Construct a Vertical Scale for a Science Assessment

• Main Authors: I-ting Kuo, 郭宜婷
• Other Authors: Bor-Yaun Twu
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/35547942128061363303

碩士 === 國立臺南大學 === 測驗統計研究所碩士班 === 98 === Scaling allows one to compare scores from different test forms, and vertical scaling is intended to support the comparison of scores obtained at each of a number of test forms of systematically different difficulty. Vertical scaling refers to the process of linking different levels of an assessment, which measure the same construct, onto a common score scale. The primary reason for creating vertical scales is to measure learning across time. Many assessments, such as ITBS, CAT, SAT, and so on, have created vertical scales in different ways, yet all of those scales appear to be functioning adequately for some of the same purposes. The data for this study were extracted from responses to the 2006 science (SCI) assessments in grade 3 through 6. The assessments were constructed with common item data collection design between adjacent grades to support the establishment of a vertical scale, and then use IRT separate estimation to independently estimate the item parameters for each level being on a separate scale. The results showed that: (1) the grade-to-grade differences in averages decrease as grade increases, (2) the standard deviations increase over grade, suggesting that the variability of scale scores increases over grade for the IRT scaling, and (3) these effect sizes decline as grade level increases, although there is a reversal when comparing the grade 4 to grade 5 growth.