ACADEMIC WRITING GUIDE
How to Write a Narrative Synthesis for a Systematic Review Dissertation (With Worked Examples)
A dissertation systematic review can reach the synthesis stage and discover that its included studies cannot responsibly be combined into one pooled effect. Perhaps the studies measure the same broad outcome with incompatible instruments, report different follow-up periods, omit usable precision estimates, or examine interventions too different to treat as a single comparison. The answer is not to abandon synthesis or simply write one paragraph about each paper. The answer is to use a transparent synthesis method suited to the available evidence—and report what the method can and cannot establish.
This guide explains how to plan and write a narrative synthesis in a systematic review dissertation, with a worked example, an evidence matrix, an example Methods paragraph, a model Results section and a structured reporting checklist. It focuses on quantitative intervention-effect studies that cannot be pooled, the setting addressed by the Cochrane Handbook's guidance on synthesis using other methods and the Synthesis Without Meta-analysis (SWiM) reporting guideline.
Teaching example: The studies, labels, instruments, observations and judgments below are entirely fictional. They do not describe actual experiments or establish whether an educational intervention works. Any wording you adapt for a dissertation must be replaced with findings supported by your own verified sources and review protocol.
What is narrative synthesis in a systematic review?
Narrative synthesis refers broadly to explaining patterns in findings across studies using structured text, tables and sometimes visual displays, rather than relying solely on a pooled effect estimate. But the label is not a sufficiently precise Methods description. Reviewers should identify the actual synthesis operations they performed: for example, grouping studies by outcome and comparator, tabulating effect estimates, describing the range of observed effects on a common scale, or—where justified—using direction-of-effect vote counting.
Cochrane expressly cautions against treating “narrative synthesis” as an unexplained substitute for a named method. Transparent reporting lets a reader reproduce the grouping and understand why some results are emphasized while others are not.
Narrative synthesis also differs from the ordinary thematic discussion of literature found in many dissertation background chapters. A systematic review has explicit eligibility criteria, a documented search and selection process, prespecified outcomes, study-level appraisal and a traceable synthesis. Our guide to writing a literature review without merely summarising sources explores a related writing skill, but it is not a substitute for systematic-review methods.
When should you synthesize without a meta-analysis?
Meta-analysis can be valuable when studies address a sufficiently coherent comparison and provide effect estimates and other information appropriate for statistical combination. It is not automatically suitable because two or more studies exist. Equally, different instruments or some heterogeneity do not automatically rule out pooling: defensible transformations or model choices may sometimes make synthesis possible. Consult an appropriate statistician or methods adviser before declaring pooling impossible.
| Observed problem | What to check first | A defensible alternative when pooling is not appropriate |
|---|---|---|
| Incompatible outcomes | Are constructs, measurement direction and follow-up genuinely comparable? | Separate outcome groups and present a structured effects matrix |
| Incomplete quantitative reporting | Can missing statistics be recovered, derived appropriately or obtained from authors? | Tabulate all available results; transparently explain missing information |
| Substantial clinical or methodological diversity | Would a pooled mean answer a meaningful question? | Group by intervention, population or context and interpret differences cautiously |
| Effect estimates available but not commensurate | Is conversion to a shared metric valid and justified? | Report estimates in meaningful, clearly separated groups |
| Direction reported but effect magnitude unusable | Is a consistent definition of benefit versus harm available? | Consider explicitly defined direction-of-effect synthesis, with its limitations |
The decision to avoid meta-analysis should be reasoned and documented, preferably using a contingency plan established in the protocol. It should not depend on whether the available studies appear to show a significant or desirable result. The Cochrane guidance on preparing studies for synthesis recommends planning how evidence will be grouped and what alternative synthesis methods will be used if the preferred method is infeasible.
A transparent synthesis begins before writing the Results chapter
A credible non-pooled synthesis has several stages, even though its final presentation is mainly text.
Step 1: Fix the review question and synthesis groups
Start with the population, intervention, comparator, outcome and follow-up period. If the dissertation includes several outcomes, avoid combining them under one broad heading such as “academic performance” without explaining important distinctions. A short-term quiz score is not identical to end-of-year course completion. Decide which measurements belong together and explain the rules.
