Measures of Central Tendency: Mean, Median and Mode Explained

The three main measures are the mean (arithmetic average, calculated by dividing the sum of values by the count), the median (middle value of ordered data), and the mode (most frequently occurring value). Each defines "center" differently and suits different data types and distribution shapes.

The mean uses every value in its calculation, making it sensitive to outliers. The median uses only the middle position, making it resistant to outliers. The mode counts frequencies and can identify multiple peaks in data. For symmetric data without outliers, all three coincide; for skewed data, they diverge systematically.

There is no universally best measure. Use the mean for symmetric interval or ratio data without extreme outliers. Use the median for skewed data or ordinal data, or when outliers are present. Use the mode for nominal data or to identify the most common value. The "best" measure is the one that honestly represents a typical observation given your specific data and question.

Use the mean when your data is measured on an interval or ratio scale, the distribution is approximately symmetric, and there are no extreme outliers distorting the average. The mean is also preferred when you need to perform further calculations (variance, standard deviation, confidence intervals), because it is algebraically tractable in ways the median and mode are not.

Use the median when the data is skewed, when outliers are present and would distort the mean, or when the data is measured on an ordinal scale. Common domains include income data, housing prices, healthcare survival times, and any situation where a few extreme values would make the mean unrepresentative of the typical case.

Use the mode for nominal (categorical) data — it is the only valid measure of central tendency in that case. It is also useful for identifying the most common value in discrete datasets, for revealing multiple peaks in bimodal distributions, and for describing what "most people" chose or experienced. Avoid it for continuous data where values rarely repeat.

The mode. Nominal data consists of categories with no meaningful numeric ordering — for example, blood type, country of origin, or favourite colour. Because these values cannot be averaged or ranked, the mean and median are meaningless. The mode simply identifies the most common category, which is the only valid summary of center for categorical data.

Outliers pull the mean toward the extreme value, sometimes dramatically. If nine employees earn between $30k and $40k and one earns $500k, the mean salary is roughly $89k — which accurately describes nobody. The mathematical reason is that the mean sums all values, so one very large or very small number carries as much weight as all the others combined when the sample is small.

No — the median is resistant to outliers. Because it depends only on the position of the middle value, changing any observation that remains on the same side of the median leaves the median unchanged. In the salary example above, replacing the $500k CEO salary with $10M would not move the median at all.

Karl Pearson observed that for moderately skewed unimodal distributions, Mode ≈ 3 × Median − 2 × Mean. This allows you to estimate one measure when the other two are known. It is an approximation, not an exact identity, and it breaks down for severely skewed or multimodal distributions.

Compute the class midpoint for each interval ((lower bound + upper bound) ÷ 2), multiply each midpoint by its class frequency, sum those products, and divide by the total frequency: x̄ = Σ(fx) ÷ Σf. The result is an estimate because the actual values within each class are unknown.

Use the interpolation formula: Median = l + ((n/2 − cƒ) ÷ f) × h. First build a cumulative frequency column and identify the median class — the first class whose cumulative frequency reaches or exceeds n/2. Then substitute l (lower class boundary), cƒ (cumulative frequency before the median class), f (frequency of the median class), and h (class width) into the formula.

Identify the modal class (the class with the highest frequency), then apply: Mode = l + ((f₁ − f₀) ÷ (2f₁ − f₀ − f₂)) × h, where f₀ is the frequency of the class before the modal class and f₂ is the frequency of the class after it. The result is an estimate of where the mode lies within the modal class.

The geometric mean is the nth root of the product of n positive values. Use it for data that involves multiplicative processes: investment returns (compound annual growth rate), population growth rates, and price index changes over time. It is always less than or equal to the arithmetic mean, and it correctly accounts for the compounding effect that the arithmetic mean ignores.

The harmonic mean, calculated as n ÷ Σ(1/xᵢ), is the correct average for rates where the denominator varies. The classic application is average speed over equal distances. If a car travels 60 km/h out and 40 km/h back, the harmonic mean (48 km/h) gives the actual average speed; the arithmetic mean (50 km/h) is wrong because more time is spent at the lower speed.

A weighted mean assigns different importance to different observations. The formula is x̄_w = Σ(wᵢ × xᵢ) ÷ Σwᵢ. Common applications include GPA (courses with different credit hours), survey estimates where sampling probabilities differ, and portfolio returns where assets have different allocations. Always verify that the weights sum to the intended total before interpreting the result.

A trimmed mean removes a fixed percentage of the extreme values from both ends of the dataset before calculating the arithmetic mean. A 10% trimmed mean drops the lowest 5% and highest 5% of observations. It sits between the mean (which includes all values) and the median (which uses only the center) in terms of outlier sensitivity. It is common in competitive scoring, such as Olympic gymnastics judging.

