对大规模读写结合写作任务的效度验证

Validation of a Large-Scale Reading-to-Writing Test:Evidence from Multi-Faceted Rasch Model Analysis

本文运用多层面Rasch模型调查高考英语(广东卷)读写结合写作任务成绩的差异来源,分析该任务的效度。利用FACETS对190份作文受试、评分员、任务和评分标准进行的模拟显示:该任务总体能够有效区分不同水平受试,且分绝大部分成绩差异可通过受试被考察的能力得到解释;但该任务相对偏难,个别评分员对评分标准的实际使用与模型预测值之间的拟合度低,有必要根据进一步研究改进评分标准、加强评分员培训。

The present study employed MFRM to collect validity evidence for a large-scale reading-to-writing test. Analyses of 190 subjects' writings with FACETS yielded the following findings:1) The task can distinguish among candidates of different reading-to-writing abilities,and score variance is largely attributed to the construct; 2) The task is difficult for the targeted candidates; 3) Differentiating rating behavior of individual raters necessitates rating rubrics improvement and further rater training.