9. Coding and Thematic Analysis

Introduction and Context

Qualitative research produces rich, extensive, and context-dependent material: interview transcripts, field notes, documents, observations, digital traces, images, and organizational records. The analytical challenge is not simply to reduce this material, but to interpret it without losing the meanings, relationships, contradictions, and situated experiences that make it valuable. Coding and thematic analysis provide a systematic means of moving from raw data to defensible interpretation.

Coding is the process of assigning concise labels to segments of data that appear analytically significant. Thematic analysis is the broader process through which patterns of meaning are identified, developed, examined, and reported. Coding is therefore an important activity within thematic analysis, but the two terms are not interchangeable. Coding may be used in grounded theory, case study research, ethnography, discourse analysis, and other approaches. Thematic analysis may be conducted inductively or deductively, at a semantic or latent level, and within different philosophical assumptions about knowledge and reality.

At doctoral level, analysis must be more than competent classification. The researcher must explain why particular interpretations were developed, how alternative interpretations were considered, how the researcher's position influenced analysis, and how themes relate to the research question, theoretical framework, and context. A theme is not merely a topic mentioned frequently. It is a patterned meaning that is relevant to the study's purpose and supported by a coherent body of evidence.

Learning Objectives

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

1. Explain the relationship between data familiarization, coding, categorization, and thematic interpretation.

2. Distinguish open coding from focused coding and inductive from theory-led analysis.

3. Apply constant comparison to qualitative data.

4. Develop, review, define, and name themes systematically.

5. Distinguish semantic from latent thematic analysis.

6. Evaluate the quality, reflexivity, credibility, and limitations of a coding process.

7. Use digital tools and collaborative procedures without confusing software organization with interpretation.

8. Address ethical and governance issues in qualitative data analysis.

From Raw Data to Analytic Meaning

Qualitative analysis is often represented as a sequence, but in practice it is iterative. A researcher may return to the transcript after developing a theme, revise an earlier code after reading a later interview, or reconsider the research question when an unexpected pattern emerges. The process can be described as a movement among four levels:

• Data extracts: specific words, statements, interactions, or observations.
• Codes: concise labels capturing what an extract appears to signify.
• Categories or code families: related codes grouped around a broader analytic concern.
• Themes: patterned and interpretive accounts that answer or illuminate the research question.

These levels should not be treated as mechanically hierarchical. A category may become a theme in one project but remain a descriptive grouping in another. The distinction depends on the analytical purpose. For example, the codes “avoids raising concerns,” “waits for senior approval,” and “uses informal channels” may form a category called “restricted voice.” A theme might then be “psychological safety is conditional on hierarchy,” which makes a more interpretive claim about organizational life.

A useful audit trail records the movement between these levels. It may include a coding framework, analytic memos, comparison tables, theme maps, decisions about discarded codes, and notes on contradictory evidence. Such records do not prove that an interpretation is true, but they make the reasoning inspectable.

Familiarization: Reading Before Coding

Familiarization is an active analytical stage rather than a preliminary administrative task. The researcher reads and, where relevant, listens to or views the material repeatedly. Initial notes may concern recurring ideas, emotional intensity, silences, contradictions, unusual metaphors, temporal sequences, and relationships among participants or events.

During familiarization, researchers should record provisional impressions without prematurely treating them as final themes. A field note such as “participants describe flexibility positively, but only when they control its boundaries” may later become a useful analytic insight. Familiarization also involves understanding the circumstances in which data were produced. An interview statement given to a researcher known to be a senior manager may carry different interpretive conditions from the same statement given to an independent researcher.

Transcription decisions matter. Verbatim transcription, selective transcription, translation, treatment of pauses, and removal of identifying details can all affect analysis. Where speech, interaction, or emotion is analytically important, pauses, laughter, interruptions, and emphases may need to be retained. Researchers should document these choices rather than presenting the transcript as a transparent mirror of experience.

Open Coding and Focused Coding

Open coding involves closely examining data and assigning provisional labels to meaningful segments. It is especially useful at the beginning of analysis because it encourages attentiveness to participants' language and reduces dependence on preconceived categories. Codes may be descriptive, process-oriented, emotional, interpretive, or expressed in participants' own words. An in vivo code preserves a distinctive participant expression, such as “quiet quitting before quiet quitting,” when that phrase captures an important local meaning.

