CAN VALIDITY RISE WHEN RELIABILITY DECLINES

Advocates of performance tests of achievement sometimes contend that t he validity of such tests typically is higher than that of multiple-ch oice tests, despite the lower reliability of the performance measures. This contention appears to contradict a deduction from classical test theory and to deny the oft-quoted principle that the reliability of a measure places an upper limit on its validity. Is this claim, therefo re, a theoretical impossibility? In this article it is demonstrated th at this ''anomaly'' can occur, even when the evidence of validity take s the form of a correlation with an acceptable criterion. Whether clea r-cut empirical examples actually can be found is an open question.