A many-facet Rasch analysis of the second language group oral discussion task William J. Bonk University of Colorado at Boulder and Gary J. Ockey University of California, Los Angeles FACETS many-facet Rasch analysis software (Linacre, 1998a) was utilized to look at two consecutive administrations of a large-scale (more than 1000 examinees) second language oral assessment in the form of a peer group dis- cussion task with Japanese English-major university students. Facets modeled in the analysis were examinee, prompt, rater, and ve rating category ‘items.’ Unidi- mensionality was shown to be strong in both datasets, and approaches to inter- preting t values for the facets modeled in the analysis were discussed. Examinee ability was the most substantial facet, followed by rater severity, and item. The prompt facet was negligible in magnitude. Rater differences in terms of severity were generally large, but this characteristic was not stable over time for individ- uals; returning raters tended to move toward greater severity and consistency, while new raters showed much more inconsistency. Analysis of the scales showed general validity in gradations of scale steps, though raters had some difculty discerning between categories at the ends of the scales for pronunciation and com- municative skills. I Introduction Performance-based second-language (L2) oral tests are increasingly prevalent, in assessments of all shapes and sizes, from classroom- based assessment to standardized prociency tests and everything in between. This increase in the use of oral tasks for assessment is likely a product of the increased interpretability of test scores, potential val- idity of the scores when linked to real-world criteria, and positive washback effects of such assessment tools. However, with the excep- tion of the oral interview and its variants (see, for example, Shohamy, 1983; Nevo and Shohamy, 1984; Bachman and Savignon, 1986; Sho- hamy et al., 1986; van Lier, 1989; Ross, 1992; Stanseld and Kenyon, 1992; Shohamy, 1994; Lazaraton, 1996; McNamara and Lumley, Address for correspondence: Gary Ockey, Department of Applied Linguistics and TESL, Uni- versity of California, Los Angeles, 3300 Rolfe Hall, PO Box 951531, 405 Hilgard Avenue, Los Angeles, CA 90095–1531; email: gockey@ucla.edu Language Testing 2003 20 (1) 89–110 10.1191/0265532203lt245oa Ó 2003 Arnold