Thematic, Content and Framework Analysis

Introduction and context

Qualitative analysis is the disciplined process of moving from raw language, observations, documents, or interactional data to an interpretation that is credible, transparent, and useful. In health research, the data may include interviews with patients, focus groups with caregivers, observations of clinical work, policy documents, complaint narratives, or digital communications. The analyst’s task is not simply to identify interesting quotations. It is to establish how meanings, experiences, practices, structures, and consequences are connected.

Thematic analysis, qualitative content analysis, and framework analysis are related but distinct approaches. All involve systematic coding and interpretation, yet they differ in their historical traditions, degree of pre-structuring, treatment of manifest and latent meaning, and preferred outputs. A strong analyst chooses an approach that fits the research question, epistemological position, data, and intended use of findings rather than treating methods as interchangeable software procedures.

In health-care management, these methods support analysis of service quality, implementation, patient safety, workforce experience, access, equity, leadership, and organizational change. They can reveal why a technically effective intervention is not adopted, how patients interpret “access,” or how formal policy differs from actual practice.

Learning objectives

By the end of this chapter, the learner should be able to:

• distinguish thematic, qualitative content, and framework analysis;
• conduct inductive, deductive, and combined coding;
• use constant comparison, analytic memos, matrices, and visualizations;
• develop, refine, and define themes defensibly;
• identify manifest and latent content;
• undertake negative-case analysis and explain divergent findings;
• manage ethical, governance, and technological issues in qualitative analysis; and
• produce an auditable interpretation for health-care decision-making.

The analytic logic of qualitative interpretation

Analysis usually proceeds through a series of recursive activities rather than a straight line:

1. Familiarization: reading or listening repeatedly, checking transcripts, and noting initial impressions.
2. Segmentation: identifying meaningful passages in relation to the research question.
3. Coding: assigning concise labels to data segments.
4. Comparison: examining similarities, differences, exceptions, and relationships across cases.
5. Categorization: grouping related codes into broader analytic concepts.
6. Theme development: constructing patterned meanings that answer the research question.
7. Interpretation: relating patterns to context, theory, power, process, and consequences.
8. Validation and reporting: testing the interpretation against the dataset and documenting decisions.

These activities overlap. New readings can alter codes; a negative case can require a theme to be narrowed; and a matrix can reveal a pattern that was not visible during line-by-line coding.

The analyst should distinguish three levels of interpretation:

• Data extract: what a participant or document literally says.
• Code: what the extract is about in a concise analytic label.
• Theme or category: a patterned, interpretive account that explains significance across multiple extracts.

For example, the statement “I had to explain my history at every appointment” might receive the code repeatedly recounting clinical history. A broader category may be fragmented continuity. A theme might be Patients perform the coordination work created by disconnected services. The theme is not merely a topic; it makes an interpretive claim.

Reflexivity and the analytic position

Qualitative analysis is interpretive, even when a team uses a structured codebook. Researchers bring assumptions about illness, professionalism, evidence, class, gender, race, disability, and organizational life. Reflexivity requires the team to examine how these assumptions shape attention, coding, and conclusions. A reflexive memo might ask: “Why did I interpret a patient’s refusal as non-adherence rather than as a rational response to previous harm?”

Reflexivity does not mean that every interpretation is equally acceptable. A defensible interpretation is grounded in data, coherent with the study design, explicit about assumptions, attentive to contradictory evidence, and appropriately modest in its claims.

Thematic analysis

Thematic analysis identifies and interprets patterned meaning across a dataset. Braun and Clarke’s widely used approach describes six flexible phases: familiarization; generating initial codes; constructing candidate themes; reviewing themes; defining and naming themes; and producing the report. These phases are guides to analytic practice, not rigid stages.

Inductive and deductive thematic analysis

Inductive analysis develops codes and themes primarily from the dataset. It is appropriate when little is known about a phenomenon, when participants’ meanings are central, or when the study seeks to avoid imposing an existing framework. Induction is never completely theory-free: the research question, sampling, language, and researcher’s standpoint still influence what becomes visible.

