Journal of Family Business Strategy 5 (2014) 116128

Contents lists available at ScienceDirect

Journal of Family Business Strategy

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

A comparative study of CB-SEM and PLS-SEM for theory development

in family rm research
Claudia Binz Astrachan a, Vijay K. Patel b, Gabrielle Wanzenried c,*
a

Witten/Herdecke University, Germany

Kennesaw State University, USA
c
Lucerne University of Applied Sciences and Arts, Switzerland
b

A R T I C L E I N F O

A B S T R A C T

Keywords:
Structural equation modeling (SEM)
Covariance-based SEM
Partial least squares-SEM
Family rms
Organizational reputation
Organizational trustworthiness

Structural equation modeling (SEM) has become the methodology of choice for many family business
researchers investigating complex relationships between latent constructs, such as family harmony or
family cohesion. Its capability to evaluate complex measurement models and structural paths involving
a multitude of variables and levels of constructs has enabled family business researchers to investigate
complex and intricate relationships that previously could not be easily untangled and examined. In many
cases, however, researchers struggle to meet some of the challenging requirements of covariance-based
SEM (CB-SEM), the most commonly used approach to SEM, such as distribution assumptions or sample
size. In this article, we point out the benets and disadvantages of CB-SEM, and present a comparison
with partial least squares-SEM (PLS-SEM) using an identical sample. We nd that even though both
methods analyze measurement theory and structural path models, there are many advantages in
applying PLS-SEM.
2014 Elsevier Ltd. All rights reserved.

1. Structural equation modeling in a nutshell

Structural equation modeling (SEM) has seen a dramatic rise in
attention and utilization across a variety of scientic disciplines
such as strategic management (Shook, Ketchen, Cycyota, &
Crockett, 2003), marketing (Chin, Peterson, & Brown, 2008) and
psychology (MacCallum & Austin, 2000) over the last decade (Hair,
Ringle, & Sarstedt, 2011b). Statistically, SEM represents an
advanced version of general linear modeling procedures (e.g.,
multiple regression analysis), and is used to assess whether a
hypothesized model is consistent with the data collected to reect [the]
theory (Lei & Wu, 2007, p. 34). While SEM is a general term
encompassing a variety of statistical models, covariance-based
SEM (CB-SEM) is the more widely used approach in SEM, and many
researchers simply refer to CB-SEM as SEM. This reference is nave,
however, because partial least squares (PLS) is also a useful and
increasingly applied approach to examine structural equation
models (Hair, Sarstedt, Ringle, & Mena, 2012).
Structural equation modeling is a multivariate analytical
approach used to simultaneously test and estimate complex
causal relationships among variables, even when the relationships
are hypothetical, or not directly observable (Williams, Vandenberg,

* Corresponding author. Tel.: +1 404 242 0803.

E-mail address: claudia.astrachan@gmail.com (G. Wanzenried).
1877-8585/$ see front matter 2014 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jfbs.2013.12.002

& Edwards, 2009). Concurrently combining factor analysis and

linear regression models, SEM allows the researcher to statistically
examine the relationships between theory-based latent variables
and their indicator variables by measuring directly observable
indicator variables (Hair, Hult, Ringle, & Sarstedt, 2014). While SEM
is similar to multiple regression in the sense that both techniques
test relationships between variables, SEM is able to simultaneously
examine multi-level dependence relationships, where a dependent
variable becomes an independent variable in subsequent relationships
within the same analysis (Shook, Ketchen, Hult, & Kacmar, 2004, p.
397) as well as relationships between multiple dependent
variables (Joreskog, Sorbom, du Toit, & du Toit, 1999).
The objective of this article is to evaluate the benets and
limitations of SEM in general, and in family business research in
particular, by directly comparing two major approaches to
structural modeling covariance based SEM (CB-SEM) and
variance-based SEM (PLS-SEM) (Sarstedt, Ringle, Smith, Reams, &
Hair, 2014; Sharma & Kim, 2013). While CB-SEM and PLS-SEM are
two different approaches to the same problem namely, the analysis
of causeeffect relations between latent constructs (Hair, Ringle, &
Sarstedt, 2011a, p. 139), they differ not only in terms of their basic
assumptions and outcomes, but also in terms of their estimation
procedures (Hair et al., 2014; Shook et al., 2004). PLS-SEM uses a
regression-based ordinary least squares (OLS) estimation method
with the goal of explaining the latent constructs variance by
minimizing the error terms [and maximizing] the R2 values of the

