Жазу · Орташа
Write a narrative review that synthesises evidence
Create an argument across studies and describe the boundaries of a narrative synthesis.
14 мин оқу · 2026 ж. 26 қаз.
Бұл бет әлі аударылмаған, сондықтан ағылшын нұсқасын оқып отырсыз.
Негізгі тұжырымдар
- Organise around ideas.
- Compare evidence and methods.
- Make literature selection transparent.
Мазмұны
Create an argument across studies and describe the boundaries of a narrative synthesis.
A narrative review teaches readers how to understand a body of knowledge. Its contribution may be an explanation of conflicting findings, a comparison of theories, a clarification of terminology, or a conceptual model connecting previously separate ideas. It is a research argument supported by literature, not a catalogue of everything you read. A strong review lets a reader answer three questions: what is known, why the evidence deserves that interpretation, and what remains uncertain.
This tutorial takes you from a broad interest to a review manuscript. The extended examples are fictional teaching scenarios. Their study labels, findings and sample descriptions are invented to demonstrate reasoning and must never be cited as real research.
Decide what the review will help a reader understand
Start with an intellectual problem. “Sleep and learning” names a topic; “Why do studies of sleep and student learning reach different conclusions when they use different sleep measures?” names a problem that a review can investigate. The narrower question gives you a reason to compare methods, populations and interpretations. It also prevents a sequence of unrelated paragraphs about everything associated with sleep.
Write a one paragraph scope statement before collecting dozens of papers. Name the audience, central question, relevant systems or populations, main concepts and exclusions. A school student might compare how classroom studies measure attention. An undergraduate engineer might explain the tradeoff between sensor accuracy and energy use. A beginning doctoral researcher might reconcile competing explanations of a particular cellular response. Difficulty comes from the depth of the question, not the number of technical words.
A useful proposed scope is: “This review examines how laboratory task measures and classroom observations support different claims about sustained attention in adolescents. It focuses on the interpretation of measurement rather than estimating the effect of a particular treatment.” This tells you to exclude papers that merely mention attention without addressing its measurement. It also tells readers what the review will not resolve.
Pautasso's practical guidance emphasises choosing a manageable topic and audience, keeping organised notes, and evaluating rather than merely listing the literature. Treat that guidance as a starting framework, then design an argument suited to your question. There is no universal minimum number of references that turns an essay into a good review. S500
Distinguish a narrative review from other review types
A narrative review can select and integrate material to develop an explanation. Its selection still needs justification. A systematic review uses explicit eligibility, search, selection, appraisal and synthesis methods to answer a defined question. A scoping review systematically maps concepts, evidence types or the extent of a field. A meta analysis is a statistical synthesis and can be part of a systematic review; it is not a synonym for a review article.
Do not choose “narrative” as permission to hide inconvenient evidence. Conversely, do not call a review systematic simply because you searched two databases or drew a selection diagram. Choose a label that accurately describes the work. If your main aim is to estimate the overall effect of an intervention, read Article 27 and seek appropriate methods support before proceeding.
SANRA is a six item appraisal scale developed for narrative reviews in a medical editorial setting. It considers the importance and aim of the review, search description, referencing, reasoning about evidence, and relevant outcome information. It is useful for checking omissions, but its development setting and limited validation mean that it should not be presented as a universal scoring rule for every discipline. S24
Search deliberately and keep an honest record
Begin with recent reviews to learn terminology and identify foundational studies. Then retrieve the primary papers relevant to the claims you plan to make. A primary paper reports the original study; a review interprets a collection of studies. You may cite a review for its synthesis, but cite the primary paper when discussing that study's design or numerical finding after checking the original report.
Use several concept groups. In a hypothetical review of classroom attention, one group might include attention, vigilance and concentration; another might include adolescent, secondary school and classroom. Combine synonyms with OR and different concepts with AND. Databases differ in subject headings, phrase matching, truncation and field codes. A search expression copied from one interface may behave differently in another.
Save the database and platform, exact expression, search date, restrictions and useful records. Follow references backward to earlier work and citations forward to later responses. Add vocabulary discovered during reading. The search log helps you explain why a literature is represented as it is, even when your review does not claim exhaustive retrieval. PRISMA S offers detailed search reporting guidance for systematic reviews; its distinction between databases, platforms and full search strategies is also a useful discipline when recording a narrative search. Applying a few of its practices does not make your review systematic. S501
An honest narrative search description might read: “We searched two disciplinary databases using terms relating to classroom attention and measurement, supplemented by reference checking of relevant reviews. We prioritised studies that compared measurement approaches or explained their interpretation. The search was intended to support a conceptual comparison rather than identify every eligible study.” In an actual manuscript, replace these general descriptions with the real databases, dates and selection rationale. Never invent a search date or imply that a search was run when it was only proposed.
