多元回归方程式.PPT

* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * 表16.10 階層迴歸分析各區組模型摘要與參數估計值 * Basic assumptions to regression Assumptions Assumptions for residuals (error scores) Zero Mean Homoscedastic Independence with predictors Normality Assumptions for specification errors Linear relationship All relevant predictors must be included No irrelevant predictors can be included Assumptions for measurement errors Relevant measurement procedures and variable selections Providence of the goodness index of measurement * Issues in Regression Multicollinearity Theoretical issues Analytic or Technical issues Measurement issues Categorical variable as predictors Effect coding Dummy coding Type of regression analysis Determination of selection procedures of predictors Simultaneous regression Stepwise regression Hierarchical regression Controlling for Type I and II error Less is more Theoretical consideration Measurement consideration * Misusage and threaten factors for correlation coefficient Conclusion of causal effects Restricting or Irrelevant range of variables Underestimation due to limited variance of variable Spurious correlation Correlation due to common cause Influences of Nuisance variables Confounding effects due to the third variables Suppressor variable: Irrelevant variance can be Suppressed Skewness to normality Influences of extreme scores Poor measurement attributes Low reliability Poor validity Irrelevant criterion variables Violation of assumption of linear relationship Misuse of non-continuous variables 線性關係分析：路徑分析Path Analysis * 路徑分析基本原理 一種用以探討多重變項之間因果結構模式的統計技術 最初由遺傳學家Wright於1921年所提出，至1960年代才廣泛受到重視 路徑分析由一系列的迴歸分析所組成，除了借用迴歸方程式的原理，並透過假設性的架構，將不同的方程式加以組合，形成結構化的模式 * 路徑分析的基本概念 結構方程式（structural equation） 構成路徑模型的數學方程式，外衍與內衍變項之間的關係係數bi，稱為路徑係數（path coefficient）。 外衍變項(exogenous variable) 模型中作為影響或解釋其他變項的變異量的變項。其變異量由不屬於路徑模型的其他變項所決定。 內衍變項(endogenous variable) 模型中被他人所影響或解釋的變項。其變異量由外衍變項及殘差（干擾）變異量兩部分。 干擾變異(disturbance