Educational assessment occurs in two major contexts. The first is the classroom. Here assessment is used by teachers and students mainly to assist learning, but also to gauge students’ summative achievement over the longer term. Second is large-scale assessment, used by policy makers and educational leaders to evaluate programs and/or obtain information about whether individual students have met learning goals.
The sharp contrast that typically exists between classroom and largescale assessment practices arises because assessment designers have not been able to fulfill the purposes of different assessment users with the same data and analyses. To guide instruction and monitor its effects, teachers need information that is intimately connected with the work their students are doing, and they interpret this evidence in light of everything else they know about their students and the conditions of instruction. Part of the power of classroom assessment resides in these connections. Yet precisely because they are individualized and highly contextualized, neither the rationale nor the results of typical classroom assessments are easily communicated beyond the classroom. Large-scale, standardized tests do communicate efficiently across time and place, but by so constraining the content and timeliness of the message that they often have little utility in the classroom. This contrast illustrates the more general point that one size of assessment does not fit all. The purpose of an assessment determines priorities, and the context of use imposes constraints on the design, thereby affecting the kinds of information a particular assessment can provide about student achievement.
To say that an assessment is a good assessment or that a task is a good task is like saying that a medical test is a good test; each can provide useful information only under certain circumstances. An MRI of a knee, for example, has unquestioned value for diagnosing cartilage damage, but is not helpful for diagnosing the overall quality of a person’s health. It is natural for people to understand medical tests in this way, but not educational tests. The same argument applies nonetheless, but in ways that are less familiar and perhaps more subtle.
In their classic text Psychological Tests and Personnel Decisions, Cronbach and Gleser (1965) devote an entire chapter to the trade-off between fidelity and bandwidth when testing for employment selection. A high-fidelity, narrow-bandwidth test provides accurate information about a small number of focused questions, whereas a low-fidelity, broad-bandwidth test provides noisier information for a larger number of less-focused questions. For a