Deductive analysis begins with concepts derived from theory, prior research, policy, or an implementation framework. For example, the Consolidated Framework for Implementation Research may guide attention to intervention characteristics, inner setting, outer setting, individuals, and implementation processes. Deduction promotes comparability and theory testing, but a rigid codebook can obscure unexpected meanings.

A hybrid approach combines both. Researchers may begin with a provisional framework, add data-driven codes, and revise the framework when evidence does not fit. This is often appropriate in applied health research.

Coding levels

Coding may be semantic or latent. Semantic codes remain close to participants’ explicit words, such as waiting for transport or fear of stigma. Latent codes interpret underlying assumptions or social processes, such as institutionalized exclusion or moralization of self-care. A study should state which level is emphasized because latent analysis involves stronger interpretive claims.

Coding can also be descriptive, process-oriented, values-based, or emotion-focused. Examples include:

• descriptive: difficulty obtaining an appointment;
• process: negotiating referral requirements;
• values: valuing privacy;
• emotion: feeling unsafe during disclosure.

A useful code is specific enough to distinguish one meaning from another and broad enough to recur across relevant extracts. Codes should not become a list of every subject mentioned. “Doctor,” “clinic,” and “phone” may describe data, but “loss of relational continuity” is more analytically useful if the research question concerns patient experience.

Constructing themes

A theme is a patterned meaning organized around a central concept. It should have:

• a clear central organizing idea;
• internal coherence among its extracts;
• distinct boundaries from other themes;
• sufficient prevalence or explanatory importance; and
• relevance to the research question.

Prevalence is not simply a vote count. A single account may be analytically important if it exposes a safety failure or challenges a dominant assumption. Conversely, a frequently repeated statement may reflect the interview schedule rather than substantive importance.

Theme names should be concise and interpretive. “Communication” is usually too broad. “Information is abundant, but responsibility for sense-making is transferred to patients” is more informative. A theme definition should explain its scope, what it includes, what it excludes, and how it answers the research question.

Qualitative content analysis

Qualitative content analysis focuses on the systematic classification and interpretation of textual material. It is particularly useful for policy documents, clinical records, complaints, interviews, and large textual datasets. Content analysis may examine manifest content, meaning what is explicitly present, or latent content, meaning underlying meanings and assumptions.

Three common orientations are:

1. Conventional content analysis: categories are derived from the data when existing theory is limited.
2. Directed content analysis: existing theory or prior research guides initial categories, while data may extend or challenge them.
3. Summative content analysis: the analyst examines the use or presence of words and then interprets their contextual meaning. Simple word counts alone are not qualitative interpretation.

Content analysis generally requires a coding frame. The frame should identify category names, definitions, inclusion and exclusion rules, examples, and relationships among categories. Categories should be mutually distinguishable, although an extract may legitimately receive more than one code when it expresses multiple meanings.

Manifest and latent content: an example

Suppose a hospital’s discharge policy repeatedly uses the terms “compliance” and “patient responsibility.” Manifest analysis can document where these terms occur and what actions they are associated with. Latent analysis can ask what model of the patient is constructed: an autonomous consumer, a responsible partner, or a potential source of risk. The latent interpretation must be supported by the policy’s language, context, and consequences rather than by the analyst’s opinion alone.

Content analysis is valuable when decision-makers need a structured account of what appears across many documents. However, reducing accounts to categories can remove chronology, context, ambiguity, and voice. Analysts should preserve contextual notes and representative extracts alongside category counts.

Framework analysis

Framework analysis is a matrix-based method developed for applied policy research and commonly used in health services research. It combines systematic case-by-case analysis with comparison across cases and themes. A framework may be established in advance from the research questions, policy domains, or an implementation model, while retaining space for emergent findings.

