Modeling differential pacing trajectories in high stakes computer adaptive testing using hierarchical linear modeling and structural equation modeling

UNCG Author/Contributor (non-UNCG co-authors, if there are any, appear on document)
Marie Huffmaster Thomas (Creator)
The University of North Carolina at Greensboro (UNCG )
Web Site:
Richard Luecht

Abstract: "This study compares two statistical methods for modeling changes in response latency (timing) patterns on a high-stakes adaptive test: (1) hierarchical linear modeling (HLM2) and (2) growth modeling using structural equation modeling (SEM). The testing context involves the NCLEX-RN®, a variable-length, computerized adaptive test used to license registered nurses in the United States and its territories. Item-level response-time data from 4,415 first-time takers of the NCLEX-RN® examination are used to create and evaluate the different "examinee pacing trajectory" models. The examinee pacing trajectory models are separately fit to examinees at three proficiency levels: (1) clear failers who take an abbreviated test of only 75 items; (2) indeterminate examinees who take a variable-length test of up to 265 items; and (3) clear passers who also take an abbreviated test of only 75 items. The HLM- and SEM-based approaches provided comparable but not identical results. The estimated intercept terms for the pacing trajectory models were different for each of the three proficiency groups, indicating a possible association between ability and pacing skill. However, the intercepts did not differ dramatically across the two modeling methods because those estimates are essentially based on empirical means. The pacing trajectory slopes varied both in scale and relative magnitude across proficiency groups and by analysis method. The HLM slope estimates reflect the average scale variances of the empirical response data across blocks of items. In contrast, SEM-based growth modeling employs arbitrary constraints imposed to statistically identify the model. These scaling differences made it difficult to directly compare the HLM-based and SEM-based slopes. Ultimately, however, it was concluded that either method (HLM or SEM) is able to model pacing trajectories in a meaningful way."--Abstract from author supplied metadata.

Additional Information

Language: English
Date: 2006
latency (timing) patterns, adaptive test, hierarchical linear modeling (HLM2), growth modeling, structural equation modeling (SEM), NCLEX-RN®, computerized, license, registered nurses, United States
Educational tests and measurements
Computer adaptive testing
Structural equation modeling
National Council Licensure Examination for Registered Nurses

Email this document to