Mean, median, and mode

For the same question "Rate from 1 to 5" you can say "the average is 3.8", "the median is 4" or "4 was chosen most often". All three numbers describe the "center" of the data, but they are calculated differently and react differently to outliers and skew.

Survey reports most often feature the mean; for scale and ordinal data it is frequently useful to look at the median and the mode as well — otherwise the picture can be incomplete or distorted.

Below is what each measure is, when to use which, and how to avoid mixing them up when interpreting results.

What the mean, median, and mode are in plain terms

The arithmetic mean (Mean) is the sum of all values divided by their count. It is sensitive to outliers: a few very large or very small answers noticeably shift the mean. The median (Median) is the value "in the middle" of an ordered series: half of the observations are below the median, half are above. Outliers at the edges have almost no effect on it. The mode (Mode) is the most frequently occurring value. For categorical and ordinal variables, the mode shows the "most popular" answer option.

Put simply: the mean is "if you spread all answers out evenly"; the median is "the middle of the list in order"; the mode is "which option was chosen most often". In a symmetric distribution without outliers the mean and median are close; with skew or rare extreme values they diverge.

When to use which measure

The mean is convenient when the data is roughly symmetric and there are no strong outliers. It is often used to calculate confidence intervals, compare groups (t-test), and run regression. In satisfaction reports an "average score of 4.2" is clear to the client — but with a skew toward "5" or a few isolated "1" answers, the mean may not reflect the typical answer.

The median is more resistant to outliers and skew. It makes sense to calculate it for ordinal and scale data (for example, on a Likert scale) when some respondents gave extreme ratings. "Median 4" means: half of the answers are no higher than 4, half are no lower. For the client the phrasing "half rated it 4 or higher" is sometimes more vivid.

The mode is useful for categorical variables (region, customer type) and when it is important to emphasize the "most frequent" answer. On a 1–5 scale the mode may coincide with the median or differ; if the mode and the mean differ greatly, that hints at skew or two "peaks" in the data. A single distribution can have two modes (bimodality) — in that case there is no single "typical" value, and it is better to describe the shape of the distribution in words or by proportions.

An example in numbers

Question "Rate the service from 1 to 5": 100 answers. Variant A: 10% — "1", 20% — "2", 40% — "3", 20% — "4", 10% — "5". The mean is about 3, the median is 3, the mode is 3 — all three measures at the center. Variant B: 5% — "1", 10% — "2", 15% — "3", 30% — "4", 40% — "5". The mean shifts to the right (about 3.9), the median is 4, the mode is 5. A single "five" from several dissatisfied respondents pulls the mean more than the median; the mode shows that "5" was chosen most often. For the full picture it is better to report both the mean and the median (and, if needed, the spread).

A third scenario — outliers: 95 people gave "4", five gave "1". The mean drops (about 3.85), the median stays at 4, the mode is also 4. A client seeing only "mean 3.85" might decide the score is "so-so"; in fact almost everyone is satisfied, and a few low ratings shifted the mean. So with this setup the report must include the median and, if desired, the proportions by option.

The median with an even N. If the number of observations is even, the median is usually taken as the half-sum of the two central values of the ordered series. For example, with 100 answers it is the average of the 50th and 51st in order. Programs and spreadsheets compute this automatically; it is important to remember that the median need not coincide with any real answer (with an even N and different central values it can be, say, 3.5).

Link to descriptive statistics and distribution

The mean, median, and mode are part of descriptive statistics as measures of central tendency. In a normal distribution they coincide; with skew they diverge — the difference between the mean and the median lets you judge the skew. So a report should not be limited to a single number: stating the median alongside the mean (and the mode if needed) helps the reader understand the shape of the data. For more on spread and interpretation, see the articles on scales, cross-tabulations, and standard deviation.

The spread of the data is described separately: standard deviation, quartiles, minimum and maximum. Measures of center without a measure of spread give an incomplete picture — "mean 4" with a spread from 1 to 5 and "mean 4" with all answers being "4" are different situations. In segmentation and group comparison the mean and median are calculated for each segment; to test differences between groups, tests are used (statistical significance).

