The mean is the arithmetic average: add all the values together and divide by how many there are.

It is the balance point of your data, the value each observation would take if the total were shared out equally. The mean uses every number in the dataset, which makes it efficient and the basis of most statistical tests, but it also means a single extreme value can drag it a long way from the typical observation. When that happens, the median is usually the better summary.

Rule of thumb: use the mean when the data are roughly symmetric with no extreme values. If the mean and median differ by more than about 10%, the data are skewed and you should report the median instead.

The mean is what most people call the average. You add up every value and divide by how many values there are. Physically it is the balance point of the data: if you placed each observation as an equal weight along a ruler, the mean is where the ruler would balance. That is also why it is so sensitive to extremes, because a weight placed far out on one end shifts the balance point a long way.

The seven averages this calculator reports:

• Arithmetic mean, the ordinary average, sum divided by count.
• Median, the middle value once sorted. Unaffected by extremes.
• Mode, the most frequent value. The only average that works for categories.
• Geometric mean, the nth root of the product. Correct for growth rates, ratios and returns.
• Harmonic mean, n divided by the sum of reciprocals. Correct for speeds, rates and averaging ratios with a fixed numerator.
• Weighted mean, where some values count more than others, as in a grade with coursework and exam components.
• Trimmed mean, the arithmetic mean after removing a percentage from each tail. A compromise between mean and median.

It also gives the midrange, sum, count, minimum, maximum, range, standard deviation, standard error and coefficient of variation, flags any outliers using Tukey fences, and tells you which average best suits the shape of your data.

A worked one-liner: for the five values 4, 7, 9, 10 and 20, the sum is 50 and the count is 5, so the mean is 10. Notice the median is 9, and only one value is above the mean, which is the signature of right skew.

Who uses which: physicists and engineers use the arithmetic mean because measurement error is usually symmetric. Economists report median income because the arithmetic mean is pulled up by a handful of very high earners. Finance uses the geometric mean for compound returns, since averaging percentage growth arithmetically overstates performance. Physics and transport use the harmonic mean for average speed. Education uses the weighted mean for final grades.

4.1 What the mean actually tells you

The mean is the value every observation would take if the total were shared out equally. That is a genuinely useful idea when the data are reasonably symmetric, because it summarises the whole dataset in one number and uses every observation to do it. No information is discarded. This efficiency is why the mean sits at the centre of nearly every statistical test you will meet, from the t-test to regression.

But that same property is its weakness. Because every value contributes, every value can move it, including one that does not belong. If one person in a room of ten earns a million, the mean income describes nobody in that room. The mean is a statement about the total, not about the typical person, and confusing the two is the most consequential mistake in applied statistics.

4.2 The mean versus median gap, your skew detector

Comparing the mean with the median is the fastest diagnostic you have, and it takes two seconds. If they are close, the data are roughly symmetric and the mean is a fair summary. If the mean is clearly higher than the median, the data are right skewed, meaning a tail of large values is pulling the mean upward. If the mean is lower, the data are left skewed.

As a working rule, a gap of more than about 10% of the median is worth investigating, and a gap of more than 20% usually means you should lead with the median. This is exactly why national statistics agencies report median household income rather than mean, and why hospital length of stay, insurance claims and response times are almost always reported as medians. The calculator computes this gap for you and states which average it recommends.

4.3 What the mode adds, and when it fails

The mode is the most frequently occurring value, and it is the only average that works on categorical data: you can find the modal eye colour but not the mean eye colour. On continuous measurements it is often unhelpful, because with enough decimal places no two values repeat and every observation is its own mode.

Where the mode earns its place is in spotting a distribution with two peaks. If the calculator reports two or more modes far apart, that is a strong hint your data are actually two different populations mixed together, and no single average will describe them honestly. Splitting the data is usually the right response.

4.4 When the arithmetic mean is simply the wrong tool

This is the part most calculators skip entirely. For some kinds of data the arithmetic mean does not merely mislead, it gives a numerically wrong answer.

Growth rates and returns need the geometric mean. If an investment gains 50% in year one and loses 50% in year two, the arithmetic mean says 0% and you broke even. You did not: 100 became 150, then 75. The geometric mean correctly reports a loss of about 13.4% per year. Any time percentages compound, the arithmetic mean overstates performance.

