Partial Credit Model (PCM / GPCM)
Partial Credit Model · Also known as: Kısmi Kredi Modeli (PCM / GPCM), Generalized Partial Credit Model, GPCM, PCM
The Partial Credit Model is an extension of the Rasch measurement framework designed for ordered polytomous items — items whose responses fall into more than two ordered categories, such as partial-credit tasks in performance assessment or open-ended scoring rubrics. Proposed by Geoff Masters in 1982 and later generalised by Eiji Muraki in 1992, the model estimates a separate threshold (step) parameter for each adjacent-category transition within every item, allowing fine-grained calibration of how much each additional credit level contributes to locating a person on the latent trait.
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
Use the Partial Credit Model when items have three or more ordered score categories that represent qualitatively distinct levels of performance or agreement — for example, open-ended items scored by rubric, partial-credit mathematics tasks, or rating-scale items where Rasch-family interval-level measurement is desired. The model requires ordered polytomous data; it is not appropriate for nominal categories without a natural order. A minimum of approximately 150 respondents is needed to obtain stable step parameter estimates; fewer items than five make calibration unreliable. If step disordering is widespread across items, consider merging categories or switching to the Graded Response Model. The GPCM is preferred when items visibly differ in how steeply their category response curves rise across the trait scale.
Strengths & limitations
- Extracts more measurement information per item than dichotomous IRT by modelling each score transition separately.
- Produces person and item estimates on the same logit scale, enabling direct interpretation of how much ability is needed to earn each additional credit level.
- The Wright map gives an intuitive visual summary of person-item alignment, revealing gaps or redundancies in the instrument.
- The GPCM extension accommodates items of differing discrimination without abandoning the polytomous IRT framework.
- Requires a substantially larger sample than dichotomous Rasch models; at least 150 respondents are recommended, and GPCM typically needs more.
- Step disordering — where a higher category is easier than a lower one — signals scoring rubric problems and must be addressed before reporting results.
- The logit scale is not inherently meaningful to non-technical audiences; communicating results requires translation into substantive terms.
- Software implementation and interpretation demand familiarity with IRT concepts that are uncommon outside specialist psychometric training.
Frequently asked
What is the difference between PCM and GPCM?
Both models estimate a step difficulty for each adjacent-category transition. The PCM (following the Rasch tradition) constrains all items to have the same discrimination — that is, the slope of each item's category response curve is fixed to 1 on the logit scale. The GPCM removes this constraint and estimates a unique discrimination parameter per item. If your goal is to preserve Rasch measurement properties such as specific objectivity and sample-independent item parameters, use PCM. If items clearly differ in how sharply they separate people, GPCM fits better at the cost of losing the Rasch property.
What is step disordering and how should I handle it?
Step disordering occurs when the estimated difficulty of a higher score category (e.g., moving from 1 to 2) is lower than that of a lower category (moving from 0 to 1). This means respondents skip the intermediate category — they almost never use it — indicating a scoring rubric problem. The standard remedy is to collapse the disordered adjacent categories into a single category and re-run the analysis.
How do I interpret the Wright map?
A Wright map (person-item map) plots the estimated person ability distribution on one side of a vertical logit scale and the estimated step difficulties on the other. Steps located in a region where few persons fall are poorly targeted; they contribute little measurement information for the sample. Gaps in the step parameter coverage mean the instrument is less precise in those trait regions. Ideally, the spread of step parameters should bracket the bulk of the person distribution.
When should I prefer the Graded Response Model over PCM or GPCM?
The Graded Response Model (GRM) is another polytomous IRT model, but it uses cumulative (boundary) response functions rather than the adjacent-category formulation of PCM/GPCM. GRM is generally preferred for Likert-type rating scales where the underlying response process is conceived as crossing successive thresholds. PCM and GPCM are more natural for partial-credit scored items where earning a specific number of points reflects distinct solution steps. In practice the two families often produce similar person estimates, but the choice should be guided by the conceptual scoring model.
Sources
- Masters, G. N. (1982). A Rasch model for partial credit scoring. Psychometrika, 47(2), 149–174. DOI: 10.1007/BF02296272 ↗
- Muraki, E. (1992). A generalized partial credit model: Application of an EM algorithm. Applied Psychological Measurement, 16(2), 159–176. DOI: 10.1177/014662169201600206 ↗
How to cite this page
ScholarGate. (2026, June 1). Partial Credit Model. ScholarGate. https://scholargate.app/en/psychometrics/partial-credit-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.
- DIF AnalysisPsychometrics↔ compare
- EFAStatistics↔ compare
- GRMPsychometrics↔ compare
- Item Response TheoryPsychometrics↔ compare
- Rasch ModelPsychometrics↔ compare