Industrial Management & Data Systems

An updated and expanded assessment of PLS-SEM in information systems research

Joe Hair, Carole L. Hollingsworth, Adriane B. Randolph, Alain Yee Loong Chong,
Article information:
To cite this document:
Joe Hair, Carole L. Hollingsworth, Adriane B. Randolph, Alain Yee Loong Chong, (2017) "An updated
and expanded assessment of PLS-SEM in information systems research", Industrial Management &
Data Systems, Vol. 117 Issue: 3, pp.442-458, https://doi.org/10.1108/IMDS-04-2016-0130
Permanent link to this document:
https://doi.org/10.1108/IMDS-04-2016-0130
Downloaded on: 21 January 2019, At: 13:16 (PT)
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

References: this document contains references to 46 other documents.

To copy this document: [email protected]
The fulltext of this document has been downloaded 2252 times since 2017*
Users who downloaded this article also downloaded:
(2018),"An assessment of the use of partial least squares structural equation modeling (PLS-SEM)
in hospitality research", International Journal of Contemporary Hospitality Management, Vol. 30
Iss 1 pp. 514-538 <a href="https://doi.org/10.1108/IJCHM-10-2016-0568">https://doi.org/10.1108/
IJCHM-10-2016-0568</a>
(2016),"Identifying and treating unobserved heterogeneity with FIMIX-PLS: part I – method",
European Business Review, Vol. 28 Iss 1 pp. 63-76 <a href="https://doi.org/10.1108/
EBR-09-2015-0094">https://doi.org/10.1108/EBR-09-2015-0094</a>

Access to this document was granted through an Emerald subscription provided by emerald-
srm:402646 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald
for Authors service information about how to choose which publication to write for and submission
guidelines are available for all. Please visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com
Emerald is a global publisher linking research and practice to the benefit of society. The company
manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as
well as providing an extensive range of online products and additional customer resources and
services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the
Committee on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for
digital archive preservation.

*Related content and download information correct at time of download.

 The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/0263-5577.htm

IMDS
117,3 An updated and expanded
assessment of PLS-SEM in
information systems research
442 Joe Hair
Mitchell College of Business, University of South Alabama, Mobile, Alabama, USA
Received 7 April 2016
Revised 1 June 2016 Carole L. Hollingsworth and Adriane B. Randolph
25 June 2016
Accepted 27 June 2016
Department of Information Systems,
Kennesaw State University, Kennesaw, Georgia, USA, and
Alain Yee Loong Chong
Nottingham University Business School China,
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

University of Nottingham Ningbo China, Ningbo, China

Abstract
Purpose – Following the call for awareness of accepted reporting practices by Ringle, Sarstedt, and Straub in
2012, the purpose of this paper is to review and analyze the use of partial least squares structural equation
modeling (PLS-SEM) in Industrial Management & Data Systems (IMDS) and extend MIS Quarterly (MISQ)
applications to include the period 2012-2014.
Design/methodology/approach – Review of PLS-SEM applications in information systems (IS) studies
published in IMDS and MISQ for the period 2010-2014 identifying a total of 57 articles reporting the use of or
commenting on PLS-SEM.
Findings – The results indicate an increased maturity of the IS field in using PLS-SEM for model complexity
and formative measures and not just small sample sizes and non-normal data.
Research limitations/implications – Findings demonstrate the continued use and acceptance of PLS-SEM
as an accepted research method within IS. PLS-SEM is discussed as the preferred SEM method when the
research objective is prediction.
Practical implications – This update on PLS-SEM use and recent developments will help authors to better
understand and apply the method. Researchers are encouraged to engage in complete reporting procedures.
Originality/value – Applications of PLS-SEM for exploratory research and theory development are
increasing. IS scholars should continue to exercise sound practice by reporting reasons for using PLS-SEM and
recognizing its wider applicability for research. Recommended reporting guidelines following Ringle et al. (2012)
and Gefen et al. (2011) are included. Several important methodological updates are included as well.
Keywords PLS, SEM, Information systems, Methods, PLS-SEM, Guidelines
Paper type General review

Introduction
Researchers at all levels seek to use standardized reporting techniques to demonstrate rigor
and allow replicability. Information systems (IS) researchers wishing to employ the method of
partial least squares structural equation modeling (PLS-SEM) are no different. This method has
been widely applied in the IS field (Gefen et al., 2011; Ringle et al., 2012) as well as other
disciplines, including marketing (Hair, Sarstedt, Ringle, and Mena, 2012), strategic management
(Hair, Sarstedt, Pieper, and Ringle, 2012), international management (Richter et al., 2016),
operations management (Peng and Lai, 2012), tourism (do Valle and Assaker, 2015), accounting
(Lee et al., 2011), group and organization research (Sosik et al., 2009), and family business
(Sarstedt et al., 2014), but does have limitations (Marcoulides et al., 2009). To ensure the method
Industrial Management & Data
Systems is properly applied and interpreted, it is important that IS researchers are familiar with the
Vol. 117 No. 3, 2017
pp. 442-458
context of the current conversation on PLS-SEM applications. Thus, it is useful to obtain a
© Emerald Publishing Limited better understanding of how PLS-SEM is being applied in the IS field following the call by
0263-5577
DOI 10.1108/IMDS-04-2016-0130 Ringle et al. (2012) for awareness and application of accepted reporting practices.
 The purpose of this paper is to review and analyze the recent applications of PLS-SEM in An updated
selected IS literature from 2010 to 2015. This paper builds on the work of Ringle et al. (2012), and expanded
which examined the use of PLS-SEM in MIS Quarterly (MISQ). We also include Industrial assessment of
Management & Data Systems (IMDS) that publishes a variety of IS studies and has broad
appeal and relevance to researchers. During the six-year period, IMDS published 58 studies PLS-SEM
using the PLS-SEM method vs 34 studies by MISQ. These findings demonstrate the
continued use and acceptance of PLS-SEM as an accepted research method within IS. Before 443
reporting our results, we reviewed recent developments in PLS-SEM.

Recent developments in PLS-SEM

Several software programs are available to execute PLS-SEM. Recent releases of the
software include options for executing multi-group analysis (Sarstedt et al., 2011), invariance
testing by means of the measurement invariance of composite models (Henseler et al., 2016),
linear and non-linear moderation, continuous moderators, confirmatory tetrad analysis
(CTA) (Gudergan et al., 2008), and partial least squares prediction-oriented segmentation
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

(PLS-POS) (Becker et al., 2013). Editors and reviewers increasingly are requesting these
types of analyses so the ability to easily execute them is a benefit to the researchers.
A somewhat surprising development is proposed approaches referred to as consistent PLS
(Bentler and Huang, 2014; Dijkstra, 2014; Dijkstra and Henseler, 2015a, b). The new approaches
adapt PLS-SEM to produce the same results as the common factor model of
CB-SEM. It is unclear why researchers would use these alternative approaches to PLS-SEM
when they could easily apply the much more widely recognized and validated CB-SEM method.
Dijkstra and Henseler (2015a, b) note that their approach supposedly corrects the deficiencies of
PLS, but that assumes there are in fact deficiencies. Indeed, scholars could just as easily label
differences in loadings and path coefficients as deficiencies of CB-SEM since the loadings are in
general lower than for PLS-SEM and the coefficients are somewhat higher. Basically, the
authors make the same mistake as many CB-SEM scholars when they assume that the common
factor model is the benchmark against which PLS-SEM should be compared – a situation
referred to as PLS bias or consistency at large, which in fact is not necessarily a bias. It is not
surprising that the PLS-SEM method produces parameter estimates that are not the same as
CB-SEM, since the algorithms are different and CB-SEM is based only on common
variance while PLS is based on total variance (Hair et al., 2016). In sum, these consistent PLS
methods are designed for situations in which the research objective is to obtain the
same results as CB-SEM, when is very seldom the objective when applying the established
PLS-SEM algorithm (Hair et al., 2016).
A final topic to summarize before discussing the results of our analysis of PLS-SEM IS
applications is the recent emergence of numerous additional reasons for choosing PLS over
CB-SEM. The rules of thumb for selecting each method are listed in Table I. Many
methodological developments have emerged for PLS-SEM in recent years and more are on
the way. These updates to the PLS-SEM method have extended its analysis capabilities
beyond CB-SEM, and as can be noted from Table I, there are quite a few situations where
PLS-SEM is the preferred method instead of CB-SEM. Moreover, several of the analysis
features of PLS-SEM are not possible with CB-SEM, including continuous moderators,
prediction with latent variable scores due to indeterminacy, and higher order constructs
with only two first-order constructs. In planning future research, we urge researchers to
consider all of the PLS-SEM analysis possibilities when specifying the research design and
selecting the method of structural modeling (Table I).

