Sopact is a technology based social enterprise committed to helping organizations measure impact by directly involving their stakeholders.
Copyright 2015-2026 © sopact. All rights reserved.
Qualitative data is the non-numerical evidence from words, images, and observation. Definition, four types, six characteristics, and eight worked examples.
Qualitative data is non-numerical evidence — words, images, observations, and narratives that capture experience, context, and meaning. A participant's written reason for an NPS score is qualitative data. An interview transcript is qualitative data. A focus group recording, a field journal, a grant report, and a written reflection are all qualitative data. So is a photograph from a service-delivery setting and a voice memo from a researcher in the field. What unites the category is form, not subject: meaning lives in the words and observation rather than in the count.
Eight fragments, eight contexts, one category. Each one captures something a number could not record on its own. The work of qualitative data is to turn this kind of evidence into pattern data the team can act on — without losing the verbatim source that made it specific.
Qualitative data sorts into four practical categories. Naming the type helps with three downstream decisions: which method captures it best, which instrument the participant interacts with, and how it should be coded when analysis begins.
Stories, written accounts, and personal reflections. The participant is the author. Narrative data captures motivation, sequence, and turning points — the parts of an experience the participant decides to foreground when they tell it.
Open-ended responses describing experience, opinion, or context. Shorter than narrative, more focused than observational. Often paired with a structured rating that gives the description its anchor.
Field notes, behavioral records, and conversation logs written by the researcher rather than the participant. Captures action and context — what people do, which is often different from what they say they do.
Pre-existing documents read as evidence. The text already exists for another purpose — the analysis layer is added by reading it through a research framework.
Most applied research uses more than one type in a single study. A workforce evaluation typically pairs narrative (intake essays) with descriptive (mid-program survey reasons) and observational (program manager field notes). A foundation portfolio review pairs documentary (grantee reports) with descriptive (grantee survey responses). The type does not determine the value — every type, well-coded against a defined framework, contributes pattern data the team can act on.
Six properties recur across every type. They are the working tests for whether you are looking at qualitative data, and they shape every downstream decision about how to collect, code, and report it.
Form is the test: words, images, observations, audio — not counts or measurements.
A confidence rating of 4 of 5 is quantitative. The sentence the participant wrote next to the rating — "the labs helped me apply what I learned" — is qualitative. The two travel together but the category boundary runs through the form.
Meaning depends on who said it, when, and in what setting.
"It was hard" said by a first-week trainee is not the same evidence as "it was hard" said by a graduating cohort. Strip the participant and the touchpoint from the response and the meaning becomes unreadable. Context is part of the data, not metadata.
One response can carry layers a single number cannot record.
A single quote — "the labs helped me apply what we covered" — encodes the participant's preferred learning mode, the relationship between lecture and practice, and a signal of growing self-efficacy. The rating that sits next to it records the magnitude. Both together explain each other.
It captures perception, not external measurement.
Qualitative data records how something was experienced, not how something objectively was. That is its strength, not its weakness. The point is to surface the participant's account — and then triangulate across many accounts to find the patterns that hold up.
Analysis requires reading, coding, and theming — not averaging.
Quantitative data goes from collection to descriptive statistics through a sequence of arithmetic operations. Qualitative data goes from collection to findings through a sequence of interpretive operations: read each response, apply the codebook, identify the patterns. The work is different, not lesser.
It surfaces what you didn't know to ask.
A structured survey can only return answers to the questions on the form. A qualitative open-ended response can name a theme nobody at the program had thought of yet. Most root-cause analyses in customer retention, employee attrition, and program drop-out start with a qualitative response that broke an assumption.
The two categories are not in competition. They answer different questions and most strong studies use both — ideally on one record per participant so a rating and the reason behind it travel together.
Every card below is a real shape of qualitative data drawn from a real research context. Each card shows the raw form, names what the response captures, and shows what the cross-participant analysis surfaces when this kind of data accumulates across a population.
Context: A 12-week trades training program runs a 40-minute intake interview with each of 200 new participants per cohort. The interview asks about goals, prior experience, and anticipated barriers.
Context: A B2B SaaS platform sends a monthly NPS survey to 3,000 customers. Every score is followed by one open-ended question: "What's the main reason for your score?"
Context: A community health nonprofit embeds a researcher in a clinic for one quarter. The researcher records field notes from morning intake.
Context: A foundation with 120 active grantees collects three narrative reports per grantee per year. Each report is coded against the funder's outcome framework.
Context: A 600-employee company runs structured exit interviews with each voluntary leaver. Interviews are transcribed and coded against an attrition framework.
Context: A scholarship program reviews 500 application essays each cycle. Essays are scored against a four-dimension rubric — academic fit, motivation, financial need, community contribution.
Context: A clinical research team collects patient-reported experience narratives across a six-week treatment arm, paired with weekly pain ratings.
Context: A community health program runs an exit focus group with each cohort. Eight clients reflect on the service in a 90-minute moderated session.
Qualitative data does not arrive as evidence. It arrives as a quote, a paragraph, or a field note — and a series of interpretive operations turns it into the patterns the report can defend. The three panels below trace what each step produces, using one workforce-training response as the running example.
No coding yet. The raw response carries meaning but no pattern. It sits next to the participant's confidence rating on the same record.
The codebook is the framework. Each tag here is anchored to the program's theory of change so it is comparable across cohorts and over time.
Across 200 mid-program survey responses, 68 percent of participants attributed their gain in confidence to the hands-on lab sessions specifically, with 41 percent describing a shift from passive recall to self-directed problem-solving. The APPLY theme was the single largest driver of positive sentiment in this cohort.
Pattern and verbatim quote travel together. The cohort-level claim is defensible because every quote clicks back to the response that produced it.