Build an evidence matrix that supports comparison
Your notes should preserve the information needed to evaluate an argument. For each source record its identifier, question, design, system or participants, measurement, analysis, main estimate or interpretation, uncertainty, limitations, and role in your review. Add a column for the exact page, figure or table supporting a claim. Keep your own interpretation separate from the authors' statements.
The following small matrix contains only fictional studies.
| Evidence label | Design and setting | What was measured | Main teaching finding | Interpretive limit |
|---|---|---|---|---|
| Study A | Laboratory task in one school sample | Errors during a brief sustained attention task | Performance differed between testing sessions | Session order and practice may matter |
| Study B | Classroom observations over several lessons | Teacher rated engagement | Ratings varied across subjects and teachers | Engagement is not identical to task accuracy |
| Study C | Interviews about attention in lessons | Students' accounts of distraction | Students described workload and belonging as relevant | Accounts explain meaning, not population prevalence |
| Study D | Instrument comparison | Agreement between two task measures | Measures ranked some students differently | Different ranking may reflect distinct constructs or measurement error |
Construct means the underlying concept you want to study, such as attention or trust. Operationalisation means the rule used to turn that concept into an observation. Measurement validity concerns whether evidence supports the intended interpretation of that observation. Reliability concerns consistency under specified conditions. A measure can be consistent while failing to represent the construct needed for your question.
This matrix reveals that “attention” has several operational meanings. That observation is a possible contribution. It prevents you from describing an interview account, a rating and a laboratory error score as though they were interchangeable measures of the same quantity.
Turn the matrix into an argument map
Group sources by the problem each section will resolve. Suitable organising principles include mechanisms, measurement approaches, competing theories, historical changes in explanation, stages of a process, and differences in context. Chronological order is useful when development over time is the argument; otherwise it can hide the comparisons readers need.
For the fictional attention review, an argument map could contain four sections: what each measure claims to represent; how context changes the observed result; what apparent disagreements can and cannot establish; and which comparisons would resolve the uncertainty. Each section has a job. “Study A”, “Study B” and “Study C” would be weak headings because they tell readers who published rather than what the review explains.
Write a provisional claim under each heading. Then attach supporting, qualifying and conflicting evidence. If a claim has only supportive sources, deliberately search for a credible alternative. If the evidence does not justify the claim, narrow or abandon it. A review's central argument should emerge from the evaluated material rather than determine which findings are allowed into the manuscript.
Write synthesis paragraphs rather than serial summaries
A useful synthesis paragraph often moves through a claim, a comparison, an explanation and a boundary. These are functions rather than a compulsory sentence formula. Some paragraphs need more description; others need more methodological analysis.
Weak fictional draft: “Study A measured attention with a laboratory task. Study B used teacher ratings. Study C interviewed students. These papers show that attention is important and more research is needed.”
Developed fictional synthesis: “The apparent disagreement between these studies partly reflects the different meanings assigned to attention. Study A evaluated performance during a tightly controlled task, whereas Study B examined visible engagement during ordinary lessons. A pupil could perform accurately on a short task while appearing disengaged in a classroom, so differences between the findings do not by themselves demonstrate that either measure is invalid. Study C adds a different kind of evidence: students described distraction in relation to workload and their willingness to ask for help. These accounts suggest contextual explanations that neither numerical measure directly tests. Taken together, the studies support treating task performance, observed engagement and reported experience as related but distinct outcomes. A study measuring all three in the same setting would help determine whether the disagreement arises from measurement, context or both.”
The opening sentence identifies the comparison. The middle explains why the methods matter. The final sentences state what the combination supports and what would test the proposed explanation. In a real review, citations would accompany the description of each study and the comparison would be checked against the original methods and results. The phrase “suggest contextual explanations” avoids claiming that an interview established a causal effect.
Evaluate strength without turning papers into votes
A review should not conclude that an intervention works because six papers report p < 0.05 and four do not. The studies may differ in sample size, design, precision, outcome or risk of bias. A large p value is not proof of no effect, and a small p value is not a measure of practical importance. Extract effect estimates and intervals where meaningful, then consider whether they address the same question.