Speeds and rates need the harmonic mean. Drive 60 km/h out and 40 km/h back over the same distance and your average speed is not 50. You spend longer at the slower speed, so the true average is the harmonic mean, 48 km/h. The same applies to averaging price-to-earnings ratios and any rate with a fixed numerator.

The calculator computes both alongside the arithmetic mean so you can see the difference immediately. When they diverge noticeably, that divergence is telling you something about the structure of your data.

4.5 The relationship between the three means

For any set of positive numbers, the harmonic mean is never larger than the geometric mean, which is never larger than the arithmetic mean. They are equal only when every value is identical. The more spread out the data, the wider the gaps between them.

That makes the spacing itself informative. If the three means are almost identical, your data are tightly clustered. If the arithmetic mean sits well above the geometric mean, the data have a long right tail. This is the same information the mean-median gap gives you, arrived at from a different direction, and the two should agree.

4.6 The trimmed mean as a middle path

A trimmed mean removes a fixed percentage of the data from each end and averages what remains. A 10% trimmed mean discards the lowest and highest tenth. It keeps most of the efficiency of the arithmetic mean while resisting extreme values, which makes it a sensible compromise when you suspect a few bad observations but do not want to fall back to the median.

Trimming is common in sports judging, where the highest and lowest scores are dropped, and in economics for constructing inflation measures. If you use one, always state the trimming percentage, because a 5% and a 25% trimmed mean can differ substantially and the reader cannot tell which you used.

4.7 Weighted means and where they go wrong

A weighted mean lets some observations count more than others. A grade might be 30% coursework and 70% exam, or a national average might weight each region by its population. The formula multiplies each value by its weight, sums those, and divides by the total weight.

Two mistakes are common. The first is forgetting to divide by the sum of the weights, which gives a number that is not an average at all. The second is applying weights that do not reflect the question being asked, which quietly produces a defensible-looking answer to the wrong question. If you weight by population you get the average person; if you weight by region you get the average region. Those are different numbers and both are correct answers to different questions.

4.8 Why the mean needs a measure of spread beside it

A mean on its own is close to meaningless. Two datasets can share a mean of 50 while one runs from 49 to 51 and the other from 0 to 100. Reporting the mean without the standard deviation, or without n, hides that difference entirely.

The convention is mean plus or minus standard deviation when describing the data, and mean with a standard error or confidence interval when describing how precisely the mean itself has been estimated. These answer different questions and are not interchangeable. The calculator gives both, along with the coefficient of variation for comparing variability across different scales.

4.9 What the mean cannot tell you

The mean says nothing about distribution shape, so symmetric and skewed data can share one. It says nothing about sample size, so a mean from five observations looks identical in form to one from five thousand and deserves far less trust. It hides bimodality completely. And it is purely descriptive, offering no test, no interval and no statement about whether two means genuinely differ.

It also cannot rescue bad data. A precisely calculated mean of a biased sample is precisely wrong, and no amount of decimal places fixes that.

4.10 Practical judgement

Compute the mean, then immediately check it against the median. If they agree, report the mean with its standard deviation and n and move on. If they disagree, ask why: an outlier, a genuine skew, or two populations mixed together each call for a different response. And before you compute anything, ask whether your data are additive at all, because growth rates and speeds need a different mean entirely and no amount of care with the arithmetic version will fix that choice.

▶ Run the analysis above to auto-fill all five examples with your results.

1. Enter your data. The default tab takes comma-separated numbers, exactly as the placeholder shows: 52, 48, 55, 61, 47, .... Newlines, tabs, semicolons and spaces all work, so a column pasted straight from Excel is fine.
2. Name each cluster. The group name field above every textarea is editable. Type something meaningful such as Site A, north slope and it flows into the results table, all four charts and the exported report.
3. Add or remove clusters. Press Add cluster to compare group means side by side. Every column has its own Clear and Remove button, and Remove is disabled on the last remaining column.
4. Or upload a file. On the Upload tab, choose a CSV or Excel file and click the column names you want. Every column you click becomes its own cluster with its own set of averages.
5. Try a sample dataset. Ten built-in datasets cover the cases that matter: symmetric data where the mean works, skewed income where it does not, growth rates needing the geometric mean, speeds needing the harmonic mean, and a bimodal set where no single average is honest.
6. Switch on weights if you need them. Set the weighted mean dropdown to On and enter one weight per value. Weights do not need to sum to 1, since the calculator divides by their total.
7. Set the trimming percentage for the trimmed mean, and choose sample or population standard deviation to match whatever you are checking against.
8. Press Calculate Mean. Nothing is computed until you do, and changing any input or setting clears the results so you never read stale numbers.
9. Read the recommendation panel first. The calculator compares the mean with the median, checks for outliers and skew, and tells you plainly which average suits your data before you commit to reporting one.
10. Check the four charts and export. Chart 1 shows all three centres on your actual data, which is usually the figure you want. Section 5 auto-fills five reporting styles, and the buttons export a text report or a print-ready PDF.

