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

net/publication/372627277

Chapter 7 Case Study Update Using SmartPLS 4

Chapter · July 2023

CITATIONS READS

0 1,493

4 authors:

Joe Hair G. Tomas M. Hult

Manchester Metropolitan University Michigan State University
55 PUBLICATIONS 17,650 CITATIONS 266 PUBLICATIONS 67,449 CITATIONS

SEE PROFILE SEE PROFILE

Christian M. Ringle Marko Sarstedt

Technische Universität Hamburg Ludwig-Maximilians-University of Munich
273 PUBLICATIONS 137,052 CITATIONS 278 PUBLICATIONS 123,099 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Christian M. Ringle on 26 July 2023.

The user has requested enhancement of the downloaded file.

 7
MEDIATION AND
MODERATION ANALYSIS
Case Study Illustration—Mediation
To illustrate the estimation of mediating effects, let us, again, consider the
extended corporate model in the SmartPLS software. If you do not have the
model readily available, please go back to the case study in the previous chap-
ter and import the SmartPLS Project file Corporate Reputation.zip by going to
Files → Import project from backup file in the SmartPLS menu. Alternatively,
you see in the main screen next to the Workspace view, under Sample projects,
the Corporate reputation—PLS-SEM book (primer) example. After clicking on
the link with the label Install next to this sample project, the Example—Cor-
porate reputation (primer) project will appear in the Workspace. Next, double-
click on Extended model, and the extended PLS path model for the corporate
reputation example will open. The model shown in Exhibit A7.1 will appear
in the Modeling window.
In the following discussion, we will further explore the relationship be-
tween the two dimensions of corporate reputation (i.e., COMP and LIKE)
and the target construct customer loyalty (i.e., CUSL). According to Festin-
ger’s (1957) theory of cognitive dissonance, customers who perceive that a
company has a favorable reputation are likely to show higher levels of satisfac-
tion in an effort to avoid cognitive dissonance. At the same time, previous re-
search has demonstrated that customer satisfaction is the primary driver of
customer loyalty. Therefore, we expect that customer satisfaction mediates
the two relationships between likeability and customer loyalty as well as com-
petence and customer loyalty. To analyze these two mediation relationships in
the corporate reputation model in more detail, we apply the procedure shown
in Exhibit 7.5.
2 A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM)

Exhibit A7.1 ■ Extended Model in SmartPLS

 Chapter 7 ■ Mediation and Moderation Analysis 3

To begin the mediation analysis, we test the significance of the indirect ef-
fects. The indirect effect from COMP via CUSA to CUSL is the product of
the path coefficients from COMP to CUSA and from CUSA to CUSL (medi-
ation path 1). Similarly, the indirect effect from LIKE via CUSA to CUSL is
the product of the path coefficients from LIKE to CUSA and from CUSA to
CUSL (mediation path 2). To test the significance of these path coefficients’
products, we run the bootstrap routine. To do so, go to Calculate → Boot-
strapping in the SmartPLS menu or click on the Calculate icon, above the
Modeling window, followed by Bootstrapping (note that you first may need to
go back to the Modeling window before the Calculate icon appears). We re-
tain all settings for the PLS-SEM algorithm and the missing value treatment
as before by selecting 10,000 subsamples, Two tailed testing, and a signifi-
cance level of 0.05. In addition, make sure to select Do parallel processing,
Most important (faster), and Fixed seed. Next, we click Start calculation mak-
ing sure you have previously ticked Open report at the bottom of the dialog
box.
After running the bootstrapping procedure, SmartPLS opens the boot-
strapping results report. The table under Final results → Specific indirect ef-
fects provides us with an overview of results, including standard errors, boot-
strap mean values, t values, and p values. Clicking on the Confidence intervals
bias corrected tab in the bootstrapping results report shows the confidence in-
terval as derived from the percentile method (with bias correction). Similarly,
the table under Final results → Path coefficients offers the corresponding re-
sults for the direct effects, which we need in the further analysis. Exhibit A7.2
summarizes the bootstrapping results for the relationships between COMP
and CUSL as well as LIKE and CUSL. Alternatively, if you are interested in
the results for indirect effects of a serial or a joint mediation, open the boot-
strapping report Total indirect effects.
We find that both indirect effects are significant, since neither of the 95%
confidence intervals includes zero (see Chapter 5 for how to use confidence
intervals for hypothesis testing). When reporting the confidence intervals, it
is not necessary to also report the t values and p values. The next step of the
mediation analysis focuses on the significance of the direct effects from

Exhibit A7.2 ■ Analysis of the Direct and Indirect Effects

Direct 95% Confi- Signifi- Indirect 95% Confi- Signifi-
Effect dence Interval cance Effect dence Interval cance
(With Bias Cor- (p < 0.05)? (via CUSA) (With Bias (p < 0.05)?
rection) of the Correction)
Direct Effect of the Indirect
Effect

