Controlling Test Overlap Rate in Automated Assembly of Multiple Equivalent Test Forms

Chuan-Ju Lin


Assembling equivalent test forms with minimal test overlap across forms is important in ensuring test security. Chen and Lei (2009) suggested a exposure control technique to control test overlap-ordered item pooling on the fly based on the essence that test overlap rate – ordered item pooling for the first t examinees is a function of test overlap rate – ordered item pooling for the previous (t-1) examinees. The exposure control procedure to control test overlap-ordered item pooling on the fly appears to meet the needs of controlling test overlap rate for tests assembled sequentially. To develop a better understanding of how well the ordered-item-pooling control method function in automated assembly of multiple forms with the WDM heuristic, this study evaluated its performance under different conditions of test length and test-content outline by comparing the outcomes to those from the corresponding baseline automated-test-assembly (ATA) conditions, where test overlap controls were not considered. The evaluation criteria included (i) the conformity to the test-assembly constraints, (ii) test parallelism in terms of the resultant psychometric properties, (iii) average test overlap rate, and (iv) distribution of item exposure rate. The results showed that the ordered-item-pooling control procedure demonstrated its effectiveness in most experimental conditions by achieving an acceptable average test overlap rate across multiple forms without compromising the conformity to the test-assembly constraints and the test equity of the assembled forms. Moreover, test security might be ensured in less supportive contexts for ATA by imposing item exposure control together with test overlap control that would be less likely to compromise test quality. More research is needed to verify this anticipation.


automated test assembly; test overlap control; ordered item pooling; test security; expected baseline test-overlap rate

Full Text: