0% found this document useful (0 votes)

355 views73 pages

SmartPLS Manual 13X

This document provides an overview of moderator analysis in partial least squares structural equation modeling (PLS-SEM). It defines a moderator variable as a third variable that changes the relationship between an independent and dependent variable. The chapter discusses how to develop hypotheses for moderator effects and presents the basic mathematical representation of moderation, including how the moderator interacts with and potentially strengthens or weakens the relationship between other variables. Key aspects covered include the need to specify the conditions under which a moderator effect occurs, and including a direct path from the moderator to the dependent variable to control for its direct impact.

Uploaded by

Nguyễn Đoàn Việt Phương

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

355 views73 pages

SmartPLS Manual 13X

Uploaded by

Nguyễn Đoàn Việt Phương

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 73

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/341357609

PLS-SEM using SmartPLS 3.0: Chapter 13: Assessment of Moderation Analysis

Chapter · July 2017

CITATIONS READS
0 8,119

5 authors, including:

T. Ramayah Cheah Jun Hwa

Universiti Sains Malaysia Universiti Putra Malaysia
632 PUBLICATIONS 15,872 CITATIONS 77 PUBLICATIONS 1,778 CITATIONS

SEE PROFILE SEE PROFILE

Francis Chuah Hiram Ting

Universiti Utara Malaysia UCSI University
51 PUBLICATIONS 684 CITATIONS 144 PUBLICATIONS 1,685 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Doctoral Degree View project

Multimedia University View project

All content following this page was uploaded by Hiram Ting on 13 May 2020.

The user has requested enhancement of the downloaded file.

Partial Least Squares

Structural Equation Modeling
(PLS-SEM) using SmartPLS 3.0
An Updated and Practical Guide to Statistical Analysis

• BASIC LEVEL •SECOND EDITION •

Only Chapter Thirteen Available:

Assessment of Moderation Analysis

Edited by

T. Ramayah
Jacky Cheah, Francis Chuah
Hiram Ting & Mumtaz Ali Memon

July, 2017

Partial Least Squares Structural Equation Modeling (PLS-SEM) using SmartPLS

3.0: An Updated and Practical Guide to Statistical Analysis
(Basic Level; Second Edition)

Table of Contents

Chapter 1 Development of Structural Equation Modeling

Chapter 2 Partial Least Squares Structural Equation Modeling (PLS-SEM)
Chapter 3 Reflection on SEM Application
Chapter 4 When should I use PLS-SEM instead of CB-SEM?
Chapter 5 Using the SmartPLS Program
Chapter 6 Understand the Measurement Model Assessment between EFA and CFA
Chapter 7 Assessment of Reflective Measurement Models
Chapter 8 Assessment of Formative Measurement Models
Chapter 9 Assessment of Goodness-of-Fit (GOF)
Chapter 10 Assessment of Structural Model
Chapter 11 Assessment of Importance and Performance Matrix Analysis (IPMA)
Chapter 12 Assessment of Mediation Analysis
Chapter 13 Assessment of Moderation Analysis

Chapter Thirteen
Assessment of Moderator Analysis

13.1 What is a Moderator Analysis?

A moderator variable can be visualized as a third variable that changes the relationship between
the independent variable and the dependent variable. It is usually called a contingent variable
(see Figure 13.1). A moderator variable is one that affects the relationship between two variables,
so that the nature of the impact of the predictor on the criterion varies according to the level or
value of the moderator (Holmbeck, 1997). In other words, a moderator interacts with the
predictor variable in such a way as to have an impact on the level of the dependent variable. Thus
a moderator specifies the conditions under which a given effect occurs, as well as the conditions
under which the direction (nature) or strength of an effect vary. Baron and Kenny (1986, pp. 1174,
1178) describe a moderator variable as follows:

A qualitative (e.g., sex, race, class) or quantitative variable . . . that affects the direction and/or
strength of a relation between an independent or predictor variable and a dependent or criterion
variable . . . a basic moderator effect can be presented as an interaction between a focal
independent variable and a factor (the moderator) that specifies the appropriate conditions for its
operation . . . Moderator variables are typically introduced when there is an unexpectedly weak or
inconsistent relation between a predictor and a criterion variable.

Figure 13.1: A Moderated Relationship

When developing moderator hypothesis, researchers should put efforts in developing

explanations of the condition of the interactions result in advance. In other words, researchers
should not only explain the existence of an interaction effect as predicted in the model, but also
its form of condition. In particular, whether a moderator increases or decreases the association
between two other variables should be specified as part of the a priori hypothesis (Dawson,
2013). Figure 13.2 and 13.3 explain how to develop hypothesis for moderator effect.

Figure 13.2: Gender as a Moderator

Figure 13.3: Task Ability as a Moderator

13.2 Basic understanding on moderation effects: basic mathematical representation

To gain an understanding of how moderating effects are modelled, consider the path model
shown in Figure 13.4. The moderating effect (d) is represented by an arrow pointing at the effect
b linking X and Y. Furthermore, when including the moderating effect in PLS path model, there
is also a direct relationship (c) from the moderator to the endogenous construct. This additional
path is crucial as it controls or the direct impact of the moderator on the endogenous construct.
How to Cite:
Ramayah, T., Cheah, J., Chuah, F., Ting, H., & Memon, M. A. (2018). Partial least squares structural equation modeling (PLS-SEM) using
SmartPLS 3.0: An updated guide and practical guide to statistical analysis (2nd ed.). Kuala Lumpur, Malaysia: Pearson.
How to Cite:
Ramayah, T., Cheah, J., Chuah, F., Ting, H., & Memon, M. A. (2018). Partial least squares structural equation modeling (PLS-SEM) using
SmartPLS 3.0: An updated guide and practical guide to statistical analysis (2nd ed.). Kuala Lumpur, Malaysia: Pearson.

If the path c were to be omitted, the effect of M on the relationship between X and Y would be
inflated. The path model in Figure 13.4 can also be expressed mathematically using the following
formula:

Main Effect:
Y = a + b. X + c. M

Interaction Effect:
Y = a + (b + d . M) . X + c. M

Y = a + b . X + c . M + d (X*M)

Figure 13.4: Moderating Effect

Based on the Figure 13.4 and Figure 13.5, it explains that:

▪ b is called the main effect when no moderator is included.

▪ b is called the simple effect when a moderator is included.
o The strength of relationship between X and Y when M is zero.
o If the level of moderator is increased by one standard deviation unit, b is expected
to change by the size of d.
▪ c accounts for variation in Y explained by M.

Figure 13.5: Moderating Effect when running Analysis

13.3 Types of Interaction Effects

In PLS-SEM 3.26, there are three approaches that researchers can employ to examine interaction
terms (Moderator Analysis). They are Product-Indicator Approach (Chin, Marcolin & Newsted,
1996; 2003), Two-Stage Approach (Henseler et al. 2012, Chin et al., 2003) and Orthogonalzing
Approach (Henseler & Chin, 2010). Subsequent sections provide a brief explanation of each step
including their selection considerations. Also, step by step guidelines are provided to perform
each approach of moderation analysis.

13.3.1 Product Indicator Approach

The product indicator approach is the conventional approach for creating the interaction term
in regression-based analyses and it also features prominently in PLS-SEM. It involves multiplying
each indicator of the exogenous construct with each indicator of the moderator (Chin et al.,
1996). Hence, the product indicators become the indicators of the interaction term. Specifically,
the notion of product indicator approach is the product term (X*Z) which is used to examine the
influence that the moderator z would have on the relationship between predictor X and the
dependent variable of interest Y. The indicator multiplication builds on the assumption that the
indicators of the exogenous construct and the moderator each stem from a certain construct
domain and are in principle interchangeable. In other words, if all constructs (IV & Moderator)
are reflective then a product indicator approach is the best (refer to Figure 13.6).

However, product indicator approach is not applicable when the exogenous construct and/or
the moderator are measured formatively. Since formative indicators do not have to correspond
to a predefined theoretical concept, multiplying them with another set of indicators will
confound the conceptual domain of the interaction term. In addition, the product indicator
approach requires the indicators of exogenous construct and the moderator variable to be
reused in the measurement model of the interaction term. This procedure, however, inevitably
introduces collinearity in the path model.

Figure 13.6: Product Indicator Approach Technique if both IV and Moderator are reflective and
continuous variables

In addition, product indicator approach can be used when researchers have categorical variable as
moderator variable, or other non-continuous variables but IV is a continuous variable (determine
moderating effects through group comparisons). However, researchers must be able to create the
dummy variables (Gujarati, 1970). For example, gender has two sub-categories (male and female),
hence, researchers need to use SPSS to recode male as 1 and female as 0 so that they can use female
as a benchmark compare to male. As the single effect of the exogenous variable describes the effect
when the moderator variable equals zero, a dummy coded moderator variable allows a
straightforward interpretation of this single effect. Therefore, dummy coding should be used for a
dichotomous (single) indicator variable instead of contrast coding. Note: If researchers have 2 groups
of nominal variable, then only 1 “dummy” variable is needed. In addition, if researchers have more
than two groups (> 2 groups) (e.g. race or education level), then they need to consider (k - 1) dummy
variables. For this example, researchers can refer to Figure 13.7 to see on how to create the dummy
variables for more than two groups of moderator, where the example indicates that m1 and m2 are
the dummy variables (refer to Figure 13.7).