COMP → CUSL 0.006 [–0.100, 0.114] No 0.074 [0.006, 0.145] Yes

LIKE → CUSL 0.344 [0.236, 0.451] Yes 0.220 [0.153, 0.292] Yes
4 A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM)

COMP to CUSL and LIKE to CUSL. As determined in Chapter 6 and shown

in Exhibit A7.2, the relationship from COMP to CUSL is very weak (0.006)
and statistically nonsignificant. Following the mediation analysis procedure in
Exhibit 7.5, we conclude that CUSA fully mediates the COMP to CUSL rela-
tionship. On the contrary, LIKE exerts a pronounced (0.344) and significant
(p e 0.05) effect on CUSL. We therefore conclude that CUSA partially medi-
ates the relationship since both the direct and the indirect effects are signifi-
cant and meaningful (Exhibit 7.5). To further substantiate the type of partial
mediation, we next compute the product of the direct effect and the indirect
effect. Since the direct and indirect effects are both positive, the sign of their
product is also positive (i.e., 0.344 · 0.220 = 0.076). Consequently, CUSA
represents complementary mediation of the relationship from LIKE to CUSL.
Our findings provide empirical support for the mediating role of customer
satisfaction in the reputation model. More specifically, customer satisfaction
represents a mechanism that underlies the relationship between competence
and customer loyalty. Competence leads to customer satisfaction, and custom-
er satisfaction in turn leads to customer loyalty. For the relationship between
likeability and customer loyalty, customer satisfaction serves as a complemen-
tary mediator. Higher levels of likeability increase customer loyalty directly,
but also increase customer satisfaction, which, in turn, leads to customer loyal-
ty. Hence, some of likeability’s effect on loyalty is explained by satisfaction.

Case Study Illustration—Moderation

To illustrate the estimation of moderating effects, let’s consider the extended
corporate model again, as shown in Exhibit A7.1 earlier in this chapter. In the
following discussion, we focus on the relationship between customer satisfac-
tion and customer loyalty. Specifically, we introduce switching costs as a mod-
erator variable that can be assumed to negatively influence the relationship be-
tween satisfaction and loyalty. The higher the perceived switching costs, the
weaker the relationship between these two constructs. We use an extended
form of Jones, Mothersbaugh, and Beatty’s (2000) scale and measure switching
costs reflectively using four indicators (switch_1 to switch_4; Exhibit A7.3),
each measured on a 5-point Likert scale (1 = fully disagree, 5 = fully agree).

Exhibit A7.3 ■ Indicators for Measuring Switching Costs

switch_1 It takes me a great deal of time to switch to another company.

switch_2 It costs me too much to switch to another company.

switch_3 It takes a lot of effort to get used to a new company with its specific
“rules” and practices.

switch_4 In general, it would be a hassle switching to another company.

 Chapter 7 ■ Mediation and Moderation Analysis 5

We first need to extend the original model by including the moderator var-
iable. To do so, enter a new construct in the model (see Chapter 2 for detailed
explanations) and name it SC (i.e. switching costs). Next, we need to assign
the indicators switch_1, switch_2, switch_3, and switch_4 to the SC construct.
Please note that in case of only one moderator variable, like income or age,
you would use a single-item construct (i.e., a construct with only one indica-
tor variable that serves as a moderator; see Chapters 2 and 4).
In the next step, we need to add a moderating relationship from SC to the
path relationship between CUSA and CUSL. To do so, click on the Moderat-
ing effect button in the toolbar. Next, left-click on the moderator (SC) and
move the cursor on the directional arrow that connect CUSA with CUSL.
Exhibit A7.4 shows the model with SC as a moderator. When calculating the
results, SmartPLS automatically adds a relationship from the moderator to
the dependent variable of the moderated relationship, regardless of whether
the relationship was included in the model or not (as in Exhibit A7.4).
To compute the moderation results, SmartPLS performs the two-stage ap-
proach, which allows to disclose the significance of a moderating effect
(which is usually the case in PLS-SEM applications). Furthermore, the two-
stage approach is the most versatile approach, as it also works when the exog-
enous construct or the moderator are measured formatively.
The evaluation of the moderator variable’s measurement model shows that
the construct measures are reliable and valid. All indicator loadings are above
0.70, and the convergent validity assessment yields an AVE of 0.705, provid-
ing support for convergent validity of the switching cost moderator (SC). The
composite reliability ρ A has a value of 0.858, indicating internal consistency
reliability. In terms of discriminant validity, SC exhibits increased HTMT
values only with COMP (0.850) and LIKE (0.802). A further analysis of these
HTMT values uses Complete (slower) bootstrapping (percentile bootstrap,
10,000 subsamples, one-tailed testing with a 0.05 significance level, and stan-
dard settings for the PLS-SEM algorithm and missing value treatment). As dis-
cussed in Chapter 4, we assume the more conservative HTMT value of 0.85 for
all relevant construct combinations except COMP and LIKE as well as CUSA
and CUSL. For these pairs of constructs, we assume the higher 0.90 threshold
because of their conceptual similarity. For the additional SC construct, we
also consider conceptional similarity with LIKE, COMP, CUSA and apply a
higher HTMT threshold such as 0.90 or even 1. The results show that the
HTMT values are significantly lower (p < 0.05) than the critical cutoff values
of 0.85, except for SC and COMP as well as SC and LIKE. However, as the
corresponding HTMT values are significantly lower than 0.9 and 1, respec-
tively, we conclude that inclusion of the SC moderator in the model does not
entail discriminant validity problems.
Due to the inclusion of additional constructs in the path model (i.e., SC
and the interaction term), the measurement properties of all other constructs
in the path model will change (even though changes will likely be marginal).
6 A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM)

