RP-WSP-U1 Writing Scientific Papers in English,
Mærsk Mc-Kinney Møller Institute, SDU.
Created by Bridget Hallam, Jan 2008, bridget AT mmmi.sdu.dk.
This web page describes the characteristics of good science papers. Basically, when you have finished a good paper, you should know:
This page is divided into the following themes:
A good paper has exactly one, specific, thing to communicate, and does this well. When you have finished the paper, you have no doubt what it was trying to say, or where its contents fit into the research picture. The results will be presented so that data can easily be understood, their significance and meaning obvious. Conclusions flow from the results, supported and justified by them.
Normally the message will be one of a few types:
Are you clear what this paper is trying to communicate?
The title and abstract should accurately summarise what the paper is about.
The introduction to a paper is vitally important. It sets the scene, showing the state of research in the tiny area covered by the work. It reminds people of the other work which impacts this, and explains where the information hole is that the paper aims to fill.
So a good introduction states clearly which research question the paper is addressing, how other people have succeeded in getting some way towards an answer, and how the work described in that paper aims to take us further forward.
A good introduction is also targetted correctly, using the right slant and the right level of explanation for the expected readership of the particular journal/conference.
Does the introduction tell you why the work was done?
Good papers should contain enough information about what was done to enable another scientist to replicate the experiments. (The idea being that only results that have been duplicated can count as "fact", unreplicated "facts" may just be an accident, the far end of a probability curve.)
We also need to be able to see for ourselves that, for example, the test groups and the control group(s) were balanced in anything that might be relevant.
In practise, many journals don't allow the space. Software journals are particularly bad about not believing that the hardware platform used affects the results(!).
Does the paper tell you enough about how the work was done, including anything that might affect the results?
The results section of the paper should contain the results(!). Actual measurements of those physical or functional properties relevant to assessing the system performance. No (or very few) interpretations, no extrapolations, no waffle, just the actual results. Not normally raw data, but data processed in standard and described ways to produce meaningful results. The processing methods should be carefully chosen and clearly laid out. See Bridget's and John's experimental methodology notes.
The paper should also tell us how much confidence we can have in the results by providing information about anything relevant, often statistical standard deviations, error bars, whatever. If any unexpected or outlying data was removed before analysis, there should be justification of this!
Of course, it helps if the results are also displayed prettily, but this is much less important.
In theory, negative results are just as important as positive results. It is just as scientifically valid and useful to state that this approach to this problem doesn't work, as it is to find an approach that does work. Unfortunately, only particularly reputable journals tend to publish negative results.
Do you now feel confident that you know exactly what the authors found out?
The discussion section is where interpretations, extrapolations, etc. should appear. Starting with concrete conclusions that come directly from the results themselves, then gradually becoming more abstract as the authors extrapolate into the future and into broader research areas.
A pure theory paper should use this space to explain exactly where the theory is immediately applicable, and then expand to other potential uses, mentioning where and why this theory is better than whatever people currently use.
Do you agree that the conclusions given are justified by the results?
Summary. Some journals like a summary, which should be similar to the abstract, but re-stated in different words.
Program. Where the journal prints programs, they should be well laid out and commented, easy for other programmers to navigate through.
Bibliography. Should be extensive, with several different authors cited, not mostly the authors' own earlier work!