygen". The name flows into every table, chart legend and reporting sentence.
3. Or load a sample dataset. Ten named datasets cover clinical, education, ecology, business, sports, survey and a deliberately messy set so you can see what an outlier does to MAD versus SD.
4. Or upload a file. On the Upload tab, choose a CSV or Excel file, then click the column names you want. Each column you select becomes its own group. Selected columns turn solid green; press "Load selected columns as groups".
5. Add more groups. Press "+ Add another group" to compare two, three or more sets side by side. Remove any group with its ✕ button.
6. Set the options. Choose decimal places, pick mean or median as the centre, and choose sample or population SD. Add a unit label such as "mg/L" so every sentence reads correctly.
7. Press Calculate MAD. Nothing is computed until you do, this keeps the page fast and stops half-typed data producing junk numbers.
8. Read the summary cards. Green is the headline MAD, blue the mean, purple the robust median absolute deviation, orange the MAD/SD ratio.
9. Study the four charts. Chart 1 shows which observations drive the MAD, chart 2 shows the ±MAD band, chart 3 shows distribution shape, chart 4 compares all four spread measures across groups.
10. Export. "Download Doc" saves a plain-text report; "Download PDF" opens the print dialogue with a clean A4 layout. Or copy any of the five reporting sentences straight into your manuscript.

Worked example: paste 52, 48, 55, 61, 47, name the group "Batch A", leave the centre on Mean, press Calculate. You should see MAD = 4.320, mean = 52.600, SD = 5.683, ratio = 0.760.

Everything this mean absolute deviation calculator does can be reproduced in Excel, and Excel is the fastest way to check the answer by hand, because it has a built-in function that does the whole job. The part that trips people up is not the formula. It is that Excel silently skips blank cells and text cells, so your n can shrink without any warning and your answer quietly changes. This section shows the one-line version, the long version that exposes every step, and the traps.

8.1 The Excel functions you need

8.2 Step by step, with screenshots

8.3 The awkward cases

• Grouped or frequency data. With values in A and frequencies in B, use =SUMPRODUCT(B2:B6,ABS(A2:A6-SUMPRODUCT(A2:A6,B2:B6)/SUM(B2:B6)))/SUM(B2:B6). AVEDEV cannot weight rows, so it gives the wrong answer on frequency tables.
• Several groups side by side. Put each group in its own column and drag one AVEDEV across. Do not select the whole block in one call, because that returns a single MAD for all groups pooled together.
• A moving MAD across time. Use =AVEDEV(A2:A13) and fill down for a rolling 12-period window.
• Mean absolute error instead. If you are scoring a forecast rather than describing a dataset, the formula is =AVERAGE(ABS(actual-predicted)), entered over two columns. It is the same arithmetic pointed at a different question.
• No p-value. Excel has no test for MAD, because MAD is descriptive. To test whether two groups differ in spread, use F.TEST or run Levene's test in R.

8.4 Why your Excel answer might differ from R or Python

8.5 Charting the result in Excel

Select the absolute-deviation column with its labels and insert a clustered bar chart. This is the chart that actually explains MAD, because the tallest bar is the observation doing most of the work. Two formatting changes make it publication ready: delete the gridlines and the legend, then add a vertical reference line at the MAD value using a second series formatted as a line. Add data labels showing one decimal place so the reader does not have to read values off an axis.

8.6 Excel errors you will probably hit

To reproduce this mean absolute deviation calculator in R, start by ignoring the function that looks like the obvious choice. R has a function called mad(), and it is almost certainly not the one you want. Base R's mad() returns the median absolute deviation multiplied by 1.4826. The mean absolute deviation has no dedicated base function, so you write it directly as mean(abs(x - mean(x))). That single line is the whole calculation, and everything below it is reporting.

No packages to install. This is deliberate: reviewers trust base R output, and a script with no dependencies still runs in five years.

