• 作者： 林妙香
• 作者服務機構： 中央研究院統計科學研究所
• 中文摘要： The junior-high school graduates in Taiwan are being selected for study in senior-high schools through two admissions channels, namely, the Joint Entrance Examination, and the Recommendation and Screen program. Both admissions processes make extensive use of admission-test data under a clear enrollment limitation. Under such a fixed- quota model, test designs and statistical models should include explicit attention to the selection ratio for achieving equity and fairness in ranking examinees. However, the current practice is to administer a too-easy admission-test battery, resulting in a ceiling effect. This study investigates the extent to which the resulting ceiling effect jeopardizes the pattern of covariance structure of the composite score via empirically analyzing the admission-test data of the 1998th annual Joint Entrance Examination. The results of study are as follows. (1) The score distributions of the admission-test battery were all negatively skewed, reflecting test items failed to discriminate among examinees falling in the upper 42% score range. (2) The resulting ceiling effect of English test was most conspicuous as compared to those of the other four tests. Thus, English test made the least contribution to the total variance of the composite score on which ranking examinees is based. (3) The low magnitude of test reliability revealed that the admission-test battery was appropriate only for measuring examinees falling in lower range of scores. Suggestions made by this study are as follows. (1) The admission-test battery for the Joint Entrance Examination should be designed to have maximum discrimination throughout the upper 42% range of scores. A platykurtic test -score distribution produced by a test with items of good discriminating power satisfies this aim. (2) Multiple sets of admissiontest battery varying in difficulty levels should be carried out under the Recommendation and Screen program, corresponding to different ability level of applicants to be screened by senior high schools. (3) The discrepancy in difficulty among test measures should be small, so that test measures have comparable weights with respect to the contribution to the variance of the composite score.
• 英文摘要： 因應多元化入學政策，北區公、私立高中主要入學方案，除現行聯招制度外，並擴大推薦甄選方案。推甄方案立意於理念層面公平性—亦即適才適所入學制度，而聯招制度則被認為具有技術層面公平性。此二入學制度皆具會考形式並以學科測驗成績作擇優錄取。因此測驗試卷命題導向是否吻合考生素質及固定名額入學模式，皆影響理念及技術二層面的公平性。本文闡述聯招及推甄二入學測驗編製的問題所在，並以八十六學年度北區高中聯招資料揭示五試卷簡易化程度及其負面效應。本文研究顯示：(1)高中年聯招五學科試卷相對於前42%錄取生群皆屬偏易（平均答對率介於73%至90%）且成績分布皆呈負偏態，偏離理想的平闊圖形；(2)天花板效應最為顯著為英文選擇、英文非選擇及數學選擇等三分測驗；(3)英文試卷對分發總分變異量的貢獻比率（9%）遠低於其他學科試卷，因而降低其對分發學校影響力。(4)五學科試卷難度水平最「適配」測試群為落榜低成就群，其次為一般高中及八所熱門高中錄取生群導致平均鑑別度降低1至2倍，因而信度亦降至1至3倍。本文建議如下：(1)高中聯招之技術層面公平性須仰賴入學測驗的二種準確度，且須將試題難易度考量對象設為錄取生群，而不是以總考生群為對象；(2) 推薦甄選之理論層面公平性須仰賴多軌因才命題及多軌並行小會考。(3)分發總分為學科試卷分數之線性組合模式，測驗編製須顧及學科試卷相對性難度的差距，以平衡學科對分發學校的決定影響力。
• 中文關鍵字： admission tests, covariance analysis, reliability, discrimination, item statistics, linear combination, ceiling effects, test score distributions
• 英文關鍵字： 入學測驗，推薦甄選，試題統計量，測驗成績分布，鑑別度，信度，線性組合模式，天花板效 應，共變數分析