Scribd
Upload
44 views

Covariance Vs Component Based Structural Equation Modeling: Prof. Dr. Imam Ghozali

This document discusses the differences between covariance-based structural equation modeling (SEM) and component-based SEM. Covariance-based SEM estimates relationships between variables by minimizing differences between the sample and estimated covariance matrices. It allows modeling of unobserved latent variables and accounts for measurement error. Component-based SEM instead focuses on explaining variance and is more suitable for prediction. Popular software for each approach is also listed.

Uploaded by

uletboeloe
Copyright
© © All Rights Reserved
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
44 views

Covariance Vs Component Based Structural Equation Modeling: Prof. Dr. Imam Ghozali

This document discusses the differences between covariance-based structural equation modeling (SEM) and component-based SEM. Covariance-based SEM estimates relationships between variables by minimizing differences between the sample and estimated covariance matrices. It allows modeling of unobserved latent variables and accounts for measurement error. Component-based SEM instead focuses on explaining variance and is more suitable for prediction. Popular software for each approach is also listed.

Uploaded by

uletboeloe
Copyright
© © All Rights Reserved
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 49

Covariance vs Component Based

Structural Equation Modeling

Prof. Dr. Imam Ghozali


Covariance Based SEM
• Structural equation modeling (SEM) also
known as covariance structured analysis,
latent variable (tidak observed) analysis,
confirmatory factor analysis or Lisrel
analysis (software name)
• SEM is an evolution from multiequation
modeling in econometrics combined with
measurement in psychology
(psychometrics) and sosiology
Characteristics of SEM

• Estimation of multiple interrelated


dependence relationship, while
multivariate analysis use separate
relationship for each set of dependent
variable.
• Able to present unobserved concept and
account measurement error in estimation
process.
Characteristics of SEM

• SEM-based procedures have substantial


advantages over first-generation
techniques such as principal components
analysis, factor analysis, discriminant
analysis, or multipleregression.
• Greater flexibility that a researcher has for
the interplay between theory and data.
SEM provides the researcher with
the flexibility to:
• (a) Model relationships among multiple
predictor and criterion variables.
• (b) Construct unobservable LVs.
• (c) Model errors in measurements for
observed variables.
• (d) Statistically test a priori
substantive/theoretical and measurement
assumptions against empirical data.
Covariance SEM Software

• Analysis of moment structured (AMOS 24)


• Lisrel 9.20
• EQS 6.2
• MPlus
• SEPATH
Types of SEM Models

• Path Analysis Models


• Confirmatory factory analysis models
• Structural regression models
• Latent change models
Path Analysis
Confirmatory Factor Analysis
Structural Model Analysis
Latent Growth Curve Model
Covariance Estimation
Bivariate Correlation
X1 X2 Y1
X1 1.0
X2 0.50 1.0
X1 B Y1 0.60 0.70 1.0

A Y1 Penyelesaian persamaan
0.50 = A
C 0.60 = B + AC
X2 0.70 = C + AB
Substitusi A = 0.50
Correlation as compound path 0.60 = B + 0.50C
Corr X1X2 = A 0.70 = C + 0.50B
Corr X1Y1 = B + AC Maka
Corr X2Y1 = C + AB B = 0.33
Jadi rX1Y1= b1= B = 0.33 C = 0.53
r X2Y1= b2 = C = 0.53
Steps in SEM
• Developed model based on theory:
1. Modeling Strategy
- Confirmatory model
- Competing model
- Model development
2. Theoretical model specification
- Casual relationship
- Avoid spesification error
3. Developed Path Diagram
- Determine exogen and endogen constructs
- Determine relationship in the path diagram
Steps in SEM
4. Convert path diagram into:
- Structural equation
- Determine measurement model
- Determine number of indicator
- Calculate construct reliability
- Indentify correlation between construct and indicators
5. Choose input matrix : correlation or covariance
6. Evaluate model estimation and Goodness-of fit
7. Interprete model and modified model
Casuality Relationship in Path Diagram
Casual Relationship Path Diagram
Independent Dependent
Variable Variable

