Grounded Theory for Open-Text Course Feedback: Building Explanation, Not Just Themes

In brief

Grounded theory (Glaser & Strauss, 1967) is a qualitative methodology for building an explanatory theory from data rather than testing a theory imposed on it. Applied to open-text course feedback, it moves you past a list of themes ("workload", "clarity") toward a model of the process that generates student reactions — for example, how a mismatch between assessment and teaching escalates into disengagement. Its three engines are constant comparison, theoretical sampling, and theoretical saturation. Used well it is the most powerful method for answering "why", but it is labour-intensive, demands methodological discipline, and is easy to do badly.

What the research says

Grounded theory originated with sociologists Barney Glaser and Anselm Strauss in The Discovery of Grounded Theory (1967), written as a corrective to a social science they saw as obsessed with testing grand theories and neglecting the generation of new ones from systematic fieldwork. Their claim was that rigorous theory could be induced from qualitative data through a disciplined procedure, not merely guessed at.

Three interlocking procedures define the method:

• Constant comparison. Every new datum is compared with existing data, codes, and emerging categories. A comment is coded, then the next comment is asked "is this the same phenomenon or a different one?" Categories are refined, split, or merged as comparison continues. This is what stops coding from collapsing into a static, pre-set frame.
• Theoretical sampling. Later data collection is directed by the emerging theory. Rather than fixing the whole sample in advance, the analyst notices a gap — "we do not understand why high achievers react differently to the group project" — and deliberately seeks data that illuminate it. Sampling chases concepts, not demographic quotas.
• Theoretical saturation. Sampling and analysis continue until new data stop yielding new properties of the categories. Saturation, not a target sample size, is the stopping rule.

Coding proceeds in layers — open coding (fracturing the data into concepts), axial or focused coding (relating concepts into categories), and selective coding (integrating around a core category that explains most of the variation). Memo-writing throughout captures the analyst's developing theoretical reasoning.

The method later split into schools that a careful QA reader should not conflate. Glaser retained the classic, more positivist emphasis on emergence and minimal preconception. Strauss and Corbin (1990) introduced a more prescriptive coding paradigm (conditions, actions, consequences), which Glaser publicly rejected. Kathy Charmaz's constructivist grounded theory (2006) reframed the whole enterprise: categories are constructed by a situated analyst interacting with participants, not "discovered" lying in the data. These are genuinely different epistemologies, and a study should declare which it follows.

Why it matters for course evaluation in practice

Most institutional open-text analysis stops at thematic description: count the comments, group them into buckets, report the biggest buckets. That answers what students mention. It rarely answers why a course produced the reaction it did — which is the question a programme team actually needs to act.

Grounded theory is the qualitative method purpose-built for that "why". Its output is not a bar chart of theme frequencies but a small, connected model: a core category and the conditions under which it operates. For example, a grounded-theory reading of comments across a struggling module might yield a process such as "assessment ambiguity → students substitute effort for understanding → a good grade feels unearned → the student discredits the whole course", where each link is grounded in specific quotations. That is a testable, actionable account. Thematic frequency counts would have shown "assessment" and "grades" as separate large buckets and missed the mechanism connecting them.

The distinction from the methods your institution may already use is real and worth stating:

• Thematic analysis describes patterns across a fixed dataset; it does not require theoretical sampling or an explanatory core category.
• Framework analysis applies a matrix, often with some a-priori categories — efficient for policy questions, but less generative.
• Phenomenography maps the qualitatively different ways students experience a phenomenon, not the process that produces an outcome.

Grounded theory is the right tool specifically when you have a puzzling, recurring pattern and need an explanation robust enough to redesign a course around.

Limitations and honest caveats

Grounded theory is frequently invoked and rarely done properly, and a sophisticated reader will be sceptical for good reasons.

• "Grounded theory lite." Many published studies cite Glaser and Strauss but in practice do ordinary thematic coding with grounded-theory vocabulary bolted on. Without genuine theoretical sampling and iteration to saturation, it is not grounded theory; calling it that overclaims rigour.
• Theoretical sampling is hard to honour on archived survey data. Classic grounded theory assumes you can go back and gather more, targeted data. End-of-term evaluation text is a fixed corpus — you cannot re-interview last term's cohort. This is a fundamental tension: with static data you can approximate the logic (comparing across strata you already hold, planning next term's targeted follow-up) but you cannot fully practise the method as designed.
• Saturation is contested. "No new codes emerged" is a judgement, not a measurement, and it depends on how hard you looked. Claims of saturation should specify the sampling and the analytic effort behind them.
• The role of prior literature divides the field. Classic grounded theory counsels delaying the literature review to avoid forcing preconceived categories; critics (and constructivists) note that no analyst is a blank slate, and that pretending to be one is itself a bias.
• Reflexivity and generalizability. Constructivist grounded theory concedes the analyst shapes the result, which raises transferability questions. A grounded theory built from one department's comments is a hypothesis about others, not a finding that automatically travels.