9.1 The complete script

# Mean Absolute Deviation (MAD) - complete script
# StatsUnlock - statsunlock.com

set.seed(42)

# ---- 1. YOUR DATA: change this one line ----
x <- c(52, 48, 55, 61, 47)
group_name <- "Batch A"
unit <- "mg/L"

# From a CSV instead:
# dat <- read.csv("mydata.csv")
# x   <- dat$value

x <- x[!is.na(x)]          # drop missing values, or mean() returns NA

# ---- 2. THE CALCULATION ----
n       <- length(x)
m       <- mean(x)
mad_val <- mean(abs(x - m))            # mean absolute deviation, divides by n
sd_s    <- sd(x)                        # sample SD, divides by n - 1
med     <- median(x)
medad   <- median(abs(x - med))        # median absolute deviation, unscaled
medad_s <- medad * 1.4826            # scaled, same as base R mad()
mad_med <- mean(abs(x - med))          # mean deviation about the median
ratio   <- mad_val / sd_s
rel_mad <- 100 * mad_val / m
cv      <- 100 * sd_s / m

# ---- 3. PRINTED SUMMARY ----
cat("\n=== Mean Absolute Deviation:", group_name, "===\n")
cat(sprintf("n                        = %d\n", n))
cat(sprintf("Mean                     = %.3f %s\n", m, unit))
cat(sprintf("MAD (about the mean)     = %.3f %s\n", mad_val, unit))
cat(sprintf("Median                   = %.3f %s\n", med, unit))
cat(sprintf("Mean dev. about median   = %.3f %s\n", mad_med, unit))
cat(sprintf("Median abs. dev (raw)    = %.3f %s\n", medad, unit))
cat(sprintf("Median abs. dev x1.4826  = %.3f %s\n", medad_s, unit))
cat(sprintf("SD (sample, n-1)         = %.3f %s\n", sd_s, unit))
cat(sprintf("MAD / SD ratio           = %.3f\n", ratio))
cat(sprintf("Relative MAD             = %.2f%%\n", rel_mad))
cat(sprintf("Coefficient of variation = %.2f%%\n", cv))

# ---- 4. PLAIN-ENGLISH VERDICT ----
spread <- if (rel_mad < 10) "tight" else if (rel_mad < 30) "moderate" else "wide"
shape  <- if (abs(ratio - 0.798) < 0.06) "close to what a normal distribution predicts"
        else if (ratio < 0.74) "below 0.798, so outliers are inflating the SD"
        else "above 0.798, so the data are flatter than normal"
cat(sprintf("\nVERDICT: values sit on average %.3f %s from the mean of %.3f.\n", mad_val, unit, m))
cat(sprintf("That is %.2f%% of the mean, a %s spread. The MAD/SD ratio is %.3f, %s.\n",
            rel_mad, spread, ratio, shape))
if (ratio < 0.74) cat("Report the median and the median absolute deviation as well.\n")

# ---- 5. PUBLICATION FIGURE AT 300 DPI ----
png("mad_figure.png", width = 2400, height = 1100, res = 300)
par(mfrow = c(1, 2), mar = c(4.2, 4.2, 2.6, 1))

barplot(abs(x - m), names.arg = seq_along(x), col = "#86efac", border = "#166534",
        main = "Absolute deviation per observation",
        xlab = "Observation", ylab = paste0("|x - mean| (", unit, ")"))
abline(h = mad_val, col = "#b45309", lwd = 2, lty = 2)
legend("topleft", sprintf("MAD = %.3f", mad_val), bty = "n", text.col = "#b45309")

stripchart(x, method = "stack", pch = 19, col = "#2563eb", offset = 0.7,
           main = "Values with the mean +/- MAD band",
           xlab = paste0("Value (", unit, ")"))
abline(v = m, col = "#166534", lwd = 2)
abline(v = c(m - mad_val, m + mad_val), col = "#b45309", lty = 2)
dev.off()

cat("\nFigure saved as mad_figure.png (300 dpi).\n")
cat("Generated by STATS UNLOCK - statsunlock.com\n")

9.2 What each part does

9.3 What the figure shows

The left panel is a bar for every observation, showing how far it sits from the mean, with a dashed line at the MAD itself. Any bar far above that line is an observation doing most of the work, which is the visual answer to "is one point driving my spread". The right panel puts the raw values on a line with the mean marked solid and the mean plus or minus MAD marked dashed, so you can see how many observations fall inside the typical distance. For the example data, four of the five sit inside the band and 61 sits well outside it.

