Extracting Evidence
The process of extracting evidence from the literature is one that can probably
never follow any set rules. Different researchers will find different ways of maximising
their efficiency. Here we describe one approach that has been successfully used
in our research projects.
1. Select papers for extraction
An initial assessment of whether the paper contains evidence is usually possible
from reading the abstract. In our experience 70% of hits from literature database
searches are irrelevant to the questions we are asking.
2. Read abstracts to filter 'relevant' literature
Read carefully to make sure the study is related to the question you are asking,
and seems to contain evidence (be prepared to be wrong occasionally).
3. Print and highlight, or annotate on-screen
For papers retained after reading the abstract, I prefer to print them out (being
as efficient as possible with paper). I need to be able to highlight and annotate
the paper as I read.
4. A focused approach to reading
Despite the effort that goes into writing it, the Introduction is not going to tell
you much about what the researchers found. It can help you to understand what question
they were asking. Concentrate on the Methods and Results, and to a lesser extent
the Discussion. This is quite a different way of reading a paper to what most of
us are used to. We specifically address the following sections.
Introduction
What question/s were the researchers asking?
Methods
Information on experimental design, number of replicates and statistical analysis
used. These allow us to assess study quality.
Results
Information on effects seen, including the effect size, dose-response, p-values
from statistical tests, etc. This is the bulk of the evidence.
Discussion
Helpful for assessing coherence of the results. Do they make sense in terms of existing
literature / theory?
5. Annotate
I use different coloured highlighters to identify different pieces of evidence in
a study – there is often more than one. One colour is used for general information
(e.g. study design, replication, climate, location, spatial scale), and then individual
colours used for each piece of evidence I find in the paper (usually concentrated
in results and discussion, but occasionally in methods as well). Highlighting different
pieces of evidence in different colours is invaluable when it comes to entering
that evidence into the database.
6. Enter extracted evidence into database
I do all the pieces of evidence at the same time. I find it easier to highlight
all the evidence in the paper and enter it later, rather than to attempt to enter
it while reading. If entered while reading, there is a greater chance of having
to back track to change misinterpretations, and there is always the possibility
of getting timed out of the database and losing all changes made to the page to
that point.
When a paper contains multiple pieces of evidence, they are usually related. Using
the 'duplicate' button on the View Evidence page can greatly reduce the amount of
work needed to input several pieces of evidence, and also make the language used
more consistent. Be careful to go through all fields in the new record to assess
which ones do and do not need to be changed.
And lastly, we recommend FILLING OUT EVERYTHING in the evidence record, unless there
really isn’t any information (e.g. it's a factorial experiment so 'dose-response'
is not relevant). It is easy to leave out information, particularly the Question
being asked, the various Detail fields, and page details for the evidence. But doing
this makes it less likely that a future user will be able to make sense of your
evidence record, and will have to repeat the extraction.