On this page
Practice testing is one of the most reliable ways to remember what you study. But in JEE, NEET-UG and board exams, the questions are rarely the ones you practised. They ask you to apply an idea to a new situation or in a new format. So the real question for students is: does practising with tests help on questions you have never seen? A 2018 meta-analysis by Steven Pan and Timothy Rickard looked at exactly this. Here is what it found.
What was already known?
The testing effect is well established for memory. In a classic study, Roediger and Karpicke (2006) had students read prose passages and then either restudy them or take recall tests. On a test five minutes later, restudying did better. But on tests two days or a week later, the students who had practised recall remembered substantially more, even though restudying made students more confident. That is retention of the same material. Transfer, using what you learned on different questions, is a harder test.
What question did the meta-analysis ask?
Pan and Rickard published Transfer of test-enhanced learning: Meta-analytic review and synthesis in Psychological Bulletin (volume 144, 2018). They asked whether learning from practice tests transfers to different contexts and, if so, when.
We could read only the abstract, so the numbers below are from the abstract.
Who and what was studied?
According to the abstract, the review covered:
192 transfer effect sizes from 122 experiments in 67 published and unpublished articles.
10,382 participants in total.
More than 40 years of research.
Each effect compared practice testing with a "nontesting reexposure" control, meaning students who saw the material again without being tested.
What did it find?
According to the abstract:
Overall, testing produced transferable learning compared with restudy, d = 0.40 (95% confidence interval 0.31 to 0.50).
Transfer was greatest across test formats (for example, practising with short answers and being tested with multiple choice), to application and inference questions, to medical diagnosis problems, and to related word cues.
Transfer was weakest to rearranged stimulus-response items (being asked the "reverse" of what you practised), to untested material seen during initial study, and to problems involving worked examples.
Three factors strongly influenced whether transfer happened: response congruency (whether the answer you practise producing matches what the new question needs), elaborated retrieval practice (retrieving with explanations, not just bare facts) and initial test performance (how well you did on the practice test).
After correcting for publication bias, the overall estimate was substantially reduced, often to no positive transfer when none of those three factors was present.
What does it not show?
We only have the abstract, so we cannot give effect sizes for each subgroup or say exactly which studies used school students.
The abstract does not say how many studies were run in real classrooms, or with school-age students. Two years of JEE preparation is a different setting from most experiments.
Worked-example problems showed weak transfer. This matters for physics and maths: testing yourself on the final answer of a solved example may not help you solve a new problem.
The bias correction matters. Without the three helpful factors, the corrected estimate often showed no transfer at all.
What does it mean for students in India?
The three factors in the abstract translate directly into how you build practice:
Elaborate when you retrieve. Don't just recall "v = u + at". Recall when it applies (constant acceleration only) and why. Writing a one-line reason with each answer is elaborated retrieval.
Aim for high initial success. If you are getting most practice questions wrong, the test is not building transferable knowledge yet. Review, then retest at a level where you can succeed most of the time.
Match the response to the exam. If the exam asks you to apply a concept to a new situation, practise application questions, not only definitions. Our post on writing your own practice questions shows how to make "what if" variations.
Mix formats. Transfer across formats was strong, so alternate between short answers, multiple choice and full numericals on the same idea.
Don't test only the answer of a worked example. Cover the solution and reproduce the reasoning, step by step.
Flashcards still help memory; see how to make flashcards that work. But add application questions on top.
When your practice question needs a full numerical rather than a recall answer, the Fermi Problem Library has worked problems with a step-by-step solver that explains the reasoning, so you can check each step of your own attempt rather than just the final number.
How to read a study like this
When a meta-analysis reports a publication-bias correction, compare the corrected number with the headline. A large drop means the average effect partly reflects which studies got published. Here, the authors say the moderator effects held up but the baseline did not, so the useful message is about conditions, not the overall average.
Next step: take ten questions you have already got right this week and, for each, write a new version that changes the situation, then solve it.