Figure 13.7: Product Indicator Approach Technique if both IV are reflective and continuous
variables but the Moderator is a Nominal Category

13.3.2 When to use product indicator approach?

To begin with, product indicator approach is known as the most frequently used technique in
many studies. Product indicator approach has always shown a higher prediction accuracy
compared to two-stage approach, especially when the sample size (≥ 200 sample size) or the
number of indicators per construct is medium to large (≥ 4 indicators). However, the biggest
issue in this approach is, it has a significantly weak statistical power (issue on Type 1 and Type 2
errors) compared to two-stage approach and orthogonalizing approach (Henseler & Chin, 2010)
due to collinearity issue when creating interaction term. Even if researchers use indicator
standardization to reduce the level of collinearity in the PLS path model, it does not fully
eliminate it. The collinearity issue will still be substantial, yielding inflated standard errors or
biased path coefficient estimates. In addition, Henseler and Chin’s (2010) study also points out
that when using product indicator approach, the path coefficient of the interaction term must
not be sued to quantify the strength of the moderating effect. Therefore, we generally advise
researchers to be cautious when they decide to use the product-indicator approach.

13.3.3 Steps on How to Perform Moderation Analysis using Product Indicator Approach (i.e.,
Moderator Variable is a Categorical)

To illustrate on how to perform on the moderator analysis using interaction effect (Moderator
Variable is a Categorical), this chapter will use the File: Burnout (n=107). The variables in the data
are Burnout (BO1 to BO2) as reflective exogenous variables and Job Satisfaction (JS2 to JS5) will
be endogenous variables (reflective measurement). Lastly, gender (Male=1 and Female=0) will
play a major role of a moderator effect (Categorical Variable).The drawing of the model is
illustrated in Figure 13.8. The moderation hypothesis can be written as follows:

H1: The negative relationship between Burnout and Job Satisfaction (JS) will be stronger for
Female when compared to Male.

Note: Researchers must do the same process of adding the moderator variable (Gender) into the
model and run the measurement model as usual to report the measurement model assessment
(Loadings, AVE and CR). This model is called the main effect model and the R2 will be noted before
introducing the interaction terms.

Figure 13.8: Burnout (n=107)

Step 1: Click on Calculate and PLS Algorithm

Step 2: Start Calculation

Step 3: You will get your R2 result

R2 result

The R2 of 0.232 indicates that the Burnout variable and the moderator which is Gender explain
23.2% of the variance in Job Satisfaction (JS). Now we are ready to create the interaction term in
this model.

Step 4: To create the interaction terms between Burnout and Gender, researchers should click
right on the Job Satisfaction construct. A dialog box will be displayed and then follow by
clicking Add Moderating Effect.

Step 5: You will get the next dialog box as follows:

Step 6: Now we will create the 1st interaction term by interacting AC and Gender

For Moderator Variable, click on

Gender; for Independent
Variable, click on Burnout.
However, based on this model,
this study intends to choose
product-indicator approach.
Lastly, click on Unstandardized
for Moderator that is Categorical
Form.

Step 7: Once Click Ok, you will get a new latent construct as follows (Moderating Effect 1):

Step 8: Name the New Construct of Moderating Effect 1 to Burnout*Gender

Step 9: Now your interaction effect is named Burnout*Gender

Step 10: From the model, it looks like the interaction effect has no indicators. To see the
indicators, we can click right on the interaction term and choose “Show Indicators of Selected
Construct”.

Step 11: Now you will see that the system has automatically created 3 interaction terms and it
is better for researchers to click back at “Hide Indicators of Selected Construct” to have a neat
and organized model.

Step 12: Once we have done with the interaction process, the full model will look like the one
shown below and we will run the analysis using PLS Algorithm.

Step 13: Look at the Algorithm result.

Check on the increase result of the

R2 change

In moderation analysis, the R2 change is important. As such, we will first look at the R2 change
from the main effect model. If you recall, the previous R2 for the main effect model is 0.232, now
in the interaction effect model, it is 0.252. The R2 change of 0.020 indicates that with the addition
of the interaction term (Burnout*Gender), the R2 has changed about 2.0% (additional variance).

Step 14: Next we need to calculate the effect size (f2) using the Effect Size Calculator

For included, insert the

interaction effect model R2
(0.252) and for excluded,
insert the R2 for the main
effect model (0.232). Thus,
result indicates small effect
size.

Now, we can follow the interpretation of the f2 by following the guidelines given in Cohen (1988,
p. 410 - 414), as follows:

• 0.02 small
• 0.15 medium
• 0.35 large

So based on the f2 of 0.040, we can conclude that the effect size is small (Cohen, 1988). Chin et
al. (2003) state that a low effect size f2 does not necessarily imply that the underlying moderator
effect is negligible: “Even a small interaction effect can be meaningful under extreme
conditions, if the resulting beta changes are meaningful, then it is important to take these
conditions into account” (Chin, Marcolin & Newsted, 2003; p. 211).

Although our result of the beta coefficient for the interaction term of Burnout*Gender beta is
0.218 (see Step 13), we would still unable to confirm whether the beta is either statistically
significant or not. Thus we will proceed to test and see which of the 2 interaction effects is
significant.

Step 15: Now, we proceed with testing the interaction effects using the bootstrapping
procedure which will give us the t-values. Click on Calculate and followed by Bootstrapping.

Step 16: Click on Start Calculation, then, we can now confirm whether the interaction terms are
significant or not.

Step 17: Bootstrap Result.

From the bootstrap result, we will only interpret the interaction term. The cut off value for this
particular test will be 1.645 (α = 0.05) and 2.33 (α = 0.01). As can be seen from the result, the
interaction term of Burnout*Gender is significant (t-value=1.886).

Step 18: Extract the Bootstrapping Result by Clicking Copy to Clipboard via Excel Format.

Original Standard
Sample T Statistics
Sample Deviation P Values
Mean (M) (|O/STDEV|)
(O) (STDEV)
Burnout -> Job Satisfaction -0.787 -0.769 0.151 5.228 0.000
Burnout*Gender -> Job Satisfaction 0.218 0.199 0.116 1.886 0.030
Gender -> Job Satisfaction -0.160 -0.143 0.180 0.889 0.187

Step 19: For reporting purpose, we can summarize the results in Table 13.1.

Table 13.1: Results of the Moderator analysis

Std.
Hypothesis Relationship Std. Beta Error t-value
H1 Burnout*Gender-> Job Satisfaction 0.218 0.116 2.272**
Note: **p< 0.01, *p< 0.05

As can be seen in Table 13.1, the interaction between Burnout*Gender is positive but it is not
entirely clear how it differs in terms of the groups (Male vs Female). In other words, the size and
precise nature of this effect is not easy to define from examination of the coefficients alone, and
it becomes even more so when one or more of the coefficients are either positive or negative,
or when the standard deviations of X and Z are very different (Dawson, 2014). Thus, Dawson

(2014) suggested that to follow up for the significant interactions, an interaction plot can be
drawn. Previously, in SPSS, drawing this interaction plot would be quite tedious as you need to
run the descriptive for the IV and the Moderator to get a value to split the variable into High/Low
before you can plot.

Step 20: Since our moderator is a categorical measurement in PLS analysis, hence, researchers
should choose interaction plot that is suitable for binary moderator (2-way with binary
moderator.xls).

Note: There are several values that are needed for the plotting and they are the beta coefficient for
independent variable, moderator and interaction.

Step 21: Researchers should recall the results of the beta coefficient that are obtained earlier.

Note: Only take consideration of the significant moderator beta coefficient result
before plotting.

Only Burnout*Gender has the interaction results. Thus, we will summarize the values from the
algorithm as follows:

Constructs Coefficients
Burnout -0.787
Gender -0.160
Burnout*Gender 0.218

Step 22: Researchers should plot the Interaction Results.

Alternatively, researchers can also look for Simple Slope Analysis Result that visualizes the two-
way interaction effect for SmartPLS 3.2.6.

First, look back on the PLS Algorithm

result that is generated with the
interaction term of Burnout*Gender.

Then, scroll to the bottom and click on

Simple Slope Analysis. Then, look at the
significant moderator result that is
Burnout*Gender.

Note: The three lines shown in the Simple Slope Analysis represent the relationship between
Burnout (x-axis) and Job Satisfaction (y-axis). The middle line represents the relationship for an
average level of the moderator variable Gender, which is not useful to be interpreted in
dichotomous or categorical data. The other two lines that are important to facilitate the
interpretation of Gender as a moderator effect between the relationship of Burnout and Job
Satisfaction are Gender at +1SD (Male = 1 ) and Gender at -1SD (Female = 0).

Step 23: Interpret the Interaction Result.

Figure 13.9: Plotting Graph Result

Just to recall the hypothesis we have written earlier:

H1: The negative relationship between Burnout and Job Satisfaction (JS) will be stronger for
Female when compared to Male.

The interpretation of the interaction plots is to look at the gradient of the slopes and the
directions. As can be seen in Figure 13.9, the line labelled for female has a steeper and negative
gradient when compared to the male (less steep and negative gradient) indicating that the
negative relationship is indeed stronger for female group. Thus, our hypothesis is supported as
what we have hypothesized before the analysis.