Methodology
Our review of PLS-SEM applications in IS consisted of empirical studies published in IMDS
and MISQ for the period 2012-2015. Papers focusing just on the PLS method and not
 IMDS PLS-SEM CB-SEM
117,3
1. The research objective is exploratory or confirmation 1. The research objective is confirmation of
of theory based on total variance well-developed structural and measurement
theory based on common variance
2. The objective of the analysis is prediction 2. The measurement philosophy is estimation with
the common factor model using only common
444 variance (covariances)
3. The measurement philosophy is estimation with the 3. The research requires a global
composite factor model using total variance goodness-of-fit criterion
4. The research objective is to explain the relationships 4. The error terms require additional specification,
between exogenous and endogenous constructs such as covariation.
5. The structural and/or measurement models are 5. The structural model specifies non-recursive
complex (many constructs ¼ 6 + and many relationships
indicators ¼ 50+)
6. Formatively measured constructs are specified 6. The structural and/or measurement models
in the research are simple (5 or fewer constructs and 50 or
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

fewer indicators)
7. Preferred method when sample size is small (no100).
But PLS is also an excellent method for larger samples
8. The data are not normally distributed
9. The scaling of responses is ordinal or nominal
10. The data is secondary/archival, particularly
single-item measures
11. The research objective is to use latent variable scores
in subsequent analyses
12. The structural model will be estimated with a higher
order construct that has only two first-order constructs
Table I. 13. The analysis involves a continuous moderator
Rules of thumb for 14. The investigation will examine the model for
choosing SEM method unobserved heterogeneity

presenting empirical results were not included in our study. The search identified a total of
92 articles (number of studies/publications) that reported the use of PLS-SEM, or
commented on the method (e.g. editorials). Note that findings reported for MISQ from prior
to 2012 are excerpted from the Ringle et al.’s (2012) assessment. This time period was chosen
because it reflects the most recent period when the applications of PLS-SEM have grown
dramatically, as noted in Figures 1 and 2.
The trends in publishing of articles using PLS-SEM are shown in Figures 1 and 2. The
number of PLS-SEM articles in IMDS increased considerably over the period, growing from
only seven in 2010 to 17 in 2015, for a total of 58. In comparison, a total of 34 PLS-SEM
articles were published in MISQ for that same period, which was the same as the previous
five-year period (2005-2009 ¼ 34). Thus, while the number of PLS-SEM articles in IMDS is
increasing, the number of similar articles in MISQ in recent years appears to be relatively
flat. This relatively flat pattern of application of PLS-SEM in MISQ may be due to
researchers exploring other types of analysis, such as hierarchical linear modeling.
The number of PLS-SEM articles as a proportion of total articles published in IMDS
from 2010 to 2015 is 13.0 percent (58 of 445), and the proportion for the same years in
MISQ is 10.9 percent (34 out of 312). Between 2010 and 2015, 29 editorials were included in
MISQ, and of those editorials four (13.8 percent) discussed PLS-SEM. Additionally,
from 1992 to 2015, there were 11.3 percent (88 of 776) PLS-SEM articles published as a
proportion of total articles in MISQ, and an MISQ special issue focused on PLS-SEM in
2009. Thus, overall there are comparable proportions of articles using PLS-SEM in both
journals during the relevant periods.
 IMDS 2010-2015 An updated
18 70
16
and expanded
60
14 assessment of

Cumulative total
50 PLS-SEM
Yearly count

12
10 40
8 30
6
445
20
4
2 10
0 0
Figure 1.
2010 2011 2012 2013 2014 2015 PLS-SEM articles in
IMDS for the period
Year
2010-2015
Number of PLS studies per year Cumulative number of PLS studies
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

MISQ 1992-2015

Cumulative
10 100
Yearly
count

total
5 50

0 0 Figure 2.
92

93

94
94
96
97
98
99
00
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
PLS-SEM articles in
19

19

19
19
19
19
19
19
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
MISQ for the period
Year 1992-2015
Number of PLS studies per year Cumulative number of PLS studies

Critical issues in PLS-SEM IS applications

Each article that applied PLS-SEM was evaluated according to a wide range of criteria,
including reasons for employing PLS-SEM, use of reflective and formative indicators,
reliability and validity assessments, multi-group analyses, supplementary analyses, and
the software used. The information was systematically recorded in a database for analysis
in this study. As a result, we are able to update and expand knowledge about applications
of PLS-SEM, as well as trace important developments in the application of PLS-SEM in IS
over the past six years. In addition, the analysis enabled us to identify critical issues and
typical misapplications of the method. The review focuses on eight critical issues in the
application of PLS-SEM identified by Ringle et al. (2012), including: reasons given for
using PLS-SEM, model descriptive statistics, sampling characteristics, technical
reporting, formative measurement metrics, reflective measurement metrics, structural
model metrics, and additional analyses such as mediation, moderation, multi-group
analyses, and common methods variance. Findings for the two journals are shown to
provide a basis of comparison. Where possible, we also suggest best practices on using
PLS-SEM as guidelines for future research.

Reasons for using PLS-SEM

For IMDS, the most common reason for using PLS-SEM is by far small sample size, with the
second most common reason being non-normal data. When compared to MISQ PLS-SEM
applications, however, model complexity and formative constructs are now the two most
common reasons, with exploratory research and small sample size tied the third. It should
also be noted that the other category for MISQ includes numerous reasons, such as:
maximizing explained variance of endogenous variables, number of interaction terms, large
number of variables, measures derived from archival data, explains an outcome of interest,
identifying relationships, second-order constructs, multiple interdependent relationships,
 IMDS multiple criterion variables, early-stage research, and mediated models. Thus, the dominant
117,3 reason for using PLS-SEM in MISQ is no longer simply small sample size. This seems to
indicate that the IS community is heeding recommendations from the Ringle et al.’s (2012)
assessment and maturing in its use of PLS-SEM.
With regard to IMDS, theory development is a widely specified reason, whereas MISQ
authors specify theory development much less often. IMDS also did not have any papers
446 that cited categorical variables, theory testing, or convergence ensured as reasons for
selecting PLS-SEM. This lack of specifying these reasons is comparable to the updated
MISQ counts. Overall, 55 of 58 papers in IMDS cite a specific reason for the choice of
PLS-SEM, while all 24 of the papers in MISQ provide a reason. Thus, continuing to model
the practices of one of the leading journals in the field by reporting specific reasons for using
PLS-SEM will be helpful for other researchers when they report their work.
A somewhat surprising finding was the low indication of choosing PLS-SEM for
prediction. The latent variable scores for CB-SEM are indeterminant – i.e., an infinite number
of different sets of latent variable scores that will fit the model equally well are possible for a
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

CB-SEM solution, which makes CB-SEM unsuitable for prediction (Hair et al., 2016, 2018).
In contrast, the PLS-SEM method always produces a single determinant score for each SEM
composite for each observation. Moreover, CB-SEM prediction as reported in the R2 is the
proportion of common variance explained, whereas PLS-SEM R2 is the proportion of total
variance explained (Hair et al., 2018). Thus, PLS-SEM is always the preferred SEM method
when the research objective is prediction and we believe this reason for selecting PLS-SEM
rather than CB-SEM will increase considerably in the future (Table II).

