Information & Management 57 (2020) 103168

Contents lists available at ScienceDirect

Information & Management

journal homepage: www.elsevier.com/locate/im

How to perform and report an impactful analysis using partial least squares: T
Guidelines for confirmatory and explanatory IS research
Jose Beniteza,b, , Jörg Henselerc,d, Ana Castillob, Florian Schuberthc
⁎

a
Rennes School of Business, Rennes, France
b
Department of Management, School of Business, University of Granada, Granada, Spain
c
Department of Design, Production and Management, Faculty of Engineering Technology, University of Twente, Enschede, the Netherlands
d
Nova Information Management School, Universidade Nova de Lisboa, Lisbon, Portugal

ARTICLE INFO ABSTRACT

Keywords: Partial least squares path modeling (PLS-PM) is an estimator that has found widespread application for causal
Partial least squares path modeling information systems (IS) research. Recently, the method has been subject to many improvements, such as
Guidelines consistent PLS (PLSc) for latent variable models, a bootstrap-based test for overall model fit, and the heterotrait-
Model validation to-monotrait ratio of correlations for assessing discriminant validity. Scholars who would like to rigorously apply
Composite model
PLS-PM need updated guidelines for its use. This paper explains how to perform and report empirical analyses
Confirmatory and explanatory information
systems research
using PLS-PM including the latest enhancements, and illustrates its application with a fictive example on
business value of social media.

1. Introduction For decades, PLS-PM has been the predominant estimator for
structural equation models in the field of information systems (IS) (e.g.,
Structural equation modeling (SEM) has become an important sta- [11–15]). IS research usually incorporates complex research problems
tistical tool in social and behavioral sciences. It is capable of modeling and questions that require conceptualization and operationalization of
nomological networks by expressing theoretical concepts through theoretical concepts, and investigation of their relationships. Current
constructs and connecting these constructs via a structural model to literature suggests two types of theoretical concepts: concepts from
study their relationships [1]. In doing so, random measurement errors behavioral sciences and concepts from design science [16]. Theoretical
can be taken into account and empirical evidence for postulated the- concepts from behavioral research are assumed to cause observable
ories can be obtained by means of statistical testing. indicators and their relationships, i.e., the theoretical concept is the
Two kinds of estimators for SEM can be distinguished: covariance- common cause of observable indicators [17]. Typically, these concepts
based and variance-based estimators. While covariance-based estima- are operationalized by a measurement model. Extant literature suggests
tors minimize the discrepancy between the empirical and model-im- two types of measurement models: the reflective [1] and the cau-
plied variance–covariance matrix of the observable indicators to obtain sal–formative measurement model [18]. Both types of measurement
the model parameter estimates, variance-based estimators create linear models assume a causal relationship between the indicators and their
combinations of the indicators as stand-ins for the theoretical concepts construct, i.e., the latent variable. In contrast, theoretical concepts from
and subsequently estimate the model parameters. A widely used var- design science, so-called artifacts, are human-made creations that are
iance-based estimator is partial least squares path modeling (PLS-PM). shaped and built by their ingredients to serve a certain goal [19]. Due to
Originally developed by Herman O.A. Wold [2] to analyze high-di- the constructivist nature of this type of theoretical concept, recent lit-
mensional data in a low-structure environment, PLS-PM has become a erature suggests to operationalize artifacts by the composite model
full-fledged estimator for SEM over the past decade [3]. Consequently, [16]. In contrast to the measurement models, in the composite model,
PLS-PM has been applied in various fields of business administration the indicators do not cause the construct, but combine to compose the
research such as strategy [4], marketing [5], operations management construct. To highlight this aspect and to pronounce the difference to
[6], human resource management [7], finance [8], tourism [9], and the latent variable, we refer to constructs that are composed of in-
family business [10]. dicators emergent variables [20,21]. In summary, IS scholars can

Corresponding author at: Rennes School of Business, Rennes, France.

⁎

E-mail addresses: jose.benitez@rennes-sb.com, joseba@ugr.es (J. Benitez), j.henseler@utwente.nl (J. Henseler), anacastillo@ugr.es (A. Castillo),
f.schuberth@utwente.nl (F. Schuberth).

https://doi.org/10.1016/j.im.2019.05.003
Received 24 October 2017; Received in revised form 29 April 2019; Accepted 18 May 2019
Available online 23 May 2019
0378-7206/ © 2019 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/BY/4.0/).
J. Benitez, et al. Information & Management 57 (2020) 103168

Table 1
Substantial changes in understanding of PLS-PM.
Traditional view of PLS-PM Up-to-date view of PLS-PM

1. PLS-PM should be used primarily for exploratory and early-stage research PLS-PM can be used for various types of research, e.g., confirmatory and explanatory or
predictive ([22], [23] [24],)
2. PLS-PM has advantages over covariance-based estimators when the sample size is PLS-PM can produce estimates even for very small sample sizes. However, as for other
small estimators, these estimates are generally less accurate than those obtained by a larger
sample ([25] [26],). Hence, the justification of using PLS-PM with small sample sizes
should be considered cautiously
3. PLS-PM can only estimate recursive structural models PLS-PM can also consistently estimate non-recursive structural models by using, e.g.,
2SLS or 3SLS instead of OLS (Dijkstra and Henseler 2015b, [27])
4. Model identification plays no role when employing PLS-PM First, PLS-PM always estimates an underlying composite model, regardless of whether
the model consists of latent variables; identification rules of composite models must thus
be taken into account ([28] [134],). Consequently, model identification is also important
in the case of PLS-PM
5. PLS-PM has greater statistical power than the maximum-likelihood (ML) estimator This statement is based on inconsistent parameter estimates and has been shown to be
invalid [30]. Furthermore, an estimator has no statistical power (one refers to its
efficiency, or accuracy, in estimating parameters, usually expressed by the standard
error); only a statistical test can be assessed in terms of its statistical power
6. Mode A can be used to consistently estimate reflective measurement models Regardless of the mode used, PLS-PM creates linear combinations of observed indicators
(composites) as proxies for the theoretical concepts ([31] [32],). Therefore, to
consistently estimate models containing latent variables, one must correct for
attenuation of the construct scores correlations. In the context of PLS-PM, this procedure
is known as PLSc (Dijkstra and Henseler 2015a, 2015b)
7. Mode B can be used to estimate causal–formative measurement models consistently Mode B is just another way to obtain weights to build composites; hence, it does not
consistently estimate causal–formative measurement models. However, causal–formative
measurement models can be estimated by means of a MIMIC model ([18] [3],)
8. The overall fit of models estimated by PLS-PM cannot be assessed The overall model can be assessed in two non-exclusive ways ([22], Dijkstra and
Henseler 2015b): (1) bootstrap-based tests for overall model fit, and (2) measures of
overall model fit. Both assess the discrepancy between the empirical and the model-
implied indicator variance–covariance matrix. While the latter is based on heuristic
rules, the former is based on statistical inferences
9. Reliability of construct scores obtained by PLS-PM should be assessed using Currently, Dijkstra–Henseler’s ρA is the only consistent reliability coefficient for PLS-PM
Cronbach’s α and Dillon–Goldstein’s ρ (also called Jöreskog’s ρ or composite construct scores [16]. Dillon–Goldstein’s ρ and Cronbach’s α indicate the reliability of
reliability) sum scores. While Cronbach’s α is based on the indicator variance–covariance matrix,
Dillon–Goldstein’s ρ is based on the factor loadings. Therefore, for the calculation of
Dillon–Goldstein’s ρ, consistent factor loading estimates should be used. Moreover,
Cronbach’s α assumes equal population covariances among the indicators of one block;
an assumption that is likely not met in empirical research. However, it can be used as a
lower bound to reliability ([33] [34],)
10. Discriminant validity should be examined using the Fornell–Larcker criterion The HTMT [35] should be considered to assess discriminant validity ([36] [37],)

operationalize a theoretical concept in three different ways in their enhancements in PLS-PM, neither of these prior PLS-PM guidelines for
model: reflective and causal–formative measurement model (usually causal research covered the full range of recent developments.
employed for behavioral concepts), and composite model (usually To fill this gap on guidelines in the current IS literature, this study
employed for artifacts). See Table 1 in Henseler (2017) for a detailed provides updated guidelines for using PLS-PM in causal research
explanation of the different types of operationalization. (confirmatory and explanatory research), employing all the most re-
In recent years, PLS-PM has become the subject of scholarly debate. cently proposed standards. In so doing, the paper addresses why and
Proponents called PLS-PM a “silver bullet” [38], while opponents cri- how to perform and report PLS-PM estimation in confirmatory and ex-
ticized PLS-PM’s inconsistency for latent variable models and the ab- planatory IS research. In confirmatory IS research, the scholar aims to
sence of a test for overall model fit (e.g. [39],). This debate has sti- understand the causal relationships between theoretical concepts of
mulated the development of several enhancements to PLS-PM. These interest for the IS community. In doing so, the scholar aims to confirm a
include consistent PLS (PLSc) to consistently estimate linear and non- postulated theory, i.e., obtain empirical evidence for his/her descrip-
linear latent variable models ([40] [41],); a bootstrap-based test to tion of the working mechanism of the world. This is tried to be achieved
statistically assess overall model fit [42]; measures of overall model fit, by imposing testable restrictions on the indicator variance–covariance
such as the standardized root mean squared residual (SRMR), based on matrix, e.g., by fixing path coefficients to a certain value, assuming that
heuristic rules to evaluate overall model fit [22]; and the heterotrait-to- the correlation between two indicators is the result of an underlying
monotrait (HTMT) ratio of correlations as a criterion to assess dis- latent variable like in the classical reflective measurement model, or in
criminant validity [35]. As a result, PLS-PM has become a full-fledged the composite model that the correlations of the indicators forming an
estimator to SEM that can deal with reflective and causal–formative emergent variable with a variable not forming the emergent variable
measurement models as well as composite models. Moreover, it can be are proportional. The dominant statistical tool in the context of con-
applied to confirmatory, explanatory, exploratory, descriptive, and firmatory research is the test for overall model fit. Testing model fit
predictive research [24]. only makes sense if the number of correlations among observable
For the field of IS to benefit from these methodological and con- variables exceeds the number of model parameters to be estimated, i.e.,
ceptual achievements in PLS-PM, IS scholars need guidelines for their it is indispensable to have a certain amount of parsimony (a positive
empirical studies that incorporate all these new developments and re- number of degrees of freedom in the sense of SEM).
cently obtained insights. Some of the guidelines papers on PLS-PM in As in confirmatory research, in explanatory IS research, the analyst
the IS literature were published before 2013 – i.e., before the debate aims to understand the causal relationship among the theoretical con-
and resulting enhancements (e.g., [12,43–45]). Although several re- cepts. However, this type of research primarily focuses on the ex-
cently published textbooks and articles (e.g. [46–48],) have provided planation of a specific phenomenon which is treated as a dependent
guidelines for causal research that cover some of the latest variable in the model. In doing so, the primary focus is on the