9.4 Common R problems and fixes

9.5 Useful R one-liners

# Mean absolute deviation, the short form
mean(abs(x - mean(x)))

# Reusable function
mad_mean <- function(v) mean(abs(v - mean(v, na.rm = TRUE)), na.rm = TRUE)

# Deviation about the median instead of the mean
mean(abs(x - median(x)))

# MAD for every group in a long data frame
tapply(dat$value, dat$group, mad_mean)

# MAD for every numeric column of a wide data frame
sapply(dat[sapply(dat, is.numeric)], mad_mean)

# Rolling 12-point MAD for a time series
sapply(12:length(x), function(i) mad_mean(x[(i-11):i]))

# Bootstrap 95% confidence interval for the MAD
quantile(replicate(5000, mad_mean(sample(x, length(x), replace = TRUE))), c(.025, .975))

# Write the results to CSV
write.csv(data.frame(n = length(x), mean = mean(x), mad = mad_mean(x), sd = sd(x)),
          "mad_results.csv", row.names = FALSE)

Reproducing this mean absolute deviation calculator in Python takes one line. The mean absolute deviation is a single NumPy expression: np.mean(np.abs(x - np.mean(x))). There used to be a pandas shortcut, Series.mad(), but it was removed in pandas 2.0, so any tutorial that still uses it will raise an AttributeError on a modern install. The script below is the same calculation as the R script in Section 9 and produces the same numbers to three decimal places.

Install once with pip install numpy scipy matplotlib pandas. NumPy alone is enough if you drop the figure and the SciPy cross-check.

10.1 The complete script

# Mean Absolute Deviation (MAD) - complete script
# StatsUnlock - statsunlock.com

import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import median_abs_deviation

rng = np.random.default_rng(42)

# ---- 1. YOUR DATA: change this one line ----
x = np.array([52, 48, 55, 61, 47], dtype=float)
group_name = "Batch A"
unit = "mg/L"

# From a CSV instead:
# import pandas as pd
# x = pd.read_csv("mydata.csv")["value"].to_numpy(dtype=float)

x = x[~np.isnan(x)]                      # drop NaN, or np.mean returns nan

# ---- 2. THE CALCULATION ----
n        = x.size
m        = np.mean(x)
mad_val  = np.mean(np.abs(x - m))        # mean absolute deviation, divides by n
sd_s     = np.std(x, ddof=1)             # sample SD; ddof=1 matches R and STDEV.S
med      = np.median(x)
medad    = median_abs_deviation(x)       # raw, scale=1.0 by default
medad_s  = medad * 1.4826
mad_med  = np.mean(np.abs(x - med))      # mean deviation about the median
ratio    = mad_val / sd_s
rel_mad  = 100 * mad_val / m
cv       = 100 * sd_s / m

# ---- 3. PRINTED SUMMARY ----
print(f"\n=== Mean Absolute Deviation: {group_name} ===")
print(f"n                        = {n}")
print(f"Mean                     = {m:.3f} {unit}")
print(f"MAD (about the mean)     = {mad_val:.3f} {unit}")
print(f"Median                   = {med:.3f} {unit}")
print(f"Mean dev. about median   = {mad_med:.3f} {unit}")
print(f"Median abs. dev (raw)    = {medad:.3f} {unit}")
print(f"Median abs. dev x1.4826  = {medad_s:.3f} {unit}")
print(f"SD (sample, ddof=1)      = {sd_s:.3f} {unit}")
print(f"MAD / SD ratio           = {ratio:.3f}")
print(f"Relative MAD             = {rel_mad:.2f}%")
print(f"Coefficient of variation = {cv:.2f}%")

# ---- 4. PLAIN-ENGLISH VERDICT ----
spread = "tight" if rel_mad < 10 else "moderate" if rel_mad < 30 else "wide"
if abs(ratio - 0.798) < 0.06:
    shape = "close to what a normal distribution predicts"
elif ratio < 0.74:
    shape = "below 0.798, so outliers are inflating the SD"
else:
    shape = "above 0.798, so the data are flatter than normal"

print(f"\nVERDICT: values sit on average {mad_val:.3f} {unit} from the mean of {m:.3f}.")
print(f"That is {rel_mad:.2f}% of the mean, a {spread} spread. "
      f"The MAD/SD ratio is {ratio:.3f}, {shape}.")
if ratio < 0.74:
    print("Report the median and the median absolute deviation as well.")

