ERIC Number: ED414297
Record Type: RIE
Publication Date: 1997-Jul
Realistic Simulation of Item Response Data. ACT Research Report Series 97-4.
Davey, Tim; Nering, Michael L.; Thompson, Tony
Potential solutions to applied measurement problems are commonly evaluated using simulated data, largely because it is often impossible, difficult, or prohibitively expensive to supply the needed volumes of real data. However, answers proved by simulation studied generalize real-world problems only to the extent that the simulated data resembles the real thing. Research has suggested that in the case of generating item responses using traditional unidimensional, logistic response models, this resemblance is less than complete. While gross features of the data such as item passing rates, score distributions, and test reliabilities are modeled adequately, subtler characteristics of real data are often absent in simulation. Accordingly, high-dimensional simulation procedures have been developed that preserve the more subtle characteristics. The simulation process begins by fitting a multidimensional latent trait to a large sample of actual data. Estimated multidimensional item response functions are used to generate data by procedures directly analogous to those used with unidimensional simulations. The procedure is illustrated with an English usage and a mathematics test as simulation templates. (Contains 1 table, 12 figures, and 18 references.) (Author/SLD)
ACT Research Report Series, P.O. Box 168, Iowa City, IA 52243-0168.
Publication Type: Reports - Evaluative
Education Level: N/A
Authoring Institution: American Coll. Testing Program, Iowa City, IA.