Open coding should remain sufficiently specific to be useful. “Problems” is usually too broad; “managerial promises not followed by resource allocation” is more analytically informative. Coding can be line by line, incident by incident, or meaning unit by meaning unit. The appropriate unit depends on the research question. A short phrase may be sufficient for a study of language, whereas a whole episode may be necessary for understanding decision-making.

Focused coding follows initial coding. The researcher identifies codes that appear particularly significant, recurrent, explanatory, or connected to the research question and applies them more selectively across the dataset. Focused coding is not merely a shorter code list. It involves comparing the scope and meaning of codes, merging overlapping labels, separating overly broad codes, and examining which codes have explanatory power.

For instance, early codes such as “working around the system,” “using personal contacts,” and “bypassing formal approval” may be consolidated into “informal navigation of bureaucracy.” The researcher should still examine whether all three extracts express the same phenomenon or whether important differences are being erased.

Inductive and Theory-Led Coding

Inductive coding develops codes from engagement with the data. It does not mean that the researcher approaches the material without assumptions; prior knowledge, disciplinary concepts, and the research question inevitably influence attention. Induction is therefore better understood as allowing concepts to emerge through close engagement rather than imposing a complete framework in advance.

Theory-led, deductive, or directed coding begins with concepts derived from existing theory, previous research, policy frameworks, or the study's conceptual model. A researcher studying organizational justice might begin with codes relating to distributive, procedural, interpersonal, and informational justice. Theory-led coding can support comparison across cases and make explicit connections with scholarship, but it may cause researchers to overlook unexpected meanings or force data into unsuitable categories.

Many doctoral studies use an abductive approach, moving back and forth between empirical material and conceptual explanation. An unexpected pattern may prompt a return to the literature; a theoretical concept may be revised after encountering evidence that does not fit it. Abduction is particularly valuable when the goal is to develop a plausible explanation rather than merely describe frequencies.

A rigorous study should state which logic predominates, how the researcher's prior concepts were handled, and what was done with data that did not fit the initial framework. Negative cases are analytically valuable. They may weaken a proposed theme, reveal conditions under which it does not apply, or lead to a more nuanced explanation.

Constant Comparison and Categorization

Constant comparison is the systematic comparison of data segments, codes, cases, contexts, and emerging interpretations. It is associated strongly with grounded theory but is useful across qualitative designs. The researcher asks questions such as: How is this experience similar to or different from another? Under what conditions does it occur? Who describes it differently? What consequences follow? What would count as a counterexample?

Comparison may occur within a single interview, across participants, between organizational units, across time, or between formal accounts and observed practice. It prevents themes from being built around a vivid quotation or a dominant participant alone. A comparison matrix can display how a developing category appears across cases, while analytic memos can explain why apparent differences matter.

Categorization groups related codes, but categories should have conceptual boundaries. A category called “technology concerns” may include privacy, surveillance, learning difficulties, and system reliability, but these may require separation if they have different causes or consequences. Categories become analytically useful when the researcher specifies their properties, dimensions, conditions, and relationships.

Developing and Reviewing Themes

A theme is a patterned response or meaning that is important in relation to the research question. It should have internal coherence and be distinguishable from other themes. Theme development commonly involves the following iterative activities:

1. Collating relevant coded extracts.

2. Grouping related codes into provisional themes.

3. Examining whether each theme has a coherent central idea.

4. Checking themes against the entire dataset.

5. Assessing relationships among themes.

6. Defining the scope and boundaries of each theme.

7. Naming themes in a precise and memorable manner.

8. Selecting extracts that demonstrate, rather than merely decorate, the analysis.

A thematic map can represent relationships among themes and subthemes. It should be more than a visual list. For example, “performing commitment,” “managing visibility,” and “protecting personal boundaries” might be related dimensions of a broader theme concerning the social negotiation of hybrid work.

Theme review asks two different questions. First, do the extracts within a theme belong together? Second, does the theme as a whole represent a meaningful pattern in the dataset? A theme can fail because it is internally inconsistent, too broad, merely a topic, or weakly connected to the research question. Theme names should communicate an analytical claim. “Communication” is weak; “Communication becomes a performance of trust” is more informative, provided the data support it.

Semantic and Latent Analysis

Semantic analysis remains close to the explicit content of participants' accounts. It asks what participants say, describe, or report. In a study of remote work, a semantic theme might be “employees value flexibility because it reduces commuting time.” Such analysis can be rigorous and valuable, particularly when the research purpose is to document experiences or perceptions.

