| Summary: | 碩士 === 國立臺中教育大學 === 教育測驗統計研究所 === 101 === Population statistics is one of the main focuses of large-scale assessments such as means and standard deviations. Since the numbers of examinees and items are enormous, large-scale assessments use equating methods to make sure that every examinee takes different booklets are on the same scale. Therefore, examinees score can be compared even if they didn’t take the same test. The scores in different year can also be compared by linking procedure. The purpose of this study is to explore the performance of different manipulate factors based on linking methods used by large-scale assessments. The results show that concurrent method outperforms when recovers the population parameters under the assumption of no difference between two assessment years. TIMSS linking method is better than the others, when there are differents in population means, and concurrent method will reduce its accuracy. When the number of examinees is increasing, the bias will decresing, but difference between 5600 people and 8400 people is not obvious. Results also show that having longer test length can raise more accuracy, but with same test length, 2 dimensions and 4 dimensions have no obvious difference.
|
|---|
