Are descriptive statistics only averages?

No. They include counts, proportions, spread, shape, and relationships, as well as graphical summaries.

Can a statistic be both descriptive and inferential?

The same numerical quantity, such as a sample mean, can describe observed data and serve as an estimator under an inferential framework.

Is the median always better than the mean?

No. They answer different questions; the choice depends on the distribution and purpose.

AI Glossary / Statistics and probability / Beginner

Descriptive (or summary) statistics

Definition

Descriptive, or summary, statistics organize and condense observed data so that their main features are easier to understand. They include counts, percentages, averages, measures of spread, and summaries of relationships, often supported by tables and graphs. Their immediate purpose is to describe the data at hand. Drawing conclusions about a wider population, future outcomes, or causes requires additional reasoning and assumptions.

What it means

A list of thousands of values can be difficult to interpret directly. Descriptive statistics answer questions such as how many observations were recorded, what values were common, how variable they were, and whether groups differed within the dataset.

Different summaries emphasize different features. The mean uses every value and can be strongly affected by extremes. The median identifies the middle of ordered data. A percentage needs a clear denominator, and a measure of spread adds information that a center alone cannot provide.

Summary choices should fit the variable type and question. Numeric category codes are labels, so averaging them usually has no useful meaning. Ordered categories, continuous measurements, and counts require different care in presentation and interpretation.

How it works

Begin by defining the observational unit, time period, variables, and inclusion rules. Check missing values, duplicated records, units, and obvious inconsistencies before calculating summaries. State how many observations contribute to each result.

For categorical data, use counts and proportions. For numerical data, consider center, spread, extremes, and shape together. Tables and plots can reveal clusters, skewness, unusual values, or time patterns that a single number hides. When summarizing groups, retain group sizes and investigate whether aggregation conceals meaningful differences.

Document weighting and computational conventions. For example, quartile algorithms can differ in small samples, and standard deviation may be calculated with different denominators depending on whether the purpose is describing a complete finite set or estimating population variability from a sample. Clear labels make these choices interpretable.

Three practical examples

A service report gives ticket counts, median resolution time, and the proportion completed within the target, with missing-time records shown separately.
A sales team summarizes revenue by product and period, checking whether a few unusually large orders dominate the overall mean.
An AI evaluation report describes task counts, observed success rates, and response-time distributions before making claims about wider deployment performance.

Think of it this way

Think of a map overview that makes patterns visible without displaying every detail. It preserves what matters for the question, while omitted details may still be important.

Worked example

Consider an original set of six completed task durations, measured in minutes: 4, 5, 5, 6, 10, and 30. The count is six and the total is 60 minutes, so the mean is 60/6 = 10 minutes. The median is the average of the third and fourth ordered values, (5+6)/2 = 5.5 minutes. The mode is 5 minutes.

The range is 30−4 = 26 minutes. Four of the six tasks finish within six minutes, giving 4/6, approximately 66.7%. The sum of squared deviations from the mean is 36+25+25+16+0+400 = 502. If these six observations are treated as a sample for estimating a common population variance, the usual sample variance is 502/5 = 100.4 and sample standard deviation is approximately 10.02 minutes.

The mean is much larger than the median because the 30-minute task contributes heavily to the total. Removing that task would change the dataset and should not be done merely to produce a more favorable average.

If a seventh assigned task has no recorded duration, report six observed durations and one missing value. The 66.7% result describes the observed six; it does not automatically describe all seven assigned tasks or future work. Missing outcomes and operational failures need explicit treatment under the reporting definition.

Illustrative case study

In this fictional case, Elmbridge Operations evaluates an AI assistant for document preparation. A dashboard initially reports only average completion time. Reviewers discover that routine documents are fast, while a small number of complex documents require much longer review.

The team adds task counts, medians, upper-tail summaries, and a breakdown by document category. It also reports unfinished tasks and explains whether time includes human checking. Managers can then see both typical work and the cases driving total workload.

The invented case illustrates reporting. Summaries clarify observed performance but do not establish causation or future performance.

Common misconceptions and mistakes

An average is not a complete description of a distribution. Two datasets can share a mean while differing greatly in spread, tails, or group composition. A chart can likewise mislead if axes, bins, or denominators are unclear.

Descriptive does not mean assumption-free or immune to errors. The data may be incomplete, selectively collected, incorrectly measured, or weighted in ways that matter. Summaries inherit those limitations.

Do not average subgroup percentages without considering their denominators when the target is the combined proportion. Nor should correlation or a before-and-after difference be presented as proof of causation. Finally, a standard deviation describes variation among observations; a standard error describes uncertainty in an estimator. Substituting one for the other changes the meaning of the report.

Business use

Businesses use descriptive statistics to understand operations, detect data-quality problems, communicate performance, and decide which questions deserve deeper analysis. Good reporting begins with a precise metric definition and a transparent population of records.

Show counts and missingness alongside key summaries. Choose measures that reflect the decision: total workload, typical experience, and extreme delays are different concerns. Preserve consistent definitions across periods, and explain any changes. For AI projects, describe evaluation-set composition and relevant subgroups so that aggregate results do not obscure weak areas. Follow descriptive findings with appropriate validation or inference when decisions require claims beyond the observed data.

Check your understanding

Check your understanding: Why is the example's mean 10 but median 5.5?

Answer: The long task strongly affects the total and mean.

What denominator produces 66.7%?

Answer: Six observed tasks.

Does a missing seventh duration count as zero?

Answer: No.

Why report several summaries?

Answer: Each reveals a different feature of the data.

Related terms

Learn next

Compare summaries and plots for datasets with the same mean. Then study sampling, confidence intervals, and how to distinguish descriptive differences from causal claims.

Mean

Sources

Sources: Australian Bureau of Statistics, Statistical terms and concepts glossary, https://www.abs.gov.au/statistics/understanding-statistics/statistical-terms-and-concepts/statistical-terms-and-concepts-glossary. NIST/SEMATECH, Measures of Scale, https://www.itl.nist.gov/div898/handbook/eda/section3/eda356.htm. The task-duration arithmetic and fictional operations case are original.

Previous: Descriptive analytics Next: Design effect