Make every percentage easy to explain
For an insights team reporting on a programme, a useful comparison connects a finding to its supporting answers. Instead of presenting “80%” on its own, show that it means eight people out of ten who answered that question. Everyone can then see the scale of the evidence before discussing a change.
Agree the question, reporting period and groups first. Keep each person in one group for this comparison, and count their answer once. State whether a percentage uses everyone who returned the survey or only those who answered that question. This number is its denominator: the total you divide by.
Read the result alongside the people counted
Synthetic demonstration, not client findings: an invented programme team asks, “Was it easy to find the session information?” The options are Yes and No; people may skip the question. Each row below represents a different session group in the same reporting period. Every answer has equal weight.
| Session | Yes / answered | Yes % | Skipped |
|---|---|---|---|
| Morning | 18 / 30 | 60% | 10 |
| Evening | 8 / 10 | 80% | 2 |
| Weekend | 0 / 0 | No answers | 3 |
The evening group has the higher share among those who answered. But changing just one of its ten answers from Yes to No would move 80% to 70%. That is a ten percentage point change. A small group can move sharply with one answer; the table alone cannot establish a reliable difference across all participants.
Skipped questions stay visible and do not count as No. For the morning group, dividing 18 by all 40 survey returns gives 45%. That answers a different question: the share of all returns containing Yes. Calling it “the share of people who answered this question” would be wrong. With no Yes or No answers in the weekend group, a percentage is undefined, not 0%.
Turn the comparison into useful follow-up
These invented records illustrate the calculation only. In a real voluntary survey, people who respond may differ from those who do not. A higher percentage does not show that the session caused a better experience. Missing answers may also change the picture. The AAPOR survey guidance covers how people are chosen to take part, missing responses and clear reporting.
Use the annotated cross-tab and synthetic source records to trace the counts and the no-answer case. Before sharing a real comparison, have your method reviewer check who was included, the missing answers and what conclusions the study supports. Ask your reviewer whether further calculations are needed to judge a group difference or describe how uncertain it is. This example includes no such calculations.
Datimore can help build a report that keeps each comparison, its answer counts and review notes together. Our community insight case shows information brought together for review; it does not prove this survey method. Bring one survey question, your group definitions and the intended use of the report. The aim is a discussion where the team can explain the finding and choose a sensible follow-up.