You stare at 47 pages of interview transcripts. And a spreadsheet that's somehow both empty and overwhelming. In real terms, highlighters in six colors. And the sinking feeling that you're doing it wrong.
Been there. We all have.
Coding qualitative data isn't magic. It's not something only "natural researchers" understand. It's a set of learnable moves — decisions you make, one segment at a time, that turn raw talk into something you can actually argue with.
The problem? So naturally, you switch mid-project. Plus, in practice, you mix them. Even so, most methods textbooks list coding types like a menu at a diner: pick one. You realize halfway through that your open codes are actually axial, and your memo from Tuesday changes everything Simple, but easy to overlook..
So let's walk through the actual landscape — not the textbook version, the working version That's the part that actually makes a difference..
What Is Coding in Qualitative Research
Coding is just labeling. That's it. You take a chunk of data — a sentence, a paragraph, a gesture in a video — and you attach a name to what's happening there Surprisingly effective..
The name can be descriptive ("participant describes onboarding frustration"). It can be theoretical ("resistance as identity work"). It can be interpretive ("powerlessness in institutional settings"). But every code is a claim: *this bit matters, and here's why.
You're not counting. That's why you're not summarizing. You're building an analytic architecture, one label at a time.
And the type of coding you choose — or stumble into — shapes what you can eventually say.
The three layers people confuse
Most frameworks talk about three "stages": open, axial, selective. They're lenses. But here's what gets missed: they're not stages. Grounded theory folks know these cold. You move between them constantly.
Open coding fractures the data. Think about it: axial coding stitches it back together. Selective coding decides what the whole thing is about.
If you treat them as a linear checklist, you'll produce a tidy coding tree that explains nothing.
Why Coding Choices Matter
Your coding strategy determines your findings. Full stop No workaround needed..
Code only for topics ("workload," "management," "tools") and you'll get a thematic inventory — useful for reporting, thin on insight. Code for process ("normalizing burnout," "performing compliance") and you start seeing mechanisms. Code for theoretical constructs ("neoliberal subjectivity," "affective labor") and you're suddenly in conversation with literature Practical, not theoretical..
None is "better." But they answer different questions Worth keeping that in mind..
And here's what most guides won't tell you: your first coding pass is almost never your real coding pass. The real work happens in the second, third, fourth rounds — when you collapse codes, split them, rename them, delete the clever ones that turned out to be noise.
How Coding Actually Works (The Moves Nobody Shows You)
Let's get practical. Here are the coding types you'll actually use, what they do, and when to deploy them.
In vivo coding — borrowing participants' words
"In vivo" means "in the living." You use the participant's exact phrase as your code Small thing, real impact..
"It's a black box — I put tickets in, nothing comes out."
Code: black box
Why it works: preserves voice, catches metaphors you'd never invent, signals respect. Participants often name their own experience better than we can.
When to use: early cycles, exploratory projects, anytime language is the phenomenon (discourse analysis, narrative work).
Trap: don't let in vivo codes pile up unexamined. "Black box" means something different to a dev than to a support agent. You still have to interpret.
Process coding — catching motion
Gerunds. Always gerunds.
"I keep escalating but nothing changes."
Codes: escalating repeatedly, experiencing stasis
Process coding sees action, not topics. It's the difference between "communication" (topic) and "withholding information," "testing boundaries," "performing transparency" (processes).
Essential for: grounded theory, ethnography, any study where change over time matters.
Values coding — what matters to them
Participants reveal values indirectly. You code for them explicitly.
"I stay late because the team depends on me."
Codes: loyalty, collective responsibility, self-sacrifice
Values coding bridges what people do and what they believe. It's inferential — you're reading between lines — so memo heavily. Why did you call that "loyalty" and not "obligation"? Your memo is your audit trail.
Descriptive coding — the workhorse
Simple. Factual. Low inference.
"Participant describes daily standup meeting structure."
Code: standup structure
Boring? Maybe. Descriptive codes build your index. Even so, they let you find every mention of standups across 30 interviews in 10 seconds. On top of that, always. That said, necessary? Without them, you're scrolling.
Use heavily in cycle one. Pare down in cycle two.
Structural coding — mapping your interview guide
If you asked everyone the same questions, structural coding tags answers by question That's the part that actually makes a difference. Simple as that..
