# 結構方程式模型

(由結構方程模型跳轉過嚟)

## 基本概念

### 迴歸分析

${\displaystyle DV=a+b(IV)}$ （線性迴歸；linear regression）；
${\displaystyle DV=a+b(IV)+c(IV)^{2}}$ （多項式迴歸；polynomial regression）；

... 等等。原則上，如果有個方法可以由過往數據嗰度搵出 ${\displaystyle a}$  同埋 ${\displaystyle b}$ 參數嘅數值，第時就可以靠條式大致上用 IV 嘅值估計 DV 嘅值；統計學同機械學習等嘅領域上有好多演算法可以用嚟搵出呢啲參數嘅數值[7]

### 因素分析

1. 想像家陣手上個數據庫有若干個被觀察咗（observed）嘅隨機變數 ${\displaystyle x_{1},x_{2},...,x_{p}}$ ，而呢柞變數嘅平均值係 ${\displaystyle \mu _{1},\mu _{2},...,\mu _{p}}$
2. 想像有 ${\displaystyle k}$ 冇被觀察到（unobserved）嘅隱藏變數（latent variable）${\displaystyle F_{j}}$ ${\displaystyle j\in 1,...,k}$ （呢柞 ${\displaystyle F_{j}}$  係所謂嘅因素）[註 1]
3. 喺做因素分析前，${\displaystyle F_{j}}$  嘅數值係未知，而因素分析嘅目的就係要搵出以下呢啲式當中嘅參數：
${\displaystyle x_{i}-\mu _{i}=l_{i1}F_{1}+\cdots +l_{ik}F_{k}+\varepsilon _{i}}$ ；當中
• ${\displaystyle i\in 1,...,p}$
• ${\displaystyle l_{ij}}$  係參數；
• ${\displaystyle \varepsilon _{i}}$ 誤差，平均值係 0，而變異數係一個有限數值，唔同 ${\displaystyle i}$ ${\displaystyle \varepsilon _{i}}$  變異數數值可以唔同[12]

## 建立模型

### 事前數據處理

${\displaystyle X_{1}}$  ${\displaystyle X_{2}}$  ${\displaystyle X_{3}}$  ${\displaystyle X_{4}}$
${\displaystyle X_{1}}$  ${\displaystyle 10}$  ${\displaystyle 6}$  ${\displaystyle 6}$  ${\displaystyle 6}$
${\displaystyle X_{2}}$  ${\displaystyle 6}$  ${\displaystyle 11}$  ${\displaystyle 6}$  ${\displaystyle 6}$
${\displaystyle X_{3}}$  ${\displaystyle 6}$  ${\displaystyle 6}$  ${\displaystyle 12}$  ${\displaystyle 6}$
${\displaystyle X_{4}}$  ${\displaystyle 6}$  ${\displaystyle 6}$  ${\displaystyle 6}$  ${\displaystyle 13}$

${\displaystyle \operatorname {cov} (X,Y)=\operatorname {E} {{\big [}(X-\operatorname {E} [X])(Y-\operatorname {E} [Y]){\big ]}}}$ ，當中
• ${\displaystyle \operatorname {cov} (X,Y)}$ ${\displaystyle x}$ ${\displaystyle y}$  呢兩個變數之間嘅協方差；
• ${\displaystyle X}$  係第 ${\displaystyle i}$  個個案嘅 ${\displaystyle x}$  數值；
• ${\displaystyle Y}$  係第 ${\displaystyle i}$  個個案嘅 ${\displaystyle y}$  數值；
• ${\displaystyle \operatorname {E} [X]}$  係啲個案喺 ${\displaystyle x}$  上嘅平均值
• ${\displaystyle \operatorname {E} [Y]}$  係啲個案喺 ${\displaystyle y}$  上嘅平均值。

### 估計模型參數

${\displaystyle {\text{Pr}}(X|\theta )}$  可以表達成[20]

${\displaystyle {\text{Pr}}(x_{1}\cap x_{2}\cap ...\cap x_{n}|\theta )}$  [註 4]

#### 最佳化

1. 睇吓自己身處嗰點周圍每個方向有幾斜，
2. 揀最能夠令自己向上爬嗰一個方向，移去嗰個方向，
3. 重複，直至某啲條件達到（例如 ${\displaystyle {\text{Pr}}(X|\theta )}$  超過咗某個特定數值）為止。

