本研究旨在發展一個題目參數的線上校準（on - line calibration）策略，以求能由線上施測的過程獲得題目的足夠資訊，以作為校準之基礎。在項目反應理論（item response theory, IRT）的研究之中，題庫建立後再加入新題目是一個較為複雜的問題，主要牽涉了參數推估及連結等化等議題。在IRT的參數估計中，一般來說都是事先進行大量的紙筆施測，再由紙筆測驗所得之結果對項目參數及使用者能力參數進行同時推估。本研究擬利用受試者（examinee）在進行適性測驗後之能力估計值，將此估計值視為真實能力值，而在能力值已知的情況下直接進行推估。
其次，本研究對於參數之連結等化亦有所考量，新題目與舊有題庫之連結方式主要是利用定錨設計（anchored item design）；首先在原始題庫中選取若干具代表性的定錨項目，在受試者進行完適性測驗後，將新題目與定錨項同時給予受試者進行傳統測驗（即利用電腦以循序方式呈現所有題目），直到獲得足夠資訊得以進行參數推估為止。當參數推估結束後，利用定錨項在原始題庫與新推估的參數間的估計值做等化的工作，至於等化模式主要是利用Stocking與Lord (1983)所提之試題特徵曲線（test characteristic curve; TCC）；轉換法，將新項目所得參數與原題庫之參數連結，以建構一個新的適性測驗題庫。至於定錨項的選擇，則有四種不同之策略，分別為：簡單設計（simple design）與鑑別度設計（discrimination design），後者分為高鑑別力（high discrimination）、中鑑別力（middle discrimination）及低鑑別力（low discrimination）。
研究結果顯示了三項事實：首先，支持了線上題目參數校準之可行性，本研究利用模擬方式發現，當系統收集到500名受試者之反應組型後，且選取9題之定錨項，則簡單設計及中鑑別力設計之參數校準將可獲致可接受之結果；其次研究結果亦顯示，對於上述兩種設計方式而言，定錨項越多則校準效果越佳；最後若採取高鑑別力設計或低鑑別力設計則五題定錨設計即可獲致較佳之效果。

The purpose of this study is to develop a procedure of on - line calibration. On - line method for calibrating new item is described. In item response theory (IRT), how to add new items to the original item bank is a complex problem, and it is involved in two issues - item parameter estimation and item parameter equating & linking. In tradition, the item parameters are estimated with ability parameter after a large amount of paper and pencil testing. This is because of the lack of the parameters. But in the above subject, the method is inapposite. Because the examinee's ability is estimated by adaptive testing, it is thought to be the real one. In the other words, the item parameter could be estimated when the ability is known. The first study of this research is to explore the feasibility of this thought.
Besides the issue, it must consider about item parameter equating and linking. For this issue, the anchored item design (Stocking, 1988) is adopted in this study. When the examinee finishes his adaptive testing, there is a nonadaptive testing to the examinee and the work is finished until receiving enough information. When the ending point is reaching, the parameter estimation is occurring. There are six designs for the number of anchored items: 5, 6, 7, 8, 9, and 10 anchored items design. Thest characteristic curve (TCC) method (Stocking & Lord, 1983) is used to implement the linking module.
Besides this, the new strategy for anchored items' selection is approached. There are two consideration about this issue: simple design and discrimination design. The first design selected anchored items with no consideration about discrimination power parameter. The second design contains three selected strategy high discrimination selects highest discriminating paramenters and middle discrimination select the middle ones and low discrimination is opposite to the high one.
The results show three facts. The first, the feasibility for on - line calibration is proved. When system collected collected above 500 response patterns of examinees, both of the design (simple and middle discrimination) in 9 anchored items design acquired acceptable results. The second, the higher anchored items design produce more efficient outcomes in simple and middle discrimination design. The last, it is more efficient for high and low discrimination design.