2
J. Benitez, et al. Information & Management 57 (2020) 103168

coefficient of determination (R²) and the significance of path coefficient measurement model imposes restrictions on the variance–covariance
estimates. Such models can be saturated (i.e., have zero degrees of matrix of indicators belonging to one latent variable. In its classical
freedom in the sense of SEM). Although the two types of research can be form, the correlations among the indicators of one block are zero when
theoretically distinguished, in empirical IS research, scholars very often controlled for the latent variable, also known as the axiom of local
combine confirmatory and explanatory IS research, e.g., testing the independence [67]. This fact is typically exploited to draw conclusions
measurement model (confirmatory research) and focusing on the ex- about the existence of the latent variable.
planation of a specific construct in structural model (explanatory re- Besides the reflective measurement model, the literature proposes
search). This paper explains why and how to perform and report em- the causal–formative measurement of behavioral concepts [68,69]. In
pirical analyses using PLS-PM in causal IS research following the latest contrast to the reflective measurement model, the causal–formative
enhancements, and illustrates this analysis with a fictive example on measurement model reverses the direction of causality between the
business value of social media. indicators and the construct and assumes that the observed indicators
cause the latent variable. This model thus does not restrict the covar-
2. Foundations of PLS-PM iances of the indicators belonging to one block. The remaining causes
not represented by the indicators are captured in an error term, which is
In its current form, PLS-PM is a full-fledged variance-based esti- by assumption uncorrelated with the causal indicators. Although a
mator for SEM that can estimate linear, non-linear, recursive, and non- violation of this assumption, i.e., omission of causal indicators, leads to
recursive structural models ([40] [42],). Moreover, it is capable of biased parameter estimates of the causal indicators, recent literature
dealing with models that contain emergent and latent variables [41], shows that the meaning of the latent variable is not affected by omitting
second-order emergent variables built by latent variables [49], and causal indicators and the remaining model parameters can be con-
ordinal categorical indicators [29]. It can incorporate sampling weights sistently estimated [70]. However, the causal–formative measurement
known as weighted partial least squares (WPLS, see [50]), deal with model is not identified on its own, i.e., the model parameters cannot be
correlated measurement errors within a block of indicators [51], and uniquely retrieved from the population indicator variance–covariance
address multicollinearity among the constructs in the structural model matrix [71,72]. To obtain an identified causal–formative measurement
[52]. It can also be used for multiple group comparison ([53], [54],), model, the latent variable must be connected to at least two other
and potential sources of endogeneity can be addressed [55]. Finally, variables not affecting the latent variable [18], for example, using a
important-performance map analysis can be used to illustrate the re- multiple-indicators, multiple-causes (MIMIC) model.
sults of the structural model [56]. For a recent overview of the meth- Typical examples of behavioral concepts in IS that have been op-
odological research on PLS-PM, we refer to [57]. erationalized by a measurement model are behavioral intention to use
information technology (IT), and IT interaction behavior. Behavioral
2.1. Model specification intention to use IT indicates the degree to which a person has for-
mulated conscious plans to perform or not to perform a specified future
To employ PLS-PM, scholars must transfer their proposed theory behavior involving IT use. This concept has been operationalized by a
into a statistical model [58]. In the context of SEM, this means that the reflective measurement model in past IS research using the following
theoretical concepts and their hypothesized relationships must be items: user’s intention, prediction, and plan to use IT in future months
transferred into a structural model. “Theoretical concepts refer to ideas (e.g., [73,74]). IT interaction behavior refers to the user’s interaction
that have some unity or something in common. The meaning of a with IT to accomplish an individual or organizational task. This concept
theoretical concept is spelled out in a theoretical definition” [59]. We has been operationalized by a causal–formative measurement model in
distinguish between two types of theoretical concepts: behavioral past IS research. For example, Barki et al. [75] employed a MIMIC
concepts, and design concepts, so-called artifacts. Typically, theoretical model to operationalize IT interaction behavior using six tasks (causes)
concepts are represented by constructs in the structural model [32]. that motivated users to interact with IT (problem solving, justifying
Although constructs and latent variables are often equated [60], we decisions, exchanging information with people, planning or following
deliberately distinguish between a latent variable, i.e., a construct that up, coordinating activities, and serving customers); and two measure-
represents a behavioral concept, and an emergent variable, i.e., a ments of this behavior using two reflective indicators (importance of IT
construct that represents an artifact. The operationalization of theore- and time invested using IT).
tical concepts, i.e., the specification of the theoretical concepts in the Emergent variables are an alternative representation of theoretical
structural model, requires special attention because estimates are likely concepts [20,21]. They have been recently referred in empirical IS re-
to be inconsistent if a concept’s operationalization is not in accordance search as “composite constructs” (e.g. [76],). Although these labels
with the concept’s nature [61]. could be used interchangeably, we recommend using the term “emer-
PLS-PM can deal with two kinds of constructs: emergent variables gent variable” to highlight that the construct emerges from the in-
and latent variables. Latent variables refer to variables that are not dicators. Emergent variables can help model artifacts [3,16]. An artifact
directly observed but instead inferred through a measurement model is a human- or firm-made object composed of its ingredients. Thus, in
from other observed variables (directly measured; [46,62]). They contrast to behavioral concepts, they are not assumed to exist in nature,
usually represent theoretical concepts of behavioral research such as but are products of theoretical thinking and/or theoretically justified
personality traits, individual behavior, and individual attitude [63]. constructions usually made to fulfill a certain purpose. To oper-
This theoretical reasoning rests on the assumption that behavioral ationalize these human- or firm-made concepts, the composite model
concepts of interest exist in nature, irrespective of scholarly investiga- can be employed [16]. Examples from the IS research are IT capability
tion [64]. The existing literature proposes two ways to measure beha- and IT ambidexterity [77,78].
vioral concepts [59]: reflective and causal–formative measurement The composite model can be understood as a recipe for how in-
model. gredients (the components) should be mixed and matched to build the
The reflective measurement model – also known as the common artifact. The composite model assumes a definitorial rather than a
factor model – is grounded in the true score theory [65]. It assumes that causal relationship between indicators and the emergent variable ([63],
a set of indicators is a measurement error–prone manifestation of an 2017). In the classical composite model, the indicators forming an
underlying latent variable [66]. Some indicators can thus be inter- emergent variable are assumed to be free of measurement errors. In
changed without altering the meaning of the latent variable. As the contrast to the reflective measurement model, the composite model
measurement errors of a block of indicators are usually assumed to be imposes no restrictions on the covariance structure of indicators be-
uncorrelated and independent of the latent variable, the reflective longing to the same construct. The reflective measurement model is

3
J. Benitez, et al. Information & Management 57 (2020) 103168

thus nested within the composite model, as the composite model relaxes Once the theoretical concepts are operationalized, the constructs
the assumption that all covariation among a block of indicators is ex- representing the theoretical concepts can be related via the structural
plained by one latent variable [22]. Yet, the composite model con- model. The structural model typically represents the core of the theory
straints the correlations between the indicators forming an emergent proposed. The structural model generally consists of a set of regression
variable and variables not forming the emergent variable, i.e., they are equations, illustrating the relationship hypothesized between the the-
proportional [29]. Similar to the causal–formative measurement model, oretical concepts. In each equation, a dependent construct is explained
the composite model is not identified when isolated in the structural by one or more independent constructs. Because a dependent construct
model. To ensure identification, a necessary condition is that each is typically not fully explained by its independent constructs, an error
emergent variable must be linked to at least one variable not forming term accounts for the remaining variance in the dependent construct.
the emergent variable [28,134]. By assumption, the error term is independent of the explanatory con-
Because the artifact as a type of theoretical concept was introduced structs of its equation. To avoid violating this assumption, in causal IS
only recently, it is helpful to illustrate this type of theoretical concept research, the scholar should make every effort to include all relevant
with an example. Based on theory, bread is made from wheat, water, constructs (those that affect the dependent construct and correlate with
salt, and yeast. Although the correlations between the amounts of at least one explanatory construct in the corresponding equation).
wheat, water, salt, and yeast in a sample of loaves of bread are likely to Otherwise, the path coefficient estimates obtained by ordinary least
be high, one would not conclude that bread is something that should be squares (OLS) suffer from omitted variable bias [87]. One potential way
measured, i.e., that bread causes (or is caused by) wheat, water, salt, to address this problem of endogeneity is to use the two-stage least
and yeast. Rather, wheat, water, salt, and yeast are the simple entities squares (2SLS) estimator for the structural model ([42] [27,55],). In the
(ingredients) combined to form the emergent variable representing the following, we consider only recursive structural models, structural
artifact we call bread. Clearly, the temporal precedence of the in- models without feedback loops, and/or correlated error terms.
gredients also suggests that bread cannot be the common cause of its
ingredients. 2.2. Parameter estimation
Because IS Science analyzes and aims at explaining how IT affects
organizations, individuals, and society, artifacts play a pivotal role in IS In its current form, PLS-PM estimates model parameters in three
research. For example, the theoretical concept IT infrastructure cap- steps. In the first, the iterative PLS-PM algorithm determines the
ability refers to a firm’s ability to use and leverage its IT resource in- weights to create scores for each construct (latent variables and emer-
frastructure for business activities [79–82]. IT infrastructure capability gent variables; [88]). As construct scores of latent variables contain
is a “human-made/firm-made” concept that can be operationalized by measurement errors, the second step corrects for attenuation in corre-
the composite model [76,81,83]. Of course, no single “true” recipe lations between latent variables. In doing so, PLSc divides the construct
exists for creating this artifact. Just as different bakeries can produce scores correlations by the geometric mean of the constructs’ reliabilities
different types of bread or different breweries produce different types of [41], making the main outcome of the second step a consistent con-
beer, different scholars can produce different recipes for the same struct correlation matrix. Finally, the third step estimates the model
concept. The beer analogy can be extraordinarily instructive. Different parameters (weights, loadings, and path coefficients). Based on the
recipes exist worldwide to design and manufacture beer. For example, consistent construct correlation matrix, OLS can be used to estimate the
Spanish breweries use one recipe, German breweries another. Recipes path coefficients of recursive structural models. In case of non-recursive
can even vary by region within a country. Such diversity makes each structural models, the 2SLS or three-stage least squares (3SLS) esti-
recipe an idiosyncratic way to understand and design beer, but all of mator can be used, instead of the OLS estimator, to obtain consistent
these recipes ultimately produce beer. path coefficient estimates ([42] [27,83],).
For example, based on Melville et al. [84] study, Ajamieh et al. [81]
define the artifact IT infrastructure capability as composed of IT tech- 2.3. Substantial changes in the understanding of PLS-PM
nological infrastructure capability, IT managerial infrastructure cap-
ability, and IT technical infrastructure capability. Further, some prior IS In recent years, PLS-PM practices have been examined, debated, and
research [85,86] considers IT capability – a concept similar to IT in- improved. The recent literature on PLS-PM has been thus substantially
frastructure capability – as composed of IT technical infrastructure, changed and improved, requiring that we identify the changes in the
human IT resources, and IT-enabled intangibles. IT infrastructure flex- understanding and practice of PLS-PM. Table 1 summarizes these
ibility and post-merger and acquisition (M&A) IT integration capability changes in the understanding of PLS-PM in the context of confirmatory
are two examples of artifacts recently considered in IS research [83]. IT and explanatory research.
infrastructure flexibility refers to the capability of the infrastructure to Traditional view 1: PLS-PM should be used primarily for ex-
adapt to environmental changes. A flexible firm IT infrastructure has ploratory and early-stage research. Although PLS-PM was originally
the following characteristics: IT compatibility, IT connectivity, mod- developed for exploratory research [2], enhancements such as PLSc and
ularity, and IT personnel skills flexibility [83]. Similarly, post-M&A IT the bootstrap-based test for overall model fit make PLS-PM suitable for
integration capability is the firm’s ability to integrate the IT technical causal research, i.e., confirmatory and explanatory research. However,
infrastructure, IT personnel, and IT and business processes of the as originally developed, PLS-PM can also be applied in descriptive and
target/acquired firm with the IT technical infrastructure, IT personnel, predictive research [23,24].
and IT and business processes of the acquirer after an M&A [83]. Thus, Traditional view 2: PLS-PM has advantages over covariance-based
post-M&A IT integration capability can be understood as an artifact estimators in the case of small sample sizes. The application of PLS-PM
built by integrating IT technical infrastructure, IT personnel, and IT and has often been justified by the size of the investigated sample [26]. It is
business processes. These are two examples of artifacts that have re- true that PLS-PM is capable of estimating models with more parameters
cently been examined in the field of IS. than observations because it only estimates partial model structures,
While this study argues for the use of the composite model to op- but as with every other statistical method, the standard errors of the
erationalize artifacts, recently it has been suggested to employ the estimates increase as the sample size decreases. Therefore, justifying
composite model to operationalize behavioral concepts [58]. This no- the use of PLS-PM due to small sample sizes should be considered
tion assumes that both latent and emergent variables serve as a proxy cautiously. In this sense, claiming that PLS-PM is particularly suitable
for behavioral concepts [61]. Following this reasoning, the validity gap for small sample sizes can be regarded as problematic [26]. However, in
occurs between the concept and its construct and not between the case of pure emergent variable models and small sample size con-
construct and the observable variables [32]. stellations, PLS-PM performs superior focusing on accuracy in the