Latent analysis examines underlying assumptions, conceptualizations, ideologies, and social conditions that may shape what is said. A latent interpretation of the same material might examine how flexibility is constructed as an individual responsibility, or how organizational expectations remain embedded in apparently autonomous work arrangements. Latent analysis requires theoretical sensitivity and should not be presented as the hidden truth behind participants' words. It is an interpretation produced through a particular conceptual lens.

Semantic and latent approaches are not mutually exclusive, but combining them requires clarity. A report should indicate when it moves from participants' explicit meanings to the researcher's interpretation of broader structures. Latent claims require careful support, reflexivity, and attention to alternative explanations.

Reflexivity, Quality, and Researcher Judgment

Coding is interpretive. Even when multiple researchers use the same codebook, they may notice different meanings or assign different significance to the same extract. Reflexivity requires researchers to consider how their identity, theoretical commitments, institutional position, relationship with participants, and expectations influence analysis.

Quality is not established by inter-coder agreement alone. Agreement can be useful in team projects, especially for clarifying code definitions and identifying inconsistency, but forced consensus may suppress legitimate interpretive difference. Depending on the methodology, quality may be supported through prolonged engagement, negative-case analysis, peer debriefing, member reflection, triangulation, transparent documentation, rich contextual description, and a clear link between claims and evidence.

A codebook should normally state the code name, definition, inclusion and exclusion criteria, examples, and relationships with other codes. It should evolve as analysis develops. Freezing it too early can prevent learning from the data; changing it without documentation undermines auditability.

Digital Tools and Collaborative Analysis

Computer-assisted qualitative data analysis software can store transcripts, attach codes, retrieve extracts, compare cases, write memos, and display relationships. It improves organization and supports large or collaborative projects. It does not determine what a theme means. A software report showing that a code occurs frequently is not itself an interpretation.

Automated text analysis and generative AI may assist with transcription, searching, clustering, translation, or preliminary suggestion of codes. These uses create risks involving confidentiality, data transfer, algorithmic bias, decontextualization, and false confidence. AI-generated codes must be treated as provisional, checked against the original data, and disclosed according to institutional and publication requirements. Sensitive organizational or personal data should not be uploaded to systems without appropriate contractual, legal, and ethical safeguards.

Collaborative teams should agree on analytic roles, version control, dispute procedures, and the status of participant quotations. Differences between coders should stimulate discussion about concepts and context, not be reduced automatically to a numerical reliability score.

Illustrative Case: Understanding Digital Transformation Resistance

Background

The following is a fictional illustrative case. Meridian Health Services, a regional healthcare organization, introduced a digital scheduling platform intended to reduce administrative workload and improve patient access. Senior leaders described resistance as a training problem.

Situation and Data

A doctoral researcher conducted interviews with administrators, clinicians, and managers, observed implementation meetings, and examined internal communications. Initial open codes included “system slows me down,” “fear of being monitored,” “training ignores actual work,” “protecting patient relationships,” and “leaders call everything resistance.”

Analysis

Focused coding grouped these into categories such as operational disruption, surveillance concerns, inadequate participation, and professional identity. Constant comparison showed that administrators mainly emphasized workflow disruption, while clinicians described threats to judgment and patient relationships. Managers often used the language of adoption and compliance. A purely semantic analysis might have produced themes concerning training, workload, and communication.

The researcher developed a latent interpretation: resistance was not simply opposition to technology but a struggle over whose knowledge counted in the redesign of work. The theme was named “Digital adoption becomes a negotiation of professional authority.” A second theme, “Participation is promised but operationally constrained,” captured the contradiction between consultation rhetoric and limited influence over system design.

Approach and Outcome

The researcher reviewed these themes against all transcripts, searched for negative cases, and retained examples of staff who welcomed the platform. Those cases showed that acceptance was higher where users had participated in configuration and could see benefits for patient care. The final report recommended co-design workshops, role-specific training, transparent monitoring policies, and post-implementation feedback loops rather than a generic resistance-management programme.

Learning Points

The case demonstrates that frequency is not equivalent to significance; a less frequent concern about professional authority may be more explanatory than repeated complaints about training. It also shows the importance of comparing occupational groups, distinguishing semantic accounts from latent interpretation, and treating contradictions as analytical evidence.

