MODULE

MULTIVARIATE ANALYSIS
STRUCTURAL EQUATION MODELING (SEM) WITH PARTIAL LEAST
SQUARES (PLS)
Using SmartPLS 3.0

MANAGEMENT STUDY PROGRAM

FACULTY OF BUSINESS, LAW AND SOCIAL SCIENCES
MUHAMMADIYAH UNIVERSITY SIDOARJO
2020
 TABLE OF CONTENTS

TABLE OF CONTENTS .................................................................................................................................. ii

FOREWORD................................................................................................................... iii
CHAPTER I BASIC CONCEPT OF PARTIAL LEAST SQUARE (PLS)....................................................1
1. History of SEM and PLS ......................................................................................................................1
2. Measurement Model and Structural Model ........................................................................................2
3. Steps to use SmartPLS .....................................................................................3
4. Estimate ............................................................................................................................................6
5. Goodness of Fit Evaluation..................................................................................................................7
CHAPTER II SMARTPLS SOFTWARE .................................................................................................12
2. Running SmartPLS 3.0............................................................................................................12
3. Tool Bar in SmartPLS Version 3.0
CHAPTER III ANALYSIS WITH MEDIATION EFFECT ........................................................................32
1. Steps of Analysis with Mediating Effects Using SmartPLS 3.0...........................33
2. Measurement Evaluation (Outer Model) ...........................................................................................38
3. Evaluation of the Structural Model (Inner Model). .......................................................................................
CHAPTER IV ANALYSIS WITH MODERATING EFFECTS......................................................47
1. Steps of Analysis With Moderating Effects Using SmartPLS 3.0. .......................49
2. Measurement Evaluation (Outer Model) ...........................................................................................55
3. Evaluation of the Structural Model (Inner Model). .......................................................................................

MULTIVARIATE ANALYSIS USING SMART 3.0 ii

 FOREWORD

Peace be upon you.

Praise and gratitude the author expresses to Allah SWT, for His grace and blessings.
It is a gift of the Multivariate Analysis Module Structural Equation Modeling (SEM) With Partial
Least Squares (PLS) using the SmartPLS 3.0 program can be completed. This module
is a continuation of the Multivariate Analysis module using SPSS and EViews, in addition to
The next module is the Multivariate Analysis module with SEM using the application
AMOS.
The author expresses gratitude to the Dean of FBHIS UMSIDA, Head of the Management Program.

FBHIS UMSIDA, the Management Study Program of FBHIS UMSIDA that has provided
opportunities and support for the Multivariate Analysis Course Module in the Management Study Program

this can be resolved.

The author realizes that this module is still not perfect, so criticism and suggestions
It is highly expected for further development for the writers or parties that
interested.
The author's hope is that this module can be useful for those in need.
"In the path of the truth, hasten to good deeds."

And peace be upon you

Sidoarjo, December 2020

Writer

MULTIVARIATE ANALYSIS USING SMART 3.0 iii

 CHAPTER I

BASIC CONCEPT OF PARTIAL LEAST SQUARE (PLS)

1. History of SEM and PLS

Currently, the method with regression testing is a method that is commonly used by
students, lecturers, and also researchers to conduct quantitative research. However
As research methods in the social field continue to develop, the analysis method with
regression is considered less capable of addressing research problems
raised by researchers, resulting in a Structural Equation Model for
overcoming the weaknesses present in regression.
Structural Equation Modeling or SEM is generally represented
by software AMOS, EQS LISREL, GSeCA, SmartPLS, WarpPLS, MPlus, etc. is
A very popular technique used in social sciences. The analysis technique is
is a combination of two methodological disciplines, namely the perspective of econometrics
which focuses on prediction and psychometrics that can describe
concept of a model with latent variables (variables that cannot be measured directly)
but measured through its indicators (manifest variables).
Herman Wold 1966 (in Ghozali, 2015) introduced two iterative methods, namely
the first to use the Least Square (LS) estimation method for a single model and
multi-component model and the second for canonical correlation using
The NILES algorithm (nonlinear iterative least squares) offers an estimation method.
for Principal Component Analysis (PCA) NILES calculates the principal components of
average iterations produced by simple linear regression.
In 1974, world introduced PLS publicly with the NIPALS algorithm.
(nonlinear iterative partial least squares) which is a development of the algorithm
previously known as NILES.
Partial Least Square is a powerful analytical method and is often referred to as
as a soft modeling because it eliminates the assumptions of OLS (Ordinary Least Squares)
regression, such as data must be normally distributed multivariate and no problems

MULTIVARIATE ANALYSIS USING SMART 3.0 1

 multicollinearity among exogenous variables (Wold, 1982). Social research, especially in business
and management in the fields of marketing management and human resource management
who measures the perception of the difficulty in obtaining distributed data
normal, and will have difficulty obtaining the Best Linear Unbiased regression equation
Estimate (BLUE) makes PLS one of the recommended analysis methods.
PLS can be used to explain the presence or absence of relationships between latent variables.
(prediction), PLS can also be used to confirm theories (Chin and Newsted)
1999). PLS uses an iterative algorithm consisting of a series of OLS (ordinary least squares)
square) so the issue of model identification is not a problem for the model
recursive (a model that has one-way causality) and avoiding problems for
a model that is non-recursive (a model that is reciprocal or mutual)
variables) that can be solved by covariance-based SEM. As an alternative analysis
covariance based SEM, variance based approach with PLS changes orientation
analysis of testing the causality model (the model developed based on theory) to
prediction model components (Chin and Newated 1999 in Ghozali, 2015).
2. Measurement Model and Structural Model
PLS-SEM analysis usually consists of two sub-models, namely the measurement model.
(measurement model) or commonly referred to as the outer model and the structural model
model) or commonly referred to as the inner model. The measurement model shows how
Manifest variables or observed variables represent latent variables to be measured.
While the structural model shows the strength of estimates between latent variables or
construct (Ghozali, 2015).

Figure 1. Structural Equation Model

MULTIVARIATE ANALYSIS USING SMART 3.0 2

 The latent variables formed in PLS-SEM can have reflective indicators.
or formative. The reflective indicator, often referred to as the American Mode, is an indicator that

is a manifestation of the construct and in accordance with Classical Test Theory

assuming that the variance in the measurement of latent variable scores is
the function of the true score plus the error. While formative indicators or often
Mode B is an indicator that defines characteristics or
explaining the construct.

Gambar 2. Konstruk dengan Indikator refleksif Gambar 3. Konstruk dengan Indikator formatif
(Mode B)
(Mode A)

Partial Least Square (PLS) with a variance-based approach means that the sample that
does not have to be large. Researchers can use small samples ranging from
30 - 100 samples. If using CBSEM, because it is based on covariance, then
the sample used must be large, ranging between 200 - 800 samples. The number of constructs and
the indicators used in both PLS and CBSEM are also different. PLS can use
up to 100 constructs and 1000 indicators, but in CBSEM it can only use
maximum 100 indicators.

