Tests of oral performance: the need for data-based criteria

It has become almost axiomatic for ‘communicative’ testing theory that the tests should contain exercises which are based on ’real-life’ communicative situations (Morrow 1979). In essence, this makes the communicative testing enterprise an exercise in content validity. The notion of ‘sampling real life’ has, however, been extensively criticized (Oiler 1979:184; Alderson 1981:57), mainly because testing has been seen as the reliable prediction of success in some behavioural performance in a non-test situation. Communicative testing theory tries to tap the performance directly, and here lies the problem. This article investigates one aspect of this problem with regard to the assessment of the English Language Testing Service,1 and attempts to uncover what appears to be a problem in the theory on which such tests are constructed—or, to put it another way, a problem in construct validity.