4
J. Benitez, et al. Information & Management 57 (2020) 103168

estimation of path coefficients compared to other variance-based esti- estimate models of this kind. First simulation studies have investigated
mators [89], i.e., generalized structured component analysis (GSCA the performance of PLSc to other estimators in this situation and in fact,
[90],) and regression with sum scores. its usage for pure latent models is considered as acceptable and its bias
Traditional view 3: PLS-PM cannot be used for non-recursive for finite samples has been evidenced of little practical relevance as-
models. Although current user-friendly software packages do not yet suming that the model is correctly specified ([42,61,93]).
implement approaches to analyzing non-recursive models, the as- Traditional view 7: Mode B can be used to estimate cau-
sumption of recursivity can be relaxed by estimating the structural sal–formative measurement models consistently. Mode B, or so-called
model parameters using 2SLS or 3SLS instead of OLS ([42,27]). Another regression weights, cannot be used to consistently estimate cau-
approach to estimate non-recursive structural models involves using the sal–formative measurement models, as this kind measurement model is
construct scores (in the case of emergent variables) or the disattenuated not identified by its own [94]. However, by the development of PLSc,
construct correlation matrix (in the case of latent variables) obtained by PLS-PM can consistently estimate causal–formative measurement
PLS-PM as input for the full-information maximum-likelihood (FIML) models by means of the MIMIC model [16].
estimator (e.g. [83],). Traditional view 8: Overall fit of models estimated by PLS-PM
Traditional view 4: Model identification plays no role when em- cannot be assessed. Due to recent developments in the context of PLS-
ploying PLS-PM. PLS-PM always employs composites to estimate the PM, the overall fit of models estimated by PLS-PM can be assessed in
model, whether or not the theoretical concepts are operationalized by a two non-exclusive ways: (1) by a bootstrap-based test for overall model
measurement model or a composite model. Therefore, the identification fit [42], and (2) by measures of overall model fit such as the SRMR
rules for composite models must be applied [28,134]. In addition to [22]. Both ways assess the difference between the empirical indicator
normalization of the weight vector such as fixing the variance of each variance–covariance matrix and the estimated model-implied counter-
composite to one, it must be ensured that each construct is connected part. While the empirical indicator variance–covariance matrix con-
(by means of a non-zero path) with at least one other construct in the tains the variances and covariances of the indicators based on the
model to ensure that the weights can be uniquely retrieved from the sample, the estimated model-implied counterpart contains the var-
population indicator variance–covariance matrix. Although all weight iances and covariances of the indicators implied by the model structure
vectors are scaled and no construct is isolated in the structural model, based on the estimated model parameters. Typically, the discrepancy
the signs of the weights of a block of indicators are still ambiguous. The between the two matrices is measured by the squared Euclidean dis-
dominant indicator approach is thus recommended to fix construct tance (dULS), the geodesic distance (dG), and the SRMR. The bootstrap-
scores’ orientation and thereby uniquely determine the weights [3]. based test for overall model fit relies on a bootstrap procedure to obtain
Additionally, the structural model must be identified. As long as one the reference distribution of the distance measures under the null hy-
only considers recursive structural models with uncorrelated error pothesis that the population indicator variance–covariance matrix
terms, identification is straightforward as they are always identified equals the model-implied counterpart [95]. Assuming a 5% level of
[1]. significance, a discrepancy value larger than the 95% quantile of the
Traditional view 5: PLS-PM has greater statistical power than the corresponding reference distribution leads to rejection of the null hy-
ML estimator. In estimating latent variable models, Reinartz et al. [91] pothesis. In addition to the bootstrap-based test for overall model fit,
claim that the power of statistical testing is higher when PLS-PM esti- the values of the distance measures can be compared to threshold va-
mates are employed than when ML estimates are used. However, these lues recommended by the literature to assess overall model fit. Mea-
findings are highly questionable, as they are based on traditional PLS- sures of fit are thus based on heuristic rules rather than on statistical
PM, which is known to produce inconsistent parameter estimates for inference. Moreover, the suggested thresholds for the measures of
latent variable models. In line with Goodhue et al. [30], who show that overall model fit in the context of PLS-PM, e.g., 0.080 for the SRMR, are
this alleged higher power goes along with an inflated type I error, we preliminary and need to be examined in more detail in future research.
conclude that preferring PLS-PM over the ML estimator due to effi- Traditional view 9: Reliability of the construct scores obtained by
ciency is not a valid argument for latent variable models. Similar PLS-PM should be assessed using Cronbach’s α and Dillon–Goldstein’s ρ
findings observed for GSCA are applicable to PLS-PM [92]. For emer- (also called Jöreskog’s ρ or composite reliability). Traditionally, the
gent variable models, however, traditional PLS-PM has shown favorable literature recommended determining the reliability of PLS-PM construct
properties among variance-based estimators, i.e., GSCA, sum scores, scores through Cronbach’s α and Dillon–Goldstein’s ρ [12]. However,
and PLS-PM [89]. considering this recommendation, several aspects of these two mea-
Traditional view 6: Mode A can be used to consistently estimate sures have been widely neglected. First, Cronbach’s α and Dillon–-
reflective measurement models. In its most modern appearance, PLS- Goldstein’s ρ both assess the reliability of sum scores (construct scores
PM can deal with models containing both emergent and latent vari- obtained by equally weighted indicators) created for the latent variable.
ables. Because PLS-PM inherently estimates composite models [31,32], However, PLS-PM allows the indicator weights used for the calculation
it is the estimator of choice for models containing emergent variables of the construct scores to vary such that indicators with a smaller
only [61]. In PLS-PM, composite models can be consistently estimated amount of random measurement error take on greater weight than in-
by Mode B [28]. To obtain consistent parameter estimates for reflective dicators containing a larger amount of random measurement error.
measurement models, PLSc should be used. Estimates obtained by tra- Consequently, the PLS-PM construct scores contain less measurement
ditional Mode A, or Mode B and C as well, suffer from the attenuation error and are generally more reliable than sum scores [22]. Second,
bias [42]. In contrast, PLSc produces consistent and asymptotically Cronbach’s α assumes tau-equivalence, i.e., equal population covar-
normal estimates for reflective measurement models by combining iances among the indicators belonging to one latent variable, an as-
Mode A estimates with a correction for attenuation [41]. Consequently, sumption that is rarely met in empirical research [34]. While Cron-
the development of PLSc [41] enables PLS-PM to analyze models con- bach’s α can be calculated based on the sample variance–covariance
taining both emergent and latent variables. However, scholars should matrix, Dillon–Goldstein’s ρ is based on factor loadings. Because tra-
use Mode B in PLS-PM (instead of PLSc) when they estimate pure ditional PLS-PM is known to produce inconsistent factor loading esti-
emergent variable models as PLSc has shown to produce biased esti- mates, Dillon–Goldstein’s ρ should be based on consistent factor loading
mates in this situation [61]. estimates obtained by PLSc. Furthermore, as the assumptions of Cron-
That being said, in the case of pure latent variable models, covar- bach’s α and Dillon–Goldstein’s ρ are likely to be violated in empirical
iance-based estimators are preferred, as they are consistent and research, their use cannot be recommended. However, the reliability
asymptotically efficient. However, the availability of asymptotically obtained by Cronbach’s α can be regarded as a lower bound [33]. To
efficient estimators does not mean that scholars cannot use PLS-PM to consistently estimate the reliability of latent variable scores obtained by

5
J. Benitez, et al. Information & Management 57 (2020) 103168

PLS-PM’s Mode A, Dijkstra–Henseler’s ρA should be used [41]. relative performance in key business processes as compared with its key
Traditional view 10: Discriminant validity should be examined by competitors [98]. Fig. 1 presents the research model of the example.
the Fornell–Larcker criterion. Although the Fornell–Larcker criterion Based on prior IS research on social media in organizations [77,99], it is
[96] had been long recommended to assess discriminant validity of assumed that social executive behavior and social employee behavior
latent variables [12], it is ineffective in combination with traditional positively affect development of a firm’s social media capability, which,
PLS-PM because it relies on consistent factor loading estimates [22]. To in turn, may positively influence firm’s business processes performance.
overcome this drawback, the HTMT was developed to assess dis- The research model represents the theory proposed by an author/
criminant validity in the case of variance-based estimators [35]. The team to be tested empirically. It illustrates how the theoretical concepts
HTMT can be assessed in two ways: (1) by comparing it to a threshold are operationalized, i.e., how the indicators are related to the constructs
value, and (2) by constructing a confidence interval to examine whether representing the theoretical concepts, and how these constructs are
HTMT is significantly smaller than a certain threshold value ([35] connected. It usually includes several hypotheses to be tested. Based on
[37],). For the first approach, simulation studies suggest a threshold prior literature and anecdotal evidence from the real world, authors
value of 0.90 if constructs are conceptually very similar or 0.85 if the should explain one by one why the hypothesized relationships are in-
constructs are conceptually more distinct ([35–37]). For the second cluded and state expectations about their signs. These explanations are
approach, prior methodological research has suggested to examine omitted from this article because theoretical explanation of the re-
whether HTMT is significantly smaller than 1 [35] or below other lationships included in the example is beyond the paper’s scope. In our
smaller values, e.g., 0.85 or 0.90 [37] [37]. conclude that HTMT is a example, the following three hypotheses are tested:
reliable tool for assessing discriminant validity, whereas the For-
Hypothesis 1 (H1). Social executive behavior has a positive impact on the
nell–Larcker criterion has limitations that do not justify its reputation
development of social media capability.
for rigor and its widespread use in empirical research.
Hypothesis 2 (H2). Social employee behavior has a positive impact on the
development of social media capability.
3. An illustrative example
Hypothesis 3 (H3). Social media capability has a positive impact on
business process performance.
3.1. Description of the example
Although prior IS studies using PLS-PM have investigated more
We provide an illustrative IS example to present the latest en- complex models (e.g., including a greater number of constructs, second-
hancements of PLS-PM. Fig. 1 displays the proposed research model to order constructs, moderation effects), the presented research model
be estimated and tested. For this purpose, we use a simulated dataset of seems reasonable for our purposes due to the following three reasons:
300 observations, where each observation represents a firm – the unit of (1) the goal of our study is to provide guidelines for using PLS-PM in
analysis in the example. Because we use a simulated dataset, the ob- causal IS research (confirmatory and explanatory), employing the most
tained results are not scientifically relevant and any comparison of our recently proposed standards. In sake of brevity, parsimony, and peda-
results to results of other empirical studies is only made for purely il- gogical illustration for IS scholars, we think, in line with Occam’s razor,
lustrative purposes. the simpler the research model is, the better. "Parsimonious yet well-
Social executive behavior is the positive/negative behavior of the fitting models are more likely to be scientifically replicable, explain-
firm’s top managers towards the firm’s use of social media for business able" [100]. “Parsimony is also regarded by many social scientists as an
activities. Social employee behavior is the positive/negative behavior of important ingredient in theory development (e.g. [101,102],), precisely
the firm’s employees towards the firm’s use of social media for business because it ‘explains much by little’ ([103]; p.153)” [100]; (2) the
activities. Social media capability refers to the firm’s ability to use and considered model contains both latent variables (ovals) and emergent
leverage external social media platforms purposefully to execute busi- variables (hexagons), and therefore, presents a situation in which PLS-
ness activities [77,97]. Business process performance is the firm’s

Fig. 1. Research model (CV = Control variables).