None of this disqualifies the method; it means the method must be reported honestly — which school, how sampling was directed, what saturation meant here, and who did the interpreting.

How Koji incorporates this

Koji is designed to support a grounded-theory-style analysis of open text, not to replace the analyst's judgement. Several mechanisms map directly onto the method's engines:

• Depth data worth theorising. Grounded theory is only as good as its data, and a single Likert number gives it nothing to work with. Koji's AI-moderated conversational interviews probe beyond the rating — following up on a vague comment to surface the process behind it ("you said the deadlines felt unfair — walk me through what happened"). This produces the rich, incident-level text from which a genuine explanatory account can be built.
• Constant comparison at scale. Koji's automatic thematic analysis provides a first-pass code structure across the full corpus, and its interface lets a human analyst compare coded segments side by side — the practical substrate for constant comparison, with the analyst refining, splitting, and merging categories rather than starting from a blank page.
• Directed follow-up as approximate theoretical sampling. Because Koji runs mid-cycle and formative collection, a gap the analyst notices this term ("we do not understand the high-achiever reaction") can become a targeted probe next cycle — the closest honest analogue to theoretical sampling that recurring institutional evaluation allows.
• Triangulation and traceability. Every emerging category can be traced back to the specific transcript segments that ground it, and cross-checked against structured items (scale, single_choice, ranking) for convergence, keeping interpretation anchored to evidence.

We frame these as support for disciplined qualitative work, not automation of it: the AI accelerates coding and comparison, but declaring a core category and claiming saturation remain human, methodological acts. Koji's core research platform at koji.so applies the same AI-moderated interview engine to product and customer research, where grounded-theory analysis of user interviews is a long-established practice.

Frequently asked questions

Is grounded theory the same as thematic analysis? No. Thematic analysis describes patterns in a fixed dataset. Grounded theory adds theoretical sampling (data collection directed by the emerging theory), iteration to saturation, and — crucially — an explanatory core category. Its goal is a process model of why, not a catalogue of what.

Can I do grounded theory on last term's end-of-course comments? Partially. You can apply constant comparison and layered coding to a fixed corpus, but you cannot fully practise theoretical sampling, which assumes you can gather more targeted data. Be honest about this: treat the analysis as generating hypotheses and plan directed follow-up in the next cycle.

Which "school" of grounded theory should we use? Declare one. Classic (Glaser) emphasises emergence and minimal preconception; Straussian (Strauss & Corbin) adds a structured coding paradigm; constructivist (Charmaz) treats categories as co-constructed by a situated analyst. They rest on different epistemologies, so mixing them silently undermines rigour.

What is theoretical saturation, and how do we know we have reached it? Saturation is the point at which new data stop revealing new properties of your categories. It is a reasoned judgement, not a number. Report the sampling and analytic effort behind the claim rather than asserting "saturation reached" on the basis of a fixed comment count.

Does grounded theory replace our Likert-scale items? No — it complements them. Numbers tell you a course scores low; grounded theory explains the process producing that score. The strongest quality cycles triangulate an explanatory qualitative account with structured quantitative evidence.

Isn't this too labour-intensive for routine evaluation? Full grounded theory is not for every module every term. Reserve it for puzzling, recurring problems where you need an explanation robust enough to redesign around. Tooling that pre-codes and organises the text lowers the cost considerably.

Related resources

• The Framework Method for Open-Text Course Feedback
• How Reliable Is Your Coding of Open-Text Feedback? Inter-Rater Reliability
• How Many Open-Text Comments Are Enough? Thematic Saturation
• The Qualitatively Different Ways Students Experience Your Course: Phenomenography
• The Critical Incident Technique for Course Feedback
• What Can Open-Text Student Comments Tell You That Likert Scores Cannot?

References

• Glaser, B. G., & Strauss, A. L. (1967). The Discovery of Grounded Theory: Strategies for Qualitative Research. Chicago: Aldine.
• Strauss, A., & Corbin, J. (1990). Basics of Qualitative Research: Grounded Theory Procedures and Techniques. Newbury Park, CA: Sage.
• Charmaz, K. (2006). Constructing Grounded Theory: A Practical Guide Through Qualitative Analysis. London: Sage.
• Glaser, B. G. (1978). Theoretical Sensitivity: Advances in the Methodology of Grounded Theory. Mill Valley, CA: Sociology Press.
• Mills, J., Bonner, A., & Francis, K. (2006). The development of constructivist grounded theory. International Journal of Qualitative Methods, 5(1), 25-35. https://doi.org/10.1177/160940690600500103