Guides / nvivo coding
Coding in NVivo
Coding is where the analysis happens; everything else in NVivo is filing around it. This covers the mechanics of coding to nodes, how to name codes so they stay useful at transcript thirty, and when to impose a hierarchy.
Open a source, select a passage, then right-click and choose Code > at New Node, or drag the selection onto an existing node. Nodes are NVivo's term for codes and each coded passage is called a reference. You can code the same passage at several nodes, and organise nodes into parent and child hierarchies once patterns emerge.
How to code
The mechanics
Coding on first contact produces codes that describe sentences rather than meaning. Read the whole transcript, then code it.
A phrase, a sentence, a paragraph — whatever carries the idea. Selecting too little loses context; selecting whole pages makes retrieval useless.
Or drag the selection onto an existing node. Ctrl+F2 on Windows codes straight to a new node.
Overlapping coding is normal, not an error. NVivo tracks each independently.
Selecting a participant's own words and choosing Code → In Vivo creates a node named with that exact phrase. Useful in early open coding when you want the codes to stay close to the participant's language.
Naming
Naming codes that survive
At transcript five you remember what every code meant. At transcript thirty you will not, and a code you cannot define is a code you will apply inconsistently.
| Weak name | Better name | Why |
|---|---|---|
| Support | Seeking support without appearing to need it | Names a process, and carries the tension |
| Time | Rationing time against competing demands | Says what is happening, not the topic |
| Negative | Describing the change as imposed | Describes content, not your reaction to it |
| Barriers | Treating constraint as external and fixed | Specific enough to apply consistently |
It is the only thing that makes coding reproducible, and it is what you will paste into your methods section when asked to describe the coding framework. Filling them in as you go costs seconds; reconstructing forty of them afterwards is a day's work.
Structure
When to build a hierarchy
NVivo lets you nest nodes under parents by dragging. The temptation is to design that structure up front — and it is worth resisting.
An early hierarchy forces material into categories you have not yet earned from the data. Let the flat list grow through the first several transcripts. Fifty to eighty codes is normal. Only when the list stops growing should you group them, and group by what they share analytically, not alphabetically or by interview question.
| Stage | What the node list looks like |
|---|---|
| Transcripts 1–5 | Flat, growing fast, some overlap and redundancy |
| Transcripts 6–15 | Growth slowing, duplicates becoming visible |
| Then | Merge duplicates, group into parents, revisit earlier coding |
| Remaining transcripts | Coding to a settled structure, adding only where genuinely new |
Dragging one node onto another merges their references rather than deleting anything. If two codes turn out to be the same idea, merge them — you lose nothing, and the retrieval improves.
Send a few transcripts and your node list. A named researcher tests the codebook against your extracts and advises on structure before you code thirty more.
See qualitative codingReviewing
Coding stripes and density
Coding stripes are the coloured bars NVivo shows down the right margin of a source, one per node applied. Turn them on from the ribbon under Coding Stripes.
Coding everything is a common early habit and it defeats the purpose — retrieval only helps if it narrows. Code what bears on your research question, and be willing to leave material uncoded.
Reliability
Checking consistency
If your method calls for inter-rater reliability, NVivo computes it directly. Explore → Coding Comparison compares two coders on the same sources and reports percentage agreement and Cohen's kappa.
| Kappa | Interpretation |
|---|---|
| < .40 | Poor to fair |
| .41–.60 | Moderate |
| .61–.80 | Substantial |
| > .80 | Almost perfect |
Note that percentage agreement on its own overstates reliability badly, because it ignores agreement expected by chance. Where one code dominates, two coders can agree 85% of the time and still have a kappa near .17. Report kappa.
For content analysis and codebook approaches, inter-rater reliability is appropriate and expected. For reflexive thematic analysis, the researcher's interpretation is the analytic instrument rather than a source of error, and Braun and Clarke argue explicitly against reporting a coefficient. Match the check to the method.
Pitfalls
Five mistakes
1. Building the hierarchy before coding
It forces the data into a structure you have not earned. Let the flat list grow first.
2. Single-word topic codes
They become containers for anything loosely related and stop discriminating.
3. Leaving node descriptions blank
Without them the coding is not reproducible and you have no codebook to describe.
4. Coding everything
Retrieval only helps if it narrows. Uncoded material is fine.
5. Reporting kappa for reflexive thematic analysis
It misrepresents what the method claims. Use trustworthiness criteria instead.
Answers
Frequently asked questions
What is a node in NVivo?
A node is NVivo's word for a code — a container holding every passage tagged with a particular idea. Each coded passage is called a reference. Nodes can be organised into parent and child hierarchies once patterns in your coding emerge.
How do I create a code in NVivo?
Select a passage in a source, then right-click and choose Code > at New Node. On Windows, Ctrl+F2 codes directly to a new node. You can also drag a selection onto an existing node in the Navigation View.
Should I build a node hierarchy before I start coding?
No. An early hierarchy forces material into categories you have not earned from the data. Let the flat list grow through the first ten to fifteen transcripts, then merge duplicates and group codes by what they share analytically.
How many codes should I have in NVivo?
Fifty to eighty after the first several transcripts is entirely normal for an inductive project. What matters is that the list stops growing as you continue, which signals you are reaching saturation, and that each code has a description precise enough to apply consistently.
Can two people code the same project in NVivo?
Yes. NVivo tracks coding by user, and Explore > Coding Comparison reports percentage agreement and Cohen's kappa between two coders on the same sources. Whether you should report that coefficient depends on your method.
What are coding stripes in NVivo?
Coloured bars in the right margin of a source showing which nodes have been applied to each passage. They make it easy to spot uncoded sections, over-coded sections, and coding density across a transcript.