The usual stages are:

1. Familiarization: understanding the dataset and study context.
2. Identifying a thematic framework: defining initial concepts and research questions.
3. Indexing: applying framework categories to the data.
4. Charting: summarizing coded data in a matrix organized by case and category.
5. Mapping and interpretation: examining patterns, explanations, differences, and relationships.

A framework matrix does not replace the original data. Each cell should contain a concise, faithful summary and, where useful, a reference to the transcript or document location. Analysts should avoid filling cells with unsupported generalizations.

A typical matrix might contain participants as rows and analytic domains as columns:

Framework analysis is especially useful when a multidisciplinary team must compare groups, when findings need to inform policy, or when the sample includes diverse roles and settings. Its risk is premature closure: a framework can make analysts search only for what they expected. An “other,” “emergent,” or “contradiction” column can protect against this problem.

Constant comparison and negative-case analysis

Constant comparison involves comparing data throughout analysis rather than only at the end. Comparisons may be made:

• within one interview, between an account of policy and practice;
• between participants with different outcomes;
• between professional groups;
• across sites or time points; and
• between confirming and disconfirming evidence.

The purpose is not to force uniformity. It is to specify conditions under which a pattern holds. For instance, “digital booking improves access” may become “digital booking improves access for patients with reliable connectivity and language-concordant support, but increases exclusion for others.”

A negative case is evidence that challenges an emerging explanation. It may be an exception, a different mechanism, or evidence that a theme is too broad. Analysts should not delete it as inconvenient. They should ask whether:

1. the case is genuinely comparable;
2. the code or theme is poorly defined;
3. a contextual condition explains the difference;
4. the case reveals an alternative pathway; or
5. the original interpretation should be rejected or qualified.

Negative-case analysis strengthens credibility because it turns contradiction into explanation. The final report should state how divergent evidence affected the interpretation.

Memos, matrices, and visualization

Analytic memos

A memo is a dated record of an analytic thought, question, decision, or uncertainty. Useful memo types include:

• descriptive memos: what appears to be happening;
• methodological memos: how coding decisions were made;
• theoretical memos: relationships between concepts;
• reflexive memos: how the researcher’s position may matter; and
• case memos: a concise account of one participant, site, or event.

A strong memo moves beyond description. For example: “Participants describe missed appointments as personal failure, but their narratives show unstable transport and inflexible work. The service’s accountability language may individualize a structural problem.”

Matrices

Matrices support comparison without pretending that qualitative data are numerical. They may compare cases, stakeholder groups, time periods, implementation sites, or themes. A matrix should preserve variation through short quotations, source identifiers, and confidence notes. It is useful to include a column for “evidence against,” “conditions,” or “unresolved question.”

Visualization

Visualization can include thematic maps, process diagrams, timelines, causal pathway sketches, and relationship maps. A thematic map might show how organizational staffing affects continuity, which affects trust, which affects disclosure and follow-up. Arrows should represent an analytic proposition, not proof of causation. Visualizations are thinking tools and should be revised when evidence changes.

A worked analytic example

Consider an illustrative dataset from interviews with 12 patients and 8 clinicians about a hospital’s telephone follow-up service after discharge.

Initial extracts include:

• “The nurse called, but I was at work and could not answer.”
• “The call made me feel that someone remembered me.”
• “My daughter answered because I cannot manage the automated menu.”
• “We are expected to call three times, but there is no interpreter.”
• “One patient refused the call because the previous admission was traumatic.”

Possible initial codes are work-time conflict, feeling remembered, family-mediated access, language barrier, protocol-driven repeated calling, and refusal as protection. Constant comparison shows that the service is not experienced uniformly. Calls can create relational continuity, but access depends on work schedules, digital literacy, language support, and prior experience.

A preliminary theme, Telephone follow-up improves continuity, is too broad. Negative-case analysis reveals that some patients experience the calls as intrusive or inaccessible. A revised thematic structure might be:

1. Follow-up can restore relational continuity after discharge.
2. A standardized contact pathway shifts coordination work onto patients and families.
3. Trust and safety determine whether contact is welcomed.