Excel has a direct function for the arithmetic mean, which makes this the easiest calculation in the whole descriptive-statistics toolkit. What trips people up is everything around it: averaging only the rows that meet a condition, handling blanks and text, weighted averages, and the fact that Excel has no built-in geometric or harmonic function on a range of growth rates without a little care. This section covers all of it with the spreadsheet shown at each stage.

8.1 The functions you need

8.2 Step by step

8.3 Weighted, geometric and harmonic means in Excel

The weighted mean has no dedicated function, so it is built from SUMPRODUCT.

With grades of 90, 80 and 70 weighted 50%, 30% and 20%, that returns 83. Dividing by SUM(B2:B17) is essential; leaving it out is the classic mistake and gives a number that is not an average at all.

For growth rates, remember that GEOMEAN needs growth factors, not percentages. A 5% gain is 1.05, not 5 or 0.05.

Store returns as 1.05, 0.90, 1.12 and so on, take GEOMEAN, then subtract 1 to get back to a percentage. Feeding raw percentages into GEOMEAN produces a meaningless answer, and feeding it a negative return makes it fail outright.

8.4 Why your Excel answer might differ from R or Python

8.5 Charting the mean in Excel

1. Select your group names and their means.
2. Insert → Charts → Clustered Column.
3. To add error bars, click the chart, press the + button, tick Error Bars, then More Options → Custom → Specify Value and point at your standard deviation or standard error cells.
4. State in the caption which one the bars represent. An unlabelled error bar is grounds for revision at most journals.
5. To draw the mean as a line across a scatter of raw data, add a second series holding the mean repeated for every row, then change its chart type to Line via Change Chart Type → Combo.

Chart 1 in section 3 of this page draws the raw data with the mean, median and mode all marked automatically, which is usually quicker than rebuilding it in Excel.

8.6 Excel errors you will probably hit

R gives you the mean in one word, mean(x), but a bare mean is rarely what you actually want to report. The code below is a complete, copy-and-run script that computes the mean, median, mode, standard deviation and standard error, prints a plain-English verdict on whether the mean is safe to use, and saves one publication-ready figure at 300 dpi. It uses base R only, so there is nothing to install.

9.1 The complete script

R, base only, saves mean_plot.png

# ============================================================
# Mean, median and mode with a single publication-ready plot
# StatsUnlock - https://statsunlock.com/mean-calculator/
# Base R only. No packages to install.
# ============================================================

# ---- 1. Your data ------------------------------------------
# Replace with your own values, or read a CSV:
#   x <- read.csv("data.csv")$column_name
#   x <- x[!is.na(x)]
x <- c(52, 48, 55, 61, 47, 50, 58, 53,
       49, 56, 51, 54, 46, 57, 50, 52)

label <- "Site A - north slope"
units <- "mm"

# ---- 2. The averages ---------------------------------------
mean_x   <- mean(x)
median_x <- median(x)

# Base R has no mode() for statistics, so define one.
# It returns every tied value, and nothing if no value repeats.
get_mode <- function(v) {
  tab <- table(v)
  if (max(tab) < 2) return(numeric(0))
  as.numeric(names(tab)[tab == max(tab)])
}
modes_x <- get_mode(x)

sd_x <- sd(x)                      # sd() always uses n - 1
se_x <- sd_x / sqrt(length(x))
gap  <- abs(mean_x - median_x) / median_x * 100

cat(sprintf("n       = %d\n", length(x)))
cat(sprintf("mean    = %.4f\n", mean_x))
cat(sprintf("median  = %.4f\n", median_x))
cat(sprintf("mode    = %s\n",
    if (length(modes_x)) paste(modes_x, collapse = ", ") else "none (no value repeats)"))
cat(sprintf("SD      = %.4f   SE = %.4f\n", sd_x, se_x))
cat(sprintf("mean vs median gap = %.1f%% -> %s\n", gap,
    if (gap < 10) "symmetric, the mean is fine" else "skewed, prefer the median"))