C.B. Astrachan et al. / Journal of Family Business Strategy 5 (2014) 116128

(target) endogenous constructs (Hair et al., 2014, p. 14; Ringle,

Sarstedt, Hair, & Pieper, 2012). CB-SEM, on the other hand, follows a
maximum likelihood (ML) estimation procedure and aims at
reproducing the covariance matrix [i.e., minimizing the difference
between the observed and estimated covariance matrix], without
focusing on explained variance (Hair et al., 2011a, p. 139). In other
words, with CB-SEM, the R2 is a by-product of the overall statistical
objective of achieving good model t (Hair et al., 2014).
Using a sample of 253 Swiss consumers surveyed in 2012
evaluating the effects of corporate expectations on the perceived
level of expertise and trustworthiness of family-owned companies,
we apply both CB-SEM and PLS-SEM to analyze the data. This
approach enables us to not only compare the requirements of each
method, the way in which the models are specied, and the
applicability and user-friendliness of available software, but also
the results and interpretations.
The remainder of this article is structured as follows: rst, we
briey highlight the most important benets of SEM. We then
summarize the results of several important articles in family
business research that utilized SEM, and point out how SEM
contributed to the ndings of these studies. Third, the research
context of the example used in this study is briey described, and
the hypotheses as well as an outline of the methodology are
presented. Fourth, we discuss the results from the CB-SEM and PLSSEM analyses. Finally, practical observations and conclusions are
provided, and limitations and suggestions for further research are
presented.
2. The benets and limitations of SEM
2.1. The benets of SEM
The question of why researchers might want to use SEM is quite
simple. The process of applying SEM enables researchers to more
effectively evaluate measurement models and structural paths,
particularly when the structural model involves multiple dependent variables, latent constructs based on multi-item indicator
variables, and multiple stages/levels of constructs in a structural
model. While there are many reasons to use SEM in social sciences
research, we consider the following to be the most relevant.
When dealing with latent constructs and complex models: Many
constructs investigated in the social sciences are latent constructs
that cannot be observed, or measured directly. Examples include
family inuence and family cohesion. Moreover, especially at the
theory development and testing stages there may be multiple
constructs and interactive effects resulting in a complex model.
While a latent construct may be measurable to some extent by
means of a directly observable indicator variable (e.g., degree of
family ownership, number of family members in management),
these indicator measures may not reect the latent variable
entirely accurately, which means the measurement will contain
error as will the results. By explicitly assessing error in the
structural model, SEM provides a powerful means of simultaneously
assessing the quality of measurement and examining causal relationships among constructs (Wang & Wang, 2012, p. 1). So while
multiple regression analysis assumes there is no error in the data,
SEM recognizes and accounts for the error in each measured item
in an effort to improve the accuracy of ndings. Additionally, the
SEM approach is designed to consider interactive effects and
complex models to nd an optimal model that reduces crossloadings and identies the higher loadings for relevant measures.
When analyzing direct, indirect, and total effects: SEM facilitates
the assessment of direct, indirect and total effects. Direct effects
include relationships between independent and dependent variables, e.g., family ownership has a direct positive effect on rm
performance. Indirect effects involve relationships between

117

independent and dependent variables that are mediated or

moderated by some other variable, e.g., the effect of family
ownership on rm performance is moderated by the owning
familys involvement in management. Total effects relate to the
sum of two or more direct or indirect effects. In comparison to
other statistical procedures such as regression, SEM enables
researchers to not only simultaneously assess the relationships
between multi-item constructs, but also to reduce the overall error
associated with the model. In contrast to multiple regression
analysis, which cannot directly deal with the measurement issues
of multi-item constructs, SEM is specically designed to improve
multi-item measurement models by directly accounting for error.
When assessing structural models: While regression also allows
researchers to evaluate structural relationships using path analysis
(examining each path separately), SEM facilitates simultaneous
analysis of all structural relationships (i.e., relationships or paths
among numerous variables, e.g., family ownership, family cohesion and performance), and is an inherently simpler approach that
leads to more accurate results. CB-SEM and PLS-SEM use different
approaches when assessing the quality of a structural model. For
example, with CB-SEM t is based on accurately estimating the
observed covariance matrix, while with PLS-SEM t is based upon
accounting for explained variance in the endogenous constructs
(Hair et al., 2014). As a result of model t requirements, however,
CB-SEM often eliminates relevant indicator variables, thereby
reducing the validity of constructs. In contrast, PLS-SEM creates
composite constructs that generally include additional theorybased indicator variables (Rigdon, 2012), while still optimizing
predictive accuracy and relevance. Also, PLS-SEM analyses can
easily incorporate single-item measures, and can obtain solutions
to much more highly complex models, i.e., models with a large
number of constructs, indicators and structural relationships (Hair
et al., 2014; Ringle, Sarstedt, & Hair, 2013).
2.2. The limitations of SEM
The fact that modern SEM software (such as AMOS, LISREL and
SmartPLS) does not require profound statistical knowledge has
made investigation of complex statistical problems accessible to
non-statisticians (Babin, Hair, & Boles, 2008; Hair, Black, Babin, &
Anderson, 2010). Yet, while ease of access to SEM has increased the
number of meaningful and valuable contributions, recent reviews
of SEM applications provide grounds for criticism of methodological aws and shortcomings in the execution of SEM in many
contributions (e.g., Hair et al., 2012; Williams et al., 2009). Being a
highly sophisticated statistical tool, insight and judgment are
crucial elements of its use (Shook et al., 2004, p. 397). Thus, to
obtain meaningful and valid results it is essential to understand
when it is appropriate to use SEM, its requirements and
interpretation, and also the potential trade-offs when compared
to other methods.
When unable to correctly identify a research model: In the case of
CB-SEM in particular, since it is a conrmatory approach, the
method requires the specication of the full theoretical model
prior to data analysis. The researcher(s) must therefore dene the
exact number of dependent (endogenous) and independent
(exogenous) variables used in the theoretical model, the relationships between these latent variables, the type of measurement
model (formative or reective), and the number of indicator
variables required to ensure a valid and reliable measure of all
constructs (e.g., Williams et al., 2009). Only when a model is
correctly specied can all parameters be estimated (Lei & Wu,
2007). Thus, if the model lacks a sound theoretical foundation, and
if the direction of the relationship between variables cannot be
determined, CB-SEM should not be the method of choice. In
contrast, PLS-SEM, which is particularly suitable for early-stage

