Second Cycle Coding: Pattern, Focused & Axial (2026)

Qualitati Research Team · 2026-10-08 · 11 min read

Short answer: Second cycle coding is the stage of qualitative analysis where you reorganize your first-cycle codes into a smaller set of categories, patterns or concepts. The most common methods are pattern coding, which groups codes into explanatory units, and focused coding, which re-codes the data with your most significant initial codes. Axial and theoretical coding serve grounded theory.

Second cycle coding is where most qualitative projects either become an analysis or stall as a long list of labels. First-cycle methods such as in vivo, descriptive or process coding break the data open. Second cycle coding puts it back together: fewer codes, more abstraction, and explicit claims about how things relate. This guide explains the main second cycle coding methods in Saldaña's framework, walks through a worked example, and gives a decision matrix and an audit checklist you can cite in a methods chapter.

It also covers the question supervisors now ask in 2026: which part of the second cycle, if any, can an AI tool do? The short version is that models are useful at proposing groupings and poor at owning them, which makes the second cycle the place where human judgment matters most.

Last updated: October 8, 2026

Key takeaways

Second cycle coding reorganizes codes, not raw data. Its input is your first-cycle codebook; its output is categories, patterns or concepts.
Pattern coding groups first-cycle codes into explanatory units: themes, causes, relationships or constructs.
Focused coding picks the most frequent or significant initial codes and uses them to re-code the whole dataset.
Axial and theoretical coding belong to grounded theory; use them only if that is your methodology.
Every merge is a claim. Record why codes were grouped, or your audit trail ends at cycle one.
AI can propose candidate clusters; the decision to merge, split or drop a code should stay with the researcher.

What is second cycle coding?

Second cycle coding is any coding method applied to the codes produced in a first pass, with the aim of reducing, classifying and abstracting them. Johnny Saldaña's The Coding Manual for Qualitative Researchers (5th edition, SAGE, 2025) separates first-cycle methods from second-cycle methods; the second-cycle set covers pattern, focused, axial, theoretical, elaborative and longitudinal coding.

The distinction is about the unit of work. In the first cycle you read transcripts and attach labels to segments. In the second cycle you mostly read your own codes, compare them, and decide which belong together and why. You still return to the data, but you return with a question about a group of codes rather than about a single line.

A June 27, 2026 review of the 5th edition in The Qualitative Report notes that the edition discusses generative AI while keeping interpretation and accountability with the researcher. That stance matters most in the second cycle, because that is where interpretation becomes visible.

The main second cycle coding methods

Most studies use one or two of the following methods, not all six. Choose by methodology and by the kind of claim you need to make.

Pattern coding

Pattern coding groups first-cycle codes into a smaller number of “meta-codes” that explain something. In Miles, Huberman and Saldaña's Qualitative Data Analysis: A Methods Sourcebook (SAGE), pattern codes typically summarize one of four things: categories or themes, causes or explanations, relationships among people, or theoretical constructs. It is the most general-purpose second cycle method and the natural bridge into thematic analysis.

Focused coding

Focused coding comes from Kathy Charmaz's constructivist grounded theory, set out in Constructing Grounded Theory (SAGE). You select the initial codes that appear most frequently or carry the most analytic weight, then use them to sift the rest of the data. It is more directed than initial coding and usually produces provisional categories. It works well outside grounded theory too, whenever the first-cycle list is large and uneven.

Axial and theoretical coding

Axial coding relates categories to their subcategories, properties and dimensions; theoretical (or selective) coding integrates categories around a core category. Both are grounded-theory procedures with specific epistemological baggage. For the full sequence, see our guide to open, axial and selective coding.

Elaborative and longitudinal coding

Elaborative coding starts from codes or constructs in a previous study and tests whether your data support, extend or contradict them, which makes it a top-down method. Longitudinal coding tracks change in participants across waves of data. Both are specialized; reach for them when the design calls for it.

Second-Cycle Method Selector

An original Qualitati decision matrix. Read across from your situation to the method that fits, and note the claim each method commits you to.