# ---- 3. One figure, two stacked panels ---------------------
png("mean_plot.png", width = 2700, height = 1950, res = 300)

layout(matrix(c(1, 2), nrow = 2), heights = c(3, 1))
par(mar = c(0.6, 5, 5, 2), family = "sans")

bins  <- max(5, ceiling(sqrt(length(x))))
brks  <- pretty(range(x), n = bins)
h     <- hist(x, breaks = brks, plot = FALSE)

plot(h, col = "#bbf7d0", border = "#16a34a", lwd = 1.6,
     main = "", xlab = "", ylab = "Frequency",
     xaxt = "n", cex.lab = 1.05, las = 1)

# mean +/- 1 SD shaded band, drawn first so it sits behind
rect(mean_x - sd_x, 0, mean_x + sd_x, max(h$counts) * 1.15,
     col = adjustcolor("#c2410c", alpha.f = 0.07), border = NA)

abline(v = mean_x,   col = "#c2410c", lwd = 2.6, lty = 2)
abline(v = median_x, col = "#9333ea", lwd = 2.6, lty = 4)
if (length(modes_x)) abline(v = modes_x, col = "#0891b2", lwd = 1.9, lty = 3)

title(main = sprintf("%s: distribution with mean, median and mode", label),
      adj = 0, cex.main = 1.15, font.main = 2, line = 3)
title(main = sprintf("n = %d   mean = %.2f %s   median = %.2f %s   SD = %.2f",
                     length(x), mean_x, units, median_x, units, sd_x),
      adj = 0, cex.main = 0.95, font.main = 1, line = 1.6)

legend("topright", bty = "n", cex = 0.92,
       lwd = c(2.6, 2.6, 1.9, 8),
       lty = c(2, 4, 3, 1),
       col = c("#c2410c", "#9333ea", "#0891b2",
               adjustcolor("#c2410c", alpha.f = 0.12)),
       legend = c(sprintf("Mean = %.2f",   mean_x),
                  sprintf("Median = %.2f", median_x),
                  if (length(modes_x))
                    sprintf("Mode = %s", paste(modes_x, collapse = ", "))
                  else "Mode = none",
                  sprintf("Mean +/- 1 SD (%.2f)", sd_x)))

# lower panel: every raw value, jittered so ties are visible
par(mar = c(4.6, 5, 0.6, 2))
set.seed(42)                       # seed = reproducible jitter
plot(x, runif(length(x), -0.32, 0.32),
     ylim = c(-1, 1), yaxt = "n", bty = "n",
     xlab = sprintf("Value (%s)", units), ylab = "",
     pch = 21, cex = 1.5, bg = adjustcolor("#16a34a", alpha.f = 0.75),
     col = "white", lwd = 1.1, cex.lab = 1.05,
     xlim = range(brks))
mtext("Raw\ndata", side = 2, line = 1.6, las = 1, cex = 0.8)
abline(v = mean_x,   col = "#c2410c", lwd = 2.2, lty = 2)
abline(v = median_x, col = "#9333ea", lwd = 2.2, lty = 4)

dev.off()
cat("\nSaved mean_plot.png\n")

Run it with Rscript mean_plot.R from a terminal, or paste it straight into the RStudio console. The figure appears as mean_plot.png in your working directory, which you can check with getwd().

9.2 What each part does

9.3 What the figure shows

The upper panel is a histogram with three vertical reference lines: a dashed orange line at the mean, a dot-dash purple line at the median, and dotted cyan lines at every mode. The shaded band spans one standard deviation either side of the mean. The lower panel plots every individual observation with a little vertical jitter, so you can see the actual data rather than just the binned summary, and the same mean and median lines carry through so the two panels line up.

Reading it takes two seconds. If the orange and purple lines sit almost on top of each other, the data are symmetric and the mean is a fair summary. If they separate, the data are skewed and the mean is being pulled away from the typical value. If the lower panel shows two distinct clumps with a gap between them, you have two populations mixed together and no single average is honest, a case the mean-median comparison alone cannot detect.

9.4 Common R problems and fixes

9.5 Useful one-liners

Python's np.mean(x) is as short as R's, and carries the same risk of being reported alone when it should not be. The script below is complete and runnable: it computes every statistic you need, prints a verdict on whether the mean is trustworthy for your data, and saves one 300 dpi figure ready to drop into a paper or a slide deck.

