09.12.2012 Views

2003 IMTA Proceedings - International Military Testing Association

2003 IMTA Proceedings - International Military Testing Association

2003 IMTA Proceedings - International Military Testing Association

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

328<br />

This finding is important from a substantive perspective, because fewer unidimensional pairings<br />

means a more fake-resistant test. As a final note, the correlations between the estimated and<br />

known thetas, ranged from .77 in the most unfavorable situation (a 20-item test with 10%<br />

unidimensional pairings) to .96 in the most favorable (80-item test with 40% unidimensional).<br />

The average correlation was about 0.9 for the 40-item tests, regardless of the percentage of<br />

unidimensional pairings.<br />

Discussion and Conclusions<br />

This paper outlines a method of constructing and scoring fake-resistant multidimensional<br />

pairwise preference items. Individual statements are administered and calibrated using a<br />

unidimensional single stimulus model. Social desirability ratings are obtained for statements<br />

using, say, a panel of judges, and fake-resistant items are created by pairing similarly desirable<br />

statements representing different dimensions. Tests are created by combining multidimensional<br />

items with a small number of unidimensional pairings needed to identify the latent metric. Trait<br />

scores are then obtained using a multidimensional Bayes modal estimation procedure based on<br />

the MUPP model developed by Stark (2002).<br />

As shown here, the MUPP approach to test construction and scoring provides accurate<br />

parameter recovery in both one- and two-dimensional cases, even with relatively few (say 15%)<br />

unidimensional pairings. Accuracy of this approach generally improves as a function of test<br />

length. Even with nonadaptive tests, good estimates may be attained using only 20 to 30 items<br />

per dimension, meaning that a 5-D test would require 100 to 150 items. If adaptive item<br />

selection were used to improve efficiency, the required number of items might decrease by as<br />

much as 40%. We are currently developing and validating a 5-D inventory, using this approach,<br />

and comparing the scores to those obtained using traditional methods.<br />

Acknowledgements<br />

We wish to thank the U.S. Army Research Institute for access to the AIM data and for supporting<br />

this research. Assistance from the Human Resources Research Organization (HumRRO) was<br />

particularly helpful with data management and recordkeeping. The Consortium of Universities<br />

of the Washington Metropolitan Area was also helpful in securing research funds. All statements<br />

expressed in this document are those of the authors and do not necessarily reflect the official<br />

opinions or policies of the U.S. Army Research Institute, the U.S. Army, the Department of<br />

Defense, HumRRO, or the Consortium of Universities.<br />

References<br />

Press, W.H., Flannery, B.P., Teukolsky, S.A., & Vetterling, W.T. (1990). Numerical<br />

recipes: The art of scientific computing. New York: Cambridge University Press.<br />

45 th Annual Conference of the <strong>International</strong> <strong>Military</strong> <strong>Testing</strong> <strong>Association</strong><br />

Pensacola, Florida, 3-6 November <strong>2003</strong>

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!