6
J. Benitez, et al. Information & Management 57 (2020) 103168

PM can leverage its full capacities; and (3) the research model is the-
oretically positioned in IS literature on business value of IT, where the
research models are usually parsimonious (e.g. [104,105],).
Theoretical concepts of behavioral research such as personality
traits, individual behavior, and individual attitude are usually re-
presented as latent variables [106]. Because social executive behavior
and social employee behavior indicate types of individual behavior and
attitude, the two theoretical concepts were operationalized by reflective
measurement models. The ovals represent the latent variables and the
connected rectangles their indicators. Social executive behavior and
Fig. 3. Modeling a nominal control variable.
social employee behavior were each measured by four indicators,
Note: Industry group 4 is the reference group.
(SEXB1-SEXB4) and (SEMB1-SEMB4), respectively. To obtain con-
sistent estimates, the reflective measurement models were estimated by
PLSc [41]. weights of the industry composite can be interpreted as a simple con-
In contrast, the theoretical concepts social media capability and trast, i.e., the difference in contribution to the total industry effect
business process performance were considered as artifacts designed by between the industry considered and the reference industry. Fig. 3
firms, executives, and/or employees. To operationalize these theore- presents how industry, a nominal control variable, was included in the
tical concepts, the composite model was employed. In doing so, social structural model. IS scholars can use the dominant or the most im-
media capability is assumed to be composed of the following in- portant industry as the reference group.
gredients: Facebook, Twitter, corporate blog(s), and LinkedIn cap-
abilities [77], which are the ingredients that shape social media cap-
abilities and are lower-order capabilities. IS scholars and analysts from 3.2. Statistical power analysis
other contexts (e.g., China) might consider the social media WeChat
(capability) as a key ingredient of social media capability and might A power analysis should typically be conducted before data col-
remove other, less relevant social media capabilities for Chinese firms. lection. It gives insight into the minimum sample size required to obtain
This illustrates the potential of including/studying different artifacts to sufficient statistical accuracy to detect effects of interest existing in the
investigate the same phenomenon of interest for firms and society. The population. The power of a statistical test is the probability of rejecting
artifact business process performance was also operationalized by a the false null hypothesis correctly, that is, of finding an effect in the
composite model. It comprises supplier relations, product and service sample if it indeed exists in the population [107]. Power analysis can be
enhancement, production and operations, marketing and sales, and conducted in two ways: (1) using heuristic rules such as Cohen’s power
customer relations (Tallon and Pinsoneault 2011). Fig. 2 illustrates how tables and the inverse square root method [108,109], and (2) con-
the artifact social media capability was operationalized. The hexagon ducting a Monte Carlo simulation study [110]. The 10-times rule [111]
represents the construct, i.e., the emergent variable, while the rec- or the minimum R2 rule is no longer recommended to estimate the
tangles represent ingredients forming the construct. minimum sample size [26,46,109].
Besides the variables of main interest, firm size and industry were To apply Cohen’s power tables for multiple regression analysis, four
included as control variables in the structural model to control for ef- parameters must be considered: effect size (the extent to which the path
fects of extraneous variables [80,81]. Firm size was modeled as a single- coefficient/weight exists in the population), power (probability of re-
indicator composite to account for the role of different firm sizes in jecting the false null hypothesis correctly), significance level (prob-
explaining business process performance through the natural logarithm ability of rejecting the true null hypothesis incorrectly), and the number
of the number of employees [76]. Due to the skewed distribution, it is of independent variables of the equation containing the considered path
advisable to also apply the logarithm when the firm size is measured coefficient/weight. Once these values are determined, Cohen’s power
through sales or total assets. Industry was incorporated as a composite tables can be used to approximate the minimum required sample size in
to control for an overall industry effect on business process performance order to achieve a certain power level. To determine the number of
and was shaped by three indicators, i.e., industry groups 1–3. The three required observations, analysts can assume a small effect size (0.020 ≤
industry group dummies indicate whether an observation belongs to f2 < 0.150) for a more conservative approximation or a medium to
industry 1, 2, or 3. Each industry assigns 0 if the observation does not large effect size (0.150 ≤ f2 < 0.350 or f2 ≤ 0.350) for a more opti-
belong to the industry and 1 if the observation does. For example, the mistic approximation of the required sample size. The statistical power
variable industry group 1 will have a value of 0 for firms that do not is usually set to 0.8, and a significance level of 0.05 is assumed [107].
belong to industry group 1 and a value of 1 for firms that belong to Often the equation with the highest number of independent variables is
industry group 1. Although the dataset consists of four different in- considered to determine the minimum number of observations to re-
dustries, industry group 4 was not included to avoid perfect multi- liably detect an effect. In our example, the composite model for busi-
collinearity. Therefore, group 4 became the reference category. The ness process performance has the highest number of independent
variables (supplier relations, product and service enhancement, pro-
duction and operations, marketing and sales, and customer relations) in
an equation. Cohen’s power tables suggest a minimum sample size of 91
observations assuming a medium effect size (f2 = 0.150), statistical
power of 0.8, and significance level of 0.05 [108]. Considering the
outcomes of the power analyses for our example, a sample size of 300
seems adequate to detect the effects of interests. The inverse square root
method assumes that the estimates are standard normally distributed,
and approximates the standard error using N . Assuming a 5% sig-
nificance level, the required sample size to obtain a statistically sig-
nificant effect (N̂ ), if it exists in the population, can be approximated by
N̂ > ( ) , where | |
2.486 2
| |min min represents the minimum magnitude of the
coefficient considered.
Fig. 2. Operationalization of the artifact social media capability. In addition to considering heuristic rules, IS scholars can conduct a

7
J. Benitez, et al. Information & Management 57 (2020) 103168

Monte Carlo simulation to examine the sample size required to reliably Table 2
detect effects that exist in the population. A population model with the Results of the confirmatory factor/composite analysis.
same structure as the estimated model must be specified and all po- Discrepancy Overall saturated model fit evaluation
pulation parameter values need to be determined. In the second step,
the model is estimated several times and the rejection rates of the null Value HI95 Conclusion
hypothesis significance test for the coefficients under examination are
SRMR 0.030 0.049 Supported
considered, i.e., the statistical power. The appealing property of this dULS 0.210 0.546 Supported
approach is that it can take into account various aspects of the model dG 0.049 0.221 Supported
and indicators incorporated, such as sample size, number of indicators,
their distribution, and magnitude of the effect. Moreover, sensitivity
analyses can be conducted by changing the assumed population model an emergent variable?” Table 2 contains the values of the discrepancy
to see how these changes affect the statistical power. While guidelines measures and 95% quantiles of their corresponding reference dis-
have been proposed for pure latent variable models in the context of tribution for our example. The value of the SRMR was below the re-
PLS-PM [110], development of guidelines for models containing commended threshold value of 0.080 [22,119]. However, the thresh-
emergent variables is still an open issue. olds for the overall model fit in the context of PLS-PM should be
considered cautiously as they are preliminary and need to be examined
3.3. Estimation in more detail in future methodological research. Moreover, all dis-
crepancy measures were below the 95% quantile of their reference
Various software packages – such as PLS-Graph [112], SmartPLS distribution (HI95). Empirical evidence was thus obtained for the latent
[113], WarpPLS [114], XLSTAT-PLS [115], and ADANCO [116] – can variables (social executive behavior and social employee behavior) as
be used to estimate the model with PLS-PM. We used ADANCO 2.0.1 well as the emergent variables (social media capability and business
Professional for Windows (http://www.composite-modeling.com/) process performance) incorporated in the model. In case of contra-
[116] to estimate the empirical example. In the following, we used dictory results for the measure of fit (SRMR) and the test of overall
Mode B to estimate composite models and PLSc to estimate reflective model fit (dULS, and dG), the test for overall model fit is preferred, as it is
measurement models. Moreover, we used the factor weighting scheme based on statistical inference rather than heuristic rules. Moreover, if
for inner weighting and statistical inferences were based on the boot- none of the discrepancies was below the 95% quantile of the corre-
strap procedure, relying on 4999 bootstrap runs. sponding reference distribution (HI95), analysts can evaluate whether
Prior to model estimation, analysts should set a dominant indicator the discrepancies are at least below the 99% quantile (HI99) before fi-
in each composite and reflective measurement model. As the signs of nally rejecting the model. In the next step, each measurement and
the weight and factor loading estimates of a block of indicators are composite model must be examined separately. Authors of future stu-
ambiguous, the dominant indicator is used to dictate the orientation of dies in IS research are encouraged to report a table like Table 2.
a construct. A dominant indicator that is expected to positively corre- Scholars should assess content validity for both kinds of constructs,
late with the construct is preferable. Face validity can be used to select i.e., latent variables and emergent variables, by carefully considering
the dominant indicator – the indicator that is theoretically most re- each type of construct and how the according concept has been oper-
levant and thus expected to positively correlate with the construct. In ationalized in prior research. In the case of emergent variables, how-
our example, we chose SEXB2, SEMB2, SMC1, and BPP4 as dominant ever, it might be desirable to modify the weighting scheme, number of
indicators. indicators, and content of the indicators as illustrated in the bread and
Before the model assessment, the researcher has to ensure that the beer example. Finally, construct validity should be assessed. Depending
estimation is technically valid, i.e., that the estimation is admissible and on the concept’s operationalization, this can be done in several non-
no Heywood case has occurred [117]. In doing so, he/she needs to exclusive ways.
investigate whether the PLS-PM algorithm has properly converged.
Additionally, in particular in the context of PLSc, he/she needs to en- 3.4.2. Assessment of the reflective measurement model
sure that the construct correlation and the model-implied indicator For reflective measurement models in which latent variables re-
correlation matrix are valid, i.e., positive semi-definite. To assess the present behavioral concepts such as social executive behavior and so-
definiteness of a matrix, user-written Excel plugins for the calculation of cial employee behavior, composite reliability, convergent validity, in-
Eigenvalues can be used. A symmetric matrix is positive semi-definite if dicator reliability, and discriminant validity should be evaluated.
all Eigenvalues are larger or equal to 0. Finally, all absolute factor Dijkstra–Henseler’s ρA should be considered in assessing composite re-
loading estimates and reliability estimates must be smaller or equal to liability (the correlation between latent variable and construct scores).
1. For our example, the solution was technically valid. A value of Dijkstra–Henseler’s ρA larger than 0.707 can be regarded as
reasonable, as more than 50% of the variance in the construct scores
3.4. Assessment of reflective measurement and composite models can be explained by the latent variable [120]. Table 3 shows that the
values of Dijkstra–Henseler’s ρA for social executive behavior and social
3.4.1. Evaluation of overall fit of the saturated model employee behavior. Both are 0.938 and 0.913, and thus above the
Table 4 summarizes the steps to assess reflective measurement and suggested threshold of 0.707, indicating reliable construct scores.
composite models. Joint assessment should begin with the evaluation of Convergent validity is the extent to which the indicators belonging
the overall fit of a model with a saturated structural model [16,118], to one latent variable actually measure the same construct. The average
that is, with confirmatory factor/composite analysis. The estimated variance extracted (AVE), typically used to assess convergent validity
model is as specified by analysts [118]. The saturated model corre- [121], indicates how much of the indicators’ variance can be explained
sponds to a model in which all constructs are allowed to be freely by the latent variable. An AVE larger than 0.5 has been suggested to
correlated, whereas the concept’s operationalization is exactly as spe- provide empirical evidence for convergent validity, as the corre-
cified by the analyst. The evaluation of the overall model fit of the sponding latent variable explains more than half of the variance in the
saturated model is useful to assess the validity of the measurement and belonging indicators, and consequently, all other latent variables ex-
the composite models, because potential model misfit can be entirely plain less than a half [96]. In our example, all AVE values are above 0.5
attributed to misspecifications in the composite and/or measurement (0.788 and 0.716), indicating convergent validity (see Table 3).
models. Therefore, empirical support can be obtained for the con- Indicator reliability can be assessed through the factor loading es-
structs, i.e., “Does a latent variable exist?”, or “Do the indicators form timates. As factor loading estimates are standardized in PLS-PM, the

8
J. Benitez, et al. Information & Management 57 (2020) 103168

Table 3
Measurement model evaluation.
Code Construct/indicator ρA AVE VIF Weight Loading

Social executive behavior (1: Strongly disagree, 5: Strongly agree) (reflective measurement model, Mode A consistent (PLSc), SEXB2 0.938 0.788
as dominant indicator)
SEXB1 Behavior of top business executives towards adoption of social media is positive 0.278*** 0.905***
SEXB2 Top business executives are positive in adopting social media for business activities 0.269*** 0.877***
SEXB3 Top business executives support adoption of social media for business activities 0.263*** 0.856***
SEXB4 Top business executives are willing to support adoption of social media in the firm 0.280*** 0.912***
Social employee behavior (1: Strongly disagree, 5: Strongly agree) (reflective measurement model, Mode A consistent, SEMB2 as 0.913 0.716
dominant indicator)
SEMB1 Employee behavior towards adoption of social media is positive 0.301*** 0.901***
SEMB2 Employees are positive to adopt social media in the firm 0.274*** 0.820***
SEMB3 Employees support adoption of social media in the firm 0.257*** 0.769***
SEMB4 Employees are willing to support adoption of social media in the firm 0.296*** 0.888***
Social media capability: My firm has deliberately used and leveraged… for business activities (1: Strongly disagree, 5: Strongly agree)
(composite model, Mode B, SMC1 as dominant indicator)
SMC1 Facebook 1.037 0.229*** 0.397***
SMC2 Twitter 1.032 0.489*** 0.627***
SMC3 Corporate blog(s) 1.059 0.601*** 0.751***
SMC4 LinkedIn 1.020 0.333*** 0.455***
Business process performance: Relative to your key competitors, what is your performance in last three years in the following
business processes (1: Significantly worse, 5: Significantly better than my key competitors) (composite model, Mode B, BPP4 as
dominant indicator)
BPP1 Supplier relations 1.022 0.285** 0.397**
BPP2 Product and service enhancement 1.134 0.553*** 0.307**
BPP3 Production and operations 1.105 0.108 0.203†
BPP4 Marketing and sales 1.064 0.609*** 0.531***
BPP5 Customer relations 1.063 0.629*** 0.591***