Confirmatory or exploratory research

When deciding whether PLS-SEM is the appropriate structural modeling approach,
researchers often question whether their research is confirmatory or exploratory research.
This is an important concept to understand when analyzing data with SEM since a general
rule of thumb for choosing CB-SEM is that it should be used for confirmatory research while

IMDS 2010-2015 MISQ 2012-2015 MISQ 1992-2011

Number Proportion Number Proportion Number Proportion
(n ¼ 58) (%) (n ¼ 24) (%) (n ¼ 65) (%)

Total 55 94.8 24 100.0 46 70.8

Specific reasons
Small sample size 23 39.7 3 12.5 24 36.9
Non-normal data 13 22.4 2 8.3 22 33.9
Formative measures 6 10.3 6 25.0 20 30.8
Focus on prediction 2 3.4 2 8.3 10 15.4
Model complexity 5 8.6 4 16.7 9 13.9
Exploratory research 3 5.2 3 12.5 7 10.8
Theory development 8 13.8 1 4.2 6 9.2
Use of categorical
variables – – – – 4 6.2
Convergence ensured – – – – 2 3.1
Theory testing 1 – – – 1 1.5
Interaction terms 4 6.9 2 8.3 1 1.5
Other/not specified 14 24.1 10 41.7 – –
Table II. Notes: Counts are based on the authors’ comments as to their explicit reasons for using PLS-SEM. If they did
Reasons for using not specify a reason, it is counted under other/not specified. Some articles specified more than one reason and
PLS-SEM all reasons are included
 PLS-SEM is preferred for exploratory research, but can also be used for confirmatory An updated
research (Hair et al., 2016, 2018). and expanded
To clarify this issue, it is useful to clarify how exploratory research differs from assessment of
confirmatory research. Exploratory research is conducted when problems have not been
clearly defined. Researchers may not have enough information to make conceptual PLS-SEM
distinctions or to propose explanatory relationships, and the approach to the problem must
be flexible (Hair et al., 2010). Exploratory research can be used to generate hypotheses from 447
qualitative methods, but it also is used to test hypotheses using quantitative research.
For example, when hypotheses are generated by research in another context, e.g., in the
USA, you may focus on testing the same or similar hypotheses in another country,
e.g., China, or Malaysia. Thus, exploratory research can address all types of research
questions, including what, when, why, and how. In contrast, confirmatory research
examines previously specified hypotheses that predict specific outcomes based on
underlying causal theory, and the hypotheses usually are derived from established causal
theories or previous studies conducted within the same context (Hair et al., 2010).
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

It is useful at this point to clarify the different types of modeling. Predictive modeling is
the process of applying a statistical model to examine data with the objective of predicting
new or future observations (Shmueli, 2010). This statistical modeling process leads to
“statistical conclusions” in terms of explained variance, statistical significance, and effect
sizes as they relate to theoretical hypotheses. It should also be noted that in social sciences
research the types of statistical models used for testing theoretical causal hypotheses are
most often correlation-based models applied to observational/survey data. Thus, statistical
modeling is not only predictive modeling, it is also explanatory modeling, or the application
of statistical modeling to data to test and explain causal hypotheses about theoretical
constructs and structural paths.
In addition to predictive and explanatory modeling, both of which can be accomplished
with either CB-SEM or PLS-SEM, a third type of modeling is descriptive. In descriptive
modeling there is no underlying causal theory, except in perhaps a limited way (Shmueli,
2010). Therefore, when researchers state they are conducting exploratory research they are
representing relationships between data structures, e.g., independent and dependent
variables, in a way that summarizes those relationships. In short, they are exploring
possible relationships not based on theoretical or causal justification, but rather searching
for potential associations that may lead to theory development.
In applying the rule of thumb for selecting the appropriate SEM method, the general
definitions of the concepts must be extended to a multivariate statistical context. SEM has
two stages – confirmatory factor analysis (CFA) and structural equation modeling. CFA
enables the researcher to test the hypothesis that theoretical relationships (sometimes
referred to as causal) actually exist between the observed indicator variables and their
underlying latent constructs. In short, a CFA assesses measurement theory. Note that when
running a CFA with PLS-SEM the process is referred to as confirmatory composite analysis
(Henseler et al., 2014). The second SEM stage, structural equation modeling, tests whether
the theoretical structural relationships (also sometimes referred to as causal) between the
constructs are meaningful and significant. In short, this second stage examines structural
theory. The hypothesized measurement and structural model relationships are based on
theory and/or previous empirical research, either qualitative or quantitative, and the
hypotheses are tested statistically.
CB-SEM and PLS-SEM are both used to conduct a CFA to assess, and perhaps confirm,
theoretical measurement theory and structural model relationships. In evaluating the results in
the CFA stage, the only difference between the two methods is CB-SEM that is assessed on the
basis of reliability, convergent validity, and discriminant validity, as well as on how well the
relationships between the indicator variables, as represented by the observed covariance matrix,
 IMDS can be reproduced. The extent to which these relationships can be reproduced is referred to as
117,3 “goodness-of-fit” (Hair et al., 2010). In contrast, PLS-SEM is not based on covariances and thus
does not have a fit measure. The CFA stage in PLS-SEM does, however, test the hypothesis that
theoretical relationships actually exist between the observed indicator variables and their
underlying latent constructs. With PLS-SEM, to confirm the CFA hypotheses for the
measurement model the only metrics applied are reliability, convergent validity, and
448 discriminant validity (Hair et al., 2016, 2018). A similar situation is present when examining the
structural model relationships. For both methods, the size and significance of the path
coefficients are assessed, but in addition when using CB-SEM researchers must also assess fit.
To summarize, when deciding whether to use CB-SEM or PLS-SEM, researchers should
understand that because CB-SEM is based on covariances (only common variance) and
requires fit, the method is suitable only for confirmatory research that is based on
well-developed theory. In contrast, PLS-SEM is based on total variance and is a useful method
for both exploratory and confirmatory research. While PLS-SEM is generally thought of as
only for exploratory research, it is also very useful for confirmatory research. The primary
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

difference is that CB-SEM is confirmed with reliability, validity, and goodness-of-fit metrics,
whereas PLS-SEM is confirmed with reliability and validity metrics only.

Model descriptive statistics

The average number of latent variables has increased in MISQ. MISQ articles continue to
include models with a large number of variables, with the Ringle et al.’s (2012) editorial
reporting a high number of 36 latent variables and the updated assessment for 2012-2015
identifying a high number of 25 latent variables. IMDS includes primarily reflectively
measured constructs, has only two papers reporting both formative and reflective constructs,
and none with formative only. MISQ has a more balanced modeling mix; i.e., quite a few
reflectively measured only papers and also both reflective and formative, but also only a small
number of formative only constructs. More recently, there are considerably more model
specifications in MISQ that are combinations of both formative and reflective constructs.
Additional comparisons show similar numbers of total indicators in the models as
numbers of control variables. Much fewer single-item measures are reported in IMDS
compared to MISQ, and this is positive, as recent research raises serious concerns about the
validity of such measures (e.g. Diamantopoulos et al., 2012; Sarstedt, Diamantopoulos,
Salzberger, and Baumgartner, 2016; Sarstedt, Diamantopoulos, and Salzberger, 2016).
Similarly, there are far few higher order constructs than in MISQ. Since higher order
constructs are becoming more prevalent, we expect their number to increase in IMDS. As a
best practice, these finding indicate authors should include item wordings, scales, scale
means, standard deviations, and a correlation matrix (Table III).

Sampling characteristics
There are differences in sample characteristics between MISQ and IMDS. Small sample size
was indicated less frequently as a primary reason for using PLS-SEM in MISQ. Reported
sample sizes in IMDS do reflect smaller samples are being used in that journal, but the
difference in sample sizes is more pronounced in the recent five-year period. In fact, in IMDS
the “ten times rule” is not always met, whereas in the more recent period for MISQ the rule is
always met. Recall that the ten times rule indicates the sample size should be equal to the
larger of ten times the largest number of formative indicators used to measure a single
construct, or ten times the largest number of structural paths directed at a particular
construct in the structural model (Hair et al., 2016). With regard to testing for non-response
bias, comments are occasionally reported, and range from 27 percent in IMDS to 37 percent
in MISQ. Not surprisingly, validation via a holdout sample is very seldom conducted in
either journal, a pattern that is unfortunately typical for most journals (Table IV).
 IMDS 2010-2015 MISQ 2012-2015 MISQ 1992-2011
An updated
Proportion Number Proportion Number Proportion and expanded
Criterion Number (n ¼ 58) (%) (n ¼ 24) (%) (n ¼ 109) (%) assessment of
Number of latent variables PLS-SEM
Mean 6.72 8.83 8.12
Median 7 – 8 – 7 –
Range (2, 12) (4, 25) (3, 36) 449
Number of structural model relations
Mean 8.15 11.54 11.38
Median 8 – 8 – 8 –
Range (1, 22) (5, 28) (2, 64)
Mode of measurement models
Only reflective 52 89.7 10 41.7 46 42.2
Only formative 0 0.0 1 4.2 2 1.8
Reflective and formative 6 10.3 12 50.0 33 30.3
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

