| Several methods have been suggested for the automated assembly of multidimensional tests. This study presents a new procedure that uses a combinatorial optimization method called quadratic knapsack programming (QKP). This new approach uses numerical integration instead of linearization methods for formulating the objective function. Additionally, a model-specific algorithm called the QKP heuristic was formulated to facilitate the selection of test items in the assembly process.; The effectiveness of the QKP procedure was illustrated using computational experiments. These experiments used simulated item pools which consisted of multidimensional item parameters and test specifications. Additionally, QKP was compared to 0-1 linear programming (LP) and random item selection (RND) procedures. Results showed that in a multidimensional context, the QKP approach assembled tests that are relatively better at maximizing information. Furthermore, QKP was more computationally efficient than the other methods when assembling longer tests from item pools with low multidimensional discrimination. Also, a unidimensional comparison of the relative efficiency of assembled tests showed that QKP-assembled tests have higher measurement precision than tests assembled with the other two methods. |