118

C.B. Astrachan et al. / Journal of Family Business Strategy 5 (2014) 116128

theory development and testing (Hair et al., 2014; Ringle et al.,

2013), permits examination of constructs and relationships in
complex structural models. Since the primary purpose in theory
development is to nd relationships, their directions and
strengths, as well as observable measures, PLS-SEM is appropriate.
The model t constraints of CB-SEM are more appropriate for
established theory testing and conrmation, but require a
substantially larger sample size, which may not be available in
general, and particularly at the early stages of theory development
in the context of family research.
When experiencing data collection constraints: Recommendations regarding the ideal sample size for SEM analysis range from
50 to 200 observations (e.g., Anderson & Gerbing, 1988; Kline,
2005). The appropriate sample size for SEM models depends rst
on the method used. Specically, CB-SEM requires larger samples
than PLS-SEM because relationships between all variables must be
assessed (i.e., a full information approach), while with PLS-SEM the
model is separated into different smaller components (a component for each construct in the model; hence the name partial least
squares). In comparison with CB-SEM, which imposes rigid sample
size restrictions on the researcher(s), PLS-SEM works efciently
with small sample sizes and complex models and makes practically no
assumptions about the underlying data [distributions] (Hair et al.,
2014; Ringle et al., 2013). This makes PLS-SEM particularly suitable
for family business research, where researchers often experience
data collection constraints and struggle with low response rates. In
PLS-SEM, the guideline is that sample size should be ten times the
number of arrows pointing at a construct (Hair et al., 2014). In
contrast, CB-SEM requires a sample size of ve times the number of
indicators included in the original model (e.g., a CB-SEM model
with 40 indicator variables on three constructs requires a sample
size of 200 (5  40), but if those 40 indicators are associated with
the same three constructs and two exogenous constructs are
predicting a single endogenous construct, then the required
sample size with PLS-SEM is 20 (2  10); i.e., arrows pointing
from the two exogenous constructs to the one endogenous
construct).
When data are not normally distributed: The CB-SEM maximum
likelihood approach, like many other multivariate statistical
methods, requires multivariate normality. In contrast, PLS-SEM
does not require normally distributed data (Hair et al., 2014), and is
therefore the more appropriate method of SEM for many social
science studies, including family business, where data are often
non-normally distributed (e.g., distribution of ownership among
US companies; Astrachan & Shanker, 2003). Moreover, when data
are categorical or ordinal (quasi-metric), or includes single item
measures, PLS-SEM can be used (Hair et al., 2014).
In sum, SEM approaches offer a range of unique benets, as
compared with rst generation statistical procedures. There may
be situations, however, where a simpler approach like regression
analysis might be adequate, or when investigating a simple model
involving two-stage (single path) models. However, regression
analysis does not directly permit assessment of measurement
characteristics so latent constructs must rst be converted to some
composite or average of individual measures, such as factor scores
from an EFA or summated scores. SEM based models inherently
include evaluation of individual measures and retention of
relevant indicators at appropriate loading levels, e.g., at a level
of .70 or higher (Hair et al., 2010).
3. SEM in family business research
Despite the fact that SEM is an increasingly popular approach in
business research and related social sciences, family rm
researchers have used the method sparingly (Wilson et al.,
2014). Several family business researchers have called for more