Not specified 0 0.0 0.0 28 25.7

Number of indicators per reflective constructa
Mean 11.87 8.75 3.58
Median 4.9 – 3.69 – 3.5 –
Range (1, 19) (1, 11) (1, 400)
Number of indicators per formative construct
Mean Only 2 included this; 5.17 3.03
1 and 13 indicators
Median – 3.3 – 3 –
Range (2, 7) (1, 11)
Total number of indicators in models
Mean 31.17 33.91 27.42
Median 29 – 30 – 26.5 –
Range (13, 65) (12, 81) (5, 1,064)
Number of models with
control variables 16 13 29
Number of control variables
Mean 2.59 – 6.00 3.69
Median 3 7 4
Range (1, 4) (3, 17) (1, 6)
Number of studies with
Single-item constructs 6 10.3 11 45.8 31 47.7
Higher order constructs 7 12.1 11 45.8 15 23.1
(i.e. hierarchical
component analysis
Non-linear relationships) 3 4.6
Model modified in the 0 0.0 3 12.5 18 27.7
course of analysis
If yes, comparison with 0 0.0 3 12.5 6 9.2
initial model?
Item wordings reported 44 75.9 21 87.5 58 89.2
Scales reported 53 91.4 23 95.8 55 84.6
Scale means and 23 39.7 20 83.3 43 66.2
standard deviations
reported
Correlation/covariance 52 89.7 20 83.3 54 83.1 Table III.
matrix Model descriptive
Note: an ¼ 56; two did not report this item statistics
 IMDS IMDS 2010-2015 MISQ 2012-2015 MISQ 1992-2011
117,3 Number Proportion Number Proportion Number Proportion
(n ¼ 58) (%) (n ¼ 24) (%) (n ¼ 109) (%)

Sample size
Mean 309.07 390.71b 238.12
Median 217.5 188b 198
450 Range (59, 736) (128, 1,512)b (17, 1,449)
Less than 100 observations 9 13.8 0 0 25 22.9
Ten times rule of thumb not met 6a 10.3 n/a 0 6 5.5
If not met, to what extent
(in percentages) was 35.0 22.5
The sample size below the
required N
According to the ten times rule?
Non-response bias 18 31.0 9 37.5 24 36.9
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

Holdout sample used 0 0.0 0 0.0 2 3.1

Missing values reported 3 5.2 0 0.0 10 15.4
Treatment of influential
observations (e.g. outliers)
reported 3 5.2 1 4.2 4 6.2
Non-normality reported
Table IV. (e.g. skewness, K-S test) 4 6.9 0 0.0 4 6.2
Sampling Notes: aIn total, 56 of the 58 articles included sample size; bSample size of 9,956 was excluded from mean,
characteristics median, and range calculations as it is an outlier

Technical reporting
A fairly high proportion of software users are violating the licensing requirements, which
specify that the specific package should be identified. Of the software packages reported,
SmartPLS (Ringle et al., 2005) is the most often one identified in IMDS but less than
50 percent report it. In contrast, there was a relatively even split between SmartPLS and
PLS graph (Chin, 2003), with both packages at about 35 percent in MISQ for the most
recent period. This suggests an increasing application of SmartPLS compared to the
previous study by Ringle et al. (2012), where PLS Graph was more widely used.
The prevalence of PLS Graph in the earlier MISQ period is a result of the fact that
SmartPLS was only available from 2005 on, whereas PLS Graph was released in the early
1990s. We expect the proportion of software users to be much higher for SmartPLS in the
future, since the current number of analysis features is much greater, but this may change
if the PLS Graph developers update that software. Slightly less than 50 percent of articles
in IMDS comment on bootstrapping, whereas a much higher proportion mentions it in
MISQ. Editors and reviewers should be urging authors to provide more details,
particularly for bootstrapping. Specific metrics to report for bootstrapping include the
number of bootstrap samples, significance levels, confidence intervals, and standard
errors (Hair et al., 2016) (Table V).

Reported formative measurement metrics

In general, formative measurement is not reported as being used very often in either journal, but
it has almost doubled in MISQ (32 to 54 percent) since the previous study by Ringle et al. (2012).
The findings for incorrect assessment of formative measurement models appear to be high, but
realistically the sample size is so small that it is risky to interpret it this way. The authors find
that quite a few individuals still do not understand that formative measurement models are not
 IMDS 2010-2015 MISQ 2012-2015 MISQ 1992-2011
An updated
Number Proportion Number Proportion Number Proportion and expanded
Number of studies reporting (n ¼ 58) (%) (n ¼ 24) (%) (n ¼ 65) (%) assessment of
Software used PLS-SEM
PLS Graph (Chin, 2003) 6 10.3 7 29.2 35 53.9
SmartPLS (Ringle et al., 2005) 32 55.2 10 41.7 2 3.1
LVPLS (Lohmöller, 1987) 0 0.0 0 0.0 1 1.5 451
Not Reported 20 34.5 7 29.2 27 41.5
Resampling method (e.g. bootstrapping)
Use mentioned 31 53.4 11 45.8 61 93.9 Table V.
Algorithmic options 0 0.0 0 0.0 24 36.9 Technical reporting

evaluated based on internal consistency measures such as composite reliability or AVE.

Downloaded by Tulane University At 13:16 21 January 2019 (PT)

The Hair et al.’s (2016) book includes a comprehensive coverage of this topic that should be
reviewed by individuals not familiar with the differences (Table VI).

Reported reflective measurement model metrics

Authors are reporting reflective measurement model metrics reasonably well. A very high
percentage of authors in both journals report indicator loadings. Further, for the
2011-2015 period well over one-half of the studies report both composite reliability
and Cronbach’s α, and for AVEs the percentage is much higher. Not surprisingly,
cross-loadings were reported more in the earlier MISQ assessment of discriminant
validity, but in the more recent assessment Fornell-Larcker is often the only criterion
in IMDS while both are typically reported in MISQ. This difference may be due to page
count limitations where IMDS limits authors to 8,000 words (which is estimated at
roughly 20 pages) while MISQ permits twice that at up to 40 pages for articles.
We recommend that authors no longer rely on cross-loadings or even Fornell-Larcker,

IMDS 2010-2015 MISQ 2012-2015 MISQ 1992-2011

Number Proportion Number Proportion Number Proportion
Empirical test criterion in PLS-SEM (n ¼ 4) (%) (n ¼ 13) (%) (n ¼ 65) (%)

Reflective criteria used to evaluate 4 100.0 5 38.5 5 14.3

formative constructs
Absolute indicator Indicator 4 100.0 11 84.6 24 36.9
contribution to the weights
construct
Significance of Standard 4 100.0 3 23.1 20 57.1
weights errors
Significance 8 61.5
levels
t-values/ 10 76.9
p-values for
indicator
weights
Multicollinearity Only VIF/ 4 100.0 11 84.6 9 25.7
tolerance Table VI.
Only condition 0 0.0 0 0.0 Reported formative
index measurement model
Both 0 0.0 0 0.0 metrics
 IMDS as both often substantially overstate the presence of discriminant validity. Instead,
117,3 authors should report the recently developed HTMT criterion for discriminant validity
(Henseler et al., 2015), particularly if space limitations are an issue (Table VII).

Reported structural model metrics

The basic structural model metrics of R2, path coefficients size, and significance are almost
452 always reported. The other important metrics including f2, Q2, and q2 are much less
frequently reported. Authors should always report these metrics to enable accurate
interpretation of the results. The small reporting frequency of confidence intervals and total
effects in both journals is a result of these metrics being required for only a few types of
analyses in previous research. For future research, there are new types of analyses, such as
mediation, that will necessitate reporting of these metrics (Table VIII).