# ---- 5. PUBLICATION FIGURE AT 300 DPI ----
fig, ax = plt.subplots(1, 2, figsize=(10, 4))

ax[0].bar(np.arange(1, n + 1), np.abs(x - m), color="#86efac", edgecolor="#166534")
ax[0].axhline(mad_val, color="#b45309", ls="--", lw=2, label=f"MAD = {mad_val:.3f}")
ax[0].set_title("Absolute deviation per observation")
ax[0].set_xlabel("Observation")
ax[0].set_ylabel(f"|x - mean| ({unit})")
ax[0].legend(frameon=False)

ax[1].scatter(x, np.zeros_like(x), s=70, color="#2563eb", zorder=3)
ax[1].axvline(m, color="#166534", lw=2, label=f"Mean = {m:.3f}")
ax[1].axvspan(m - mad_val, m + mad_val, color="#fde68a", alpha=0.45, label="Mean +/- MAD")
ax[1].set_yticks([])
ax[1].set_title("Values with the mean +/- MAD band")
ax[1].set_xlabel(f"Value ({unit})")
ax[1].legend(frameon=False, loc="upper left")

fig.tight_layout()
fig.savefig("mad_figure.png", dpi=300, bbox_inches="tight")

print("\nFigure saved as mad_figure.png (300 dpi).")
print("Generated by STATS UNLOCK - statsunlock.com")

10.2 What each part does

10.3 What the figure shows

The left panel gives one bar per observation, its height being the distance from the mean, with a dashed line at the MAD. The right panel places the raw values on a number line, marks the mean with a solid line, and shades the mean plus or minus MAD band. Reading the two together answers the only two questions people really have about spread: how far is a typical value from the centre, and is one observation doing most of the work.

10.4 Common Python problems and fixes

10.5 Useful Python one-liners

# Mean absolute deviation, the short form
np.mean(np.abs(x - np.mean(x)))

# NaN-safe version
np.nanmean(np.abs(x - np.nanmean(x)))

# Deviation about the median instead of the mean
np.mean(np.abs(x - np.median(x)))

# pandas, post-2.0 (Series.mad was removed)
(df["value"] - df["value"].mean()).abs().mean()

# MAD per group
df.groupby("group")["value"].apply(lambda v: (v - v.mean()).abs().mean())

# MAD of every numeric column
df.select_dtypes("number").apply(lambda v: (v - v.mean()).abs().mean())

# Rolling 12-point MAD for a time series
df["value"].rolling(12).apply(lambda v: np.mean(np.abs(v - np.mean(v))), raw=True)

# Bootstrap 95% confidence interval for the MAD
np.percentile([np.mean(np.abs(s - s.mean()))
               for s in rng.choice(x, (5000, x.size))], [2.5, 97.5])

Running it: save as mad.py and run python mad.py from a terminal. In Jupyter, drop the savefig line and add plt.show() instead. All three sections, Excel, R and Python, use the same five values and return the same MAD of 4.320, mean of 52.600 and SD of 5.683.

11.1 Converting between spread measures (normal data)

Use these factors only when the data are approximately normal. Read across: multiply the row measure by the factor to approximate the column measure.

Conclusion: if your observed MAD/SD ratio is far from 0.798, your data are not normal and these conversions should not be used.

11.2 Interpreting relative MAD (MAD as % of the mean)

Conclusion: these bands are conventions, not laws. Always compare against a threshold from your own field before calling variability "high".

11.3 MAD/SD ratio as a distribution-shape diagnostic

Conclusion: the ratio is a fast, free normality screen that needs no test statistic. For a uniform distribution the theoretical ratio is 0.866; for a Laplace distribution it is 0.707.