#### 結構模型

${\displaystyle x_{i}-\mu _{i}=l_{i1}F_{1}+\cdots +l_{ik}F_{k}+\varepsilon _{i}}$

## 模型評估

### 適合度

• 卡方檢定（Chi-squared test，χ2[註 6]：呢種做法將「個模型係正確嘅」當做 ${\displaystyle H_{0}}$ 虛無假說），並且攞「個模型嘅協方差矩陣」同「實際觀察到嘅協方差矩陣」做卡方檢定，如果卡方檢定嘅數值（χ2）愈大，就表示兩個矩陣之間嘅差異愈大－研究者就愈有理由相信個模型係錯嘅[29][30]
• 近似值根均方誤差（Root Mean Square Error of Approximation，RMSEA）：一個數值愈低愈好嘅適合度指標；RMSEA 最細嘅可能數值係 0，而一般認為，RMSEA 數值喺 0.1 或者以上嘅話個模型嘅適合度就算係低到唔可以接受[31]
• 標準化根均殘差（Standardized Root Mean Residual，SRMR）：另一個數值愈低愈好嘅適合度指標；一般認為，SRMR 嘅數值最好係喺 0.1 以下[2]，亦都有統計學家主張 SRMR 數值要喺 0.08 以下個模型先可以算係有充足嘅適合度[29]
• 比較適合指數（Comparative Fit Index，CFI）：一個主要反映數據當中嘅統計相關嘅大細嘅適合度指標，所以數值係愈高愈好；一般嚟講，CFI 嘅數值過到 0.95，個模型就算係可以接受[29]

... 等等。

## 多組分析

### 量度不變特性

• ${\displaystyle c_{1}}$  表示由喺美國得到嘅數據造出嚟嘅 ${\displaystyle c}$  量度模型、
• ${\displaystyle c_{2}}$  表示由喺歐洲得到嘅數據造出嚟嘅 ${\displaystyle c}$  量度模型、同埋
• ${\displaystyle c_{3}}$  表示由喺日本得到嘅數據造出嚟嘅 ${\displaystyle c}$  量度模型、

### 約束分析

• 首先俾個程式完全自由噉建立一個結構模型先，得出一個 χ2 值；
• 跟住佢作出約束，要求個程式喺「假設兩組嘅嗰一個 ${\displaystyle \beta _{xy}}$  數值一樣」嘅情況下建立一個結構模型，又得出一個 χ2 值，
• 如果呢個 χ2 值明顯大過[註 7]打前嗰個，就表示「假設兩組嘅嗰一個 ${\displaystyle \beta _{xy}}$  數值一樣」會搞到個結構模型嘅適合度明顯變差－噉個分析者就有理由相信「兩組嘅嗰一個 ${\displaystyle \beta _{xy}}$  數值有差異」，可以作出「『屬實驗組定對照組』呢個變數會對 ${\displaystyle X}$ ${\displaystyle Y}$  之間嘅關係有調節效應」嘅推論[36]

## 統計功效

${\displaystyle {\text{power}}=\Pr {\big (}{\text{reject }}H_{0}\mid H_{1}{\text{ is true}}{\big )}}$

• 大過 200 [40]
• 係要估計嘅參數數量嘅最少 10 倍[41]、而且
• 係變數數量嘅最少 10 倍[42]

## 註釋

1. ${\displaystyle a\in b}$ 」意思係「${\displaystyle a}$ ${\displaystyle b}$  呢個入面」。
2. 可以睇吓迴歸分析（regression analysis）。
3. 又有做法會事先將數據庫入面每個數據嘅數值減以相應嘅平均值，噉 ${\displaystyle \operatorname {cov} (X,Y)}$  就會簡化變成 ${\displaystyle \operatorname {E} {{\big [}XY{\big ]}}}$
4. 喺實際應用上，考慮咁多極細嘅數值可能會引起算術下溢（指要處理嘅數值細過部電腦能夠表示嘅最細值），所以「要點樣計 ${\displaystyle {\text{Pr}}(X|\theta )}$ 」呢家嘢有一定嘅學問。
5. 順帶一提，通徑分析（path analysis）基本上就係冇量度模型、齋用觀察到嘅變數嚟建立結構模型嘅 SEM。
6. 卡方檢定喺樣本大細大過要估計嘅參數數量（${\displaystyle df>0}$ ）嗰陣先會有用，而且當樣本大細大得滯嗰陣，卡方檢定會變得靠唔住。
7. 點樣算係「明顯大過」都有學問。

