When to Use Mean, Median, or Mode

Quick Decision Table

Use this table to pick the right measure in under 30 seconds. When in doubt, report two — mean and median together give a much richer picture than either alone.

When to Use the Mean

The arithmetic mean is the right choice when data is approximately symmetric and does not contain extreme outliers. For symmetric distributions (like test scores, manufacturing measurements, or heights), the mean, median, and mode are all similar — and the mean is preferred because it uses all data points and enables downstream statistical tests (t-tests, ANOVA, confidence intervals).

Use the mean when: (1) you plan to run parametric statistical tests that assume normality; (2) you need to report total quantities (total sales ÷ number of days = mean daily sales); (3) data is ratio or interval scale without skew.

Example where mean is correct: A teacher records exam scores: {72, 75, 78, 80, 82, 85, 88}. The distribution is roughly symmetric. Mean = 80, Median = 80. Both agree — use the mean so you can later run a t-test to compare the class to a standard.

Example where mean fails: A company reports "average employee salary = $95,000." In reality: 9 employees earn $50,000 and the CEO earns $590,000. Mean = ($50,000×9 + $590,000) / 10 = $104,000 — not representative of any actual employee. The median ($50,000) tells the true story.

When to Use the Median

The median is the right choice for skewed data or data with outliers. Because the median is the middle value of a sorted list, it is not affected by extreme values — a single outlier worth $1 million has the same effect on the median as a value of $1.

Use the median when: (1) data is skewed (income, real estate, web response times, medical costs); (2) the dataset contains known or suspected outliers; (3) data is ordinal (ranked categories like "satisfied/neutral/dissatisfied"); (4) reporting to a non-technical audience who will interpret "average" as "typical."

The rule of thumb: when mean > median, the distribution is right-skewed (pulled up by high values). When mean < median, the distribution is left-skewed. The bigger the gap, the more important it is to report the median rather than the mean.

Example: US median household income in 2024 was approximately $80,000, while the mean was approximately $105,000. The $25,000 gap exists because a small number of very high-income households drag the mean upward. The median correctly captures what a "typical" American household actually earns.

When to Use the Mode

The mode is the only appropriate measure of central tendency for categorical (nominal) data. You cannot compute the mean of {Red, Blue, Blue, Green, Red, Red} — but the mode is Red (3 occurrences).

Use the mode when: (1) data is categorical (colours, product categories, countries, job titles); (2) you need to know the most popular choice or product; (3) data is discrete and you want to know the most common count.

The mode is also valuable for continuous data when the distribution is multimodal (has two or more peaks). A clothing retailer tracking shoe sizes might find mean=9.3 and median=9 — but these are useless for inventory because customers want specific integer sizes. The mode reveals that sizes 8 and 11 are most popular (a bimodal distribution), which is actionable for stock planning.

Mode limitations: (1) a dataset can have no mode (all values appear once), one mode (unimodal), or multiple modes (bimodal, multimodal); (2) it ignores most of the data; (3) it is sensitive to how data is grouped — two slightly different values that are nearly identical may appear as separate modes.

Common Mistakes

Mistake 1: Reporting only the mean for skewed data. Income, house prices, response times, and medical costs are almost always right-skewed. Reporting only the mean makes "average" sound higher than what most people experience. Always check if mean >> median before choosing which to report.

Mistake 2: Using arithmetic mean for rates and ratios. If you drive 60 km/h for one hour and 20 km/h for one hour, your average speed is not (60+20)/2 = 40 km/h. It is 80 km / 2 hours = 40 km/h in this case — but for equal distances, the harmonic mean applies. For investment returns and growth rates, use the geometric mean. The arithmetic mean consistently overstates the average for multiplicative quantities.

Mistake 3: Computing the mean of ordinal data without care. A 5-point Likert scale (Strongly Disagree to Strongly Agree) is ordinal — the intervals between points are not guaranteed to be equal. Computing a mean (e.g., "average rating = 3.7") is common in practice and often acceptable, but strictly the median or mode is more defensible. When in doubt, report both the median and the mean.

Mistake 4: Ignoring the distribution shape entirely. The single best habit: plot your data (histogram or box plot) before computing any summary statistic. A distribution with two peaks, a long tail, or gaps changes which measure is meaningful. No formula replaces looking at the data first.

Best Practice: Report Both Mean and Median

For any dataset where the mean and median differ significantly (more than 10% relative), best practice is to report both. The gap itself is informative — it tells you the data is skewed and by how much.

A simple reporting template: "The median [variable] was X (mean: Y, SD: Z, n = N)." This gives readers everything they need: the typical value (median), the average for calculations (mean), the spread (SD), and the sample size.

When presenting to decision-makers who need a single number: use the median for describing "typical" and save the mean for total/aggregate calculations. For example: "Our typical customer takes 4.2 days to pay (median). Across 1,200 invoices last month, we processed an average of 8.7 days outstanding (mean) — indicating a small number of very late accounts are inflating the average."