Note: †p < 0.10, *p < 0.05, **

p < 0.01, ***
p < 0.001, one-tailed test.

squared factor loading estimate equals the estimated indicator relia- While weights show the relative contribution of an indicator to its
bility. It is generally advisable for factor loadings to be greater than construct, composite loadings represent the correlation between the
0.707, indicating that more than 50% of the variance in a single in- indicator and the corresponding emergent variable; a loading shows the
dicator can be explained by the corresponding latent variable. In this absolute contribution of an indicator to its construct [122]. As weights
context, also the significance of the factor loading estimates should be show the degree of importance of each indicator (ingredient) to the
investigated. Somewhat lower values are not really problematic as long construct, analysts should examine whether all indicator weight esti-
as the construct validity and reliability criteria are met. Table 3 presents mates are significant. For indicators with non-significant weight esti-
the factor loading estimates from our example. They range from 0.769 mates, one must investigate whether composite loading estimates are
to 0.912 and are all significant on a 1‰ level, suggesting that the statistically significant and consider dropping any indicators with non-
measures are reliable. significant weight and loading estimates. However, content validity
Discriminant validity entails that two latent variables that are meant must be considered as well, because dropping an indicator may alter the
to represent two different theoretically concepts are statistically suffi- meaning of the emergent variable. IS scholars can thus decide to keep
ciently different. To obtain empirical evidence for discriminant validity, an indicator with non-significant weight and loading to preserve the
IS scholars should consider the HTMT [35]. The HTMT should be lower construct’s content validity [46].
than 0.85 (more strict threshold) or 0.90 (more lenient threshold) or Table 3 shows that the VIF values for the indicators of the composite
significantly smaller than 1 ([36,37]). In our example, the HTMT of models range from 1.020 to 1.134, suggesting that multicollinearity is
social executive behavior to social employee behavior is 0.322, and not a problem in our data. Moreover, all weight and composite loading
thus below the recommended threshold of 0.85 (and of 0.90). More- estimates show the expected sign and are significant at a 5% sig-
over, the one-sided 95% percentile confidence interval of HTMT does nificance level except one (estimated weight of the indicator production
not cover 1, that is, it is significantly different from 1. Scholars can also and operations of the construct business process performance). The
follow [37] suggestion to test whether the HTMT is significantly smaller weight estimate of this indicator is 0.108, and its composite loading
than 0.85 or 0.90. estimate is 0.203† (close to be significant). Considering content validity,
the indicator production and operations may include some of the firm’s
key business processes. Therefore, we decided to keep the indicator in
3.4.3. Assessment of the composite model the empirical analysis to preserve content validity and avoid altering
The composite model requires an evaluation sui generis – an ex- the meaning of the emergent variable business process performance. In
amination of the composite model with respect to multicollinearity, this type of situation, analysts can also repeat the analysis, dropping the
weights, composite loadings, and their significances [43,76,77,83,122]. questionable indicators to explore whether the decision to keep or drop
As composite models are typically estimated by Mode B (regression these indicators affects the results. We dropped BPP3 and repeated the
weights) in PLS-PM, collinearity among indicators forming an emergent empirical analysis. The results obtained were qualitatively identical,
variable should be investigated by means of the variance inflation suggesting that this decision does not affect the research findings. One
factor (VIF), as high multicollinearity can lead to insignificant estimates might ask why BPP3 should be included when the results do not
and unexpected signs of the weights. Traditionally, VIF values above 5 change. This is theoretically justified, as it is difficult to imagine a
are regarded as indications of problematic multicollinearity [38,46]. company’s business process performance without production and op-
Yet, typical phenomena of multicollinearity can also occur in case of erations processes, as these often are the heart of a company. Because
VIF values far below 5. For weights estimated by Mode A, an assessment all reflective measurement and composite models from our example
of multicollinearity is not necessary as these equal scaled covariances, show desirable properties, we proceed to evaluate the structural model.
and therefore, ignores multicollinearity [32].

9
 J. Benitez, et al.

Table 4
Steps to assess common factor and composite models.
Steps Type of construct Description Assessment Decision criterion Interpretation
criterion

Testing the adequacy of reflective Latent and emergent Evaluate the overall fit of the model with a saturated SRMR SRMR < 0.080 SRMR < A SRMR value smaller than 0.080 indicates an acceptable
measurement and composite variable structural model by investigating discrepancy between HI95 model fit [22]; however, these thresholds are preliminary and
models empirical and model-implied indicator variance–covariance need to be investigated in more detail
matrix dULS dULS < HI95 The null hypothesis that the population indicator
dG dG < HI95 variance–covariance matrix equals the model-implied
counterpart is not rejected. Hence, empirical evidence for the
model is given when the value of the discrepancy measure is
below the 95% quantile of its corresponding reference
distribution
Evaluating content validity Latent and emergent How the corresponding theoretical concepts have been Flexibility in the case of artifacts represented by an emergent variable (bread and beer analogy)
variable operationalized (measured or built) in prior research
Evaluating reliability of construct Latent variable Evaluating whether the construct scores reliably represent ρA ρA > 0.707 More than 50% of the variance in the construct scores can be

10
scores the underlying construct explained by the underlying latent variable
Evaluating indicator reliability Evaluating whether indicators are reliable Factor loading Factor loading estimates > More than 50% of the indicator’s variance is explained by the
estimates 0.707 latent variable
Factor loading Significant at 5%
significance significance level
Evaluating convergent validity Latent variable Evaluating the share of variance in the indicators that is AVE AVE > 0.5 More than 50% of indicators’ variance is explained by the
explained by the underlying latent variable underlying latent variable
Evaluating discriminant validity Latent variable Evaluating whether two latent variables are statistically HTMT HTMT < 0.85 (or whether Factors are statistically different and thus have discriminant
different the HTMT is significantly validity
smaller than 1)
Multicollinearity Emergent variable Evaluating how the standard errors of the weight estimates VIF VIF < 5 If the estimates suffer from multicollinearity, weights obtained
(estimated by Mode are affected by the correlations of the indicators by Mode A or predetermined weights can be used
B)
Weights Emergent variable Evaluating relative contribution of an indicator to its Weights’ value and Significant at 5% Each indicator contributes significantly to the emergent
construct significance significance level variable
Loadings Emergent variable Evaluating absolute contribution of an indicator to its Loading Significant at 5% Each indicator contributes to the emergent variable in a
construct significance significance level statistically significant way
Information & Management 57 (2020) 103168
J. Benitez, et al. Information & Management 57 (2020) 103168

3.5. Assessment of the structural model Table 6

Construct correlation matrix.
In evaluating the structural model, the analyst should examine the 1 2 3 4 5 6
overall fit of the estimated model, the path coefficient estimates, their
significance, the effect sizes (f2), and the coefficient of determination 1. Social executive 1.000
behavior
(R2, [3,123]). Analysts should focus specifically and primarily on
2. Social employee 0.322 1.000
overall model fit in confirmatory research and primarily on R2, the path behavior
coefficient estimates, and the effect sizes in explanatory research [24]. 3. Social media capability 0.550 0.532 1.000
Table 7 summarizes the steps to follow in evaluating the structural 4. Business process 0.216 0.309 0.515 1.000
model. performance
5. Firm size −0.025 −0.048 −0.014 0.016 1.000
6. Industry 0.069 0.073 0.010 0.036 0.038 1.000

3.5.1. Evaluation of the overall fit of the estimated model

First, analysts should evaluate the overall fit of the estimated model 3.5.2. Evaluation of path coefficients and their significance levels
through the bootstrap-based test of overall model fit and the SRMR as a The path coefficient estimates are essentially standardized regres-
measure of approximate fit to obtain empirical evidence for the pro- sion coefficients, whose sign and absolute size can be assessed. These
posed theory. Analysis in confirmatory research without assessing the coefficients are interpreted as the change in the dependent construct
overall model would be incomplete as this means ignoring empirical measured by standard deviations, if an independent construct is in-
evidence for and also against the proposed model and the postulated creased by one standard deviation while keeping all other explanatory
theory [124]. Without assessing the model fit, a researcher would not constructs constant (ceteris paribus consideration). For example, in-
obtain any signal if he or she had incorrectly omitted an important creasing social media capability by one standard deviation will increase
effect in the model. Because the test for overall model fit was in- business process performance by 0.515 standard deviations if all other
troduced only recently in the context of PLS-PM, the vast majority of variables are kept constant. Statistical tests and confidence intervals can
models estimated by PLS-PM in past IS research has not been evaluated be used to draw conclusions about the population parameters. For
in this respect. However, because the overall model fit can now be confidence intervals, the percentile bootstrap confidence interval is
tested in the context of PLS-PM, we encourage IS scholars to take this recommended [125]. As shown in Fig. 4, the path coefficient estimates
evaluation very seriously in causal research. In our example, all values for the hypothesized relationships included in the example range from
of discrepancy measures were below the 95% quantile of their corre- 0.396 to 0.515, and are all significant at a 5% significance level except
sponding reference distribution (HI95), indicating that the estimated the effect of the two control variables, firm size and industry. A path
model was not rejected at a 5% significance level (see Table 5). coefficient estimate is considered as statistically significant different
Moreover, the SRMR was below the preliminary suggested threshold of from zero at a 5% significance level when its p-value is below 0.05 or
0.080, indicating an acceptable model fit. This result suggests that the when the 95% bootstrap percentile confidence interval constructed
proposed model is well suited for confirming and explaining the de- around the estimate does not cover the zero.
velopment of social media capability and business process performance
among firms. While the model fit suggests that there is a possibility that
3.5.3. Evaluation of effect sizes
the world functions according to the specified model, the model can
The practical relevance of significant effects should be investigated
still be misspecified in the sense of over-parameterization, i.e., the
by considering the effect sizes of the relationships between the con-
model contains superfluous zero-paths [22]. Neither the bootstrap-
structs. The effect size is a measure of the magnitude of an effect that is
based test of model fit nor the SRMR punishes for unnecessary paths,
independent of sample size. The f2 values ranging from 0.020 to 0.150,
i.e., neither of them rewards parsimony. Regardless of whether one
0.150 to 0.350, or larger or equal to 0.350, indicating weak, medium, or
conducts confirmatory or explanatory research, it remains indis-
large effect size respectively [108]. Just as all actors in a movie cannot
pensable to assess all path coefficients and their significance. Table 6
play a leading role, it is unusual and unlikely that most constructs will
presents the construct correlation matrix.
have a large effect size in the model. We provide this clarification be-
cause scholars often expect/self-demand that all/most of their effect

Table 5
Structural model evaluation.
Relationship Path coefficient

Social executive behavior → Social media capability (H1) 0.422*** (8.830) [0.327, 0.512]
Social employee behavior → Social media capability (H2) 0.396*** (8.052) [0.300, 0.490]
Social media capability → Business process performance (H3) 0.515*** (10.232) [0.426, 0.609]
Firm size → Business process performance (control variable) 0.022 (0.305) [-0.128, 0.160]
Industry → Business process performance (control variable) 0.030 (0.312) [-0.161, 0.174]
Endogenous variable R2
Social media capability 0.443
Business process performance 0.267
Overall fit of the estimated model Value HI95
SRMR 0.032 0.049
dULS 0.232 0.558
dG 0.052 0.222
Effect size f2
Social executive behavior → Social media capability (H1) 0.286
Social employee behavior → Social media capability (H2) 0.252
Social media capability → Business process performance (H3) 0.362
Firm size → Business process performance (control variable) 0.001
Industry → Business process performance (control variable) 0.001

Note: t-values (one-tailed test) are presented in parentheses. Percentile bootstrap confidence intervals are presented in brackets.

