Descriptive Analysis

Summarising what happened: counts, averages, breakdowns.

Also known as: descriptive analysis, descriptive statistics in practice, summarising data

Descriptive analysis answers “what happened?” with summaries: totals, averages, breakdowns by segment, distributions over time. It is the foundation every other analysis type stands on — you cannot diagnose, predict or prescribe before you can describe accurately.

raw events → totals, means, medians → breakdowns (by segment, by time)
           → distributions (shape, spread, outliers)

The work is choosing summaries that do not lie by omission. A mean without the distribution hides skew; a total without the time window hides pace; a breakdown by one dimension hides the dimension that actually explains it. Report the centre and the spread, the total and the rate, the overall and the segments.

The classic mistakes:

Mean-only reporting. Revenue per user “averages $40” while the median is $3 and one whale pays thousands. Show the distribution or pick the typical value honestly.
Breakdowns that stop one level too soon. Overall conversion is flat while every segment moves — Simpson’s paradox lives in undescribed data. Cut by the dimensions that could change the story.
Precision without context. “12,483 users” with no comparison, no window, no definition of “user”. Every number needs its denominator and its timeframe.
Describing dirty data beautifully. Summaries of unexamined data launder its problems. Profile first (exploratory analysis).

Good description is underrated because it looks easy. It is easy to do and easy to do misleadingly — the difference is whether each number carries its context with it. See data visualisation for presenting it.