How to understand a paper’s methodology
Methods sections are not rituals. They are the authors telling you what would have to be true for the result to mean what they say it means. Read them as a set of decisions, not as a list of tools.
Four questions, in order
- Design. What is compared with what? Same people over time, two groups, a model versus a baseline, a proof versus an assumption? If you cannot draw the comparison, you do not have the design.
- Sample or scope. Who or what entered, who was excluded, and how many remained? In a theory paper, what class of cases is claimed?
- Measures. What was counted or labeled as the outcome? A questionnaire, a lab value, a benchmark, an annotation scheme — name it.
- Analysis. What turned those measures into the headline number? A regression, a qualitative codebook, a training split, a proof step.
Write four short answers in your own words. If any answer is “they used AI” or “they did stats,” you have a brand name, not a method.
The sentence that usually matters
Somewhere in methods is a sentence that is easy to skip and expensive later: eligibility, randomization (or its absence), the primary outcome, the train/test split, the annotation agreement, the stopping rule. Search for “primary,” “excluded,” “random,” “split,” “inter-rater,” “powered.” Read that paragraph twice.
What would change the result?
A practical test: pick one decision and invert it. If they measured self-report, what if they had used a record? If they dropped dropouts, what if those people differed? If the benchmark is toy-sized, what happens on a messy file? You do not need to run the study. You need to know which inversion would embarrass the claim.
Jargon is often a pointer, not a wall
When a clause is packed (“afterload is set by aortic impedance”), do not start with a textbook chapter. Stay in the paper: the previous sentence and the next figure usually define the term for this text. Select the clause if you need a restatement, then check that the restatement still uses the paper’s comparison, not a generic one.
PDF Lens can explain a methods sentence beside the page. It cannot see a figure that never became text, and it does not OCR scans. If the method lives in a diagram, you still have to read the diagram.
Methods vs. marketing
Introductions sell a gap. Methods either close it or do not. If the introduction promised “real-world” and the sample is 18 undergraduates in one hour, you have learned something important about the paper — before any p-value.
Next: limitations and extracting evidence.