The mean, in its most common form the arithmetic mean, is the sum of all values divided by the number of values. In statistics, it is the standard measure of central tendency: a single number that represents the typical level of a dataset and serves as the starting point for variance, standard deviation, correlation and most inferential methods.
The mean is the balance point of the data: the deviations of all values from it add up to exactly zero. That property makes it efficient and mathematically convenient, but it also makes the mean highly sensitive to outliers and skew, which is why it is often reported together with the median.
How the arithmetic mean is calculated
The arithmetic mean adds every observation and divides by the count. For a sample it is written x̄ = (x1 + x2 + ... + xn) / n, and for a whole population it is written with the Greek letter μ (mu) and divided by the population size N. The sample mean x̄ is an estimate of the usually unknown population mean μ.
For the values 2, 4, 4, 4, 5, 5, 7 and 9, the sum is 40 and the count is 8, so the mean is exactly 5. The deviations from 5 are -3, -1, -1, -1, 0, 0, 2 and 4, and they sum to zero, which illustrates the balance-point property.
Two further properties explain why the mean dominates statistical theory:
- It minimizes the sum of squared deviations. No other single number gives a smaller total of squared distances to the data, which links the mean directly to variance and least-squares regression.
- It is unbiased and precise. The sample mean equals the population mean on average, and its standard error is σ divided by the square root of n, so quadrupling the sample size halves the uncertainty.
- It uses every value. This makes it efficient for symmetric data, but also means a single extreme value can move it far.
Types of mean and when each applies
The word mean covers several averages, and choosing the wrong one produces systematically wrong answers. The right type depends on whether the quantities add, multiply or represent rates.
| Type | Calculation | Correct use | Example |
|---|---|---|---|
| Arithmetic mean | Sum divided by count | Additive quantities | Average order value |
| Weighted mean | Sum of value times weight, divided by total weight | Values of unequal importance | Grade point average by credits |
| Geometric mean | nth root of the product | Growth rates and ratios | Average annual return |
| Harmonic mean | Count divided by the sum of reciprocals | Rates over equal amounts | Average speed over equal distances |
| Trimmed mean | Arithmetic mean after removing a share of extremes | Data with outliers | Mean response time without the top and bottom 5% |
The differences are not cosmetic. An investment that gains 50% and then loses 50% has an arithmetic mean return of 0%, yet it ends at 0.75 of its starting value; the geometric mean of about -13.4% per period describes what actually happened. A car that drives equal distances at 60 km/h and 40 km/h averages 48 km/h, the harmonic mean, not 50 km/h.
For positive values that are not all equal, the three classical means always follow the same order: harmonic mean < geometric mean < arithmetic mean. This inequality is a quick sanity check when results are compared.
Mean compared with median and mode
The mean, median and mode answer different questions about the center of the data. The median is the middle value of the sorted data, and the mode is the most frequent value; for a perfectly symmetric distribution all three coincide.
| Property | Mean | Median | Mode |
|---|---|---|---|
| Uses every value | Yes | Only the order | Only frequencies |
| Effect of one extreme outlier | Can shift strongly | Almost none | None |
| Breakdown point | 0% | 50% | Rarely analyzed this way |
| Works for categories | No | Only for ordered categories | Yes |
| Best for | Symmetric data, totals, modeling | Skewed data, typical case | Most common category |
The breakdown point is the share of data that can be corrupted before an estimate becomes arbitrarily wrong. The mean has a breakdown point of zero, because one sufficiently extreme value can move it anywhere, while the median tolerates corruption of almost half the data.
In a right-skewed distribution such as household income or house prices, a long tail of high values pulls the mean above the median. Reporting only the mean in such cases overstates what a typical household earns.
Choosing the right average for the data
The choice of average should follow the shape of the data and the question being asked, not habit. The decision table below covers the most frequent situations in analysis and reporting.
| Situation | Recommended measure | Reason |
|---|---|---|
| Roughly symmetric measurements | Arithmetic mean | Most precise and supports further inference |
| Strongly skewed values such as income | Median, optionally with mean | Reflects the typical case |
| A total must be recovered | Arithmetic mean | Mean times count equals the total |
| Compound growth over several periods | Geometric mean | Growth multiplies rather than adds |
| Ratios such as price per unit over equal spend | Harmonic mean | Averages rates correctly |
| Latency or performance data with spikes | Median and high percentiles | Tail behavior matters more than the center |
A useful rule is that the mean is right when totals matter. A retailer planning stock needs the mean daily demand, because mean multiplied by the number of days gives the expected total, whereas the median would underestimate it for skewed demand.
Common mistakes when using the mean
Errors with the mean are rarely arithmetic; they come from applying it where its assumptions do not hold. The following mistakes appear regularly in reports and dashboards:
- Averaging averages. The mean of two group means equals the overall mean only when the groups are the same size; otherwise a weighted mean by group size is required.
- Reporting a mean without spread. A mean of 50 says little without the standard deviation or a confidence interval.
- Averaging percentages or rates directly, which ignores the different denominators behind each figure.
- Ignoring heavy tails. For some distributions, such as the Cauchy distribution, the mean is not even defined, and sample means never settle down.
- Treating a mean as a real individual. A mean of 2.4 children per family describes the group, not any actual family.
In simple terms
The mean is what most people call the average. Add up all the numbers and divide by how many numbers there are. If three friends have 2, 4 and 6 apples, they have 12 apples together, and an even share is 4 apples each.
Picture the numbers as weights placed along a ruler. The mean is the point where the ruler balances. That is also its weakness: one very heavy weight far out at the end tips the balance point toward it, which is why a few extreme values can make the mean misleading.
A practical example
A web team measures the load time of a checkout page for ten visits: 0.8, 0.9, 0.9, 1.0, 1.0, 1.1, 1.1, 1.2, 1.3 and 9.7 seconds. The values sum to 19.0 seconds, so the mean load time is 1.9 seconds, which suggests the page is slow for everyone.
The median is only 1.05 seconds, and without the single 9.7-second visit the mean of the remaining nine visits drops to about 1.03 seconds. One user on a poor mobile connection nearly doubled the mean.
The team therefore reports the median together with the 95th percentile for user experience, and keeps the mean only for capacity planning, where the total server time across all visits is what matters. Choosing the measure by purpose turned a misleading single number into an accurate picture.