Common Problems, Risks, and Ethical Responsibilities

Several errors recur in qualitative analysis. Researchers may code too much without developing an argument, create an excessive number of overlapping codes, treat interview questions as themes, select only quotations that support a preferred conclusion, or equate repetition with importance. Another problem is premature closure: themes are named before the researcher has examined variation, context, and negative cases.

Ethically, analysis must preserve confidentiality. Small organizational settings can make participants identifiable even after names are removed. Quotations should be edited only when necessary for clarity or anonymity, and the meaning must not be altered. Researchers should consider whether publishing a latent interpretation could stigmatize a group or expose information that participants did not reasonably expect to be made public.

Data governance includes secure storage, controlled access, retention periods, destruction procedures, and compliance with applicable institutional and legal requirements. Participants should be informed, within the limits of the study design, about how their data will be analyzed and quoted. Researchers also have a responsibility not to convert participants' accounts into managerial conclusions without acknowledging power relations and interpretive limits.

Best Practices for Doctoral Researchers

A defensible coding and thematic analysis process generally includes the following practices:

• Begin with the research question and epistemological position, not with software functions.
• Familiarize yourself deeply with the full dataset before making strong claims.
• Use precise, revisable codes and document changes.
• Compare cases, groups, time periods, and contradictory evidence.
• Distinguish description from interpretation and semantic from latent claims.
• Use theory as a resource while remaining open to empirical challenge.
• Preserve an audit trail of memos, decisions, versions, and rejected interpretations.
• Treat reflexivity as an analytical practice rather than a personal biography section.
• Use quotations as evidence for an argument, not as substitutes for analysis.
• Explain how themes answer the research question and what they cannot establish.

Current Developments

Contemporary qualitative analysis increasingly involves multimodal and digital data, including online communities, platform interactions, video, emojis, and organizational collaboration records. These materials require attention to context, audience, persistence, and the difference between public accessibility and ethical appropriateness. Computational tools can help identify patterns in large datasets, but interpretation remains dependent on theoretical framing and contextual knowledge.

There is also growing interest in reflexive thematic analysis, participatory analysis, and approaches that involve participants or community partners in interpreting data. These approaches can broaden interpretive insight while raising questions about power, ownership, disagreement, and representation. Doctoral researchers should select an approach because it fits the research problem and philosophical position, not because it is fashionable or supported by a particular software package.

Chapter Summary

Coding transforms meaningful segments of qualitative data into analytically useful labels, while thematic analysis develops patterned interpretations connected to a research question. The process begins with deep familiarization, proceeds through open and focused coding, and uses categorization and constant comparison to examine relationships, variation, and negative cases. Themes must be coherent, distinctive, evidence-based, and interpretive rather than merely frequent topics. Inductive, theory-led, and abductive strategies each have strengths and risks. Semantic analysis stays close to explicit accounts; latent analysis examines underlying assumptions and structures. Reflexivity, ethical data governance, transparent documentation, and careful use of digital tools are essential to credible doctoral research.

Key Terms

Coding: Assigning analytic labels to meaningful segments of qualitative data.

Open coding: Initial, exploratory coding that remains close to the data.

Focused coding: Selective use and refinement of the most significant initial codes.

In vivo code: A code using a participant’s own distinctive words.

Category: A grouping of related codes with conceptual boundaries.

Theme: A patterned meaning that is relevant to the research question.

Thematic map: A visual representation of relationships among codes, categories, and themes.

Constant comparison: Systematic comparison across data, cases, codes, and contexts.

Inductive analysis: Analysis in which concepts are developed primarily through engagement with data.

Deductive or theory-led analysis: Analysis guided by prior theory, concepts, or frameworks.

Abductive analysis: Iterative movement between empirical evidence and conceptual explanation.

Semantic analysis: Interpretation focused on explicit, surface meanings in the data.

Latent analysis: Interpretation of underlying assumptions, structures, or ideologies.

Negative case: Evidence that challenges, qualifies, or does not fit an emerging interpretation.

Reflexivity: Critical examination of how the researcher influences knowledge production.

Audit trail: Documentation of analytical decisions, changes, and supporting evidence.

Codebook: A structured record of code definitions, boundaries, and examples.

Member reflection: Inviting participants to respond to or reflect on interpretations.

Computer-assisted qualitative data analysis software: Digital software used to organize and retrieve qualitative data and codes.

