On this page
Some tests change as you take them. Get a question right and the next one tends to be harder; get it wrong and it tends to be easier. This is computer-adaptive testing. The GRE uses a version of it, and so do many practice platforms. Knowing how it works helps you read your scores properly and stops a run of hard questions from rattling you.
How does an adaptive test choose the next question?
ETS, which runs the GRE, describes a question-level adaptive test like this: the first question is typically at about middle difficulty; if you answer correctly, the computer is likely to give a harder question, and if you answer incorrectly, an easier one. As you go, it builds up a pattern of responses and estimates your ability from it. In ETS's words, "high scores equal hard questions that were answered correctly."
The GRE itself adapts by section rather than by question. The first Verbal and Quantitative sections are of average difficulty, and the difficulty of the second section depends on your performance on the first. Within a section you can skip, mark and return to questions. ETS notes that this kind of preview and review is not possible in a question-level adaptive test.
What is item response theory?
Item response theory (IRT) is the maths underneath. Frank Baker's free primer, The Basics of Item Response Theory, explains it in three ideas.
Ability is a number on a scale. Each test taker has an ability, written θ.
Each question has a curve. The item characteristic curve shows the probability of a correct answer at each ability level. It is S-shaped: near zero for low ability, near one for high ability.
Two properties describe the curve. Difficulty is where the question "functions" on the ability scale, a location. Discrimination is how steep the curve is in the middle, which shows how well the question separates students just below that point from those just above it.
Baker's three-parameter model adds a guessing parameter c, "the probability of getting the item correct by guessing alone":
P(θ) = c + (1 − c) × 1 / (1 + e^(−a(θ − b)))
Here b is difficulty, a is discrimination and c is guessing. Baker also explains why adaptive tests pick questions near your level: an item measures ability with greatest precision at the ability level corresponding to its difficulty. A question far too easy or far too hard tells the computer very little about you.
A worked example
Take an illustrative question with b = 0, a = 1 and c = 0.25 (a four-option question where blind guessing gets one in four right).
A student at θ = 0: the logistic part is 1/(1 + e^0) = 0.5, so P = 0.25 + 0.75 × 0.5 = 0.625.
A student at θ = 2: 1/(1 + e^−2) ≈ 0.881, so P ≈ 0.25 + 0.75 × 0.881 ≈ 0.91.
A student at θ = −2: 1/(1 + e^2) ≈ 0.119, so P ≈ 0.25 + 0.75 × 0.119 ≈ 0.34.
The question separates these three students well. For a student at θ = 3, almost any easy question would be answered correctly and would tell the test little, which is why the computer moves on to harder ones.
Is adaptive testing accurate?
David Weiss and Gage Kingsbury described classroom uses of adaptive testing in 1984. For deciding whether a student has mastered a topic, they reported evidence that the adaptive procedure gave more accurate mastery classifications than conventional mastery tests while using fewer questions. That efficiency is a big part of the appeal: fewer wasted questions.
What does this mean for how you practise?
Read the right number. Percentage correct means less in an adaptive test, because a strong student sees harder questions. Look at the ability estimate or level the platform reports.
Check your real exam's format. Patterns change. For JEE Main or NEET-UG, read the current NTA information bulletin rather than assuming a format. If your exam is a fixed paper, practise full fixed papers under time too, as covered in our guide to using previous-year papers.
Use adaptive practice to find your edge. Questions near your level are where you learn most about your gaps. Log every miss and review it, as in our mock test data guide.
Don't game it. Deliberately missing early questions only gives the test a worse estimate of you.
For JEE Main and Advanced students, Fermi Testprep includes adaptive mocks and a study plan. Use the results as a map of what to fix next, then go back to the questions you missed.
