順序カテゴリデータへの確認的因子分析に基づく 信頼性係数の評価

Based on recent discussions of the use of reliability coefficients, many psychometricians have recommended using model-based reliabilities. Green & Yang (2009) proposed that nonlinear SEM coefficients be used as model-based reliability for scales with ordered category data. However, very few published studies have evaluated nonlinear SEM coefficients. In order to use SEM coefficients in applied research, how they perform when models are misspecified should be investigated. The present study used a Monte Carlo method to evaluate nonlinear SEM coefficients in conditions of model misspecification. The results indicated that, in most of the conditions of the simulation, nonlinear SEM coefficients performed very well when the models were correctly specified, whereas the coefficients were severely biased when the models were misspecified. Biases in the coefficients were parallel to the extent of misspecification of the models. Based on the results of this simulation, the discussion proposes future directions for the use of nonlinear SEM coefficients in applied research.