13.4 Two-Stage Approach

The idea of the two-stage approach was initially suggested by Chin et al. (2003) and elaborated
by Fassot, Henseler and Coelho (2016) as well as Henseler and Fassott (2010). Specifically, if the
exogenous variable or the moderator variable is formative, the pairwise multiplication of
indicators is not feasible. “Since formative indicators are not assumed to reflect the same
How to Cite:
Ramayah, T., Cheah, J., Chuah, F., Ting, H., & Memon, M. A. (2018). Partial least squares structural equation modeling (PLS-SEM) using
SmartPLS 3.0: An updated guide and practical guide to statistical analysis (2nd ed.). Kuala Lumpur, Malaysia: Pearson.
How to Cite:
Ramayah, T., Cheah, J., Chuah, F., Ting, H., & Memon, M. A. (2018). Partial least squares structural equation modeling (PLS-SEM) using
SmartPLS 3.0: An updated guide and practical guide to statistical analysis (2nd ed.). Kuala Lumpur, Malaysia: Pearson.

underlying construct (i.e., can be independent of one another and measuring different factors),
the product indicators between two sets of formative indicators will not necessarily tap into the
same underlying interaction effect” (Chin et al., 2003). Instead of using the product indicator
approach, Fassot et al. (2016) and Henseler and Fassott (2010) similarly articulated the two-stage
PLS approach for estimating moderating effects in particular when formative constructs are
involved. The two-stage approach makes use of PLS path modeling’s advantage of explicitly
estimating latent variable scores. The two stages are built up as follows (refer to Figure 13.10 and
Figure 13.11):

Stage 1: In the first stage, the main effect PLS path model is run to obtain estimates for the latent
variable scores. The latent variable scores (standardized) are calculated and saved for further
analysis. In other words, researchers should run the algorithm to calculate the latent variable
scores (standardized).

Figure 13.10: Stage 1

Stage 2: In the stage 2, the interaction term X*M is built up as the element wise product of the
latent variable scores of X and M. This interaction term as well as the latent variable scores of X
and M are used as independent variables in a multiple linear regression on the latent variable
scores of Y. In other words, the latent variables scores from Stage 1 will become the indicators
for the exogenous, endogenous and moderator.

Figure 13.11: Stage 2

13.4.1 When to use two-stage approach

In terms of the two-stage approach, it is applicable when formative measures are involved in the
model (regardless it is an exogenous construct and/or the moderator) (Henseler & Chin, 2010).
Although Chin et al. (2003) and Fassott et al. (2016) limited the usage of the two-stage approach
to cases when the exogenous or the moderator variable or both are formative, this limitation is
not mandatory. It can also be applied to models with interaction effect among all constructs that
are measured by reflective indicators. This notion was highlighted by Hair et al., (2017) and
Henseler and Chin (2010) that the choice of this method depends on the aim of the analysis.
When the objective is to determine whether or not the moderator exerts a significant effect on
the relationship, the two-stage approach is preferred. This approach yields the most accurate
estimates of the single effects and it has higher statistical power test (Type 1 error and Type 2
error) when compared to product indicator approach or orthoganalizing approach. However, a
clear disadvantage of the two-stage approach is that the moderating effect is not taken into
account when estimating the latent variable scores. The fact that the two-stage approach is a
limited-information approach is the key reason for Chin et al. (2003) to prefer the product
indicator approach and also the prediction accuracy of this approach is low when compared to
product indicator approach or orthoganalizing approach (Henseler & Chin, 2010).

13.4.2 Steps on How to Perform Moderation Analysis using the Two-Stage Approach (i.e., IV is
Reflective and Moderator is Formative or the opposite measurement model).

To illustrate on how to perform on the moderator analysis using two-stage approach, this
chapter will use the File: Two-Stage (n=1400). The variable in the data is Behavioral Intention - BI
(BI1 to BI3) as reflective exogenous variables. Use-Task (USE3 and USE4) will be endogenous

variable (formative measurement). Lastly, Facilitating Condition (FC1 to FC3) will play a major role
of a moderator variable (formative measurement). The drawing of the model is illustrated in
Figure 13.12. The moderation hypothesis can be written as follows:

H1: The positive relationship between Behavioral Intention (BI) and Use-Task will be stronger for
higher facilitating condition (FC).

Note: Researchers must always do the same process of adding the moderator variable (FC) into
the model and run the measurement model as usual to report the reflective measurement model
assessment (Loadings, AVE and CR) as well formative measurement assessment (Concurrent
Validity, Weights, VIF and t-values). This model is called the main effect model and the R2 will be
noted before introducing the interaction terms.

Figure 13.12: File:Twostage (n=1400)

Step 1: Click on Calculate and PLS Algorithm.

Step 2: Start Calculation.

Step 3: You will get your measurement model result for reflective and formative
measurements.

Step 4: Assessing the measurement model for both reflective models for BI.

Delete BI3

Note: Researchers should check on the reflective measurement result (loadings, AVE, CR and Discriminant
Validity). Initially, BI3 indicates poor loading that will affect the poor result of AVE and CR, thus, it must be
removed from the model.

Step 5: Re-run PLS Algorithm after deleting BI3 indicator from the construct of Behavioural
Intention

Step 6: Click on Start Calculation.

Step 7: Check on the AVE and CR results to ensure they meet the satisfactory threshold value.

First, Click PLS Algorithm

Second, Click Construct Reliability

and Validity and make sure the result
of BI has met the threshold value of
AVE > 0.50 as well as CR>0.70

Note: There is no discriminant validity issue due to one reflective measurement construct in the model,
which is BI.

Step 8: Check on the outer weights for formative measurement.

Note: In the initial stage, the redundancy analysis needs to be performed to assess the concurrent
validity for formative measurement. In this example, redundancy analysis is not shown as the early
step to assess formative measurement.

Step 9: Check the collinearity issue. Click Collinearity Statistics (VIF) and then click Outer VIF
Values.

Click on Collinearity Statistics (VIF)

and then click Outer VIF Values.
Researchers should look for
indicators of FC and USE

Based on the result, there is no problem with the VIF values result as all formative indicators of FC1
to FC3 as well as USE3 and USE4 are < 5. Thus, the result indicates no collinearity issue.

Step 10: Check the relative contribution for formative measurement for FC and Use-Task by
clicking Bootstrapping.

Note: The subsample must change to any number (in this case subsamples about 2000) that is larger
than the actual observation of 1400.

All the formative indicators have met the t-value result of >1.96. Therefore, relative contributions
are established on FC and Use-Task.

Step 11: In previous Step 7 and Step 8, you will also get your R2 result.

R2 result

The R2 of 0.283 indicates that the variables of Behavioral Intention (BI) and the moderator which
is Facilitating Condition (FC) explain 28.3% of the variance in Use-Task. Now we are ready to
include the interaction term. Since this model has only one IV, hence, we have to create only
interaction term.

Step 12: Now create the interaction terms between BI and FC, researchers should click right on
the Use-Task Construct. A dialog box will be displayed and then followed by clicking Add
Moderating Effect.

Note: Due to the sophistication of SmartPLS 3.2.6 when compared to SmartPLS 2.0, researchers can
now skip the procedure of obtaining the latent variable scores (LVSs) and save it for further
moderator analysis. In other words, researchers can straightaway proceed with performing the
interaction term.

Step 13: You will get the next dialog box as follows:

Step 14: Now we will create the interaction term by interacting BI and AC.

For Moderator Variable, click

on FC and for Independent
Variable, click on BI. Also,
based on this model, this study
must choose the Two-Stage
Approach. Lastly, click on
Mean Centered for Moderator.

Step 15: Once you Click Ok, you will get a new latent construct as follows (Moderating Effect 1):

Step 16: Name the New Construct of Moderating Effect 1 to BI*FC.

Step 17: Now your interaction effect has the name of BI*FC.

Step 18: From the model, it looks like the interaction effect has no indicators. To see the
indicators, we can click right on the interaction term and choose “Show Indicators of Selected
Construct”.

Step 19: Now you will see that the system has automatically created only one interaction terms
using the technique of two-stage approach. However, it is better for researchers to click back
at “Hide Indicators of Selected Construct” to have a neat and organized model.

Step 20: Now we will run the analysis using PLS Algorithm.

Step 21: Look at the Algorithm result.

Check on the increase result of the

R2 change

In moderation analysis the R2 change becomes an important issue. As such, we will first look at
the R2 change from the main effect model. If you recall, the previous R2 for the main effect model
is 0.283, now in the interaction effect model the R2 is 0.296. The R2 change of 0.013 indicates that
with the addition of one interaction term, the R2 has changed about 1.3% (additional variance).

Step 22: Next we need to calculate the effect size (f2) using the Effect Size Calculator

For included, insert the

interaction effect model R2
(0.296) and for excluded, insert
the R2 for the main effect model
(0.283).

Now, we can follow the interpretation of the f2 is by following the guidelines given in Kenny’s
(2016), as follows:

• 0.005 small
• 0.01 medium
• 0.025 large

Again, this is used when researchers fail to pass the minimum or small effect size suggested by
Cohen’s (1988) criterion (see the Effect Size Calculator result that exhibits none effect size).
Based on the effect size result, the f2 of 0.0185 result indicates medium effect size as suggested
by Kenny (2016).

Now, although our result of the beta coefficient for the interaction effect of BI*FC beta is 0.103
(see Step 21), we would still unable to confirm whether the beta is either statistically significant
or not. Thus we will proceed to test and see whether the interaction effects are significant or
not.

