|Abstract: ||在今日,資料庫的使用已是十分普遍。然而,有關資料庫設計方法論的實證研究仍非多見。本研究以探索中外文獻中有關資料庫設計的實證研究開始,而後採用實驗室實驗的策略,配合問卷的資料蒐集方法。以學校之大二資管系學生為實驗對象,來模擬初學的資料庫設計師。要求最終實施的資料庫為「符合第三正規化的關連表格」,來比對二種方法:(1)先繪實體關係(ER)圖,再轉換成關連表格;(2)直接採用正規化的過程,分解大關連,由第一、二而至第三正規化。實驗舉行二次,實驗I有101位學生參與。實驗II有98位同學參與。所有學生之電腦基礎訓練課程成績均予收集。認知型態以Slocum與Hellriegel(1983)的量表加以事先度量。實驗I有兩個個案:A是以文字敘述,也較複雜;B是以表格為主,也較簡單。實驗II也有兩個個案,不過其資訊需求相同(較實驗I之表格個案複雜,但較其文字個案簡單),而只是分為文字及表格形式。在兩個實驗中,每位學生均要求使用兩種方法。但使用何種方法於那個個案,則係考慮其認知型態,以求各組各種認知型態人數相同,再依隨機原則指派。衡量的應變數,則包含設計正確性(實驗I有12個指標,實驗II有8個指標)、建模時間,以及事後以類似Batra、Hoffer與Bostrom(1990)的量表度量之知覺易用性與偏好。 實驗結果顯示:資訊評估方式對設計績效有些有限影響。思考型表現較佳。在實驗I中,對複雜的文字敘述個案,ER方法比正規化分解方法績效較好;而在簡單的表格表示個案中則相反。但「ER方法適用於文字敘述」、「正規化分解適用於表格表示」的論點,在實驗II中,僅有不顯著的方法論差異,並未得到充分的統計支持,不過,ER是比正規化顯著費時,另外,實驗I、II均顯示,受測者大體上認為ER技術較易用、也較偏好,但要求很多心智努力。而且,對此感受與其正確性績效間並沒有相關。|
Databases have been widely used for many years. However, there are few empirical studies on database design methodologies. This research began with surveying empirical literature on database design. Then laboratory experiment strategy with questionnaires were adopted. The experiment subjects were MIS sophomore students to simulate novice database designers. The final outputs were third normal form relational tables. Two methods were compared: (1) first drawing Entity-Relationship (ER) diagram, and then transforming to normalized tables; (2) directly decomposing a "large" relational tables into first, second, third normalized tables. There were two experiments: 101 students in experiment I and 98 students in experiment II. The computer literacy backgrounds of students were collected. Their cognitive styles were measured by questionnaires of Slocum and Hellriegel (1993). Both experiments had two cases. In experiment I, complicated one (Case A) was described by words, the simple one (Case B) was in tables. Those two cases in experiment II had same information requirements, which were more complicated than Case B, but simpler than Case A of experiment I. They were different in formats: word description vs. tables. In both experiments, every student was required to apply different methods on different cases. The assignments were random, but trying to keep the numbers of the different cognitive styles in different teams equal. The dependent variables included correctness (12 indicators in experiment I, 8 in experiment II), modelling time, perceived ease-of-use and method preference (referencing to Batra, Hoffer and Bostrom (1990)). The results indicated that information evaluation styles had a little influence on design performance (thinking style was better than feeling). In Case A of the experiment I, the result correctness of the ER method was better than the decomposition method. But, the reverse was true in Case B of the experiment I. However, the propositions, "ER method suitable to table format" and "decomposition method suitable to table format", were not statistically significantly supported in experiment II. But ER method took more time significantly. In addition, both experiments indicated that in average subjects perceived ER method easy-to-use and had more preference, but thought that ER methods need more mental works. These perception was irrelevant to correctness.