Review Questions

Short Answer Questions

1. What is the difference between a code, a category, and a theme?

2. Why is familiarization an analytical rather than merely preparatory activity?

3. Define constant comparison and identify one way it can be used.

4. How does focused coding differ from open coding?

5. What distinguishes semantic from latent analysis?

Long Answer Questions

1. Explain how a doctoral researcher can move from raw interview data to a defensible thematic interpretation.

2. Critically compare inductive, theory-led, and abductive coding strategies.

3. Discuss the role of reflexivity and the audit trail in establishing qualitative research quality.

4. Explain how negative cases and contradictory evidence can improve theme development.

5. Evaluate the benefits and risks of using qualitative analysis software and AI-assisted tools.

Analytical / Application Questions

1. A researcher finds that employees frequently mention training problems but occasionally describe the technology as a threat to professional judgment. How should the researcher decide whether these form separate themes or dimensions of a broader theme?

2. Design an initial coding and comparison strategy for studying how managers implement a controversial performance-monitoring system across three organizational units.

3. A research team has achieved high agreement by forcing all disagreements into a single codebook. Critically assess this procedure and propose a more reflexive alternative.

22. Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes a theme from a category?

A. A theme is always more frequent than a category.

B. A theme presents a patterned meaning relevant to the research question, while a category groups related codes.

C. A category is interpretive, whereas a theme is always descriptive.

D. A theme is created before coding, while a category is created afterward.

Correct Answer: B. A theme presents a patterned meaning relevant to the research question, while a category groups related codes.

Question 2

A researcher initially labels extracts as “waiting for approval,” “using personal contacts,” and “bypassing formal procedures.” What is the most appropriate next step in focused coding?

A. Delete all three because they use different wording.

B. Treat each quotation as a separate final theme.

C. Examine whether they can be refined into a coherent code such as informal navigation of bureaucracy.

D. Count the words in each extract and select the longest one.

Correct Answer: C. Examine whether they can be refined into a coherent code such as informal navigation of bureaucracy.

Question 3

Which practice best illustrates constant comparison?

A. Reading only the most articulate participant’s interview.

B. Comparing how a developing code appears across roles, cases, or contexts.

C. Using software to calculate the total number of codes.

D. Removing all data that contradict the initial framework.

Correct Answer: B. Comparing how a developing code appears across roles, cases, or contexts.

Question 4

A latent analysis of employees’ accounts of flexible work would most likely examine:

A. Only the number of times flexibility is mentioned.

B. The explicit hours employees report working.

C. Underlying assumptions about autonomy, responsibility, and organizational control.

D. Spelling differences in interview transcripts.

Correct Answer: C. Underlying assumptions about autonomy, responsibility, and organizational control.

Question 5

What is a principal risk of purely theory-led coding?

A. It makes comparison impossible.

B. It prevents the use of categories.

C. It may force unexpected data into pre-existing concepts.

D. It always produces latent themes.

Correct Answer: C. It may force unexpected data into pre-existing concepts.

Question 6

In the illustrative Meridian Health Services case, the most explanatory latent theme concerned:

A. The number of software functions available.

B. Professional authority and the limited influence of users over system design.

C. The length of the interview transcripts.

D. The organization’s geographic location.

Correct Answer: B. Professional authority and the limited influence of users over system design.

Question 7

Which item would most appropriately appear in a codebook?

A. The researcher’s preferred final conclusion only.

B. A code definition, inclusion and exclusion criteria, and illustrative examples.

C. Only the frequency of each participant’s speech.

D. A list of quotations with no interpretation.

Correct Answer: B. A code definition, inclusion and exclusion criteria, and illustrative examples.

Question 8

Why should negative cases be retained during thematic analysis?

A. They automatically invalidate every theme.

B. They can qualify, refine, or challenge an emerging interpretation.

C. They increase the apparent frequency of a code.

D. They eliminate the need for reflexivity.

Correct Answer: B. They can qualify, refine, or challenge an emerging interpretation.

Question 9

What is the most defensible use of AI-generated codes in doctoral qualitative research?

A. Accepting them as objective findings.

B. Using them provisionally, checking them against original data, and documenting their use.

C. Uploading confidential data without governance review.

D. Replacing researcher interpretation with automated frequency counts.

Correct Answer: B. Using them provisionally, checking them against original data, and documenting their use.

Question 10

Which proposed theme is most analytically developed?

A. Communication

B. Meetings

C. Technology concerns

D. Communication becomes a performance of trust

Correct Answer: D. Communication becomes a performance of trust