For example, a review of structured instructor feedback in undergraduate courses might prespecify separate syntheses for course completion, assessment performance and student confidence. It could also group studies by whether the comparator is usual instruction or another active feedback approach. Grouping on the basis of favorable results after you see them is a major source of selective interpretation.
Our guide to conducting a systematic literature search addresses how to establish a traceable evidence base before synthesis. Study selection can then be documented using a PRISMA 2020 flow diagram.
Step 2: Build a study-by-study extraction matrix
Record enough information to support every comparative claim. At minimum, identify study design, population, intervention, comparator, outcome definition, time point, effect estimate if available, uncertainty information, result direction, risk-of-bias judgment and missing data. Record which study result is eligible for each synthesis: one study may supply a valid result for one outcome and not another.
A matrix prevents a common error: treating the same study as though it were several independent confirmations because it reports several measures. Where multiple correlated outcomes or follow-ups exist, explain which result is used in the primary synthesis and how alternatives will be addressed.
Step 3: Choose the synthesis method that answers your question
There is no single standardized statistical procedure called “narrative synthesis.” Common approaches include structured tabulation, describing the distribution or range of effect estimates on a comparable scale, displaying individual effects without a pooled diamond, and direction-of-effect synthesis where only minimal data are available. Each has different data requirements and limitations.
| Method | What you can report | Important limitation |
|---|---|---|
| Structured tabulation and text | Study results grouped by meaningful comparisons and outcomes | A descriptive pattern is not itself a pooled treatment effect |
| Range or distribution of effects | Variation across estimates on a comparable scale | Does not establish a weighted average; selective availability may distort the range |
| Forest plot without pooled estimate | Individual estimates and uncertainty visually | The plot alone does not settle heterogeneity or certainty |
| Direction-of-effect vote counting | Pattern of results favoring one direction or another | Ignores magnitude and usually precision; should not be based on significance |
| Explicitly structured narrative comparison | How findings differ by method, context and risk of bias | Requires transparent grouping to avoid cherry-picking |
Do not use vote counting based on whether each paper reports a statistically significant p-value. The Cochrane Handbook identifies significance-based vote counting as unacceptable because it confounds sample size and precision with effect direction. If direction-of-effect vote counting is used, define how positive, negative, null or indeterminate directions are handled; state the synthesis unit; and report its limited inferential value rather than treating the count as an effect-size estimate.
Step 4: Appraise limitations before drawing a pattern
A table of apparently consistent results is not enough. Examine whether a study's findings may be affected by allocation problems, missing outcomes, selective reporting, measurement limitations or confounding, as appropriate for its design. Weighting the interpretation toward more credible studies is reasonable only if the approach and reasons are explained, rather than quietly ignoring inconvenient findings.
Our risk-of-bias assessment guide explains the difference between structured appraisal and unsupported judgments. Risk of bias informs the strength of a synthesis; it does not produce the synthesis method by itself.
Step 5: Present the evidence before interpreting it
Your Results section should first allow readers to understand which studies contributed to each group and what each reported. Then describe patterns and exceptions. Reserve broader interpretation, applicability and recommendations mainly for the Discussion chapter. A clear visual display or matrix is often more informative than long prose that repeatedly names studies.
Worked example: feedback interventions in undergraduate education
Imagine a dissertation asks: How do structured weekly instructor-feedback approaches, compared with usual teaching, affect undergraduate assessment performance? The reviewer has included several hypothetical intervention studies. They evaluate broadly related assessments but use different scoring conventions and report inconsistent precision information. After consulting the protocol and an appropriate methods adviser, the reviewer decides that a pooled estimate would be insufficiently interpretable.
