本研究旨在運用項目反應理論(item response theory; IRT)、電腦化適性測驗(computerized adaptive testing; CAT)理論以及品質管理的技術，為IRT-Based題庫建立品質管理模式，提供合理、有效的品質管理策略，以篩選出品質不佳之試題，減輕題庫維護工作之負荷，並維護題庫精確測量之品質。研究之結果將可做為電腦化適性測驗及網路化適性測驗(networking adaptive testing)之動態題庫(dynamic item bank)規劃、設計、開發及維護上的參考。本研究可分為下列三步驟：首先進行文獻探討，將品質管理模式應用於IRT-Based題庫管理流程，建立題庫品質管理系統，並採用品質管理技術，以Urry、Baker及MicroCAT三種管制條件，搭配寬鬆、中度及嚴格三種不同管制程度，提出九種題庫品質管理策略；其次，進行品質評估系統之設計、開發、實作，以及模擬實驗研究；最後，再依據研究成果評估各題庫品質管理策略的效能。根據模擬實驗結果之差異分析，發現各品質管理策略中，以Baker所建議三參數範圍做為管制條件配合寬鬆管制程度，較能有效篩選IRT-Based題庫中試題訊息量較低之不良試題。倘若採用嚴格管制程度，配合三種管制條件，皆極易發生評估過於嚴苛，而產生誤將使用頻率頗高之試題評估為不合格的現象，中度管制程度配合三種管制條件之管理策略，亦有上述之情況出現。此外，在真實題庫之應用上，建議分別修正Baker 管制條件中之a參數範圍上限至3.0，以及MicroCAT管制條件中c參數之下限至零，才能有較佳的品質管理效果。在模擬題庫方面，則需調整試題缺點數計算公式中a參數之權數(weight)，才能提高品質管理策略之有效性。本研究為IRT-Based 題庫提出新的維護模式與簡單、經濟的題庫品質管理策略，經由模擬實驗之資料分析結果，驗證題庫品質管理策略的有效性，亦即管理策略可有效篩選IRT-Based題庫中試題訊息量較低之不良試題，不僅可增加題庫儲存空間之利用率，提高CAT選題時之計算效率，還可維護IRT-Based題庫之品質，對CAT實務上之應用極具參考價值。

The purpose of this study is to apply item response theory(IRT), computerized adaptive testing (CAT) theory and the techniques of quality management to develop aprocedure and proffer some useful strategies to screen useless items and maintain the IRT-based item banks. The results of the research will provide a planning ,designing ,developing, implementing and maintaining guideline of dynamic item bankfor computerized adaptive testing and even networking adaptive testing. The methodology of the study consists of three stages. First, according to the result of paper review plan and design the procedure of maintaining IRT-based item banks.Base on the technology of quality management, adopting three conditions (Urry's,Baker's and MicroCAT's suggestions) and three degrees (loose ,medium and strict) ofquality control to elaborate nine strategies. Second, the quality evaluating system was designed and implemented. Then, some simulation experimentations were proceeded.At the last, evaluate the effect of the procedure and nine quality control strategies. According to the analytical results of the simulation experimentations, Baker-loosestrategy is better than the others in screening the items with lower item informationfrom IRT-Based item banks. When adopting Urry-strict , Baker-strict and MicroCAT-strictstrategies to maintain item banks, some useful items misjudged. Because the criterionis too strict. Similarly, the similar result appears when the medium strategies were applied.Moreover, to modify the upper bound of a-parameter from 2.0 to 3.0 in Baker's suggestionsand the lower bound of c-parameter form 0.05 to 0 in MicroCAT's suggestions, it will makebetter efficiency in maintaining work of real item banks. Adjusting the weight of a-parameterin the item-defect formula will increase the utility of quality control strategy in virtual item banks. The research proposes a new maintaining procedure and some simple , economic , and useful quality control strategies. The results of the research present the availability of quality control strategies in IRT-Based item banks. In other words, the qualitycontrol strategy could screen useless items , increase the usage of space , upraise theefficiency of item selections and keep the quality of IRT-Based item banks. It will bevery useful for computerized adaptive testing and even networking adaptive testing.