The short answer: the mean adds the values and divides by how many there are. For 1, 2, and 9 the mean is (1 + 2 + 9) ÷ 3 = 4. The median is the middle value once the list is sorted, which is 2 here, and the weighted mean multiplies each value by its weight before dividing by the total weight. Outliers move the mean far more than the median.

Key takeaways

  • Mean uses every value. The sum divided by the count is the arithmetic mean.
  • Median resists extremes. Sorting and taking the middle value ignores how large the outliers are.
  • Weighted mean respects importance. Multiply each value by its weight, then divide by the total weight.
  • Skewed data needs care. Incomes, prices, and waiting times often deserve a median.
  • An outlier is a signal. Investigate it before deleting it.
  • Never average group averages without weights. Unequal group sizes distort the result.

Three different averages

“Average” is a family of measures, not one number. The mean is the sum divided by the count. The median is the middle value of a sorted list. The mode is the value that occurs most often. They answer related but distinct questions: what the total would be if shared equally, what the middle case looks like, and what is most common.

For a symmetric list without extreme values, the three measures usually land close together, and the choice hardly matters. For skewed data they can diverge sharply. That divergence is not a paradox; it is information about the shape of the data, and choosing which one to report is an editorial decision about what “typical” should mean.

The arithmetic mean

mean = (x₁ + x₂ + … + xₙ) ÷ n

The mean uses every value exactly once. It has a useful balancing property: the signed deviations from the mean sum to zero. For 1, 2, and 9, the mean is 4, and the deviations are −3, −2, and +5, which add to zero. The mean is the point where the list would balance if each value were a weight placed on a number line.

That same property makes the mean sensitive. A single very large value pulls the balance point toward it, because the mean must move until the deviations cancel again. The mean is also the natural summary when the total matters: average revenue per customer multiplied by the number of customers recovers total revenue, which is not true of the median.

Worked example: when the mean misleads

Five salaries of 40,000, 42,000, 45,000, 48,000, and 500,000 have a total of 675,000 and a mean of 135,000. The median is 45,000. Almost everyone in the group earns less than the mean, which is why “average salary” can feel misleading. The mean is not wrong; it is answering “what would each person get if the total were split equally,” and that is rarely the question a job applicant is asking.

The gap between the mean and the median measures skew. When the mean is much larger than the median, the right tail is heavy. When the mean is much smaller, the left tail is heavy. Reporting both is a compact way to show that the data are not symmetric.

The median as a robust alternative

To find the median, sort the values from smallest to largest. With an odd count, take the middle value. With an even count, average the two middle values. For 2, 4, 8, and 10, the two middle values are 4 and 8, so the median is 6. For 1, 2, and 9, the median is 2 regardless of how large the 9 becomes.

The median depends on rank, not magnitude, so extreme values cannot drag it. That robustness is why incomes, house prices, and response times are often reported as medians. The trade-off is that the median ignores information: changing 9 to 9,000 leaves the median at 2, even though the data have changed dramatically.

Weighted mean

weighted mean = (w₁x₁ + w₂x₂ + … + wₙxₙ) ÷ (w₁ + w₂ + … + wₙ)

A weighted mean is appropriate when some observations count for more than others. Course grades are the classic case. If homework is 20 % of the grade at 90, the final exam is 50 % at 80, and the project is 30 % at 85, the weighted mean is 0.20 × 90 + 0.50 × 80 + 0.30 × 85 = 18 + 40 + 25.5 = 83.5. The unweighted mean of 90, 80, and 85 would be 85, which overstates the result because the lowest score carries the most weight.

Grade point averages work the same way with credits as weights. Grades of 3.7 in a 4-credit course, 3.0 in a 3-credit course, and 4.0 in a 2-credit course give (3.7 × 4 + 3.0 × 3 + 4.0 × 2) ÷ 9 = 31.8 ÷ 9 ≈ 3.53. If the weights are normalized to sum to 1, the denominator is 1 and the formula simplifies, but the result is unchanged.