The example is illustrative, not evidence. It uses qualitative labels rather than invented effect sizes so it demonstrates classification and reporting without suggesting that a statistical analysis was actually carried out.
| Study label | Outcome and follow-up | Available result | Main comparability issue | Appraisal note |
|---|---|---|---|---|
| Cedar | End-of-course standardized assessment | Reported estimate favors feedback; precision available | Comparable construct, but unusual assessment format | Some concerns about missing outcome data |
| Brook | Mid-term instructor-designed quiz | Descriptive result favors feedback; precision not reported | Shorter follow-up and local measure | Selective reporting cannot be excluded |
| Willow | Final-course grade | Reported result is near no difference; uncertainty broad | Composite grade combines several performance components | Allocation process insufficiently described |
| Vale | Subject-specific skills assessment | Reported result favors usual teaching | Different subject and skills measurement | Outcome assessor aware of study group |
| Rowan | End-of-course standardized assessment | Reported estimate favors feedback; precision available | Closest match to Cedar's outcome | Lower concern in the assessed domains |
How to use this table: Each label represents a fictional distinct study. The table deliberately avoids numeric estimates because none were calculated from underlying data. A real dissertation should include effect sizes, denominators and uncertainty intervals where available, and never replace numerical evidence with labels simply to make a table easier to write.
First synthesis: comparable end-of-course assessments
Cedar and Rowan use approximately aligned assessment constructs and time points. The reviewer presents these studies together, including their actual effect estimates and confidence intervals if they exist in the underlying reports. Both fictional studies favor feedback, but Cedar has concerns about missing outcomes. The Results paragraph should therefore state that the two observations point in the same direction while emphasizing that the magnitude and reliability must be assessed from the real underlying data.
Second synthesis: other assessment measures
Brook, Willow and Vale are not simply added to the first group. They measure different constructs or time points and have additional reporting limitations. The reviewer reports these as a separate, heterogenous group: one descriptive result favors feedback, one is near no difference, and one favors usual teaching. That pattern does not support a claim of a uniform benefit.
A crucial insight follows: the apparent overall story changes depending on which outcomes are treated as comparable, not merely on how many individual studies appear positive. This is why grouping must be planned and transparent.
Model Results paragraph: defensible language
The synthesis was organized by outcome comparability and follow-up. Studies using broadly comparable end-of-course assessments reported results in the direction of benefit from structured feedback, although one study raised concerns about missing outcomes. Studies employing mid-term quizzes, composite grades or subject-specific skills measures showed a less consistent pattern, and their outcome differences limited direct comparison. Because measures and reporting were insufficiently compatible for a meaningful pooled estimate, no overall effect size was calculated. These findings describe the available study results rather than establishing a common intervention effect.
This text is a template illustrating appropriate caution. In a real dissertation, name the contributing studies, give the actual estimates and explain how the prespecified synthesis approach leads to the stated interpretation. Avoid claiming that the fictional pattern is generalizable research evidence.
Write a Methods section readers can reproduce
A dissertation Methods chapter needs more than “a narrative synthesis was conducted.” State the circumstances that prevented pooling, how studies were grouped, the synthesis method for each outcome, how effect directions or estimates were handled, and how uncertainty and bias were evaluated.
Model Methods paragraph
Eligible intervention studies were grouped by prespecified outcome construct, comparator and follow-up period. We assessed whether effect estimates could be meaningfully combined by examining the definitions, measurement scales, populations, time points and availability of precision estimates. For the assessment-performance outcome, these conditions did not support an interpretable pooled estimate. We therefore tabulated individual results by outcome group, reported available effect estimates and uncertainty measures, and compared the direction and consistency of findings in structured prose. We did not infer effectiveness from the number of statistically significant findings. Risk-of-bias judgments and missing outcome information were considered when interpreting apparent patterns. The synthesis and its limitations were reported with reference to Cochrane methodological guidance and the SWiM reporting checklist.
Modify that paragraph to match what was actually done. For example, do not say you reported uncertainty measures if the included papers supplied none, and do not imply that SWiM is a statistical algorithm. It is a reporting guideline, not an analysis method.