3. Steps to use SmartPLS

a. Designing the Structural Model (inner model)
The first step taken is to design the structural model (inner model).
that is designing the relationships between latent variables in PLS based on
the research hypothesis that has been created.
b. Designing the Measurement Model (outer model)
The second step is to design the measurement model (outer model) which is to design
the relationship between latent variables and their indicators. In PLS, the design of the outer model

very important, reflexive or formative.

MULTIVARIATE ANALYSIS USING SMART 3.0 3

c. Path Diagram Construction
The third step is to construct a path diagram based on the outer design.
The result of the design of the outer model and inner model is stated.
in the form of a flowchart to make it easier to understand before analysis.
d. Convert Path Diagram to System of Equations
1) Outer Model
Outer model, which is the specification of the relationship between latent constructs and their indicators.

Outer model, also known as the outer relational measurement model,

defining the characteristics of the construct with its manifest variables. Equation
The outer model of the reflective indicator model (mode A) is:
x = ⋀x + εx
y = ⋀y η + εy
where:
x and y are manifest variables or indicators related to
exogenous construction ) and endogenous (η).
⋀xand⋀yis a loading matrix that illustrates the regression coefficients
simple that connects latent variables with their indicators.
εxand εyis the residual of the measurement error.
Meanwhile, for the outer formative model (mode B), the equation is for
The formative indicator model can be written as follows:

= ∏! x +δ!

η = ∏η + δη

where:
( ) and (η) are exogenous and endogenous latent constructs.
x and y are the manifest variables or indicators for the exogenous latent construct.
( ) and endogenous (η).
∏" and ∏# is the multiple regression coefficient for latent variables and blocks

indicator.
δxandδyis the residual of the regression.

MULTIVARIATE ANALYSIS USING SMART 3.0 4

Inner Model
Inner model, which is a specification that shows the relationships or
the strength of estimation between latent variables (structural model) or constructs

based on the substantive theory. The inner model equation can be written as
the following:
η= 0+ η +Γξ +ς
where
η is an endogenous latent construct vector.
ξ is an exogenous latent construct vector.
ς is a vector of residual variables (unexplained variance)

Basically, PLS is designed for recursive models (models that have one)
(causal path), then the relationship between exogenous latent variables and each
exogenous latent variables to each endogenous latent variable are referred to as
causal chain system that can be specified as follows:

η j= Σ i hi ηj +Σiγjb ξb +ςj

where:
dan γ is
yes jb
the path coefficient that connects the endogenous latent variables.

(η) with exogenous (ξ).

and b is a range of indices.
j is an inner residual variable.

3) Weight Relation
Weight relation, estimation of latent variable case values. Inner and outer
the model provides specifications followed by estimated weight relations in
PLS algorithm. Case values for each latent variable estimated in PLS
following:
ξb= Σkb kb xkb

η.b= Σwho who ywho

MULTIVARIATE ANALYSIS USING SMART 3.0 5

 Where:
kb and is the weight used to estimate latent variablesξb
what

andη.b .
Jadi, estimasi variabel laten adalahlinier aggregatedari indikator yang nilai
the weights were obtained through the PLS estimation procedure specified by inner
danoutermodel where η is the endogenous variable vector and ξ is the vector
exogenous variable,ς is the residual vector andΓ is a matrix
path coefficient.

4. Estimation
The estimation of parameters in PLS is obtained through a three-stage iterative process and at each

stage of producing estimates. Here are the steps for estimating PLS parameters:
a. Weight estimate
The first stage produces a weight estimate (weight estimate) wjh, used for
creating a latent variable score. The estimated weight wjh is obtained through two ways, namely
mode A and mode B. Mode A is designed to obtain weight estimates with
reflective indicator type, while mode B is designed to obtain estimates
weight with a formative indicator type.
b. Path estimate
The second stage produces estimates of the paths that connect the latent variables.
and the estimation of loading between latent variables and their indicators. This estimation is obtained
through the estimation of the inner model and outer model.

c. Mean estimates and parameter location (constant)

d. At this stage, the estimation is based on the original data matrix and the results of the weight estimation and

the path coefficient in the second stage, the goal is to calculate the average and location
parameter.

MULTIVARIATE ANALYSIS USING SMART 3.0 6

5. Goodness of Fit Evaluation

a. Model Measurement Evaluation (Outer Model)

Convergent Validity
Convergent validity measures the magnitude of the correlation between constructs and variables.

late. In the evaluation of convergent validity from the individual item examination

Reliability can be seen from the standardized loading factor. Standardize loading
the factor represents the magnitude of the correlation between each measurement item (indicator)

with its construction. Correlation can be said to be valid if the loading factor has
> 0.7.
2) Discriminant Validity
Comparing the discriminant validity and the square root of average variance
extracted(AVE). The measurement model is assessed based on the cross-loading measurement.

with constructs. If the correlation of the construct with each of its indicators is greater than

on other construct sizes, the latent construct predicts its indicators more
better than other constructs. If the AVE value is higher than the correlation value
between constructs, good discriminant validity is achieved.
It is recommended if the AVE value is greater than 0.5 (Tasha Hoover, 2005 in
Sofyan Yamin, 2009). Here is the formula for calculating AVE:

(∑λ2i) var F
AVE=
(∑λ2i) + var F+ ∑θii

where:

% is factor loading
F is the variance factor
θiiis heror variance

In addition, there is another way to test discriminant validity, which is by

comparing the square root of AVE for each construct with the correlation values
The good discriminant validity is indicated by the root.

MULTIVARIATE ANALYSIS USING SMART 3.0 7

 The square of AVE for each construct is greater than the correlation between constructs in
model (Fornell and Larcker, 1981).

3) Composite Reliability
Reliability testing is conducted to prove accuracy, consistency, and precision.
instrument in measuring constructs. In PLS-SEM using
SmartPLS 3.0, to measure the reliability of a construct with indicators.
Reflective can be done with two methods, namely: with Cronbach’s Alpha and
Composite Reliability is also known as Dillon Goldstein's.
The Rule of Thumbyang is usually used to assess the reliability of a construct.
nilaiComposite Reliabilityharus lebih besar dari 0,7 untuk penelitian yang bersifat
confirmatory and 0.6 - 0.7 for exploratory research.

(∑ % )& var F
pc=
(∑ % )& var F + ∑θii

where

% is factor loading
F is the factor variance
θiithe hero of variance

In PLS, the reliability test is strengthened by the presence of Cronbach Alpha.

The consistency of each answer is tested. Cronbach's alpha is considered good if α > 0.5
and it is said to be sufficient when α > 0.3. According to (Ghozal, 2015) Rule of Thumb

is usually used to assess the reliability of a construct, namely Cronbach's alpha

Alpha is greater than 0.7 for confirmatory research and
0.6 - 0.7 for exploratory research. Summary of Rule of Thumbs
The reliability test of the construct with reflective indicators can be seen in the table below.

