Abstract: Practice in psychological and educational measurement often needs to use the technique of test linking to analyze the difference or trend of the examinees’ traits. When implementing test linking, we need to choose a data collection design. Among several designs, the Non-Equivalent Groups with Anchor Test (NEAT) design is the most frequently used one. With this design, the anchor items, the set of items shared by all the examinee groups, are unique carrier to achieve test linking. Given the importance of the anchor items, except for their parameters, what we must pay attention to is the relationship pattern of the anchor items and their related test forms/levels. This paper was aimed to explore the relationship pattern based upon a concept clarification and a literature review about the representativeness of the anchor items. First, it indicated that the meaning of the representativeness of the anchor items is different between the equating research field and the vertical scaling field, the two most important sub-areas of linking. In equating, it involves the statement that the anchor test be a “mini version” of the tests to be equated. In present, it is widely believed that the anchor used in equating should be constructed according to the same specification with the total test, including the content and statistic characteristics, to accurately reflect the group difference. In vertical scaling, with the grade-to-grade definition of growth which corresponds to the NEAT design, achievement growth is defined on the content in the test level(s) appropriate for the specific grade level(s). According to this growth definition, the anchor which serves as the way to quantify the growth should also come from the test level(s) appropriate for the specific grade level(s). Therefore, the representativeness of the anchor items in vertical scaling can be defined as the representativeness of the anchor items to test level(s) they are chosen from. It is worthy to note that the definitions of the representativeness of the anchor items do not necessarily mean that the anchor items should be a “mini version” of the test levels they are chosen from. That is why the representativeness of the anchor items is called a topic here rather than a property or nature in test linking. Based on the above statement, this paper examined the existing research on the representativeness of the anchor items in the fields of both equating and vertical scaling, summarized the results of the studies, and systematically analyzed the relationship pattern between the anchor items and the related test forms/levels in terms of the content, item format, and statistic characteristic. It revealed the following general results. (1) The content representativeness of the anchor items might be necessary in test linking, though some lack of representativeness can also get fine results. (2) When linking the test forms/levels composed of dichotomous items and polytomous items, the representativeness of the anchor items in item format is necessary. However, when linking the test forms/levels composed of discrete items and passage based items, the research in equating area indicated that only employing the discrete items might lead to better results. (3) With respect to the difficulty level of the anchor items, the researches in equating support the representativeness of the anchor items. (4) With respect to the difficulty range of the anchor items, the research in equating indicated that anchor with a difficulty range less than that of the total test might lead to better results than the mini anchor; however, the research in vertical scaling conformed the traditional view of the representativeness of the anchor items. Finally, this paper generalized the probably optimal proposal with respect to the current practice of construction of the anchor items and analyzed the future directions of the research on the representativeness of the anchor items.

Key words: test linking, equating, vertical scaling, Non-Equivalent Groups with Anchor Test design, the representativeness of the anchor items

摘要： 本文旨在以“锚题代表性”这一研究命题切入，探索在非等组锚测验设计下，作为实现测验链接的重要载体，锚题和相关的测验试卷/水平之间究竟应该有什么关系。本文首先指出锚题代表性这一概念在等值和垂直量尺化领域中具有不同的含义，并给出其在垂直量尺化中的含义。通过考察测验链接中有关锚题代表性的既有研究，系统总结相关研究成果，本文概括出了当前锚题构建实践的可能优化方案，分析了锚题代表性研究的未来方向。

关键词: 测验链接 等值 垂直量尺化 非等组锚测验设计 锚题代表性