The Average Salary Is $72,000. The Median Is $48,000. One of These Tells You Almost Nothing.
At a company where a dozen executives earn between $300,000 and $800,000, and forty staff members earn between $38,000 and $55,000, the mean salary might be reported as $72,000. That number is mathematically correct. It is also completely misleading. It describes no actual employee's experience and would give a job applicant a profoundly false picture of what working there pays.
The mean, median, and mode all claim to answer the same question: what is the central value of this dataset? They produce different answers because they define "central" differently. Those differences are not technical details to be left to statisticians. They determine whether the number you are reading reflects your actual situation or someone else's.
This tension has a long history. In the 1880s, the British statistician Francis Galton argued that the arithmetic mean was routinely abused in anthropometric and social data because a few extreme values pulled it away from the experience of most people. He began using what he called the "middle-most value" in 1869, eventually formalizing the term "median" in 1881 as a deliberate alternative to the mean for skewed distributions. The debate Galton was having about which measure to trust has not been resolved. It plays out every time someone reads a salary figure, a housing report, or a national income statistic.
Mean: What the Formula Produces and What It Hides
The mean is the sum of all values divided by the count of values. Its mathematical simplicity is its main appeal: it uses every data point in the calculation, and for symmetrically distributed data with no extreme outliers, it produces a reliable center estimate.
The problem surfaces in skewed distributions. If a dataset contains 40 values clustered between $38,000 and $55,000, and 5 values between $300,000 and $800,000, the five high earners do not just nudge the mean upward slightly. They pull it far above the cluster. A company with this pay structure might report a mean salary of $72,000 when only a small fraction of employees earn within $20,000 of that figure.
The Bureau of Labor Statistics has been explicit about this problem for decades. When reporting national wage data, the BLS uses median figures rather than mean figures because the department recognizes that extreme earners distort the mean and misrepresent what typical workers are paid. In 2024, the mean annual salary in the United States was approximately $66,600, while the median was approximately $62,000. The $4,600 gap between those two numbers represents the upward pull of high earners on the mean. In income data, the mean is almost always higher than the median for this reason, and the gap widens as income inequality increases.
The mean is appropriate when data is roughly symmetric and outliers are not a concern: exam scores in a class, daily temperature readings, product dimensions on an assembly line. In those cases, every data point is a legitimate representative of the distribution. It is the wrong tool when a small number of extreme values are doing outsized work.
Median: The Number That Describes Most People
The median is the middle value in an ordered dataset. Sort all values from lowest to highest and find the one that splits the group exactly in half. For an even number of values, average the two middle entries. Every data point above or below the median is irrelevant to the calculation; only position matters. This makes the median resistant to distortion in a way the mean never can be.
The National Association of Realtors reports median home prices, not mean home prices. This is a deliberate methodological choice. A single sale at $10 million in a market where most homes sell for $350,000 to $500,000 would significantly inflate the mean. The median would barely move. Real estate analysts want to describe where most buyers transact, and for that question, the median is the more honest answer.
Consider a concrete case. A neighborhood sells 11 homes in a month: 10 properties close between $280,000 and $420,000, and one closes at $3.8 million. The mean sale price for that month is around $625,000. The median is around $350,000. A buyer trying to understand what purchasing in that neighborhood costs would be badly misled by the mean. The median describes the actual market.
The same logic applies to any skewed dataset. When mean and median diverge substantially, the data is sending a signal: there are outliers operating far from the center, and the mean is being moved by them while the median is not. Seeing both numbers together is almost always more informative than seeing either alone.
Mode: When the Most Common Value Is the Right Question
The mode is the value that appears most frequently in a dataset. For many real-world questions, the mode is the most directly useful measure and the other two are irrelevant.
A clothing manufacturer trying to decide how many units of each size to produce does not need to know the mean or median clothing size. Neither of those values will tell the production team whether to make more size mediums or size larges. The mode, the size purchased most often, tells them exactly what they need. Similarly, a survey asking customers to rate a product on a scale of 1 to 5 might have a mean of 3.8, a median of 4, and a mode of 5. The mode tells the manufacturer that the most common response was the highest possible rating, a fact that neither the mean nor the median communicates as clearly.
For categorical data where values have no inherent order (colors, product categories, voting choices), the mode is the only measure that applies at all. The mean color of a dataset is undefined. The median category in an unordered list is meaningless. But the most common category is always a valid question, and mode answers it.
When all three measures are equal, the data is approximately symmetric: a normal distribution with no meaningful skew. When they diverge, particularly when mean and median pull apart, the data has a tail, and that tail is worth understanding before trusting any single number as the summary.
Conclusion
A single central tendency number almost never tells the full story. The salary example that opens this article illustrates the problem exactly. A job candidate told that the average salary is $72,000 might accept an offer of $55,000 thinking she is below average. If the median salary is $48,000, she is actually being paid above the level that most employees earn, and the gap is driven by executive compensation, not by a general salary floor near $72,000.
The most common statistical abuse is reporting the mean when it flatters, reporting the median when it obscures, and never mentioning both. Recognizing when a number is a mean versus a median is the first step to reading it correctly. The gap between the two measures is the skew in the data made visible.
A statistics calculator that computes mean, median, and mode together, alongside standard deviation and range, gives you enough to see the shape of a distribution rather than just a single summary. Running all three on your data before accepting any central tendency claim is a simple habit that makes it considerably harder to be misled by a well-chosen average.
Frequently Asked Questions
When should you use median instead of mean?
When the data contains significant outliers or is skewed. Home prices, salary data, and income distributions are typically reported using the median because a few extreme values would distort the mean.
What does it mean when mean and median are very different?
It indicates a skewed distribution with outliers pulling the mean in one direction. A mean significantly higher than the median suggests a few very large values. A mean significantly lower suggests a few very small values.
When is mode the most useful average?
For categorical data (colors, sizes, categories) where mean and median are not meaningful. Also useful in large numerical datasets where knowing the most common value matters more than the mathematical center.