Mean, Median, Mode: Choosing the Right Average for Decisions

Three Averages for Three Questions

"Average" is a sloppy word. It hides three very different statistics that answer three very different questions. The mean answers "what is the total divided by the count?" The median answers "what value sits in the middle when everything is sorted?" The mode answers "what value appears most often?" All three are legitimate; the mistake is reaching for the wrong one when the data has a particular shape.

The mean is the only one of the three that uses every data point in its calculation. That makes it efficient — it captures all available information — but also fragile, because a single extreme value can drag it anywhere. The median ignores how far values are from the center and only cares about rank, which makes it robust. The mode ignores both magnitude and rank and only counts frequency, which makes it the right tool for categorical data where arithmetic on the values themselves would be meaningless.

The Mean: Definition and Achilles' Heel

For a dataset x1, x2, ..., xn, the arithmetic mean is μ = (Σ xi) / n. The mean is the value that minimizes the sum of squared deviations — a property that falls out of calculus and is the reason least-squares regression works. It is also the average that pops out of every introductory physics lab, every school gradebook, and every sports statistic.

The mean's weakness is sensitivity to outliers. Consider nine employees earning \$50,000 each and one CEO earning \$1,000,000. The mean salary is (9 × 50,000 + 1,000,000) / 10 = \$145,000. That number accurately describes the total payroll divided by headcount, but it describes nobody in the company. A new hire reading the figure would form a wildly misleading impression of typical compensation. The median, by contrast, is \$50,000 — the salary of the person in the middle when the list is sorted — and that is the number that actually represents what most employees earn.

Whenever a distribution has a long tail — incomes, house prices, city populations, website visit counts, insurance claim sizes — the mean gets pulled toward the tail and the median becomes the more honest summary. This is why the U.S. Census Bureau reports median household income rather than mean household income, and why the Bureau of Labor Statistics reports median weekly earnings in its usual weekly earnings series.

The Median: Robust by Construction

The median is the middle value when the data is sorted. For an odd count, it is the value at position (n + 1) / 2. For an even count, it is the mean of the two middle values at positions n/2 and n/2 + 1. Either way, the median splits the dataset into two equal halves: 50 percent of observations lie above it, 50 percent below.

The median's defining strength is that it is robust to outliers. In the salary example above, replacing the CEO with someone earning \$50,000,001 changes the mean by roughly \$950,000 but leaves the median exactly where it was. This is the same property that makes the interquartile range (the range between the 25th and 75th percentiles) a better measure of spread than the full range, and why box plots use median and quartiles rather than mean and standard deviation when distributions are skewed.

The median is the right choice for housing prices, salary surveys, survival times, and any quantity whose distribution is bounded at zero on one side but unbounded on the other. It is also the right choice for ordinal data — survey responses on a 1-to-5 Likert scale, for example — where the gap between adjacent values is not necessarily equal and arithmetic on the values themselves is questionable.

The Mode: Champion of Categorical Data

The mode is the most frequent value in the dataset. A dataset can have one mode (unimodal), two modes (bimodal), several modes (multimodal), or no mode at all if every value appears exactly once. The mode is the only average that makes sense for nominal data — categories with no inherent ordering.

Imagine a shoe store tracking daily sales. The mean shoe size sold might be 9.4, but nobody wears a size 9.4 shoe. The median might be 9.5, but the store cannot stock 9.5 units of inventory based on a fractional size. The mode — say, size 10 — tells the store manager exactly which size to overstock. The same logic applies to survey responses on favorite color, most-used programming language, or most-frequent customer complaint.

The mode is also useful as a sanity check on continuous data. A bimodal distribution of heights in a mixed-gender sample — one peak around 162 cm, another around 178 cm — tells you that you are looking at two distinct populations lumped together, and that averaging them into a single number would hide the most important feature of the data.

A Worked Example: Salaries in a Small Tech Company

Consider ten engineers at a small startup with the following annual salaries in dollars: 62,000; 65,000; 68,000; 72,000; 75,000; 78,000; 82,000; 88,000; 95,000; 240,000. The last figure is a senior architect who joined through an acquisition.

The mean is (62 + 65 + 68 + 72 + 75 + 78 + 82 + 88 + 95 + 240) × 1000 / 10 = 925,000 / 10 = \$92,500. The median is the average of the 5th and 6th values: (75,000 + 78,000) / 2 = \$76,500. There is no mode because every value appears exactly once.

The mean overstates the typical salary by roughly \$16,000 — about 21 percent — because one outlier drags it upward. A job candidate negotiating on the basis of "average salary at this company is \$92,500" would be disappointed; one negotiating on the basis of "median salary is \$76,500" would be much closer to reality. This single example explains why every reputable salary survey reports medians, not means.

Two Useful Variations

The trimmed mean drops the top and bottom k percent of observations before averaging. The 10 percent trimmed mean of the salary data above drops the \$240,000 architect and the \$62,000 junior, then averages the remaining eight values to \$77,875 — close to the median but still slightly higher. Trimmed means are standard in Olympic judging (the highest and lowest scores are dropped) and in robust statistical methods where you want some of the efficiency of the mean without its full outlier sensitivity.

The weighted mean multiplies each value by a weight before summing. A student's GPA is a weighted mean where each course's grade is weighted by its credit hours. An inflation index is a weighted mean where each good's price change is weighted by its share of typical household spending. The weights themselves encode information about importance that the unweighted mean ignores.

Which Average Should You Report?

The decision tree is short. If the data is roughly symmetric and you care about totals, report the mean. If the data is skewed, ordinal, or has outliers, report the median. If the data is categorical, report the mode. If the data has multiple peaks, report all three and explain why they differ — the differences themselves are often the most informative part of the analysis. A reporting style that lists mean and median together, the way the U.S. Census does for household income, lets the reader see the skew at a glance.

Conclusion

"Average" is a word that hides three different statistics, each suited to a different kind of question. The mean captures total-over-count and is the right tool for symmetric distributions and totals-driven analysis. The median captures the middle value and is the right tool for skewed distributions, ordinal data, and any situation where outliers would distort the mean. The mode captures frequency and is the only sensible average for categorical data. Choosing the wrong one is not a minor error — it is the source of countless misleading headlines, disappointed expectations, and bad decisions. The right average, paired with a quick look at the distribution's shape, is usually all the statistical machinery an everyday decision needs. Written by the HT99 Tools Editorial Team.