Descriptive Statistics

Descriptive statistics summarizes the data actually in hand, making no claim about anything beyond it — the mean, median, and standard deviation of a specific dataset, exactly as collected, with no attempt to generalize to a broader population. This is the narrower and more modest of the field's two branches: where inferential statistics uses a sample to estimate something about a population it didn't directly observe, descriptive statistics stays entirely within the bounds of what was actually measured, which makes it, by construction, a more certain and less contested kind of statement — there's no sampling uncertainty to account for when a number simply describes the data that exists.

Descriptive statistics organizes around three core questions about a dataset: where is the center, how spread out is it, and what shape does it take. Measures of central tendency — the mean, median, and mode — each answer "where is the center" differently and aren't interchangeable: the mean is sensitive to extreme values and can be pulled substantially away from where most data points actually sit, the median is the middle value and resistant to that pull, and the mode is simply the most frequent value, the only one of the three that's meaningful for categorical data. Measures of spread — the range, variance, standard deviation, and interquartile range — answer "how spread out is it," distinguishing a dataset where most values cluster tightly around the center from one where they're widely scattered, even when both datasets happen to share the same mean. Measures of shape — skewness (whether the distribution leans toward one tail) and kurtosis (how heavy the distribution's tails are relative to a normal distribution) — answer the third question, and both matter because two datasets can share an identical mean and standard deviation while looking completely different when actually plotted.

Reporting only a single summary statistic in isolation — a mean without any accompanying measure of spread — is one of the most common and consequential ways descriptive statistics gets misused in practice, because the mean alone can describe two wildly different underlying datasets equally well. Two datasets can share an identical mean of 50 while one clusters tightly between 45 and 55 and the other spans from 0 to 100 — information the mean alone simply doesn't carry, which is why a standard deviation or interquartile range is conventionally reported alongside any mean rather than left implicit, and why "average" by itself, without any sense of the underlying spread, is widely regarded among statisticians as an incomplete and potentially misleading summary.

Anscombe's quartet — four small datasets constructed by statistician Francis Anscombe in 1973, each sharing nearly identical means, variances, and correlation coefficients despite looking completely different when actually plotted (one roughly linear, one curved, one with a single outlier distorting an otherwise perfect relationship, one with almost no actual relationship at all) — remains the most frequently cited illustration in statistics education of why descriptive summary statistics alone are insufficient and visualization is a necessary complement, not an optional nicety, before trusting any conclusion drawn purely from a table of numbers.

The five-number summary — minimum, first quartile, median, third quartile, and maximum — is the compact descriptive summary that underlies the box plot, giving a quick sense of both center and spread (via the median and interquartile range) and the presence of extreme values (via the minimum and maximum) in a single, standardized format that's become one of the most common ways to report a distribution's shape without needing a full histogram.

Read descriptive statistics as the discipline of characterizing exactly what a dataset contains, nothing more: its core measures of center, spread, and shape together build a fuller picture than any single number can on its own, Anscombe's quartet is the field's standard, decades-old reminder that identical descriptive statistics can hide entirely different underlying data, and the discipline's real value lies precisely in that modesty — it claims only what the data actually shows, leaving the harder, riskier work of generalizing beyond it to inferential statistics.