Q: "Tell me about onboarding." → Code: onboarding experience
This isn't analysis. It's organization. But it saves weeks. You can pull "all onboarding segments" instantly and then code them analytically Most people skip this — try not to..
Don't confuse structural codes with analytic ones. They serve different masters.
Pattern coding — the meta-move
Pattern codes group other codes. They're second-cycle work.
You notice: escalating repeatedly, bypassing hierarchy, documenting everything, CC'ing managers — these keep appearing together That alone is useful..
Pattern code: defensive communication practices
Pattern coding is where analysis happens. It's you saying: these distinct behaviors are the same strategy.
Theoretical coding — bringing the big guns
You connect your patterns to existing theory. Not "my theory" — published theory.
Pattern: defensive communication practices
Theoretical code: institutional isomorphism (DiMaggio & Powell)
Or: performativity of compliance (Butler)
Or: bureaucratic ritual (Weber via Graeber)
This move lets you speak beyond your dataset. It's also where hubris lives. Force a theory too early and you'll miss what's actually weird in your data.
Emotion coding — affect as data
"I dread Monday's standup."
Code: dread (not "standup anxiety" — dread)
Emotion coding treats affect as analytic signal, not noise. Consider this: people's feelings about their work shape their work. Coding "frustration," "pride," "resignation," "cynicism" as distinct states — not just "negative affect" — reveals texture That's the whole idea..
Pair with process coding: dreading → avoiding → rationalizing → normalizing
That's a trajectory. That's a finding.
Simultaneous coding — one segment, multiple lenses
Real talk: a single paragraph usually carries five things at once And that's really what it comes down to..
"My manager says she values transparency but hides decisions until they're final, so we stop asking."
Codes: espoused values, information hoarding, decision opacity, learned helplessness, resistance withdrawal
Simultaneous coding means applying multiple codes to the same segment. Here's the thing — your software supports this. Use it. Forcing "one code per chunk" flattens complexity And that's really what it comes down to..
Provisional coding — starting with a list
Sometimes you have a framework going in. A literature review. A logic model. A funder's priorities.
Provisional coding means: here's my starter list. I'll use these, but I'll also add, delete, rename, split Nothing fancy..
Honest provisional coding leaves room for surprise. Dishonest provisional coding forces data into
predefined boxes. The difference is whether you’re discovering or confirming.
Reflexive coding — making space for yourself
Your presence in the data matters. Reflexive coding asks: What’s my role here? Are you an observer? A participant? A critic? Code your own assumptions. Example:
"I kept noticing I flinched when someone said ‘synergy.’ Code: cringe-worthy jargon."
This isn’t navel-gazing. It’s rigor. Acknowledging your interpretive lens doesn’t weaken analysis—it strengthens it But it adds up..
Iterative coding — cycles over linearity
Coding isn’t a one-shot sprint. It’s iterative. You’ll revisit segments, refine codes, merge categories, delete dead ends. A code might start as “meeting fatigue” and evolve into “ritualized disengagement.” Let your data reshape your framework. Tools like Atlas.ti or NVivo track these shifts.
Triangulation coding — cross-checking lenses
After coding, test your findings. Apply three lenses:
- Process: How does this happen? (e.g., "defensive communication practices")
- Structure: What systems enable it? (e.g., "flat hierarchy with no accountability")
- Outcome: What’s the human cost? (e.g., "emotional exhaustion")
Converging insights across lenses validates robustness. Divergence signals gaps—dig deeper.
Visualization coding — seeing the story
Turn codes into maps. Use tools to cluster themes spatially or temporally. A heatmap might show "decision opacity" spikes in Q3. A timeline could reveal "learned helplessness" growing after leadership changes. Visuals expose patterns words alone miss.
Conclusion
Coding is not a finish line—it’s a dialogue. Between data and theory, between structure and meaning, between observation and interpretation. The best analyses emerge when you treat codes as hypotheses, not labels. When you code “defensive communication practices” as a pattern, then tie it to institutional isomorphism, then trace its emotional toll, you’re not just describing a workplace. You’re revealing a system in motion Simple as that..
Your findings will never be perfect. But they’ll be alive—if you let the data speak, the theories challenge, and the emotions resonate. That’s where insight lives: in the friction of making sense of the messy, human work of work.
