How to Read a Research Paper
Learn to scan research papers in under five minutes.
The key to reading many research papers is to know how to scan a journal article in under five minutes for its most salient ideas and conclusions. Most papers are not written with the intent of providing the conclusions upfront, so you have to learn how to perform research paper hopscotch. Otherwise you waste hours reading papers that lead nowhere.
A generic research paper looks roughly like this:
Research paper hopscotch
Before you read anything, you might want to check the journal whether it is a reputable or familiar one. Predatory journals typically charge fees to publish without proper peer review. They are often not indexed in PubMed or Scopus. Moreover, their editorial boards tend to be anonymous. Avoid such journals.
Preprints may lack peer review, but in certain fields (e.g. physics or computer science), preprint servers are where the latest ideas are found. Just because a study is peer-reviewed also does not mean it is guaranteed to replicate.
Similarly, if standard terms are redefined, the methodology is vague, there are no links to code or data, or the results are too good to be true, beware of potentially wasting your time.
1. Read the title: < 5 seconds (mandatory)
Obviously, the title is the first thing that grabs your attention. Read it and decide whether the article might be of interest to you.
Note that many authors are terrible at coming up with good titles for their research papers, so do not be dismayed if it sounds unintelligible or bland. That’s why you have to continue to the abstract.
2. Scan the abstract: < 1 minute (mandatory)
The abstract ought to contain the main conclusion, ideally as the first line, but in most cases it is the last or close to it. If the abstract opens with any variation of “In this paper, we…“ or “Based on recent work…“ you can be sure to go to the last line of the abstract and read upwards. The same applies when it opens with a generic statement about a popular technology, such as “Quantum computing / AI stands to transform industries.”
Note that some authors refuse to provide a summary and instead do what most LLMs do when asked for one: they state what was done instead of what was found. For instance, they write: “We provide a rigorous analysis of the impact of…“. All such authors and LLMs achieve is to infuriate readers as they—but perhaps it is only I—wish to exclaim: And what was the result of your damned rigorous analysis? It’s a bit similar to the various skincare products that have been dermatologically tested. Just because the cream was tested in a lab does not mean its results were any decent.
Anyway, if by now you do not know if the research paper is of interest to you, you have three possibilities:
- It is not of any interest to you, so you can move on.
- It might be of interest, but you lack the background knowledge to appreciate it.
- It is relevant, but the authors are terrible writers, so they make their ideas needlessly hard to grasp.
My recommendation is to skip the paper in all three cases. If you lack the background to appreciate the article, you are not going to learn much from it without following up on introductory texts: go to the introduction for more information and such references. Foundational or review papers are key sources to bone up on knowledge.
But if your background is not the problem, yet neither the authors nor the editors cared to be clear, it is generally not worth your time either. There may be rare exceptions, but it is a sensible rule of thumb.
3. Jump to the conclusion: < 5 minutes (mostly mandatory)
The conclusion is where the main ideas are summarized and placed in context. Read this section carefully to understand what the authors are claiming. For theoretical and mathematical papers, read the main theorems, as they are the equivalent of the conclusions. Just skip the proofs for now.
When the conclusion is not relevant to your inquiry, drop the paper immediately. Otherwise, go on.
4. Check for limitations (optional)
If the conclusion is surprising or you wish to know more about the limitations of the study, check the discussion. Mathematical research papers may not have such a section, as the main ideas are available in lemmas and theorems; the axioms contain the limitations by exclusion.
5. Analyse the method (optional)
The experimental setup or details of the analysis are crucial to determine the validity of the conclusions. Again, for theoretical or mathematical papers, you might not encounter such a section. Instead, go through the proofs of the theorems.
For research on humans (e.g. psychology, sociology, medicine, economics), please review the selection of subjects carefully. You might find out that the conclusions are not really well supported because the sample size is too small or the selection procedure is highly biased (e.g. psychology undergraduates). Likewise, medical research papers might claim of an amazing effect in curing a horrible disease… in mice. Beware!
The experimental setup is also crucial to determine whether the results can be replicated in another lab or not. If only one lab in the world has the tools to verify, be very sceptical, especially when the results sound almost too good to be true and/or come from a corporation’s research division in a market full of hype. CERN’s Large Hadron Collider (LHC) and Webb Space Telescope (JWST) are somewhat similar in that not many organizations can replicate their equipment, except studies that rely on these facilities are usually conducted by hundreds of researchers with no commercial agenda. Microsoft, on the other hand, has a habit of making grandiose yet unconvincing claims with regard to topological quantum computers that fail upon scrutiny or basic replication. More details are always to follow rather than already shared.
6. Check the background (optional)
Whereas the introduction is primarily a history lesson, the background lists key terms, symbols, and assumptions. If you are familiar with the field, you can skip the introduction, though usually the background is still useful to read through.
7. See the larger context (optional)
The introduction is as much a (bad) sales pitch as a literature review. Read it to spot spin. It usually contains a description of the prior research, so it answers the question: What has been done/tried before? It is almost always in prose and heavy on references to ensure the editors and readers can see the authors have done their homework and to boost citation counts needed for tenure. After all, many research findings are false because of perverse incentives in academia. Plug the DOI into Connected Papers to see whether you are in a citation echo chamber or not.
What about the references?
The references are there for you to follow up on related research. The only time you really want to check is whether authors mostly cite themselves. If that’s the case, you might be dealing with crackpots.
Can’t I use an LLM?
Sure you can! But you must be acutely aware that they have a tendency to exaggerate results or refuse to admit ignorance, especially as models become larger. Just because the LLM says so, does not mean it is so.