sophisticated and rigorous statistical analysis techniques, such as

SEM (e.g., Debicki, Matherne, Kellermanns, & Chrisman, 2009; Dyer
& Dyer, 2009; Westhead & Howorth, 2006). One assessment of
empirical articles published in family business research revealed
that only 13 empirical studies investigating family businesses
published between 1989 and 2013 used SEM methodologies, seven
of which (from a total of 183 empirical articles) were published in
Family Business Review. Interestingly, a broader EBSCO database
search using the keywords family business and structural
equation modeling resulted in considerably larger numbers
however, many of these contributions only point out in their
discussion or contribution sections that using a SEM approach
would provide additional insights, and that further research should
look into applying these methodologies. The low number of actual
applications using SEM methods mostly CB-SEM based is a
particularly unfortunate shortcoming given the possibilities these
methods offer to family business research, as some of the widely
cited examples presented below illustrate (Wilson et al., 2014).
Aspects related to causality: Mustakallio, Autio, and Zahra
(2002), analyzing a sample of 192 Finnish family rms, explored
the effects of both contractual (formal control) and relational
(social control) governance systems on strategic decision quality
and commitment. Using CB-SEM, the authors evaluated the t of
the overall measurement model as well as the strength and
signicance of the relationships (or paths) between the exogenous
and endogenous variables. For instance, the results suggest that
family size has a negative effect on the degree of social interaction
within the family, i.e., the larger the family the fewer family
members interact with each other. Moreover, while the relationship between board monitoring and the board commitment to
strategic decisions was hypothesized to be positive the results did
not conrm the relationship. In sum, SEM can shed light on the
theoretical causality of relationships between latent and observable variables, and can help researchers decide whether to accept
or reject hypothesized relationships.
Theory testing and scale development: In addition to analyzing
relationships between multiple variables or constructs, SEM is
particularly useful for testing theoretical models with nonexperimental data (Bagozzi, 1980). Astrachan, Klein, and Smyrnios
(2002; see also Klein, Astrachan, & Smyrnios, 2005) used CB-SEM
when developing their continuous (rather than dichotomous) FPEC scale of family inuence, which today is one of the few widely
accepted, measureable, and validated conceptualizations of family-owned business (see also Holt, Rutherford, & Kuratko, 2010).
The F-PEC scale is an index of family inuence, measured by three
dimensions (power, experience, culture), which include nine
subscales with 23 corresponding indicator variables (Power: 4;
Experience: 6; Culture: 13). The authors used CB-SEM when
developing the original scale to conrm the theoretically developed
model with data (Klein et al., 2005, p. 327). In comparison to other
statistical procedures, SEM models enable researcher(s) to
evaluate complex models with regard to their compatibility (t)
with all the relationships (covariances) in the data set. By
calculating a range of goodness-of-t statistics, CB-SEM can assess
whether the theoretical model is conrmed. While scale development is possible based on exploratory factor analysis (EFA), an
inherent advantage of SEM is that it includes Conrmatory Factor
Analysis (CFA), which is considered a superior approach to scale
development (Hair et al., 2010). SEM based modeling enables more
precise evaluation of indicator variable loadings as well as
reliability and validity of measurement models.
Inclusion of mediating/moderating effects: SEM approaches are
particularly useful when examining mediating and moderating
effects (Hair et al., 2010). Using a sample of 163 Swiss companies,
Memili, Eddleston, Kellermanns, Zellweger, and Barnett (2010)
investigate the mediating effects of entrepreneurial risk taking

C.B. Astrachan et al. / Journal of Family Business Strategy 5 (2014) 116128

(willingness to undertake high risks) and family rm image

(promotion of the rms family background) on the relationships
between the degree of family ownership and the owning familys
identication with the rm (independent variables) and rm
performance (dependent variable). Using a CB-SEM approach, and
comparing the t indices of both a fully and partially mediated
model enabled the authors to show that a fully mediated model t
our data best, showing, for example, that the possible relationship
between family expectations and family rm performance was fully
mediated by family rm image and risk taking (Memili et al., 2010,
p. 206).
Besides the three examples presented above, other interesting
possible applications in family business research might include the
examination of group differences such as differences between
family and non-family rms but also, and possibly more
important, within the family rm group cross-cultural comparisons (e.g., family rms in Germany vs. the United States), or the
investigation of differences between generations, for example in
terms of attitudes, values, or expectations. Despite the fact that
family rms are far from being a homogeneous group of
organizations, and numerous calls for within-group comparisons,
most studies thus far focus on the differences between family and
non-family companies. Given that family business researchers
often experience theory specication and data collection constraints, SEM approaches and in particular, PLS-SEM may be a
valuable tool for research in the family business context.
4. Research context
To illustrate how the applicability and the results of CB-SEM
and PLS-SEM compare, we applied both SEM approaches to the
same research context. In this example, we examine the
relationship between organizational reputation and corporate
credibility. Specically, we investigate whether two distinct
dimensions of reputation, namely social expectations and business
expectations, lead to organizational trust, i.e., the degree to which
individuals consider an organization to be trustworthy. Furthermore, we test if perceived expertise acts as a mediating factor.
Corporate credibility refers to the expertise and trustworthiness a potential customer attributes to an organization, or in other
words, the extent to which consumers feel that the rm has the
knowledge or ability to fulll its claims and whether the rm can be
trusted to tell the truth (Newell & Goldsmith, 2001, p. 235).
Corporate credibility has been shown to inuence customer
attitudes and ultimately purchase decisions and therefore rm
performance (Fombrun, 1996). Being viewed as credible (i.e., as
experienced and trustworthy), is therefore a source of competitive
advantage for companies. This phenomenon may be particularly
relevant in the context of family rms as this type of governance
structure has repeatedly been characterized by its ability to create
long-term, trust-based relationships (Tagiuri & Davis, 1996; Ward,
1997).
Recent research has shown a growing scholarly interest in the
areas of family rm reputation and branding, and ndings indicate
that stakeholders are likely to perceive family-owned businesses
differently, and possibly view them in a more positive light as
compared with publicly listed companies (e.g., Binz, Hair, Pieper, &
Baldauf, 2013; Carrigan & Buckley, 2008; Craig, Dibrell, & Davis,
2008; Kashmiri & Mahajan, 2010). Several studies have suggested
that a distinct family rm brand, which refers to the active
promotion of a rms family background (e.g., SC Johnson: A family
company), may lead to superior organizational reputation, and
that such distinct family rm reputation could be a unique asset
that family rms can leverage to obtain a competitive advantage
(e.g., Craig et al., 2008; Zellweger, Kellermanns, Eddleston, &
Memili, 2012). While the research is inconclusive as to what leads

