Analysis of standardized achievement tests by classical test theory and genetic factor analysis models

The purpose of this study was to analyze standardized achievement test scores of twin and non-twin pupils. These scores were analyzed by genetic factor analysis in consideration of classical test theory. Samples were 953 pupils ; 100 monozygotic (MZ) twin pairs, 25 dizygotic (DZ) twin pairs, and 703 non-twins. On the basis of results by Toyoda & Muraishi (1998), non-twins' scores were used for stabilizing covariance structure. Four subjects (Japanese, social studies, mathematics, and science) were analyzed through the model. Evidence of (i) genetic influences, (ii) common environment, and (iii) non-shared environment showed that large differences existed in the structure of achievements on four subjects. The ratios of variances for achievement of Japanese were as follows : (i) 0.0%, (ii)64.5%, (iii) 2.9% ; those of social studies were (i) 52.3%, (ii) 17.0%, (iii) 4.7% ; those of mathematics were (i) 0.0%, (ii) 47.7%, (iii) 10.4%, while those of science were (i) 56.1%, (ii) 0.0%, (iii) 13.3%, respectively. Comparing fitness among various models of classical test theory showed the best fit for congeneric measurement. It warns us against the use of the α-coeffecient that assumes τ-equivalent measurement in calculating coefficient of reliability.