Additional considerations and supplementary analyses

Downloaded by Tulane University At 13:16 21 January 2019 (PT)

For several years, an assessment of the presence of common methods variance has often
been required by the editors and reviewers in many journals. When common methods
variance assessment has been reported, it typically refers to procedures recommended by
Podsakoff et al. (2003). Almost 80 percent of articles in MISQ comment on common methods
variance, whereas only about 44 percent comment in IMDS. Assessments of common
methods bias may be reported in articles less often in the future, however, as recent research
by Fuller et al. (2016) and Babin et al. (2016) indicate common methods variance is not
present nearly as often as suggested in the past (Podsakoff et al., 2003). Moreover, this recent
research also concludes that the Harman’s (1976) one-factor test is in fact an acceptable
method of assessing common methods bias, so when reported the Harman test is likely to be
considered as an acceptable assessment tool for common methods variance.
An interesting finding is the increase in applications of mediation in the recent
assessment of MISQ (35.7 percent). In contrast, the previous MISQ assessment reported
mediation much less often (23.1 percent), but slightly more than the current applications in
IMDS (19.7 percent). We expect the application of mediation in PLS studies to increase in all

IMDS 2010-2015 MISQ 2012-2015 MISQ 1992-2011

Number Proportion Number Proportion Number Proportion
Empirical test criterion in PLS-SEM (n ¼ 58) (%) (n ¼ 24) (%) (n ¼ 79) (%)

Indicator Indicator 52 89.7 18 75.0 70 88.6

reliability loadings
Internal Only composite 8 13.8 6 25.0 45 57.0
consistency reliability
reliability
Only 7 12.1 0 0.0 8 10.1
Cronbach’s α
Both 42 72.4 15 62.5 22 27.9
Convergent AVE 1 53 91.4 20 83.3 70 88.6
validity
Other 0 0.0 0 0.0 9 11.4
Discriminant Only 42 72.4 8 33.3 29 36.7
validity Fornell-Larcker
criterion
Table VII. Only 3 5.2 3 12.5 7 8.9
Reported reflective cross-loadings
measurement model Both 6 10.3 11 45.8 33 41.8
metrics Other 0 0.0 0 0.0 3 3.8
 Empirical test IMDS 2010-2015 MISQ 2012-2015 MISQ 1992-2011
An updated
criterion in Number Proportion Number Proportion Number Proportion and expanded
Criterion PLS-SEM (n ¼ 61) (%) (n ¼ 24) (%) (n ¼ 109) (%) assessment of
Coefficient of R2 58 95.1 22 91.7 105 96.3 PLS-SEM
determination
f2 effect size 10 16.4 13 54.2 13 11.9
Predictive Cross-validated 0 0.0 1 4.2 0 0.0 453
relevance
2
Redundancy Q 14 23.0 0 0.0
q2 effect size 4 6.6 1 4.2 0 0.0
Path coefficients Absolute values 48 78.7 22 91.7 107 98.2
Significance of Standard errors, 56 91.8 22 91.7 107 98.2
path coefficients significance levels,
t-values, p-values
Confidence Table VIII.
intervals – 0 0.0 2 8.3 0 0.0 Reported structural
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

Total effects – 3 4.9 2 8.3 4 3.7 model metrics

IS journals as editors and reviewers are increasingly requesting it. In addition, the PLS-SEM
approach to examining mediation overcomes previous shortcomings of the Baron and
Kenny’s (1986) method, noted by Zhao et al. (2010), and is superior to the Preacher and
Hayes’ (2008) process approach that relies on multiple regression.
There are quite a few other analyses that should be considered for future applications of
PLS-SEM. One of the primary issues to focus on is the possible presence of unobserved
heterogeneity, which if overlooked can be a threat to the validity of PLS-SEM findings.
Observed heterogeneity is a situation in which on an a priori basis potential characteristics
of subgroups, such as male or female, or country of origin, are known and can be examined.
In contrast, unobserved heterogeneity is the opposite situation in which there are
unobservable characteristics that cause differences in subgroups and thus the theoretical
model cannot be examined as a single homogeneous population. To identify unobserved
heterogeneity, there are multiple approaches generally referred to as latent class techniques.
We suggest you to refer to the most recent discussions of this topic that summarize the
benefits of combining FIMIX-PLS and PLS-POS (e.g. Sarstedt, Ringle and Gudergan, 2016).
Another topic that may be useful for researchers is confirmatory tetrad analysis in
PLS-SEM (CTA-PLS) (Gudergan et al., 2008). CTA-PLS is a method of empirically testing
and evaluating the cause-effect relationships for latent variables as well as the specification
of indicators in measurement models (Hair et al., 2016). This test assists in avoiding
misspecification of formative and reflective indicators (Table IX).

Best practices recommendations on reporting PLS-SEM results

PLS-SEM is a widely applied tool in the IS literature. Its attractiveness for IS scholars and
practitioners can be attributed to several characteristics. First, as a limited information
approach (Dijkstra, 1983), there are few assumptions about the population or scale of
measurement (Fornell and Bookstein, 1982) and therefore nominal, ordinal, interval, and
ratio scaled variables can be used when interpreted according to guidelines for this type of
measurement. Second, PLS-SEM achieves high levels of statistical power with small sample
sizes and complex models (Chin and Newsted, 1999; Reinartz et al., 2009). The general PLS
path modeling algorithm is based on ordinary least squares regression for separate
subparts of the path model. Therefore, the complexity of the overall model has minimal
influence on sample size requirements. Third, PLS-SEM is preferred over CB-SEM in many
 IMDS IMDS 2010-2015 MISQ 2012-2015 MISQ 1992-2011
117,3 Number Proportion Number Proportion Number Proportion
Criterion (n ¼ 61) (%) (n ¼ 24) (%) (n ¼ 65) (%)

Common method variance 27 44.3 19 79.2 26 40.0

Mediator analysis 12 19.7 9 37.5 15 23.1

454 Multi-group analysis

Continuous moderator analysis 0 0.0 2 8.3 8 12.3
Categorical, observed (multi-
group comparison) 2 3.3 0 0.0 16 24.6
Categorical, unobserved
Table IX. (model-based) 0 0.0 0 0.0 0 0.0
Additional Segmentation techniques;
considerations and e.g., FIMIX-PLS)
supplementary Measurement model invariance 0 0.0 0 0.0 3 4.6
analyses Tetrad analysis 0 0.0 0 0.0 1 1.5
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

research contexts (see Table I), particularly when the statistical objective is prediction.
Fourth, PLS-SEM readily incorporates constructs that are measured either reflectively or
formatively, making the approach particularly appealing for success factor studies
(Albers, 2010). Finally, user-friendly software with a graphical user-interface and many
options for advanced analyses, like SmartPLS (Ringle et al., 2015) and PLS Graph (Chin,
2003), have contributed to the attractiveness of PLS-SEM.
Our review of PLS-SEM applications in IS suggests that PLS-SEM’s methodological
properties continue to be misunderstood, and at times this has led to misapplication of the
technique. Table X includes guidelines for best practices in the application of the PLS-SEM
to IS research.

Conclusions
This paper heeds the call for awareness and application of accepted reporting practices by
Ringle et al. (2012). It reviews and analyzes the use of PLS-SEM from 2010 to 2015 in two
representative journals for the field of IS: IMDS and MISQ. Findings indicate an increased
maturity in the application of PLS-SEM by IS researchers. Specifically, PLS-SEM is being
chosen less often due to small sample sizes and non-normal data, and much more often as a
result of model complexity and the use of formative measurement approaches. At the same
time, exploratory research and theory development are increasingly listed as reasons for
choosing PLS-SEM, particularly in IMDS. The IS field should continue to exercise sound
practice by improved reporting of PLS-SEM assessment metrics and recognizing its
wider applicability for research. Future studies should expand this assessment by
examining the application of PLS-SEM in other journals and including metrics on the
recently emerging analyses, such as mediation, moderation, invariance, unobserved
heterogeneity, and multi-group analysis.
There are many new options for further PLS-SEM analyses that will lead to better
understanding of data relationships. These new options provide additional reasons to apply
PLS to explain SEM relationships. The traditional CB-SEM approach has several
unique situations where it is the preferred SEM approach. Overall, however, PLS-SEM has
many situations where it is the preferred SEM approach. And a number of these situations,
such as continuous moderators and higher order constructs with only two first-order
constructs, cannot be executed using the CB-SEM method.
As a final thought, authors should remember that PLS-SEM and CB-SEM differ from a
statistical point of view, are designed to achieve different objectives, and rely on different
 Criterion Rules of thumb
An updated
and expanded
Characteristics of the data assessment of
Sample size Use “ten times rule” as rough guideline for minimum sample size; adjust
sample size considering power PLS-SEM
Sample distribution Very robust when applied to highly skewed data
Use of holdout sample Minimum of 30% of original sample
Number of missing values PLS-SEM is robust with approximately 10% missing at random data 455
Missing value treatment Do not use mean replacement; delete case-wise or carry out nearest neighbor
or EM algorithm imputation; code missing data where appropriate
Measurement scales All types OK, except categorical scale for ultimate dependent variable
Algorithm settings and software
Starting values of weights for Use 1 as an initial value for each outer weight
initial approximation of latent
variable scores
Weighting scheme Use path weighting scheme
Sum of the outer weights’ changes between two iterationso 10−5
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