MULTIVARIATE ANALYSIS USING SMART 3.0 8

 Table 1. Summary of Rule of Thumb Model Evaluation (Mode A)
Validity and Reliability Parameter Rule of Thumbs

Convergent Validity Loading Factor • 0.70 for Confirmatory

Research
• 0.60 for Exploratory
Research
• 0.50 for Exploratory
Research(Chin, 1998)
Average Variance 0.50 for Confirmatory or
Extracted (AVE) Exploratory Research
Communality 0.50 for Confirmatory or
Exploratory Research
Discriminant Validity Cross Loading 0.70 for each variable
Square root of AVE and Square root of AVE > Correlation among
Correlation between Constructs Late Construction
Late
Reliability Cronbach's Alpha • 0.70 for Confirmatory
Research
• 0.60 is still acceptable for
Exploratory
Research
Composite Reliability • 0.70 for Confirmatory Research
• 0.60 - 0.70 is still acceptable
for exploratory
Research
Source: Ghozali (2015:76)

If the construct is in a formative format (mode B), then evaluate the measurement model.

done by looking at the significance of its weight. So validity testing and testing
reliability is not required. To obtain significance, weights must be through
resampling procedures (jackknifing or bootstrapping). In addition, the multicollinearity test
For absolute formative construction, it is essential to calculate the value of variance.

Inflation Factor (VIF) and its counterpart Tolerance. If a significance value is obtained

if the T-statistic weight > 1.96 then it can be concluded that the construction indicator (mode

B) is valid. For VIF it is recommended <10 or <5 and tolerance > 0.10
or > 0.20. Summary of the rule of thumb for evaluating measurement models (mode B) can be

as seen in the following table:

Multivariate Analysis Using SMART 3.0 9

 Table 2. Summary of Rules of Thumb for Model Measurement Evaluation (Mode b)

Criteria Rule of Thumbs

Significance of Weight • 1.65 (significance level 10%)
• 1.96 (significance level 5%)
• 2.58 (significance level 1%)

Multicollinearity • VIF <10 or <5

• Tolerance >0,1 atau >0,2
Source: Ghozali (2015:78)

b. Evaluation of Measurement Model (Inner Model)

In assessing the structural model with PLS, it is evaluated by looking at the percentage.
The explained variance is seen through the R-Square for the endogenous latent construct.
as a predictive strength of the structural model. Its interpretation is the same as
interpretation in OLS regression. Changes in the R-Squared value can be used for
explaining the influence of certain exogenous latent variables on endogenous latent variables
does it have a substantive influence. R-Square values of 0.75, 0.50, and 0.25
It can be concluded that the models are strong, moderate, and weak. The results of the PLSR-Square

represents the amount of variance of the constructs explained by the model.

Besides looking at the R square value, it can also be done by looking at the Q value.2predictive

relevance. This technique can represent the synthesis of cross validation and
fitting function. Value Q2>0 has predictive relevance while if
2
value of Q The model has less predictive relevance. Q value2predictive
relevance 0.02, 0.15, and 0.35 means being weak, moderate, and strong.
(Ghozali and Latan, 2015:81)
Table 3. Summary of Rule of Thumbs for Structural Model Evaluation
Criteria Rule of Thumbs
R-Square According to (Chin 1998)
• 0.67 indicates a strong model
• 0.33 indicates a moderate model
• 0.19 indicates a weak model
According to (Hair et al. 2011)
• 0.75 indicates a strong model
• 0.50 indicates a moderate model
• 0.25 shows a small model
According to (Widarjono, 2015)
• ≥ 0.75 good model

MULTIVARIATE ANALYSIS USING SMART 3.0 10

Effect size f2 • 0.02 small
• 0.15 currently
• 0.35 large
Q2predictive relevance • Q20 model has predictive relevance
• Q2< 0 model has less predictive relevance
q 2predictive relevance • 0.02 weak
• 0.15 moderate
• 0.35 strong
Significance (two tailed) t-value 1.65 (signifikan level = 10%, 1.96 (significance level = 5%),
and 2.58 (significance level = 1)
Source: Ghozali (2015:81)

Hypothesis Testing (resampling bootstrapping)

Hypothesis testing (β, γ, and λ) was conducted using the bootstrap resampling method.
The statistical test used is the t statistic or t test, with statistical hypotheses.
as follows:
1) Statistical hypothesis for the outer model:
H0 : λi = 0
H1: λi ≠ 0
2) Statistical hypothesis for the inner model: exogenous latent variables affecting endogenous:

H0γi = 0
H1γi ≠ 0
The application of resampling methods allows for the validity of distributed data.
free (distribution free), does not require normal distribution assumptions, and does not
requires a large sample (minimum sample of 30).

MULTIVARIATE ANALYSIS USING SMART 3.0 11

 CHAPTER II

SOFTWARE SMARTPLS

Software Structural Equation Modeling (SEM) to analyze component based PLS first
Kali was developed by Jan-Bernd Lohmoller (1984, 1987, 1989) under DOS and is called LVPLS.
version 1.8 (Latent Variable Partial Least Squares). Then further developed by Wynne
In China (1998, 1999, 2001) became under windows with graphical interface and additional
improvement of validation techniques by incorporating bootstrapping and jackknifing. Software that

developed by Chin is named PLS GRAPH version 3.0. still a beta version, version
students can be downloaded for free. The PLS SmartPLS 3.0 software developed in
The University of Hamburg in Germany can be downloaded for free [Link](Ghozali, 2015).
Using the free student version has its limitations because it is not the full version.
among others: maximum sample of 100, output cannot be saved, copied, and cannot be exported
into Excel format.

2. Running SmartPLS 3.0

Before starting to run the SmartPLS application, first prepare the research data.
If the research uses a questionnaire or survey, it must be tabulated into Excel.
However, if the questionnaire is distributed using Google Forms, you need to download it.
in Excel format and then rearranged so that the data is ready for analysis. To run
smartPLS 3.0 data must be in CSV (Comma delimited) or Comma Separated Value format
(CSV), so the data that has been prepared in Excel format then is saved as CSV.
can be seen in the image below:

MULTIVARIATE ANALYSIS USING SMART 3.0 12

a. Open the SmartPLS Application.
The image below shows the Main Window of SmartPLS Version 3.0

Window place
The project that was created
that will
Window place
worked on
Structural model in
image

Create New Project

StepFileà Newatau click in the file name, enter the name of the file that will be

helloà Ok.

MULTIVARIATE ANALYSIS USING SMART 3.0 13

c. Import data by clicking Double click to import data.

Search for the Marketing Research [Link] (in the directory where the file is saved) and then click

OpenFile.

MULTIVARIATE ANALYSIS USING SMART 3.0 14

The window appears as shown below. The Green Marketing project contains files.
Green Marking (100 records).

Click Marketing Research and the worksheet below will appear:

From the display above, it appears that the indicators and working area for

describing the model and tools to draw latent variables and

the indicator.
MULTIVARIATE ANALYSIS USING SMART 3.0 15
3. Tool Bar on SmartPLS Version 3.0

Modelling Mode
Three types of modeling that can be used for drawing or designing and
changing the model, namely: selection mode, drawing mode, and connection mode, more
the details can be seen below:

The select function to select and move objects

on the drawing board. Press (SHIFT) key for multiple
selection objects. Objects that are on the drawing board
can be edited by double-clicking the left mouse button.

