Constrained multidimensional adaptive testing without intermixing items from different dimensions
Multidimensional adaptive testing (MAT) can improve the efficiency of measuring traits that are known to be highly correlated. Content balancing techniques can ensure that tests fulfill requirements with respect to content areas, such as the number of items from various dimensions (target rates). However, content balancing does not restrict the order in which items are selected from dimensions. If multiple dimensions are measured with MAT, intermixing items from different dimensions might invalidate properties of those items, which are known from calibration studies without mixed item content. To avoid this, the known correlations between traits can be used to increase efficiency of the ability estimation only without intermixing items from different dimensions. In this simulation study, MAT allowing items to be intermixed between dimensions is compared to Constrained MAT (CMAT) that does not allow intermixing items between dimensions for items with between-item multidimensionality. As expected, MAT achieved the greatest reliability for equal target rates; however, CMAT with items administered in a pre-specified order dimension by dimension was not disadvantageous for unequal target rates.
Kroehne, Ulf; Goldhammer, Frank; Partchev, Ivailo: Constrained multidimensional adaptive testing without intermixing items from different dimensions, in: Psychological Test and Assessment Modeling, 56 (2014) 4, 348-367