Reading tables of data
3 min read
Core idea
A table is a communication device, not a storage cabinet. Databases store data; tables on a page exist to make patterns visible to a human reader. The corollary is harsh: any table whose layout obscures rather than reveals is failing its job. Effective tables rank rows by what matters, round numbers to as few digits as defensible, replace raw counts with rates when comparison demands it, and tell the reader exactly which population, geography, and definition is in view via title, source line, and footnotes.
Why it matters
Most readers either skip tables entirely or skim them so quickly they miss the story. The work of designing a clear table is the work of meeting them halfway: putting the most interesting column first, sorting rows by the dimension you want them to compare, and stripping every digit that is not load-bearing. Done well, a single table replaces a paragraph of prose. Done badly, it absorbs page space while transmitting nothing.
Mental model
Precision is not accuracy
The two words are routinely confused. Precision measures reproducibility — how tightly repeated measurements cluster. Accuracy measures truth — how close the centre of those measurements is to the real value. A bathroom scale that reads 70.000 kg every morning is precise; if you really weigh 75 kg, it is wildly inaccurate. Quoting a table to four decimals when the underlying measurement is good to one suggests false confidence in numbers that do not deserve it.
The redesign pipeline
A messy raw table rarely becomes a clear one in a single edit. Treat the cleanup as a pipeline: each stage strips, regroups, or recomputes one thing, and you compare the result against the question that motivated the table in the first place.
Practical application
-
Re-read the title and footnotes first. Does the table cover the United Kingdom, Great Britain, or just England and Wales? Are part-time workers in or out? The body is uninterpretable without these definitions.
-
Decide what comparison the reader needs. If group sizes differ wildly, raw counts mislead — convert to rates per 1000, percentages, or per-capita figures.
-
Round aggressively. A table about percentage urban population to one decimal is fine for the body; a glance summary in the abstract should round to whole numbers.
-
Reorder rows by the column of interest, biggest first. Alphabetical sort is good for lookup tables; sorted by value is good for storytelling.
-
Group related rows visually. Use whitespace or a faint divider to separate European countries from the rest of the world. Add a median row in italics to anchor the reader.
-
Sanity-check with totals. Row totals and column totals should sum to the same grand total. Mismatched totals reveal data-entry errors before the reader finds them.
Example
Consider a table of road fatalities by country: Germany 2,800; Luxembourg 35; United States 38,000. Read literally, the United States looks 1,000 times more dangerous than Luxembourg. Convert to rate per 100,000 inhabitants and the picture rearranges: Germany 3.4; Luxembourg 5.8; United States 11.6 — Luxembourg jumps ahead of Germany; the United States is dangerous, but only roughly three times more so than Germany rather than fifteen times.
Now sort the table by the rate, biggest first, and add a regional median row. The reader's eye lands first on whatever country tops the list, learns the median in passing, and can place every other country in context within ten seconds. The raw count table required minutes of calculation by the reader to reach the same conclusion — and most readers never bothered.
The principle generalises: whenever you compare across groups of different sizes, the absolute number is a trap and the rate is the truth.
Related lessons
Related concepts
- Precision vs. Accuracylinked concept