Перейти к содержанию

С чего начать · Начальный

Search the literature and read papers critically

Build a search, read methods critically and record evidence without losing its limitations.

11 мин чтения · 26 окт. 2026 г.

Эта страница ещё не переведена, поэтому вы читаете английскую версию.

Главное

  • Record searches and source versions.
  • Compare methods as well as findings.
  • Build an evidence matrix.
Содержание

Build a search, read methods critically and record evidence without losing its limitations.

A literature search is a method for finding evidence, not a hunt for sentences that support your preferred conclusion. Your aim is to understand what is known, how it was established, where findings conflict and which uncertainty your paper can address. Reading critically means examining the connection between a claim and its evidence rather than deciding whether you like the conclusion.

This tutorial takes you from a research question to a search log, an evidence matrix and a synthesis paragraph. The searches and study comparisons below are teaching examples. They have not been run as a complete review and do not represent actual findings about mentoring, plants or engineering systems.

Decide what the search must accomplish

An introduction needs a balanced, relevant account of the work that motivates the question. A narrative review needs a transparent and defensible way of selecting and organising literature. A systematic review needs a planned, reproducible process intended to identify eligible evidence comprehensively within a stated scope. Calling a search “systematic” because you used several keywords does not establish that standard.

A record is a database entry. A report is a document describing research, such as an article or conference paper. A study is the underlying investigation. One study can produce several reports, and one report can describe several studies. This distinction matters because counting every document as independent evidence can count the same participants or experiment twice. Cochrane emphasises identifying studies through their reports. S20

Write a short search purpose before opening a database. For example: “Identify research comparing structured undergraduate mentoring with usual support, and understand how research skill was measured.” This sentence helps you notice when a search drifts toward mentoring satisfaction or employment outcomes instead.

Translate the question into concepts

Break the question into two or three core concepts. List alternative terms for each. Include spelling variants, older terminology and relevant controlled vocabulary, meaning indexing terms assigned within a database.

Concept Illustrative terms Decision to record
Population undergraduate, university student, college student Whether postgraduate students are eligible
Activity research mentoring, research supervision, research apprenticeship Whether informal mentoring belongs in scope
Outcome research skill, proposal quality, research competence Whether requiring outcome terms would miss relevant papers

Combine alternatives with OR, which broadens a concept, and combine different concepts with AND, which requires both. Parentheses make the intended grouping explicit. An initial conceptual search might be (undergraduate OR "university student" OR "college student") AND ("research mentoring" OR "research supervision").

This is a starting structure, not a validated database query. Phrase searching, truncation, proximity operators and field tags vary by platform. In PubMed, for example, quoting terms or using field tags can change automatic term mapping. Check the search details and the current official instructions rather than assuming that the database interpreted your words exactly as intended. S105

Do not add every detail of the question immediately. A study may measure the desired outcome without naming it in its title or abstract. Overly narrow searches can therefore omit relevant evidence. Conversely, an unrestricted term such as “stress” can refer to psychological experience, mechanical loading or cellular responses. Examine the first results to identify which ambiguity needs correction.

Choose sources for the discipline

The source set should reflect where relevant work is published. Biomedical research often requires biomedical databases; education and psychology require attention to their specialist literature; engineering may rely heavily on conference proceedings and technical reports. A general scholarly search engine can help discovery, but its coverage, ranking and export limitations may make it insufficient as the sole source for a comprehensive review.

Ask a librarian or information specialist which resources cover your question. Access varies by institution, so record the actual database and platform used rather than listing resources you could not search. Explain justified language, date or document restrictions. “Only freely available papers” describes an access limitation, not an evidence quality criterion.

Use backward citation searching, examining the references of a relevant paper, and forward citation searching, locating later papers that cite it. These routes can reveal terminology missed by the initial query. They can also keep you within one research community, so use them as complements to broader searching rather than proof of completeness.

For systematic intervention reviews, Cochrane recommends combining appropriate databases and other sources, documenting search decisions, and obtaining specialist search expertise. Its detailed resource choices are designed for intervention evidence and should be adapted to the actual field rather than copied automatically into an engineering review. S20

Find a small set of clearly relevant papers through initial reading or supervisor advice. Check whether the candidate search retrieves them. If it misses one, investigate why: terminology, indexing, date restrictions or an omitted database may be responsible. Retrieving these known papers is a useful check, but it does not prove that the search is comprehensive.

A search log should preserve the database, platform, exact query, date, filters, number of records and export filename. Record revisions as new entries rather than overwriting the earlier query. For citation searching, record the seed papers, direction, tool and date. For a review, preserve the exported records and the duplicate removal decisions.

Hypothetical log entry. “Database X, institutional platform Y, searched on the recorded date; exact query saved in search file 03; publication dates unrestricted; 184 records exported.” The numbers and platform labels here are placeholders for a teaching example. In a real paper, supply actual names and counts. A statement such as “we searched several online databases” gives readers almost nothing they can reproduce.

Read each paper in three passes

During the first pass, identify the question, design, main figures and claimed answer. This helps determine relevance. Do not treat the abstract as sufficient evidence for a detailed claim.

During the second pass, examine the Methods, figure legends and supplementary material. Identify who or what was studied, how units were selected, what was independently assigned or sampled, what was measured, and when. Look for planned outcomes, exclusions and handling of missing data.

During the third pass, compare the Results with the conclusions. Record the effect estimate, its units and uncertainty. Ask whether the paper answers the question it introduced, whether the analysis reflects the design, and whether alternative explanations remain credible.

For a qualitative paper, adapt these questions: inspect the sampling logic, context, researcher role, data collection, analytic approach, supporting excerpts and treatment of contradictory material. A confidence interval is not the appropriate quality criterion for every form of evidence.

