Start with the claim’s job
Before judging a chart or a study, ask what its central sentence is trying to do. “Most respondents in this survey prefer recorded lectures” describes a group. “Students who watch recorded lectures are more likely to pass” predicts an outcome from an observed characteristic. “Offering recorded lectures increases pass rates” says that changing one thing changes another. These sentences may concern the same topic, but they need different evidence.
This distinction matters because evidence can be strong for one job and weak for another. A well-run survey might describe students’ preferences without showing what would happen if a school changed its teaching. A model might forecast pass rates accurately without explaining why some students pass. Matching evidence to a claim means checking both what was measured and what conclusion the evidence is capable of supporting.
Description: Is the picture faithful to the group?
A descriptive claim reports what was observed: a count, proportion, average, or pattern. Its first test is scope. If 200 volunteers at one campus answer a questionnaire, their responses describe those volunteers. They do not automatically describe every student at the campus, much less students elsewhere. To make a broader claim, ask how participants were selected, who was left out, and whether the question was understood consistently.
Measurement matters as much as sampling. “Time spent studying” could mean scheduled class time, independent reading, or any hour with a textbook open. Missing responses can also change the picture if people who skip a question differ from those who answer it. When evaluating a descriptive statement, look for a clearly defined group, a clear measure, and a time period. If those details are narrow, the wording of the conclusion should be narrow too.
Prediction: Does the pattern work on new cases?
A predictive claim says that information available now helps estimate something not yet known. For example, a course team might use attendance and early assignments to identify students who may need extra support. The useful question is not merely whether those measures lined up with final grades in past records. It is whether the approach performs well for students whose final grades were not used to build it.
A fair test keeps evaluation cases separate from the cases used to develop the prediction. It also avoids using information that would not be available at the moment of prediction, such as a late assignment to forecast an earlier outcome. Compare the approach with a simple baseline, like predicting from an early quiz alone, and examine mistakes as well as correct predictions. If the students, course, or grading system change, past performance may not carry over.
Prediction does not require a causal explanation. Attendance may help forecast a grade even if requiring attendance would not improve learning. Other factors could influence both. A responsible predictive claim states what outcome is predicted, for whom, using information available when, and how its performance was checked.
Causation: What would happen if something changed?
A causal claim asks about an intervention: would pass rates change if recorded lectures were offered? A simple comparison between students who use recordings and those who do not cannot settle that question. Students may choose recordings because of work schedules, illness, or difficulty with the material. Those differences could affect grades independently of the recordings.
One way to investigate causation is to compare groups assigned to different conditions by chance, when that is feasible and appropriate. Other study designs can also provide useful evidence, but they require a convincing account of why the comparison groups would have had similar outcomes without the change. Ask whether the proposed cause came before the outcome, what alternative explanations remain, and whether the result applies to the setting in the claim. Even persuasive causal evidence may support a limited conclusion, such as an effect in one type of course rather than every classroom.
Make the conclusion no bigger than the evidence
When reviewing a source, underline its main claim and label it description, prediction, or causation. Then write down the evidence in ordinary terms: who or what was observed, how measurements were made, what was compared, and which cases were used to test the result. If the claim shifts from “was associated with” to “improved,” pause; the second wording asks for evidence about an intervention that the first does not.
Good evaluation is not a contest to find a flaw in every study. It is a way to use each piece of evidence for what it does well. A small survey can reveal a concern worth exploring, a tested forecast can help plan support, and a credible comparison can inform a decision about change. The key is to keep the conclusion aligned with the question the evidence actually answers.
