ABSTRACT
The focus of this chapter picks up on a perspective voiced by Davidson (2000) when he warned against statistical determinism when validating language assessment measures:
Any new development in language test statistics should not divert language testers away from the far more difficult and fundamental challenges of making valid assessments. Statistical determinism is a deceptively accessible Weltanschaung (world view); it is far too easy to allow items to survive trialing or editing based solely on their “p” or “d” value, just as it is far too easy to judge the work by an oral interview rater based solely on inter-rater reliability correlation coefficients. Statistical evidence of educational quality should be fused with and weighed against evidence that derives from other sources.
