Summary: | 碩士 === 國立臺南大學 === 教育學系測驗統計碩士班 === 101 === Many countries and international organizations are dedicated in developing large-scale assessments in order to provide useful information for educational policies. These large-scale assessments, such as PISA, TIMSS, NAEP and TASA, are focus on comparing the achievement across different groups or tracking the growth trends across different grades or ages. Therefore, large-scale assessments need linking technique to deal with these issues.
In practice, common-item nonequivalent groups design is widely used to collect data in test linking. Under this design, the selection of the common items is an important task. In equating, a popular belief is that common items should be a mini version of the tests being equated. Since the test settings for vertical scaling are different from those for equating, the requirement of a mini version of a test may not be met in practice. Although many studies
have discussed the issue of the content and statistical representation requirements for common items in equating, few of them have discussed the issue under the test settings for vertical scaling.
This study used the three-parameter logistic model to simulate item response data and investigated the performance of vertical scaling under different common-item data sets. The results indicated that the related statistical characteristics of common items, such as length, discrimination and difficulty, are critical factors in vertical scaling. Different combination of
these factors would produce different linking results.
|