The Common European Framework in its political and educational context What is the Common European Framework?

No 12.
Item response theory (IRT) or ‘latent trait’ analysis: IRT offers a family of
measurement or scaling models. The most straightforward and robust one is
the Rasch model named after George Rasch, the Danish mathematician. IRT is
a development from probability theory and is used mainly to determine the
difficulty of individual test items in an item bank. If you are advanced, your
chances of answering an elementary question are very high; if you are
elementary your chances of answering an advanced item are very low. This
simple fact is developed into a scaling methodology with the Rasch model,
which can be used to calibrate items to the same scale. A development of the
Appendix A: developing proficiency descriptors
210

---

approach allows it to be used to scale descriptors of communicative
proficiency as well as test items.
In a Rasch analysis, different tests or questionnaires can be formed into an
overlapping chain through the employment of ‘anchor items’, which are
common to adjacent forms. In the diagram below, the anchor items are
shaded grey. In this way, forms can be targeted to particular groups of
learners, yet linked into a common scale. Care must, however, be taken in this
process, since the model distorts results for the high scores and low scores on
each form.
No 12.
The advantage of a Rasch analysis is that it can provide sample-free, scale-free
measurement, that is to say scaling that is independent of the samples or the
tests/questionnaires used in the analysis. Scale values are provided which
remain constant for future groups provided those future subjects can be
considered new groups within the same statistical population. Systematic
shifts in values over time (e.g. due to curriculum change or to assessor
training) can be quantified and adjusted for. Systematic variation between
types of learners or assessors can be quantified and adjusted for (Wright and
Masters 1982; Lincare 1989).
There are a number of ways in which Rasch analysis can be employed to
scale descriptors: