论文部分内容阅读
没有信度的测试不可能有效度。在语言运用测试中,考试分数因受多方面因素影响而无法只用简单的一维相关衡量其信度。本文基于某考点的一次大学英语四、六级口语考试(CET-SET)的考生成绩,用多层面Rasch模型对这次考试的信度进行了研究。研究结果发现,考官的严厉度、任务难度、评分标准和量表等因素都可能产生一定的测量误差,从而导致考生的成绩差异;多层面Rasch模型作为经典Rasch模型的延伸,能够综合分析这些误差来源及误差大小,根据各层面的统计结果对考生的原始成绩进行相应的调整和补偿,并为确保考试信度提供有效的反馈信息。
Tests without credibility can not be valid. In language proficiency testing, test scores can not be measured by simple one-dimensional correlations due to many factors. Based on the results of CET-SET examinations of a test center, this paper studies the reliability of the exam using the multi-level Rasch model. The results showed that the severity of the examiner, the difficulty of the task, the score standard and the scale and other factors may produce some measurement errors, resulting in differences in the performance of candidates; multi-level Rasch model as an extension of the classic Rasch model can be comprehensive analysis of these errors Source and error size, based on the statistical results at all levels of the original results of the candidates to make the appropriate adjustments and compensation, and to ensure the reliability of the test to provide effective feedback.