To try to ensure your strategy is retrieving all the relevant studies you can create a test set of records. This is a group of records you would want (and expect) your search strategy to retrieve. This set is usually decided by the reviewers or from the studies included in an earlier review.
Before you start you MUST carefully consider what words you need to include in your search. If the searches aren’t good enough, the whole thing is flawed.
Think about...
Your search is likely to be complex and involve multiple steps to do with different subjects, what are often called "strands" or "strings" in the search. Look at the appendices of existing reviews for an idea of what's involved in creating a comprehensive search.
Most people should start by finding all the articles on Topic A, then moving on to Topic B, then Topic C (and so on), then combining those strands together using AND (see Combining your terms: search operators below). This will then give you results that mention all those topics.
You will then need to adapt (or "translate") your strategy for each database depending on the searching options available on each one. A core of terms is used across multiple databases - this is the "systematic" part - BUT with additions and subtractions as necessary. While the words in the title and abstract might remain the same, it's highly likely the thesaurus terms (if they exist) will be different across the databases. You may need to leave out some strings completely; for example, let's say you are doing a study that needs to find Randomised Controlled Trials (RCTs) on a particular disease and its treatment. You will be looking in multiple databases for words to do with the condition, and also words that are used for RCTs. But when you are looking in databases that are composed entirely of RCTs (trials registers), the part of a search looking for RCTs doesn’t need to be included as it's redundant.
The techniques described below will help ensure you cover everything. Contact your Subject Librarian if you would like guidance on constructing your search.
This video from the University of Reading gives a good overview of literature searching tips and tricks:
Jump to 01:45 for truncation and 05:46 for wildcards.
Most searches have two elements - the "keywords" part and the "subject headings" part - for each topic. When you are initially constructing your search and trialling it in a database, you are likely to just add your keywords, click Search, and see how many that retrieves. But after that, for any type of comprehensive search, you should look at limiting your keywords to looking in specific search fields.
A field in this context is where the database only looks at one aspect of the information about the article. Common examples are the Author, Title, Abstract, and Journal Name. More esoteric ones could be fields like the CAS Registry Entry or Corporate Author.
In complex reviews like systematic, scoping and rapid reviews, the accepted wisdom is to limit these "keyword" searches to the Title and Abstract fields, plus (if available, and the search is looking to be comprehensive) any available "Author Keyword" or "Contributed Indexing" fields. It is vital that the keywords you use in these fields are identical - you are using the same words in the Title, Abstract and any related fields - and that you combine them using OR (see Combining your terms: search operators below)
Using keyword searching limited to the Title/Abstract/Keywords fields should reduce the number of results which are retrieved in error or are only on the periphery of your subject. If you do this, please be aware that you will need to ensure that you have definitely also included all relevant subject headings in your search strategy (in databases that use controlled vocabulary) otherwise you risk missing out on useful results. It *is* quite possible that there will be no relevant subject headings in a particular search.
Although some databases will automatically search for variant spellings, mostly they will just search for the exact letters you type in. Use wildcard and truncation symbols to take control of your search and include variations to widen your search and ensure you don't miss something relevant.
A truncation symbol (*) retrieves any number of letters - useful to find different word endings based on the root of a word:
africa* will find africa, african, africans, africaans
agricultur* will find agriculture, agricultural, agriculturalist
A wildcard symbol (?) replaces a single letter. It's useful for retrieving alternate spelling spellings (i.e. British vs American English) and simple plurals:
wom?n will find woman or women
behavio?r will find behaviour or behavior
Hint: Not all databases use the * and ? symbols - some may use different ones (! instead of *, for example), or not have the feature at all, so check the online help section of the database before you start.
Search operators (also called Boolean operators) allow you to include multiple words and concepts in your searches. This means you can search for all of your terms at once rather than carrying out multiple searches for each concept.
There are three main operators:
Using OR will bring you back records containing either of your search terms. It will return items that include both terms, but will also return items that contain only one of the terms.
This will give you a broader range of results.
OR can be used to link together synonyms. These are then placed in brackets to show that they are all the same concept.
women AND AfricaUsing AND will find items that contain both of your search terms, giving you a more specific set of results.