119

to a superior reputation, it has been suggested that it may be the

owning familys dedication to the companys ongoing success and
survival that strengthens the rms reliability and increases
stakeholders trust in the organization (Dyer & Whetten, 2006;
Miller, McLeod, & Young, 2001; Tagiuri & Davis, 1996).
Another driver of stakeholder trust may be the continuity and
stability that a family rms long-term existence implies. The fact
that a company has been around for decades, implying that
knowledge and experience have been accumulated and transferred
across generations, creates a perception of expertise (Miller & Le
Breton-Miller, 2005; ODonnell, Carson, & Gilmore, 2002; Zahra,
Hayton, Neubaum, Dibrell, & Craig, 2008). Accordingly, familyowned companies are assumed to be viewed as more trustworthy
than publicly-owned rms, which might lead to higher levels of
customer satisfaction, loyalty and trust (e.g., Carrigan & Buckley,
2008; Dyer & Whetten, 2006; Memili et al., 2010; Orth & Green,
2009; Tagiuri & Davis, 1996; Ward, 1997).
For this study, we draw from and extend the ndings from a
previous study examining the effects of distinct family rm
reputation on customer preferences (Binz et al., 2013). In this
research, we use the adapted reputation construct with two
dimensions, which was identied in the previous study based on
exploratory factor analysis. The social expectations dimension
refers to how a company conducts its business, while the business
expectations dimension refers to what a company does in order to
be successful. As suggested by prior research, we assume that
satisfaction of customer expectations leads to trust, e.g., if a
company claims to have high quality standards, and customers are
satised with the quality of the products they purchased, the
company fullled their claim and is deemed trustworthy.
Furthermore, we assume that a high level of perceived expertise
(i.e., the company is skilled, has substantial experience, and broad
expertise) strengthens the relationship between an individuals
expectations (e.g., this company is a loyal employer, the company
develops innovative products and services) and the degree to which
they trust that an organization will fulll their claims.
The literature and its synthesis suggest the conceptual model
shown in Fig. 1 and the following hypotheses.
H1. Business expectations are positively related to organizational
expertise.
H2. Business expectations are positively related to organizational
trustworthiness.
H3. Social expectations are positively related to organizational
expertise.
H4. Social expectations are positively related to organizational
trustworthiness.

Fig. 1. Theoretical model and hypotheses.

120

C.B. Astrachan et al. / Journal of Family Business Strategy 5 (2014) 116128

H5. Organizational expertise is positively related to organizational

trustworthiness.