Stop criterion
Software Report full citation as required by license agreements
Parameter settings
Bootstrapping sign changes Use individual sign changes
option
Number of bootstrap samples 5,000; must be greater than the number of observations
Size of bootstrap samples Equal to the number of observations
Blindfolding Use cross-validated redundancy
Omission distance d Number of valid observations divided by d must not be an integer; 5 ⩽ d ⩽ 10
Multi-group comparisons Use non-parametric approach to assess significance
Model characteristics
Inner model description Structural model should display all inner model relationships
Outer model description A list of all measurement model indicators (questions)
Latent variable Endogenous latent variables should not be modeled formatively; evaluate
measurement mode measurement mode with CTA-PLS (confirmatory tetrad analysis)
Number of items on In general avoid single-item measures
latent variable
Outer model evaluation: reflective measurement models
Indicator loadings size Standardized indicator loadings ⩾ 0.70
Construct reliability Composite reliability ⩾ 0.70 (in exploratory research 0.60 to 0.70 is
considered acceptable)
Convergent validity AVE ⩾ 0.50
Discriminant validity – HTMT Values lower than 0.85 for conceptually distinct constructs and below 0.90
for conceptually similar constructs; confidence intervals should not include a
value of 1
Measurement model Consider executing CTA-PLS to empirically evaluate model specification
specification
Outer model evaluation: formative measurement models
Significance of weights Report t-values and p-values
Multicollinearity Examine for VIFo5
Absolute vs relative indicator Report indicator weights and loadings; assess significance
contributions
Inner model evaluation
R2 Research context determines acceptable level
Effect size f2 0.02, 0.15, 0.35 for weak, moderate, strong effects
Path coefficient estimates Assess significance and confidence intervals Table X.
Best practices:
reporting PLS-SEM
(continued ) results
 IMDS Criterion Rules of thumb
117,3
Predictive relevance Q2 and q2 Use blindfolding; q2: 0.02, 0.15, 0.35 for weak, moderate, strong predictive
relevance
Observed and unobserved Consider categorical or continuous moderating variables using a priori
heterogeneity information, FIMIX-PLS or PLS-POS

456 Other
Multi-group analysis When comparing groups specified with observed or unobserved
heterogeneity, group sizes should be comparable. When group sizes are
quite different (e.g. ratio ⩾ 2:1), randomly withdraw observations from larger
Table X. group so group sizes are comparable

measurement philosophies – total variance for PLS-SEM vs common variance only for
CB-SEM. Overall, neither of the methods is generally superior to the other and neither
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

method is appropriate for all situations. In general, the strengths of PLS-SEM are CB-SEM’s
limitations, and vice versa. It is important that researchers be aware of the different
applications for which each approach was developed and apply the methods accordingly.

References
Albers, S. (2010), “PLS and success factor studies in marketing”, in Vinzi, V.E., Chin, W.W., Henseler, J.
and Wang, H. (Eds), Handbook of Partial Least Squares: Concepts, Methods and Applications in
Marketing and Related Fields, Springer, Berlin, pp. 409-425.
Babin, B.J., Griffin, M. and Hair, J.F. (2016), “Heresies and sacred cows in scholarly marketing
publications”, Journal of Business Research, Vol. 69 No. 8, pp. 3133-3138.
Baron, R.M. and Kenny, D.A. (1986), “The moderator-mediator variable distinction in social
psychological research: conceptual, strategic, and statistical considerations”, Journal of
Personality and Social Psychology, Vol. 51 No. 6, pp. 1173-1182.
Becker, J.-M., Rai, A., Ringle, C.M. and Völckner, F. (2013), “Discovering unobserved heterogeneity in
structural equation models to avert validity threats”, MISQ, Vol. 37 No. 3, pp. 665-694.
Bentler, P.M. and Huang, W. (2014), “On components, latent variables, PLS and simple methods:
reactions to Rigdon’s rethinking of PLS”, Long Range Planning, Vol. 47 No. 3, pp. 136-145.
Chin, W.W. (2003), PLS Graph 3.0, Soft Modeling Inc., Houston, TX.
Chin, W.W. and Newsted, P.R. (1999), “Structural equation modeling analysis with small samples using
partial least squares”, in Hoyle, R.H. (Ed.), Statistical Strategies for Small Sample Research,
Sage, Thousand Oaks, CA, pp. 307-341.
Diamantopoulos, A., Sarstedt, M., Fuchs, C., Wilczynski, P. and Kaiser, S. (2012), “Guidelines for
choosing between multi-item and single-item scales for construct measurement: a predictive
validity perspective”, Journal of the Academy of Marketing Science, Vol. 40 No. 3, pp. 434-449.
Dijkstra, T. (1983), “Some comments on maximum likelihood and partial least squares methods”,
Journal of Econometrics, Vol. 22 Nos 1/2, pp. 67-90.
Dijkstra, T.K. (2014), “PLS’ Janus face – response to Professor Rigdon’s ‘Rethinking partial least squares
modeling: in praise of simple methods’ ”, Long Range Planning, Vol. 47 No. 3, pp. 146-153.
Dijkstra, T.K. and Henseler, J. (2015a), “Consistent and asymptotically normal PLS estimators for linear
structural equations”, Computational Statistics & Data Analysis, Vol. 81, pp. 10-23.
Dijkstra, T.K. and Henseler, J. (2015b), “Consistent partial least squares structural equation modeling”,
MISQ, Vol. 39 No. 2, pp. 297-316.
do Valle, P.O. and Assaker, G. (2015), “Using partial least squares structural equation modeling in
tourism research: a review of past research and recommendations for future applications”,
Journal of Travel Research, Vol. 54 No. 1, pp. 36-51.
 Fornell, C.G. and Bookstein, F.L. (1982), “Two structural equation models: LISREL and PLS applied to An updated
consumer exit-voice theory”, Journal of Marketing Research, Vol. 19 No. 4, pp. 440-452. and expanded
Fuller, C., Dickerson, M., Atinc, G., Atinc, Y. and Babin, B. (2016), “Common method variance detection assessment of
in business research”, Journal of Business Research, Vol. 69 No. 6, pp. 1183-1202.
Gefen, D., Straub, D.W. and Rigdon, E.E. (2011), “An update and extension to SEM guidelines for
PLS-SEM
administrative and social science research”, MISQ, Vol. 35 No. 2, pp. iii-xiv.
Gudergan, S., Ringle, C.M., Wende, S. and Will, A. (2008), “Confirmatory tetrad analysis in PLS path 457
modeling”, Journal of Business Research, Vol. 61 No. 12, pp. 1238-1249.
Hair, J.F., Anderson, R.E., Babin, B.J. and Black, W.C. (2010), Multivariate Data Analysis: A Global
Perspective, Vol. 7, Pearson, Upper Saddle River, NJ.
Hair, J.F. Jr, Hult, G.T.M., Ringle, C. and Sarstedt, M. (2016), A Primer on Partial Least Squares
Structural Equation Modeling (PLS-SEM), 2nd ed., Sage, Thousand Oaks, CA.
Hair, J.F., Sarstedt, M., Pieper, T.M. and Ringle, C.M. (2012), “The use of partial least squares structural
equation modeling in strategic management research: a review of past practices and
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