11
J. Benitez, et al. Information & Management 57 (2020) 103168

magnitude be large – an unrealistic expectation. This cautionary note

Value of discrepancy measure below the 95% quantile of

Effect of independent variables on dependent variables is

words, it is possible that the empirical data stem from a
extends to supervisors’ expectations for their Ph.D. students (as illu-

empirical evidence for the postulated model. In other

Degree of variance explained for phenomenon under

strated by [126]). In our sample, the f2 values for the hypothesized
the corresponding reference distribution provides relationships range from 0.252 to 0.362 (medium to large).

world that functions as theorized by the model

3.5.4. Evaluation of R2
R2 is used to assess goodness of fit in regression analysis [87]. In the
case of models estimated by OLS, the R2 value gives the share of var-

Degree of strength of an effect

iance explained in a dependent construct. Thus, it provides insights into
a model’s in-sample predictive power [127]. Moreover, R2 forms the
statistically significant

basis for several innovative model selection criteria ([37,100]). Re-

porting R² makes PLS-PM research future-proof in this regard, because
the new model selection criteria can still be calculated ex post as long as
Interpretation

investigation
the R² values are given.
The expected magnitude of R2 depends on the phenomenon in-
vestigated. As some phenomena are already quite well understood, one
would expect a relatively high R². For phenomena that are less well
understood, one would expect a high R². When the

understood, a lower R² is acceptable. The R² values should be judged

phenomena are not yet well understood, a lower
f2 < 0.020: no substantial effect 0.020 ≤ f2 <
Significant at 5% significance level, i.e., p-value

medium effect size f2 ≥ 0.350: large effect size

0.150: weak effect size 0.150 ≤ f2 < 0.350:

relative to studies that investigate the same dependent variable. In our

When the phenomena are already quite well

example, the R2 values for social media capability and business process
performance are 0.443 and 0.267, respectively. The study of social
media in organizations is in its initial stages [77]. Braojos et al. [128]
SRMR < 0.080 SRMR < HI95

report an R2 value of 0.541 for social media capability. In our example,

social executive behavior and social employee behavior explain 44.3%
of variance in development of social media capability, using two un-
Suggested threshold

explored exogenous variables for social media capability (social ex-

ecutive behavior and social employee behavior). Considering explained
R² is acceptable
dULS < HI95

variance in prior IS research and the originality of our two exogenous

dG < HI95

variables in influencing social media capability, an R2 of 0.443 seems to

< 5%

be an excellent value.
The models of [129,130]) explain 49% and 43.9% of the variance in
business process outcomes. In our example, social media capability,
firm size, and industry explain 26.7% of the variance in business pro-
estimates and their

cess performance. Although this R2 value is somewhat smaller than

significance level
Path coefficient

those obtained by [129,130]), it can be considered as satisfactory be-

cause our model is the first using social media capability to explain
Criterion

f2 value

business process performance individually. The independent variables

SRMR
dULS

explaining business process outcomes in [129,130]) work refer to other

dG

R2

IT resources (e.g., IT assets, enterprise resource planning capabilities)

sample size. Give an indication about the practical relevance of
variance–covariance matrix and its model-implied counterpart

Standardized regression coefficients are interpreted as change

different from social media capability. This subsection illustrates by our

while all other independent variables in the equation remain

Measure of the magnitude of an effect that is independent of

independent variable is increased by one standard deviation
Evaluating overall fit of the estimated model by evaluating

fictive example how analysts can report and compare their R2 values.
in standard deviations of the dependent variable if an

4. Discussion and conclusions

Explained variance of an dependent construct
discrepancy between the empirical indicator

IS research often tackles complex research problems and questions

that require conceptualization and operationalization of different types
of theoretical concepts, i.e., behavioral concepts and artifacts, as well as
the estimation of their relationships. PLS-PM is a suitable estimator for
this purpose. How can one perform and report an impactful analysis
Steps to follow in performing structural model evaluation.

using PLS-PM in IS research following the recent improvements in PLS-

PM? This study provides thorough guidelines on PLS-PM in the fra-
mework of causal (confirmatory and explanatory) research, employing
the latest standards recommended. In doing so, it addresses the why
Description

and how to perform and report a PLS-PM estimation in confirmatory

an effect
constant

and explanatory IS research, illustrated by a fictive example on business

value of social media. This is the key contribution of this paper to the
methodological literature in IS empirical research.
In the last five years, methodologists have overcome major weak-
Overall fit of estimated model

nesses of traditional PLS-PM, such as its inconsistency for latent vari-

Consider path coefficient

Consider effect sizes (f2)

able models and lack of a test for overall model fit. To benefit from all
estimates and their
significance levels

these enhancements, IS scholars need new guidelines for empirical

studies that incorporate all these recent new developments and insights,
as most of the guidelines papers on PLS-PM in the IS research were
Evaluate R2

published before 2013 (e.g., [12,43–45]). Although several recent

Table 7

Steps

scholarly textbooks and articles (e.g. [46–48],) have provided guide-

lines for causal research that cover some of the latest enhancements to

12
J. Benitez, et al. Information & Management 57 (2020) 103168

Fig. 4. Results of the structural model.

Note: ***p < 0.001.

PLS-PM, neither of these PLS-PM guidelines for causal research covered methodological toolbox.
the full range of recent developments, nor did they introduce any new The adjusted coefficient of determination (adjusted R2) has been
framework for applying PLS-PM and reporting its outcomes. To address proposed by prior methodological research [3,46,123] as a criterion to
this shortcoming in the existing IS literature, this paper provides up- assess the structural model in explanatory research [24]. Using a Monte
dated guidelines on the use of PLS-PM in assessment of reflective Carlo study [100], find that R2, adjusted R2, goodness-of-fit index, and
measurement models, composite models, and structural models. To the Q2 are not appropriate criteria to compare competing latent variable
best of our knowledge, the proposed guidelines take into account all models based on the same dataset (same sample). Their argument is
recent enhancements. An application of the guidelines is illustrated that these criteria improve with greater model complexity, and there-
using a parsimonious IS research example on business value of social fore, favor more complex models (e.g., the saturated model). To avoid
media. the shortcoming, they propose to employ the Geweke–Meese criterion
In contrast to prior guidelines [11,12], our article introduces the (GM) and Bayesian information criteria (BIC) to compare alternative
artifact – a human-made/firm-made object – as a new kind of theore- latent variable models, and find support for their proposal in a simu-
tical concept and shows how this type of theoretical concept can be lation study. “BIC and GM should be used due to their high model se-
operationalized by means of the composite model. Because a significant lection accuracy and ease of use” ([100], p. 7). Future methodological
proportion of theoretical concepts in IS research are human-made/firm- research should study the performance of adjusted R2 in comparing
made, one can expect the composite model to become the dominant different estimated models based on different samples as well as ex-
conceptualization in IS research in the coming years. Against this tending the simulation study of [100] to emergent variable models.
background, we highlight the usefulness of model testing in con- PLS-PM has become a valuable statistical tool for empirical research
firmatory and explanatory research using PLS-PM. Without considering in IS and many other disciplines of business and social sciences re-
its results, it is hardly possible to obtain empirical evidence for or search. To meet the ever-growing scholarly demand in terms of scien-
against a scholar’s proposed theory. Finally, we strongly recommend tific rigor, methodologists have continuously been improving PLS-PM.
that scholars employ consistent estimators, using PLSc when the theo- This paper helps disseminate these improvements, and enables users of
retical concept is operationalized by a measurement model. PLS-PM to be aware of and fulfill the contemporary methodological
As our article about the use of PLS-PM for causal research is limited standards.
to linear, recursive models containing only first-order constructs, future
IS research should develop additional updated guidelines incorporating Acknowledgments
recent developments for more complex models, such as models con-
taining moderation effects, second-order emergent variables of emer- This research was sponsored by the European Regional
gent variables, and for composite models that account for more com- Development Fund (European Union) and the Government of Spain
plex relationships between the indicators and the emergent variable. (Research Project ECO2017-84138-P, and Research Grant FPU14/
Although some steps have been made using PLS-PM to deal with en- 01930), the Regional Government of Andalusia (Research Project P11-
dogeneity in the form of omitted variables (e.g., [27,55]), the problem SEJ-7294), the COVIRAN-Prodware Chair of Digital Human Resource
of endogeneity requires more attention in the field of IS and in the Strategy at the University of Granada, and the Endowed Chair of
context of PLS-PM. Business Value of IT of the Green, Demand, Digital Driven-supply Chain
Although this study focuses only on PLS-PM for confirmatory and Excellence Area at Rennes School of Business. Jörg Henseler acknowl-
explanatory purposes, PLS-PM can be used for different types of re- edges a financial interest in ADANCO and its distributor, Composite
search [24]. A further promising application of PLS-PM in IS research is Modeling.
predictive modeling, which aims to produce accurate forecasts [23].
These assessment criteria differ from those employed in confirmatory Appendix A. Supplementary data
and explanatory research [131,132], and models of causal research do
not necessarily perform well when it comes to prediction purposes and Supplementary material related to this article can be found, in the
vice versa. online version, at doi:https://doi.org/10.1016/j.im.2019.05.003.
As part of the progress in methodology-related research, scholars
continuously suggest new approaches and new validity criteria. For References
instance, [133] recommend conducting redundancy analysis to assess
the convergent validity in the context of causal–formative measurement [1] K. Bollen, A new incremental fit index for general structural equation models,
and composite models. As soon as there is sufficient evidence (e.g., by Sociol. Methods Res. 17 (3) (1989) 303–316.
[2] H. Wold, Path models with latent variables: the NIPALS approach, in: H. Blalock,
means of Monte Carlo simulations) for the efficacy of the new sugges- A. Aganbegian, F. Borodkin, R. Boudon, V. Capecchi (Eds.), Quantitative
tion, scholars may consider adding this suggestion to their Sociology: International Perspectives on Mathematical and Statistical Modeling,

