Abstract
The article argues for a broad base for construct validation which, alongside statistics, employs other areas of mathematics, notably mathematical logic, as well as qualitative methods for building a cognitive performance model for a language test. A model is constructed and experimentally validated for gap‐filling tests, in particular the cloze test. The testee's mind is modelled as a system of switches ‐ a model that is demonstrated to be essentially non‐linear. Truth‐value tables prove indispensable for coping with the exploding number of possible states of the system. The tables allow the tester to become aware of factors potentially harmful to validity, with a view to eliminating them, as well as estimate a test's relative validity.
Notes
Address correspondence to: Krista Vogelberg, Chair of English Philology, University of Tartu, 18a Ülikooli, Tartu EE 2400, Estonia. Tel. : +3727 470405. Fax: 3727 441 252. E‐mail: [email protected]