recommendations for future applications”, Long Range Planning, Vol. 45 Nos 5/6, pp. 320-340.
Hair, J.F., Sarstedt, M., Ringle, C.M. and Gudergan, S.P. (2018), Advanced Issues in Partial Least Squares
Structural Equation Modeling (PLS-SEM), Sage, Thousand Oaks, CA.
Hair, J.F., Sarstedt, M., Ringle, C.M. and Mena, J.A. (2012), “An assessment of the use of partial least
squares structural equation modeling in marketing research”, Journal of the Academy of
Marketing Science, Vol. 40 No. 3, pp. 414-433.
Harman, H.H. (1976), Modern Factor Analysis, 3rd ed., The University of Chicago Press, Chicago, IL.
Henseler, J., Ringle, C.M. and Sarstedt, M. (2015), “A new criterion for assessing discriminant validity in
variance-based structural equation modeling”, Journal of Academy of Marketing Science, Vol. 43
No. 1, pp. 115-135.
Henseler, J., Ringle, C.M. and Sarstedt, M. (2016), “Testing measurement invariance of composites using
partial least squares”, International Marketing Review, Vol. 33 No. 3, pp. 405-431.
Henseler, J., Dijkstra, T., Sarstedt, M., Ringle, C., Diamantopoulos, A., Straub, D., Ketchen, D., Hair, J.,
Hult, G.T. and Calantone, R. (2014), “Common beliefs and reality about PLS: comments on
Ronkko & Evermann (2013)”, Organizational Research Methods, Vol. 17 No. 2, pp. 182-209.
Lee, L., Petter, S., Fayard, D. and Robinson, S. (2011), “On the use of partial least squares path modeling
in accounting research”, International Journal of Accounting Information Systems, Vol. 12 No. 4,
pp. 305-328.
Lohmöller, J.-B. (1987), “PLS-PC: latent variables path analysis with partial least squares – version 1.8
for PCs under MS-DOS”.
Marcoulides, G.A., Chin, W.W. and Saunders, C. (2009), “A critical look at partial least squares
modeling”, MISQ, Vol. 33 No. 1, pp. 171-175.
Peng, D.X. and Lai, F. (2012), “Using partial least squares in operations management research:
a practical guideline and summary of past research”, Journal of Operations Management, Vol. 30
No. 6, pp. 467-480.
Podsakoff, P.M., MacKenzie, S.B., Lee, J.-Y. and Podsakoff, N.P. (2003), “Common method biases in
behavioral research: a critical review of the literature and recommended remedies”, Journal of
Applied Psychology, Vol. 88 No. 5, pp. 879-903.
Preacher, K.J. and Hayes, A.F. (2008), “Asymptotic and resampling strategies for assessing and
comparing indirect effects in multiple mediator models”, Behavioral Research Methods, Vol. 40
No. 3, pp. 879-891.
Reinartz, W.J., Haenlein, M. and Henseler, J. (2009), “An empirical comparison of the efficacy of
covariance-based and variance-based SEM”, International Journal of Market Research, Vol. 26
No. 4, pp. 332-344.
Richter, N.F., Sinkovics, R.R., Ringle, C.M. and Schlägel, C. (2016), “A critical look at the use of SEM in
international business research”, International Marketing Review, Vol. 33 No. 3, pp. 376-404.
 IMDS Ringle, C.M., Sarstedt, M. and Straub, D. (2012), “A critical look at the use of PLS-SEM in MIS
117,3 Quarterly”, MIS Quarterly, Vol. 36 No. 1, pp. iii-xiv.
Ringle, C.M., Wende, S. and Becker, J.-M. (2015), SmartPLS 3, SmartPLS, Bönningstedt.
Ringle, C.M., Wende, S. and Will, A. (2005), “SmartPLS release: 2.0 (beta)”, SmartPLS, Hamburg,
available at: www.smartpls.de
Sarstedt, M., Diamantopoulos, A. and Salzberger, T. (2016), “Should we use single items? Better not”,
458 Journal of Business Research, Vol. 69 No. 6, pp. 1224-1231.
Sarstedt, M., Diamantopoulos, A., Salzberger, T. and Baumgartner, P. (2016), “Selecting single items to
measure doubly-concrete constructs: a cautionary tale”, Journal of Business Research, Vol. 69
No. 6, pp. 1202-1223.
Sarstedt, M., Henseler, J. and Ringle, C.M. (2011), “Multigroup analysis in partial least squares (PLS)
path modeling: alternative methods and empirical results”, Advances in International Marketing,
Vol. 22 No. 1, pp. 195-218.
Sarstedt, M., Ringle, C.M. and Gudergan, S.P. (2016), “Guidelines for treating unobserved heterogeneity
in tourism research: a comment on Marques and Reis (2015)”, Annals of Tourism Research,
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

Vol. 57, pp. 279-284.

Sarstedt, M., Ringle, C.M., Smith, D., Reams, R. and Hair, J.F. (2014), “Partial least squares structural
equation modeling (PLS-SEM): a useful tool for family business researchers”, Journal of Family
Business Strategy, Vol. 5 No. 1, pp. 105-115.
Shmueli, G. (2010), “To explain or predict?”, Statistical Science, Vol. 25 No. 3, pp. 289-310.
Sosik, J.J., Kahai, S.S. and Piovoso, M.J. (2009), “Silver bullet or voodoo statistics? A primer for using
the partial least squares data analytic technique in group and organization research”,
Group & Organization Management, Vol. 34 No. 1, pp. 5-36.
Zhao, X., Lynch, J.G. and Chen, Q. (2010), “Reconsidering Baron and Kenny: myths and truths about
mediation analysis”, Journal of Consumer Research, Vol. 37 No. 2, pp. 197-206.

Further reading
Rönkkö, M. and Evermann, J. (2013), “A critical examination of common beliefs about partial least
squares path modeling”, Organizational Research Methods, Vol. 16 No. 3, pp. 425-448.

Corresponding author
Alain Yee Loong Chong can be contacted at: [email protected]

For instructions on how to order reprints of this article, please visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details: [email protected]
 This article has been cited by:

1. Jeong-Yeol Park, Robin M. Back, Diego Bufquin, Valeriya Shapoval. 2019. Servicescape, positive
affect, satisfaction and behavioral intentions: The moderating role of familiarity. International
Journal of Hospitality Management 78, 102-111. [Crossref]
2. Hossein Shabanali Fami, Lusine H. Aramyan, Siet J. Sijtsema, Amir Alambaigi. 2019.
Determinants of household food waste behavior in Tehran city: A structural model. Resources,
Conservation and Recycling 143, 154-166. [Crossref]
3. ChongWoo Park, Dong-gook Kim, Sunyoung Cho, Hyo-Joo Han. 2019. Adoption of multimedia
technology for learning and gender difference. Computers in Human Behavior 92, 288-296.
[Crossref]
4. GreenKenneth W., Kenneth W. Green, InmanR. Anthony, R. Anthony Inman, SowerVictor E.,
Victor E. Sower, ZelbstPamela J., Pamela J. Zelbst. 2019. Impact of JIT, TQM and green supply
chain practices on environmental sustainability. Journal of Manufacturing Technology Management
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

30:1, 26-47. [Abstract] [Full Text] [PDF]

5. Christian Nedu Osakwe. 2019. Customer centricity: an empirical analysis in the micro-sized firm.
Journal of Strategic Marketing 17, 1-14. [Crossref]
6. Mousa Albashrawi, Hasan Kartal, Asil Oztekin, Luvai Motiwalla. 2019. Self-Reported and
Computer-Recorded Experience in Mobile Banking: a Multi-Phase Path Analytic Approach.
Information Systems Frontiers 47. . [Crossref]
7. Mostafa Al-Emran, Vitaliy Mezhuyev, Adzhar Kamaludin. PLS-SEM in Information Systems
Research: A Comprehensive Methodological Reference 644-653. [Crossref]
8. Lisa Perkhofer, Othmar Lehner. Using Gaze Behavior to Measure Cognitive Load 73-83.
[Crossref]
9. R. Merli, M. Preziosi, A. Acampora, M.C. Lucchetti, F. Ali. 2019. The impact of green practices
in coastal tourism: An empirical investigation on an eco-labelled beach club. International Journal
of Hospitality Management 77, 471-482. [Crossref]
10. M.Rosario González-Rodríguez, M.Carmen Díaz-Fernández, Simonetti Biagio. 2019. The
Perception Of Socially And Environmentally Responsible Practices Based On Values And Cultural
Environment From A Customer Perspective. Journal of Cleaner Production . [Crossref]
11. . Prelims i-xviii. [Citation] [Full Text] [PDF] [PDF]
12. Hengky Latan. Chapter 4 PLS Path Modeling in Hospitality and Tourism Research: The Golden
Age and Days of Future Past 53-83. [Citation] [Full Text] [PDF] [PDF]
13. HouAvus C.Y., Avus C.Y. Hou, ShiauWen-Lung, Wen-Lung Shiau, ShangRong-An, Rong-
An Shang. The involvement paradox. Industrial Management & Data Systems, ahead of print.
[Abstract] [Full Text] [PDF]
14. Inma Rodríguez-Ardura, Antoni Meseguer-Artola. 2018. The playfulness of Facebook – Shaped by
underlying psychological drivers and gender differences. Telematics and Informatics 35:8, 2254-2269.
[Crossref]
15. Garry Wei-Han Tan, Voon-Hsien Lee, Jun-Jie Hew, Keng-Boon Ooi, Lai-Wan Wong. 2018. The
interactive mobile social media advertising: An imminent approach to advertise tourism products
and services?. Telematics and Informatics 35:8, 2270-2288. [Crossref]
 16. Thomas Clauss, Peter Harengel, Marianne Hock. 2018. The perception of value of platform-
based business models in the sharing economy: determining the drivers of user loyalty. Review of
Managerial Science 13. . [Crossref]
17. ChangYounghoon, Younghoon Chang, WongSiew Fan, Siew Fan Wong, EzeUchenna, Uchenna
Eze, LeeHwansoo, Hwansoo Lee. The effect of IT ambidexterity and cloud computing absorptive
capacity on competitive advantage. Industrial Management & Data Systems, ahead of print.
[Abstract] [Full Text] [PDF]
18. UsakliAhmet, Ahmet Usakli, KucukerginKemal Gurkan, Kemal Gurkan Kucukergin. 2018. Using
partial least squares structural equation modeling in hospitality and tourism. International Journal
of Contemporary Hospitality Management 30:11, 3462-3512. [Abstract] [Full Text] [PDF]
19. Russell Torres, Anna Sidorova, Mary C. Jones. 2018. Enabling firm performance through business
intelligence and analytics: A dynamic capabilities perspective. Information & Management 55:7,
822-839. [Crossref]
20. Girish Prayag, Mesbahuddin Chowdhury, Samuel Spector, Caroline Orchiston. 2018.
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

