Influence of Fallible Item Parameters on Test Information during Adaptive Testing

摘要

Computer simulation was used to assess the effects of item parameter estimation errors on different item selection strategies used in adaptive and conventional testing. To determine whether these effects reduced the advantages of certain optimal item selection strategies, simulations were repeated in the presence and absence of item parameter estimation errors. Results showed that item parameter estimation errors had little effect on the efficiency and measurement precision of the adaptive test item selection strategies studied. Strategies that explicitly made optimal use of item parameters for item selection were superior to a less optimal strategy, even when item parameters were fallibly estimated. It appears that errors in the item parameter estimates do not reduce the psychometric advantages of these 'optimal' strategies. Item selection strategies that explicitly employ optimization criteria should be regarded as preferable to simpler strategies that do not. Further development of psychometric procedures for the CAT system should focus on the former type of strategy. (Author)