Three stages, one piece of evidence. The transformation works at the scale of a single response or at the scale of two hundred. The architectural property that makes it work at scale is that every coded fragment retains its link back to the source — so cohort claims have verbatim quotes underneath them, and verbatim quotes are not cherry-picked because their cohort-level frequency is right there next to them.
Most qualitative data problems are not analysis problems. They are setup problems that turned into analysis problems by the time anyone noticed. The six cards below name the failures that come up most often, with the substitute that prevents each one.
Quotes get pulled into the deck without the cohort-level pattern underneath them. A vivid quote becomes the evidence. The pattern that would have either backed it up or contradicted it never gets surfaced.
Run the codebook across every response. Report the pattern frequency next to the verbatim quote — a quote is illustration, not proof.
Open-ended responses arrive in a spreadsheet without a participant ID, without demographics, without a session reference. The response exists; the context does not. Disaggregation by site, gender, cohort, or plan tier becomes impossible.
Assign a persistent participant ID at first contact. Carry it through every later touchpoint. Demographics go on the record at intake, not retrofitted later.
The team reads the responses as they come in, gets a general sense, and never applies a systematic framework. The general sense is the analysis. The report writes itself out of recent memory, not out of the data.
Write the codebook before collection. Apply it to every response. The codebook is the analysis prompt, not a post-hoc cleanup pass.
Ratings live in SurveyMonkey. Open-ended responses live in a second tool. Demographics live in the CRM. The connection between a rating and the reason behind it has to be reassembled by hand for every report.
Pair the rating and the reason on the same form. Keep demographics, ratings, and open-ended responses on one record per participant from intake forward.
The most quotable response goes into the funder report. It is unrepresentative — the cohort actually said something different — but it is on-brand. The quote is doing the work the pattern data should be doing.
Surface the cohort-level theme distribution. Choose the verbatim quote that illustrates the dominant theme, not the most marketable outlier.
The team runs 50 interviews because 50 sounded thorough. The transcripts pile up. The first ten get read in detail. The other forty are summarized in a single sentence. The cost of collection was not budgeted against the cost of analysis.
Match collection volume to analysis capacity. Either reduce the count to what the team can actually code, or use an analysis approach that scales with volume.
Qualitative data is non-numerical evidence: words, images, observations, and narratives that capture experience, context, and meaning. A participant's written reason for an NPS score, an interview transcript, a focus group recording, a field journal, and a grant report all count. Qualitative data answers why and how, where quantitative data answers how many and how much.
Four types cover most qualitative data in practice: narrative data (stories, written accounts, reflections), descriptive data (open-ended responses describing experience or opinion), observational data (field notes, behavioral records, conversation logs), and documentary data (pre-existing documents read as evidence — reports, case notes, applications, journals). Most studies use more than one type.
Common examples include a workforce training participant's written reason for a confidence rating, a B2B customer's open-text response to an NPS survey, a community health field journal, a foundation grantee's annual narrative report, an employee's exit interview transcript, a scholarship application essay, a clinician's patient observation notes, and a focus group recording with a program cohort. Every one of these carries meaning that a number alone cannot.
Six characteristics distinguish qualitative data. It is non-numerical (words, images, observations rather than counts). It is contextual (meaning depends on who, when, and where). It is rich and dense (one response can carry layers a single number cannot). It is subjective by nature (it captures perception). It is interpretive (analysis requires reading, coding, and theming). And it is exploratory (it surfaces what you did not know to ask).
Quantitative data gathers numbers — counts, ratings, scores, measurements. It answers how many and how much. Qualitative data gathers words, images, and narratives. It answers why and how. They are not in competition: most strong studies pair them in the same instrument so a rating and the reason behind it sit on the same record per participant. The pairing produces evidence neither method gives alone.
Qualitative data is described by what it captures, how it was collected, and what it surfaced. A complete description names the method (interview, focus group, open-ended survey, observation, document analysis), the source (who provided it and in what context), the codebook (the framework used to identify themes), and the patterns the analysis produced. Verbatim quotes belong in the description alongside the pattern-level findings.
Qualitative data means evidence captured in non-numerical form — typically in words, but also in images, sound, and observation. The meaning the data carries is interpretive: it surfaces experience, context, and reasoning rather than measurement. A participant saying transportation was the hardest part of the program is qualitative data; the same participant's attendance rate is quantitative data. The two together explain each other.
Qualitative data is analyzed by reading, coding, and theming. The analyst applies a codebook to each response, assigning theme codes and capturing patterns across responses. Modern workflows apply the codebook automatically as responses arrive, with theme distributions and verbatim quotes produced in the same pass. The output is cross-participant pattern data with citation chains back to the source response, not a stack of unread transcripts.
Qualitative data is evidence in words rather than numbers. It includes interviews, open-ended survey responses, focus group transcripts, field notes, and written documents. It captures meaning, context, and experience — the parts of what is happening that numbers alone cannot describe.
A survey can produce either kind, depending on the question. Multiple-choice and rating questions produce quantitative data — numbers and categories. Open-ended questions produce qualitative data — written reasons, descriptions, and narratives. Most useful surveys combine both: a structured rating immediately followed by an open-ended question that captures the reason behind the rating.
Three tests. First, can you read it? If it is words, images, or recorded behavior rather than numbers, it is qualitative. Second, does meaning depend on context? If removing the participant or setting changes what the data means, it is qualitative. Third, does it require interpretation? If you have to read, code, and theme it to produce findings — rather than count, average, or correlate — it is qualitative.
A twenty-minute working session takes a real sample of your qualitative data — interview transcripts, open-ended survey responses, or grantee report excerpts — and shows what theming on arrival looks like against a codebook you define. No procurement decision required. The point is to see how the work changes when collection and analysis share one record.