13
J. Benitez, et al. Information & Management 57 (2020) 103168

Academic, New York, USA, 1975, pp. 307–357. marketing: an analysis, causes for concern, and proposed remedies, J. Acad. Mark.
[3] J. Henseler, G. Hubona, P. Ray, Using PLS path modeling in new technology re- Sci. 44 (1) (2016) 119–134.
search: updated guidelines, Ind. Manage. Data Syst. 116 (1) (2016) 1–20. [37] G. Franke, M. Sarstedt, Heuristics versus statistics in discriminant validity testing:
[4] J. Hair, M. Sarstedt, T. Pieper, C. Ringle, The use of partial least squares structural a comparison of four procedures, Internet Res. (in press), pp. 1-32.
equation modeling in Strategic Management research: a review of past practices [38] J. Hair, C. Ringle, M. Sarstedt, PLS-SEM: indeed a silver bullet, J. Mark. Theory
and recommendations for future applications, Long Range Plann. 45 (5/6) (2012) Pract. 19 (2) (2011) 139–151.
320–340. [39] M. Rönkkö, J. Evermann, A critical examination of common beliefs about partial
[5] J. Hair, M. Sarstedt, C. Ringle, J. Mena, An assessment of the use of partial least least squares path modeling, Organ. Res. Methods 16 (3) (2013) 425–448.
squares structural equation modeling in Marketing research, J. Acad. Mark. Sci. 40 [40] T. Dijkstra, K. Schermelleh-Engel, Consistent partial least squares for nonlinear
(3) (2012) 414–433. structural equation models, Psychometrika 79 (4) (2014) 585–604.
[6] D. Peng, F. Lai, Using partial least squares in Operations Management research: a [41] T. Dijkstra, J. Henseler, Consistent partial least squares path modeling, Mis Q. 39
practical guideline and summary of past research, J. Oper. Manage. 30 (6) (2012) (2) (2015) 297–316.
467–480. [42] T. Dijkstra, J. Henseler, Consistent and asymptotically normal PLS estimators for
[7] C. Ringle, M. Sarstedt, R. Mitchell, S. Gudergan, Partial least squares structural linear structural equations, Comput. Stat. Data Anal. 81 (1) (2015) 10–23.
equation modeling in HRM research, Int. J. Hum. Resour. Manage. (2018) 1–27. [43] W. Chin, Issues and opinion on structural equation modeling, Mis Q. 22 (1) (1998)
[8] N. Avkiran, C. Ringle, Partial Least Squares Structural Equation Modeling, vii–xvi.
Springer, Cham, Switzerland, 2018. [44] D. Gefen, D. Straub, A practical guide to factorial validity using PLS-Graph: tu-
[9] T. Müller, F. Schuberth, J. Henseler, PLS path modeling: a confirmatory approach torial and annotated example, Commun. Assoc. Inf. Syst. 16 (1) (2005) 91–109.
to study tourism technology and tourist behavior, J. Hosp. Tour. Technol. 9 (3) [45] G. Marcoulides, W. Chin, C. Saunders, A critical look at partial least squares
(2018) 249–266. modeling, Mis Q. 33 (1) (2009) 171–175.
[10] M. Sarstedt, C. Ringle, D. Smith, R. Reams, J. Hair, Partial least squares structural [46] J. Hair, G. Hult, C. Ringle, M. Sarstedt, A Primer on Partial Least Squares
equation modeling (PLS-SEM): a useful tool for family business researchers, J. Structural Equation Modeling (PLS-SEM), 2nd ed., Sage, Thousand Oaks, USA,
Fam. Bus. Strategy 5 (1) (2014) 105–115. 2017.
[11] G. Marcoulides, C. Saunders, PLS: a silver bullet? Mis Q. 30 (2) (2006) iii–ix. [47] j. hair, c. hollingsworth, a. randolph, a. chong, an updated and expanded assess-
[12] C. Ringle, M. Sarstedt, D. Straub, A critical look at the use of PLS-SEM in MIS ment of pls-sem in information Systems research, Ind. Manage. Data Syst. 117 (3)
Quarterly, Mis Q. 36 (1) (2012) iii–xiv. (2017) 442–458.
[13] G. Polites, N. Roberts, J. Thatcher, Conceptualizing models using multi- [48] M. Sarstedt, J. Hair, C. Ringle, Partial least squares, in: C. Homburg, M. Klarmann,
dimensional constructs: a review and guidelines for their use, Eur. J. Inf. Syst. 21 A. Vomberg (Eds.), Handbook of Market Research, Springer, Berlin, Germany,
(1) (2012) 22–48. 2017.
[14] P. Lowry, J. Zhang, C. Wang, M. Siponen, Why do adults engage in cyberbullying [49] A. Van Riel, J. Henseler, I. Kemeny, Z. Sasovova, Estimating hierarchical con-
on social media? An integration of online disinhibition and deindividuation effects structs using consistent partial least squares: the case of second-order composites
with the social structure and social learning model, Inf. Syst. Res. 27 (4) (2016) of common factors, Ind. Manage. Data Syst. 117 (3) (2017) 459–477.
962–986. [50] J. Becker, I. Ismail, Accounting for sampling weights in PLS path modeling: si-
[15] L. Rueda, J. Benitez, J. Braojos, From traditional education technologies to student mulations and empirical examples, Eur. Manage. J. 34 (6) (2016) 606–617.
satisfaction in Management education: a theory of the role of social media ap- [51] M. Rademaker, F. Schuberth, T. Dijkstra, Measurement error correlation within
plications, Inf. Manage. 54 (8) (2017) 1059–1071. blocks of indicators in consistent partial least squares: Issues and remedies,”
[16] J. Henseler, Bridging Design and Behavioral research with variance-based struc- Internet Res. (in press), pp. 1-40.
tural equation modeling, J. Advert. 46 (1) (2017) 178–192. [52] S. Jung, J. Park, Consistent partial least squares path modeling via regularization,
[17] H. Reichenbach, The Direction of Time, University of California at Los Angeles, Front. Psychol. 9 (1) (2018) 1–10.
Los Angeles, USA, 1956. [53] M. Sarstedt, J. Henseler, C. Ringle, Multigroup analysis in partial least squares
[18] A. Diamantopoulos, Incorporating formatives measures into covariance-based (PLS) path modeling: alternative methods and empirical results, Measurement and
structural equation models, Mis Q. 35 (2) (2011) 335–358. Research Methods in International Marketing, Emerald, Enschede, The
[19] H. Simon, The Sciences of the Artificial, MIT Press, Cambridge, USA, 1969. Netherlands, 2011, pp. 195–218.
[20] D. Cole, S. Maxwell, R. Arvey, E. Salas, Multivariate group comparisons of variable [54] M. Klesel, F. Schuberth, J. Henseler, B. Niehaves A test for multigroup comparison
systems: MANOVA and structural equation modeling, Psychol. Bull. 114 (1) in partial least squares path modeling,” Internet Res. (in press), pp. 1-40.
(1993) 174–184. [55] T. Hult, J. Hair, D. Proksch, M. Sarstedt, A. Pinkwart, C. Ringle, Addressing en-
[21] G. Hancock, F. Lawrence, J. Nevitt, Type I error and power of latent mean methods dogeneity in International Marketing applications of partial least squares struc-
and MANOVA in factorially invariant and noninvariant latent variable systems, tural equation modeling, J. Int. Mark. 26 (3) (2018) 1–21.
Struct. Equ. Model. 7 (4) (2000) 534–556. [56] C. Ringle, M. Sarstedt, Gain more insight from your PLS-SEM results: the im-
[22] J. Henseler, T. Dijkstra, M. Sarstedt, C. Ringle, A. Diamantopoulos, D. Straub, portance-performance map analysis, Ind. Manage. Data Syst. 116 (9) (2016)
D. Ketchen, F. Hair, G. Hult, R. Calatone, Common beliefs and reality about PLS: 1865–1886.
comments on Rönkkö and Evermann (2013), Organ. Res. Methods 17 (2) (2014) [57] G. Khan, M. Sarstedt, W. Shiau, J. Hair, C. Ringle, M, M. Fritze, Methodological
182–209. research on partial least squares structural equation modeling (PLS-SEM): An
[23] G. Shmueli, S. Ray, J. Estrada, S. Chatla, The elephant in the room: predictive analysis based on social network approaches,#xp#, Internet Res. (2019) 1–40.
performance on PLS models, J. Bus. Res. 69 (10) (2016) 4552–4564. [58] E. Rigdon, M. Sarstedt, C. Ringle, On comparing results from CB-SEM and PLS-
[24] J. Henseler, Partial least squares path modeling: quo vadis? Qual. Quant. 52 (1) SEM: five perspectives and five recommendations, Mark. Zfp 39 (3) (2017) 4–16.
(2018) 1–8. [59] K. Bollen, Evaluating effect, composite, and causal indicators in structural equa-
[25] D. Goodhue, W. Lewis, R. Thompson, Does PLS have advantages for small sample tion models,”, Mis Q. (35:2) (2011) 359–372.
size or non-normal data? Mis Q. 36 (3) (2012) 981–1001. [60] R. Bagozzi, Y. Yi, Specification, evaluation, and interpretation of structural
[26] E. Rigdon, Choosing PLS path modeling as analytical method in European equation modeling, J. Acad. Mark. Sci. 40 (1) (2012) 8–34.
Management research: a realist perspective, Eur. Manage. J. 34 (6) (2016) [61] M. Sarstedt, J. Hair, C. Ringle, K. Thiele, S. Gudergan, Estimation issues with PLS
598–605. and CBSEM: where the bias lies!, J. Bus. Res. 69 (10) (2016) 3998–4010.
[27] J. Benitez, J. Henseler, J. Roldan, How to address endogeneity in partial least [62] D. Borsboom, G. Mellenbergh, J. Van Heerden, The theoretical status of latent
squares path modeling, Proceedings of the 22th Americas Conference on variables, Psychol. Rev. 110 (2) (2003) 203–219.
Information Systems (2016) 1–10. [63] J. Henseler, Is the Whole More Than the Sum of Its Parts? On the Interplay of
[28] T. Dijkstra, A perfect match between a model and a mode, in: H. Latan, R. Noonan Marketing and Design Research, University of Twente, Enschede, The Netherlands,
(Eds.), Partial Least Squares Path Modeling: Basic Concepts, Methodological Issues 2015.
and Applications, Springer, Cham, Switzerland, 2017, pp. 55–80. [64] K. Bollen, R. Hoyle, Latent variables in structural equation modeling, in: R. Hoyle
[29] F. Schuberth, J. Henseler, T. Dijkstra, Partial least squares path modeling using (Ed.), Handbook of Structural Equation Modeling, Guilford, New York, USA, 2012,
ordinal categorical indicators, Qual. Quant. 52 (1) (2018) 9–35. pp. 56–67.
[30] D. Goodhue, W. Lewis, R. Thompson, A multicollinearity and measurement error [65] C. Spearman, General intelligence, objectively determined and measured, Am. J.
statistical blind spot: correcting for excessive false positives in regression and PLS, Psychol. 15 (2) (1904) 201–292.
Mis Q. 41 (3) (2017) 667–684. [66] K. Jöreskog, A general approach to confirmatory maximum likelihood factor
[31] T. Dijkstra, J. Henseler, Linear indices in nonlinear structural equation models: analysis, Psychometrika 34 (2) (1969) 183–202.
best fitting proper indices and other composites, Qual. Quant. 45 (6) (2011) [67] P. Lazarsfeld, Latent structure analysis, Psychol.: Study Sci. 3 (1) (1959) 476–543.
1505–1518. [68] A. Diamantopoulos, Formative indicators: introduction to the special issue, J. Bus.
[32] E. Rigdon, Rethinking partial least squares path modeling: in praise of simple Res. 61 (12) (2008) 1201–1202.
methods, Long Range Plann. 45 (5/6) (2012) 341–358. [69] K. Bollen, A. Diamantopoulos, In defense of causal-formative indicators: a min-
[33] L. Guttman, A basis for analyzing test-retest reliability, Psychometrika 10 (4) ority report, Psychol. Methods 22 (3) (2017) 581–596.
(1945) 255–282. [70] M. Aguirre-Urreta, M. Rönkkö, G. Marakas, Omission of causal indicators: con-
[34] K. Sijtsma, On the use, the misuse, and the very limited usefulness of Cronbach’s sequences and implications for measurement, Meas. Interdiscip. Res. Perspect. 14
alpha, Psychometrika 74 (1) (2009) 107–120. (3) (2016) 75–97.
[35] J. Henseler, C. Ringle, M. Sarstedt, A new criterion for assessing discriminant [71] K. Bollen, R. Lennox, Conventional wisdom on measurement: a structural equation
validity in variance-based structural equation modeling, J. Acad. Mark. Sci. 43 (1) perspective, Psychol. Bull. 110 (2) (1991) 305–314.
(2015) 115–135. [72] M. Rönkkö, C. McIntosh, J. Antonakis, On the adoption of partial least squares in
[36] C. Voorhees, M. Brady, R. Calantone, E. Ramirez, Discriminant validity testing in Psychological research: caveat emptor, Pers. Individ. Dif. 87 (1) (2015) 76–84.

