Systematic and structured literature search
Systematic and structured literature search
Although there are countless different ways to find academic literature, structured literature searches in bibliographic databases still constitute a central methodological element in most literature reviews.
On these pages, you will find a brief, pragmatic introduction that will enable you to conduct good, structured searches in such databases. On this main page, we briefly describe a 5-step working method that will take you from the research question to the completed search. Along the way, we explain some important concepts.
The 5-step method for developing good literature searches is essentially the same, whether you are conducting a limited literature study at the bachelor level or a full-fledged, publishable systematic literature review. The difference between these two primarily lies in the breadth of the research question, the number of databases searched, and the sensitivity of the searches constructed.
The 5-step method is also, in a sense, an abstraction, in that the steps in practice may overlap significantly, and one often needs to return to an earlier step as a consequence of what is learned
At the bottom of the page, you will find a collection of level-adapted examples that show how the 5-step process is applied.
The 5-step method
Here you will find the general description of the 5-step method for building good literature searches in reference databases. It may be wise to alternate a bit between the general description of the 5-step method on this page and the level-adapted examples on the tiles below. The general description can be too general and abstract, and will be easier to understand when you refer to a concrete example. The examples provide all the details about how things look and are done, but also build on the explanation of the 5-step method on this page.
In the type of bibliographic databases we primarily work with in thorough searches for literature reviews, it is not effective to search using long phrases in natural language that intuitively describe what we are looking for. To build good, structured database searches, we must first identify the main thematic elements of the research question for which we are seeking literature.
A structured database search consists of search keywords organized into groups or "boxes," where each box corresponds to a main thematic element or searchable concept. Within each box, we combine synonymous or nearly synonymous search keywords using the Boolean operator OR, while the groups or boxes are ultimately combined with each other using the operator AND. Abstractly, this looks like this:
A simple example for illustration: Let's imagine we are looking for research literature on measures or support schemes for gifted children with learning disabilities. In this research question, there are a number of words that carry meaning, but not all are equally suitable for attempting to build a literature search around. The best starting points are the concepts 'giftedness' and 'learning disabilities.' These become our main elements. In our box model, it might look something like this:
Don't worry about the details of the search keywords. They will be explained later and demonstrated in the examples. It's the structure we should note here, and perhaps especially the simplicity of the structure.
It is very important to take this first step seriously. A common mistake is to get lost in overly complex and unstable structures when starting a literature search.
Perhaps you have heard of PICO or other such problem formulation frameworks. In some cases, they can be useful tools for identifying the main thematic elements of our research question. However, our experience is that they often do more harm than good, precisely because they tempt us to start with an overly complex structure.
Once we have identified the thematic main elements of our research question, we must then find search keywords that match each of them. There are countless different ways to come up with search keywords. Here, we will describe a handful that we find particularly useful.
But before we proceed, we need to know that there are two different types of search keywords: controlled search keywords and "free" text keywords. See our explanation of controlled and "free" search keywords.
A general piece of advice before we delve into the specific methods: Take good notes during this process! There is little point in identifying good search keywords if you do not collect and systematize them. It may also be wise to keep an eye on how often a given keyword appears in the literature you consider most relevant. A good way to keep track is to draw a box for each of the main elements you identified in Step 1, and then continuously write in the keywords you find. Feel free to use this template.
The first and most obvious method for finding search keywords is to use your own and other knowledgeable people's ideas. What do we ourselves believe are the words used in the literature for our main elements? This method works better the more knowledge one has about the topic.
The second method involves using the database's controlled search vocabulary. In reference databases with controlled search vocabulary ("subject headings", thesaurus, synonym dictionary), there is always a function to search for keywords from the controlled vocabulary. It tries to "map" the keywords you enter to keywords from the controlled vocabulary. The suggestions sometimes fit well and sometimes less so. Once you get a suggestion, you can click on it and click around in the vocabulary to look at superior, subordinate, and related keywords, read definitions, and the like. You should invest some time in this so that you can make good decisions about which controlled search keywords to use.
The database interface for the controlled search vocabulary can also enable us to find usable search keywords that are not part of the controlled vocabulary as such, i.e., free text words. When we have found a controlled search keyword we want to investigate and click on it, we often get an explanation of how that keyword is used to index articles in the database. Sometimes there is a definition, and quite often also a list of synonyms. We usually find them under the heading "entry terms" or "used for". They can be handy to note down, to later use as free search text words when we build our search.