11.4 Worked mini-datasets you can verify by hand

Conclusion: use row 3 to confirm the calculator is working. Row 4 shows how a single extreme value inflates SD faster than MAD.

11.5 Which spread measure to report

Conclusion: MAD is a description, not an inference tool. Pair it with the measure your downstream analysis actually requires.

Eight complete worked cases, each with the full step-by-step arithmetic, a complete statistics panel, an inline figure and a copy-paste cross-check for Excel, R and Python. Every number below is reproducible in the calculator above. Together the set covers a clean textbook case, an outlier, a zero-variance column, a tiny sample, a two-group comparison, skewed data, negative values and a classroom example.

Excel   =AVEDEV(A1:A5)                     → 4.320
R       x <- c(52, 48, 55, 61, 47)
        mean(abs(x - mean(x)))              → 4.320
        mad(x)                              → 5.930   # median-based, scaled
Python  import numpy as np
        x = np.array([52, 48, 55, 61, 47])
        np.mean(np.abs(x - x.mean()))       → 4.320

Excel   =AVEDEV(A1:A5)                     → 179.040
R       x <- c(52, 48, 55, 610, 47)
        mean(abs(x - mean(x)))              → 179.040
        mad(x)                              → 5.930   # median-based, scaled
Python  import numpy as np
        x = np.array([52, 48, 55, 610, 47])
        np.mean(np.abs(x - x.mean()))       → 179.040

Excel   =AVEDEV(A1:A6)                     → 0.000
R       x <- c(5, 5, 5, 5, 5, 5)
        mean(abs(x - mean(x)))              → 0.000
        mad(x)                              → 0.000   # median-based, scaled
Python  import numpy as np
        x = np.array([5, 5, 5, 5, 5, 5])
        np.mean(np.abs(x - x.mean()))       → 0.000

Excel   =AVEDEV(A1:A3)                     → 21.333
R       x <- c(310, 352, 298)
        mean(abs(x - mean(x)))              → 21.333
        mad(x)                              → 17.791   # median-based, scaled
Python  import numpy as np
        x = np.array([310, 352, 298])
        np.mean(np.abs(x - x.mean()))       → 21.333

Excel   =AVEDEV(A1:A5)                     → 5.200
R       x <- c(42, 58, 50, 45, 55)
        mean(abs(x - mean(x)))              → 5.200
        mad(x)                              → 7.413   # median-based, scaled
Python  import numpy as np
        x = np.array([42, 58, 50, 45, 55])
        np.mean(np.abs(x - x.mean()))       → 5.200

Excel   =AVEDEV(A1:A9)                     → 53.778
R       x <- c(28, 31, 33, 35, 38, 41, 44, 52, 310)
        mean(abs(x - mean(x)))              → 53.778
        mad(x)                              → 8.896   # median-based, scaled
Python  import numpy as np
        x = np.array([28, 31, 33, 35, 38, 41, 44, 52, 310])
        np.mean(np.abs(x - x.mean()))       → 53.778

Excel   =AVEDEV(A1:A6)                     → 1.167
R       x <- c(-2.1, 0.4, 1.8, -0.9, 0.6, -1.2)
        mean(abs(x - mean(x)))              → 1.167
        mad(x)                              → 1.334   # median-based, scaled
Python  import numpy as np
        x = np.array([-2.1, 0.4, 1.8, -0.9, 0.6, -1.2])
        np.mean(np.abs(x - x.mean()))       → 1.167

Excel   =AVEDEV(A1:A5)                     → 12.000
R       x <- c(60, 95, 88, 70, 87)
        mean(abs(x - mean(x)))              → 12.000
        mad(x)                              → 11.861   # median-based, scaled
Python  import numpy as np
        x = np.array([60, 95, 88, 70, 87])
        np.mean(np.abs(x - x.mean()))       → 12.000

Study design. The mean absolute deviation assumes a single set of independent measurements on one continuous variable, all recorded in the same unit on the same measurement scale.

