Skip to content

>>> Blog

Search Strategy for a Literature Review: Terms, Databases and Grey Literature

12 min readLittview

Your protocol says you will search "relevant databases using a comprehensive strategy". It was an easy sentence to write. Now you have a page of keywords in a notebook, a Google Scholar tab open, and three questions the sentence does not answer: which databases, which words, and how would anyone know you found everything?

Here is the short answer. A search strategy is the written plan for finding every study that could answer your question: the concepts you search, the words and subject headings for each, how they are combined, where you run them, and what you record so someone else could run them again. Build it from the main concepts of the question, test it on papers you already know belong in the review, adapt it to each database, add grey literature and citation searching, and record each search as you run it.

That is the full version a systematic or scoping review needs. The search strategy in a narrative literature review can be shorter, so take as much of this as your time allows. The map below follows one invented review through all eight steps: how do warehouse workers experience shift schedules set by an algorithm?

Systematic review search strategy example: one question, eight steps

StepWhat you doIn the warehouse review
1. ConceptsPick the ideas every relevant paper must mentionWarehouse work; algorithmic scheduling
2. TermsGather keywords and subject headings for eachwarehous*, "distribution centre", "algorithmic management"
3. CombineOR within a concept, AND between conceptsWarehouse terms AND scheduling terms
4. TestCheck it finds papers you know belongSix known studies; five found at first
5. DatabasesRun it, adapted, in several databasesScopus, Business Source Complete, Sociological Abstracts
6. Grey literatureSearch where reports and theses liveUnion and government reports, theses, working papers
7. Citations and peopleFollow references both ways, ask expertsReference lists and citing papers of included studies
8. RecordKeep every search as run, with dates and countsOne row per database, pasted from the search history

Each section below opens one row of the map.

1.Break the question into concepts

The question comes first. The Campbell guide to searching, written for the social, behavioural and educational sciences, bases the structure of a search on the main concepts of the review, and calls it usually unnecessary, even undesirable, to search on every aspect of the question.

The reason is that every concept you add is one more condition a paper must meet to be found. Some parts of a question, a particular setting or outcome for instance, may not be well described in titles and abstracts, and are often poorly indexed by the databases.

ConceptSearch it?Why
Warehouse workYesIt is the review's subject, named in titles and abstracts
Algorithmic schedulingYesIt is what the review examines
Workers' experienceNoMay not appear in titles and abstracts; screening checks it

Your inclusion and exclusion criteria tell you which concepts matter. The search finds candidates; screening decides.

2.Systematic review search terms: keywords and subject headings

Each concept gets two kinds of terms, because many databases can be searched two ways. Keywords are words in the title or abstract. Subject headings are the terms a database assigns to each record from its own thesaurus. Use both together.

Many databases have their own thesaurus, such as the ERIC Thesaurus in ERIC, the APA Thesaurus of Psychological Index Terms in APA PsycInfo, and MeSH in MEDLINE. A heading in one may have no equivalent in another.

Where terms come fromIn the warehouse review
Titles, abstracts and keywords of papers you already know"algorithmic management", "shift scheduling"
The database's thesaurusIts headings for warehouses and for scheduling
Search strategies in related reviewsTerms another review used for platform work
Spelling variantscentre and center, labour and labor
Someone who knows the field"workforce management software"

Truncation finds word endings, so warehous* finds warehouse, warehouses and warehousing. Cut too short and it backfires, since cent* finds century. Quotation marks keep a phrase together. Check each database's help page for its symbols.

Leave out words the database already assumes. Searching "education" in ERIC, an education database, adds nothing and can narrow the results.

3.Combine the terms with OR and AND

Boolean logic sounds technical, and it comes down to two rules. OR joins the terms within a concept, so a paper needs only one of them. AND joins the concepts, so a paper must match each one. Parentheses keep the groups apart.

Search lineWhat it does
1. warehous* OR "distribution centre*" OR "distribution center*" OR "fulfilment centre*" OR "fulfillment center*" OR "order picker*"Warehouse keywords
2. The database's headings for warehousesWarehouse headings
3. 1 OR 2Concept 1
4. "algorithmic management" OR "algorithmic scheduling" OR "automated scheduling" OR "shift scheduling"Scheduling keywords
5. The database's headings for schedulingScheduling headings
6. 4 OR 5Concept 2
7. 3 AND 6Both concepts

Run the keyword lines in the title, abstract and keyword fields. It is a more specific search than searching every field, and when the terms for each concept are complete, a paper that matches none of them there is unlikely to be relevant.

Avoid NOT where possible. A search for female NOT male throws away every record about both. Proximity operators (NEAR, N3, adj) sit between a phrase and AND, finding two words close to each other.

Every limit needs a reason. Campbell's advice is ideally no language limit; a date limit when the thing you study did not exist before a given year, with that reason written down; and the database's built-in limiting commands avoided, except for publication date, because it is not always clear how they work.

4.Test it on papers you already know

Before you trust a search, see whether it finds what it should. Seed articles are studies you already know meet your criteria, collected from early exploratory searches, earlier reviews on the topic, or experts. The set should cover the range of what you expect to include, in field, region, time and design.

Use the seeds twice. First, harvest their titles, abstracts, author keywords and subject headings for terms. Then run the finished search and check that it finds them. Ideally every seed turns up somewhere in the combined results. When one is missing, open its record and find the word it used that you did not. In the warehouse review, the sixth study never said warehouse. It said "fulfilment centre", and that phrase went into line 1. Report the test and the seed list in your methods.

