Summary statistics without misleading yourself

Every summary statistic throws information away on purpose. The skill is knowing which part it discards, because a single number reported without its shape is how honest data becomes a misleading chart.

Centre depends on the distribution

The mean uses every value and is dragged by outliers. The median only uses the ordering and stays where most observations are. For skewed data—latency, salary, prices—the median describes the typical case and the mean describes the total divided by the count.

Spread is not optional

Two datasets with identical means can behave completely differently. Standard deviation, range, or a pair of percentiles tells the reader whether the centre is representative. Performance work reports p95 and p99 for exactly this reason.

Percentages need a stated base

"Up 50%" and "up one percentage point" can describe the same move from a 2% to a 3% rate. Both are correct, so fix the convention before you see which number flatters the result, and include the absolute values.

Small samples exaggerate

With a handful of observations, a single value moves every statistic dramatically. Report the count alongside the summary so readers can calibrate how much to trust it.

Compute and compare

Get the full set of summaries with Statistics Calculator, work through relative change with Percentage Calculator, and read Mean vs Median for the case that matters most.