Assembly model. The assembly model describes what the test
as a whole will look like. Assembly models are expanded versions
of detailed test specifications. An assembly model contains the
information necessary to build parallel forms of the test. (The
algorithms used to generate computer assemblies of tests are part
of the assembly model for such tests.) Desired statistical attributes
of the test as a whole should be indicated in the assembly model,
including such data as the target mean difficulty and mean
discrimination, the distribution of task difficulty, the desired
reliability of the reported score(s), and the intended level of
speededness. Ideally, the assembly models should be specific
enough that the test forms generated by the same model are
interchangeable.
To make the linkages among claims, evidence, and tasks explicit,
the assembly models indicate the KSA(s) and claim(s) for which each
task provides evidence. Different assembly models can be used with
a single pool of tasks to produce a “family” of tests. For example,
different assembly models could be used to produce diagnostic tests
that serve as study guides and summative tests for decision-making
purposes about test takers. (Note that different evidence models
would likely be needed as well.)