Framework Analysis: the matrix method for applied research

In most thematic approaches the analytic object becomes the theme: all the material about trust, gathered from everyone. What is lost is the case — you can no longer easily see what one participant said across all themes, or how their account hangs together.

Framework analysis keeps both axes. Each row is a case, each column a theme or subtheme, and each cell holds a summary of that case's data on that theme, with the source reference retained. You can read down a column for the range of views on one theme, or across a row for one person's whole account, and you can always get back to the transcript.

That two-way readability is why the method suits applied research. A commissioner asking "what did the rural practices say about referral delays?" gets an answer that is traceable to specific cases rather than to an aggregate. It also makes team working practical: analysts can chart different transcripts into the same framework and the results are comparable, which is much harder to achieve with open coding.

The method was developed by Jane Ritchie and Liz Spencer at the UK's National Centre for Social Research (then SCPR) for government-commissioned policy work. Their foundational chapter, Qualitative Data Analysis for Applied Policy Research, sets out the five stages still used today.

It reached a much wider audience through Gale and colleagues' Using the framework method for the analysis of qualitative data in multi-disciplinary health research, which is the most practical published account and the one most health researchers work from.

Spencer, Ritchie and O'Connor's In Pursuit of Quality addresses how quality in this tradition should be judged, which matters because the method's procedural clarity is sometimes mistaken for a guarantee of analytic quality.

A note on terminology: framework analysis is this method. It is not the same as using a theoretical framework, and it is not the "Framework" feature in a particular software package, though several implement the matrix.

The four rows above are a fragment of a 34-case chart from a study of a new referral pathway. Reading down the awareness column shows a pattern that a coded-fragment analysis would have muddled: awareness is not a rural-urban split but an employment one — the locum had never been briefed, and locums were a fifth of the sample.

Reading across GP-02's row shows something the column view hides. That case is internally coherent: an isolated practitioner who knows the pathway exists, cannot use it as designed because of a genuine geographic constraint, and has built a personal workaround that works for them and is invisible to the system.

Comparing the workaround column against the barriers column produced the study's central finding: every workaround was a rational response to a specific, identifiable constraint, and none of them was recorded anywhere. The pathway's monitoring data showed high compliance because the workarounds all terminated in a compliant-looking referral.

That is what the matrix is for. The finding came from reading columns against each other and rows for coherence — an operation that is natural in a chart and awkward in a list of coded extracts.

• Quoting instead of summarising. Cells full of pasted transcript make the chart a reorganised dataset, not an abstraction.
• Summarising to nothing. "Positive", "negative", "mixed" strips out everything that made the case worth reading.
• Losing the source reference. The audit trail is the method's main claim. A cell you cannot trace back has forfeited it.
• A framework fixed too early. Built from three transcripts and never revisited, it will not fit the other ninety.
• Stopping at the chart. Charting is stage four of five. The analysis happens in stage five.
• Treating procedural transparency as analytic quality. A perfectly documented framework can still be superficial. This is the criticism the method most needs to answer.
• Using it where the question is about meaning-making. Framework analysis summarises content well and is a poor fit for how someone made sense of something.

The distinguishing quality question for framework analysis is whether the chain from transcript to conclusion is genuinely traceable — cell to source, theme to framework, finding to matrix. A reader should be able to pick any claim in the report and follow it back.

Beyond that: was the framework revised as the data demanded, or applied unchanged? Are deviant cases identified and discussed rather than averaged away? Does the report show cross-case interpretation rather than theme-by-theme summary? And, in team studies, is there any evidence that the analysts meant the same thing by the same category?

Framework analysis is descriptive by temperament. It is very good at organising what people said and much less naturally suited to theory generation — grounded theory approaches exist because this kind of structured summarising does not readily produce new concepts.

The matrix also imposes a comparability assumption: every case gets every column. Where accounts are genuinely incommensurable, that structure forces material into cells it does not belong in, and the empty cells stop being informative.

And the summarising step loses the texture of how something was said. For questions where the language is the data, this is a serious loss and a different method is required.

Template analysis is the closest relative — also a structured coding framework applied and revised across a dataset — but organised hierarchically around themes rather than into a case-by-theme matrix.

Directed content analysis starts from theory-derived categories in a similar spirit, with less emphasis on the case axis.

Reflexive thematic analysis trades the audit trail for interpretive depth and does not attempt case-level comparability.

Qualitative Comparative Analysis is the formal next step when the matrix suggests that combinations of conditions, rather than single themes, explain the outcome.

This is the method most improved by tooling, because the matrix is genuinely painful to maintain by hand. Thirty-four cases by twenty subthemes is 680 cells, each needing its source reference kept live, and each needing to update when the framework is revised at case sixty.

A tool that holds the matrix, keeps the link from cell to transcript, and lets a team chart in parallel into one shared framework removes most of the method's administrative cost without touching its analytic core. Evidano supports framework analysis as a named methodology and builds the case-by-theme matrix with source references retained. The summarising in each cell is analytic work — that is where the abstraction happens, and it is the analyst's judgement.