Your situation	Method	What you will claim	Main risk
Generic qualitative or thematic analysis, mixed first-cycle codes	Pattern coding	These codes form a theme, cause or relationship	Topic buckets dressed up as themes
Large first-cycle list (100+ codes), uneven frequency	Focused coding	These codes best capture the data	Frequency mistaken for significance
Grounded theory, building a process model	Axial, then theoretical coding	Categories relate in this structure around a core	Imposing a paradigm model the data do not support
Testing or extending a prior framework	Elaborative coding	Our data confirm, extend or contradict X	Confirmation bias toward the prior framework
Panel or multi-wave interviews	Longitudinal coding	Participants changed in this way over time	Treating attrition as change
Fewer than about 30 first-cycle codes	Often none needed	Go straight to categories and themes	Adding ceremony without adding analysis

Worked example: from first-cycle codes to a pattern code

Continuing the hypothetical teacher-workload study from our in vivo coding guide, suppose the first cycle across ten interviews produced these codes, among others:

"Sundays are basically school days" (in vivo)
"carrying it around in your head" (in vivo)
Answering parent emails late at night (descriptive)
Ruminating about work (process)
"I just don't open the laptop on Saturdays" (in vivo)
Setting device boundaries (process)
Marking at the kitchen table (descriptive)

A topic-level grouping would put all seven under Weekend work. That is a category, not a pattern; it tells a reader where the data sit, not what they mean.

A pattern code makes a claim. Comparing the codes, two patterns appear: work that has no clear end point spills into rest time and into the head (Unbounded tasks colonize rest), and some participants push back with physical rules about devices (Boundary-making as self-protection). The second pattern exists only in relation to the first, which is an explanatory link worth memoing.

Focused coding would take a different route: notice that "carrying it around in your head" recurs in seven of ten interviews in different words, promote it to a focused code (Mental carrying), and re-read all transcripts to see where it applies, where it does not, and what it hides. The deviant cases are where the analysis gets interesting; see negative case analysis.

For how pattern codes then become themes, see codes vs categories vs themes.

Can AI do second cycle coding?

Partly. Grouping codes is a clustering-shaped task, and language models are fluent at proposing clusters with plausible labels. The difficulty is that a plausible label is not the same as a defensible claim, and the second cycle is where qualitative research makes claims.

Recent work points at the same split. In a preprint submitted September 10, 2026, Kim and Mitchell studied multi-agent LLM coding, where AI coders code independently, debate and reconcile. They report that accuracy depended on codebook length, data similarity and the amount of agent disagreement, that intense and unresolved debates were associated with higher accuracy, and that the models imitated human discussion behaviors but lacked adaptive responsiveness to context. Read for the second cycle, that suggests disagreement is a signal to preserve, not noise to smooth away. In a CHI 2026 study of ChatQDA, researchers trusted an on-device model for explicit content extraction more than for deeper interpretation; that is a four-participant study, so treat it as a signal about failure modes.

A practical division of labor:

Delegate: proposing candidate groupings, flagging near-duplicate codes, listing every segment under a proposed pattern, and counting how many participants each candidate covers.
Keep: deciding whether a grouping is a pattern or just a topic, naming the pattern, deciding what to drop, and handling cases that do not fit.

When you prompt a model for candidate pattern codes, ask it to return for each candidate the member codes, the participant IDs covered, one counter-example, and an explicit note on whether the grouping is topical or explanatory. Ask it to list codes it could not place rather than forcing them into a cluster. Those leftovers are often the most important part of the output. For more on keeping AI-assisted steps traceable, see audit trails for AI qualitative analysis.

The Pattern Code Audit Checklist

An original Qualitati asset. Run it on every pattern or focused code before it becomes a theme. Each “no” is a repair item.

Claim test. Can you state the code as a sentence with a verb (“Unbounded tasks colonize rest”), not only as a noun (“Weekend work”)?
Membership test. Is every first-cycle member code listed, with a one-line reason for its inclusion?
Spread test. Does the pattern appear across participants, not only in one or two articulate interviews?
Counter-example test. Have you looked for and recorded data that cut against it?
Leftover test. Are first-cycle codes that fit no pattern listed and explained, not silently dropped?
Data-return test. Did you re-read the underlying segments, not just the code labels, before merging?
Provenance test. If AI proposed the grouping, is that recorded, with what you accepted, changed and rejected?
Memo test. Is there a dated analytic memo explaining the decision?

