How to Find and Write Themes Across Multiple Studies in a Literature Review

First, a theme is a claim, not a topic

Most weak literature reviews are not short of themes. They are short of claims. “Online learning,” “student wellbeing” and “barriers” are topics: they tell the reader which box a paper goes in, and nothing about what the papers in that box found. Braun and Clarke (2006) describe a theme as something that captures an important pattern in relation to your research question. In a literature review, the pattern is across studies, and the theme should say what it is.

A quick test: write each theme as a one-sentence thesis statement, the way you would state the argument of an essay. If you can't, you probably have a topic. That doesn't make topics useless. They make good folders while you read. They just can't be the headings of the finished review. If the code/theme distinction itself is fuzzy, our guide on the difference between a code and a theme walks through it.

Step 1: Fix the question and what you will code

Themes only exist in relation to a question. “What does the literature say about first-generation students?” will give you thirty candidate themes and no way to choose between them. “What shapes first-generation undergraduates' sense of belonging in their first year?” gives you a filter: every finding either helps answer it or doesn't.

Then decide what counts as data. In a thematic synthesis you code what studies found. Thomas and Harden (2008) took this to mean all text labelled “results” or “findings”. Journal articles often restate and interpret findings in the discussion and conclusion as well, so most reviewers code those too. Leave the introduction and literature review of each paper out. Those sections summarise other people's work, and coding them means counting the same finding twice. Methods sections are useful too, just not as findings: you will need them in step 3.

How many studies? Enough to cover the concepts your question needs. Thomas and Harden made the point that a synthesis doesn't change if ten studies rather than five report the same concept, so what matters is whether new papers still add new ideas. Our guide to thematic analysis for a literature review covers sample size and saturation in more depth.

Step 2: Code each study's findings

A paper is too big a unit to compare. “Okafor and Reid agree with Haddad” is almost never true across a whole paper; it is true about one finding. So break each study down. Read the findings and give every distinct result a short label, a code, of two to five words. Thomas and Harden call this line-by-line coding. For a dissertation, coding each sentence or paragraph that carries one finding is usually enough.

Two habits save weeks later. First, keep the evidence with the code: the quote, or a close paraphrase, and the page number. When your supervisor asks which studies back a theme, you want to answer in seconds, not spend an afternoon re-reading PDFs. Second, reuse codes across papers. When a finding in paper four matches a code from paper one, use the same label. That reuse is how patterns start to show.

Here is what that looks like for one illustrative study (the six studies in this guide are invented for teaching, though they reflect patterns common in real research on first-generation students):

The mechanics are the same as coding interview data, so if you haven't coded before, our step-by-step guide to coding interview transcripts covers code length, code types and building a codebook.

Step 3: Build a theme-by-study matrix

Once you have coded four or five studies, group codes that seem to belong together into candidate themes and build a grid: studies down the side, candidate themes across the top. This is the literature review matrix, also called a synthesis matrix or literature review table. Webster and Watson (2002) argued for exactly this kind of concept matrix as the way to turn an author-by-author review into a concept-centred one.

Each cell gets a short note on what that study found about that theme, with a page reference. Two rules matter more than the layout. Leave a cell blank if the study says nothing on the theme, because blanks are data: a column that is mostly empty is either a weak theme or a gap. And add columns for method, sample and setting. They look like admin, but they are what explains disagreement later.

Illustrative studies. In a real matrix, keep a column for every candidate theme, including the ones you later drop.

A spreadsheet is enough for up to twenty or so studies. Keep one sheet per review question, freeze the header row, and resist the urge to write a paragraph in each cell. Two lines and a page number is plenty. If the matrix gets wider than ten theme columns, some of them are probably codes that belong under a bigger heading.

Step 4: Read down the columns to see how studies relate

Reading across a row tells you about one paper, which you already knew. Reading down a column is where synthesis starts. For each column, ask what the studies have in common, where they part company, and whether together they show something none of them says alone.

Noblit and Hare (1988), who developed meta-ethnography, gave names to the three ways studies can relate, and they are worth borrowing even if you are not doing a meta-ethnography:

• Reciprocal: the studies say broadly the same thing in different words. “Everyone else had the manual” and “a performance with rules learned at school” are the same finding, which is why they share a column.
• Refutational: one study contradicts another. Moreno's commuter students worked long hours and still felt they belonged, against Okafor and Reid and Haddad et al.
• Line of argument: each study shows part of a larger picture. Put the columns together and a bigger claim appears: belonging is built from time and tacit knowledge that universities assume students already have.

The refutational cases need the most care, and most guides stop at “note where studies disagree.” Disagreement comes in different kinds, and each kind means something different for the theme. This is the classification our tool uses when it compares studies, and it works just as well by hand:

The design column earns its place here. Once you look at it, Moreno's result stops looking like a contradiction. Moreno studied a commuter campus where students didn't expect a campus social life in the first place. That is a difference of scope, and it changes the theme from “paid work harms belonging” to something more precise.

Step 5: Test each candidate theme before you write it

A column in a spreadsheet is not yet a theme. Before a candidate becomes a section of your review, put it through these checks:

1. Is it a claim? Write it as one sentence. “Family” fails. “Family anchors first-generation students and pulls against them at the same time” passes.
2. Does more than one study support it? A theme built on one paper is a finding from that paper. Fold it into a broader theme, or flag it as something the literature has barely looked at.
3. Is it carried by weak evidence? Look at which studies fill the column. GRADE-CERQual, the approach Cochrane reviewers use to judge confidence in qualitative synthesis findings, looks at methodological limitations, coherence, adequacy of data and relevance (Lewin et al., 2018). You don't need the full procedure for a dissertation, but asking those four questions of each theme will stop you overstating one.
4. Does it answer the review question? Interesting themes that don't bear on the question go in a footnote or nowhere.
5. Is it distinct from the other themes? If two themes keep citing the same studies for the same point, merge them.