Install with pip install numpy matplotlib if you do not already have them.

10.1 The complete script

Python 3, saves mean_plot.png

# ============================================================
# Mean, median and mode with a single publication-ready plot
# StatsUnlock - https://statsunlock.com/mean-calculator/
# ============================================================

import numpy as np
import matplotlib.pyplot as plt
from collections import Counter

# ---- 1. Your data ------------------------------------------
# Replace this list with your own values, or read a CSV:
#   import pandas as pd
#   x = pd.read_csv("data.csv")["column_name"].dropna().to_numpy()
x = np.array([52, 48, 55, 61, 47, 50, 58, 53,
              49, 56, 51, 54, 46, 57, 50, 52], dtype=float)

label = "Site A - north slope"
units = "mm"

# ---- 2. The averages ---------------------------------------
mean_x   = np.mean(x)
median_x = np.median(x)

counts   = Counter(x)
top      = max(counts.values())
modes    = sorted(float(v) for v, c in counts.items() if c == top) if top > 1 else []

sd_x = np.std(x, ddof=1)          # ddof=1 -> sample SD, matches Excel STDEV.S
se_x = sd_x / np.sqrt(x.size)
gap  = abs(mean_x - median_x) / median_x * 100

print(f"n       = {x.size}")
print(f"mean    = {mean_x:.4f}")
print(f"median  = {median_x:.4f}")
mode_txt = ", ".join(f"{m:g}" for m in modes) if modes else "none (no value repeats)"
print(f"mode    = {mode_txt}")
print(f"SD      = {sd_x:.4f}   SE = {se_x:.4f}")
print(f"mean vs median gap = {gap:.1f}%  ->",
      "symmetric, the mean is fine" if gap < 10 else "skewed, prefer the median")

# ---- 3. One figure, two stacked panels ---------------------
fig, (ax_hist, ax_dot) = plt.subplots(
    2, 1, figsize=(9, 6.5), sharex=True,
    gridspec_kw={"height_ratios": [3, 1], "hspace": 0.08})

bins = max(5, int(np.ceil(np.sqrt(x.size))))
ax_hist.hist(x, bins=bins, color="#bbf7d0", edgecolor="#16a34a",
             linewidth=1.4, zorder=2)

ax_hist.axvline(mean_x,   color="#c2410c", lw=2.4, ls="--", zorder=4,
                label=f"Mean = {mean_x:.2f}")
ax_hist.axvline(median_x, color="#9333ea", lw=2.4, ls="-.", zorder=4,
                label=f"Median = {median_x:.2f}")
for i, m in enumerate(modes):
    ax_hist.axvline(m, color="#0891b2", lw=1.8, ls=":", zorder=3,
                    label=f"Mode = {mode_txt}" if i == 0 else None)

ax_hist.axvspan(mean_x - sd_x, mean_x + sd_x, color="#c2410c",
                alpha=0.07, zorder=1, label=f"Mean ± 1 SD ({sd_x:.2f})")

ax_hist.set_ylabel("Frequency", fontsize=11)
ax_hist.set_title(f"{label}: distribution with mean, median and mode\n"
                  f"n = {x.size}   mean = {mean_x:.2f} {units}   "
                  f"median = {median_x:.2f} {units}   SD = {sd_x:.2f}",
                  fontsize=12, fontweight="bold", loc="left")
ax_hist.legend(frameon=False, fontsize=9.5)
ax_hist.grid(axis="y", alpha=0.25, zorder=0)
ax_hist.spines[["top", "right"]].set_visible(False)

rng = np.random.default_rng(42)                    # seed = reproducible jitter
ax_dot.scatter(x, rng.uniform(-0.32, 0.32, x.size),
               s=52, color="#16a34a", alpha=0.75,
               edgecolor="white", linewidth=0.9, zorder=3)
ax_dot.axvline(mean_x,   color="#c2410c", lw=2.0, ls="--", zorder=2)
ax_dot.axvline(median_x, color="#9333ea", lw=2.0, ls="-.", zorder=2)
ax_dot.set_ylim(-1, 1)
ax_dot.set_yticks([])
ax_dot.set_xlabel(f"Value ({units})", fontsize=11)
ax_dot.set_ylabel("Raw\ndata", fontsize=9, rotation=0,
                  ha="right", va="center")
ax_dot.grid(axis="x", alpha=0.25, zorder=0)
ax_dot.spines[["top", "right", "left"]].set_visible(False)

fig.savefig("mean_plot.png", dpi=300, bbox_inches="tight")
print("\nSaved mean_plot.png")

Run it with python mean_plot.py. In Jupyter, delete the savefig line and add plt.show() instead, or keep both to display and save at once.