X1 X2 Y1 X1 X2

Y1

X1 X2 Y1 X1 X2
X2 Y1 Y2
Y1 Y2

X1 X2 Y1 X1 X2 X3
X2 X3 Y1 Y2 Y2
Y1 Y2 Y3 Y1 Y2

Y3
Structural Equation

Endogen = Exogen + Endogen + Error


Variable Variable Variable
------------- ------------------------- ---------------------- ------
Y1 X1 X2 X3 Y1 Y2 Y3 e1

Y1 = b1X1 + b2X2 + e1

Y1 = b1X1 + b2X2 + e1
Y2 = b3X2 + b4Y1 + e2

Y1 = b1X1 + b2X2 + e1
Y2 = b3X2 +b4X3 + b5Y1 +b6Y3 + e2
Y3 = + b7Y1 +b8Y2 + e3
Model Identification
• Only identified model that can be estimated by
SEM
• Model is identified if it has only one unique
solution for each estimation
• Model identification:
- Overidentified if data point>number of
parameters
- Justidentified if data point=number of
parameters
- Underidentified if data point<number of
parameters
Model Estimation
• Population parameter is estimated by
minimazing the difference betrween estimated
covariance matrix and sample covariance matrix
• That is minimazing Q function
Q = (s – s(Q))’W(s – s(Q))
Where s is a vectorized sample covariance
matrix, s is a vectorized estimated matrix
and Q indicates that s is estimated from
the parameters and W is a weight matrix
Model Estimation Procedure
• There are 6 estimation procedures that can be
used:
1. ULS (Unweighted least square)
2. GLS (Generalized least square)
3. ML (Maximum likelihood) default AMOS
4. WLS (Weighted least square)
5. ADF (Asymptotically distribution free)
6. Satorra-Bentler Scaled Chi-square (corrected
ML estimate for non-normality of data)
Goodness fit Model

• Chi-squares=kecil sehingga p>0.05


• GFI, AGFI,TLI, CFI >0.90
• RMSEA <0.08
Model Modification

• Modification Index
Covariances: (Group number 1 - Model A)

M.I. Par Change

eps2 <--> eps4 13.161 3.249

eps2 <--> eps3 10.813 -2.822

eps1 <--> eps4 11.968 -3.228

eps1 <--> eps3 9.788 2.798


Model Evaluation
• Measurement Model:
1. Convergent validity – Loading >0.07
2. Discriminant validity – Akar AVE > korelasi
3. Composite reliability (CR) >0.7
4. Average variance extracted (AVE)>0.50
• Structural Model
1. t statistics test - sigmnifikan
2. R2
Second Order Factor Model
SEM and Moderating Variable

• Y = a + b1X + b2Z + B3X*Z


x1 X y1
Y
x2 y2

Z
z1

X*Z
z2

(x1 + x2) (z1 +z2)


Moderating Spesification
• Loading factor from the interaction latent
variable
λinteraction=(λx1+ λx2) (λz1 + λz2)
• Error variance of indicator from the
interaction latent variable
θq=(λx1+ λx2)2VAR(X)(θz1+ θz2)+
(λz1+ λz2)2VAR(Z)(θx1+ θx2)+
(θx1+ θx2) (θz1+ θz2)
Formative vs. Reflective
Indicators

• Yet a common and serious mistake often


committed by researchers is to
inadvertently apply formative indicators
(also known as cause measures) in an
SEM analysis.
Reflective and Formative Indicator

Stress Stress

Stomach Headache tension Divorce Retired Bankcrupt


ache

Reflective Indicator Formative Indicator


Formative Indicator

• An example is socio-economic status (SES),


where indicators such as education,
income, and occupational prestige are
items that cause or form the LV SES. If an
individual loses his or her job, the SES
would be negatively affected.
• Another example of formative measures
would be the amount of beer, wine, and hard
liquor consumed as indicators of mental
inebriation.
• For the formative measures, an increase in
beer consumption does not imply similar
• Increases in wine or hard liquor consumption.
Thus, while it may occur, formative indicators
need not be correlated nor have high internal
consistency such as Cronbach's alpha (Bollen
1984; Bollen and Lennox 1991).
Component Based SEM
Component SEM Software

• VPLS under DOS


• SmartPLS 3
• PLS Graph 3.0
• Visual PLS 1.8
• PLS GUI
• XLStat PLSM
• Warppls 5.0
Thank You!

Q&A

You might also like