## 軟件

• LISREL
• R 程式語言當中都有軟件包做結構方程式模型。
• SPSS 當中嘅軟件包「AMOS」專係整嚟做結構方程式模型嘅。

## 文獻

### 一般文獻

• Bowen, N. K., & Guo, S. (2011). Structural equation modeling. Oxford University Press.
• Elliott, M. R. (2003). Causality and how to model it (PDF). BT technology journal, 21(2), 120-125.
• Kaplan, D. (2008). Structural Equation Modeling: Foundations and Extensions (2nd ed.). SAGE. ISBN 978-1412916240.
• Kline, Rex (2011). Principles and Practice of Structural Equation Modeling (3rd ed.). Guilford. ISBN 978-1-60623-876-9.

### 社科文獻

• Bagozzi, Richard P; Yi, Youjae (2011). "Specification, evaluation, and interpretation of structural equation models". Journal of the Academy of Marketing Science. 40 (1): 8–34.
• Hox, J. J., & Bechger, T. M. (1998). An introduction to structural equation modeling. Family Science Review, 11, 354-373.
• Hu, Li‐tze; Bentler, Peter M (1999). "Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives". Structural Equation Modeling: A Multidisciplinary Journal. 6: 1–55. doi:10.1080/10705519909540118. hdl:2027.42/139911.
• MacCallum, Robert; Austin, James (2000). "Applications of Structural Equation Modeling in Psychological Research". Annual Review of Psychology. 51: 201–226. doi:10.1146/annurev.psych.51.1.201. PMID 10751970. Retrieved 25 January 2015.
• Quintana, Stephen M.; Maxwell, Scott E. (1999). "Implications of Recent Developments in Structural Equation Modeling for Counseling Psychology". The Counseling Psychologist. 27 (4): 485–527. doi:10.1177/0011000099274002.
• Tarka, P. (2018). An overview of structural equation modeling: its beginnings, historical development, usefulness and controversies in the social sciences (PDF). Quality & quantity, 52(1), 313-354.