The variable is used to create or

adding latent variables in drawing mode,
by clicking with the left mouse button on
drawing area to create new latent variables with
standard label (latent variable 1, latent variable 2, etc.
following the number of latent variables created). For
change the variable's latent name by right-clicking
on the variable layer and select rename then click OK.

Connection Mode is used to establish a relationship

between latent variables, by pressing the mouse button
drag and pull to the target latent variable that will be given
relationship.

MULTIVARIATE ANALYSIS USING SMART 3.0 16

 • Delete: used to delete variables
late or indicator.
• Rename: used to change the name or
label variable.
• Switch between formative / reflective: used
to change indicators into a formative format
reflective or vice versa
• Hide indicators of selected construct:
used to hide variables
indicator of the model (used if the model
complex with too many indicators
so that the drawing area is insufficient)
• Align: used to place indicators,
or change the layout of the indicator whether it is above
(top), dibawah (bottom), di kanan (right), di kiri
(left), from the latent variables.

Drawing the Model

How to create a variable latent with indicators, first create
latent variable with click drawing mode When it is on, then click on
area drawing, highlight variable indicators from the data area of variable indicators, then drag
then drag and drop with the mouse to the latent variable according to the variables that have been created.

Create latent variables and provide relationships as below:

MULTIVARIATE ANALYSIS USING SMART 3.0 17

Rename Latent Variable
Double click on latent variable 1 and rename it to Buying Interest, then double click again.
change the time of variable 2 and rename it to Purchase Decision.

So the result is as follows:

MULTIVARIATE ANALYSIS USING SMART 3.0 18

Drag each construct indicator into the Latent Variable so that it is visible.
as follows:

The model is ready to be estimated by clicking the main menu and then selecting calculate.à PLS
algorithm

MULTIVARIATE ANALYSIS USING SMART 3.0 19

A window like the one below appears:
Algorithm Setting
• Weight Shceme : path
• Maximum Interation : 300

Click Start calculation and the output will appear in the form of a path diagram on the drawing.

the area as follows:

From the data diagram, it can be seen that there is one indicator that has a loading value.
factors with outer loading below 0.70 or not valid, namely KP2, which is
indicator from the Purchasing Decision variable, because it does not meet the criteria
convergent validity means that the KP2 indicator is removed from the model, then it is calculated

repeat. So that the result is like below:

MULTIVARIATE ANALYSIS USING SMART 3.0 20

After recalculating, all loading factors or outer loadings above 0.70 are all,
or none is below 0.70.

• If you want to save the diagram path, then click file.à export as image to file.
another option is to click export in the theme and select PNG.
• To view the analysis results click Open this report or by clicking PLS
Algorithm:

• To save the overall analysis results, use save as, and if you want to
save in excel format then click PLS Algorithm then click export to excel.

MULTIVARIATE ANALYSIS USING SMART 3.0 21

 • So the output display in the form of excel as below, you can take advantage of
navigation to view the Report value of your analysis results.

Interpretation
A. Evaluation of Measurement (Outer Model)

1) Convergent Validity
The analysis results show that there is one indicator with a loading factor value below
0.70 or not valid which is KP2 that serves as an indicator of the Decision variable
Pembelian, Oleh karena tidak memenuhi kriteriaconvergent validity maka
The KP2 indicator is removed from the model.

Table 4. Convergent Validity (AVE)

Average Variance Description
Extracted (AVE)
Interest in Buying 0.661 Valid
Purchase Decision 0.810 Valid
Source: Processing output with SmartPLS 3.0.

MULTIVARIATE ANALYSIS USING SMART 3.0 22

 The AVE output results from Smart PLS in table 4, produced by all
the construct has a value > 0.50, thus each construct has met the requirements
requirements.

2) Discriminant Validity
The level of construct validity can be measured in the second way, which is
using discriminant validity, presented in the table …… below:

Table 5. Results of Discriminant Validity Test (Fornell-Larcker Criterion)

Decision Interest
Purchase Buying
Purchase Decision 0.900
Interest in Buying 0.518 0.813

Source: Processing output with SmartPLS 3.0.

From table 5. it can be seen that the values on the diagonal that are bolded are the roots

the square of AVE and the values below it are correlations between constructs, the root value

the square of AVE is greater than the correlation between constructs, it can be concluded

that the model is valid because it meets discriminant validity.

3) Composite Reliability
Composite reliability and Cronbach's alpha for each construct
presented in table 6. below:

Table 6. Results of Composite Reliability and Cronbach's Alpha Test

Cronbach's Composite
Description
Alpha Reliability
Reliable
Purchase Decision 0.766 0.895

Interest in Buying 0.741 0.853 Reliable

Source: Processing output with SmartPLS 3.0.

MULTIVARIATE ANALYSIS USING SMART 3.0 23

 The Cronbach's Alpha values produced for all constructs are very good because the value

above >0.70 so it can be concluded that the construction indicator is reliable

or it can be said to meet the reliability test. The Composite Reliability value
all constructs are produced very well because above >0.70 so that it can
It is concluded that all construct indicators are reliable or meet the test.
reliability.

B. Evaluation of the Structural Model (Inner Model)

Estimation has met the Outer Model criteria, the next step is to test the model.
structural (Inner model). According to (Ghozali, 2011), the purpose of evaluating the structural model

(Inner model) to predict the relationships between latent variables. In evaluating the inner

model with SmartPLS seen from the R-Square value ( & for each latent variable
endogenous. Coefficient of determination R-square ( & indicates how large the variable is
Exogenous explains its endogenous variables. The R-Square value ( & is zero until
with one.

1. R-Square Test ( )
Table 7. R-Square Test Results( &)

R Square Adjusted R Square

Purchase Decision 0.268 0.260

Source: Processing output with SmartPLS 3.0.

From the table above, the coefficient of determination (R can be seen2 0.268 for

The purchasing decision variable means that the buying interest variable explains the variance.

the purchasing decision variable is 26.8%.

2. Path Coefficient
To obtain the Path Coefficient value or t-statistic, it must first be
estimate bootstrapping by clicking the main menu and then selecting calculateà
bootstrapping.

MULTIVARIATE ANALYSIS USING SMART 3.0 24

 Bootstrapping Setting
• Subsample: 500
• Sign Changes : No Sign Changes

Click Start calculation and the output will appear in the form of a path diagram on the drawing.

the area as follows

MULTIVARIATE ANALYSIS USING SMART 3.0 25

• If you want to save the diagram path, click fileà export as image to file. Ada
another event is by clicking export in the theme then select PNG.
• Untuk melihat hasil analisis klikOpen this reportatau dengan klikBootstrapping:

• To save the overall analysis results, use save as, and if

to save in Excel format, click bootstrapping then click export to
excel.

MULTIVARIATE ANALYSIS USING SMART 3.0 26

• So the output display in the form of excel as below, you can
Make use of the navigation to view the value of the report from your analysis.

• After obtaining the coefficient path values, the next step is

interpreting, as below:

Table 8. Path Coefficient

Original Sample Standard
T Statistics
Sample Mean Deviation P Values
(|O/STDEV|)
(O) (M) (STDEV)
Interest in Buying
0.518 0,523 0.080 6,496 0.000
Purchase Decision
Source: Output processing with SmartPLS 3.0.