Step 23: Now, we proceed with testing the interaction effects using the bootstrapping
procedure which will give us the t-values. Click on Calculate and followed by Bootstrapping.

Step 24: Click on Start Calculation once the bootstrapping results are out, we can then confirm
whether the interaction terms are significant or not.

Step 25: Bootstrap Result.

From the bootstrap result, we can now interpret the interaction effect. The cut off value for this
particular test will be 1.645 (α = 0.05) and 2.33 (α = 0.01). As can be seen from the bootstrap
result, BI*FC is significant (t-value=5.422).

Step 26: Extract the Bootstrapping Result by Clicking Copy to Clipboard or Export to Excel.

First, click the

Bootstrapping Result.
Next, extract the
bootstrapping result
by clicking Excel
format.

Original Sample Standard T Statistics P Values

Sample Mean Deviation (|O/STDEV|)
(O) (M) (STDEV)
BI -> USE-TASK 0.431 0.432 0.022 19.177 0.000
BI*FC -> USE-TASK 0.103 0.103 0.019 5.422 0.000
FC -> USE-TASK 0.246 0.248 0.023 10.813 0.000

Step 27: For reporting purpose, we can summarize the results in Table 13.2.
Table 13.2: Results of the Moderator analysis
Std. Std
Hypothesis Relationship Beta Error t-value
Behavioral Intention (BI) *Facilitating Condition
H1 (FC)-> Loyalty 0.109 0.019 5.422**
Note: **p< 0.01, *p< 0.05
As we can see in Table 13.2, the interaction between Behavioral Intention (BI)*Facilitating
Condition (FC) is positive. Thus we can say that the positive relationship between BI and Use-
Task would be stronger when FC is higher. However, it is not entirely clear how it differs. If you
get a positive coefficient, the positive coefficient of the interaction term suggests that it
becomes more positive as Facilitating Condition increases; however, the size and precise nature
of this effect is not easy to define from examination of the coefficients alone, and it becomes
even more so when one or more of the coefficients are negative, or when the standard
deviations of X and Z are very different (Dawson, 2014). Thus, Dawson (2014) suggested that to
follow up for the significant interactions, an interaction plot can be drawn.
How to Cite:
Ramayah, T., Cheah, J., Chuah, F., Ting, H., & Memon, M. A. (2018). Partial least squares structural equation modeling (PLS-SEM) using
SmartPLS 3.0: An updated guide and practical guide to statistical analysis (2nd ed.). Kuala Lumpur, Malaysia: Pearson.
How to Cite:
Ramayah, T., Cheah, J., Chuah, F., Ting, H., & Memon, M. A. (2018). Partial least squares structural equation modeling (PLS-SEM) using
SmartPLS 3.0: An updated guide and practical guide to statistical analysis (2nd ed.). Kuala Lumpur, Malaysia: Pearson.

Step 28: Since our moderator is continuous measurement and the beta in PLS analysis is
standardized beta, hence, researchers should choose the standardized beta coefficients (2-way
standardised.xls) for Interaction Plot.

Note: There are several values that are needed for the plotting and they are the beta coefficient for
independent variable, moderator and interaction.

Step 29: Researchers should recall the results of the beta coefficient that are obtained earlier.

Note: Only take consideration of the significant moderator beta coefficient result.

BI*FC has the interaction results. Thus, we will summarize the values from the algorithm as
follows:

Constructs Coefficients
BI 0.431
FC 0.246
BI*FC 0.103

Step 30: Researchers should plot the Interaction Results.

Alternatively, researcher can also look for Simple Slope Analysis Result that visualizes the two-
way interaction effect for SmartPLS 3.2.6.

First, look back on the PLS Algorithm

result that is generated with the
interaction term of Bi*FC.

Then, scroll to the bottom and click on

Simple Slope Analysis. Then, look at the
significant moderator result that is BI*FC.

Note: The three lines shown in the Simple Slope Analysis represent the relationship between BI
(x-axis) and Use-Task (y-axis). The middle line represents the relationship for an average level of
the moderator variable FC. The other two lines represent the relationship between BI and Use-
Task for higher (i.e., mean value of FC plus one standard deviation unit) and lower (i.e., mean
value of FC minus one standard deviation unit) levels of the moderator variable FC.

Step 31: Interpret the Interaction Result.

Figure 13.13: Interaction Plot

Just to recall the hypothesis we have written earlier:

H1: The positive relationship between Behavioral Intention (BI) and Use-Task will be stronger
for higher facilitating condition (FC)

The interpretation of the interaction plots is to look at the gradient of the slopes. As can be seen
in Figure 13.13, the line labelled for high FC has a steeper gradient when compared to low FC,
indicating that the positive relationship is indeed stronger when FC is high. Thus, our hypothesis
is supported as what we have hypothesized before the analysis.

13.5 Orthogonalization Approach

The orthoganalizing approach is an extension of the use of residual centering for moderated
multiple regressions as described by Lance (1988). It is an extension of the product indicator
approach. Little, Bovaird, and Widaman (2006) developed the approach to address two issues
that are the result of the standardization of variables as implemented in the product indicator
approach. First, while indicator standardization or mean-centering reduces the level of
collinearity in the PLS path model, it does not fully eliminate it. A certain degree of correlation
between the interaction term and the original variables remains. So despite the standardization
or mean-centering, collinearity in the path model may still be substantial, yielding inflated
standard errors or biased path coefficient estimates. To eliminate this remaining correlation,
residual centering using orthogonalizaing approach can be used to represent the interaction
effect. Second, when the variables are standardized, one cannot readily compare the direct
effect between IV and DV when no interaction term is included (also known as the main effect),
with the effect between IV and DV when the interaction term is included (also known as the
simple effect).

The orthogonalizing approach builds on the product indicator approach (extension of the
product indicator approach) and requires creating all product indicators of the interaction term.
Looking at the Figure 13.6, this would mean creating four product indicators: x1*m1, x1*m2,
x2*m1, and x2*m2. Next, it is to regress each product indicator on all indicators of the exogenous
construct and the moderator variable. For the example in Figure 13.10, we would need to
establish and estimate the following four regression models:

x1*m1 =b1,11*x1+b2,11*x2+b3,11*m1+b4,11*m2+e11
x1*m2=b1,12*x1+b2,12*x2+b3,12*m1+b4,12*m2+e12
x2*m1 =b1,21*x1+b2,21*x2+b3,21*m1+b4,21*m2+e13
x2*m2 =b1,22*x1+b2,22*x2+b3,22*m1+b4,22*m2+e14

In each regression model, a product indicator (e.g., x1*m1) represents the dependent variable,
while all indicators of the exogenous construct (i.e., x1 and x2) and the moderator (m1 and m2)
act as independent variables. When looking at the outcomes, researchers should be interested
in the residual term e but not the regression coefficient b. The orthogonalizing approach uses
the standardized residuals e as indicators for the interaction term, as shown in Figure 13.14.
How to Cite:
Ramayah, T., Cheah, J., Chuah, F., Ting, H., & Memon, M. A. (2018). Partial least squares structural equation modeling (PLS-SEM) using
SmartPLS 3.0: An updated guide and practical guide to statistical analysis (2nd ed.). Kuala Lumpur, Malaysia: Pearson.
How to Cite:
Ramayah, T., Cheah, J., Chuah, F., Ting, H., & Memon, M. A. (2018). Partial least squares structural equation modeling (PLS-SEM) using
SmartPLS 3.0: An updated guide and practical guide to statistical analysis (2nd ed.). Kuala Lumpur, Malaysia: Pearson.

The variance of this new orthogonalized interaction term contains the unique variance that fully
represents the interaction effect, independent of the first-order effect variance (as well as
general error or unreliability) (Little et al., 2006, p. 500). That way, it is ensured that the indicators
of the interaction term do not share any variance with any of the indicators of the exogenous as
well as the moderator variable. In other words, the interaction term is orthogonal to the other
two constructs (IV and M), precluding any collinearity issues among the constructs involved.
Another vital consequence of the orthogonality is that the path coefficient estimates in the
model without the interaction term are identical to those with the interaction term. Specifically,
this approach avoids the changes in the standardized beta coefficient result of IV when running
the interaction. Therefore, this approach as described by Little et al. (2006), has a correlated
error structure that is required to provide unbiased estimates, which indirectly increases the
interpretability of the overall result of the moderator analysis. In other words, the characteristics
of orthogonalization approach will greatly facilitate the interpretation of the moderating effect’s
strength compared with the product indicator approach. However, because of its reliance on
product indicators, the orthogonalization approach is only applicable when the exogenous
construct and the moderator variable are measured reflectively.