## 攷

1. Kaplan, D. (2008). Structural Equation Modeling: Foundations and Extensions (2nd ed.). SAGE. p. 79-88.
2. Kline, Rex (2011). Principles and Practice of Structural Equation Modeling (Third ed.). Guilford.
3. Hancock, Greogry R (2003). "Fortune Cookies, Measurement Error, and Experimental Design". Journal of Modern Applied Statistical Methods. 2 (2): 293–305.
4. Salkind, Neil J. (2007). "Intelligence Tests". Encyclopedia of Measurement and Statistics.
5. Westland, J. Christopher (2015). Structural Equation Modeling: From Paths to Networks. New York: Springer.
6. Hox, J. J., & Bechger, T. M. (1998). An introduction to structural equation modeling. Family Science Review, 11, 354-373.
7. Linear Regression with example. Towards Data Science.
8. Seber, G. A., & Lee, A. J. (2012). Linear regression analysis (Vol. 329). John Wiley & Sons.
9. YangJing Long (2009). "Human age estimation by metric learning for regression problems". Proc. International Conference on Computer Analysis of Images and Patterns: 74–82.
10. Regression in Machine Learning. Towards Data Science.
11. Thompson, B.R. (2004). Exploratory and Confirmatory Factor Analysis: Understanding Concepts and Applications. American Psychological Association.
12. Child, Dennis (2006), The Essentials of Factor Analysis (3rd ed.), Continuum International.
13. Mayes, S. D., Calhoun, S. L., Bixler, E. O., & Zimmerman, D. N. (2009). IQ and neuropsychological predictors of academic achievement. Learning and Individual Differences, 19(2), 238-241.
14. Kline, P. (2014). An easy guide to factor analysis. Routledge.
15. Rice, John (2007). Mathematical Statistics and Data Analysis. Belmont, CA: Brooks/Cole Cengage Learning.
16. Maydeu-Olivares, A. (2017). Maximum likelihood estimation of structural equation models for continuous data: Standard errors and goodness of fit 互聯網檔案館歸檔，歸檔日期2021年2月16號，. (PDF). Structural Equation Modeling: A Multidisciplinary Journal, 24(3), 383-394.
17. Rossi, Richard J. (2018). Mathematical Statistics : An Introduction to Likelihood Based Inference. New York: John Wiley & Sons. p. 227.
18. Lee, S. Y., & Zhu, H. T. (2002). Maximum likelihood estimation of nonlinear structural equation models. Psychometrika, 67(2), 189-210.
19. Preedy, V. R., & Watson, R. R. (2009) Handbook of Disease Burdens and Quality of Life Measures. New York: Springer.
20. Factor Loadings. Salkind, N. J. (2010). Encyclopedia of research design (Vols. 1-0). Thousand Oaks, CA: SAGE Publications, Inc. doi: 10.4135/9781412961288.
21. Singh, R. (2009). Does my structural model represent the real phenomenon?: a review of the appropriate use of Structural Equation Modelling (SEM) model fit indices. The Marketing Review, 9(3), 199-212.
22. MacCallum, Robert; Austin, James (2000). "Applications of Structural Equation Modeling in Psychological Research". Annual Review of Psychology. 51: 201–226.
23. Xia, Y., & Yang, Y. (2019). RMSEA, CFI, and TLI in structural equation modeling with ordered categorical data: The story they tell depends on the estimation methods. Behavior research methods, 51(1), 409-428.
24. Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 6, 1-55.
25. Deciding Between Competing Models: Chi-Square Difference Tests (PDF). Introduction to Structural Equation Modeling with LISREL.
26. Browne, M. W.; Cudeck, R. (1993). "Alternative ways of assessing model fit". In Bollen, K. A.; Long, J. S. (eds.). Testing structural equation models. Newbury Park, CA: Sage.
27. Loehlin, J. C. (2004). Latent Variable Models: An Introduction to Factor, Path, and Structural Equation Analysis. Psychology Press.
28. MacCallum, Robert (1986). "Specification searches in covariance structure modeling". Psychological Bulletin. 100: 107–120.
29. Sarstedt, M. , Henseler, J. and Ringle, C. (2011), "Multi-group analysis in partial least squares (PLS) path modeling: alternative methods and empirical results", Advances in International Marketing, Vol. 22 No. 1, pp. 195-218.
30. Vandenberg, Robert J.; Lance, Charles E. (2000). "A Review and Synthesis of the Measurement Invariance Literature: Suggestions, Practices, and Recommendations for Organizational Research". Organizational Research Methods. 3: 4–70.
31. Takayama, Akira (1985). Mathematical Economics (2nd ed.). New York: Cambridge University Press. p. 61.
32. Hoenig; Heisey (2001). "The Abuse of Power". The American Statistician. 55 (1): 19–24.
33. Quintana, Stephen M.; Maxwell, Scott E. (1999). "Implications of Recent Developments in Structural Equation Modeling for Counseling Psychology". The Counseling Psychologist. 27 (4): 485–527.
34. Westland, J. Christopher (2010). "Lower bounds on sample size in structural equation modeling". Electron. Comm. Res. Appl. 9 (6): 476–487.
35. Boomsma, A. (1985). Nonconvergence, improper solutions, and starting values in LISREL maximum likelihood estimation. Psychometrika, 50(2), 229-242.
36. Bentler, P. M., & Chou, C. P. (1987). Practical issues in structural modeling (PDF). Sociological methods & research, 16(1), 78-117.
37. Nunnally J.C. (1967). Psychometric theory. New York, NY: McGraw-Hill.
38. Wolf, E. J., Harrington, K. M., Clark, S. L., & Miller, M. W. (2013). Sample size requirements for structural equation models: An evaluation of power, bias, and solution propriety (PDF). Educational and psychological measurement, 73(6), 913-934.
39. Chou, C. P.; Bentler, Peter (1995). "Estimates and tests in structural equation modeling". In Hoyle, Rick (ed.). Structural equation modeling: Concepts, issues, and applications. Thousand Oaks, CA: Sage. pp. 37–55.