14
J. Benitez, et al. Information & Management 57 (2020) 103168

[73] F. Davis, Perceived usefulness, perceived ease of use, and user acceptance of in- J. Mark. Res. 16 (1) (1979) 64–73.
formation technology, Mis Q. 13 (3) (1989) 319–340. [107] J. Cohen, A power primer, Psychol. Bull. 112 (1) (1992) 155–159.
[74] V. Venkatesh, S. Brown, L. Maruping, H. Bala, Predicting different con- [108] J. Cohen, Statistical Power Analysis for Behavioral Sciences (2nd Edition),
ceptualizations of system use: the competing roles of behavioral intention, facil- Erlbaum, Hillsdale, USA, 1988.
itating conditions, and behavioral expectation, Mis Q. 32 (3) (2008) 483–502. [109] N. Kock, P. Hadaya, Minimum sample size estimation in PLS-SEM: the inverse
[75] H. Barki, R. Titah, C. Boffo, Information system use-related activity: an expanded square root and gamma-exponential methods, Inf. Syst. J. 28 (1) (2018) 227–261.
behavioral conceptualization of individual-level information system use, Inf. Syst. [110] M. Aguirre-Urreta, M. Rönkkö, Sample size determination and statistical power
Res. 18 (2) (2007) 173–192. analysis in PLS using R: an annotated tutorial, Commun. Assoc. Inf. Syst. 36 (3)
[76] J. Benitez, J. Llorens, J. Braojos, How information technology influences oppor- (2015) 33–51.
tunity exploration and exploitation firm’s capabilities, Inf. Manage. 55 (4) (2018) [111] W. Chin, How to write up and report PLS analyses, in: V. Esposito, W. Chin,
508–523. J. Henseler, H. Wang (Eds.), Handbook of Partial Least Squares: Concepts,
[77] J. Benitez, A. Castillo, J. Llorens, J. Braojos, IT-enabled knowledge ambidexterity Methods and Applications, Springer, Berlin, Germany, 2010, pp. 655–690.
and innovation performance: the moderator role of social media capability, Inf. [112] W. Chin, The partial least squares approach for structural equation modelling, in:
Manage. 55 (1) (2018) 131–143. G. Marcoulides (Ed.), Modern Methods for Business Research, Lawrence Erlbaum
[78] J. Benitez, Y. Chen, T. Teo, A. Ajamieh, Evolution of the impact of e-business Associates, London, UK, 1998, pp. 295–336.
technology on operational competence and firm profitability: a panel data in- [113] C. Ringle, S. Wende, J. Becker, SmartPLS 3, SmartPLS, Boenningstedt, Germany,
vestigation, Inf. Manage. 55 (1) (2018) 120–130. 2015www.smartpls.com.
[79] J. Benitez, R. Walczuch, Information technology, the organizational capability of [114] N. Kock, WarpPLS 3.0 User Manual, ScriptWarp Systems, Laredo, USA, 2012.
proactive corporate environmental strategy and firm performance: a resource- [115] Addinsoft, Xlstat 2009, Addinsoft, Paris, France, 2009http://www.xlstat.com/en/
based analysis, Eur. J. Inf. Syst. 21 (6) (2012) 664–679. products/xlstat-plspm/.
[80] Y. Chen, Y. Wang, S. Nevo, J. Benitez, G. Kou, IT capabilities and product in- [116] J. Henseler, ADANCO 2.0.1. Professional for Windows, Composite Modeling
novation performance: the roles of corporate entrepreneurship and competitive GmbH & Co., Kleve, Germany, 2017www.compositemodeling.com.
intensity, Inf. Manage. 52 (6) (2015) 643–657. [117] J. Henseler, Partial least squares path modeling, in: P. Leeflang, J. Wieringa,
[81] A. Ajamieh, J. Benitez, J. Braojos, C. Gelhard, IT infrastructure and competitive T. Bijmolt, K. Pauwels (Eds.), Advanced Methods for Modeling Markets, Springer,
aggressiveness in explaining and predicting performance, J. Bus. Res. 69 (10) Heidelberg, Germany, 2017, pp. 361–381.
(2016) 1–31. [118] D. Gefen, E. Rigdon, D. Straub, An update and extension to SEM guidelines for
[82] Y. Chen, Y. Wang, S. Nevo, J. Benitez, G. Kou, Improving strategic flexibility with Administrative and Social Science research, Mis Q. 35 (2) (2011) iii–xiv.
information technologies: insights for firm performance in an emerging economy, [119] L. Hu, P. Bentler, Y. Kano, Can test statistics in covariance structure analysis be
J. Inf. Technol. 32 (1) (2017) 10–25. trusted? Psychol. Bull. 112 (2) (1992) 351–362.
[83] J. Benitez, G. Ray, J. Henseler, Impact of information technology infrastructure [120] J. Nunnally, I. Bernstein, Psychometric Theory, McGraw-Hill, New York, USA,
flexibility on mergers and acquisitions, Mis Q. 42 (1) (2018) 25–43. 1994.
[84] N. Melville, K. Kraemer, V. Gurbaxani, Information technology and organizational [121] C. Fornell, J. Cha, Partial least squares, Adv. Methods Market. Res. 407 (1) (1994)
performance: an integrative model of IT business value, Mis Q. 28 (2) (2004) 52–78.
283–322. [122] R. Cenfetelli, G. Bassellier, Interpretation of formative measurement in
[85] A. Bharadwaj, A resource-based perspective on information technology and firm Information Systems research, Mis Q. 33 (4) (2009) 689–707.
performance: an empirical investigation, Mis Q. 24 (1) (2000) 169–196. [123] J. Hair, M. Sarstedt, L. Hopkins, V. Kuppelwieser, Partial least squares structural
[86] R. Santhanam, E. Hartono, Issues in linking information technology capability to equation modeling (PLS-SEM): an emerging tool in Business research, Eur. Bus.
firm performance, Mis Q. 27 (1) (2003) 125–153. Rev. 26 (2) (2014) 106–121.
[87] J. Wooldridge, Introductory Econometrics: a Modern Approach, 5th ed., Cengage [124] L. Hayduk, Shame for disrespecting evidence: the personal consequences of in-
Learning, Mason, USA, 2013. sufficient respect for structural equation model testing, BMC Med. Res. Methodol.
[88] J. Lohmöller, Predictive vs. Structural modeling: PLS vs. ML, Latent Variable Path 14 (124) (2014) 1–10.
Modeling With Partial Least Squares, Physica-Verlag, Hidelberg, Germany, 1989, [125] M. Aguirre-Urreta, M. Rönkkö, Statistical inference with PLSc using bootstrap
pp. 199–226. confidence intervals, Mis Q. 42 (3) (2018) 1001–1020.
[89] J. Hair, G. Hult, C. Ringle, M. Sarstedt, K. Thiele, Mirror, mirror on the wall: a [126] X. Zhao, J. Lynch, Q. Chen, Reconsidering Baron and Kenny: myths and truths
comparative evaluation of composite-based structural equation modeling about mediation analysis, J. Consum. Res. 37 (2) (2010) 197–206.
methods, J. Acad. Mark. Sci. 45 (5) (2017) 616–632. [127] J. Becker, A. Rai, E. Rigdon, Predictive validity and formative measurement in
[90] H. Hwang, Y. Takane, Generalized structure component analysis, Psychometrika structural equation modeling: embracing practical relevance, Proceedings of the
69 (1) (2004) 81–99. 34th International Conference on Information Systems (2013) 1–19.
[91] W. Reinartz, M. Haenlein, J. Henseler, An empirical comparison of the efficacy of [128] J. Braojos, J. Benitez, J. Llorens, How do small firms learn to develop a social
covariance-based and variance-based SEM, Int. J. Res. Mark. 26 (4) (2009) media competence? Int. J. Inf. Manage. 35 (4) (2015) 443–458.
332–344. [129] J. Karimi, T. Somers, A. Bhattacherjee, The impact of ERP implementation on
[92] J. Henseler, Why generalized structured component analysis is not universally business process outcomes: a factor-based study, J. Manage. Inf. Syst. 24 (1)
preferable to structural equation modeling, J. Acad. Mark. Sci. 40 (3) (2012) (2007) 101–134.
402–413. [130] J. Karimi, T. Somers, A. Bhattacherjee, The role of information systems resources
[93] Y. Takane, H. Hwang, Comparisons among several consistent estimators of in ERP capability building and business process outcomes, J. Manage. Inf. Syst. 24
structural equation models, Behaviormetrika 45 (1) (2018) 157–188. (2) (2007) 221–260.
[94] A. Diamantopoulos, P. Riefler, K. Roth, Advancing formative measurement [131] J. Evermann, M. Tate, Assessing the predictive performance of structural equation
models, J. Bus. Res. 61 (12) (2008) 1203–1218. model estimators, J. Bus. Res. 69 (10) (2016) 4565–4582.
[95] R. Beran, M. Srivastava, Bootstrap tests and confidence regions for functions of a [132] G. Cepeda, J. Henseler, C. Ringle, J. Roldan, Prediction-oriented modeling in
covariance matrix, Ann. Stat. 13 (1) (1985) 95–115. Business research by means of PLS path modeling, J. Bus. Res. 69 (10) (2016)
[96] C. Fornell, D. Larcker, Evaluating structural equation models with unobservable 4545–4551.
variables and measurement error, J. Mark. Res. 18 (1) (1981) 39–50. [133] J. Cheah, M. Sarstedt, C. Ringle, T. Ramayah, H. Ting, Convergent validity as-
[97] J. Braojos, J. Benitez, J. Llorens, How do social commerce-IT capabilities influence sessment of formatively measured constructs in PLS-SEM: On using single-item
organizational performance? Theory and empirical evidence, Inf. Manage. 56 (2) versus multi-item measures in redundancy analyses, Int. J. Contemp Hosp.
(2019) 155–171. Manage. 30 (11) (2019) 3192–3210.
[98] P. Tallon, A. Pinsonneault, Competing perspectives on the link between strategic [134] F. Schuberth, J. Henseler, T. Dijkstra, Confirmatory composite analysis, Frontiers
information technology alignment and organizational agility: insights from a in Psychology 9 (1) (2018) 1–14.
mediation model, Mis Q. 35 (2) (2011) 463–486.
[99] S. Aral, C. Dellarocas, D. Godes, Social media and business transformation: a Jose Benitez is Full Professor of IS at Rennes School of Business, France. Jose also holds
framework for research, Inf. Syst. Res. 54 (1) (2013) 3–13. the COVIRAN-Prodware Chair of Digital Human Resource Strategy at the University of
[100] P. Sharma, M. Sarstedt, G. Shmueli, K. Kim, K. Thiele, PLS-based model selection: Granada, Spain. Jose is also Instructor of PLS-PM at the PLS School. His research interests
The role of alternative explanations in IS Research,”, J. Assoc. Inf Syst. 20 (4) cover the study of how the firm’s portfolio of IT capabilities affects organizational cap-
(2019) 346–397. abilities and firm performance, and the development of PLS-PM in the field of IS. His
[101] S. Gregor, The nature of theory in information systems, Mis Q. 30 (3) (2006) research has been published in top IS journals such as MIS Quarterly, Information &
611–642. Management, European Journal of Information Systems, Journal of Information Technology,
[102] H. Simon, Science seeks parsimony, not simplicity: searching for pattern in phe- and Journal of Business Research. He currently serves as Associate Editor for Information &
nomena, in: A. Zellner, Keuzenkamp H, M. McAleer (Eds.), Simplicity, Inference Management, European Journal of Information Systems, and Decision Support Systems, and as
and Modelling: Keeping It Sophisticatedly Simple, Cambridge University Press, Guest Editor for Decision Sciences. He is also a member of the Editorial Review Board of
Cambridge, UK, 2001, pp. 32–72. the Journal of the Association for Information Systems. Jose obtained a PhD in Business
[103] M. Friedman, The methodology of positive economics, Philos. Econ.: Anthol. 2 (1) Administration (with concentration in IS) from the University of Granada, Spain. Jose can
(1994) 180–213. be contacted at jose.benitez@rennes-sb.com and joseba@ugr.es.
[104] A. Rai, R. Patnayakuni, N. Seth, Firm performance impacts of digitally enabled
supply chain integration capabilities, Mis Q. 30 (2) (2006) 225–246.
[105] N. Saraf, C. Schlueter, S. Gosain, IS application capabilities and relational value in Jörg Henseler holds the Chair of Product-Market Relations in the Faculty of Engineering
Technology at the University of Twente, the Netherlands, and he is Visiting Professor at
interfirm partnerships, Inf. Syst. Res. 18 (3) (2007) 320–339.
[106] J. Churchill, A paradigm for developing better measures of Marketing constructs, NOVA Information Management School, Universidade Nova de Lisboa, Portugal. His

15
J. Benitez, et al. Information & Management 57 (2020) 103168

broad-ranging research interests encompass empirical methods of Marketing and Design media capabilities to pursue knowledge management and innovation activities. She holds
research as well as the management of design, products, services, and brands. His work a master’s degree in Management from the University of Granada. Her research has been
has been published in Computational Statistics and Data Analysis, European Journal of published in Information & Management. She has presented her research in the Academy of
Information Systems, International Journal of Research in Marketing, Journal of the Academy Management Annual Meeting, European Conference on Information Systems, Americas
of Marketing Science, Journal of Supply Chain Management, MIS Quarterly, Organizational Conference on Information Systems, and Pacific Asia Conference on Information Systems.
Research Methods, and Structural Equation Modeling-An Interdisciplinary Journal, among
others. He chairs the Scientific Advisory Board of ADANCO, software for variance-based Florian Schuberth obtained his PhD in Econometrics in the Faculty of Business
SEM (www.composite-modeling.com). Management and Economics at the University of Wuerzburg, Germany. Currently, he is
Assistant Professor in the Faculty of Engineering Technology at the University of Twente,
Ana Castillo is a PhD Student of IS in the Department of Management at the University of the Netherlands. His main research interests are focused on SEM, in particular on com-
Granada, Spain. In her doctoral dissertation, she examines how firms leverage social posite-based estimators and their enhancement.

16