Your radar chart is lying by about 15%: assessment report design that people act on

Nobody asks whether the radar chart works. They ask how many axes it should have, pick eight because the reference picture had eight, and ship it. Then nobody acts on the report and the conclusion is that people don't read long reports.

People read long reports. They don't read shapes. A radar chart is an index, not an argument, and most assessment report design treats it as the whole deliverable. Here is what the geometry does to your scores, and what has to sit underneath the picture before anyone acts on it.

The shape is doing the talking, and the shape exaggerates

A radar chart encodes score as radius. Your eye reads area. Area scales with the square of the radius, so the two disagree on every single axis.

Score a dimension 4 out of 5 instead of 5 and it renders with 0.64 of the area — a 20% gap that looks like a 36% gap. Every weak dimension gets punished harder than it earned, and the reader's felt sense of their own report runs more negative than their numbers justify.

The axis ordering problem is worse, because it is invisible. Take five dimensions scoring 80, 30, 75, 35 and 70 on a 100-point scale. Those numbers are fixed. Rotate which axis sits where and the enclosed area swings about 15% between the worst and best arrangement — and sorting highest-to-lowest instead of leaving model order inflates the polygon roughly 12%. Nobody changed an answer. The chart just got bigger.

Two rules fall out of that:

• Never let two radar charts be compared by size. Not respondent against peer, not this quarter against last. If you want a comparison, put the numbers in a table where a 5-point gain reads as 5 points.
• Fix the axis order and write down why. Model order or alphabetical, decided once, identical on every report. Sorting by score is a silent inflation that also destroys the reader's ability to find the same axis twice.

None of this means drop the chart. It means the chart's job is orientation — here are your dimensions, here is roughly where the dip is — and nothing more.

How many dimensions before the shape stops meaning anything

A dimension needs at least 3 questions before its score is anything but noise, and 4 if the questions are self-reported. That multiplier is the real constraint, not the drawing.

The length column is arithmetic, not a benchmark — 12 seconds per single-select question is a placeholder. Substitute your own measured pace and rerun the multiplication; the ranking of the rows won't change.

Five to seven is the working range. Below five, bars communicate better and cost you nothing. Above seven, either cut or nest: group twelve sub-dimensions into five parent axes, chart the parents, and let the sub-scores live in the table underneath.

The honest exception is a category set your audience already knows cold. The classic eight-spoke life wheel survives the clutter because nobody has to learn what "Health" or "Finances" means — that's the structure behind the Wheel of Life Scorecard, and it works because none of the reader's attention goes to decoding labels. If your axes each need a sentence of explanation, you don't get that exemption.

All of this assumes you want a dimensional scorecard rather than a single score or a result type. If that's still open, settle it first — the three quiz formats and what each is for sits upstream of every choice on this page.

Name the axis after the behavior, not the construct

Construct names are how you think about the model. Behavioral names are how the respondent checks whether the score is fair. Ship the second kind.

The test takes ten seconds per axis. Read the label aloud and ask: if someone disagreed with this score, would they be arguing about their behavior or about what the word means? Arguing about the word means you have a construct, not a measurement, and the report is going to lose that argument.

Practical constraint: the label printed on the chart has to survive a narrow screen, so keep it under about 24 characters — two or three words. The behavioral sentence goes underneath, as the subhead for that dimension. A multi-dimension instrument like the Team Health Scorecard has to carry both: the axis label is what the reader sees, the behavioral sentence is what makes the score arguable.

Where a baseline can honestly come from

The dashed ring is where assessment reports get people into trouble. Four defensible sources, in the order you should reach for them:

1. The respondent's own median across their axes. Always available, n=1, zero fabrication risk. Turns "you scored 62 on Delegation" into "Delegation is your weakest axis by 19 points" — more actionable than any peer comparison, and true the day you launch.
2. Their previous attempt. The strongest baseline that exists, and the reason retakes are worth engineering early. Same instrument, same person, movement you can attribute.
3. Your own measured sample, with n printed next to the line. Don't draw a segment median on fewer than ~100 responses without showing the count, and don't claim a percentile under a few hundred — percentiles imply a distribution shape you don't have yet. The counting discipline is the same one in the quiz funnel benchmark piece.
4. A target you set and defend. "70 is the line for a team past Series A." That's a standard, not data. Label it as a standard, on the chart, in words.

What never goes on the chart is an industry average you did not measure. Finish the sentence "this line is ___, measured on ___ people, on ___." If you can't fill all three blanks, delete the line. Lifting a number off a vendor's marketing page and rendering it as a dashed ring converts their claim into your liability — and it comes back in a sales call eighteen months later as something you have to defend.

Draw the baseline in a different visual language: dashed stroke, no fill. Two filled translucent polygons stacked on one radar is the most common unreadable chart in this category.

What has to sit underneath the chart

The chart orients. The body does the work. Order matters, because attention runs out fast:

1. One sentence naming the shape — highest axis and lowest axis, by name. Not the average. The average of five dimensions is a number that describes nobody.
2. The lowest axis, expanded. What the score means, and which specific answers drove it. Quoting the respondent's own answers back is the single cheapest credibility move available and almost nobody does it.
3. The highest axis, one short paragraph. Explicit permission to stop optimizing it. Reports that only list problems get closed.
4. One action per axis. One. With a field for owner and a date. Three suggestions per axis across six axes is eighteen suggestions, which is zero suggestions.
5. The score table, raw numbers. This is what gets screenshotted and pasted into Slack. The chart doesn't survive copy-paste; the table does.

Resist ranking all seven axes best-to-worst in prose. The reader will absorb two items from that list and it won't be the ones you care about.

The twenty-minute audit

Run this on a report you already shipped:

• Open it at 375px wide. Can you read every axis label without rotating the phone?
• Count questions per dimension. Anything under 3 is decoration with a number attached.
• Is the axis order fixed and documented — or sorted by score?
• Read each label aloud. Construct, or behavior?
• Finish the baseline sentence: "this line is ___, measured on ___ people, on ___." Can't? Delete the line.
• Does the chart repeat underneath as a table of actual numbers?
• Count total recommended actions. More than one per axis?
• Hand the report to someone who didn't take the assessment. Give them 30 seconds and ask what this person should do first.

That last one is the whole test. If a stranger can't name the first action in 30 seconds, the report isn't finished — and adding a prettier chart won't finish it.

Photo by Clint McKoy on Unsplash