10.2 What each part does

10.3 What the figure shows

Exactly the same figure as the R script produces, drawn by a different engine. The upper panel is the histogram with the mean, median and every mode marked, plus a shaded one-standard-deviation band. The lower panel shows every raw observation with jitter, sharing the same x-axis so the reference lines carry straight through.

The point of showing both panels is that the histogram tells you the shape while the strip tells you the truth. Binning can conceal ties, gaps and small clusters that are obvious in the raw values. When a distribution turns out to be bimodal, it is almost always the lower panel that reveals it, not the summary statistics.

10.4 Common Python problems and fixes

10.5 Useful one-liners

11.1 Which average should you use?

Conclusion: the choice is driven by the shape of your data and by what the numbers represent, never by which is easiest to compute.

11.2 How one outlier moves the mean but not the median

Conclusion: changing one value from 12 to 60 nearly doubled the mean and left the median untouched. That is the whole argument for the median in one line.

11.3 Growth rates: why the arithmetic mean is wrong

Conclusion: compounding the geometric mean over four years reproduces the true 24.74% growth exactly, while the arithmetic mean would predict 27.4%. The arithmetic mean always overstates compound growth unless every return is identical.

11.4 Speeds: why the harmonic mean is right

Conclusion: you spend more time travelling at the slower speed, so it deserves more weight. The arithmetic mean always overstates average speed over equal distances.

11.5 The three means and how spread pulls them apart

Conclusion: harmonic never exceeds geometric, which never exceeds arithmetic, and they are equal only when every value is the same. The wider the gap between them, the more spread out your data are.

11.6 Weighted mean worked two ways

Grades of 90, 80 and 70 with coursework, midterm and participation weightings.

Conclusion: weights of 0.5, 0.3, 0.2 and weights of 5, 3, 2 give exactly the same answer, because the formula divides by the total weight. The unweighted mean of those three scores would be 80, so the weighting is worth three marks here.

Study design: the arithmetic mean assumes a set of independent measurements of one quantity on an interval or ratio scale, where the values are meaningfully additive.

1. Check the values are additive before you collect anything. Lengths, masses, counts, times and money add up sensibly. Growth rates and speeds do not, and averaging them arithmetically gives a numerically wrong answer. Decide now whether you need the geometric or harmonic mean instead.
2. Define the population and the unit of observation precisely. Be explicit about whether a measurement is one specimen, one repeat reading of the same specimen, or an average of several. Mixing these inflates n and makes the mean look more precise than it is.
3. Randomise the selection. The mean of a convenience sample is a precise summary of the wrong group. No amount of care with the arithmetic fixes a biased sampling frame.
4. Fix the instrument, units and resolution and record them in the protocol. Changing instruments mid-study shifts the mean for procedural reasons that have nothing to do with the subject.
5. Decide how zeros will be recorded. A genuine zero and a missing measurement are different things, and if both end up as 0 in the spreadsheet the mean will be pulled down. Use a blank or a code for missing data, never a zero.
6. Write the outlier policy in advance. Because the mean is not robust, one bad value can dominate it. State how outliers will be identified and whether they will be excluded, before you see the data.
7. Plan the sample size. Aim for at least 15 to 20 per group for a stable mean, and 30 or more if the data are skewed. Below about 10 the mean moves noticeably when a single value changes.
8. Record whether weights apply. If some observations represent more units than others, capture the weight as its own column at collection time. Reconstructing weights afterwards is guesswork.
9. Record raw data in one column per group, exactly as the layout below shows, ready to paste or upload here.

One column per group, one row per unit, numbers only below the header row. Blank cells are ignored, so unequal group sizes are fine.

Pre-registration: record which average you will report, the outlier rule, and whether the standard deviation will be the sample or population version, before looking at the data. Switching from the mean to the median after seeing that the mean was inconvenient is a real and common form of unintentional bias.

Common collection mistakes:

• Recording missing values as zero, which drags the mean down without warning.
• Averaging percentages or rates arithmetically when they should compound.
• Treating repeat readings of one specimen as independent observations.
• Rounding aggressively at collection, which shifts the mean and manufactures a false mode.
• Pooling two genuinely different populations into one column, giving a mean that describes neither.