Build an evidence matrix that supports writing

An evidence matrix is a structured set of notes that makes papers comparable. Use one row per underlying study, with links to all associated reports when necessary.

Field Record Why the information is useful
Research aim The actual question and proposed explanation Distinguishes similar labels from similar questions
Design and setting Allocation, recruitment, location and timing Identifies the available claim and its context
Sample structure Independent units, clusters and repeated observations Prevents inflated sample size comparisons
Measurement Outcome definition, instrument and scoring Reveals whether results can be compared meaningfully
Findings Estimate, uncertainty and relevant qualitative interpretation Preserves more information than “significant”
Appraisal Specific threats to inference and strengths Explains how much weight the study deserves
Source location Page, table, figure or supplement Allows every note to be checked

Keep exact quotations visibly separate from your own notes. Copying a sentence into a notes file without quotation marks makes later accidental plagiarism easier. Record the page or section when you save it. Reference managers can organise records and produce bibliographies, but imported author names, titles, dates and identifiers still require checking. Zotero’s official documentation describes bibliography creation and supported export workflows. S19

Work through an extended critical comparison

Imagine three fictional mentoring studies. Study A surveys volunteers who chose mentoring and classmates who did not. Study B allocates eligible students randomly to mentoring or usual support. Study C interviews students who completed mentoring about their experiences. All three discuss “benefit”, but they contribute different evidence.

Study A reports a difference in confidence scores. Check whether mentoring participants already differed in prior experience or motivation, whether the scale was appropriate, and how many invited students responded. An adjusted association may be informative, but adjustment does not automatically remove unmeasured differences.

Study B assesses scored research proposals. Check the allocation process, whether assessors knew group membership, whether students were analysed consistently with the planned question, and whether missing assessments differed by group. Its design may better support a causal comparison, but the result concerns proposal quality in the studied setting, not every possible benefit of mentoring.

Study C describes how participants gained access to informal advice while also feeling pressure to appear competent. Check whether the sampling captured varied experiences and whether the interpretation is supported by the material. It can explain a process that numerical scores overlook. It should not be converted into a percentage of all students unless the design supports that estimate.

A weak review sentence would be: “Three studies prove mentoring works.” A stronger synthesis would be: “The fictional evidence addresses different aspects of mentoring. The survey links participation with confidence but is vulnerable to selection; the allocated comparison addresses proposal performance; and the interviews suggest a possible mechanism involving access to advice. Their findings cannot be combined as three measurements of a single outcome.”

The stronger version explains why the studies belong together and why their agreement has limits. In a real review, attach the relevant study citations to each clause and replace this invented evidence with verified findings.

Evaluate credibility without shortcuts

Check the original report, supplementary methods, corrections and retraction status. A DOI identifies an object; it does not certify the object’s conclusions. Journal reputation and citation count can guide discovery, but neither replaces appraisal of the study.

A preprint is a publicly shared manuscript whose review status must be checked. Record its version and date. If a later journal article exists, determine whether the data or conclusions changed and cite the version you actually use. A conference abstract may contain too little detail to support a strong methodological judgment, even when it identifies relevant ongoing work.

Separate reporting limitations, information missing from a paper, from design limitations, weaknesses in how the study was conducted. If allocation is not described, write “the allocation procedure was not reported”, rather than asserting that allocation was biased. Similarly, an unexplained discrepancy deserves investigation, not an immediate accusation of misconduct.

Measurement deserves particular attention in social science. Two studies may both use the word “engagement” while measuring attendance in one and a self report scale in the other. Flake and Fried argue for transparent measurement choices; the practical lesson is to compare what was measured rather than assuming that shared terminology guarantees a shared construct. S104

Turn notes into a synthesis plan

Group the evidence by a question, mechanism, method or source of disagreement. For a plant study, groups might concern species, growth conditions and outcome timing. For an engineering review, groups might concern benchmark difficulty, computational budget and evaluation leakage. For education research, groups might concern outcome measurement, selection and duration of follow up.

Write a topic sentence that states the comparison your paragraph will make. Follow it with the most relevant evidence, explain the difference between studies, and end with the uncertainty that remains. Avoid one isolated paragraph per author unless the history of a small set of contributions is itself the subject.

Practice. Two fictional studies report p values of 0.04 and 0.08 for similar estimated effects. Is this sufficient evidence that their findings conflict?

Model answer. No. Obtain the estimates, uncertainty intervals, designs and sample sizes. Crossing different sides of a threshold does not establish that the effects differ. A useful comparison examines the estimated difference between results and the reasons their precision or methods differ. Do not replace that assessment with counting “significant” studies.

Before drafting, identify credible evidence that challenges your expected conclusion. If every paper in your matrix supports your initial belief, revisit the search terms and selection decisions. Use Template 3 to preserve the search and evidence record, and Article 26 or Article 27 for the appropriate review workflow.

Sources

S20: Lefebvre, C., Glanville, J., Briscoe, S., and colleagues. (2025). Chapter 4: Searching for and selecting studies. In Cochrane Handbook for Systematic Reviews of Interventions, version 6.5.1; chapter updated March 2025.

S105: US National Library of Medicine. PubMed User Guide. Official search documentation, accessed 11 November 2026.

S19: Zotero. Creating Bibliographies. Official documentation, accessed 11 November 2026.

S104: Flake, J. K., and Fried, E. I. (2020). Measurement Schmeasurement: Questionable Measurement Practices and How to Avoid Them. Advances in Methods and Practices in Psychological Science, 3(4), 456 to 465. DOI: 10.1177/2515245920952393.