項目反応理論における無答処理とニューラルネットワークによる能力母数の推定

概要

論文の詳細を見る
In the item response theory, if omitted responses (no answers) are detected in the analysis data of a test, the no answers must be handled appropriately. The omitted responses are usually processed by "ignoring them" or "handling them as wrong answers," or "handling them according to the Lord Method" if the data is composed of answers from a multiple choice test. In this study, some test data sets were simulated, and they were analyzed using the no-answer handling method proposed by Fujimori (1994). The analysis results aren't overall if the ability parameters are estimated using a neural network. Findings suggest that the method may be more appropriate for a long test, but that it isn't useful for a short test. In addition, this study also includes an analysis of no-answer processing by the method when the ability parameters are estimated by the maximum likelihood method. Results from this analysis show that the method using the maximum likelihood estimates is effective for any test length.
文教大学の論文
2000-12-20