Numbered collection steps

1. Define the population. State exactly what you want the spread to describe, for example "dissolved oxygen in the lower reach of the Cauvery during March 2026", not just "water quality".
2. Define the sampling frame. List the units you can actually reach (sites, patients, batches, students) and note anything systematically excluded.
3. Choose the selection method. Simple random, systematic or stratified sampling. Convenience sampling is acceptable for teaching but must be declared, because MAD describes only the sample you took.
4. Fix the instrument and unit. One instrument, one calibration, one unit for the whole dataset. Never mix mg/L with ppm or kg with lb.
5. Record the resolution. Note the smallest increment the instrument reports; your MAD should never be given to more decimals than that.
6. Set the measurement schedule. Fix the times, dates and order of measurement in advance so drift does not masquerade as spread.
7. Decide the outlier and missing-value rules before collecting. Write down in advance what you will do with an implausible reading. Deciding afterwards is how bias gets in.
8. Record raw data in one long column per group, exactly matching the layout below, with a header row containing the group name.
9. Keep the raw file untouched. Do all cleaning in a copy so the original is auditable.
10. Log the metadata: who measured, when, with what instrument, and any deviations from protocol.

Required data layout

One column per group, one observation per row, header row with the group name and unit. Upload this file on the Upload tab and click each column to load it as its own group.

Minimum sample size

MAD is descriptive and has no formal power requirement, but its stability does depend on n. As a working rule use n ≥ 10 for a usable estimate, n ≥ 30 for one you would publish without caveats, and always report the raw values when n < 10. For inferential planning, where you need a target n before collecting anything, use the StatsUnlock sample size calculator. To summarise the same dataset in other ways, browse the full set of descriptive statistics calculators.

Pre-registration and ethics

Write down the analysis plan, which centre, which spread measure, which outlier rule, before you look at the data, and store it with a timestamp (OSF, AsPredicted or a dated lab-notebook entry). If your measurements involve people or animals, obtain ethics approval and record the approval number alongside the dataset.

Common collection mistakes

• Measuring the same unit twice and treating both rows as independent observations.
• Switching instruments or observers midway through, so drift is recorded as natural spread.
• Rounding raw readings before entry, which shrinks the MAD artificially.
• Silently dropping readings that "looked wrong", which shrinks the MAD a great deal.
• Mixing two genuinely different populations into one column, producing a bimodal set where no single MAD is meaningful.

This free mean absolute deviation calculator is designed for anyone who needs a plain-units description of how scattered a set of numbers is, students meeting spread for the first time, researchers writing a descriptive-statistics table, analysts checking process consistency, and teachers preparing worked examples.

Decision checklist

• ✅ Your variable is continuous or at least interval-scaled (temperature, mass, score, time, concentration).
• ✅ All values share one unit and one measurement scale.
• ✅ You want a spread figure expressed in the original units, not squared units.
• ✅ You want something you can explain to a non-statistician in one sentence.
• ❌ Do not use it as the input to a t-test, ANOVA or confidence interval, those need the standard deviation.
• ❌ Do not use it on nominal categories (blood group, species name), distances have no meaning there.
• ❌ Do not rely on it alone when a single extreme value dominates, use the median absolute deviation.
• ❌ Do not use it to compare datasets in different units, use the relative MAD instead.

Real-world examples

1. Education. A teacher compares two classes with identical means and shows that one is far more consistent, the MAD makes that visible without introducing squares.
2. Ecology and wildlife. A researcher reports the mean absolute deviation of daily camera-trap detection rates to describe how variable animal activity was across the survey period.
3. Clinical and laboratory work. A lab reports the MAD of repeated control-sample assays as a plain-units precision figure that technicians can compare directly to the tolerance.
4. Business and operations. A supply-chain analyst reports the mean absolute deviation of daily order volumes to size buffer stock, because the figure is in units the warehouse actually counts.