The mean is the default average for good reasons, but it has a narrower valid range than most people assume.

Use the arithmetic mean when:

• ✓ The data are roughly symmetric with no extreme values.
• ✓ The values are genuinely additive, so their total means something.
• ✓ The scale is interval or ratio, not ordinal or categorical.
• ✓ You intend to run a t-test, ANOVA or regression, all of which are built on means.
• ✓ You need every observation to contribute to the summary.

Use a different average when:

• ✗ The data are skewed or contain outliers, where the median is more honest.
• ✗ The values are growth rates, returns or ratios, where you need the geometric mean.
• ✗ The values are speeds or rates over equal distances, where you need the harmonic mean.
• ✗ The data are categorical, where only the mode applies.
• ✗ The data are ordinal, such as Likert ratings, where the median is safer.
• ✗ The distribution has two peaks, where no single average is honest and you should split the data.

Real-world examples:

1. Laboratory measurement, averaging replicate readings where error is symmetric. The mean is ideal.
2. Ecology, mean body mass or tree diameter per plot, usually mildly right skewed but still workable.
3. Economics, household income, where the median is reported precisely because the mean misleads.
4. Finance, annual returns, which need the geometric mean because they compound.
5. Transport, average speed over a route, which needs the harmonic mean.
6. Education, final grades from weighted components, which need the weighted mean.
7. Clinical research, hospital length of stay, always reported as a median because of the long right tail.

Decision rule: check the values are additive, compute the mean, then compare it with the median. If they agree, report the mean with its standard deviation and n. If they disagree, find out why before choosing.

Assumptions

1. The data are additive. Check what the values represent. Growth rates and speeds are not additive, and their arithmetic mean is simply wrong rather than merely imprecise.
2. The scale is interval or ratio. Check the measurement scale. Averaging ordinal codes such as Likert responses produces a computable but questionable number.
3. Observations are independent. Check the design. Repeated readings of the same unit are not independent and make the mean look more precise than it is.
4. The distribution is roughly symmetric. Check the mean-median gap and chart 1. Under skew the mean stops representing a typical value.
5. No dominant outliers. Check the outlier flags. The mean has a breakdown point of zero, meaning a single arbitrary value can move it anywhere.
6. The sample is large enough. Check n. Below about 10 observations the mean shifts noticeably when any single value changes.
7. The data come from one population. Check chart 3 for two peaks. A bimodal dataset produces a mean that describes neither group.

Limitations

• The mean is not robust. Its breakdown point is zero, the worst possible: one bad value out of a million can move it arbitrarily far. Use the median or a trimmed mean when that is a concern.
• It reveals nothing about distribution shape, and cannot detect bimodality even when the mean-median check passes.
• It carries no measure of uncertainty by itself. Report the standard deviation to describe the data and the standard error or a confidence interval to describe the estimate.
• It is purely descriptive. Two different means are not thereby significantly different; that requires a t-test, ANOVA or an equivalent.
• The geometric mean requires strictly positive values and the harmonic mean requires positive non-zero values. Both are reported as undefined otherwise rather than silently returning a wrong number.
• The weighted mean is only as good as the weights. Weights that do not reflect the question produce a defensible-looking answer to the wrong question.
• This tool assumes a simple unweighted sample unless you supply weights. Stratified, clustered and survey designs need design-consistent estimators.
• It does not compute a confidence interval for the median, or run any hypothesis test. Those live in other tools.

A mean calculator does something almost trivially simple, adding up numbers and dividing by how many there are, and yet the mean is the statistic most often reported when it should not be. The arithmetic is never the hard part. The hard part is deciding whether the arithmetic mean is the right average for your data at all, and that decision takes about ten seconds if you know what to look for.

The single most useful habit is to compute the median at the same time and compare the two. When they sit close together the data are roughly symmetric, the mean is a fair summary, and you can report it with confidence. When they pull apart, something is going on: an outlier, a genuine skew, or two populations mixed together. Each of those calls for a different response, but all of them start with noticing the gap, and noticing costs you nothing.

The second habit is to ask whether your numbers are additive before you average them. This is where the mean does not merely mislead but gives an answer that is numerically wrong. Averaging growth rates arithmetically overstates compound returns, as anyone who has gained 50% and then lost 50% and found themselves down 25% can confirm. Averaging speeds arithmetically overstates how fast you actually travelled. Neither error is subtle once you know to check, and neither is detectable by looking at the answer.