Exhibit A7.4 ■ Moderator Analysis Results in SmartPLS

Reanalyzing all measurement models provides support for the measures’ reli-
ability and validity. Note that the measurement model results shown in the
Modeling window stem from Stage 1 of the two-stage approach. The structur-
al model results, however, stem from Stage 2 of the two-stage approach when
all constructs are measured with single items.
Our next concern is with the size of the moderating effect. As can be seen
in Exhibit A7.4, the interaction term has a negative effect on CUSL (–0.071),
whereas the simple effect of CUSA on CUSL is 0.467. Jointly, these results
suggest that the relationship between CUSA and CUSL is 0.467 for an aver-
age level of switching costs. For higher levels of switching costs (e.g., SC is in-
creased by one standard deviation unit), the relationship between CUSA and
CUSL decreases by the size of the interaction term (i.e., 0.467 – 0.071 =
0.396). On the contrary, for lower levels of switching costs (e.g., SC is de-
creased by one standard deviation unit), the relationship between CUSA and
CUSL becomes 0.467 + 0.071 = 0.538. To better comprehend the results of
the moderator analysis, go to Final results → Simple slope analysis. The simple
 Chapter 7 ■ Mediation and Moderation Analysis 7

slope plot that follows visualizes the two-way interaction effect (Exhibit
A7.5).
The three lines shown in Exhibit A7.5 represent the relationship between
CUSA (x-axis) and CUSL (y-axis). The middle line represents the relationship
for an average level of the moderator variable SC. The other two lines repre-
sent the relationship between CUSA and CUSL for higher (i.e., mean value of
SC plus one standard deviation unit) and lower (i.e., mean value of SC minus
one standard deviation unit) levels of the moderator variable SC. As we can
see, the relationship between CUSA and CUSL is positive for all three lines as
indicated by their positive slope. Hence, higher levels of customer satisfaction
go hand in hand with higher levels of customer loyalty.

Exhibit A7.5 ■ Simple Slope Plot in SmartPLS

In addition, we can analyze the moderating effect’s slope in greater detail.

The upper line, which represents a high level of the moderator construct SC,
has a flatter slope while the lower line, which represents a low level of the
moderator construct SC, has a steeper slope. This makes sense since the inter-
action effect is negative. As a rule of thumb and an approximation, the slope
of the high level of the moderator construct SC is the simple effect (i.e.,
0.467) plus the interaction effect (–0.071), while the slope of the low level of
the moderator construct SC is the simple effect (i.e., 0.467) minus the inter-
action effect (–0.071). Hence, the simple slope plot supports our previous
discussion of the negative interaction term: higher SC levels entail a weaker
relationship between CUSA and CUSL, while lower levels of SC lead to a
stronger relationship between CUSA and CUSL.
 8 A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM)

Next, we assess whether the interaction term is significant. For this pur-
pose, we run bootstrapping with the Percentile bootstrap procedure, 10,000
Subsamples, Two tailed testing, and a significance level of 0.05. In addition,
make sure to select Do parallel processing, Most important (faster), and Fixed
seed. The analysis yields a p value of 0.024 for the path linking the interaction
term SC × CUSA and CUSL. Similarly, the 95% percentile bootstrap bias
corrected confidence interval of the interaction term’s effect is [–0.134,
–0.010]. As the confidence interval does not include zero, we conclude that
the interaction effect is significant, providing support for the existence of a
moderating effect. Overall, these results provide clear support that SC exerts a
significant and negative effect on the relationship between CUSA and CUSL.
The higher the switching costs, the weaker the relationship between customer
satisfaction and customer loyalty.
For the completeness of the results representation, the final step addresses
the moderator’s f 2 effect size. Recall that Kenny (2018) defines interaction
term effect sizes of 0.005, 0.01, and 0.025 as small, medium, and large. By
going to Quality criteria → f-square in the SmartPLS algorithm results report,
we learn that the f 2 effect size of the interaction term (i.e., CUSA * SC) has a
value of 0.014 and, thus, a medium effect.

View publication stats