The themes that survive are usually what Thomas and Harden call descriptive themes: they stay close to what the studies reported. The strongest reviews take one more step and build analytical themes, which answer your question in a way no single study does. In their own worked example, 36 codes from eight studies became 12 descriptive themes and then 6 analytical ones. In ours, the line-of-argument reading from step 4 is the analytical theme: belonging depends on time and tacit knowledge the institution assumes students bring with them. That sentence is not in any of the six papers. It is yours, and it is what makes the review worth reading.

Step 6: Write each theme up

Each theme becomes a section, or in a short review a paragraph or two. A pattern that works for most of them:

1. Open with the claim, in your words, not an author's name.
2. Show the convergence. Cite the studies that support it together, and say what kind of studies they are (three qualitative studies in three countries carries more weight than three from one department).
3. Deal with the divergence, using the type you identified in step 4.
4. Say what it means for your question, and where the evidence runs out.

Here is the same material written both ways. First, the version that gets sent back:

Summary

Okafor and Reid (2021) interviewed 24 first-generation students at a UK university and found they were unsure how seminars and office hours worked. Haddad et al. (2022) ran focus groups in the US and found similar results. Moreno (2023) interviewed 15 commuter students and found that paid work did not affect their belonging. Chen and Abara (2022) found that family was important.

Every sentence starts with an author. Nothing connects one study to the next, and the reader is left to work out whether Moreno disagrees with the others or not. Now the synthesised version:

Synthesis

First-generation students' belonging seems to depend less on wanting to take part than on knowing how. In qualitative studies from the UK, the US and Australia, students described seminars, office hours and assessment as run on rules nobody explained (Okafor and Reid, 2021; Haddad et al., 2022; Chen and Abara, 2022), and a survey of 1,140 Swedish students found the same gap on a hidden-curriculum scale (Lindqvist, 2020). Paid work makes this worse where it removes the informal time in which such rules are usually picked up (Okafor and Reid, 2021; Haddad et al., 2022). Moreno's (2023) commuter students are the exception: they worked long hours and still reported belonging, but located it in the classroom and at home rather than in campus social life. The difference looks like one of scope rather than a contradiction, and it exposes a gap. Almost all of this evidence comes from residential campuses, so it is not clear that universities serving commuter students should be building belonging in the same places.

Try deleting every citation from the second version. It still reads as an argument, because the claims are yours and the studies are evidence for them. Delete the citations from the first version and nothing is left. That test is quicker than any checklist.

Sentence starters that signal synthesis

Putting the themes in order

The order of themes is itself an argument. A simple rule for a dissertation: go from what is settled, to what is contested, to what is missing, so the last theme hands over to your study. In the worked example, that means unwritten rules (well supported), then paid work (supported, with a scope limit), then family (mixed mechanisms), and finally the analytical theme and the commuter-campus gap. Other ways to organise a thematic review, such as hanging the themes on a theoretical framework, are covered in our complete guide to thematic literature reviews.

Our guide on how to find a research gap covers the different kinds of gap and how to argue for one, and the dissertation guide shows how the review connects to the rest of the thesis.

Does the type of review change anything?

The process above works for most reviews, but the expectations around it differ. Grant and Booth (2009) identified 14 types of review, and Snyder (2019) groups the common ones into systematic, semi-systematic and integrative approaches. What that means for finding themes:

• Narrative or traditional review (most dissertation chapters): the steps here as written. You choose the studies, so say briefly how you chose them.
• Systematic review of qualitative studies: the same steps, but with a documented search and screening process, quality appraisal of every study, and a named synthesis method such as thematic synthesis or meta-ethnography.
• Scoping review: maps what research exists. Themes tend to stay descriptive, and that is acceptable, because the aim is coverage rather than new interpretation.
• Integrative review: mixes qualitative and quantitative studies. The matrix helps most here, because the design column stops you treating a survey correlation and an interview theme as the same kind of evidence.

If you are choosing between synthesis methods, our overview of qualitative analysis frameworks compares thematic synthesis with the alternatives.

Where AI fits, and where it doesn't

Steps 2 and 3 are where the weeks go. Coding the findings of twenty papers and copying each finding into the right cell with a page number is slow, mechanical work, and it is the part AI tools do reasonably well. Steps 5 and 6 are different. Deciding that Moreno's result is a scope difference rather than a contradiction, or seeing the line of argument across the columns, is the analysis, and it has to be yours. A chatbot summary of your PDFs skips straight to prose and gives you citations you cannot check, which is the worst of both. Our review of AI tools for literature reviews covers what the research says about accuracy.

thematicanalysis.ai was built for the middle of this process. Its literature review synthesis reads the findings, discussion and conclusion of each paper you upload, codes them, and gives you cross-study themes with a theme-by-study matrix. For every theme it lists where studies differ and labels each difference as emphasis, scope, degree, mechanism, methodological or contradiction, the same types as the table above. Every code links to a quote checked against the paper it came from. Treat it as a first draft of your matrix, then do steps 5 and 6 yourself. If your supervisor asks how you used it, our guide to reporting AI-assisted analysis has wording for the methods section.

Upload three papers below and compare its matrix with your own. The first three studies are free.

Frequently asked questions