The coefficient value of 0.518 indicates a positive influence between

Buying Interest towards Purchase Decision. The higher the buying interest

the higher the purchase decision, with a t-statistic value of 6.496

significant (the significant t-table at 5% = 1.96), because the t-statistic value is greater than t

Table 1.96.

MULTIVARIATE ANALYSIS USING SMART 3.0 27

3. Predictive Relevance Test

To obtain the Q² value in the Predictive Relevance Test, first...

must be estimated by Blindfolding by clicking the main menu and then selecting calculateà

Blindfolding.

Blindfolding Setting
Omission Distance : 7

MULTIVARIATE ANALYSIS USING SMART 3.0 28

Click Start calculation and an output will appear in the form of a path diagram on the drawing.

area as follows:

• If you want to save the diagram path, then click file.à export as image to [Link]
another event is by clicking export located in the theme then select PNG.
• To see the analysis results click Open this report or by clicking Blindfolding:

• To save the results of the analysis as a whole, then save as, and if
to save in Excel format, click Blindfolding then click export to
excel.

MULTIVARIATE ANALYSIS USING SMART 3.0 29

 • So the output display in the form of excel as below, you can
utilize the navigation to view the value of the report from your analysis.

• After obtaining the Q² value, the next step is to interpret it.

as below:
Table 9. Predictive Relevance Test
SSO SSE Q² (=1-SSE/SSO)
Purchase Decision 200,000 158,605 0.207
Interest in Buying 300,000 300,000
Source: Output processing with SmartPLS 3.0.

Based on the calculation of Predictive Relevance(Q& ) in the table above indicates

The value of 0.207 indicates that the model has a relevant predictive value.

c) Model Fit Evaluation

The model fit evaluation in this study uses the Normal Fit Index testing model.
(NFI).
Table 10. Model Fit Evaluation Test
Saturated Model Estimated Model
NFI 0.668 0,668

Source: Processing output with SmartPLS 3.0.

MULTIVARIATE ANALYSIS USING SMART 3.0 30

The research results indicate that the values of the saturated model (measurement) fit and the value

The estimate model (structural model) has the same value. From the results, it shows that
It can be concluded that the model in the study has a normal fit index (NFI) value.
which shows that the model in this study is 66.8% better than the model
null model.

The NFI value can be seen in the PLS algorithm Model Fit Output.

MULTIVARIATE ANALYSIS USING SMART 3.0 31

 CHAPTER III

ANALYSIS WITH MEDIATING EFFECT

Structural Equation Modeling (SEM) with mediating effects is the relationship between constructs.
exogenous with endogenous through intermediate or linking variables. According to Sugiyono (2016),
Intervening or mediating variables are variables that theoretically influence relationships.
between the independent variable (exogenous) and the dependent variable (endogenous) forms a relationship

which is indirect and cannot be observed or measured. The influence of exogenous latent variables
on endogenous latent variables can be directly or also through intermediate variables or
mediation. According to Ghozali (2015), testing the mediation effect uses SmartPLS 3.0
using the procedure developed by Baron and Kenny 1986. There are three stages
model for testing mediation effects (Baron and Kenny 1986), the stages include:
a. The first model tests the influence of exogenous variables (X) on endogenous variables (Y)
and must be significant at t-statistic >1.96.
b. The second model examines the effect of exogenous variables (X) on the mediating variable (M)

and must be significant on the t-statistic >1.96.

c. The third model tests the simultaneous effect of exogenous variables (X) and variables
mediation (M) towards the endogenous variable (Y). in the final testing phase is expected
the influence of the exogenous variable (X) on the endogenous variable (Y) is not significantly moderate

the influence of the mediating variable (M) on the endogenous variable (Y) must be significant at t-

>1.96 statistics.
To better understand SEM mediation effects, let's practice with a research model.
as follows:

H1 H4

H3

H2 H5

Figure 4. Green Marketing Research Model

MULTIVARIATE ANALYSIS USING SMART 3.0 32

1. Steps for Analysis with Mediation Effects Using SmartPLS 3.0.

A. Open the SmartPLS 3.0 application by double-clicking, the display window will look like below
this:

[Link] New Project, Steps Fileà New click ,

in the file name, enter the name of the file to be processedà Ok.

MULTIVARIATE ANALYSIS USING SMART 3.0 33

C. Import data by clicking Double click to import data.

Find the Marketing Research [Link] file (in the directory where the file is stored) and then click Open.

File.

MULTIVARIATE ANALYSIS USING SMART 3.0 34

The window display appears as shown below. The Green Marketing project contains the Green file.
Marking (100 records).

Click Marketing Research and the worksheet below will appear:

[Link] create a latent variable with indicators, first create the latent variable.
with drawing mode click When it is on, then click in the drawing area, highlight.
indicator variable from the data area of the indicator variable, then drag and drop with the mouse to
latent variables according to the variables that have been created. Create latent variables, rename
variable, give a relationship with click , and enter the indicator. Thus forming
as below:

MULTIVARIATE ANALYSIS USING SMART 3.0 35

[Link] siap di estimasi dengan cara klik menu utama lalu pilihcalculateà PLEASE
algorithm

oWeight Scheme: path

Maximum Iteration: 300

MULTIVARIATE ANALYSIS USING SMART 3.0 36

I. Click Start calculation to display the output in the form of a path diagram in the drawing area.

as follows:

From the data diagram, it can be seen that there is one indicator that has a loading factor value.
below 0.70 or not valid, which is GM3, an indicator of the green variable
marketing, and KP2 which is an indicator of the Purchase Decision variable, By
because it does not meet the criteria of convergent validity, the indicators GM3 and KP2

removed from the model. More details on the invalid indicators can be seen on the outer.
loading the following red value:

MULTIVARIATE ANALYSIS USING SMART 3.0 37

 [Link] GM3 and KP2 are removed from the model by right-clicking on the indicator.
then delete, the next step is to recalculate. So the result is as follows:

After recalculating all loading factors or outer loadings above 0.70, or not
There are values below 0.70. It can be seen from the outer loading output that the values are no longer

ada yang berwarna merah.

2. Measurement Evaluation (Outer Model)

a. Convergent Validity
The analysis results show that there are two indicators with loading factor values below 0.70 or
invalid is GM3 which is an indicator of the green marketing variable, and KP2

MULTIVARIATE ANALYSIS USING SMART 3.0 38

 which is an indicator of the Purchase Decision variable, because it does not
to meet the criteria of convergent validity, the indicators GM3 and KP2 are removed from
model or dropped from the analysis.

Table 11. Convergent Validity (AVE)

Average Description
Variance
Extracted
HAIL
Green Marketing 0.700 Valid
Purchase Decision 0.810 Valid
Environment and Health 0,580 Valid
Interest in Purchasing 0.661 Valid
Source: Output processing with SmartPLS 3.0.

