潛在特質理論與其應用於適性測驗之評估研究

林世華(Sieh-Hwa Lin)

《教育心理學報》 20期 (1987/05) Pp. 131-182

https://doi.org/10.6251/BEP.19870601.8

摘要

本研究主要的目的在於探討：(一)潛在特質理論與傳統測驗理論二者之間的關係。(二)潛在特質理論應用於適性，對於心理測量所産生的助益。
研究者使用電腦模擬的方式，模擬受試者對項目的反應，並就適性測驗θ已知與θ求知的兩種情況模擬反應的過程，從事系列性的追蹤觀察。
結果發現：
一、潛在特質理論的a參數、b參數及θ參數分別與傳統測驗理論的r(下標 bls)、P及個人得分等統計數乏間均有很高的相關。
二、θ已知時的翱生測驗路徑，根據b參數進行羣聚分析結果可以得到三種類型。這三種類型的路徑，也就是高、中、低三種不同能力水準受試者的路徑。這些路徑主要是依照能力與項目難度相匹配的原則所形成。就測量的功能而言，使用題庫中不到50%的題目所測量的結果與使用所有題目所測量的結果是一樣的。而且隨著題數逐步增加，側量結果也逐斯趨於穩定。
三、對θ未知的適性測驗路徑進行軍聚分析的結果顯示：根據a參數可以得到兩類，根據b參數可以得到三類，根據c參數則可以得到四類。每一種類型的路徑都代表某個能力水準範圍內受試者的路徑。路徑的形成主要還是依照能力與項目難度相匹配的原則。惟，在測量的功能上則呈不穩定的現象；測量的結果並不會因題數的增加而趨於穩定。
根據本研究的結果可以得到本研究的結論是：
一、潛在特質理論與傳統測驗理論二者之基木觀念上是相通一致的。二、將潛在特質理論應用於適性測驗是可行的，而且可以改善測量工作。但是因為目前使用的方怯產生的偏差仍舊相當大，所以在技術上仍有待進一步研究突破。
本文最後研究者根據結論提出，估計θ有關技術的改善建議，以利適性測驗的實際應用。

並列摘要

The purposes of this study were:
(1) to investigate the relationship between latent trait theory (LTT) and classical test theory (CTT), and
(2) to evaluate the efficiency of latent trait theory applied to the adaptive testing.
Computer simulation was used in this study. The response patterns of the ”examinee” were first generated, and the procedures of adaptive testing were simulated, supposing θ known and θ unknown. The testing paths of each examinee were traced, recorded, and analyzed.
The results were summarized as follows:
(1) The a, b, θ parameters in LTT were highly correlated with r(subscript bla), P, and obtained score in CTT respectively.
(2) whenθwas supposed to be known, the adaptive testing paths could be classified into three patterns according to b parameter in testing path. These three paths patterns belonged to examinees of high, middle and low ability level respectively. Apparantly, the paths were formulated by matching examinee's ability and item difficulty. when less then 50% of items in the pool were taken, the efficiency of measurement was the same as when all items in the pool were taken.
(3) whenθwas suppsed not to be known, the adaptive testing paths could be classified into two patterns according to a paramter, into three patterns according to b parameter, and into four patterns according to c parameter in path. The Paths were also formulated by matching examinee's ability and item difficulty. However, the efficiency of measurement was not stable.
Based on the results of this study, the conclusions were drawn:
(1) The basic concepts of LTT and CTT were similar and consistent.
(2) It was practicable to apply LTT to adaptive testing, however the techniques of estimatingθshould be improved furthur.