本研究實徵性地探討根據不同的選題策略所編製的測驗，應用於不同測驗目換時的適合程度。研究的方法是以一份45題的國中酸鹼概念全測（universe test, UT)對598位國中三年及的學生樣本施測，測試結果經古典測驗理論與試題反應理論之試題分析，提供試題特性的訊息，再依據不同的選題策略，從全測中各選取20題組成三份分測驗。
第一份分測驗是選取古典鑑別度指數最高的20題所組成的古典測驗（classical test, CT )，第二份分測驗是選取最能代表等距樂許（Rasch）量尺的20題所組成的樂許測驗(Rasch test, RT)，第三份是隨機取樣方式選取20題所組成的抽樣測驗（sampled test, ST)。分別比較這些測驗應用在區辨（differentiation)、定位(location)、與選擇(selection)三種不同測驗目的時之效能。
茲將研究的主要結果摘述如下：
一、當測驗的目的在區辨學生時，三份分測驗區辨學生的效能有顯著的差異，其中以古典鑑別度指數的選題策略區辨高低能力學生的效果較佳，但進一步分析顯示，此種選題策略會造成高估高分組學生的得分，同時也低估低分組學生的得分。
二、當測驗的目的在將學生定位在酸鹼的知識量尺上時，CT、RT、ST與UT估計學生樣本能力之平均值並無顯著差異。也就是根據本研究的三種選題策略所編製的測驗，經樂許模型（Rasch model）所估計的能力，均能有效地將學生定位在酸鹼知識量尺上。
三、當測驗的目的在選擇學生時，三份分測驗均能有效地預測學生是否通過酸鹼概念測驗之全測，但整體而言，仍以CT的效果較佳。進一步分析顯示，對通過者的判定，以CT和RT之結果較準確；對未通過者的判定以CT為佳。

The proposition of different item selection strategies which contribute to the development of tests that optimally serve different purpose were empirically exlored. An universe test (UT), consisting of 45 multiple choice problem of acid-base concept, was given to a sample of 598 junior high school students. The statistical results of this test provided the information of item characteristics by the analysis of classical test theory and item response theory. Then, three subtests of 20 items were selected from universe test according to different item selection strategies.
The first subtest, classical tset (CT), was composed of 20 items with the highest classical item discrimination indexes. The second subtest, Rasch test (RT), which was also 20 items selected on the basis of Rasch difficulty in equal intervals. The third subtest, sampled test (ST), which was consist of 20 items randomly selected form universe test. The study compared the efficiency of the three subtests, if the purpose of such testing was to the differentiation, location, and selection of students, respectively.
The main conclusions of this study were summarized as follows:
1.When the purpose of testing is the differentiation of students, the effectiveness of three item selection strategies reveals significant difference. The selection of items on the basis of their classical discrimination indexes gave effectively in differentiation between high and low ability students. However, advanced analyses indicated that this item selection strategy was to overestimate the scores of students in the high-score groups and underestimate that in the low-score groups.
2.When the purpose of testing is the location of students on a continum of acid-base knowledge, none of the UT、CT、RT and ST could provide significantly different ability estimates for the sample of students. In other words, it could be concluded that tests developed according to three item selection strategies examined in this study are equally effective at locating students on a continum of acid-base knowledge when estimates of student ability along the continum are based on the Rasch model.
3.When the purpose of testing is the selection of students, each of three subtests could effectively predict whether students pass the universe test or not. But CT is still better than others in all. Advanced analyses indicated that CT and RT could predict more effectively in the judgement of the passing groups, whereas CT could predict more effectively in the judgement of the nonpassing groups.