How to visualize survey results with the right chart

Most confusing survey charts don't come from bad data. They come from a chart that doesn't fit the question. A pie chart of a checkbox question, an agreement scale sorted by size, or a percentage built on four people can each mislead a reader who trusts what they see.

The fix is to choose the chart from the question type, not from taste. Use sorted bars for single-choice questions and bars that can add past 100% for checkbox questions. Use stacked or diverging bars for agreement scales, and a table with counts when groups are small. A few habits then keep every chart honest.

Start from the question type

Before you open any chart menu, sort your questions by what kind of answer they collect. If you need a refresher on the types, the guide to survey question types covers them. For charting, four situations cover most surveys:

• Single choice: each person picks one answer, so the answers add up to everyone. Use a sorted horizontal bar chart.
• Checkbox (multi-select): each person can pick several, so the answers add up to more than everyone. Use bars showing the percent of respondents, and say that totals pass 100%.
• Agreement and rating scales: the answers have a natural order. Use stacked or diverging bars that keep that order.
• Small groups: whatever the question, use a table with counts once a group gets small.

Single-choice questions: sorted horizontal bars

Horizontal bars are the default for a reason. Answer labels are often long ("Through a friend or colleague"), and a horizontal layout gives them room without tilting text. Bar length is also the easiest visual quantity to compare.

Sort the bars from largest to smallest so the ranking is visible at a glance. Then move catch-all answers such as "Other", "None of these" and "Don't know" to the bottom, whatever their size. They aren't competing with the real options, and putting them in the middle of the ranking muddles it.

Here is a worked example. Suppose 200 people answered this:

The answers were search 74, a friend or colleague 58, social media 41, a podcast 15 and Other 12. That adds up to 200. As sorted bars that's 37%, 29%, 21% and 8%, with Other at 6% sitting last.

When a pie chart is fine, and when it isn't

A pie works when there are two or three slices and the message is "part of a whole". An example is 62% yes and 38% no. Past that, people struggle to compare angles. Five similar slices of 18% to 23% look almost identical in a pie but are easy to rank as bars. If you have more than three answers, use bars.

Checkbox questions: bars that can add up past 100%

Checkbox questions trip up more reports than anything else. You have two possible denominators, and only one answers the question people actually ask.

Say 150 people answered "Which features do you use?" and could tick as many as they liked. The counts were dashboards 105, exports 87, alerts 66, integrations 48 and the mobile app 34. That's 340 selections from 150 people.

• Percent of respondents: dashboards 70% (105 of 150), exports 58%, alerts 44%, integrations 32%, mobile app 23%. These add up to 227%, and that's correct.
• Percent of selections: dashboards 31% (105 of 340). This answers "what share of all ticks went to dashboards", which almost nobody wants to know. Read carelessly, it suggests only 31% of people use dashboards.

Chart the percent of respondents as sorted bars. Then add a note under the chart that explains the total, so no reader thinks your arithmetic is broken:

Never put a checkbox question in a pie. A pie claims its slices make one whole, and here they don't.

Agreement scales: stacked or diverging bars

The answers to "Strongly disagree" through "Strongly agree" are ordinal data. They have an order, but the steps between them aren't guaranteed to be equal. That shapes the chart in three ways.

First, keep the scale order. Don't sort the answer categories by size, because a chart that runs "Agree, Neutral, Strongly agree, Disagree" destroys the meaning. Second, color them as a gradient. Use two directions of one palette for disagree and agree, with a muted gray for the neutral point. Third, be wary of reducing the scale to an average. Picture two statements that both average 3.0 on a five-point scale. One got neutral from everyone, and the other got half "Strongly disagree" and half "Strongly agree". Those are very different findings, and only the full distribution shows it.

Stacked bars for one or two statements

A 100% stacked bar shows the whole distribution in one row, from strongly disagree on the left to strongly agree on the right. It works well for a single item, or for one item compared across two groups.

Diverging bars for many statements

When you have a matrix of eight or ten statements, a diverging bar chart is easier to read. Disagreement extends left of a center line and agreement extends right. Sort the statements by total agreement so the strongest and weakest stand out. Decide on purpose what to do with neutral answers. You can split them evenly across the center line or show them in a separate column. Whichever you choose, use it the same way on every chart.

For more on building and analyzing these scales, see the guide to Likert scales.

Small groups: switch to a table with counts

Percentages imply a stability that small groups don't have. If 3 of a team's 4 respondents agree, that's 75%. One person answering differently would move it 25 points. A bar of 75% next to a bar of 61% from a group of 300 looks like a real difference, and it may be noise.

When you break results down by team, region or plan in a cross-tabulation, show a table once groups get small. Put the counts alongside the percentages, and include the group size in each column header.

This lets readers judge for themselves how much weight a number can carry.

In employee and other sensitive surveys, small groups raise a second problem: people can be identified. Plumeform lets any survey set a minimum group size of 3, 5, 7 or 10. Any result describing fewer people than that is hidden, including a filter to one team, a cross-tab group and shared results links. It hides summaries, not individual responses, which people in the workspace can still open.

Habits that apply to every chart

Show the number of responses

Put the base, such as "n = 212", on every chart, not just once at the top of the report. Optional questions and skip logic mean each question can have a different base. A chart that says 40% without saying 40% of whom invites the wrong reading.

Order categories on purpose

Sort unordered answers by size. Keep ordered answers, such as scales, age bands and frequency, in their natural order. If the same question appears in several charts, such as one per region, fix the category order across all of them so readers compare like with like.

Don't cut axes short

Bar charts encode values as length, so the axis must start at zero. Suppose you compare 52% with 48% on an axis running from 45 to 55. The bars are 7 and 3 units long, and the first looks more than twice the size of the second for a four-point gap. If a small difference matters, state it in words and show its uncertainty. Don't stretch the bars.

Label directly and say what you found

Put values at the end of each bar instead of making readers trace gridlines. Write the chart title as the finding, not the question number.

Putting the report together in Plumeform

In Plumeform, the Report tab is where you turn results into something you can share. You pick the chart for each question, add findings that state what each result means, and attach chart notes. The chart note is the natural home for the "can add up past 100%" line or a caution about a small base. When the report is ready, you can print it or save it as a PDF. You can find an overview on the reports and dashboards page.

Whatever tool you use, the order of work stays the same. Identify the question type, choose the chart that fits it, then check the base, the category order and the axis before anyone else sees it.