目的:探討不同迴歸分析方法對預測女于手球賽得失分差的攻守技術之差異，同時進行迴歸診斷，以找出最佳的預測模式，並比較迴歸分析與教練評價之間的相關。方法:根據北京2008年奧運會女子手球賽，官方的攻守紀錄，分成18個攻守技術變項，以得失分差為效標(Y)，應用逐步迴歸分析與c (P)法合併R2法進行分析，比較兩者問選入變項之差異;並訪談25位國家 級手球教練對影響比賽得失分差的評價，以斯皮爾曼等級相關考驗其與迴歸分析選入變項順序的相關。結果:逐步迴歸分析與c(P)法合併R2法都選入 相同的9個變項，可解釋女子手球賽得失分差的變異量為70.6%;教練評價與兩種迴歸分析所選入的9個變項有8個相同，兩者問具有顯著的相關(r主 = .91 ' 1.975(7)= 5.99 ' p < .05)。結論:逐步迴歸分析與C (P)法和R2法所選 入的9個變項，都可以作為預測女子手球比賽得失分差的最佳組合，且具有良好的評估品質。

Purpose: By using various regression analysis methods， this study delved into how different methods may have had different prediction about the attack/defense skills of the net game scores in women's handball games at Beijing 2008 Olympics Games; at the same time， there was a regression diagnosis undergoing to find the best predictive model. Besides， this study also compared the correlation between the regression analysis and the coaches' evaluations. Method: Based on the official records of women's handball competitions at Beijing 2008 Olympics Games， 18 attack/defense skills were chosen as variables and the net game scores were taken as (Y) to conduct stepwise regression analysis under the Mallow's C (P) and R2 methods， comparing the difference between the variables chosen by the two methods. 25 national handball coaches were interviewed to share their professional evaluations on what competition skills may be crucial for the net game scores. Moreover， the Spearman rank correlation was used to see the correlation between the order of the chosen variables of the aggression analysis and the coaches' opinions. Results: The stepwise regression analysis and the Mallow's C (P) method had both entered 9 same variables， which could explain that the variance of the net game scores in the women's handball games was 70.6%. Of the 9 variables from the two methods， 8 were unanimous with the coach variables， and there was significant correlation between the variables entered by the two methods and the ones entered by the coaches (rs = .91 ' t.975 (7)= 5.99 ' P < .05). Conclusion: The 9 variables entered by the stepwise regression analysis and the Mallow's C (P) and R2 methods are the best combination to predict the net game scores in women's handball games and provide ve可r valuable evaluations.