The output results of AVE from Smart PLS in the table above produced by all constructs.
having a value > 0.50, so each construct has met the requirements
convergent validity.

b. Discriminant Validity
The level of construct validity can be measured in a second way, namely by using
discriminant validity, presented in the following table 12:

Table 12. Results of Discriminant Validity Test (Fornell-Larcker Criterion)

Green Decision Environment and Interest in Buying

Marketing Purchase Health
Green Marketing 0.837
Decision
-0.195 0.900
Purchase
Environment and
-0.172 0.565 0,762
Health
Interest in Buying -0.104 0.510 0.581 0.813
Source: Output processing with SmartPLS 3.0.

From table 12, it can be seen that the thick printed diagonal values are square roots.
AVE and its values below are the correlations between constructs, the square root value of AVE is higher

greater than the correlation between constructs, it can be concluded that the model is valid
because it fulfills the discriminant validity.

MULTIVARIATE ANALYSIS USING SMART 3.0 39

 There are other methods that can be used to test discriminant validity, namely
by looking at the cross loading table.

From the table above, it can be seen that the value of the loading factor of each item against
Its structure is larger than its cross loading value. Therefore, it can be concluded that
there are no problems with discriminant validity.

c. Composite Reliability
The composite reliability and Cronbach's alpha for each construct are presented.
in the following table:

Table 13. Results of Composite Reliability and Cronbach's Alpha Test

Cronbach's Composite
Description
Alpha Reliability
Green Marketing 0.892 0.920 Reliable
Purchase Decision 0.766 0.895 Reliable
Environment and
0,763 0.847 Reliable
Health
Interest in Buying 0.741 0.854 Reliable
Source: Processing output with SmartPLS 3.0.

MULTIVARIATE ANALYSIS USING SMART 3.0 40

 The Cronbach's Alpha values produced for all constructs are very good because the values are above >0.70.

it can be concluded that the construction indicator is reliable or can be said to be

fulfill the reliability test. The Composite Reliability value produced by all constructs is very
good because above >0.70 so it can be concluded that all construct indicators are
reliable or meets reliability testing.

3. Evaluation of the Structural Model (Inner Model).

a. R-Square Test ( )
Table 14. Results of R-Square Test( & )
R Square Adjusted R Square
Purchase Decision 0,379 0.359
Interest in Buying 0.338 0.324
Source: Output processing with SmartPLS 3.0. From the table above, the value can be seen.

coefficient of determination (R2 0.379 for the purchase decision variable, which means

green marketing variables, environment and health, as well as buying interest explain

The variance of the purchase decision variable is 37.9%, while the green marketing variable,

environment and health, explaining the variance of the Buying Interest variable of 33.8%.

b. Path Coefficient
To obtain the Path Coefficient value or t-statistic, it must first be
bootstrapping estimation by clicking the main menu and then selecting calculateà bootstrapping.

MULTIVARIATE ANALYSIS USING SMART 3.0 41

 Bootstrapping Setting
• Subsample: 500
• Sign Changes : No Sign Changes

Click Start calculation to display the output in the form of a path diagram on the drawing.

area as follows

• After obtaining the coefficient path values, then open this report choose path
the next step is to interpret, as follows:

MULTIVARIATE ANALYSIS USING SMART 3.0 42

 Table 15. Path Coefficient
Original Sample Standard T Statistics P Values
Sample Mean Deviation (|O/STDEV|)
(O) (M) STDEV
Green Marketing -0.100 -0.102 0.096 1,037 0.300
Purchase Decision
Green Marketing -> Minat -0.004 -0.011 0.097 0.046 0.963
Buy
Environment and Health 0.389 0.397 0.098 3,989 0.000
Purchase Decision
Environment and Health 0.581 0.582 0.082 7,123 0.000
Purchase Interest
Minat Membeli -> Keputusan 0.273 0.267 0.115 2,379 0.018
Purchase
Source: Output processing with SmartPLS 3.0.

From the table above, it can be seen that: the environment and health have a positive influence.

significant to the purchasing decision with a t-statistic value > 1.96, the environment and

health has a significant positive effect on purchase interest with a t-statistic value

1.96, buying interest has a significant positive influence on purchasing decisions

with a t-statistic value >1.96. However, there are two that have no effect, namely: Green

marketing towards buying decisions and green marketing towards purchase intent

because the t-statistic value < 1.96 and p-value > 0.05.

c. Indirect Effect or Mediation Test

Table 16. Test of Indirect Effect or Mediation

Original Sample Standard T Statistics P Values

Sample Mean Deviation
Green Marketing ->
-0.001 -0.007 0.030 0.041 0.967
Purchase Decision
Green Marketing -> Minat
Buy
Environment and Health -
0.159 0.154 0.069 2,298 0.022
Purchase Decision
Environment and Health
Interested in Buying
Intent to Purchase
Purchase Decision
Source: Processing output with SmartPLS 3.0.

MULTIVARIATE ANALYSIS USING SMART 3.0 43

 From the table above, it can be concluded that there is an indirect influence, namely: Environment

and Health -> Purchase Interest -> Purchase Decision of 0.159 with t value
2.298 and significant at 5%

In addition, Green Marketing -> Purchase Intention -> Purchase Decision is -0.001
with a t value of 0.041 and significance of 0.967, because significance > 0.05, it can be
it is concluded that there is no mediating relationship.

d. Predictive Relevance Test

To obtain the Q² value in the Predictive Relevance Test, it must be first ...
Estimation by blindfolding by clicking the main menu and then selecting calculate.à Blindfolding.

Blindfolding Setting
Omission Distance : 7

MULTIVARIATE ANALYSIS USING SMART 3.0 44

 Click Start calculation and the output will appear in the form of a path diagram on the drawing.

the area as follows:

• After obtaining the Q² value, the next step is to interpret it.

as follows:
Table 17. Predictive Relevance Test
SSO SSE Q² (=1-SSE/SSO)
Green Marketing 500,000 500,000
Purchase Decision 200,000 143,274 0.284
Environment and Health 400,000 400,000
Interest in Buying 300,000 238,808 0.204
Source: Processing output with SmartPLS 3.0.

Based on the calculation of Predictive Relevance(Q& ) in the table above indicates the value
0.284 and 0.204, it can be concluded that the model has relevant predictive value.

e. Model Fit Evaluation

The model fit evaluation in this study uses the Normal Fit Index testing model.
(NFI).
Table 18. Model Fit Evaluation Test
Saturated Model Estimated Model
NFI 0.685 0.685

Source: Processing output with SmartPLS 3.0.

MULTIVARIATE ANALYSIS USING SMART 3.0 45

The research results show that the value of the saturated model (measurement) fit and the value

the estimate model (structural model) has the same value. From these results,
it can be concluded that the model in the research has a normal fit index (NFI) value
which shows that the model in this study is 68.5% better than the model
null model.

Explanation: The NFI value can be seen in the PLS algorithm Model Fit Output.

MULTIVARIATE ANALYSIS USING SMART 3.0 46

 CHAPTER IV

ANALYSIS WITH MODERATING EFFECT

Structural Equation Modeling (SEM) with moderating effects shows variable interactions.
exogenous with moderating variables in influencing endogenous variables (Baron and Kenny)
1986 in Henseler and Fassott 2010). The moderator variable is a variable that influences
(strengthening and weakening) the relationship between exogenous and endogenous (Sugiyono, 2015).