Choosing the right average

Use the mean when the data are roughly symmetric and the total is meaningful. Use the median when the distribution is skewed or contains outliers, or when you want the typical case rather than the total. Use the mode for categorical or repeated discrete values, such as the most common shoe size. Use a weighted mean whenever the observations represent groups of different sizes or importance.

Growth rates need a geometric mean, not an arithmetic one. A value that grows 10 % and then 20 % has a multiplier of 1.10 × 1.20 = 1.32, so the average annual factor is √1.32 ≈ 1.149, about 14.9 % per year. The arithmetic mean of 10 % and 20 % is 15 %, which is close here but wrong in general; with a 50 % gain and a 50 % loss, the arithmetic mean is 0 % while the actual result is a 25 % loss.

Detecting outliers without deleting them

An outlier is a value that sits far from the rest. A common screening rule uses the interquartile range: flag values below Q1 − 1.5 × IQR or above Q3 + 1.5 × IQR, where Q1 and Q3 are the first and third quartiles. The NIST/SEMATECH handbook also describes the modified z-score, 0.6745 × (x − median) ÷ MAD, with absolute values above 3.5 worth labelling as potential outliers.

A flag is a reason to look, not a reason to delete. An outlier may be a data-entry error, a broken instrument, or a genuine extreme event that is the most important observation in the set. The NIST guidance is to investigate first, correct if the value is provably wrong, and otherwise use a robust method such as a trimmed or winsorized mean rather than silently dropping the point.

Common mistakes

  • Reporting the mean for skewed data. One large value can place the mean above most of the list.
  • Forgetting to sort before taking the median. The middle of an unsorted list is meaningless.
  • Mishandling an even count. Average the two middle values after sorting.
  • Averaging percentages or rates directly. Weight by the underlying counts, or use a geometric mean for growth.
  • Combining group averages without weights. A 90 % rate in a group of 10 and a 50 % rate in a group of 100 is not a 70 % overall rate.
  • Deleting outliers to improve the result. That changes the question being answered.

Practical caveats

Check the units and the count before trusting any average. Mixing dollars and cents, or minutes and hours, produces a number with no meaning. A missing value is not a zero, and treating it as one drags the mean down; decide explicitly whether to exclude incomplete records or impute a value, and report which you did.

Precision matters as well. A mean of 4.333333 implies more certainty than three integer observations can support, so round the reported average to the precision of the inputs. Finally, remember that an average is a summary, not the data. Two very different lists can share a mean of 10, and a mean says nothing about the spread. Pair it with a measure of variation, such as the standard deviation or the interquartile range, before drawing conclusions.

Frequently asked questions

Is “average” the same as “mean”?

Colloquially, usually yes. Technically, average can refer to the mean, median, or mode, so name the measure when the distinction matters.

How do I find the median with an even number of values?

Sort the values and average the two middle ones. For 2, 4, 8, and 10, the median is (4 + 8) ÷ 2 = 6.

Can a list have more than one mode?

Yes. If two or more values tie for the highest frequency, the list is bimodal or multimodal. If every value appears once, there is no mode.

Why does one large value move the mean so much?

The mean uses every value’s magnitude, so the total rises with the outlier while the count stays the same. The median only uses rank and barely moves.

When should I use a weighted mean?

When observations represent different amounts, sizes, or importance, such as course credits, group sizes, or sales volumes. Weight each value by its share.

Can an average include negative numbers?

Yes. Negative values are summed like any others, and the mean can be negative. The interpretation depends on what the numbers measure.

Further reading

Keep exploring

Read next Significant Figures: Counting, Rounding, and Precision Significant figures record how precisely a value was measured. Learn the counting rules, the addition and multiplication rules,… Read next Logarithm: Base, Product Rule, and Scale A logarithm is an exponent in disguise. Learn the definition, the product, quotient and power rules, change of… Read next Order of Operations: PEMDAS and Ambiguous Expressions PEMDAS is a shared convention, not a law of mathematics. Learn the real precedence rules, the exponent and…

Decision path

Use this guide in a decision

Open Math Decision Center Open companion solver