Computerized Adaptive Testing with the Rasch Model (CAT-Rasch)
Computerized Adaptive Testing with the Rasch Model · Also known as: CAT-Rasch, Rasch-based CAT, adaptive Rasch testing, computerized adaptive measurement
Computerized adaptive testing with the Rasch model selects items in real time based on each examinee's evolving ability estimate, so that every person receives a test precisely calibrated to their proficiency level. The result is a shorter, more efficient measurement instrument that loses none of the precision of a full-length fixed-form test.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
CAT-Rasch is the right choice when you need precise, efficient measurement of a unidimensional latent trait and can invest in building a calibrated item bank of adequate size (typically 100+ items). It excels in large-scale educational assessment, clinical outcome measurement, and patient-reported outcomes where administration time is a practical constraint. Do not use it when the construct is multidimensional without modification, when your item bank is too small to sustain meaningful selection across the ability range, when the Rasch model's equal-discrimination assumption is seriously violated by your items, or when examinees lack access to computers or the testing context does not support adaptive delivery.
Strengths & limitations
- Achieves the same measurement precision as a full fixed-form test with roughly 50 percent fewer items, reducing respondent burden.
- Provides an individualized, tailored assessment experience that keeps items at the appropriate difficulty level for each person.
- Yields a conditional standard error of measurement for every examinee, making the precision of each score transparent.
- The Rasch framework offers strong testable assumptions (fit statistics, invariance checks) that allow systematic quality control of the item bank.
- Enables real-time scoring and immediate feedback at the end of the session.
- Requires a large, well-calibrated item bank, which demands substantial upfront investment in item writing, piloting, and IRT calibration.
- The equal-discrimination constraint of the Rasch model may be too restrictive for some item sets; violation inflates measurement error.
- Item exposure must be actively managed to prevent high-information items from being overused, which adds algorithmic complexity.
- Security of the item bank is a persistent operational challenge in high-stakes settings.
Frequently asked
How is CAT-Rasch different from CAT based on the 2PL or 3PL model?
The Rasch model assumes all items share the same discrimination and (for multiple-choice) guessing is negligible, so items differ only in difficulty. The 2PL and 3PL models add discrimination and pseudo-guessing parameters respectively. Rasch-based CAT produces simpler, more interpretable item banks and imposes stronger, testable invariance requirements; 2PL/3PL CAT allows more flexible item behavior but requires larger samples for calibration and more complex item selection logic.
How many items does a CAT-Rasch test typically administer?
Item count depends on the stopping criterion and trait distribution of examinees. In practice, CAT-Rasch tests commonly administer 10 to 30 items to achieve a conditional standard error below 0.30 logits, compared with 50 to 100+ items for a fixed-form test of equivalent precision.
What sample size is needed to calibrate a Rasch item bank for CAT?
Stable Rasch item calibration typically requires at least 200 to 500 respondents per item under conventional guidelines; Wright and Stone suggest a minimum of 100 but with wider standard errors. Larger banks for high-stakes CAT are usually calibrated with samples of several thousand to ensure stable difficulty estimates across the full ability range.
Can CAT-Rasch be used for polytomous items such as Likert scales?
Yes. The Partial Credit Model and the Rating Scale Model are Rasch-family extensions for polytomous items and are routinely incorporated into CAT item banks for patient-reported outcomes and attitude measurement, using analogous item-selection and ability-estimation logic.
How is model fit evaluated before deploying a CAT-Rasch bank?
Standard Rasch fit statistics — infit and outfit mean-square statistics — are examined for each item. Values between about 0.7 and 1.3 are generally acceptable; values outside this range signal items that behave more erratically or less discriminatingly than the model expects and should be revised or dropped before the bank is used adaptively.
Sources
- Wainer, H. (Ed.). (2000). Computerized Adaptive Testing: A Primer (2nd ed.). Lawrence Erlbaum Associates. ISBN: 978-0805835113
- Wright, B. D. & Stone, M. H. (1982). Best Test Design: Rasch Measurement. MESA Press. link ↗
How to cite this page
ScholarGate. (2026, June 3). Computerized Adaptive Testing with the Rasch Model. ScholarGate. https://scholargate.app/en/psychometrics/computerized-adaptive-test-rasch-model
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Confirmatory factor analysisPsychometrics↔ compare
- Differential Item FunctioningPsychometrics↔ compare
- EFAStatistics↔ compare
- Item Response TheoryPsychometrics↔ compare
- Rasch ModelPsychometrics↔ compare