Decision tree

Need a spread measure? ├─ Feeding an inferential test (t, F, CI)? ────────► Standard deviation ├─ Outliers present or strongly skewed data? ─────► Median absolute deviation ├─ Comparing across different units or scales? ───► Relative MAD / coefficient of variation ├─ Scoring forecast accuracy? ────────────────────► Mean absolute error (MAE) └─ Want plain units, easy to explain, no squaring ► MEAN ABSOLUTE DEVIATION (this tool)

Assumptions

1. The variable is at least interval-scaled. Check that differences between values are meaningful. If the data are nominal categories, MAD is meaningless and no amount of arithmetic will fix it.
2. All values share one unit. Check the source column headers. Mixing units does not produce an error message, it silently produces a wrong, usually inflated, MAD.
3. Observations are distinct units. Check that no subject appears twice. Duplicated rows pull the mean toward the duplicated value and shrink the MAD downward.
4. The data come from one population. Check the histogram in chart 3 for two peaks. A bimodal mixture produces a large MAD that describes neither subgroup.
5. Missing values are absent, not zero. Check that blanks were not filled with 0. Zeros entered as placeholders drag the mean down and inflate MAD substantially.
6. The recorded precision is adequate. Check the instrument resolution. Heavy rounding truncates real variation and biases MAD downward.

Limitations

• MAD is purely descriptive. It has no p-value, no confidence interval by default, and supports no hypothesis test. For inference on spread, use Levene's test or a bootstrap interval.
• MAD uses the mean as its centre, and the mean is not robust. For genuinely outlier-resistant spread, use the median absolute deviation, which this tool also reports.
• MAD is not mathematically convenient. It has no simple additive decomposition, which is why variance and SD, not MAD, underpin ANOVA, regression and least-squares theory.
• This tool assumes simple random or convenience sampling. It does not handle survey weights, clustering, stratification or repeated-measures structure. For those designs use a survey package (R survey, Stata svy).
• No bootstrap confidence interval for MAD is computed. If you need one, resample your data 5,000 times in R or Python and take the 2.5th and 97.5th percentiles of the resampled MADs.
• Group comparisons here are descriptive only. To test whether two spreads differ, use Levene's or the Brown–Forsythe test rather than eyeballing two MADs.

This mean absolute deviation calculator turns a column of numbers into a single, honest statement about how scattered those numbers are. It finds the mean, measures how far each value sits from it, ignores the direction of that gap, and averages the result. Because nothing is squared, the answer stays in your original units, which is why a mean absolute deviation of 4.32 mg/L needs no translation for a reader who has never taken a statistics course.

The single most important interpretation rule is to read MAD against the mean, not on its own. A MAD of 4 means very different things when the mean is 50 and when the mean is 4,000. Divide MAD by the mean, express it as a percentage, and you have a figure you can compare across studies, instruments and scales, provided all your values are positive and the mean is not close to zero.

The most common mistake is quiet and easy to miss: forgetting the absolute-value bars. Signed deviations from the mean always cancel to exactly zero, so an "average deviation" computed without absolute values is always 0, for every dataset ever collected. The second most common mistake is dividing by n − 1. Bessel's correction exists to remove bias from the variance estimator; it has no role in the mean absolute deviation, which always divides by n.

Remember that MAD is a description, not a decision. It cannot tell you whether two groups differ, whether an effect is real, or whether an extreme value should be removed. It is the number you put in a descriptive table, in a figure caption, and in the sentence that tells a reader whether your mean is worth trusting. When outliers are present, quote the median absolute deviation beside it and let the reader see both.

Next, load your own data into the mean absolute deviation calculator above, name your groups, read the four charts, and copy whichever of the five reporting sentences matches the journal or report you are writing for. If your data are skewed or contain a genuine extreme value, take the robust route and lead with the median and the median absolute deviation instead.

Each of these takes an output from this page as its input, so you can carry the same dataset through a full descriptive analysis.

Ten multiple-choice questions on the mean absolute deviation calculator and the statistics behind it. Pick an answer to see instant feedback and a short explanation. Your score updates as you go.

The following peer-reviewed references support the statistical methods used in this mean absolute deviation calculator, covering measures of dispersion, robust statistics, effect size and variability reporting, and best practice in descriptive data analysis for research.