By segments and subgroups

The mean, median, and mode are calculated not only across the whole sample but also within segments: by region, customer type, age. In cross-tabulations and group breakdowns it is important to state the size of each subgroup — otherwise "mean 4.5" for 15 respondents and for 200 will be perceived the same way, even though their reliability differs. For small subgroups the median is often more reliable than the mean as a "typical" value.

When weighting answers, a weighted mean is calculated: each value is multiplied by the respondent's weight (inverse to the probability of being included in the sample, or a calibration weight), then divided by the sum of the weights. The median and the mode can also be calculated from weighted frequencies when weighting — in that case the report states that weighted measures are shown.

Common mistakes

Reporting only the mean for ordinal data. On 1–5 scales, with skew or outliers the median is often more informative. If the report shows a single "mean 3.2" while 60% gave "4" or "5", the client may draw the wrong conclusion.

Confusing the median and the mean when comparing groups. "Group A has a higher mean" and "Group A has a higher median" are not the same thing when the shape of the distribution differs. To compare groups, tests are used (on means or on ranks); when describing, it is better to state both measures.

Looking for the mode in continuous data without grouping. If the variable is essentially continuous (time, amount), the "most frequent value" may be unique for each respondent. The mode is then calculated from grouped data (intervals).

Forgetting about the sample size. Both the mean and the median based on 20 answers are less stable than those based on 500. Alongside a measure of center you should state the number of observations and, if needed, the spread.

Interpreting the mean as "most people answered this way". A mean of 3.5 can result both when everyone gave 3 or 4 and when half gave 1 and half gave 6. The "typical" answer is better described by the median or the mode; the mean answers the question "what is the average value", not "how did the majority answer".

Visualization

A histogram or bar chart of the distribution of answers immediately shows whether the data is symmetric, whether there is one peak (mode) or two, and whether the distribution is shifted to the left or to the right. From such a picture it is easier to decide whether one mean is enough or whether you should add the median and an explanation. Survey reports often include both a table with proportions by option and the mean (plus the median in case of skew) — so the client sees both the "center" and the shape of the distribution.

How it looks in SurveyNinja

In reports, by default each question shows the number of answers and the proportions by option; for scales the mean is displayed. The median and the mode are not computed in the interface — you can get them after exporting the answers to CSV/XLSX and calculating them in a spreadsheet or a statistical package. When preparing a report for the client it is convenient to add the median next to the mean if the scale data is skewed or there are outliers.

Practical recommendations

For scale and ordinal questions report both the mean and the median where possible. If they differ noticeably, briefly explain it (for example, "skew toward high ratings") or show the distribution.

With outliers rely on the median for the "typical" value, or explicitly state that the mean may be biased by isolated extreme answers.

For categorical variables (gender, region, customer type) only frequencies and the mode are appropriate; the arithmetic mean is not calculated for them.

When all three measures are close. If the mean, median, and mode almost coincide (for example, all around 4 on a 1–5 scale), the distribution is close to symmetric and without strong outliers. In such cases it is enough to state the mean; when the measures diverge, add the median and, if needed, explain the shape of the distribution.

How to phrase it in a report. Instead of a dry "mean 3.8" you can write "average score 3.8 out of 5 (median 4)" or "half of the respondents rated it 4 or higher". For the client "option '4' was chosen most often" (the mode) is often clearer than the mean alone. If the mean and the median differ greatly, add a short explanation: "skew toward high ratings" or "a few extreme answers shifted the mean".

In brief

• The mean — the sum of values divided by the count; sensitive to outliers.
• The median — the value "in the middle" of the series; resistant to outliers, convenient for ordinal and scale data.
• The mode — the most frequent value; suitable for categorical variables and when the "most popular" answer matters.
• In a survey report it is better to give both the mean and the median for scale questions; with skew or outliers, rely on the median and proportions.

The mean, median, and mode answer the question "where is the center of the data" in different ways. In surveys with ordinal and scale answers it is useful to look not only at the mean but also at the median (and the mode if needed), so as not to distort the picture with outliers and skew of the distribution.