Figure 13.14: The Orthogonalizing Approach

13.5.1 When to use orthogonalizing approach

Importantly, when the primary concern is minimizing estimation bias, the orthogonalizing
approach should be preferred as it performs best when compared to product indicator and two-
stage approaches (Henseler & Chin, 2010). The reason is that both the product indicator and the
two-stage approaches will provide an interaction term which can be correlated to both the
independent and the moderator variable. As a consequence, the typical phenomena of
multicollinearity may occur, such as unexpected signs of coefficients or increased standard
errors. Although multicollinearity caused by interaction terms is not a problem per se, it can
How to Cite:
Ramayah, T., Cheah, J., Chuah, F., Ting, H., & Memon, M. A. (2018). Partial least squares structural equation modeling (PLS-SEM) using
SmartPLS 3.0: An updated guide and practical guide to statistical analysis (2nd ed.). Kuala Lumpur, Malaysia: Pearson.
How to Cite:
Ramayah, T., Cheah, J., Chuah, F., Ting, H., & Memon, M. A. (2018). Partial least squares structural equation modeling (PLS-SEM) using
SmartPLS 3.0: An updated guide and practical guide to statistical analysis (2nd ed.). Kuala Lumpur, Malaysia: Pearson.

hamper the interpretation (Echambadi & Hess, 2007). Thus, this negative consequence can be
avoided by adapting the use of residual centering or also known as orthogonalizing approach.
Similarly, when the aim is to maximize prediction of the endogenous construct, researchers
should apply the orthoganalizing approach as this approach yields high prediction accuracy
(Henseler and Chin, 2010). In addition, this approach is useful for researchers who are interested
to quantify or interpret the strength of the moderating effect (Henseler & Chin, 2010) because
the path coefficient estimates in the model without the interaction term are identical to those
with the interaction term. However, this approach works best in the case of small sample size (≤
200 sample size) and few indicators per constructs (≤ 4 indicators) (Henseler & Chin, 2010).
However, the biggest issue in this approach is, it has a significantly weak statistical power (issue
on Type 1 and Type 2 error) when compared to two-stage approach only (Henseler & Chin, 2010).

13.5.2 Steps on How to Perform Moderation Analysis using Orthogonalization Approach (i.e.,
Moderator Variable is a Continuous Variable)

To illustrate on how to perform on the moderator analysis using interaction effect (Moderator
Variable is a Continuous Variable), this chapter will first use the MODC Data Set (n=413). The
variables in the data are Privacy (PP1 to PP3) as reflective exogenous variable, while Trust (FP1
to FP5) will be endogenous variable. Lastly, Authentication (AUT1 to AUT5) will play a major role
of moderator effect (Reflective Continuous Variable). The reason this variable is used as a
moderator is because we would test if the negative relationship (Direct relationship of Privacy
on Trust) is actually dependent on Authentication. For moderator analysis not only should the
existence of an interaction effect be predicted, but also its form. In particular, whether a
moderator increases or decreases the association between two other variables should be
specified as part of the priori hypothesis (Dawson, 2014). The moderation hypotheses for Figure
13.15 can be written as follows:

H1: The negative relationship between Privacy and Trust will be stronger when Authentication is
lower.

Note: Researchers must add in the moderator (Authentication) into the model and run the
measurement model as usual to report the measurement model assessment (Loadings, AVE and
CR). This model is called the main effect model and the R2 will be noted before introducing the
interaction terms.

Figure 13.15: MODC Data Set (n=413)

Step 1: Click on Calculate and then followed by clicking PLS Algorithm to check the R2 that we
get and the result is in shown in Step 3.

Step 2: Click on Start Calculation.

Step 3: You will get your measurement model result for reflective constructs.

Step 4: Assessing the measurement model for both reflective models for BI.

Delete PP2

Note: Researchers should check on the reflective measurement result (loadings, AVE, CR and
Discriminant Validity). Initially, PP2 indicates poor loading that will affect the poor result of AVE
and CR, thus, it must be removed from the model.
Step 5: Re-run PLS Algorithm after deleting pp2 indicator from the construct of Privacy to get
the R2 result.

Step 2: Click on Start Calculation.

Step 7: Check on the AVE, CR and Discriminant Validity result again to ensure it meets the
satisfactory threshold value.

First, click PLS Algorithm

Second, Click Construct Reliability

and Validity and make sure the result
of BI has met the threshold value of
AVE > 0.50 as well as CR>0.70

Third, click Discriminant Validity to

make sure the result of each
constructs correlation is below the
threshold value of 0.85 (Kline, 2015).

Step 8: Once satisfactory measurement model result is achieved, you will need to get the R2
result of this model.

R2 result

The R2 of 0.318 indicates that the Privacy (exogenous variable) and the Authentication
(moderator variable) explain 31.8% of the variance in Trust. Now we are ready to create the
interaction term in this model.

Step 9: To create the interaction term between Privacy and Authentication, researchers should
click right on the Trust (DV) construct. A dialog box will be displayed and then followed by
clicking at Add Moderating Effect.

Step 10: You will get the next dialogue box as follows.

Step 11: Now we will create the 1st interaction term by interacting Privacy and Authentication.

For Moderator Variable, click on

Authentication and for Independent
Variable, click on Privacy. Also, choose
either product indicator approach or
orthoganalization approach. However,
based on this model, this study intends to
try on orthoganalization approach. Lastly,
click on Mean-Centered/ Standardized.

Again, we would like to highlight that when the moderator variable is a continuous variable i.e.
actual measurement (objective) or scaled items (Likert scale), we need to use the “Interact
effect term generation” option of “mean-center indicator values before multiplication”. This is
done to reduce a problem called multicollinearity. Multicollinearity is a well-known problem that
arises in the context of modeling moderating effects through multiplicative terms, which can
lead to serious computational problems (Cohen 1978; Pedhazur 1982). However, in contrast to
multicollinearity between two theoretically different variables, which has to be acknowledged
by the researchers, “[t]he multicollinearity in the context of regression with higher order terms
[which comprise interaction terms; the authors] is due to scaling, and can be greatly lessened by
centering variables” (Aiken and West 1991, p. 35). This is also clearly discussed by Marquardt
(1980), Smith and Sasaki (1979), and Tate (1984).

The second reason to center the indicators of the independent and the moderator variable lies
in the interpretation of the single effects. The coefficient b represents the slope of the regression
of X on Y when M has a value of zero. If zero is not an existing value on the scale of M, the
reference point would not be a particularly sensible choice. Centering is an appropriate means
of shifting the reference point to the mean and facilitating the interpretation of the parameters
(Aiken and West 1991; Finney et al. 1984).

Step 12: Once you Click Ok, you will get a new latent construct as follows (Moderating Effect 1):

Step 13: Name the New Construct of Moderating Effect 1 to Priv*Aut.

Step 14: Now interaction effect has the name of CCA*CB.

Step 15: From the model, the interaction effect has no indicators. To see the indicators, we can
click right on the interaction term and choose “Show Indicators of Selected Construct”.

Step 16: Now, researchers will see that the system has automatically created 10 interaction
terms and it is better for researchers to click back at “Hide Indicators of Selected Construct” to
have a neat and organized model.

Step 17: Now we will run the analysis using PLS Algorithm again.

Step 18: Look at the Algorithm result of R2

Check on the increase result of the R2 change

In moderation analysis, the R2 change becomes an important issue. As such, we will first look at
the R2 change from the main effect model. If you recall, the previous R2 for the main effect model
is 0.318, now in the interaction effect model, the R2 is 0.350. The R2 change of 0.049indicates that
with the addition of one interaction term, the R2 has changed about 4.9% (additional variance).
Next we need to calculate the effect size (f2) using the formula given below:

𝑹𝟐 𝒊𝒏𝒄𝒍𝒖𝒅𝒆𝒅 𝒎𝒐𝒅𝒆𝒓𝒂𝒕𝒐𝒓 − 𝑹𝟐 𝒆𝒙𝒄𝒍𝒖𝒅𝒆𝒅 𝒎𝒐𝒅𝒆𝒓𝒂𝒕𝒐𝒓

𝒇𝟐 =
𝟏 − 𝑹𝟐 𝒊𝒏𝒄𝒍𝒖𝒅𝒆𝒅 𝒎𝒐𝒅𝒆𝒓𝒂𝒕𝒐𝒓

We can calculate the f 2 by filling in the two R2 values which we have already noted down or
researchers can use the Effect Size Calculator or the result produced in SmartPLS, which will be
explained in the 2nd example.

𝟎. 𝟑𝟓𝟎 − 𝟎. 𝟑𝟏𝟖
𝒇𝟐 = = 𝟎. 𝟎𝟒𝟗𝟐
𝟏 − 𝟎. 𝟑𝟓𝟎

Check on the increase

result of the R2 change
via F square

Now, we can follow the interpretation of the f2 is by following the guidelines given in Cohen
(1988, p. 410 - 414), that values of 0.02, 0.15 and 0.35 represent small, medium and large effect
size. However, Aguinis, Beaty, Boik and Pierce (2005) have shown that the average effect size in
tests of moderation is only 0.009. Against this background, Kenny (2016) proposes that 0.005,
0.01, and 0.025 constitute more realistic standards for small, medium, and large effect sizes,
respectively, but also points out that even these values are optimistic given Aguinis et al., (2005)
review. Thus, it still depends on researchers to follow their appropriate guideline, when
reporting effect size.

So based on the f2 of 0.0492 we can conclude that the effect size is small (Cohen, 1988). Chin et
al. (2003) state that a low effect size f2 does not necessarily imply that the underlying moderator
effect is negligible: “Even a small interaction effect can be meaningful under extreme
conditions, if the resulting beta changes are meaningful, then it is important to take these
conditions into account” (Chin, Marcolin & Newsted, 2003; p. 211).

Although our result of the beta coefficient for the interaction effect of CCA*CB beta is 0.089 (see
Step 18), we would still unable to confirm whether the beta is either statistically significant or
not. Thus, we will proceed to the bootstrapping procedure to see the whether the interaction
effect is significant.

