变异数分析.ppt

變異數分析 Analysis of Variance 11.1 引言 壹、變異數分析（ANOVA） 變異數分析（ANOVA）是一般線性迴歸模型的特例，最常用來分析由實驗所蒐集的資料。 檢定某特定因子(例如價格)或處理(例如新藥)對於應變數(銷售或治癒率)的影響(效果)。 　　例子：比較實驗組新藥和對照組舊藥的治 療率是否有顯著差異 因子或處理通常是離散數值(不連續，如高、中、低) 可同時檢定多個因子對於應變數的影響，以及因子間的交互作用　 　　　 貳、共變數分析（ANCOVA） 在ANOVA分析中，納入一個連續(metric)自變數（又稱共變數）來解釋無法控制的變異 參、多變量變異數分析 包含一個以上的因變數 MANOVA和MANCOVA 11.1.1 應用案例 背景： 　Ofir和Simonson(2001)研究125位顧客，其在特定商店消費前是否事先告知參與評分顧客服務(expecting to evaluate)及購物前預期滿意度(the expectations of shopping experience)對於消費後商店滿意度評分的影響。 Two Factors Evaluation expectation Expected Not expected Experience expectations Positive Negative Positive expectations condition The employees of the store had received special training on how to provide good customer service. The store employed a special consultant who worked with the management to improve customer service. A recent overall evaluation conducted by an independent expert had rated the store as excellent on various dimensions related to customer service. Customers interviewed by the store management concurred with the independent expert’s judgment. Negative expectations condition The store was understaffed because of employee absenteeism. A few new employees had been hired but were still untrained. A recent overall evaluation conducted by an independent expert had rated the store as poor on various dimensions related to customer service. Customers interviewed by the store management concurred with the independent expert’s judgment. 結論： 　1、正面預期的消費者對於商店滿意度的評價都高於負面預期者 2、正面預期的消費者在事先告知參與評價時對於消費滿意度評價相對於未事先告知者顯著較低；但對於負面預期的消費者影響不顯著。 11.2如何使用ANOVA和ANCOVA─ Wildt和 Ahtola(1978)的研究 壹、 ANOVA 例子：檢定在芬蘭鄉間城市實施賣酒政策對交通事故發生數目的影響。 樣本來源：選取12個城市，隨機將其中4個城市指派為允許有照酒店和餐廳販可售酒，隨機指派4個城市是只允許有照酒店販售，隨機指派4個城市是不允許販售。樣本期間是一年。本例為one factor balanced design. 資料列於表11.1。 結果： 1. 從表11.1和圖11.2中發現交通事故發生次數在兩個實驗組中都較高( =206.5， =214.5)，對照組 =186.5。 2. 每一組的變異程度都很大，S1=26.2、S2=13.7和S3=18.7。因此，無法直覺地判斷出應變數的變動是否受到實驗因子顯著影響。 NOTATION: ONE-FACTOR DESIGN 假設檢定：檢定m組間應變數的平均是否有差異。虛無假說設定各組平均數接相等，即沒有處理的效果。m是nu