Why a skipped survey question is not a 'no', and how to count it right
The most common mistake in reading your own survey results isn't a math error — it's a silent assumption about the denominator. "60% said yes" means nothing until you know 60% of what: everyone you asked, or only the people who bothered to answer that particular question. Those are different numbers, and a blank answer is neither of them — it is not a "no," not a missing "yes," and not something you get to fill in from a pattern in someone else's row.
The method
- For each question separately, count how many people answered it at all. This is the question's own base — not the total number of respondents, because people skip different questions for different reasons.
- Count blanks on their own line. A blank is data: it tells you someone saw the question and didn't answer, which is worth knowing and worth keeping visible, not folding into either side of a yes/no split.
- Compute every percentage over the answered base for that question, not the full respondent count. State the base next to the percentage — "75% of 8 respondents" carries information that "75%" alone throws away.
- For a rating scale (1–5, 0–10), report the full distribution of counts per value, not just a mean. The mean is a convention for summarizing an order, not a measured quantity, since the gap between a 1 and a 2 isn't established to equal the gap between a 4 and a 5.
- Free-text answers get listed, not scored or grouped — counting and tallying is mechanical, but judging what people meant in their own words is a human job a tally can't shortcut.
A worked example
Ten people were sent a two-question survey. Two skipped both questions entirely; eight answered.
"Would you recommend us?" (yes/no) — of the 8 who answered: 6 said yes, 2 said no.
- Yes: 6 ÷ 8 = 0.75 → 75.0% of 8 respondents.
- No: 2 ÷ 8 = 0.25 → 25.0% of 8 respondents.
- Skipped: 2 ÷ 10 (the full group, not the answered base) = 0.20 → 20.0% of 10 sent.
Notice the yes/no percentages are taken over 8, and the skip percentage is taken over 10 — mixing those bases is exactly the error that makes "75% recommend us" quietly misrepresent 6 out of 10 people as something closer to 3 out of 4.
"Rate your satisfaction" (1–5) — the same 8 people answered: 5, 4, 2, 5, 3, 4, 5, 1.
- Distribution: one 1, one 2, one 3, two 4s, three 5s.
- Sum: 5+4+2+5+3+4+5+1 = 29. Mean: 29 ÷ 8 = 3.62.
The mean of 3.62 is a real number, correctly computed — but it hides that three of the eight answers were a 5 and one was a 1. "Average satisfaction 3.62" and "three people are delighted, one is unhappy, four are in the middle" describe the same eight numbers very differently, and only one of those descriptions tells you there's a problem worth finding.
What this doesn't tell you
None of this says whether the eight respondents represent the other two who skipped, or anyone who was never asked at all. A tally counts what's in front of it and stops there; any claim about people who didn't answer has to come from somewhere else, with the reasoning shown, not from the percentage alone. The Survey Builder & Results Tally builds the printable and plain-text survey plus a CSV header row to collect against, then runs this exact base-per-question counting on whatever responses you paste back in — blanks kept on their own line, every percentage labeled with the count it was taken over.