Many-Facet Rasch Model
The Many-Facet Rasch Model (MFRM) extends Rasch measurement to performance assessments where the score depends on more than examinees and items. Developed by John Michael Linacre in his 1989 University of Chicago dissertation under Benjamin Wright and published the same year by MESA Press, the model treats raters, tasks, criteria, and any additional sources of variation as separate facets whose effects can be estimated jointly on a common logit scale.
Specification
In its standard rating-scale form, the log-odds of an examinee receiving a particular score is modelled as a function of examinee ability, item or task difficulty, rater severity, criterion difficulty, and the threshold structure of the rating scale. Each facet contributes additively in the log-odds metric, so estimates are independent of the particular sample of raters or tasks observed — the central appeal of the Rasch family for performance testing. The model is most commonly fitted with the FACETS software developed by Linacre.
Use in language testing
MFRM has become the standard analytic tool for high-stakes speaking and writing tests. Reported outputs include rater severity measures, infit and outfit mean-square statistics that flag rater misfit, fair-average scores adjusted for the severity of the raters who happened to mark a candidate, and bias analyses that test whether a rater behaves differently with particular task types or candidate subgroups. Studies of IELTS, TOEFL iBT, and Cambridge English speaking and writing modules routinely apply MFRM to monitor rater consistency and inform rater training.
Compared with generalizability theory, MFRM provides individual rater and task estimates rather than only variance components, but assumes a probabilistic Rasch model and unidimensionality. The two frameworks are often used together rather than as substitutes.
References
- Linacre, J. M. (1989). Many-Facet Rasch Measurement. MESA Press.
- McNamara, T. (1996). Measuring Second Language Performance. Longman.
- Eckes, T. (2015). Introduction to Many-Facet Rasch Measurement (2nd ed.). Peter Lang.