For quantitative evidence, distinguish a mean difference in original units from a standardised mean difference, which expresses a difference relative to a chosen standard deviation. Standardisation can help when compatible constructs use different scales; it cannot make different constructs identical. Relative effects also need context: a large relative change may correspond to a small absolute change when the baseline rate is low. Appropriate numerical synthesis requires explicit assumptions about comparability and dependence. S21
For qualitative evidence, examine the coherence of the question, sampling, context and interpretation. Do not grade an interview study poorly because it lacks a p value. For engineering evidence, compare benchmark conditions, measurement calibration, baseline strength and external validation. A simulated improvement under one noise model does not directly establish performance in physical deployment.
Use specific criticism. “The study is small” does not explain the consequence. “The interval includes both negligible and substantial changes, so the study does not estimate the magnitude precisely” does. “The work is biased” is less useful than explaining how selective recruitment could change the observed relationship. State what a limitation prevents the reader from concluding.
Work through a second example in engineering
Suppose three fictional sensor studies all report that a new design is “more accurate”. Study E evaluates steady temperatures indoors, Study F evaluates rapid changes outdoors, and Study G evaluates the same sensor after a calibration procedure that the other studies do not use. A superficial review might average the reported percentages or rank the devices. A better review first asks what error each paper calculated and under what reference measurement.
A suitable synthesis could read: “The reported accuracy improvements cannot yet be interpreted as a common performance ranking. Study E evaluates error after conditions stabilise, while Study F emphasises response during rapid temperature changes. Study G additionally changes the calibration procedure, making it difficult to separate sensor design from calibration. The evidence therefore supports three narrower questions: performance at equilibrium, response to changing conditions, and sensitivity to calibration. Future comparisons should report these outcomes separately under shared reference conditions.”
This paragraph teaches a transferable habit: explain disagreement using the evidence before declaring one paper correct. It also identifies a specific next experiment rather than ending with an empty request for more research. No effect sizes have been pooled because the fictional comparisons do not define a common quantity.
Draft each section around its purpose
The title should identify the subject and review type without promising a definitive solution. The abstract states the problem, scope, approach to selecting and organising literature, main synthesis and its boundary. Write the abstract after the argument is stable, and include search details proportionate to the review type and journal requirements.
The introduction explains why the conceptual problem matters, what previous reviews have covered, and what this review adds. Avoid a long history that never reaches the question. A distinct approach or methods section explains search and selection choices. The main body develops comparisons under informative headings. The discussion integrates the argument across sections, considers rival explanations and identifies implications. The conclusion answers the review question within the limits of the evidence.
A review may need a structured synthesis without numerical pooling. Explain how you formed groups, which evidence informed each conclusion and how limitations changed your interpretation. SWiM was developed to improve reporting of intervention reviews using synthesis methods other than meta analysis. Its scope is narrower than all narrative reviews, but it demonstrates why saying “we used narrative synthesis” without explaining the actual method is insufficient. S502
Use tables, figures and references as parts of the argument
A table should allow a comparison that would be difficult in prose. Put the outcome definition and main interpretive limitation beside the study description. A table with only author, year and conclusion reproduces abstracts without helping readers judge the evidence. Label unavailable details as not reported rather than guessing them.
A conceptual figure can distinguish established relationships, proposed mechanisms and unresolved links. Define those categories in the legend and cite the evidence supporting the diagram. An original arrangement of ideas still requires attribution for the underlying claims. Check permission or licence conditions before reproducing another publication's image. A redraw that closely copies a published figure may still require attention to rights and attribution.
Audit each citation at sentence level. Does the source support the exact claim, population and direction stated? Did you read the cited version? Are corrections or later updates relevant? Avoid attaching a long reference list to a sentence when only one source supports it. Do not cite a famous review as though it performed the experiment you describe.
Practice and model answer
Task: Your draft says, “Most studies show that peer feedback improves writing, so schools should use it.” You have a mixture of short laboratory exercises, classroom interventions and interviews. Write a more defensible argument and identify the next information needed.
Model answer: “The included studies address different aspects of peer feedback. Classroom interventions can inform changes in assessed writing under the conditions studied, while interviews can explain how students interpret and use feedback. Before drawing a recommendation, the review should compare outcome definitions, study designs, effect estimates and classroom contexts, and examine whether benefits depend on feedback training or implementation. The present collection supports a question about the conditions under which feedback is useful more clearly than a universal claim about effectiveness.”
The answer does not assume that the fictional evidence establishes improvement. It separates evidence functions and names the missing comparison. Now write your own review outline without using author names as headings. Under each heading, state the comparison readers will learn to make. Use Template 8 for the manuscript and Template 3 for the evidence matrix.