Use SWiM to make reporting complete
The SWiM guideline was developed to improve reporting of quantitative intervention-effect syntheses conducted without meta-analysis. It contains nine reporting items and complements PRISMA rather than replacing it. Its scope matters: a qualitative meta-synthesis or a broad non-systematic literature review should not automatically claim SWiM compliance.
| SWiM reporting item | What the dissertation should make explicit |
|---|---|
| 1. Grouping studies for synthesis | Criteria for grouping studies and why the groupings answer the question |
| 2. Standardized metric and transformations | Common effect metric if used, and how any conversions were performed |
| 3. Synthesis methods | Specific non-meta-analytic method, justification and handling of missing data |
| 4. Criteria for prioritizing results | Why certain outcomes, study results or analyses were emphasized |
| 5. Investigation of heterogeneity | How differences between study findings were examined or why they were not |
| 6. Certainty of the evidence | How confidence or certainty was assessed, with limits |
| 7. Data presentation methods | Which tables, plots and displays show the findings |
| 8. Reporting results | Which studies contributed and what the synthesis actually found |
| 9. Limitations of the synthesis | What the method cannot resolve and how this affects conclusions |
The original SWiM guideline article gives detailed explanations and examples. Work through its full checklist while writing, rather than treating the summary table above as a formal substitute. Also follow the PRISMA 2020 checklist for review-wide reporting, including study selection, bias assessment, synthesis methods and findings.
Explain heterogeneity without exaggerating what it shows
When study estimates differ, explore whether population characteristics, educational setting, intervention delivery, outcome definition, follow-up or study quality offer plausible explanations. Describe those differences as possibilities unless supported by a valid comparison. Finding that one type of study reports a larger benefit does not demonstrate that the study type caused the difference.
In a small synthesis without comparable effect estimates, a numeric I² statistic may be unavailable or inappropriate. Do not invent one. If a meta-analysis is conducted for a legitimate subset of studies, report the model and heterogeneity methods for that subset; do not imply the subset's pooled result applies to studies omitted for outcome incompatibility.
For reviews that have enough information to make an appropriate certainty judgment, our GRADE and Summary of Findings guide explains how to distinguish what the evidence appears to show from how confidently it can be believed. A narrative presentation does not make certainty assessment optional where the review protocol and applicable guidance require it.
Report limitations honestly
A careful synthesis discusses limitations of both the evidence and the synthesis method. These are related but not identical.
Evidence limitations include small or selective samples, incomplete data, imprecise estimates, confounding and biased outcome measurement. Synthesis limitations include inability to compare effect magnitude across incompatible scales, lack of a weighted summary estimate, incomplete coverage of results, possible selective outcome grouping and weak ability to explain observed heterogeneity.
A useful limitations paragraph might read:
Interpretation is limited by differences in outcome construction and follow-up, incomplete reporting of effect estimates in some studies and variation in risk-of-bias concerns. The structured synthesis preserves these distinctions rather than producing an overall pooled estimate. As a result, its conclusions concern the direction and distribution of reported findings within the specified groups; they do not quantify a common intervention effect or establish causal explanations for between-study differences.
If the included evidence is too weak or inconsistent to support a directional conclusion, say so. A dissertation is stronger when it clearly identifies uncertainty than when it manufactures agreement across studies.
Common mistakes—and how to correct them
Mistake: writing one summary paragraph per study
A sequence of individual summaries can be useful for documenting studies but does not explain how they fit together. Group studies by the question they answer and compare differences and similarities within that group. For practical writing techniques, see our guide to synthesizing academic sources.
Mistake: treating significance as a vote
Saying “most studies were statistically significant” privileges studies with greater precision or larger samples and can obscure an important effect in an underpowered study. If direction-of-effect synthesis is appropriate, report the method explicitly and distinguish direction from magnitude, uncertainty and certainty.
Mistake: assuming no meta-analysis means no systematic method
The search, screening, extraction, appraisal and synthesis still require transparent rules. “Narrative” does not mean informal or unverifiable.
Mistake: collapsing distinct outcomes into one favorable conclusion
Test performance, course completion, attendance and satisfaction are not interchangeable. If you combine distinct constructs conceptually, justify the interpretation and avoid implying numerical equivalence.
Mistake: omitting non-significant or unfavorable findings
Readers need the whole eligible evidence picture. Record why any study result was excluded from a particular synthesis group and report meaningful contradictions rather than relegating them to silence.
