Opening Remarks
Synthesis Process
You now have a screened, appraised set of papers, the raw material for a review. Turning that material into a review is the work still ahead of you.
Search and screening are necessary logistics; synthesis is the unit’s actual intellectual work, turning a set of individually-read papers into a single, defensible argument about what the field currently knows, and that argument is the actual contribution your review makes.
Learning Outcomes
By the end of this unit, you will be able to:
- Build an extraction template and a coding scheme, deductive, inductive, or abductive, chosen to fit your review type rather than a generic default.
- Build and read a concept-centric matrix, by column rather than by row, to synthesize what a corpus of papers collectively knows rather than what each paper individually said.
- Diagnose a contradiction between two papers as a clue to an unnamed contingency factor, and state the boundary condition that resolves it as your review’s synthetic contribution.
Data Extraction
Extraction Judgment
Your extraction template encodes a theoretical decision about what matters in these papers.
What you choose to extract already encodes a theory of what matters about these papers, well beyond mechanically copying the author, year, method, and main finding into a spreadsheet: an extraction template with a column for “method” but not for “boundary conditions” decides, implicitly and before you have read a single paper’s findings closely, that method matters to your synthesis and boundary conditions do not.
Build your extraction template from your research question and your review type decision, not from a generic template. What would you need recorded, from every paper, to actually answer your question once you have read them all? That list of fields is your template.
Coding Schemes
Deductive, Inductive, Abductive
The same three modes of theorizing you already know, applied to a corpus instead of a single observation.
What is Theorizing introduced these as general reasoning modes. Applied to coding a literature:
- Deductive coding: you bring a pre-existing framework (a theory, a typology) and code each paper against its categories. Fits when your review type is testing or extending a specific, named theoretical lens.
- Inductive coding: you let codes emerge from the papers themselves, building categories bottom-up as patterns appear. Fits scoping and critical reviews of a young or conceptually unsettled domain, where imposing a pre-existing framework would obscure exactly the disagreement you are trying to surface.
- Abductive coding: you start with a rough framework, but revise it as the corpus surprises you, an iterative back-and-forth between your categories and the data. The most common mode in practice, and usually the most honest one to report.
Your choice here should follow directly from your SLR I review type decision, a meta-analysis has little use for inductive coding, a scoping review has little use for a rigid deductive scheme borrowed from a theory the field hasn’t agreed on yet.
The Concept Matrix
Concept-Centric Review
This is the single most consequential move in your synthesis, and the one most teams miss.
Webster & Watson (2002) distinguish two ways to organize a review’s presentation. An author-centric review walks through the literature study by study: “Smith (2020) found X. Jones (2021) found Y. Lee (2022) found Z.” A concept-centric review is organized around the concepts and constructs themselves: “Regarding construct A, three studies find X, while two studies, conducted in [context], find not-X instead.”
The author-centric version is easier to write, it mirrors the order you read the papers in. It is also close to useless as a synthesis: it tells the reader what each study said, but not what the field, taken together, actually knows. The concept-centric version forces you to have actually synthesized, because you cannot write it without first grouping findings by what they are claims about, not by who made them.
Build a matrix: rows are your included papers, columns are the concepts or constructs your extraction identified as relevant. Fill each cell with what that paper says about that concept. Once the matrix is filled, read it by column, not by row, and write your synthesis from that column-wise reading. That reading order is what makes the difference between an author-centric summary and a genuine concept-centric synthesis.
Filled Matrix
A small worked matrix, three invented papers, two concepts, before you build your own:
| Governance model | Adoption driver | |
|---|---|---|
| P1 | Centralized IT governance board approves every AI agent deployment | Regulatory pressure named as the primary driver |
| P2 | Federated governance, business units decide independently | Competitive pressure from early-adopting rivals |
| P3 | No formal governance structure reported | Cost reduction named as the primary driver |
Read this by column, not by row. The “Governance model” column, read down, already tells a synthesis story: three papers report three different structures, from centralized to entirely absent, which is a finding about the corpus, not three isolated facts about P1, P2, and P3. The “Adoption driver” column shows the corpus disagrees on why organizations adopt at all, regulation, competition, or cost, exactly the kind of disagreement the contradictory-findings exercise below asks you to explain rather than average away.
Compare this to reading by row: “P1 has centralized governance and cites regulatory pressure” is a fact about P1 alone. It is not yet a synthesis of what the corpus knows about governance, or about adoption drivers.
Synthesis Techniques
Technique Selection
- Thematic synthesis: identify recurring themes across the corpus, inductively, and organize the narrative around them. Common for qualitative systematic and scoping reviews.
- Framework synthesis: apply an existing conceptual framework as the organizing structure for the synthesis, useful when your review is deductively testing or extending a specific theory.
- Narrative synthesis: a structured, but non-statistical, textual synthesis, common when the included studies are too heterogeneous in method or context to pool numerically, but a purely thematic organization would lose important comparative structure.
Each of these three is a deliberate technique with its own logic, matched to a review type from SLR I, not a fallback for “not doing a meta-analysis.” A critical review typically needs a narrative or framework synthesis that can carry an evaluative argument. A scoping review typically needs a thematic synthesis that maps rather than argues. Pick the technique your review type actually calls for, and say so explicitly in your write-up, the same way you already stated and justified your review type.
Contradictory Findings
Contingency Factors
A contradiction in the literature is a clue that points toward a contingency factor. Resist the urge to average it away.
You met this idea in What is Theorizing, where Bacharach (1989)’s own criterion for a good theory was that it reconciles contradictory findings by discovering the contingent factors that explain why one study found X and another found not-X; your review is where you apply that criterion in practice rather than treat it as an abstract standard.
When two papers in your corpus disagree, resist the instinct to report both and move on, or to quietly treat the majority finding as “the” finding. Ask instead: what differs between these two studies, context, sample, operationalization, time period, that might explain why they found different things? If you can name that contingency factor, you have made a genuine synthetic contribution, exactly the kind that improves the field’s explanation, one of the three routes to contribution from Knowledge, Science and Epistemology.
Synthetic Contribution
Contribution Threshold
Noticing a pattern only describes what the literature says. Explaining why the pattern holds, and where it stops holding, is what makes it a contribution.
“Twelve of the fifteen papers in our corpus find X” is an observation; it becomes a contribution once you can say something about why, what mechanism would produce that regularity, and where the boundary is, under what conditions you would expect the pattern to break down, and whether the remaining three papers that found not-X actually sit at that boundary.
This is the same move as the contingency-factor exercise above, generalized: a synthesis is finished only once you can state a claim, with its boundary conditions, that the field did not already have stated this clearly before your review put the pieces together.
Homework
Produce your concept matrix, version one.
Create your full concept matrix, and for each concept column, one written synthesis statement, including, where the corpus disagrees, your best account of the contingency factor that explains the disagreement.