How to Read a Research Paper—and Decide Whether to Trust It

Before highlighting sentences, inspect the title page and publication record. Is this an original study, a review, a theoretical argument, a methods paper, an editorial, or a preprint? These documents do different jobs. An editorial can offer a valuable interpretation without presenting a new experiment; a review can synthesize many studies without collecting new participant data.

The Cornell University Library: Peer review and article types explains peer review and common article types. Check the item’s actual status rather than assuming that a scholarly-looking PDF has been peer reviewed. A preprint may later have a published version with changed analyses or wording. Record which version you read and use the publication record to look for a later version.

Write your own reading question before continuing: ‘Can this paper explain my concept?’, ‘Does it justify this method?’, or ‘Can it support this particular claim?’ This keeps relevance separate from prestige. A highly cited paper may address a different population, outcome, or context from the one in your paragraph.

S. Keshav: How to Read a Paper describes a staged approach to reading papers. Use an initial orientation, a closer examination, and a deeper check when the paper is central to your work. The workflow below is an adaptation for thesis reading; it does not require a fixed number of minutes or an identical reading order for every discipline.

On the first pass, inspect the abstract, section headings, figures, and conclusion to map the question and proposed answer. Mark unfamiliar terms. On the second, examine the evidence and methods closely enough to explain how the authors reached the result. On the third, challenge the interpretation, inspect important supplementary details, and compare relevant sources. Return to the abstract: does it fairly represent what you found?

Decide how much depth the paper deserves. A background source may need a careful check of one relevant argument. A paper underpinning your central method or conclusion requires deeper engagement. If you cannot evaluate a decisive technical step, record the uncertainty and seek an explanation from a methods adviser or an appropriate reference.

Many empirical papers separate introduction, methods, results, and discussion, but labels vary. Qualitative, theoretical, historical, and mathematical papers may organize their reasoning differently. Follow the function of a section rather than treating a particular heading as compulsory.

Find the boundary between the authors’ evidence and their interpretation. The introduction may repeat another researcher’s claim. The discussion may propose an explanation that was not directly tested. Both can be useful, but neither should be attributed to the paper as a demonstrated result without checking what the study actually did.

Identify the unit of analysis and the actual evidence. Were the authors studying people, classes, organizations, documents, or repeated events? What was selected, what was excluded, and what comparison was possible? A large number of observations does not remove a systematic sampling problem or make every observation independent.

Examine the measurement. If the paper asks whether students learned more, did it measure assessed performance, self-reported confidence, or satisfaction? Each can be meaningful, but they answer different questions. Notice when the discussion uses a broad concept that is narrower in the instrument or dataset.

Match causal language to the design. An association can have several explanations, including selection or a third variable. Experimental allocation may strengthen a causal comparison, while questions about missing outcomes, implementation, and transfer to another setting remain. For interviews or interpretive work, examine how cases were selected and how the interpretation is grounded in the material. Use the research-design guide when you need more detail on those distinctions.

Start with the caption, then identify the axes, units, groups, denominator, and time period. Check whether an axis starts at zero or uses a logarithmic scale, whether points are individual observations or summaries, and what any error bars represent. A standard deviation, standard error, and confidence interval describe different quantities; the figure or methods should identify which is shown.

For a numerical comparison, write down the magnitude in its actual units. Ask whether the comparison is absolute or relative and whether the baseline is visible. Then consider uncertainty and practical meaning. A graph can faithfully display a result while a sentence about it implies a much broader benefit.

The American Statistical Association: Statistical significance and p-values explains why a p-value should not be treated as the probability that a hypothesis is true or as a measure of an effect’s importance. Read it alongside the estimate, uncertainty, study design, and analytical choices. A threshold alone cannot tell you whether a finding is useful for your question.

Turn each limitation into a consequence for interpretation. A single setting may constrain transfer. An imperfect measure may leave part of the concept unobserved. Missing follow-up can change which participants are represented in the final result. Avoid converting these observations into an automatic pass/fail score; decide which uncertainty matters to the specific claim you intend to cite.

Read funding and competing-interest statements as context for scrutiny. They do not, by themselves, establish that a finding is valid or invalid. Examine whether the design, reporting, access to data or materials where appropriate, and independent evidence let you evaluate the claim. Publication venue and citation counts are signals to investigate, rather than substitutes for reading.

Open the publisher’s current article page and look for linked notices. Crossref: Checking publication updates with Crossmark can expose registered corrections, retractions, and other updates where it is implemented. Crossref explicitly cautions that the badge is not a guarantee. An absent badge or search result does not establish a clean publication record. If a notice exists, read it and determine what it changes before relying on the source.

For practice, open Salganik, Dodds, and Watts (2006): Experimental study of an artificial cultural market. This published Science study provides a concrete example of connecting a question, comparison, outcome, and limitation. The locators below refer to printed pages 854–855, which are the first two pages of the linked PDF.

Before highlighting more text, close the paper and explain its contribution in a few sentences. Reopen it and compare your explanation with the evidence. This reveals gaps that a page full of highlights can hide. Keep exact quotations visibly separate from your paraphrases and interpretations.

Use the worksheet below to record a source’s role in your project. Fill it from the publication and your own reading. If a detail is missing or you have read only an abstract, record that limit rather than completing the field with a guess. A note is useful when it lets you return to the evidence and understand why you saved the paper.

Finish with a decision: use the paper for a specified purpose, read further before deciding, or leave it out. A credible source can be irrelevant to your claim; a relevant source can have a limitation that prevents the use you intended. Narrowing your sentence is often more responsible than searching for another citation to decorate an overstatement.

For a broader evaluation framework, see Cornell University Library: Evaluating information. When moving from a reading note into your draft, consult the citation guide for attribution and the literature-review guide for synthesis across papers. The goal here is to arrive at those writing tasks with an accurate understanding of the individual source.