If you're getting too many results, using AND can be a good way to narrow your search.
women NOT AfricaUsing NOT will find articles containing a particular term, but will exclude articles containing your second term.
Use this with caution - by excluding results you might miss out on key resources.
Sometimes your search may contain common words (i.e. development, communication) which will retrieve too many irrelevant records, even when using an AND search. On many databases, including Google, to look for a specific phrase, use inverted commas:
Your search will only bring back items containing these exact phrases.
Some databases automatically perform a phrase search if you do not use any search operators. For example, "agriculture africa" is not a phrase used in English so you may not find any items on the subject. Use AND in between your search words to avoid this.
On Scopus to search for an exact phrase use { } e.g. {agricultural development}. Using quotes on Scopus will find your words in the same field (e.g., title) but not necessarily next to one another. In this database, you need to be very careful with those brackets - {heart-attack} and {heart attack} will return different results because the dash is included. Wildcards are searched as actual characters, e.g. {health care?} returns results such as: Who pays for health care?
Some databases use proximity operators, which are a more advanced search function. You can use these to tell the database how close one word must be to another and, in some cases, in what order. This makes a search more specific and excludes irrelevant records.
For instance, if you were searching for references about women in Africa, you might retrieve irrelevant records for items about women published in Africa. Performing a proximity search will only retrieve the two words in the same sentence, making your search more accurate.
Each database has its own way of proximity searching, often with multiple ways of doing it, so it's important to check the online help before you start. Here are some examples of the variety of possible searches:
After completing your keywords search on a topic, you can move on to looking for appropriate subject headings.
Most databases have this controlled vocabulary feature (standardised subject headings or thesaurus terms - a bit like standard tags) which can help ensure you capture all the relevant studies; for example, MEDLINE, CENTRAL and PubMed use the exact same headings, which are called MeSH (Medical Subject Headings). Some of these headings will be the same in other related databases like CINAHL, but many of them will be slightly different, could be the same but have subtly different meanings, or not be there at all.
Not all databases have these types of subject headings - Web of Science and JSTOR don't allow you to search for subject headings like these, although you can of course search for subjects in them.
The easiest way to search for a subject heading is to go to the relevant area in the database that searches specifically for them; this might be called something like Thesaurus, Subject Headings, or similar. Then search for some of the words to do with your topic - not all of them at once, just a word on its own or a very simple phrase. Does this bring anything up? When you read the description, are you talking about the same thing?
You can then tell it to search for everything listed under that subject heading, then move on to looking for another subject heading. It's quite common for one topic to have several relevant headings.
Once you have found all the relevant headings, and made the database run searches for them, you will then combine them together using OR.
After you have found all the title/abstract/keywords for Topic A, and then all the relevant subject headings, you then combine those together using OR. You may need to go into the Search History section of the database to do this, and work out whether you can tick boxes next to your various searches to combine them, or have to type out something like "#1 OR #2".
This gives you a "super search" with everything in the database on that topic. It's likely to be a lot!
It might be that adding them together gives no extra results than the amount in either the keywords or the subject headings on their own. This is unusual, but not impossible:
You now go back to the start and for Topic B do the same title/abstract/keywords searches, then the relevant subject headings searches, then combine them as above. Then Topic C, and so on. Again, each of these super searches may have very high numbers - possibly millions.
Finally, you then combine all these super searches together, but this time using AND; they need to mention all the topics. It's possible that there are no articles in that database that mention all those things - the more subjects you AND together, the less results you are likely to find. However, it's also possible to end up with zero as there is a mistake in your search, and in most cases having zero results won't allow you to write your paper or thesis. So contact us if you think you have too few or two many results, and we can advise.
Methodological search filters are search terms or strategies that identify a topic or aspect. They are predefined, tried and tested filters which can be applied to a search.
Examples:
Study types: 'systematic reviews', 'Randomised Controlled Trials'
Age groups: 'children', 'elderly'
Language: 'English'
They are available to select via the results filters displayed alongside your results and are normally applied at the very end of your search. For instance, on PubMed after running your results it is possible to limit by 'Ages' which gives predefined groupings such as 'Infant: birth-23 months'. These limits and filters are not always the same across the databases, so do be careful.