Table 1
Exploratory factor analysis.
Variables

5. Methodology
5.1. Measures
To investigate the relationship between distinct family rm
reputation and perceived trustworthiness of family businesses, a
standardized questionnaire was developed based on two established scales. One scale was the Fombrun, Gardberg, and Sever
(2000) Reputation Quotient Scale, which consists of six dimensions of corporate reputation, namely emotional appeal, products
and services, vision and leadership, workplace environment, social
and environmental responsibility, and nancial performance,
measured with 7-point Likert scales. The original wording of
the scale items was adapted by replacing the term organization
with family rm in each question to t the family business
context (see also Holt et al., 2010; Zellweger, Nason, & Nordqvist,
2012).
The second scale was Newell and Goldsmiths (2001) Corporate
Credibility Scale, a self-report scale designed to measure
corporate credibility or the amount of expertise and trustworthiness that consumers perceive in a corporation (p. 235). The scale
consists of two dimensions (4 items each), namely expertise and
trustworthiness, which were both assessed by 7-point-Likert
scales. In addition to the two scales described above, respondents
were asked to provide basic demographic information, including
whether they had previously worked in a family or non-family rm
(see Binz et al., 2013 for details).
5.2. Sample prole
An invitation to participate in the online survey on unipark.de
was sent to 480 potential respondents, all of which were personal
and professional acquaintances of 24 lecturers working at Lucerne
University of Applied Sciences in Switzerland. Two follow-up
emails were sent after 14 and 21 days, respectively, and 266
respondents followed the link and completed the questionnaire.
After eliminating respondents that failed to complete the
questionnaire, a total of 174 usable responses remained, representing an overall response rate of 36.25%. The sample size exceeds
the minimum required for the application of either CB-SEM or PLSSEM (Hair et al., 2014; Hair et al., 2010). The sample diversity was
satisfactory with 51% of all respondents being male. The average
age of the sample was 38 years. A test for non-response bias
(Armstrong & Overton, 1977) did not reveal signicant differences
between early and late respondents.
5.3. Initial measurement model evaluation
In the Binz et al. (2013) study, CFA was used to examine the
dimensionality, reliability and validity of the reputation constructs. When the CFA did not achieve acceptable t, and thus the
data did not reect the six dimensions proposed by Fombrun et al.
(2000), it was necessary to re-assess the theoretical foundation of
the scales. Subsequently, an EFA was executed, and after several
iterations and the removal of weaker items, an empirically
validated two-factor solution emerged (see Table 1). Based on a
qualitative assessment of the loadings, the new constructs were
named social expectations (SE, related to how a company does
business) and business expectations (BE, related to what a business
does in order to be successful), which differs slightly from the
original wording used in the previous study (see Appendix for list

I have a good feeling about family rms

I trust family rms
I admire and respect family rms
Family rms stand behind their products
and services
Family rms look like good companies
to work for
Family rms support good causes
Family rms are environmentally friendly
Family rms have high standards with
employees
Family rms develop innovative products
and services
Family rms offer high quality products
and services
Family rms offer good value for money
Family rms have a clear vision for
their future
Family rms recognize and take advantage
of market opp.
Family rms look like they have good
employees
Family rms have a strong record
of protability
Family rms tend to outperform
their competitors
Family rms have strong prospects
for future growth
Eigenvalue
% of variance
Total variance explained

Factor 1:
Social
expectations

Factor 2:
Business
expectations

.881
.884
.883
.747
.645
.705
.658
.812
.759
.557
.563
.741
.838
.549
.663
.580
.638

6.88
40.4

2.65
15.6
56.0

Note: N = 253. Varimax rotation. Factor loadings higher than .35 shown. Kaiser
MeyerOlkin measure of sampling adequacy = .917. The KMO measures the
sampling adequacy, which should be greater than .5 for a satisfactory factor
analysis to proceed (Hair et al., 2010).

of questions). The SE and BE constructs along with the Expertise

and Trust constructs were then used to run the CB-SEM and PLSSEM analyses.
6. Results from the SEM analyses
In this section we discuss the results from applying the CB-SEM
and PLS-SEM methods separately to examine the theoretical model
and hypotheses. We present an overview of our approach and
ndings as well as comparative results. We also discuss the specic
ndings to evaluate the theoretical model and delineate the
strengths and limitations of the two SEM approaches, as indicated
by this study.
As a preliminary step the data was examined for kurtosis and
skewness to obtain insights about the distributional characteristics. This step is particularly important for CB-SEM since it
assumes normality in the data, but not for PLS-SEM since normality
is not assumed. Where both Kurtosis and Skewness fall within a
range of 1 to 1, data are considered within an acceptable range
(Hair, Celsi, Money, Samouel, & Page, 2011). In this case, Kurtosis
for 5 of 25 parameters fell outside the normal range, while
skewness for the sample was generally acceptable. The data were
therefore somewhat close to a normal distribution but a note of
caution about checking distribution normality is necessary for the
CB-SEM analysis, and for this type of analysis in general.
6.1. CB-SEM
Conrmatory factor analysis (CFA) was undertaken to further
assess the factor structure and validate the scales (Hair et al., 2010;