A good search finds a lot that you will throw away. Systematic searches aim for sensitivity and accept low precision. At Campbell's estimated reading rate of two abstracts a minute, a thousand records is about a working day. If the count is more than you can screen, tighten with proximity operators before you add a third concept.

Then ask someone who searches for a living to read it. PRISMA-S, the reporting guideline for searches, says authors "should strongly consider having the search strategy peer reviewed by an experienced searcher, information specialist, or librarian", and the PRESS guideline sets out what such a review checks.

5.Choose the databases

One database is never the whole literature, and Campbell does not consider a search of one database alone adequate. The JBI Manual sets no number, notes that most reviews search at least three, and asks that the databases fit the question by topic and by region.

A useful pattern is the main subject database for your field, the neighbouring fields the topic borrows from, and one multidisciplinary database.

FieldDatabases to consider
Any fieldScopus, Web of Science, OpenAlex
EducationERIC
PsychologyAPA PsycInfo
SociologySociological Abstracts
Business and managementBusiness Source Complete, ABI/INFORM
EconomicsEconLit
Computing and softwareACM Digital Library, IEEE Xplore
Health and nursingMEDLINE, CINAHL

OpenAlex, in the first row, is an open index that is free to search, so a reader without a library subscription can rerun your search there. Exporting the results needs a free account.

Finish the strategy in the database you expect to hold most of the studies, then adapt it to each of the others: their subject headings, their field codes, their syntax. The same database can come through several platforms (ERIC through EBSCO, ProQuest or Ovid), each with its own commands, so note which one you used. For how an open index differs from a bibliographic database, and for the other kinds of search tools, see the tool map.

6.Grey literature search: where to look

Some of the studies you need never reached a journal. The JBI Manual defines grey literature as essentially anything not controlled by the traditional, usually peer-reviewed, academic publishing market. Campbell warns that failing to find what is reported in conference proceedings and other grey literature might bias a review's results.

The JBI Manual names BASE and CORE among grey literature databases. For the rest, you work through kinds of source.

KindWhere to look
Theses and dissertationsProQuest Dissertations & Theses Global, university repositories
Working papers and preprintsIDEAS/RePEc for economics, arXiv, OSF Preprints
ReportsWebsites of governments, charities, professional bodies and other organisations in your field
Conference papersWeb of Science's conference proceedings indexes, Scopus
Registered and ongoing studiesOSF Registries, the AEA RCT Registry for economics, ClinicalTrials.gov for health
Unpublished studiesAuthors of included studies, and experts in the field

Older lists still name OpenGrey and the DART-Europe theses portal; both have closed.

Many grey literature sites offer only a simple search, so run a short version of your terms, several times, in different combinations. Decide before you start which sites you will search and why, since there is no natural end point. Save a copy of every document you keep, because web pages move and disappear.

7.Citation searching and asking people

Two more routes find what keywords miss.

RouteWhat you do
BackwardRead the reference lists of included studies and earlier reviews
ForwardFind the papers that cite your included studies, in Web of Science, Scopus, OpenAlex or Google Scholar
PeopleSend authors and experts your list of included studies and criteria, and ask what is missing

For complex questions the JBI Manual considers forward and backward citation searching mandatory. Authors may cite selectively, though, so reference lists back up the database search and never replace it.

8.Record it: a systematic review search strategy template

Write the search down while you run it. The Cochrane Handbook notes that a search record can be nearly impossible to recreate after the fact. Copy each strategy exactly as run from the database's search history; retyping can introduce errors.

FieldWhat to writeIn the warehouse review
DatabaseIts nameSociological Abstracts
PlatformThe interface you usedProQuest
DateThe day the search ran14 March
Search as runEvery line, pasted from the search historyLines 1 to 7, in this database's syntax
LimitsEach limit and its reason, or noneNone
ResultsRecords found and records exported412 found, 412 exported
SearcherWho ran itThe searcher's name

For a website, also record the address, the date, the pages browsed or searched, the terms, and any decision about how many results to screen. PRISMA-S says the last search should ideally be less than 6 months before publication. If yours will be older than that when the review appears, run the searches again before you submit.

PRISMA 2020 asks for every source with the date it was last searched (item 6) and the full strategies for all databases, registers and websites, with any filters and limits (item 7); the checklist walk-through covers both. PRISMA-S breaks the search into 16 items and suggests putting the full documentation in a repository that issues a DOI, such as OSF or Zenodo.

In Littview, each database's export comes in as its own file (BibTeX, RIS, PubMed or EndNote XML), and duplicates across the files are found in one detection run before screening. The databases and registers you list, with the record count you enter for each, appear in the PRISMA flow diagram and the Methods draft.

FAQ

Two questions come up most.

Can I use Google Scholar for grey literature? Yes, as a supplementary source. The JBI Manual counts Google and Google Scholar as useful sources of grey literature, and a comparison of 28 academic search systems rated Google Scholar suitable as a supplementary source but inappropriate as the principal search system; one problem was that repeating the same query did not always return the same results. Its ranking can vary with your location and search history, so decide in advance how many results to screen (Campbell's example is the first 100) and record it. It finds citing papers, but not a paper's references.

Is the search strategy for a scoping review different? Not in how it is built. The JBI Manual says the conduct and reporting of a scoping review search align with those of a systematic review, and points to the same three-step search. The difference is one of emphasis, since it sees more opportunity for a scoping review to include grey literature. PRISMA-S was written for scoping reviews too.

For this week: write your question at the top of a page and circle the two or three concepts every relevant paper must mention. Under each, list every word for it you can find, starting from five papers you already know belong in the review. Run that in one database, check that the five come back, and book half an hour with your librarian before you run it anywhere else.