中華心理衛生協會

This study demonstrates how to obtain an interval scale and to check whether test construct remains stable over time in longitudinal studies through analyzing a longitudinal dataset of the SCL-90-R collected by Wu (1997, 1998, 1999), where 1101 seven-graders took four subtests of the SCL-90-R in four years, once a year. Items administered were not identical over the four time points. To obtain an interval scale for each subtest and to put scores of the four time points onto the same scale so that changes in psychological disorders could be measured, the Rasch model (Rasch, 1960) as well as its extensions was employed. Generally, the four subtests fitted the model's expectation fairly well so that interval scales could be obtained. However, the items were far too difficult for these subjects, which reduced test reliability. Five-point scales could be collapsed to two-point scales without scarifying their reliabilities substantially. The test construct remained unchanged over time so that change measurement was applicable. On Somatization and Anxiety, the mean magnitudes were reduced from seven- to eight-grade, but were increased from eight- to nine-grade; On Depression, thy remained unchanged from seven- to eight-grade but were increased from eight- to nine-grade; On Hostility, thy were increased from seven- to eight-grade and then remained unchanged from eight- to nine-grade. After graduation from junior high schools (the fourth time point), the mean magnitudes on all the four subtests were decreased significantly, especially on Hostility. The correlations between the four time points for the four subtests were all positively moderate, ranging from .42 to .76.