从多元概化理论看高考综合能力测试的改进

通过多元概化理论的研究发现，高考综合能力测试（2001，广东）的总体信度达到了可以接受的水平（0.784）。但测验中各部分对总方差的贡献程度与预定的赋分比例有较大差距。其中，地理和政治的贡献度偏低，化学和历史的贡献度偏高。这表明有（历史和化学）偏科特点的考生得到了较高的综合分数。又经决策(D)研究发现，增加地理部分的题量会反常地降低测验的总体信度，这说明有不少高分考生答错或主动放弃了地理科题目。因此，如何有效控制各部分的实际贡献程度、避免负面导向是当前高考综合能力测试亟待解决的问题。

MULTIVARIATE GENERALIZABILITY ANALYSIS OF THE CHINESE COLLEGE ENTRANCE COMPREHENSIVE EXAMINATION

The Chinese College Entrance Comprehensive Examination (CCECE), which consists of Geography, History, Political Studies, Biology, Chemistry and Physics, is a comprehensive achievement test to measure an examinee's academic readiness for tertiary education. This study used multivariate generalizability theory to estimate generalizability (reliability) coefficients of the CCECE. The G-study and D-study results suggested that the generalizability coefficient of the whole test was 0.78, which was considered moderate but acceptable. Doubling the items for all the subtests would increase generalizability coefficient to 0.88. Except for geography and political studies, reliability estimates for the subtests were acceptable. However, a composite universe score approach was not ideal both because variance distributions among subtests were inconsistent with the planned score weighting and because correlations among some subtests were low. For example, chemistry and history accounted for more variance than other subtests combined. Among the problematic subtests, geography contributed negatively to reliability of the whole test.