Step 19: Now, we can proceed with the interaction effects test using the bootstrapping
procedure which will give us the t-values. Click on Calculate and followed by Bootstrapping.

Step 20: Click on Start Calculation. Once the bootstrapping results are out, we can then confirm
whether the interaction terms are significant in one-tailed test.

Step 21: Bootstrapping Result

From the bootstrapping result, we will only interpret the interaction term of Priv*Aut. The cut
off value for this particular test will be 1.645 (α = 0.05) and 2.33 (α = 0.01). As can be seen from
the result, the interaction term of Priv*Aut is significant (t-value=2.521).

Step 22: Extract the Bootstrapping Result by Clicking Copy to Clipboard via Excel Format.

Original Sample Standard T Statistics P Values

Sample Mean (M) Deviation (|O/STDEV|)
(O) (STDEV)
Authentication -> Trust 0.566 0.568 0.040 14.305 0.000
Priv*Aut -> Trust 0.089 0.095 0.035 2.521 0.006
Privacy -> Trust -0.056 -0.032 0.066 0.839 0.201

Step 23: For reporting purpose, we can summarize the results in Table 13.3.

Table 13.3: Results of the Moderator analysis

Hypothesis Relationship Std. Beta Std. Error t-value

H1 Priv*Aut -> Trust 0.089 0.095 2.521**
Note: **p< 0.01, *p< 0.05

As we can see in Table 13.3, the interaction between Priv*Aut is positive. Thus, we can say that
the negative relationship between Privacy and Trust would be stronger when Authentication is
lower. However, it is not entirely clear how it differs. If you get a positive coefficient, the positive
coefficient of the interaction term suggests that it becomes more negative as Authentication
decreases; however, the size and precise nature of this effect is not easy to define from
examination of the coefficients alone, and it becomes even more so when one or more of the
coefficients are positive, or when the standard deviations of X and Z are very different (Dawson,
2014).

Thus, Dawson (2014) suggested that to follow up for the significant interactions, an interaction
plot can be drawn. Previously in SPSS, drawing this interaction plot would be quite tedious as
you need to run the descriptive for the IV and the Moderator to get a value to split the variable
into High/Low before you can plot.

Recently there have been some advances to this effect and there are several templates and
software that are available to make this process much simpler. One of them is by Professor
Jeremy Dawson which can be downloaded from this URL: www.jeremydawson.co.uk/slopes.htm

As you can see from above, there are 3 excel templates available. The first one is when you want
to do a continuous moderator and you want to use the unstandardized beta coefficients (2-way
unstandardised.xls). The second one is still the same, the moderator is continuous but you would
like to use the standardized beta coefficients (2-waystandardised.xls). The third one is when
your moderator variable is binary (2-way with binary moderator.xls)

Step 24: Since our moderator is continuous measurement and the beta in PLS analysis is
standardized beta, hence, researchers should choose the standardized beta coefficients (2-
waystandardised.xls) for Interaction Plot.

Note: There are several values that are needed for the plotting and they are the beta coefficient for
independent variable, moderator and interaction.

Step 25: Researchers should recall the results of the beta coefficient that are obtained earlier.

Note: Only take consideration of the significant moderator beta coefficient result before
plotting.

In this example, Priv*Aut has the significant interaction result. Thus, we will summarize the
values from the algorithm as follows:

Constructs Coefficients
Privacy -0.056
Authentication 0.566
Priv*Aut 0.089

Step 26: Researchers should plot the Interaction Results.

Alternatively, researchers can also look for Simple Slope Analysis Result that visualizes the two-
way interaction effect for SmartPLS 3.2.6.

First, look back on the PLS Algorithm

result that is generated with the
interaction term of Priv*Aut.

Then, scroll to the bottom and click on

Simple Slope Analysis. Then, look on the
significant moderator result that is
Priv*Aut.

Note: The three lines shown in the Simple Slope Analysis represent the relationship between
Privacy (x-axis) and Trust (y-axis). The middle line represents the relationship for an average level
of the moderator variable Authentication. The other two lines represent the relationship
between Privacy and Trust for higher (i.e., mean value of Authentication plus one standard
deviation unit) and lower (i.e., mean value of Authentication minus one standard deviation unit)
levels of the moderator variable Authentication.

Step 27: Interpret the Interaction Result

Figure 13.16: Graphing Plot Result

Just to recall the hypothesis we have written earlier:

H1: The negative relationship between Privacy and Trust will be stronger when Authentication is
lower.

The interpretation of the interaction plots is to look at the gradient of the slopes. As can be seen
in Figure 13.16, the line labelled Low Authentication has a steeper gradient compared to the High
Authentication indicating that the negative relationship of Privacy and Trust is indeed stronger
when Authentication is low. Thus, our hypothesis is supported with the indication of the
significant result of t-value.

13.6 Guidelines for Researchers and Reviewers

Figure 13.17 summarizes the guidelines for all three major approaches of moderation analysis.
Moreover, Table 13. 4 provides a brief description of each of the problems that researchers need
to be pre-cautious before executing moderating effects. Note, the simultaneous presence of
two or more of these problems as stated in Table 13.4, usually precludes researchers from
reaching accurate conclusions about the presence of moderating effects (e.g., Aguinis,
Culpepper, & Pierce, 2010; Aguinis & Stone-Romero, 1997). Thus, information in this table can be
used as a resource for researchers conducting studies that involve moderation and a checklist
for reviewers who evaluate manuscripts reporting tests of moderation.

Figure 13.17: Guidelines for Creating the Interaction Term

Source: Extended Primer on Partial Least Squares

Structural Equation Modeling (PLS-SEM) Book (Hair et al., 2017) by author

Table 13.4: Problems Regarding Moderation Assessment, Detrimental Consequences for

Substantive Conclusions, and Proposed Solutions.
No Problems Detrimental Consequences for Proposed Solution
Substantive Conclusions
1 Lack of attention to Less than perfect reliability can lead Report reliability estimates for all
measurement error in tests of to incorrectly dismissing moderating predictors (including those
moderation. effects and to underestimating hypothesized to play the role of
existing ones. moderator variables); this practice is
particularly necessary when a
hypothesized moderating effect is not
found.
How to Cite:
Ramayah, T., Cheah, J., Chuah, F., Ting, H., & Memon, M. A. (2018). Partial least squares structural equation modeling (PLS-SEM) using
SmartPLS 3.0: An updated guide and practical guide to statistical analysis (2nd ed.). Kuala Lumpur, Malaysia: Pearson.
How to Cite:
Ramayah, T., Cheah, J., Chuah, F., Ting, H., & Memon, M. A. (2018). Partial least squares structural equation modeling (PLS-SEM) using
SmartPLS 3.0: An updated guide and practical guide to statistical analysis (2nd ed.). Kuala Lumpur, Malaysia: Pearson.
2 Tests of moderation assume Range restriction has an adverse Attempt to capture the full range of
that variable distributions impact on the accuracy of population scores of all variables
include the full range of possible substantive conclusions: Moderating involved in the analysis and, if not
values. effects can go undetected, and feasible and moderating effects are not
when detected, they are found or found to be small, information
underestimated. on sample and population variances
should be provided to rule out range
restriction as a plausible explanation for
the results obtained.
3 When testing hypotheses about Even if total sample size is very Collect data such that the number of
categorical moderators, a large large, unequal sample sizes across firms within each moderator based
total sample size is assumed to the moderator-based subgroups subgroup is similar (but keep in mind
be a sufficient condition for decrease statistical power and lead that this oversampling strategy may lead
adequate statistical power even to an underestimation of to an unrepresentative sample).
in the presence of unequal moderating effects.
sample size across moderator
based categories.
4 Statistical power is assumed to Statistical power is often insufficient A priori statistical power is necessary
be adequate due to small sample size, before collecting data to plan study
measurement error, range design, and post hoc statistical power
restriction, unequal sample size should be calculated in all cases when a
across moderator-based subgroups, moderating effect is not found to rule
and other artifacts and leads to false out the possibility that insufficient
negative decisions regarding power has led to the no-moderator
moderating effects. conclusion.

