IRT

• assumes that every test item has a difficulty, and different items have different difficulties (Embretson & Reise, 2000)
• Unfortunately, IRT generally assumes that test items are conditionally independent given the student’s competence. This is seldom true of the raw measures collected at the step level by tutoring systems
• Although IRT has powerful features, such as calibration algorithms that empirically determine item difficulties and other parameters, considerable work is needed before it can be applied to tutoring systems.