Mistake: copying a model paragraph without adapting the method
The worked example illustrates transparent reporting, not a recipe that validates every dissertation's design. Your Methods and Results chapters must match your actual protocol, included data, study design and institution's reporting requirements.
Frequently asked questions
Is narrative synthesis the same as a systematic literature review?
No. A systematic review is a research design involving structured identification, selection, appraisal and synthesis of evidence. Narrative synthesis is a broad term for approaches used to present and interpret study results when a pooled statistical analysis is not performed or is not the whole story.
Can I write a narrative synthesis if I have only a few studies?
Possibly, provided the review question and study designs support a meaningful synthesis. A small number of studies does not automatically rule out meta-analysis, and it does not justify unsupported claims. Explain the actual rationale for your chosen method.
Does SWiM mean I have conducted a narrative synthesis correctly?
No. SWiM improves reporting transparency. It does not certify the quality of your search, grouping decisions, study appraisals or statistical method. Good reporting and appropriate methodological conduct are separate requirements.
Can I include a forest plot without a pooled effect?
Yes, when studies report compatible effect metrics and uncertainty information, an appropriately labeled forest plot may show individual study estimates without a pooled summary. Explain why no overall estimate is presented. If metrics differ so much that the display would mislead, use separate plots or a table instead.
Should I use GRADE in a review without meta-analysis?
It may be appropriate or required for intervention-effect evidence, depending on the review question and applicable methods. Certainty can be evaluated without a pooled meta-analysis, but the procedures and reporting should fit the underlying evidence. See our GRADE guide for outcome-level judgments.
Can I write that an intervention is effective if several studies favor it?
Not automatically. Direction alone tells you little about effect size, uncertainty, risk of bias or real-world importance. State exactly what the evidence supports, include opposing results, and explain limitations before drawing a conclusion.
Dissertation-ready checklist for a synthesis without meta-analysis
Before submitting, confirm that your chapter does the following:
Defines the review question, comparisons, outcomes and follow-up periods.
Documents why the relevant results were not pooled and whether alternative statistical approaches were considered.
States grouping rules established before interpreting findings, including how exceptions were handled.
Gives a traceable source-to-result matrix with real study identifiers and available effect estimates.
Names and justifies the specific synthesis method for each outcome group.
Distinguishes effect direction, magnitude, statistical uncertainty and certainty of evidence.
Explains how risk of bias and missing information affected interpretation.
Reports counterexamples and inconsistent findings without selective emphasis.
Includes appropriately captioned tables and visuals.
Addresses relevant SWiM items and the broader PRISMA reporting requirements.
Avoids claiming a pooled effect, universal benefit or causal subgroup explanation that was not estimated.
Presents transparent limitations and appropriately cautious conclusions.
Final takeaway
A strong narrative synthesis is not a weaker version of a meta-analysis or a series of disconnected article summaries. It is a planned, traceable and candid account of what the included studies collectively show—and where comparison is limited. Use explicit groupings, study-level displays, suitable synthesis methods and clear explanations of uncertainty. Your dissertation's credibility will depend less on whether it contains a single pooled number than on whether readers can follow and evaluate each interpretive step.
If you need careful, human academic help with a review's organization, methods wording, evidence tables or the clarity of its Results and Discussion sections, explore Acadelyra's Research Support and Dissertation & Thesis Support services. Support should strengthen your understanding, source handling and academic integrity; the research decisions, actual evidence and submitted work remain your responsibility.
Authoritative references and further reading
Cochrane. Cochrane Handbook for Systematic Reviews of Interventions, Chapter 9: Summarizing study characteristics and preparing for synthesis.
McKenzie, J. E., & Brennan, S. E. Cochrane Handbook, Chapter 12: Synthesizing and presenting findings using other methods.
Campbell, M., et al. (2020). Synthesis without meta-analysis (SWiM) in systematic reviews: Reporting guideline. BMJ, 368, l6890. Full text.
Page, M. J., et al. (2021). The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. PRISMA 2020 checklist.
Cochrane. Cochrane Handbook, Chapter III: Reporting the review.