The framework matrix may then compare patients and clinicians across domains. Clinicians emphasize failed contact rates and protocol completion; patients emphasize timing, language, emotional safety, and whether the caller understands their circumstances. The interpretation is therefore not that the service is simply effective or ineffective. Rather, its benefit is conditional on adaptable scheduling, interpreter availability, trauma-sensitive communication, and alternatives to automated contact.

Detailed applied case: implementing a medication reconciliation intervention

Background

This illustrative case concerns a regional hospital introducing medication reconciliation at discharge for older adults taking multiple medicines. The intervention requires a pharmacist or trained nurse to compare admission medicines, inpatient changes, and the discharge list, then explain the final plan to the patient and primary-care provider.

Problem

After three months, managers observe that documentation completion is high, but readmissions related to medication confusion have not declined. The evaluation team conducts interviews with 10 patients, 8 ward nurses, 4 pharmacists, and 5 community physicians, alongside observations and document review.

Analysis approach

The team uses a hybrid framework. Deductive domains are workflow, communication, patient understanding, interprofessional coordination, and implementation context. Inductive coding is used for unexpected meanings. Analysts write case memos, compare wards, and maintain a negative-case log.

The framework matrix shows three patterns:

Theme development

The initial theme Medication reconciliation is inconsistently implemented is descriptive. Through memoing and comparison, the team develops a more explanatory theme: Formal completion conceals unequal opportunities for understanding and action. A second theme is Coordination depends on relationships, not only on information transfer. A negative case—a ward with low documentation compliance but strong pharmacist-patient interaction—challenges the assumption that documentation percentage is a sufficient implementation indicator.

Approach and solution

The team recommends three changes: protected time for reconciliation, mandatory teach-back for high-risk medicines, and a risk-based communication protocol with community clinicians. It also recommends auditing patient understanding and actionability rather than only form completion.

Outcome

The qualitative findings do not establish that the revised intervention will reduce readmissions; that requires outcome evaluation. They do, however, explain why the original implementation measure was misleading and identify modifiable mechanisms.

Learning points

• Implementation indicators can measure administrative completion rather than meaningful practice.
• The same intervention can operate differently across wards.
• Negative cases can reveal that apparently “poor” performers possess valuable relational practices.
• Qualitative findings are particularly useful for explaining how and why an intervention produces variable outcomes.

Quality, rigor, and defensibility

Quality criteria should fit the study’s paradigm, but several practices are broadly valuable:

• Credibility: prolonged engagement, attention to disconfirming evidence, peer debriefing, and participant engagement where appropriate.
• Dependability: an audit trail showing changes to the codebook, framework, and interpretations.
• Confirmability: evidence that findings are grounded in data and reflexively examined.
• Transferability: thick description of setting, participants, intervention, and conditions so readers can judge relevance elsewhere.

Intercoder agreement can be useful for training and checking a shared coding frame, especially in large applied projects. It should not be treated as proof that one interpretation is objectively correct. If agreement statistics are used, the team must explain the unit of coding, coding instructions, missing data, and how disagreement was resolved. In interpretive thematic analysis, discussion and reflexive engagement may be more appropriate than maximizing agreement.

A transparent report should describe the analytic approach, researchers and reflexive positions, data sources, coding process, treatment of contradictory evidence, and relationship between extracts and claims. Software can facilitate retrieval, coding, memos, and visualization, but it does not perform interpretation.

Ethical, governance, and regulatory considerations

Qualitative data may contain identifiable health histories, names of clinicians, rare conditions, or descriptions of unsafe care. De-identification must consider deductive disclosure: a reader may identify a person from a combination of role, location, and event even after names are removed. Quotations should be edited only when necessary for confidentiality, and substantive meaning should not be distorted.

Teams should establish access controls, retention periods, version management, and procedures for unexpected disclosures of harm. If analysis reveals an immediate safety risk, researchers need a protocol consistent with consent, institutional governance, and applicable law. Participants should understand whether quotations may be published and whether raw data will be shared.