C.B. Astrachan et al. / Journal of Family Business Strategy 5 (2014) 116128

Hinkin, 1998) using the AMOS 20 software. As a preliminary step a

congeneric model was examined for model t, reliability, and
convergent validity and discriminant validity. The model consisted
of four constructs with 25 indicators Business expectations
(BE) = 9 indicators; Social expectations (SE) = 8 indicators; Expertise (EXP) = 4 indicators; and Trust (TRU) = 4 indicators, as shown
in Fig. 2.
The results of the initial CFA revealed a lack of t (x2 = 556.4;
DF = 269; p = .000, CFI = .857; RMSEA = .079). A systematic process
of examining the loadings and removing indicators with loadings
below .70 was followed (Hair et al., 2010). To achieve acceptable
model t it was necessary to eliminate 16 of the original 25 indicator
variables, including the reduction of the expertise construct to a
single item measure. An interim CFA model with a three-indicator
expertise construct was examined but two indicators exhibited
squared loadings below .40 and an AVE >.50 could not be achieved.
The chi-square for the nal 9 indicator, three construct model was
44.912 with 24 degrees of freedom, and a p = .006 (since the
expertise construct had only a single item, it was not included in the
nal CFA). The comparative t index (CFI) was .973 and RMSEA was
.071. Acceptable ranges for CFI are .9 or higher, and for RMSEA .08 or
less. The overall model t for the measurement model was therefore
within recommended ranges (Byrne, 2010; Hair et al., 2010).
Convergent validity and reliability are shown in Table 2.
Scale items loaded on their respective factors with loadings
ranging from .69 to .87 (Hair et al., 2010). The average variance
extracted (AVE) ranged from .54 to .63, conrming convergent

121

Table 2
CB-SEM convergent validity, reliability, and discriminant validity.
Variables

Business
expectations

Social
expectations

Trust

BE_1
BE_2
BE_9
SE_1
SE_2
SE_4
TRU25
TRU26
TRU27

.72
.72
.75

Average variance extracted

.54

.63

.62

Composite reliability
Cronbach alpha

.78
.78

.84
.83

.83
.83

.81
.87
.69
.87
.70
.77

Item
reliabilities
.53
.52
.57
.66
.75
.48
.76
.49
.60

FornellLarcker criterion*

BE

SE

Trust

BE
SE
Trust

.537
.425
.482

.63
.819

.617

validity and implicitly, content validity. Composite reliabilities

ranged from .78 to .84 demonstrating reliability for all constructs.
Table 2 also displays the results for the FornellLarcker procedure
(Fornell & Larcker, 1998) to assess discriminant validity. Discriminant validity is satisfactory for all constructs except the relationship between Trust and SE. The result was not unexpected,

Fig. 2. CB-SEM CFA model with 25 indicators.

122

C.B. Astrachan et al. / Journal of Family Business Strategy 5 (2014) 116128

Fig. 3. Structural model for CB-SEM based hypotheses tests.

however, since it represents a relationship between an exogenous

construct and an endogenous construct. Examination of the
indicators for these constructs shows the content in general is
distinct from a face validity perspective as well as based on the
literature (Fombrun et al., 2000; Newell & Goldsmith, 2001). In
sum, the three-construct model was considered satisfactory in
terms of content and convergent validity, discriminant validity,
and composite reliability.
The next step in CB-SEM is to analyze the structural model.
Fig. 3 shows the model tested and the path coefcients as well as R2
for the endogenous constructs. The chi-square for the structural
model is 58.99 with DF 30 resulting in normed x2 of 1.966 and a
p = .001. A normed x2 of 2 or less suggests the p = .001 is due to
sample size and not to lack of t. The CFI is .969 (.9 or greater
recommended) and RMSEA is .075 (.08 or less recommended).
These t measures are comparable to those obtained with the CFA.
The path coefcients are all signicant at the p = .000 level, with
the exception of the Business expectations to Trust path at p = .129,
and the Business expectations to Expertise path which is p = .037.
The SE to Expertise and Trust path coefcients of .54 and .61,

respectively, are relatively stronger than the paths from BE to Trust

of .12, BE to Expertise of .21, and Expertise to Trust of .32. The R2 for
the single indicator Expertise is .48 showing that Expertise is a
meaningful mediator. The overall R2 for Trust, the dependent
variable, is .89, which is considered strong however, the high R2 is
likely an artifact of the reduced measurement model that the
constraints of CB-SEM force on the researcher. Table 3 summarizes
the hypotheses tests based on the CB-SEM analysis. Four of the ve
hypotheses are supported at a signicance level of p < .05. The BE
to Trust relationship is rejected.
6.2. PLS-SEM
The 25 indicator theoretical model was run using the SmartPLS
version 2 software, as shown in Fig. 4. Note that with PLS-SEM the
analysis begins with a structural model and the CFA results are part
of the initial calculations. Recall that CB-SEM executes and conrms
the CFA before moving on to examine the structural model.
The initial output includes metrics to assess the measurement
characteristics of the outer model, which is what the constructs

Table 3
Results of hypotheses tests based on CB-SEM model.
Hypothesis

Hypotheses paths

Path coefcients

p values

Accept/reject signicance

H1
H2
H3
H4
H5

BE ! Trust
BE ! Expertise
SE ! Trust
SE ! Expertise
Expertise ! Trust

.12
.21
.61
.54
.32

.129
.037
.000
.000
.000

Reject
Accept
Accept
Accept
Accept

C.B. Astrachan et al. / Journal of Family Business Strategy 5 (2014) 116128

123

Fig. 4. Structural model for PLS-SEM with 25 indicators.

and their indicators are called in PLS-SEM (Sarstedt et al., 2014).