Organizational resilience and financial performance. Annals of Tourism Research 73, 193-196.
[Crossref]
21. Meena Pradhan, Jungjoo Oh, Hwansoo Lee. 2018. Understanding Travelers’ Behavior for
Sustainable Smart Tourism: A Technology Readiness Perspective. Sustainability 10:11, 4259.
[Crossref]
22. Jun-Hwa Cheah, Hiram Ting, T. Ramayah, Mumtaz Ali Memon, Tat-Huei Cham, Enrico
Ciavolino. 2018. A comparison of five reflective–formative estimation approaches: reconsideration
and recommendations for tourism research. Quality & Quantity 42. . [Crossref]
23. Adedapo Oluwaseyi Ojo, Chris Niyi Arasanmi, Murali Raman, Christine Nya-Ling Tan. 2018.
Ability, motivation, opportunity and sociodemographic determinants of Internet usage in Malaysia.
Information Development 65, 026666691880485. [Crossref]
24. Mohammad I. Merhi, Punit Ahluwalia. 2018. Examining the impact of deterrence factors and
norms on resistance to Information Systems Security. Computers in Human Behavior . [Crossref]
25. MüllerTobias, Tobias Müller, SchuberthFlorian, Florian Schuberth, HenselerJörg, Jörg Henseler.
2018. PLS path modeling – a confirmatory approach to study tourism technology and tourist
behavior. Journal of Hospitality and Tourism Technology 9:3, 249-266. [Abstract] [Full Text] [PDF]
26. ChuahStephanie Hui-Wen, Stephanie Hui-Wen Chuah, RauschnabelPhilipp A., Philipp A.
Rauschnabel, TsengMing-Lang, Ming-Lang Tseng, RamayahT., T. Ramayah. 2018. Reducing
temptation to switch mobile data service providers over time. Industrial Management & Data Systems
118:8, 1597-1628. [Abstract] [Full Text] [PDF]
27. Majid Javari. 2018. Comparing causal techniques for rainfall variability analysis using causality
algorithms in Iran. Heliyon 4:9, e00774. [Crossref]
28. Joseph F. Hair, Christian M. Ringle, Siegfried P. Gudergan, Andreas Fischer, Christian Nitzl, Con
Menictas. 2018. Partial least squares structural equation modeling-based discrete choice modeling:
an illustration in modeling retailer choice. Business Research 19. . [Crossref]
29. PratonoAluisius Hery, Aluisius Hery Pratono. 2018. Does firm performance increase with risk-
taking behavior under information technological turbulence?. The Journal of Risk Finance 19:4,
361-378. [Abstract] [Full Text] [PDF]
 30. ZiembaEwa, Ewa Ziemba. The contribution of ICT adoption to sustainability: households’
perspective. Information Technology & People, ahead of print. [Abstract] [Full Text] [PDF]
31. Afris Widya-Hasuti, Abbas Mardani, Dalia Streimikiene, Ali Sharifara, Fausto Cavallaro. 2018.
The Role of Process Innovation between Firm-Specific Capabilities and Sustainable Innovation in
SMEs: Empirical Evidence from Indonesia. Sustainability 10:7, 2244. [Crossref]
32. Heetae Yang, Hwansoo Lee. 2018. Understanding user behavior of virtual personal assistant devices.
Information Systems and e-Business Management 24. . [Crossref]
33. AmegbeHayford, Hayford Amegbe, OsakweChristian Nedu, Christian Nedu Osakwe. 2018.
Towards achieving strong customer loyalty in the financial services industry. International Journal
of Bank Marketing 36:5, 988-1007. [Abstract] [Full Text] [PDF]
34. IlmudeenAboobucker, Aboobucker Ilmudeen, BaoYukun, Yukun Bao. 2018. Mediating role of
managing information technology and its impact on firm performance. Industrial Management &
Data Systems 118:4, 912-929. [Abstract] [Full Text] [PDF]
35. Keng-Boon Ooi, Jun-Jie Hew, Voon-Hsien Lee. 2018. Could the mobile and social perspectives of
Downloaded by Tulane University At 13:16 21 January 2019 (PT)

mobile social learning platforms motivate learners to learn continuously?. Computers & Education
120, 127-145. [Crossref]
36. OsakweChristian N., Christian N. Osakwe, AnazaNwamaka A., Nwamaka A. Anaza. 2018.
Understanding marketing resources and size in agro-based enterprises. Marketing Intelligence &
Planning 36:2, 230-244. [Abstract] [Full Text] [PDF]
37. Jooyoung Hwang, Sangwon Park, Mina Woo. 2018. Understanding user experiences of online
travel review websites for hotel booking behaviours: an investigation of a dual motivation theory.
Asia Pacific Journal of Tourism Research 23:4, 359-372. [Crossref]
38. AliFaizan, Faizan Ali, RasoolimaneshS. Mostafa, S. Mostafa Rasoolimanesh, SarstedtMarko,
Marko Sarstedt, RingleChristian M., Christian M. Ringle, RyuKisang, Kisang Ryu. 2018. An
assessment of the use of partial least squares structural equation modeling (PLS-SEM) in
hospitality research. International Journal of Contemporary Hospitality Management 30:1, 514-538.
[Abstract] [Full Text] [PDF]
39. Christian M. Ringle, Marko Sarstedt, Rebecca Mitchell, Siegfried P. Gudergan. 2018. Partial least
squares structural equation modeling in HRM research. The International Journal of Human
Resource Management 14, 1-27. [Crossref]
40. Necmi K. Avkiran. Rise of the Partial Least Squares Structural Equation Modeling: An Application
in Banking 1-29. [Crossref]
41. Christian Nitzl. Management Accounting and Partial Least Squares-Structural Equation Modelling
(PLS-SEM): Some Illustrative Examples 211-229. [Crossref]
42. Nur Ainna Ramli, Hengky Latan, Gilbert V. Nartea. Why Should PLS-SEM Be Used Rather
Than Regression? Evidence from the Capital Structure Perspective 171-209. [Crossref]
43. Adedapo Oluwaseyi Ojo, Murali Raman, Rameswaran Vijayakumar. Cognitive Determinants of IT
Professional Belief and Attitude Towards Green IT 1006-1015. [Crossref]
44. Pratyush Nidhi Sharma, Galit Shmueli, Marko Sarstedt, Nicholas Danks, Soumya Ray. 2018.
Prediction-Oriented Model Selection in Partial Least Squares Path Modeling. Decision Sciences .
[Crossref]
 45. Amparo Cervera-Taulet, Carmen Pérez-Cabañero, Walesska Schlesinger. 2018. Experience
management as an innovative approach in emerging Mediterranean destinations. Journal of Business
Research . [Crossref]
46. Chang-Tang Chiang, Chiou-Fong Wei, Kevin R. Parker, Bill Davey. 2017. Exploring the drivers
of customer engagement behaviours in social network brand communities: towards a customer-
learning model. Journal of Marketing Management 33:17-18, 1443-1464. [Crossref]
47. Chen Lisha, Chin Fei Goh, Sun Yifan, Amran Rasli. 2017. Integrating guanxi into technology
acceptance: An empirical investigation of WeChat. Telematics and Informatics 34:7, 1125-1142.
[Crossref]
48. Marko Sarstedt, Christian M. Ringle, Joseph F. Hair. Partial Least Squares Structural Equation
Modeling 1-40. [Crossref]
49. Marko Sarstedt, Christian M. Ringle, Joseph F. Hair. Treating Unobserved Heterogeneity in PLS-
SEM: A Multi-method Approach 197-217. [Crossref]
Downloaded by Tulane University At 13:16 21 January 2019 (PT)