Technology and AI

Computer-assisted qualitative data analysis software can support organization, retrieval, coding, linking, and audit trails. Automated transcription requires human checking because accents, clinical abbreviations, overlapping speech, and emotional pauses may be misrepresented.

Generative AI may assist with administrative tasks such as suggesting preliminary labels or identifying repeated terms, but it creates risks involving confidentiality, hallucinated summaries, automation bias, and loss of contextual nuance. Sensitive health data should not be uploaded to an external system without approved governance, contractual safeguards, and a lawful basis. Any AI assistance should be documented, independently checked against the source, and treated as a support tool rather than an analyst. Human researchers remain responsible for interpretation and accountability.

Common challenges and best practices

Premature coding: Beginning detailed coding before understanding the whole dataset can fragment meaning. Read broadly first and write a provisional map.

Code proliferation: Hundreds of overlapping codes make comparison difficult. Merge codes when their analytic consequences are the same, while preserving meaningful distinctions.

Topic lists mistaken for themes: A list such as “staffing, communication, technology” does not explain a pattern. Ask what connects these topics and what they mean in context.

Overreliance on frequency: Frequent mention is not equivalent to importance. Consider intensity, explanatory power, distribution, and consequences.

Framework rigidity: Deductive categories should be open to revision. Include an emergent category and actively search for evidence that does not fit.

Decontextualized quotations: A quotation should be connected to participant characteristics, setting, and analytic claim without compromising confidentiality.

Confusing association with causation: Qualitative data can suggest mechanisms and plausible explanations, but causal claims require appropriate design and evidence.

Best practice is to maintain a versioned codebook, use regular analytic meetings, record disagreements rather than erase them, preserve links between extracts and claims, examine subgroup differences, and write interpretations throughout the project rather than at the end.

Chapter Summary

Thematic analysis develops patterned meanings across qualitative data; qualitative content analysis systematically classifies manifest or latent content; and framework analysis organizes comparison through a structured matrix. Inductive coding allows concepts to emerge from data, while deductive coding applies theory or prior knowledge; hybrid designs combine both. Constant comparison specifies similarities, differences, and conditions, while negative-case analysis tests and refines interpretations. Memos, matrices, and visualizations make analytic reasoning visible and support team-based inquiry. Defensible qualitative findings require reflexivity, an audit trail, attention to contradictory evidence, ethical protection of participants, and claims proportionate to the data. Technology can support organization but cannot replace contextual human interpretation.

Key Terms

Analytic memo: A dated record of an analytic insight, decision, question, or reflexive observation.

Audit trail: Documentation showing how data, codes, categories, and interpretations developed.

Charting: Summarizing indexed data in a framework matrix.

Code: A concise label attached to a meaningful segment of data.

Codebook: A structured list of codes with definitions, boundaries, and examples.

Constant comparison: Systematic comparison of data, codes, cases, and themes throughout analysis.

Content analysis: Systematic classification and interpretation of textual or visual material.

Deductive coding: Coding guided by existing theory, concepts, or research questions.

Framework analysis: A matrix-based approach for comparing cases and themes in applied research.

Inductive coding: Developing codes from close engagement with the dataset.

Latent content: Underlying assumptions, meanings, or social processes in data.

Manifest content: Explicitly stated or directly observable content.

Negative case: Evidence that challenges, qualifies, or contradicts an emerging interpretation.

Reflexivity: Examination of how the researcher’s position and assumptions shape the study.

Semantic coding: Coding focused on explicit meanings and words used by participants.

Thematic analysis: Identification and interpretation of patterned meaning across data.

Theme: A coherent, significant pattern organized around a central interpretive idea.

Thick description: Detailed contextual account enabling readers to assess transferability.

Review Questions

Short Answer Questions

1. What is the difference between a code, a category, and a theme?
2. Distinguish inductive from deductive coding.
3. What are manifest and latent content?
4. Why is negative-case analysis important?
5. State two functions of an analytic memo.

