실험인자의 개수, 랜덤성 및 지분성에 의한 일반화 가능도 설계의 유형화 및 응용

Abstract

Generalizability Design (GD) is a most practical tool to improve the reliability of behavioral measurement by controlling multiple sources of error variations. The paper presents the classified models of Generalizability Design (GD) regarding four aspects: the number of factors, the randomness of factors, the nestness of factors and the types of errors. In addition to classified models, the study develops mixed and combined models of (p x r:g): (i x h) which include random, crossed and nested factors. Furthermore, the research displays the implementation steps as explained below:1) Identification of data structure model included in variation source by Venn diagram. 2)Performance of approximate F-test with Satterthwaite’s degree of freedom considering Expected Mean Square (EMS). 3) Evaluation of reliability using Generalizability Index (GI) and Dependability Index (DI) by Expected Variance Component (EVC). The Generalizability Design (GD) model anticipated in this study can be applied to improve the reliability of evaluation for industrial safety training program.