The calculation is identical: sum all values and divide by the count. The difference is notation and statistical interpretation. The population mean uses the symbol μ (mu) and N (total population size). The sample mean uses x̄ and n (sample size). The distinction matters for statistical inference — sample means are used to estimate population means, with associated uncertainty — but not for the arithmetic itself.

In a positively skewed (right-skewed) distribution, the ordering is Mode < Median < Mean. The long tail extends to the right, pulling the mean in that direction while the mode remains at the most frequent (leftward) peak and the median sits between them. Income distribution is the most cited real-world example.

In a negatively skewed (left-skewed) distribution, the ordering is Mean < Median < Mode. The long tail extends to the left, pulling the mean downward, while the mode remains at the peak on the right side of the distribution. Age at retirement and scores on an easy exam are common examples.

A bimodal distribution has two distinct peaks — two values or ranges that appear with higher frequency than the values between them. When data is bimodal, no single measure of central tendency adequately summarizes it. Both modes should be reported, and the data should be investigated to determine whether subgroups (for example, two distinct populations or processes) are producing the two peaks.

Yes. When every value in a dataset appears exactly once, there is no mode by definition. This is common with continuous data (measurements rarely repeat exactly) and with small datasets containing all distinct values. In these cases, the mean and median are the appropriate summary measures.

Central tendency measures where a dataset is centered (mean, median, mode). Dispersion measures how spread out the values are around that center (standard deviation, variance, interquartile range, range). Both are necessary for a complete descriptive summary. Two datasets can share the same mean but have completely different spreads — reporting only the center hides that difference.

Report mean with standard deviation when data is approximately symmetric and interval/ratio-level. Report median with interquartile range (IQR) when data is skewed or contains outliers, or when it is ordinal-level. Report mode with relative frequency for nominal data. These pairings are internally consistent: mean and SD are both sensitive to all values; median and IQR are both position-based and outlier-resistant.

Always report the median for income data. Income distributions are right-skewed in virtually every country: a small number of very high earners pull the arithmetic mean well above what most people earn. Government agencies, the U.S. Census Bureau, and international organizations like the OECD all use median household income as the standard measure of typical living standards.

The safest choice is the median, because Likert scale data is ordinal — the intervals between "Agree" and "Strongly Agree" are not guaranteed to be equal. The mode is also informative, showing the most common response. Mean is widely reported in practice and in published research, but this involves the assumption of equal intervals, which is contested. When in doubt, report all three and show the full distribution of responses.

Three standard approaches: (1) statistics.mean(data) from Python's built-in statistics module; (2) numpy.mean(data) from NumPy; (3) sum(data) / len(data) using no imports. For grouped data, use numpy.average(midpoints, weights=frequencies) to compute the weighted mean of class midpoints.

Use =MEDIAN(range). Excel handles both odd and even n automatically: for odd n, it returns the middle value; for even n, it averages the two middle values. The function accepts up to 255 separate number arguments or ranges and ignores text and logical values within a range.

A substantial gap between mean and median is a signal that the data is skewed or contains influential outliers. If mean > median, the distribution is right-skewed (high values pulling the mean up). If mean < median, it is left-skewed. In either case, graph the data with a histogram before deciding which measure to report. For skewed data, the median is almost always the more informative summary of a typical observation.

Key properties: (1) The sum of deviations from the mean equals zero: Σ(xᵢ − x̄) = 0. (2) The mean minimizes the sum of squared deviations. (3) Adding a constant to all values shifts the mean by that constant. (4) Multiplying all values by a constant scales the mean by that constant. (5) The mean is unique — there is exactly one per dataset. (6) It is algebraically tractable, making it the foundation for variance, standard deviation, and regression.

Rarely, and with caution. Truly continuous measurements (like height or weight to many decimal places) rarely repeat exactly, so no mode exists. However, if continuous data is rounded to a practical precision and one value appears noticeably more often, reporting it may be informative. More commonly, the mode is reserved for discrete or categorical data where repetition is meaningful and expected.

The midrange is (minimum + maximum) ÷ 2. It is easy to compute but highly sensitive to outliers — even more so than the mean — because it uses only the two extreme values and ignores all others. It is rarely reported in practice. For the midrange in more detail, see midrange.

In a perfect normal distribution, mean = median = mode, and all three fall at the axis of symmetry. This equivalence is why the mean is the natural center measure for normally distributed data, and why testing assumptions of normality is important before applying procedures that assume mean-centered summaries. The empirical rule (68-95-99.7) describes how data clusters around that shared center.

Use the arithmetic mean when values combine additively: total scores, total quantities, averaged measurements. Use the geometric mean when values combine multiplicatively: growth rates, ratios, investment returns, index numbers. The arithmetic mean of percentage returns overstates performance when there is volatility; the geometric mean gives the true compound rate. For symmetric data without multiplicative structure, the arithmetic mean is appropriate.