The construct modeling approach was developed by the Berkeley Evaluation and Assessment Research (BEAR) group (Wilson & Sloane, 2000; Wilson, 2005) and is essentially a coherent and integrated assessment system (BEAR Assessment System—or “BAS”). In contrast to summative one-shot assessments, construct modeling is an approach that takes a developmental perspective on learning and describes learning as progress along dimensions of interest to the teacher or researcher. Like other approaches construct modeling starts off by hypothesizing the existence of constructs that describe latent abilities of learners such as their understandings of specific ideas or practices in a domain. Of course, in the process of building assessments for this construct, we may decide that there is a better way to conceptualize it. For example, understanding the concept of density can be seen as a construct. Depictions of how the understandings embodied in the construct develop over time are termed construct maps. Construct maps thus describe hypothetical learning paths for a particular construct.
Construct modeling is an integrated assessment approach comprised of four building blocks or steps beginning with a construct map. The map is used to guide the design of assessment items and to describe the range and sophistication of responses to the item, what is termed the outcome space for the item (akin to a coding scheme for item responses). A statistical measurement model (in this case a multidimensional IRT model) is used to relate what is observed (outcome space data) to what is measured (construct map) (Wilson, 2005).