Sei sulla pagina 1di 22

Partial Least Squares Structural

Equation Modeling (PLS-SEM)


Nikunj Kumar Jain
FPM (IIM Indore)
Asst Prof (IMI Delhi)
nikunj.jain@imi.edu, f11nikunjk@iimidr.ac.in
Mob. 9827440301, 9999453770
Organization of Multivariate Methods*
Primarily Exploratory Primarily Confirmatory
First-generation techniques Cluster analysis Analysis of variance
Exploratory factor Logistic regression
analysis Multiple regression
Multidimensional scaling
Second-generation PLS-SEM CB-SEM, including
techniques confirmatory factor
analysis

*Hair Jr, J.F., Hult, G.T.M., Ringle, C. and Sarstedt, M., 2016. A primer on partial
least squares structural equation modeling (PLS-SEM). Sage Publications.
SEM?
• More than one dependent variable
• CFA
• Types of SEM
– CB-SEM
– PLS-SEM
CB-SEM vs PLS-SEM
CB-SEM PLS-SEM
Theory-testing Theory-building
Confirmatory Exploratory
Simple model Complex model
Sufficiently large sample size Low sample size
Normally distributed data Non-normal data
CFA
• Model Assessment
• Model Fit (Chi-square, df/cmin, p-value)
• Fit Indices (CFI, GFI, AGFI, RMSEA)
Hu and Bentler (1999)
Key Characteristics of PLS-SEM
• Sample size
• Distribution
– No distribution assumptions
– Non-parametric method
• Missing values
• Scale of measurement
A Systematic Procedure for Applying
PLS-SEM
Specifying the Structural Model

Specifying the Measurement Models

Data Collection and Examination

PLS Path Model Estimation


PLS-SEM

Assessing PLS-SEM Results of the Reflective


Measurement Models

Assessing PLS-SEM Results of the Formative


Measurement Models

Assessing PLS-SEM Results of the Structural


Model

Advanced PLS-SEM Analyses

Interpretation of Results and Drawing


Conclusions
Systematic Evaluation PLS-SEM
Results
Reflective Measurement Models Formative Measurement Models
Internal Consistency (composite Convergent Validity
reliability)
Indicator reliability Collinearity among indicators
Convergent validity (average variance Significance and relevance of outer
extracted) weights
Discriminant validity
Systematic Evaluation PLS-SEM
Results
Evaluation of the Structural Model
Coefficients of determination (R2)
Predictive relevance (Q2)
Size and significance of path coefficients
f2 effect sizes
q2 effect sizes
Reliability and Validity
• Indicator Loadings
– 0.7
• Internal Consistency (Composite Reliability)
– .708
– 0.6 – 0.7(exploratory)
• Average Variance Extracted
– 0.5
• Discriminant Validity (Fornell-Lacker Criteria)
Reliability and Validity
• AVE=Average lamda square
• Composite Reliability=(Square of summation
of lamda)/(Square of summation of
lamda+(summation of 1-lamda square))
Structural Path Coefficients
• Standardized Regression Weights, p-value
• Squared Multiple Correlations- R square Value
• Bootstrapping Technique
Common Method Bias
• Procedural
• Statistical
– Harmon’s Single Factor Test
– Common Latent Factor Approach
– Common Variable Factor Approach
Reflective and Formative Constructs
Refer Appendix 1 of following paper:
Coltman, T., Devinney, T. M., Midgley, D. F., &
Venaik, S. (2008). Formative versus reflective
measurement models: Two applications of
formative measurement. Journal of Business
Research, 61(12), 1250-1262.
Moderation
• Moderation
• Multi-group analysis
• Interaction
The Parametric Approach to MGA

p (1)  p (2)
t
n  1 n  1
(1) 2 (2) 2
1 1
* se( p )  (1) (2)
(1) 2
* se( p ) * (1)  (2) (2) 2

(n  n  2)
(1) (2)
(n  n  2) n n
Welch-Satterthwait Approach to MGA

If the standard errors are unequal, then the t-test is computed as follows:

p p (1) (2)

t
n 1(1)
n 1 (2)

(1)
* se( p )  (2) * se( p )
(1) 2 (2) 2

n n

2
 n(1)  1 n(2)  1 (2) 2 
 (1) * se( p )  (2) * se( p ) 
(1) 2

df   (1) 
n n
n 1 n (2)
 1
(1)2
* se( p (1) ) 4  (2)2 * se( p (2) ) 4
n n
Henseler’s MGA

Henseler’s MGA can be used for single-tail test, p-value smaller than 0.05 or
larger than 0.95 is significant at the 5% probability of error level for a certain
difference of group-specific path coefficients (Hair Jr et al., 2013; Henseler,
Ringle, & Sinkovics, 2009; Sarstedt, Henseler, & Ringle, 2011).
Mediation
• Sobel’s test (1982)
• Baron and Kenny (1986)
• Mediation Analysis using bootstrapping
approach
• Mediation technique (Hayes, 2009; Preacher
and Hayes, 2008)
Mediation Analysis using bootstrapping
approach

To assess how much of the direct path is absorbed, variation accounted factor is calculated
as
VAF=Ratio of Indirect Effect to Total Effect (i.e. direct effect +indirect effect): VAF=
(p12*p23) / (p13+ p12*p23).
Based on the value of VAF, following conditions of mediator effect is given by Hair, et al.
(2013, p. 224):
(i) if 0< VAF <0.2, No Mediation
(ii) (ii) if 0.2< VAF< 0.8 Partial Mediation
(iii) (iii) if VAF>0.8, Full Mediation.
Hayes Mediation Technique
• Mediation technique (Hayes, 2009; Preacher and
Hayes, 2008)
– Bootsrapping resamples the data multiple times (5000
resamples in our case) to calculate the path
coefficients.
– To test the significance of indirect effect, 95%
confidence interval is obtained though the bias
corrected percentile method.
– The indirect relationship (βindirect=0.062, 95%
CI=0.0062, 0.13) with 95% confidence interval did not
include zero showed that the mediation relationship
was statistically significant.

Potrebbero piacerti anche