查詢結果分析
相關文獻
- 電腦化適性測驗題庫擴增研究
- 中文研究生學術性向測驗之編製
- 線上題庫系統之研發
- The Impact of Pool Modifications on the Exposure Control Parameters of CAT Items
- 試題反應理論參數估計電腦程式之簡介
- 試題反應理論的介紹(11)--題庫的建立
- 單向度試題反應理論之可能值方法於等化設計下之模擬實驗探究
- Design of the IRT-based Evaluation System Supporting with Material Recommendation
- 以a-鄰近法為選題策略之電腦化適性測驗模擬研究
- 傳統紙筆測驗與線上電腦化測驗試題參數估計差異之比較
頁籤選單縮合
題 名 | 電腦化適性測驗題庫擴增研究=The Research of Expanding Item Bank for Computerized Adaptive Test |
---|---|
作 者 | 吳慧珉; 蘇少祖; 趙佑軒; | 書刊名 | 測驗統計年刊 |
卷 期 | 17(下) 2009.12[民98.12] |
頁 次 | 頁19-53 |
分類號 | 521.3 |
關鍵詞 | 電腦適性測驗; 題庫; 試題反應理論; 參數估計; Computerized adaptive test; Item bank; Item response theory; Parameter estimation; |
語 文 | 中文(Chinese) |
中文摘要 | 摘要 本研究透過模擬研究,分析題庫更新時,如何兼顧試題參數估計的精準度及 更新的速率。在試題參數估計穩定後,是否能藉由適性測驗試題與新試題的作答 反應,增加能力參數估計的精準度。題庫的維護需耗費許多人力與成本,因此在 題庫擴充時,若依照題庫建立之初,對新的試題重新舉行大規模的預試,將加重 成本的負擔,本研究主要探討實施電腦適性測驗時,將未校準的試題逐次加入已 校準的題庫中,以試題反應理論為基礎,估計未校準的試題參數和能力參數,在 不增加成本的情況下,線上更新擴大題庫。 結果顯示如下: 一、新試題題數是 5 題或10 題,對於新試題參數估計並沒有太大的影響。 二、隨著受試者人數的增加,能力參數和試題參數之估計誤差逐漸降低。 三、受試者能力值呈現常態與雙峰分佈時,藉由加入新試題估計能力值,除了可 以估計出新試題參數之參數值外,更可估計出更精準的受試者的能力值。但 若受試者能力真值呈現偏態分佈時,無法估算出較精準新試題之參數與受試 者能力值。 四、受試者能力值呈現常態與雙峰分佈時,當加入新試題題數是 10 題時,其能 力估計誤差比加入新試題題數是5 題時較低。 |
英文摘要 | Abstract A simulation study was conducted to evaluate the accuracies of estimating un-calibrated item parameters and how the ability estimate were affected by different factors while expanding the item bank in the process of administrating computerized adaptive test. The main idea of this research was to make examinees answer not only calibrated items but also un-calibrated ones. The abilities obtained from calibrated items were used to estimate un-calibrated items, so the item bank could be expanded without pretest procedures. In the simulation study, three factors and their varied conditions were considered: different distributions of abilities, different number of examinees, and different number of un-calibrated items. The major findings of this study were summarized as follows: 1. The root mean square errors (RMSEs) of the un-calibrated item parameters were not affected by the number of items. 2. The RMSEs of the item and the ability parameters decrease as the number of the examinees crease. 3. Under normal or binomial distribution for ability, adding un-calibrated items to estimate ability could decrease the RMSEs of ability; under skewed distribution for ability, the RMSEs of both item and ability parameter were higher than normal and binomial distributions. 4. Under normal or binomial distribution for ability, the RMSEs of ability estimated by adding 10 un-calibrated items were lower than the RMSEs of ability estimated by adding 5 un-calibrated items. |
本系統中英文摘要資訊取自各篇刊載內容。