How to Evaluate the Evidence in a Research Article

Finding the results in a research article is the easy part. They are in the figures, in the tables, in the section helpfully labeled “Results.” The hard part is the question that comes next: is this evidence actually good enough to support what the authors are claiming?

Most readers answer that question by proxy. The journal is reputable, the authors are at a good university, the methods are reasonable, the statistics look sophisticated, and so the evidence is probably fine. This guide replaces the proxies with a short set of questions you can ask of the evidence in any article, in any field, without being an expert in the methods. It assumes you have already found the article’s main Claim. If you have not, start with How to Read a Scientific Paper.

First, separate the evidence from everything around it

Evidence is the curated set of data that support the main Claim. It includes what this study observed, measured, modeled, or collected. It is not what other studies found (that is Context), not what the authors conclude from it (that is the Claim), not the argument connecting the two (that is Reasoning), and not what would happen if the claim is true (that is Implications). Keeping these apart may seem pedantic, but it is the single most useful habit in critical reading, because articles routinely blur and occasioankky omit them. An abstract sentence like “We find that X increases Y, using a sample of 1,200 patients” fuses a Claim with compressed Evidence. A discussion sentence like “Our results, consistent with Smith et al., suggest…” blends Evidence, Context, and Reasoning in eleven words.

So begin by listing the evidence on its own. For the main Claim, which figures, tables, or reported numbers directly support it? Write each one down as a plain statement of what was observed: “Group A scored 14 points higher than group B on the test, with overlapping error bars in two of the four trials.” No interpretation yet.

The three questions that matter most

1. What does the evidence actually show?

Be precise, and be stubborn about it. If the authors cite their Figure 3 as showing an effect, look at Figure 3 and decide for yourself what it shows. Read the axes. Find the units. Notice what the comparison is: treated versus untreated, before versus after, this population versus that one. Notice how much variation there is around the averages or the level of uncertainty. Is the signal greater than the noise? A surprising number of Claims rest on a difference that is visible in the summary statistic and invisible in the raw data.

Then check how the evidence was produced. You do not need to understand every step of the methods, but you do need the answers to four questions: What was measured? In whom or what? How many? And by what instrument or procedure? The answers set the limits on what the evidence can show. Self-reported food intake can show what people remember eating; it cannot show what they ate, and it can be subject orecall bias. A survey of first-year students at one university can show something about that group; whether it generalizes to another group or context is a separate question the evidence cannot answer on its own.

2. Is the evidence relevant to the Claim?

This is where the Claim you wrote down earns its keep. Put it next to the evidence and ask whether they are about the same thing. The gaps are often small and consequential. The Claim is about anxiety; the evidence is about time spent in the open arm of a maze. The Claim is about learning; the evidence is about test scores one week later. The Claim is about a drug’s effect in patients; the evidence comes from cells in a dish.

None of these gaps is necessarily fatal. Science proceeds by measuring proxies. But each gap is a step the authors have to argue across, and that argument belongs to the Reasoning, not the Evidence. If you spot a gap and the discussion section never addresses it, you have found a real weakness.

3. Is the evidence sufficient?

Sufficiency has several parts, and the most useful checks are ones that need no statistical training:

  • How many? Small samples produce noisy results and overestimate effects. There is no universal threshold, but if the number of subjects, trials, or observations surprises you as low, it probably surprised the reviewers too, and the authors should say why it was enough.
  • Compared with what? A result without a comparison group is a description, not a test. Look for the control condition, and ask whether it was a fair one.
  • How big, and how uncertain? A difference can be statistically significant and practically trivial. Look for the size of the effect in real units and for the error bars, confidence intervals, or ranges around it. If the paper reports only p-values, that is itself worth noting.
  • How many ways was it checked? Strong articles test a Claim more than one way: a second measure, a second population, a replication, a model that predicts something new. One figure supporting a big Claim is thin.
  • What did the authors themselves flag? Read the limitations paragraph carefully. Authors are usually honest about the weaknesses they know of. However, the ones they omit are the ones to think hardest about, for example under what conditions would the evidence support a different claim?

Behind all of these sits a broader point. Across many fields, systematic efforts to repeat published studies have found that a substantial fraction do not reproduce, and that effects which do reproduce tend to be smaller than first reported. A single article is a report of one attempt, not a settled fact. Reading it well means holding the Claim at the level of confidence the evidence has earned, which is often “promising” rather than “shown.”

A quick test: can you write 1-3-1?

Here is a check that catches most problems. Try to write five sentences: one stating the Claim, up to three stating the specific pieces of Evidence, and one stating the Reasoning that connects them.

If you cannot fill the three evidence sentences, the Claim is resting on less than it appears to. If you can fill them but the reasoning sentence comes out as “and therefore the Claim,” with nothing in between, the authors have passed the essential logic step to the reader. If the reasoning sentence is easy to write and the evidence sentences are specific, you are probably looking at a sound argument, whatever its field.

Rating the evidence

When you compare articles, or when you have to decide how much weight to give one in your own work, a simple four-level scale is enough:

  • Strong. Directly relevant to the Claim, adequate sample and comparison, effect size and uncertainty reported, supported more than one way.
  • Adequate. Relevant and reasonably sized, but resting on a single measure or line of evidence.
  • Weak. A real gap between what was measured and what is claimed, a small or unrepresentative sample, or missing uncertainty.
  • Missing. The Claim is asserted but the article contains no evidence for it; it is being carried by citations or by assumption.

You will find that “missing” is more common than you expect, especially for secondary claims tucked into discussion sections. Noticing it is not cynicism. It is the difference between reading an article and being read to by one.

Evidence is one element of five

Evaluating evidence is the second of the five CERIC questions: Claim, Evidence, Reasoning, Implications, Context. The free one-page cheat sheet lists all five, where to find each in a typical article, and the question to ask. The method is taught step by step in the free Foundations Course 1, and in full, with worked examples from chemistry to astrophysics to education research, in How to Critically Read the Scientific Research Literature (Cambridge University Press, 2026). Running a journal club or seminar? See How to Run a Journal Club That Actually Teaches Critical Reading.