Data Visualisation and Chart Selection

Which chart type should I use for my data?

Start from the question rather than the data. Comparison between categories goes to a bar chart, ranking to an ordered bar, change over time to a line, distribution to a histogram or box plot, relationship between two measures to a scatter, composition to a stacked or divided bar, a sequence with drop-off to a flow or funnel, segment differences across many measures to small multiples, qualitative themes to a thematic framework with participant counts, and relationships that are not measured to a labelled conceptual diagram. The underlying principle is perceptual: encode the comparison that carries your finding in the most accurately read channel available, which is position, then length, then angle, then area, then colour intensity.

Are pie charts bad?

They are usually the wrong choice and should never be the default. Angle and area are read poorly, comparing slices is unreliable, comparing two pies is worse, and more than about four slices requires a legend, which defeats the purpose. A pie is acceptable in a narrow set of cases: two or three categories, a genuine part-to-whole relationship, where the point is an approximate magnitude such as "about half" rather than a precise comparison, and where the audience's convention expects one. If the finding is a ranking, a pie is never correct. An ordered bar chart answers the same question better.

When is it acceptable to truncate a chart axis?

It depends on the encoding, not on a blanket rule. Bar charts must start at zero without exception, because length encodes value and a truncated bar shows a false ratio. Line charts may start above zero, because position rather than length encodes value, and forcing a line to zero can flatten meaningful variation into nothing. The test to apply is whether the reader would draw the same conclusion from the untruncated version. If not, the truncation is doing the arguing, and it is not defensible. Where you do truncate, make it visible in the axis labels and state the range.

Which chart types are misleading?

Four recur. Dual axes, where the apparent relationship between two series is created entirely by the scaling choices and can be manufactured or destroyed by moving either axis. Three-dimensional effects, where perspective makes near elements read larger. Circles or icons sized by radius rather than area, which exaggerate a ratio quadratically. And comparisons of middle or top segments across stacked bars, where only the bottom segment shares a baseline, so every other comparison is unreliable.

Should I put base sizes on charts?

Always, on the chart itself rather than in the surrounding text. Charts are the most extracted element of any report: they end up on slides, in emails and in other people's decks without the paragraph that qualified them. A chart that has lost its base has lost its evidence status. Show n per category where bases vary, flag any base below 100, and do not chart percentages at all on a base below 30, where counts or verbatim are the honest form.

How do I make charts accessible?

Never rely on colour alone to distinguish anything: add direct labels, position or pattern. Check every chart in greyscale, which simulates the worst case and catches most failures. Avoid palettes that separate only in the red and green channels. Keep contrast adequate between adjacent categories and against the background. Around one in twelve men has some form of colour vision deficiency, so in any audience of size this is a certainty rather than an edge case, and designing for it costs nothing and improves the chart for everyone.

What should a chart title say?

The finding, not the variables. "Satisfaction by service issue experience" is a label and leaves the reader to work out what the chart says. "Satisfaction falls sharply after a customer's first service issue" is a claim, and a claim can be checked and disagreed with. The title carries the same discipline as a report headline: it must be true of that chart and nothing more, with no causation the design does not license, no behaviour inferred from stated preference, and no change over time in a single-wave study.

How many charts should a research report have?

One per finding that needs one, and no more. Findings that read faster as a sentence get a sentence. A useful discipline is to write the question and the claim for every proposed chart before building any of them: in most sets, a third have neither and can be deleted before anyone spends time on them. A report with six charts that each answer a question is stronger than one with twenty that display data, and it is faster to check.

How do I visualise qualitative research findings?

Use a thematic framework showing themes with participant counts, illustrative evidence attached, and prevalence stated as counts rather than percentages, because qualitative bases do not support percentages. Use a journey or process diagram where the sequence is the finding. Where a diagram shows relationships that were not measured, label it explicitly as the analyst's model, so a reader cannot mistake a synthesis for a measurement. The failure mode to avoid is borrowing the visual grammar of data charts, which implies a precision qualitative work does not claim.

What is the difference between chart selection and report design?

Chart selection decides what is encoded and how: the question, the form, the ordering, the scale, the labelling, the base. Report design decides how it looks within a house system: typography, palette definition, grid, template and layout. A beautiful chart that is wrongly encoded is a selection failure, and a correctly encoded chart in the wrong typeface is a design one. Get the encoding right first, because no amount of styling fixes a chart that claims more than the data.