Using Multidimensional Analysis to Investigate the Extrapolation Inference of a High-Stakes Test

The degree of correspondence of situational characteristics between test tasks and target domains is believed to affect how similar language use on tests is to the target domains (i.e., extrapolation). Multidimensional analysis (MDA) has gained in popularity in extrapolation research because it not only interprets language use functionally associated with situational characteristics but also provides a quantitative method to investigate co-occurring linguistic features. Using the automatic tool Multidimensional Analysis Tagger (MAT), the present paper applies MDA to explore the linguistic features between writing of the National Entrance Test of English for MA/MS (NETEM) and disciplinary writing on four dimensions in Biber's (1988) seminal work. Overall, NETEM writing differed significantly from all four disciplinary domains on Dimension 1 and Dimension 4, and showed similarities with Humanities and Arts papers but differed from the other three domains on Dimension 2 and Dimension 3. NETEM writing was weak in its extrapolation to real academic contexts.