The IRT model in practice
Item Response Theory replaces the raw test score with a probabilistic view: for any person and any question, what is the likelihood they answer correctly? The sigmoid curve captures the intuition that ability and difficulty interact. A student with ability 1.5 might have a 75% chance on a moderate item, but only 10% on a hard item.
This probability model lets you compare people and questions on the same scale, even when they answered different items. A student with high ability can be estimated from just their hard items; another with low ability estimated from their easy items. The math disentangles person ability from question difficulty.
Why IRT matters for assessment design
Traditional scoring sums right and wrong counts, treating all items as equal. IRT recognizes that a correct answer to a hard item carries more information about ability than a correct answer to an easy item. This unlocks adaptive testing: show harder items to able students, easier items to struggling students, and converge on an accurate ability estimate faster.
Test makers also use IRT to spot biased items. If two groups with equal ability have different probabilities on the same question, that signals something is unfair or ambiguous about the wording. IRT analysis catches these before tests go live.