According to Baron and Kenney (1986) in (Hartono and Abdillah, 2014), testing moderating effects
can only be done if the main effect of the exogenous variable on the endogenous variable is
significant. If not met, then the moderation effect testing does not need to be continued because
the result will also not be significant. This assumption also applies to testing on the model that
there is a mediating variable within it. According to Ghozali (2015) Moderated Regression Analysis
(MRA) is a common method that can be used in multiple linear regression analysis with
introducing a third variable in the form of the multiplication of two exogenous variables as a variable

moderating. This will create a non-linear relationship leading to measurement errors and
the estimation coefficient of MRA when using latent variables becomes inconsistent and biased. For
To address the aforementioned issue, one can use the Structural Equation Modeling (SEM) model, because SEM

should correct measurement errors by incorporating the effect of interactions.

the model we created.
There are four ways to conduct a moderation effect test using SmartPLS 3.0.
Is the exogenous construct and moderator reflective or formative? The methods include the following:
a. Product Indicator Approach
If the exogenous and moderator constructs are reflective, then the appropriate method
using the Product Indicator Approach, by multiplying
indicator of exogenous variables with moderators to form interaction constructs.
b. Two-Stage Approach
Digukan if the exogenous construct or moderator has a formative type. In the following way:
1) Testing the main effects of the PLS model to obtain latent variable scores. Scores
latent variables are calculated and stored for further analysis.

MULTIVARIATE ANALYSIS USING SMART 3.0 47

 2) Create interactions by multiplying exogenous latent variable scores and moderators.
then regressed as an interaction term variable against the endogenous variable.
c. The Gybrid Approach
To test the non-linear PLS model. This approach considers the model in the form
quadratic and non-linear relationships between latent variables (heseler et al. 2012)
[Link] Orthogonalizing Approach
It is an expansion of residual centering for moderated multiple regression. Residual
centering using the two-stage OLS procedure.

In order to better understand SEM moderation effects, let's practice with a research model.
which has been modified as follows:

Figure 5. Human Resource Research Model

MULTIVARIATE ANALYSIS USING SMART 3.0 48

Steps of Analysis with Moderating Effects Using SmartPLS 3.0.

A. Open the SmartPLS 3.0 application by double-clicking, the display window will appear as below.
this:

Creating a new project, the steps are Fileà Newatau click ,

in the file name, enter the name of the file to be processedà Ok.

MULTIVARIATE ANALYSIS USING SMART 3.0 49

C. Import data by clicking Double click to import data.

[Link] the data Moderating [Link] (in the directory where the file is saved) and then click
OpenFile.

MULTIVARIATE ANALYSIS USING SMART 3.0 50

The window display looks like the one below. The SmartPLS project with moderation contains
file Moderating SmartPLS (141 records).

[Link] with Moderation will display the worksheet below:

To create a latent variable with indicators, one must first create the latent variable.
with drawing mode click When it's turned on, then click in the drawing area, highlight
indicator variable from the data area of the indicator variable, then drag and drop with the mouse to

latent variables according to the created variables. Create latent variables, rename
Variable, establish a connection with ClickConnect, and input indicators. Thus formed
as below:

MULTIVARIATE ANALYSIS USING SMART 3.0 51

Creating an interaction variable between motivation and employee performance as a variable
moderator by selecting employee performance construct then right-click the mouse and create
interaction effect or choose moderating effect, on the moderator variable content Style
leadership, on the independent variable of motivation content, and on the calculation of Product content

Indicator because all indicators are reflexive. Adjust as below:

I. Thus, the results of the moderating effect have been formed as follows:

MULTIVARIATE ANALYSIS USING SMART 3.0 52

The model is ready to be estimated by clicking the main menu and then selecting calculate.à PLS
algorithm

oWeight Shceme : path

Maximum Iteration: 300

MULTIVARIATE ANALYSIS USING SMART 3.0 53

[Link] Start calculation will produce output in the form of a path diagram in the drawing area
as follows:

From the data diagram, it can be seen that many indicators have outer loading values.
below 0.70 or not valid or does not meet the criteria of convergent validity then
indicators - invalid indicators are removed from the model. More details on the indicators
invalid can be seen on the outer loading value indicated in red below:

MULTIVARIATE ANALYSIS USING SMART 3.0 54

 L. How to remove invalid indicators from the model by right-clicking on
then delete the indicator, the next step is to recalculate. So that the result is like
below:

After recalculating all loading factors or outer loadings above 0.70, or not
There are those below 0.70. It can be seen from the outer loading output that the values are no longer

there is something red.

2. Measurement Evaluation (Outer Model)

d. Convergent Validity
The analysis results show that there are two indicators with loading factor values below 0.70 or
not valid, namely GM3 which is an indicator of the green marketing variable, and KP2
which is an indicator of the Purchase Decision variable, because it does not
to meet the criteria of convergent validity, the indicators GM3 and KP2 are excluded from
model or dropped from analysis.

MULTIVARIATE ANALYSIS USING SMART 3.0 55

 Table 19. Convergent Validity (AVE)
Average Variance Description
Extracted (AVE)
Leadership Style 0.562 Valid
Employee Performance 0.577 Valid
Moderating Effect 1 0.602 Valid
Motivation 0.711 Valid
Career Development 0.647 Valid
Extrinsic Reward 0.662 Valid
Intrinsic Reward 0.646 Valid
Source: Processing output with SmartPLS 3.0.

The output results of AVE from Smart PLS in the table above produced by all constructs.
having a value > 0.50, so each construct has met the requirements
convergent validity.

e. Discriminant Validity
The level of construct validity can be measured in the second way, which is by using
discriminant validity, presented in the table ...... below:

Table 20. Results of the Discriminant Validity Test (Fornell-Larcker Criterion)

Gaya Performance Moderating Development Reward Reward

Motivation
Leadership Employee Effect 1 Career Extrinsic Intrinsic
Gaya
0,749
Leadership
Performance
0.625 0.759
Employee
Moderating
-0,314 -0.164 0.776
Effect 1
Motivation 0,339 0.502 -0.341 0.843
Development
0.362 0.492 -0.090 0.389 0.804
Career
Reward
0.291 0.352 -0.096 0.269 0.582 0,814
Extrinsic
Reward
0.288 0.436 -0.116 0.398 0.529 0.324 0.803
Intrinsic
Source: Output processing with SmartPLS 3.0.

From the table above, it can be seen that the bolded diagonal values are the roots.
The AVE square and the values below it are correlations between constructs, the square root value.

AVE is greater than the correlation between constructs, it can be concluded that the model
valid because it meets the discriminant validity.

MULTIVARIATE ANALYSIS USING SMART 3.0 56

 f. Composite Reliability
The composite reliability and Cronbach's alpha for each construct are presented.
in the following table:

Table 21. Results of Composite Reliability and Cronbach's Alpha Test

Cronbach's
Composite Reliability
Alpha
Leadership Style 0.740 0.837
Employee Performance 0.633 0,803
Moderating Effect 1 0.917 0.923
Motivation 0.594 0.831
Career Development 0.818 0.880
Extrinsic Reward 0.503 0.795
Intrinsic Reward 0.453 0.784

Source: Processing output with SmartPLS 3.0.

