How to visualise survey results without losing the answers

The first question is not “which chart?” but “what kind of answer did the question produce?”. A single-choice question, a multi-select question and a five-point rating scale need different treatment, even when they sit in the same report.

Single-choice questions

Shares add up to 100%, so a 100% stacked bar shows the mix directly and compares across groups. A bar chart of counts answers a different question — how many people — and is the better choice when the size of the groups matters more than the mix.

Keep “don’t know” and “prefer not to say” as visible categories. Hiding them changes the denominator and makes the percentages look more decisive than the data is.

Multi-select questions

When people can choose several options, the percentages do not add to 100% — they add to the average number of choices per person. A pie chart is therefore wrong by construction. Use a bar chart of “% of respondents who selected this option”, and state the base (n = …) in the title or subtitle.

Ratings and Likert scales

A five-point scale is ordered, and the middle is a real answer, not missing data. Show the distribution — a 100% stacked bar across “strongly disagree” to “strongly agree” — and keep the category order the same in every chart so the eye can compare.

The average of a rating scale is easy to compute and easy to misread. “3.8 out of 5” can come from almost everyone choosing 4, or from half choosing 5 and half choosing 2.5 — and those are different findings. If you report an average, show the distribution next to it; if you report a distribution, say what the scale was.

Do not collapse “neutral” into “positive” or drop it without saying so. Both change the story more than they change the chart.

Net Promoter Score

NPS is a single number: the percentage of promoters (9–10) minus the percentage of detractors (0–6). Passives (7–8) do not count. The score is useful for tracking over time; it is not a description of the answers. Show the three groups’ distribution alongside the score so the reader can see what produced it.

With small samples, a few points of NPS difference is noise. Under roughly 100 responses, treat the score as a rough signal, not a ranking.

Small samples

• Show n. “62% said yes (n = 13)” tells the reader something very different from “62% (n = 1,300)”.
• Be careful with subgroups. Splitting 100 responses by region can leave four groups of 25, where one person moves the percentage by four points.
• Do not rank on tiny differences. If two options differ by two percentage points with n = 40, the order may not mean anything.

Open-text answers

Code the responses into a small number of themes, count them, and show the counts with representative quotes. Report how many responses were uncoded. A word cloud looks impressive and communicates almost nothing about frequency or importance.

Common mistakes

• A pie chart for eight options — angles stop being comparable.
• An average rating with no distribution behind it.
• Percentages without the base size.
• Dropping “neutral” or “don’t know” without saying so.
• Comparing subgroups with very different sample sizes as if they were equally reliable.

If you want to build the charts from your own response table, the Excel-to-chart guide explains the layout, and any chart on the chart directory can be produced from a pasted table.