1 / 22

Structural Equation Modeling

Structural Equation Modeling. Estimates and Tests in SEM. 資料來源 : 結構方程式 LISREL 的理論技術與應用 2003 邱皓政. 外衍觀測變數.  觀測變數. 內衍觀測變數. 外衍變數 exogenous variables (X).  潛在變數. 內衍潛在變數. 外衍潛在變數. 內衍變數 endogenous variables (Y). 外衍觀測變數.  觀測變數. 內衍觀測變數. Terminology.

lacy
Télécharger la présentation

Structural Equation Modeling

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Structural Equation Modeling Estimates and Tests in SEM

  2. 資料來源:結構方程式 LISREL 的理論技術與應用 2003邱皓政

  3. 外衍觀測變數  觀測變數 內衍觀測變數 外衍變數 exogenous variables (X) 潛在變數 內衍潛在變數 外衍潛在變數 內衍變數 endogenous variables (Y) 外衍觀測變數  觀測變數 內衍觀測變數 Terminology 資料來源:結構方程式 LISREL 的理論技術與應用 2003邱皓政

  4.              y y      x x Parameter in Structural and Measurement Model ■ 結構模型方程式 y ■ 變項測量模型方程式 x ■ 變項測量模型方程式 資料來源:結構方程式 LISREL 的理論技術與應用 2003邱皓政

  5. 資料來源:結構方程式 LISREL 的理論技術與應用 2003邱皓政

  6. Statistical Theory of Covariance 資料來源:結構方程式 LISREL 的理論技術與應用 2003邱皓政

  7. Generating the Covariance Matrix 資料來源:結構方程式 LISREL 的理論技術與應用 2003邱皓政

  8. Variance Covariance 資料來源:結構方程式 LISREL 的理論技術與應用 2003邱皓政

  9. cov(v1,v1) cov(v2,v1) cov(v3,v1) cov(v4,v1) cov(v5,v1) cov(v6,v1) cov(v1,v2) cov(v2,v2) cov(v3,v2) cov(v4,v2) cov(v5,v2) cov(v6,v2) Σ= cov(v1,v3) cov(v2,v3) cov(v3,v3) cov(v4,v3) cov(v5,v3) cov(v6,v3) cov(v1,v4) cov(v2,v4) cov(v3,v4) cov(v4,v4) cov(v5,v4) cov(v6,v4) cov(v1,v5) cov(v2,v5) cov(v3,v5) cov(v4,v5) cov(v5,v5) cov(v6,v5) cov(v1,v6) cov(v2,v6) cov(v3,v6) cov(v4,v6) cov(v5,v6) cov(v6,v6) Estimation of Parameters in Covariance Matrix Variables: v1 v2 v3 v4 v5 v6 p: number of variables in covariance matrix p(p+1)/2: non-redundant elements in the sample covariance matrix q: number of free parameters q<=p(p+1)/2 T- rule

  10. Simultaneous equations x+y=5 ---------- (1) 2x+y=8----------(2) x+2y=9----------(3) • Only (1)  (c, 5-c) • (1) and (2)  (3,2) • (1) , (2) , (3)  no exact solution under-identified just identified over-identified

  11. Smallest absolute difference from the constants (2.273,3.273) (2.333,3.333) Smallest squared difference (3,2) Y criterion X

  12. Example :T rule Variables: v1 v2 v3 v4 v5 v6 Simultaneous Equations Cov(v1,v1)=λ12+θ1 Cov(v3,v3)= λ32+θ3 Cov(v1,v2)=λ1λ2 Cov(v3,v4)= λ3λ6φ21 Cov(v1,v3)= λ1λ3 Cov(v3,v5)= λ3λ5φ21 Cov(v1,v4)= λ1λ4φ21 Cov(v3,v6)= λ3λ6φ21 Cov(v1,v5)=λ1λ5φ21 Cov(v4,v4)= λ42+θ4 Cov(v1,v6)=λ1λ6φ21 Cov(v4,v5)= λ4λ5 Cov(v2,v2)= λ22+θ2 Cov(v4,v6)= λ4λ6 Cov(v2,v3)= λ2λ3 Cov(v5,v5)= λ52+θ5 Cov(v2,v4)= λ2λ4φ21 Cov(v5,v6)= λ5λ6 Cov(v2,v5)= λ2λ5φ21 Cov(v6,v6)= λ62+θ6 Cov(v2,v6)= λ2λ6φ21 6(6+1)/2 =15 13 Estimate parameter λ1,λ2λ3λ4λ5λ6φ21θ1 θ2 θ3 θ4 θ5 θ6 13<=6(6+1)/2  T rule

  13. Estimation methods • Different estimation methods will yield different results for estimates and model tests. (Chapter 5 ) • Normal distribution assumption • ML:maximum likelihood • GLS: generalized least squares • ADF: asymptotic distribution free method --- non-normal distribution most commonly used

  14. Fitting Functions • The criterion selected for parameter estimation is AKA the discrepancy function. • F=F(S, Σ(hat θ))

  15. Maximum likelihood • 觀察數據都是從母體中抽取得到的資料,而所抽出的樣本必須是所有可能樣本中被選擇的機率的最大者,若能符合此一假設,估計的參數即能反應母體的參數。 • FML=log| Σ(θ)| + Trace[Σ(θ)-1S] – log|S| - p p : number of measured variable if Σ(θ)=S then Trace[Σ(θ)-1S] =Trace(I)=p  FML=0

  16. Generalized Least Squares • 使用差異平方和的概念,只是在計算每一個差異值時,同時計算了一個特定的權數用以整合個別的比較值。 資料來源:結構方程式 LISREL 的理論技術與應用 2003邱皓政

  17. Asymptotic Distribution Free • 一種無須常態假設為基礎的參數估計法,由於不需考慮常態分配的問題,因此稱為分配自由(free)。利用W-1權數,來消除多變量常態假設的影響。 資料來源:結構方程式 LISREL 的理論技術與應用 2003邱皓政

  18. estimated Population (sample) Goodness-of-fit Similarity Σ( hat θ) ~ Σ H0: Σ(θ)= Σ

  19. Testing(1) T=Min(F)*(N-1) ~ X2 distribution • q parameters p(p+1)/2 equations • df=(q-p(p+1)/2 ) >0  testable model • H0: Σ(θ)= Σ  significance  reject H0 • Large sample assumption

  20. Testing(2) • Scaled test (non-normal distribution) • SCALED T=c-1T • T : function of standard goodness-of-fit X2

  21. Practical problems • Covariance matrix is not positive definite– linear dependency among observed variables. • Non-convergence – iterative process is terminated before the predetermined criteria have been satisfied.

  22. Thank you The End

More Related