The progressive tradition has long advocated teaching and testing through "realistic" projects instead of through separate subject matters, and has long rejected tests that probe isolated knowledge and skills.
Realistic performance assessments, it is claimed, have a number of advantages over multiple-choice tests, which include being more informative, more motivational, and fairer to minorities and nonverbal students. These claims are often plausible, particularly when performance tests are used as teaching and monitoring devices in the classroom context; for instance, in a course on writing, it is clearly preferable to use writing tasks as tests rather than to use multiple-choice tests.
However, performance tests are only one of many monitoring devices in classroom teaching, and they have been shown to be ineradicably subjective and arbitrary in grad¬ing. They are not appropriate for large-scale, high-stakes testing because no one has been able, even in theory, to make such tests fair and accurate at reasonable cost in money and time.
To serve democratic ends, American educators have pioneered the creation of fair and accurate multiple-choice tests that probe a wide variety of knowledge and skills. The consensus among psychometricians is that these objective tests, rather than performance tests, are the fairest and most accurate achievement tests available. Performance tests, while important as one tool for classroom use, should not play a decisive role in high-stakes testing, where fairness and accuracy are of paramount importance.
and Why We Don't Have Them.