A third important method is to start with relevant articles we already have or that we have found in intuitive phrase searches in more intuitive search tools. Academic search engines like Google Scholar, discovery tools like Oria, Basic search in Ovid databases (Ovid Medline, Ovid Embase and Ovid PsycINFO), and some other tools (e.g. Keenious) can in slightly different ways take longer phrases and more natural language as input, and based on that, find the most relevant articles for us.
Once we then have found a handful of articles we know are releveant to our research question, we can use them to find good keywords for more extensive keyword searches in reference databases. Look for words in the title and see which keywords or key phrases the authors or publishers have chosen for the article. Additionally, you can look up the articles individually in a database, click into the complete database entry, and note which controlled search keywords the article is indexed with. We have created a separate little guide for how you can find search keywords from articles.
All these methods are illustrated in the examples below.
Now we are ready to construct our first keyword search in a proper bibliographic database. First, we choose the database we believe best covers our research question. Based on the main elements, which we have organized and illustrated in our box structure, we select the database we think best covers our research question. If you are unsure which one it is, please contact us at the library. The examples below can give you an idea about which databases are the first choice for various topics.
In a first attempt to build a good search, it is most important to proceed step by step: One main element of the research question at a time, and one search keyword at a time. The reference databases we use for structured searches all have an interface that lets us see all our previous searches in a given session – a so-called search history. When we search one keyword at a time, we can see how many results each individual search keyword contributes, and we can later use the search history to combine all the different search keywords with OR and AND, according to the structure and the main elements we have come up with.
It is crucial to stick to this step-by-step way of working. It is a common mistake to try to enter the entire search with all the keywords in combination as one search string. If we do that, we quickly lose control, and it becomes impossible for both ourselves and others to make a good assessment of the search quality.
In all our examples, we demonstrate this step-by-step way of working.
In a first attempt to build a search, it is less important that we include all conceivable relevant search keywords. We first use the most obvious ones from Step 2. This can give us an impression of how large the literature is in the area, and we can, both through the process of building the first search and by looking a bit in the results lists, get ideas for more good search keywords, so that we can build an even better search in the next round.
Our first attempt to build a well-structured keyword search in a good reference database will almost never be entirely optimal. Developing good searches is an iterative process. What we learn from working through and seeing the results of the first attempt will give us a good starting point for improving the search.
The assessment of how successful our first attempt is based primarily on how well the search succeeds in capturing the relevant literature. This requires that we skim through the search results to form an impression. When making such an assessment, you must be aware that the perceived relevance of the results is worse than what you are used to from more intuitive search tools like Google Scholar, since the result lists are typically not sorted by relevance.
Additionally, you can check if your search has captured the examples of relevant literature you had beforehand, or which you have picked out from phrase searches in the more intuitive tools.
The most common improvement of a first attempt is probably that we need to supplement with more search keywords. Sometimes we find that certain keywords we thought were good are not, and therefore should be removed. Often, the new keywords we add will be free text words.
Since free text words do not have a fixed form, like the controlled keywords do, we must be especially attentive to variations in the form and composition of words. We can account for variations using techniques such as truncation, wildcards, and proximity operators. In addition, we must also consider which metadata fields we want our search keywords to match with. If we only search for them in the title and the authors' own keywords, we get a narrower and more focused search than if we also search in the abstract.
Exactly how a search is improved will vary greatly, so we recommend that you look at some of the examples below, so you can get a sense of both what it can look like in practice, and what the variations can involve. We do want to note, however, that the improvements we demonstrate in the examples are somewhat abbreviated, compared to many typical search building processes, where you might need to make improvements in both two and three rounds.
No reference database contains absolutely everything. To ensure that we do not miss important relevant sources, we usually need to search in multiple databases.
The number of databases that should be searched will depend on the purpose of the search and your level of study. For most literature studies at the bachelor's level, it is often sufficient to use one database. For a master's thesis, it is usually necessary to search in more than one database. Discuss the need to conduct your literature search in multiple databases with your supervisor.
If you choose to search in more than one reference database, the trick is to keep the search as similar as possible between databases. The "free" text words you searched for in the first database can usually be reused when you search in other databases. However, some aspects of the original search must be "translated" to the other database. Not all reference databases use the same controlled vocabulary, so an important step when we "translate" a search strategy is to look up the corresponding controlled search keywords in the relevant controlled vocabulary. We must also be careful with special characters and operators like wildcards. These can vary from database to database and must also be "translated".
Examples
Here you will find a collection of level-adapted examples that demonstrate how the 5-step method is applied. The examples also show the practical search construction in the databases. The examples are based on different research topics from various subjects, and with different levels of ambition regarding delimitation and sensitivity.