5 Moderator-based subgroups are This practice results in information The practice of artificially dichotomizing
created by categorizing loss, reduced statistical power to or polychotomizing continuous variables
continuous variables into detect moderating effects, and a should be discontinued.
subgroups such as ‘‘high’’ and downward bias in the size of
‘‘low’’ or categories above and estimated moderating effects.
below a distribution’s median
value (i.e., median split
procedure)

Source: Extracted from Improving Our Understanding of Moderation and Mediation in Strategic Management
Research by Aguinis, Edwards & Bradley (2016)

References
Aguinis, H., Beaty, J. C., Boik, R. J., & Pierce, C. A. (2005). Effect size and power in assessing moderating effects of
categorical variables using multiple regression: a 30-year review. Journal of Applied Psychology, 90(1), 94.
Aguinis, H., Culpepper, S. A., & Pierce, C. A. (2010). Revival of test bias research in preemployment testing. Journal
of Applied Psychology, 95(4), 648.
Aguinis, H., Edwards, J. R., & Bradley, K. J. (2016). Improving our understanding of moderation and mediation in
strategic management research. Organizational Research Methods, 1094428115627498.
Aguinis, H., & Stone-Romero, E. F. (1997). Methodological artifacts in moderated multiple regression and their
effects on statistical power. Journal of Applied Psychology, 82(1), 192.
Aiken, L. S., West, S. G., & Reno, R. R. (1991). Multiple regression: Testing and interpreting interactions. Sage.
Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinction in social psychological research:
Conceptual, strategic, and statistical considerations. Journal of personality and social psychology, 51(6), 1173.
Chin, W. W., Marcolin, B. L., & Newsted, P. R. (2003). A partial least squares latent variable modeling approach for
measuring interaction effects: Results from a Monte Carlo simulation study and an electronic-mail
emotion/adoption study. Information systems research, 14(2), 189-217.
Chin, W. W., Marcolin, B. L., & Newsted, P. R. (1996). A partial least squares latent variable modeling approach for
measuring interaction effects: Results from a Monte Carlo simulation study and voice mail emotion/adoption

How to Cite:
Ramayah, T., Cheah, J., Chuah, F., Ting, H., & Memon, M. A. (2018). Partial least squares structural equation modeling (PLS-SEM) using
SmartPLS 3.0: An updated guide and practical guide to statistical analysis (2nd ed.). Kuala Lumpur, Malaysia: Pearson.
How to Cite:
Ramayah, T., Cheah, J., Chuah, F., Ting, H., & Memon, M. A. (2018). Partial least squares structural equation modeling (PLS-SEM) using
SmartPLS 3.0: An updated guide and practical guide to statistical analysis (2nd ed.). Kuala Lumpur, Malaysia: Pearson.
study. In J. I. DeGross, S. L. Jarvenpaa, & A. Srinivasan (Eds.), Proceedings of the Seventeenth International
Conference on Information Systems (pp. 21–41). Cleveland, OH: Association for Information Systems.
Cohen, J. (1978). Partialed products are interactions; partialed powers are curve components. Psychological
Bulletin, 85(4), 858.
Cohen, J. (1988). Statistical Power Analysis for the Behavioral Science (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum
Associates.
Dawson, J. F. (2014). Moderation in management research: What, why, when, and how. Journal of Business and
Psychology, 29(1), 1-19.
Echambadi, R., & Hess, J. D. (2007). Mean-centering does not alleviate collinearity problems in moderated multiple
regression models. Marketing Science, 26(3), 438-445.
Fassott, G., Henseler, J., & Coelho, P. S. (2016). Testing moderating effects in PLS path models with composite
variables. Industrial Management & Data Systems, 116(9).
Finney, J. W., Mitchell, R. E., Cronkite, R. C., & Moos, R. H. (1984). Methodological issues in estimating main and
interactive effects: Examples from coping/social support and stress field. Journal of Health and Social behavior,
85-98.
Gujarati, D. (1970). Use of dummy variables in testing for equality between sets of coefficients in two linear
regressions: a note. The American Statistician, 24(1), 50-52.
Hair Jr, J. F., Hult, G. T. M., Ringle, C., & Sarstedt, M. (2017). A primer on partial least squares structural equation
modeling (PLS-SEM). Sage Publications.
Henseler, J., & Fassott, G. (2010). Testing moderating effects in PLS path models: An illustration of available
procedures. In Handbook of partial least squares (pp. 713-735). Springer Berlin Heidelberg.
Henseler, J., & Chin, W. W. (2010). A comparison of approaches for the analysis of interaction effects between latent
variables using partial least squares path modeling. Structural Equation Modeling, 17(1), 82-109.
Holmbeck, G. N. (1997). Toward terminological, conceptual, and statistical clarity in the study of mediators and
moderators: examples from the child-clinical and pediatric psychology literatures. Journal of consulting and
clinical psychology, 65(4), 599.
Kenny, D. A. (2016). Moderation. Retrieved from http://davidakenny.net/cm/moderation.htm
Lance, C. E. (1988). Residual centering, exploratory and confirmatory moderator analysis, and decomposition of
effects in path models containing interactions. Applied Psychological Measurement, 12(2), 163-175.
Little, T. D., Bovaird, J. A., & Widaman, K. F. (2006). On the merits of orthogonalizing powered and product terms:
Implications for modeling interactions among latent variables. Structural Equation Modeling, 13(4), 497-519.
Marquandt, D. (1980). You should standardize the predictor variables in your regression models. Discussion of: A
critique of some ridge regression methods. Journal of the American Statistical Association, 75(369), 87-91.
Nunnally, J. C., & Bernstein, I. H. (1994). The assessment of reliability. Psychometric theory, 3(1), 248-292.
Pedhazur, E. J. (1982). Multiple regression and behavioral science. Explanation and Prediction, 2.
Smith, K. W., & Sasaki, M. S. (1979). Decreasing Multicollinearity a Method for Models with Multiplicative
Functions. Sociological Methods & Research, 8(1), 35-56.
Spiller, S. A., Fitzsimons, G. J., Lynch Jr, J. G., & McClelland, G. H. (2013). Spotlights, floodlights, and the magic number
zero: Simple effects tests in moderated regression. Journal of Marketing Research, 50(2), 277-288.
Tate, R. L. (1984). Limitations of centering for interactive models. Sociological Methods & Research, 13(2), 251-271.
Wedel, M., & Kamakura, W. A. (2002). Introduction to the special issue on market segmentation. Intern. J. of
Research in Marketing, 19, 181-183.

Suggested Reading:
Ting, H., de Run, E.C., Cheah, J.-H. and Chuah, F. (2016), "Food neophobia and ethnic food consumption intention:
An extension of the theory of planned behaviour", British Food Journal, Vol. 118 No. 11, pp. 2781-2797.
Lim, X.-J., Cheah, J.-H., Waller, D.S., Ting, H. and Ng, S.I. (2019), "What s-commerce implies? Repurchase intention
and its antecedents", Marketing Intelligence & Planning, https://doi.org/10.1108/MIP-03-2019-0145
Jun-Hwa Cheah, Hiram Ting, Park Thaichon, Kim-Shyan Fam & Martin Bazylewich (2020) Can positioning strategies
help influence willingness to pay for office space? Evidence on the moderating effect of office space grade
and industry sector for occupiers of leased office space, Journal of Strategic Marketing, DOI:
10.1080/0965254X.2020.1733049