Finally, never let a mean travel alone. On its own it tells you nothing about how variable the data are, how many observations produced it, or what shape they took. Mean, standard deviation and n together are the minimum any reader needs to judge what you found, and adding the median costs one extra number while making your reporting substantially more honest.

Paste your numbers into the calculator above, read the recommendation panel before you read the answer, and check chart 1 to see where the mean sits relative to your actual data. Then come back to this mean calculator whenever you meet an unfamiliar dataset, because knowing which average to reach for matters far more than being able to compute any of them.

Cards marked Coming soon are not yet published, so they are shown without a link. Wire them up once each page goes live.

The following peer-reviewed references support the methods used in this mean calculator, covering central tendency, robust alternatives to the mean, the correct use of geometric and harmonic means, and best practice in reporting.

1. Altman, D. G., & Bland, J. M. (2005). Standard deviations and standard errors. BMJ, 331(7521), 903. https://doi.org/10.1136/bmj.331.7521.903
2. Altman, D. G., & Bland, J. M. (1996). Statistics notes: Presentation of numerical data. BMJ, 312(7030), 572. https://doi.org/10.1136/bmj.312.7030.572
3. 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
4. Wilcox, R. R. (2016). Introduction to robust estimation and hypothesis testing (4th ed.). Academic Press. https://doi.org/10.1016/C2010-0-67044-1
5. Huber, P. J., & Ronchetti, E. M. (2009). Robust statistics (2nd ed.). Wiley. https://doi.org/10.1002/9780470434697
6. 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
7. Fleming, P. J., & Wallace, J. J. (1986). How not to lie with statistics: The correct way to summarize benchmark results. Communications of the ACM, 29(3), 218–221. https://doi.org/10.1145/5666.5673
8. Spizman, L., & Weinstein, M. A. (2008). A note on utilizing the geometric mean: When, why and how the forensic economist should employ the geometric mean. Journal of Legal Economics, 15(1), 43–55. https://ssrn.com/abstract=1518803
9. Vogel, R. M. (2022). The geometric mean? Communications in Statistics: Theory and Methods, 51(1), 82–94. https://doi.org/10.1080/03610926.2020.1743313
10. Ferger, W. F. (1931). The nature and use of the harmonic mean. Journal of the American Statistical Association, 26(173), 36–40. https://doi.org/10.1080/01621459.1931.10503148
11. Manikandan, S. (2011). Measures of central tendency: The mean. Journal of Pharmacology and Pharmacotherapeutics, 2(2), 140–142. https://doi.org/10.4103/0976-500X.81920
12. Manikandan, S. (2011). Measures of central tendency: Median and mode. Journal of Pharmacology and Pharmacotherapeutics, 2(3), 214–215. https://doi.org/10.4103/0976-500X.83300
13. Weissgerber, T. L., Milic, N. M., Winham, S. J., & Garovic, V. D. (2015). Beyond bar and line graphs: Time for a new data presentation paradigm. PLOS Biology, 13(4), e1002128. https://doi.org/10.1371/journal.pbio.1002128
14. Cumming, G., Fidler, F., & Vaux, D. L. (2007). Error bars in experimental biology. Journal of Cell Biology, 177(1), 7–11. https://doi.org/10.1083/jcb.200611141
15. Wan, X., Wang, W., Liu, J., & Tong, T. (2014). Estimating the sample mean and standard deviation from the sample size, median, range and/or interquartile range. BMC Medical Research Methodology, 14, 135. https://doi.org/10.1186/1471-2288-14-135
16. Tukey, J. W. (1977). Exploratory data analysis. Addison-Wesley. https://search.worldcat.org/title/3058187
17. Weisstein, E. W. (n.d.). Arithmetic mean. MathWorld, A Wolfram Web Resource. https://mathworld.wolfram.com/ArithmeticMean.html
18. American Psychological Association. (2020). Publication manual of the American Psychological Association (7th ed.). https://doi.org/10.1037/0000165-000
19. NIST/SEMATECH. (2013). e-Handbook of statistical methods. National Institute of Standards and Technology. https://www.itl.nist.gov/div898/handbook/
20. R Core Team. (2024). R: A language and environment for statistical computing. R Foundation for Statistical Computing. https://www.R-project.org/