The new vision of mathematics requires that assessment reinforce a new conceptualization that is both broader and more integrated.
Tests have traditionally been built from test blueprints, which have often been two dimensional arrays with topics to be covered along one axis and types of skills (or processes) on the other.1 The assessment is then created by developing questions that fit into one cell or another of this matrix. But important mathematics is not always amenable to this cell-by-cell analysis.2 Assessments need to involve more than one mathematical topic if students are to make appropriate connections among the mathematical ideas they have learned. Moreover, challenging assessments are usually open to a variety of approaches, typically using varied and multiple processes. Indeed, they can and often should be designed so that students are rewarded for finding alternative solutions. Designing tasks to fit a single topic and process distorts the kinds of assessments students should be able to do.
Assessment developers need characterizations of the important mathematical knowledge to be assessed that reflect both the necessary coverage of content and the interconnectedness of topics and process. Interesting assessment tasks that do not elicit important mathematical thinking and problem solving are of no use. To avoid this, preliminary efforts have been made on several fronts to seek new ways to characterize the learning domain and the corresponding assessment. For example, lattice structures have recently been proposed as an improvement over matrix classifications of tasks.3 Such structures provide a different and perhaps more interconnected view of mathematical understanding that should be reflected in assessment.
The approach taken by the National Assessment of Educational Progress (NAEP) to develop its assessments is an example of the effort to move beyond topic-by-process formats. Since its inception, NAEP has used a matrix design for developing its mathematics assessments. The dimensions of these designs have varied over the years, with a 35-cell design used in 1986 and the design below for the 1990 and 1992 assessments. Although classical test theory strongly encouraged the use of matrices to structure and provide balance to examinations, the designs also were often the root cause of the decontextualizing of assessments. If 35 percent of the items on the assessment were to be from the area of measurement and 40 percent of those were to assess students' procedural