Definition
Descriptive statistics is the branch of the discipline that organises, summarises, and presents data without attempting to draw conclusions beyond the cases observed. It includes the single-number measures of central tendency (mean, median, mode), the measures of dispersion (range, variance, standard deviation, interquartile range), the shape descriptors (skewness, kurtosis), and the visual displays (histograms, bar charts, scatter plots, box plots) that turn a table of numbers into something a reader can grasp at a glance.
It contrasts with inferential statistics, which uses sample data to estimate properties of an unobserved population and to test hypotheses about that population. Descriptive work answers "what does the data show?" Inferential work answers "what does the data let us conclude about the world beyond it?" Almost every analysis begins with the descriptive stage; skipping it is the surest way to chase noise.
Why it matters
How it works
A typical descriptive workflow has three phases. The first is to summarise each variable on its own. For numeric data, that means reporting the mean and median (centre), the standard deviation and interquartile range (spread), and a histogram or density plot (shape). For categorical data, it means frequency tables and bar charts. Outliers are flagged, missing values are counted, and any obvious data-quality problems are surfaced before any modelling begins.
The second phase examines pairwise relationships: scatter plots for numeric pairs, cross-tabulations for categorical ones, side-by-side box plots when one variable is categorical and the other numeric. Correlation coefficients quantify the strength of linear association. The third phase examines higher-dimensional structure with techniques such as parallel coordinate plots, principal-component scatters, or small-multiple plots that show how relationships shift across groups. Throughout, the goal is to develop an honest, internalised picture of what the data actually contains before drawing any inferential or causal conclusions from it.