View publication stats

How To Use Spss A Stepbystep Guide To Analysis and Interpretation 4th Edition Brian Cronk PDF Download
No ratings yet
How To Use Spss A Stepbystep Guide To Analysis and Interpretation 4th Edition Brian Cronk PDF Download
23 pages
Pls-Sem Using Smartpls 3.0: Chapter 12: Assessment of Mediation Analysis
No ratings yet
Pls-Sem Using Smartpls 3.0: Chapter 12: Assessment of Mediation Analysis
29 pages
37610-Hair6 Im
100% (3)
37610-Hair6 Im
201 pages
(Ebook PDF) Introductory Econometrics: A Modern Approach 6th Edition PDF Download
100% (2)
(Ebook PDF) Introductory Econometrics: A Modern Approach 6th Edition PDF Download
48 pages
Monte Carlo Simulation
No ratings yet
Monte Carlo Simulation
21 pages
Smart Pls
No ratings yet
Smart Pls
23 pages
Sampling Theory and Method
No ratings yet
Sampling Theory and Method
22 pages
Malhotra Mr05 PPT 20
100% (1)
Malhotra Mr05 PPT 20
41 pages
Christer Thrane - Doing Statistical Analysis. A Student's Guide To Quantitative Research-Routledge (2023)
No ratings yet
Christer Thrane - Doing Statistical Analysis. A Student's Guide To Quantitative Research-Routledge (2023)
266 pages
Confirmatory Factor Analysis
100% (1)
Confirmatory Factor Analysis
38 pages
Romantizam
No ratings yet
Romantizam
283 pages
Smart PLS
100% (2)
Smart PLS
54 pages
Free Online Course On PLS-SEM Using SmartPLS 3.0 - Introduction
0% (1)
Free Online Course On PLS-SEM Using SmartPLS 3.0 - Introduction
73 pages
Globalization As An Experience
100% (1)
Globalization As An Experience
2 pages
SEM Boot Camp Day 1 Morning: Basics & Data Screening: James Gaskin James - Gaskin@byu - Edu
No ratings yet
SEM Boot Camp Day 1 Morning: Basics & Data Screening: James Gaskin James - Gaskin@byu - Edu
38 pages
Country of Origin Effects On Consumer Behavior
No ratings yet
Country of Origin Effects On Consumer Behavior
406 pages
2017 Book PartialLeastSquaresPathModelin
100% (1)
2017 Book PartialLeastSquaresPathModelin
434 pages
Smart PLS 2 Module
100% (1)
Smart PLS 2 Module
104 pages
Pls-Sem Using Smartpls 3: School of Management Research Seminar
No ratings yet
Pls-Sem Using Smartpls 3: School of Management Research Seminar
50 pages
Malhotra Mr05 PPT 18
No ratings yet
Malhotra Mr05 PPT 18
47 pages
2021 - Marketing Analytics Herramientas Estadísticas para Marketing y Comportamiento Del Consumidor Usando SPSS - Carvalho
No ratings yet
2021 - Marketing Analytics Herramientas Estadísticas para Marketing y Comportamiento Del Consumidor Usando SPSS - Carvalho
211 pages
Soal Bahasa Inggris Kelas Xi Semester 2
100% (10)
Soal Bahasa Inggris Kelas Xi Semester 2
8 pages
Sri Guru Tegh Bahadur Institute of Management & Information Technology (Ggsipu)
No ratings yet
Sri Guru Tegh Bahadur Institute of Management & Information Technology (Ggsipu)
63 pages
Strctural Equation Modeling
100% (1)
Strctural Equation Modeling
69 pages
Electromagnetic Testing-EMT Chapter 01 04 PDF
No ratings yet
Electromagnetic Testing-EMT Chapter 01 04 PDF
180 pages
Ggplot2 Elegant Graphics For Data Analysis (2016, Springer) PDF
No ratings yet
Ggplot2 Elegant Graphics For Data Analysis (2016, Springer) PDF
281 pages
Sas Semma
100% (1)
Sas Semma
39 pages
PLS-SEMs Most Wanted Guidance
No ratings yet
PLS-SEMs Most Wanted Guidance
26 pages
Factor Analysis by Diagrams PDF
No ratings yet
Factor Analysis by Diagrams PDF
6 pages
WS 7 IG I Chemistry
100% (1)
WS 7 IG I Chemistry
6 pages
Chemical Calligraphy
No ratings yet
Chemical Calligraphy
11 pages
(Memon, Ramaya Et Al., 2017) - Editorial - SEM Application in Malaysia
No ratings yet
(Memon, Ramaya Et Al., 2017) - Editorial - SEM Application in Malaysia
14 pages
Protective Relaying Principles and Applications 4 Ed Blackburn
No ratings yet
Protective Relaying Principles and Applications 4 Ed Blackburn
312 pages
Health Econometrics Using Stata 1nbsped 1597182281 9781597182287 Compress
No ratings yet
Health Econometrics Using Stata 1nbsped 1597182281 9781597182287 Compress
374 pages
Difference PLS-SEM and Covariance-Based CB-SEM
No ratings yet
Difference PLS-SEM and Covariance-Based CB-SEM
1 page
Sample Size and Sample Power Considerations in Factor Analysis
No ratings yet
Sample Size and Sample Power Considerations in Factor Analysis
24 pages
Driver Analysis and Product Optimization Using Bayesian Networks
No ratings yet
Driver Analysis and Product Optimization Using Bayesian Networks
27 pages
PLS Sem
No ratings yet
PLS Sem
45 pages
Partial Derivatives: Chapter 2: Function of Several Variables
No ratings yet
Partial Derivatives: Chapter 2: Function of Several Variables
26 pages
Customer Satisfaction Towards Patanjali Ayurvedic Products
No ratings yet
Customer Satisfaction Towards Patanjali Ayurvedic Products
42 pages
Best SEM STATA Menu StataSEMMasterDay2and3 PDF
No ratings yet
Best SEM STATA Menu StataSEMMasterDay2and3 PDF
58 pages
STM32Cube Firmware
No ratings yet
STM32Cube Firmware
42 pages
7871 - 管理运筹建模与求解 - 基于Excel VBA与Matlab
No ratings yet
7871 - 管理运筹建模与求解 - 基于Excel VBA与Matlab
333 pages
Calculating Basic Statistical Procedures in SPSS
No ratings yet
Calculating Basic Statistical Procedures in SPSS
161 pages
PLS-SEM or CB-SEM: Updated Guidelines On Which Method To Use
No ratings yet
PLS-SEM or CB-SEM: Updated Guidelines On Which Method To Use
17 pages
Dr. Anwar Al Assaf
No ratings yet
Dr. Anwar Al Assaf
24 pages
Structural Equation Modeling (Sem) : Kassa T. (PHD) Email: Tel
No ratings yet
Structural Equation Modeling (Sem) : Kassa T. (PHD) Email: Tel
76 pages
Finnish 2
No ratings yet
Finnish 2
10 pages
asset-v1-IIMBx QM901x 3T2015 Type@asset Block@w02 - C03
No ratings yet
asset-v1-IIMBx QM901x 3T2015 Type@asset Block@w02 - C03
6 pages
Jean-Michel Josselin_Benoît Le Maux - STATISTICAL TOOLS FOR PROGRAM EVALUATION_ methods and applications to economic policy, public ... health, and education (2018, Springer International Publishing, Cham) - libgen
No ratings yet
Jean-Michel Josselin_Benoît Le Maux - STATISTICAL TOOLS FOR PROGRAM EVALUATION_ methods and applications to economic policy, public ... health, and education (2018, Springer International Publishing, Cham) - libgen
622 pages
Structural Equation Model
No ratings yet
Structural Equation Model
46 pages
Factor Analysis Xid-2898537 1 BSCdOjdTGS
No ratings yet
Factor Analysis Xid-2898537 1 BSCdOjdTGS
64 pages
4-1 - How To Use SmartPLS Software - Structural Model Assessment - 3-5-14
No ratings yet
4-1 - How To Use SmartPLS Software - Structural Model Assessment - 3-5-14
23 pages
Factor Analysis
No ratings yet
Factor Analysis
36 pages
Multivariate Data Analysis Joseph F. Hair Jr. William C. Black Barry J. Babin Rolph E. Anderson Seventh Edition
0% (1)
Multivariate Data Analysis Joseph F. Hair Jr. William C. Black Barry J. Babin Rolph E. Anderson Seventh Edition
7 pages
Free Online Course On PLS-SEM Using SmartPLS 3.0 - Mediator
No ratings yet
Free Online Course On PLS-SEM Using SmartPLS 3.0 - Mediator
24 pages
Scaffolding and Reading Comprehension
No ratings yet
Scaffolding and Reading Comprehension
8 pages
Issues and Procedures in Adopting Structural Equation Modelling T
No ratings yet
Issues and Procedures in Adopting Structural Equation Modelling T
9 pages
Tutorial IBM SPSS Statistics
No ratings yet
Tutorial IBM SPSS Statistics
22 pages
Slave Narratives
100% (1)
Slave Narratives
180 pages
Customer Service Survey
No ratings yet
Customer Service Survey
3 pages
Factor Analysis
No ratings yet
Factor Analysis
8 pages
Ijem 8 4 625
No ratings yet
Ijem 8 4 625
13 pages
Using R For Introductory Econometrics
No ratings yet
Using R For Introductory Econometrics
378 pages
Ordinance Creating Peoples Law Enforcement Board
No ratings yet
Ordinance Creating Peoples Law Enforcement Board
3 pages
An Intelligent Student Advising System: A Spanning Tree Approach
No ratings yet
An Intelligent Student Advising System: A Spanning Tree Approach
30 pages
Date Sheet BEd DEd June 2010
No ratings yet
Date Sheet BEd DEd June 2010
2 pages
A Christian Lifestyle in The Last Days
No ratings yet
A Christian Lifestyle in The Last Days
16 pages
Multiple Regression Analysis Using SPSS Laerd
No ratings yet
Multiple Regression Analysis Using SPSS Laerd
14 pages
Sample Size Dr. Karmran
No ratings yet
Sample Size Dr. Karmran
5 pages
Structural Equation Modeling in Stata
No ratings yet
Structural Equation Modeling in Stata
10 pages
Abell Model-Business Modeling - (Chapter 2 MSO)
No ratings yet
Abell Model-Business Modeling - (Chapter 2 MSO)
35 pages
Quantitative and Qualitative Methods in Medical Education Research AMEE Guide No 90 Part II
No ratings yet
Quantitative and Qualitative Methods in Medical Education Research AMEE Guide No 90 Part II
12 pages
Structural Equation Models: The Basics
No ratings yet
Structural Equation Models: The Basics
15 pages
Impactofpeerinfluence
No ratings yet
Impactofpeerinfluence
12 pages
PLS SEMPrimer BookFlyer
No ratings yet
PLS SEMPrimer BookFlyer
3 pages
NCM 107 Reviewer
No ratings yet
NCM 107 Reviewer
16 pages
Nforme de Investigación Sobre El Tiempo Pasado en Inglés
No ratings yet
Nforme de Investigación Sobre El Tiempo Pasado en Inglés
4 pages
Activity 6
No ratings yet
Activity 6
2 pages
De Cuong On Tap Tieng 8 HK1!23!24
No ratings yet
De Cuong On Tap Tieng 8 HK1!23!24
7 pages
Chapter 3 - Logic Gates v2
No ratings yet
Chapter 3 - Logic Gates v2
21 pages
Fall-Rise Tone Rising Tone Falling Tone Level Tone
No ratings yet
Fall-Rise Tone Rising Tone Falling Tone Level Tone
1 page
Samantha Reference Letter
No ratings yet
Samantha Reference Letter
2 pages
Homework 4
No ratings yet
Homework 4
4 pages
Vikramaditya VI of Vengi
No ratings yet
Vikramaditya VI of Vengi
3 pages
Ibsen's Portrayal of Nora in ADH
No ratings yet
Ibsen's Portrayal of Nora in ADH
1 page
I Had Long Been Conquered. I Had Long Been Placed in The Position of The Admiring Child.
No ratings yet
I Had Long Been Conquered. I Had Long Been Placed in The Position of The Admiring Child.
7 pages
PLS-SEMPrimer BookFlyer
No ratings yet
PLS-SEMPrimer BookFlyer
3 pages
Makalah Intervening (Path Analysis)
No ratings yet
Makalah Intervening (Path Analysis)
22 pages
Structural Equation Modelling
No ratings yet
Structural Equation Modelling
36 pages

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.