Long Answer Questions

1. Explain the stages of thematic analysis and discuss why they should be treated as recursive rather than strictly linear.
2. Compare thematic analysis, qualitative content analysis, and framework analysis in relation to purpose, structure, and outputs.
3. Explain how a framework matrix can support comparison without reducing qualitative data to numbers.
4. Discuss the role of reflexivity and audit trails in establishing qualitative rigor.
5. Evaluate the benefits and risks of using AI-assisted tools in qualitative health research.

Analytical / Application Questions

1. A hospital reports 95% completion of a discharge checklist, but patients still misunderstand medication changes. Design a qualitative analysis using coding, a matrix, and negative-case analysis to investigate the discrepancy.
2. Interviews suggest that a digital appointment system improves access for some patients but excludes others. Develop an inductive-deductive coding strategy that could explain the conditions producing these different experiences.
3. A policy document frequently describes patients as “responsible” for managing chronic disease. Propose a manifest and latent content analysis and explain what evidence would be needed to support your interpretation.

Multiple Choice Questions with Answers

Question 1

Which statement best describes inductive coding?

A. Applying only categories specified by an implementation framework

B. Developing codes through close engagement with the dataset

C. Counting the number of words in each transcript

D. Removing all contradictory data before interpretation

Correct Answer: B. Developing codes through close engagement with the dataset

Question 2

What is the central purpose of constant comparison?

A. To ensure every participant receives the same interview questions

B. To replace interpretation with frequency counts

C. To examine similarities, differences, and conditions across data

D. To guarantee that all codes have identical frequency

Correct Answer: C. To examine similarities, differences, and conditions across data

Question 3

Which is the strongest example of an interpretive theme?

A. Communication

B. Telephone calls

C. Patients

D. Standardized contact shifts coordination work onto patients and families

Correct Answer: D. Standardized contact shifts coordination work onto patients and families

Question 4

In framework analysis, charting refers to:

A. Designing a survey instrument

B. Summarizing indexed data in a matrix

C. Calculating inter-rater reliability

D. Deleting data that do not fit the framework

Correct Answer: B. Summarizing indexed data in a matrix

Question 5

What is latent content analysis concerned with?

A. Only the exact words appearing in a document

B. Underlying meanings and assumptions in the data

C. The number of interview participants

D. The duration of each interview

Correct Answer: B. Underlying meanings and assumptions in the data

Question 6

A negative case is best understood as:

A. A participant who must be excluded from analysis

B. Evidence that challenges or qualifies an emerging interpretation

C. A transcript with technical recording problems

D. A code that occurs only once and is always irrelevant

Correct Answer: B. Evidence that challenges or qualifies an emerging interpretation

Question 7

Which practice most directly strengthens an analytic audit trail?

A. Changing codes without recording the reason

B. Documenting revisions to codes, frameworks, and interpretations

C. Reporting only quotations that support the final theme

D. Allowing software to determine the final themes

Correct Answer: B. Documenting revisions to codes, frameworks, and interpretations

Question 8

Why should qualitative software not be treated as the analyst?

A. It cannot store text files

B. It cannot perform retrieval or coding

C. It organizes data but does not provide contextual human interpretation

D. It prevents researchers from writing memos

Correct Answer: C. It organizes data but does not provide contextual human interpretation

Question 9

Which action is most appropriate when a deductive framework fails to capture an important recurring idea?

A. Force the idea into the nearest existing category

B. Delete the idea because it was not anticipated

C. Add or revise an emergent category and document the change

D. Stop analysis immediately

Correct Answer: C. Add or revise an emergent category and document the change

Question 10

What does transferability depend on most directly?

A. A claim that findings apply everywhere

B. Thick description that allows readers to judge relevance elsewhere

C. A high number of codes per transcript

D. Using only quantitative outcome measures

Correct Answer: B. Thick description that allows readers to judge relevance elsewhere