Coding Frequency Counts: When Quantifying Qualitative Data Helps (and When It Misleads) (2026)

"This theme appeared in 34% of responses" feels more rigorous than "this theme was common" - a precise-sounding number carries an authority a vague adjective doesn't. That added precision is only trustworthy to the extent the number is actually measuring what it appears to measure, and coding frequency counts get misread more often than most people producing them realize, in ways that don't require any dishonesty - just an unexamined assumption about what a count of coded responses actually represents.

Table of Contents¶

1. What a Frequency Count Actually Adds
2. The Denominator Problem
3. Frequency Isn't the Same as Importance
4. Double-Counting in Multi-Coded Data
5. When to Quantify and When to Resist
6. A Worked Example
7. FAQ

What a Frequency Count Actually Adds¶

A frequency count genuinely earns its place when a reader needs to compare the relative prevalence of themes against each other - is pricing a bigger concern than onboarding, has a theme grown or shrunk between two waves, does one segment mention a theme more than another. None of these comparisons are meaningfully possible with qualitative description alone; "several respondents mentioned pricing" and "many respondents mentioned onboarding" don't actually tell a reader which is more common, while 31% and 18% do. This is real, legitimate value, and it's exactly why frequency counts have become standard practice even within thematic analysis, which is fundamentally an interpretive rather than counting-based method to begin with.

The Denominator Problem¶

The single most common way a frequency count misleads is an unclear or shifting denominator - what, exactly, is the percentage a share of. "34% mentioned pricing" could mean 34% of everyone who answered the survey, 34% of everyone who left a substantive open-ended comment (a meaningfully smaller, different group), or 34% of responses that were actually coded into any category at all (excluding vague or off-topic responses entirely). Each of these produces a different, defensible-looking number from the same underlying dataset, and a reader has no way to tell which one they're looking at unless it's stated explicitly. Stating the denominator plainly - "34% of the 240 substantive comments" rather than just "34%" - is a small addition that closes a real, easy-to-fall-into gap between what a number appears to claim and what it actually measures.

Frequency Isn't the Same as Importance¶

A theme mentioned by 8% of respondents isn't automatically less important than one mentioned by 40%, and treating frequency as a direct proxy for importance is one of the most common ways a quantified qualitative finding misleads a reader into the wrong priority. A rare but severe safety concern, a specific and highly actionable piece of product feedback, or an emerging complaint that's small today but growing wave over wave can all deserve more attention than their raw frequency alone would suggest - the same point covered from a quote-selection angle in our guide on choosing quotes without cherry-picking, here applied to the count itself rather than the illustrative example. A frequency table is a useful input to prioritization, not a substitute for it - it needs to sit alongside a judgment about severity, actionability, and trend, not stand alone as the only signal driving what gets acted on.

Double-Counting in Multi-Coded Data¶

When a coding scheme allows a single response to be coded into more than one theme - a deliberate, often appropriate choice covered from a survey-analytics angle in our guide on multi-label vs. single-label classification - the resulting frequency percentages across all themes will sum to more than 100%, and that's expected, not an error. The mistake is failing to disclose this and letting a reader assume, by habit, that the percentages should sum to a clean 100% the way a single-choice survey question's would. A brief note stating that responses could be coded into more than one theme, and that percentages therefore don't sum to 100%, prevents a reader from either being confused by the arithmetic or, worse, silently "correcting" it in their own head by assuming the numbers must represent something they don't.

When to Quantify and When to Resist¶

Quantifying is worth doing whenever a genuine comparison is the point - across themes, across segments, across time. It's worth resisting, or at least handling carefully, in a few specific situations: when the underlying sample is small enough that a percentage implies more precision than the data actually supports (a theme appearing in "3 of 12 responses" is more honestly reported as a raw count than dressed up as "25%," which sounds more authoritative than the tiny underlying sample deserves); when the coding scheme itself is still evolving or under-tested, since a frequency count computed from an unstable codebook is measuring something that's still shifting; and when the finding's real value is in its specific content rather than its prevalence - a single, highly detailed, actionable piece of feedback doesn't need a frequency count attached to justify its inclusion in a report.

A Worked Example¶

A nonprofit's program evaluation team codes 85 open-ended responses from a participant survey and reports that "42% of participants mentioned wanting more one-on-one support." Before finalizing the report, a colleague asks what the 42% is a share of - the 85 total respondents, or a smaller subset who left a substantive comment at all. It turns out 20 of the 85 responses were blank or non-substantive, meaning the 42% figure was actually calculated against the 65 substantive responses, not the full 85 - a real, defensible choice, but one the original draft hadn't stated, leaving a reader to reasonably (and incorrectly) assume it referred to the full sample. The revised report states both numbers explicitly - "42% of the 65 substantive responses (35% of all 85 participants surveyed)" - giving the reader an accurate, unambiguous picture instead of a single number that looked precise while leaving its actual basis unstated.

FAQ¶

Should every qualitative finding include a frequency count?
No - reserve quantification for findings where a comparison (across themes, segments, or time) is actually the point. A single, specific, highly actionable finding often doesn't need a percentage attached to justify its inclusion.

How small a sample is too small to report as a percentage?
There's no universal cutoff, but below roughly 20-30 coded responses, reporting a raw count ("4 of 18 responses") alongside or instead of a percentage is more honest than a percentage alone, which can imply more precision than a small sample supports.

Is it dishonest to report percentages that sum to more than 100%?
No, as long as it's disclosed - multi-coded responses naturally produce this, and the fix is a brief explanatory note, not avoiding multi-coding or forcing responses into artificial single categories just to make the arithmetic look cleaner.

Should frequency alone ever determine what gets acted on first?
Not on its own - pair frequency with a judgment about severity, actionability, and trend direction. A rare but severe or rapidly growing theme can reasonably outrank a common but low-stakes one.

For related guidance, see Content Analysis vs. Thematic Analysis: What's Actually Different and Multi-Label vs. Single-Label Classification.