1. Gorard, S. (2005). Revisiting a former idea: The mean deviation. British Journal of Educational Studies, 53(4), 417–430. https://doi.org/10.1111/j.1467-8527.2005.00304.x
2. Gorard, S. (2015). Introducing the mean absolute deviation "effect" size. International Journal of Research & Method in Education, 38(2), 105–114. https://doi.org/10.1080/1743727X.2014.1002090
3. Pham-Gia, T., & Hung, T. L. (2001). The mean and median absolute deviations. Mathematical and Computer Modelling, 34(7–8), 921–936. https://doi.org/10.1016/S0895-7177(01)00109-1
4. Leys, C., Ley, C., Klein, O., Bernard, P., & Licata, L. (2013). Detecting outliers: Do not use standard deviation around the mean, use absolute deviation around the median. Journal of Experimental Social Psychology, 49(4), 764–766. https://doi.org/10.1016/j.jesp.2013.03.013
5. Rousseeuw, P. J., & Croux, C. (1993). Alternatives to the median absolute deviation. Journal of the American Statistical Association, 88(424), 1273–1283. https://doi.org/10.1080/01621459.1993.10476408
6. Huber, P. J. (1964). Robust estimation of a location parameter. The Annals of Mathematical Statistics, 35(1), 73–101. https://doi.org/10.1214/aoms/1177703732
7. Tukey, J. W. (1977). Exploratory data analysis. Addison-Wesley. https://archive.org/details/exploratorydataa0000tuke
8. Fisher, R. A. (1920). A mathematical examination of the methods of determining the accuracy of an observation by the mean error, and by the mean square error. Monthly Notices of the Royal Astronomical Society, 80(8), 758–770. https://doi.org/10.1093/mnras/80.8.758
9. Willmott, C. J., & Matsuura, K. (2005). Advantages of the mean absolute error (MAE) over the root mean square error (RMSE) in assessing average model performance. Climate Research, 30(1), 79–82. https://doi.org/10.3354/cr030079
10. Chai, T., & Draxler, R. R. (2014). Root mean square error (RMSE) or mean absolute error (MAE)? Geoscientific Model Development, 7(3), 1247–1250. https://doi.org/10.5194/gmd-7-1247-2014
11. Hyndman, R. J., & Koehler, A. B. (2006). Another look at measures of forecast accuracy. International Journal of Forecasting, 22(4), 679–688. https://doi.org/10.1016/j.ijforecast.2006.03.001
12. Wilcox, R. R., & Keselman, H. J. (2003). Modern robust data analysis methods: Measures of central tendency. Psychological Methods, 8(3), 254–274. https://doi.org/10.1037/1082-989X.8.3.254
13. Erceg-Hurn, D. M., & Mirosevich, V. M. (2008). Modern robust statistical methods. American Psychologist, 63(7), 591–601. https://doi.org/10.1037/0003-066X.63.7.591
14. Cousineau, D., & Chartier, S. (2010). Outliers detection and treatment: A review. International Journal of Psychological Research, 3(1), 58–67. https://doi.org/10.21500/20112084.844
15. Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Lawrence Erlbaum Associates. https://doi.org/10.4324/9780203771587
16. American Psychological Association. (2020). Publication manual of the American Psychological Association (7th ed.). https://doi.org/10.1037/0000165-000
17. Field, A. (2018). Discovering statistics using IBM SPSS statistics (5th ed.). SAGE Publications. https://uk.sagepub.com/en-gb/eur/discovering-statistics-using-ibm-spss-statistics/book257672
18. NIST/SEMATECH. (2013). e-Handbook of statistical methods, Measures of scale. National Institute of Standards and Technology. https://www.itl.nist.gov/div898/handbook/eda/section3/eda356.htm
19. R Core Team. (2024). R: A language and environment for statistical computing. R Foundation for Statistical Computing. https://www.R-project.org/
20. Virtanen, P., Gommers, R., Oliphant, T. E., Haberland, M., Reddy, T., Cournapeau, D., … SciPy 1.0 Contributors. (2020). SciPy 1.0: Fundamental algorithms for scientific computing in Python. Nature Methods, 17, 261–272. https://doi.org/10.1038/s41592-019-0686-2