Coding data in qualitative reviews is the process of reading the findings of included primary studies and attaching short, meaningful labels, called codes, to segments of text, so that recurring concepts can be grouped, compared across studies, and built into the descriptive or analytical themes that form a qualitative synthesis. It is the qualitative counterpart to numerical extraction: instead of pulling numbers into a form, you pull meaning into a coding structure.
How coding differs from numerical extraction
In a quantitative review you extract fixed values with a piloted form, as we cover in data extraction. Qualitative coding is interpretive: the unit of analysis is a passage of text, the author’s findings, participant quotes, or the themes the original authors reported, and the “data” you carry forward is meaning rather than a measurement. That interpretive step is why coding needs its own discipline, an audit trail, and usually more than one coder, even though it sits in the same extraction slot of the wider systematic review process.
The coding workflow, step by step
Decide what counts as data
Qualitative reviews usually code the findings sections of primary studies, including the authors’ interpretations and the participant quotes they present. Settle this scope in your protocol, the same place you fix your review question and frame it with a structure such as SPIDER or PCC, so every coder knows which text is in scope.
Choose inductive, deductive, or both
Inductive coding lets codes emerge from the text, useful when you want the synthesis to stay close to what studies actually said. Deductive coding starts from a pre-defined framework and sorts text into it, useful when you are testing or extending an existing model. Many reviews blend the two, beginning deductively and adding inductive codes for what the framework misses. State which approach you used so the method is transparent.
Build a coding frame and apply it
As codes accumulate, organise them into a coding frame: a structured list of codes with definitions and examples, grouped into categories. This is the qualitative equivalent of a data extraction form template and its codebook, and like that form it should be piloted on a few studies and refined before you code the full set.
Keeping qualitative coding trustworthy
Code with more than one reviewer
Because coding is interpretive, a second coder guards against one person’s reading dominating. Have two reviewers code a shared sample independently, compare how they applied the codes, and discuss divergences, the same duplicate-and-reconcile logic that underpins duplicate full-text screening. The aim is not forced uniformity but a defensible, agreed reading.
Resolve and document disagreements
Treat coding disagreements the way you treat screening conflicts: discuss them, agree a resolution, and record it. Keeping a clear audit trail of how codes evolved and how disputes were settled is what lets a reader judge the rigour of the synthesis.
Move from codes to themes
Once coding is complete, cluster related codes into categories and lift them into themes that answer your review question. This thematic step is where qualitative synthesis parts company with a pooled estimate, and it sits alongside the narrative synthesis family rather than a meta-analysis of effect sizes.
Matching coding to the synthesis method
Coding is not a single technique; it is shaped by the qualitative synthesis method you committed to in the protocol, and choosing the method first keeps the coding coherent. Three families dominate:
- Thematic synthesis codes the findings line by line, groups codes into descriptive themes, and then develops more interpretive analytical themes that go beyond the primary studies. It is the most widely taught approach and pairs well with inductive coding.
- Framework synthesis starts from an existing conceptual framework and charts coded data into a matrix, which suits deductive coding and questions where a model already exists to test or extend.
- Meta-ethnography translates the concepts of one study into those of another to build a higher-order interpretation, demanding a richer, more conceptual coding than a flat label.
These methods sit within the broader family of review designs and produce a synthesis closer to a mapping or interpretive review than to a pooled effect. Naming the method up front tells coders how granular and how interpretive their codes should be.
Practical coding decisions that shape the result
Set the unit of coding
Decide whether you code at the level of the line, the sentence, the paragraph, or the complete finding, and apply that unit consistently. Coding a whole paragraph under one label loses nuance; coding every clause fragments the meaning. Most reviews settle on the discrete finding or the participant quote as the natural unit.
Judge when coding is saturated
As you work through studies, new codes appear quickly at first and then taper. When several consecutive studies add no new codes, you are approaching thematic saturation, a signal that the coding frame is stable. Recording where saturation occurred is useful evidence that the analysis was complete rather than cut short.
Use software, but let the analysis lead
Qualitative analysis software such as NVivo, ATLAS.ti, or the open-source Taguette helps store codes, retrieve coded segments, and track how the frame evolves, which is invaluable for the audit trail. The tool organises the coding; it does not perform the interpretation, and a tidy software project is no substitute for a defensible, agreed reading between coders.
Common pitfalls in qualitative coding
- Coding only the abstract or discussion rather than the full findings, which strips out the participant voice the synthesis depends on.
- Letting the frame sprawl into dozens of near-duplicate codes that never cluster, instead of merging them as patterns emerge.
- Forcing premature consensus between coders, which buries a genuine interpretive difference that should have been explored.
- Treating frequency as importance, so a concept mentioned in many studies outranks a rarer but more meaningful finding.
- Losing the chain of evidence, so a coded theme can no longer be traced back to identifiable text in the primary studies, which breaks the audit trail a reader needs to trust the synthesis.
Reporting the coding process
Whatever method you use, report it transparently: the coding approach, who coded, how the frame was developed, and how disagreements were resolved. Appraising the underlying studies still matters too, with the appropriate quality appraisal for qualitative research, so the confidence in your themes reflects the strength of the evidence behind them rather than the coder’s preferences. A well-reported synthesis also states the confidence in each finding using an approach such as GRADE-CERQual, which judges how much trust to place in a review finding given the methodological limitations, coherence, adequacy, and relevance of the studies behind it. If a qualitative or mixed-methods evidence synthesis is more than your team can carry, our evidence synthesis support covers the coding, the second-coder reconciliation, and the write-up.