Where Qualitati fits

Qualitati is a European, budget-friendly qualitative research platform for universities and research companies, with data privacy and GDPR as central priorities. Two parts of it map onto second cycle coding.

QDA Workspace supports AI-assisted inductive and deductive coding, codebook generation and theme visualization. It is where a first-cycle codebook can be reviewed, grouped and refined under your control.
ThemeLens runs a map-reduce thematic analysis across up to 100 transcripts at once, mapping codes to research questions and synthesizing themes with participant-anchored quotes, so a proposed pattern can be checked back against what participants actually said.

Upstream, AI-moderated interviews, focus groups and conversational surveys collect the material, in 10 languages. Qualitati publishes per-credit usage rates and starts free with 30 monthly credits, no credit card required.

Who this is for, and when not to use it

Who this is for: PhD students and academic researchers moving from first-cycle codes to findings; research labs that need a documented, repeatable merge procedure across coders; qualitative and market research firms turning large interview sets into defensible themes; UX and insights teams synthesizing discovery research.

When not to use a formal second cycle: when your first-cycle codebook is already small and conceptual; when you are applying a fixed deductive framework, where the categories exist before coding (see inductive vs deductive coding); or when you are running rapid analysis to a fixed deadline, where a structured matrix may serve better (see rapid qualitative analysis).

Limitations and methodological concerns

Methods are not interchangeable. Focused and axial coding carry grounded-theory assumptions. Using their vocabulary in a reflexive thematic analysis can draw reviewer criticism about methodological coherence; see thematic analysis vs grounded theory.

Frequency is not significance. Focused coding often starts from frequent codes, but a code mentioned once by a key informant can matter more than one mentioned by everyone. Say how you weighed this.

AI groupings converge on the obvious. Models tend to propose clusters that match common categories in their training data. That is useful as a baseline and risky as a finding. Human-review note: if your methods section says AI assisted second cycle coding, describe what it proposed and what you decided, and verify on your own data; the published evidence on AI in this step is recent and small.

FAQ

What is the difference between first and second cycle coding?

First cycle coding attaches initial labels to segments of raw data. Second cycle coding works on those labels, grouping, reducing and abstracting them into categories, patterns or concepts. The first cycle opens the data up; the second organizes it into claims.

What is pattern coding in qualitative research?

Pattern coding groups first-cycle codes into a smaller set of explanatory meta-codes. Pattern codes typically summarize a theme, a cause or explanation, a relationship among people, or a theoretical construct. It is the most widely used second cycle method.

What is focused coding?

Focused coding, from Charmaz's constructivist grounded theory, selects the most frequent or analytically significant initial codes and uses them to re-code the dataset. It produces provisional categories and tests whether those codes hold across all the data.

Is axial coding a second cycle method?

Yes. Saldaña lists axial coding as a second cycle method. It relates categories to subcategories and is specific to grounded theory, so it is usually not the right choice for a thematic analysis.

Do I always need second cycle coding?

No. If your first-cycle codes are already few and conceptual, or you are applying a predefined framework, you can move directly to categories and themes. Second cycle coding earns its time when the first-cycle list is large or mixed.

Can ChatGPT or another LLM do pattern coding?

It can propose candidate groupings and labels quickly. It should not decide which groupings count as findings. Ask for member codes, coverage, counter-examples and unplaced codes, and keep the merge decisions and memos yourself.

Bottom line

Second cycle coding turns a first-cycle codebook into analysis. Pattern coding makes explanatory claims about groups of codes; focused coding tests which initial codes really carry the data; axial and theoretical coding serve grounded theory. Whichever you use, every merge is a claim that should be written down. AI can speed up the proposing; the deciding is still the researcher's job.

Start free with 30 credits and take a first-cycle codebook through a second cycle in QDA Workspace, or view transparent pricing. Comparing tools? See how Qualitati sits against NVivo, ATLAS.ti, MAXQDA and AI-native research platforms, or create an account and start coding today.