大学英语机考条件下随机生成试卷的概化理论研究

A Generalizability Study of Randomly Generated Computer-Based College English Test Papers

机考卷子一般通过对题库中的题目进行随机组合生成。概化理论工具软件m GENOVA实证结果显示:考生与试题的交互效应对成绩的影响远远高于考生对成绩的影响;考试的概化系数和可靠性指数远远低于理想水平;在合理范围内调整试题数量和赋分权重能使概化系数和可靠性指数得到一定提高,但未能达到理想水平。随机组卷导致了不同考生试卷的非等值性,因而降低了测量信度。

Computer-based test papers are generally generated through random combinations of questions from test question banks. The empirical results obtained by applying the generalizability theory tool mG ENOVA show that the effects of interaction between candidates and test papers is far higher than the candidates' impact on the test results; the generalizability coefficient and the reliability index of the test is much lower than the ideal level; adjusting the number of questions and the weighting of various tasks within a reasonable limits can raise the generalizability coefficient and reliability index to a certain extent but fall short of the ideal level. Randomly generated test papers lead to the non-equivalence of different examinees' test papers,thus reducing the reliability of the test.