科學能力的建構反應評量之發展與信效度分析：以自然科光學為例   全文下載

Developing and Validating a Constructed-Response Assessment of Scientific Abilities: A Case of the Optics Unit

由於建構反應試題較選擇題更適於評估學生的高階認知能力，本研究目的係在發展科學能力的建構反應評量，建立評分規準，並分析信度與效度。全評量包含「科學知識的記憶與瞭解」、「科學程序的應用與分析」、「科學邏輯的論證與表達」，以及「問題解決的評估與創造」四個分評量，共計32題。分析結果顯示，評分者內之Cronbach’s α 與評分者間之Kendall ω 和諧係數值均大於 .90，表示評分者內與評分者間的一致性良好。再者，評分者嚴苛度之多面向Rasch測量模式之卡方考驗未達顯著水準，表示評分者間的嚴苛度未有差異存在，infit與outfitMNSQ 介於1 ± 0.5，顯示無論嚴格或寬鬆的評分者，均能有效區分高、低能力的學生。另RSM 與PCM 模式比較的卡方考驗達顯著水準，將所估計的Deviance進行BIC 轉換，結果發現RSM較為適配，顯示評分者間有相同的評分閾值。此外，全評量之Cronbach’s α在 .85 以上，顯示具有不錯的信度。驗證性因素分析結果顯示，「科學知識的記憶與瞭解」、「科學程序的應用與分析」、「科學邏輯的論證與表達」，以及「問題解決的評估與創造」所檢測四個一階潛在因素，可被二階因素之「科學能力」解釋的變異量分別為 .92、 .56、.46、.46，實徵資料尚且支持「科學能力的建構反應評量」的理論構念模式，係為一項精確測量科學能力的工具。

This study aimed to develop and validate a constructed-response assessment of scientific abilities and an accompanying rubric. The assessment included 32 open-ended test items that were categorized into four subscales—Remembering and understanding scientific knowledge, application and analysis of scientific procedures, argumentation and expression of scientific logic, and evaluation and innovation during problem solving. The analysis revealed the following results: First, the Cronbach’s α values were higher than .90, indicating high intrarater consistency. Second, Kendall’s coefficient of concordance was higher than .90 and its p value was less than .001, denoting a consistent scoring pattern between raters. In addition, many-facet Rasch measurement (MFRM) analysis revealed no significant difference in rater severity, whereas a comparison of the rating scale model (RSM) and partial credit model (PCM) indicated that each rater had a unique rating scale structure. The infit and outfit mean squares of the MFRM were 1 ± 0.5, which suggested that both severe and lenient raters could effectively distinguish high and low-ability students. The Deviance values estimated by the RSM and PCM were converted to Bayesian information criterion values, and the RSM was viewed to fit the empirical data appropriately compared with the PCM. Therefore, the severity thresholds of the raters were the same. Third, Cronbach’s α coefficients of the four subassessments and the full assessment were higher than .85, indicating that the constructed-response assessment of scientific abilities (CRASA) provided a high internal-consistency reliability. Finally, confirmatory factor analysis revealed acceptable goodness-of-fit for the CRASA. These results suggested that the CRASA is a useful tool for accurately measuring scientific abilities.