The Cronbach's Alpha values produced for several constructs are not good because of
some are <0.70 and the rest is very because the Cronbach's Alpha values are >0.70. so
to improve the value of Cronbach's Alpha to make it reliable, it can be excluded
the outer loading indicator is the smallest. The Composite Reliability value generated is all
the construct is very good because it is above >0.70 so it can be concluded that all
The construct indicator is reliable or meets the reliability test.

3. Evaluation of the Structural Model (Inner Model).

a. R-Square Test ( )
Table 22. R-Square Test Results( &)

R Square Adjusted R Square

Employee Performance 0.537 0,519
Motivation 0.204 0.187
Source: Output processing using SmartPLS 3.0. From the table above, the value can be seen.

coefficient of determination (R2 0.537 for the employee performance variable, meaning the variable
extrinsic reward, motivation, career development and leadership style explain
the variance of employee performance variables is 53.7%, which falls into the moderate category,
on the other hand, extrinsic reward variables, intrinsic rewards, and career development
explaining the variance of the motivation variable of 20.4%, which falls into the weak category.
MULTIVARIATE ANALYSIS USING SMART 3.0 57
b. Path Coefficient
To obtain the Path Coefficient value or t-statistic, it must first be
bootstrap estimation by clicking the main menu and then selecting calculateà bootstrapping.

Bootstrapping Setting
• Subsample: 500
• Sign Changes : No Sign Changes

Click Start calculation and the output will appear in the form of a path diagram on the drawing

area as follows

MULTIVARIATE ANALYSIS USING SMART 3.0 58

• After obtaining the coefficient path value, then open this report select path
the next step is to interpret the coefficients, as follows:

Table 23. Path Coefficient

Original Sample Standard T Statistics P Values
Sample Mean Deviation (|O/STDEV|)
(O) (M) (STDEV)
Gaya Kepemimpinan -> Kinerja
0.480 0,477 0.089 5,427 0.000
Employee
Moderating Effect 1 -> Kinerja
0.172 0.137 0.150 1,146 0.253
Employee
Motivasi -> Kinerja Karyawan 0.291 0.280 0.079 3,681 0.000
Pengembangan Karir -> Kinerja
0.197 0.191 0.082 2,413 0.016
Employee
Pengembangan Karir -> Motivasi 0.214 0.214 0.113 1,893 0.059
Reward Ekstrinsik -> Kinerja
0.030 0.038 0.084 0.350 0.727
Employee
Reward Ekstrinsik -> Motivasi 0.059 0.067 0.119 0.497 0.620
Reward Instrinsik -> Motivasi 0.265 0.263 0.090 2,944 0.003
Source: Processing output with SmartPLS 3.0.

MULTIVARIATE ANALYSIS USING SMART 3.0 59

 From the table above, it can be seen that: leadership style has a significant positive influence.
towards employee performance with a t-statistic value >1.96, motivation has a positive influence
significant to employee performance with t-statistic value >1.96, career development
significantly positive influence on employee performance with a t-statistic value >1.96,
Intrinsic reward has a significantly positive effect on motivation with a t-statistic value.
1.96, career development has no effect on motivation with a t-statistic value <
1.96 and P value >0.05, extrinsic rewards do not affect employee performance.
with a t-statistic value < 1.96 and P value > 0.05, extrinsic rewards have no effect
towards motivation with a t-statistic value < 1.96 and P value > 0.05.
For the moderating variable between motivation and leadership style on performance
the employee obtained a t-statistic value of 1.146 < 1.96 and a P value > 0.05, thus it can be
it can be concluded that the leadership style variable is not a moderator variable,
or cannot moderate the relationship between motivation and employee performance.

c. Predictive Validity Test

To obtain the Q² value in the Predictive Relevance Test, it must first be ...
estimate Blindfolding by clicking the main menu and then selecting calculateà Blindfolding.

Blindfolding Setting
Omission Distance : 7

MULTIVARIATE ANALYSIS USING SMART 3.0 60

Click Start calculation and the output will appear in the form of a path diagram on the drawing.

area as follows:

• After obtaining the Q² value, the next step is to interpret it.

as below:
Table 18. Predictive Relevance Test
SSO SSE Q² (=1-SSE/SSO)
Leadership Style 564,000 564,000
Employee Performance 423,000 301,042 0,288
Moderating Effect 1 1,128,000 1,128,000
Motivation 282,000 250,130 0.113
Career Development 564,000 564,000
Extrinsic Reward 282,000 282,000
Intrinsic Reward 282,000 282,000
Source: Processing output with SmartPLS 3.0.

MULTIVARIATE ANALYSIS USING SMART 3.0 61

 Based on the calculation of Predictive Relevance(Q& ) in the table ..... indicates the value
0.288 and 0.113, it can be concluded that the model has relevant predictive value.

d. Model Fit Evaluation

Model fit evaluation in this study uses the Normal Fit Index testing model.
(NFI).
Table 24. Model Fit Evaluation Test
Saturated Model Estimated Model
NFI 0.454 0.450

Source: Processing output with SmartPLS 3.0.

The research results indicate that the value of the saturated model (measurement) fit and the value

The estimate model (structural model) has slight differences. From those results that
it can be concluded that the model in the study has a normal fit index (NFI) value
which shows that the model in this research is 45.4% better than the model
null model.

Explanation: The NFI value can be seen in the PLS algorithm Model Fit Output.

MULTIVARIATE ANALYSIS USING SMART 3.0 62

 REFERENCES

Aprilia, K. and Imam G. 2013. Generalized Structured Component Analysis (GSeCA) Model
Component-Based Structural Equation. Semarang: Diponegoro University.

Fornell, C., and Larcker, D.F. 1981. Evaluating Structural Equation Models with Unobservable
Variable and Measurement Error. Journal of Marketing Research (18.1), pp. 39-50.

Henseler J and Fassott G (2010) Testing Moderating Effects in PLS Path Models: an illustration
of available procedures. In Handbook of Partial Least Squares: Concepts, Methods, and
Applications. Vol. II of Computational Statistics (ESPOSITO VINZI V, CHIN WW,
HENSELER J and WANG H, Eds). Springer, Heidelberg. pp 713–735,

Ghozali, Imam and Hengky Latan. 2015. Partial Least Squares Concepts, Techniques, and Applications
SmartPLS 3.0 For Empirical Research. Semarang: Diponegoro University.

Irwan dan K. Adam. 2015. Metode Partial Least Square (PLS) dan Terapannya: Studi Kasus
Customer Satisfaction Analysis of PDAM Camming Unit Services, Bone District. Journal
Teknosains. Volume 9 Number 1, pages 53 - 68.

Sugiyono. [Link] Penelitian Kuantitatif, Kualitatif dan R&[Link]: Alfabeta.

Widarjono, A. 2015. Applied Multivariate Analysis with SPSS, AMOS, and

SMARTPLS. Yogyakarta: UPP STIM YKPM.

MULTIVARIATE ANALYSIS USING SMART 3.0 63