The standard metrics provided by SmartPLS are the indicator
loadings, Cronbach alphas and composite reliability, convergent
validity (AVE), path coefcients, cross loadings, interconstruct
correlations, latent variable scores, t-values, and more. A
systematic process of examining the loadings and removing
indicators with loadings below .70 was followed (Hair et al., 2010).
The result was the elimination of eight of the original 25 indicators.
Note that the Trust and Expertise constructs in the nal model both

had three indicators, the business expectations construct had ve

indicators, and the social expectations construct had six indicators.
Fig. 5 shows the nal structural model used for hypotheses testing.
The rst step in evaluating a PLS-SEM model is to examine the
outer model in an effort to validate the measurement model (Hair
et al., 2014). To do so, relationships between the constructs and
their indicators are assessed. Note that with PLS-SEM the Expertise
construct survived as a 3-indicator construct. As shown in Table 4,
composite reliability ranged from .86 to .90 for the four constructs,

Fig. 5. Structural model for PLS-SEM hypotheses tests.

C.B. Astrachan et al. / Journal of Family Business Strategy 5 (2014) 116128

124

Table 4
PLS-SEM average variances extracted, composite reliability and R2 for endogenous
constructs.
Construct

AVE

Composite
reliability

Business expectations
Expertise
Social expectations
Trust

.56
.67
.56
.75

.86
.86
.88
.89

R2

.42
.69

Table 5
PLS-SEM FornellLarcker test for discriminant validity.
Business
expectations

Cronbach alpha
Business expectations
Expertise
Social expectations
Trust

.81
.76
.84
.83

.56
.31
.52
.42

Expertise

.67
.39
.54

Social
expectations

.56
.57

Trust

.75

Diagonal values in bold are AVEs and off-diagonal values are squared interconstruct
correlations.

exceeding the minimum requirement of .7. The average variance

extracted (AVE) for the measurement models exceeded .56 for all
constructs, while the cutoff is .50 (Hair et al., 2012), thus indicating
convergent validity for all constructs. Table 5 shows the AVEs on
the diagonal and the squared interconstruct correlations off the
diagonal. The FornellLarcker criterion (Fornell & Larcker, 1998)
demonstrated that all AVEs were higher than the squared
interconstruct correlations, except SE which exhibited an AVE of
.56 and a shared variance with Trust of .57. At the same time, the
AVE of Trust was .74 so it met the guideline for discriminant
validity. An alternative assessment of discriminant validity using
cross loadings was also examined. All indicator loadings were
higher than their respective cross loadings, providing further
evidence of discriminant validity (Hair et al., 2014). Thus, overall
discriminant validity was achieved with the PLS-SEM analysis.
After the constructs have been conrmed as reliable and valid,
the next step is to assess the structural model results to identify
patterns in the data relationships. Before assessing the structural
model, we examined the model for collinearity, an important rst
step since the estimation of the path coefcients is based on OLS
regressions and they may be biased if multicollinearity is present
(Hair, Ringle, & Sarstedt, 2013). Minimum multicollinearity was
found so we then examined the models structural relationships.
The key criteria are the size and signicance of the path
coefcients, the level of the R2 values, and the predictive relevance
as measured by Q2 (Hair et al., 2014). We rst examined the sizes
and signicance of the path coefcients that represent the
hypothesized relationships. To obtain the signicance levels the
bootstrapping option was run using 5000 subsamples (Hair et al.,
2014). Table 6 shows the coefcients, T statistics, and summarizes
the results of the hypotheses tests.
An analysis of path coefcients and levels of signicance shows
that all ve hypotheses were accepted. Hypothesis 1 predicts a
positive relationship between BE and Trust, and it was accepted at
p = .05 (note that Hypothesis 1 was rejected with CB-SEM). The
relationship between BE and Expertise (H2) was also accepted
(p = .05). Hypothesis 3, which predicts a positive relationship
between Social Expectations and Trust is accepted at p = .01. The
positive relationship between Social Expectations and Expertise
(H4) was accepted at p = .01. Finally, the relationship between
Expertise and Trust (H5) was accepted (p = .01). The sizes of the
structural coefcients for the accepted hypotheses were all
considered meaningful for interpretation purposes (Hair et al.,
2014).

We next examined the R2 values for the two endogenous

constructs Expertise and Trust. R2 can be classied into one of
three categories for social science research: weak (.25), moderate
(.50), or substantial (.75) (Hair et al., 2010). Prediction of Trust, the
primary outcome measure of the model, was close to substantial,
with an R2 = .69. Prediction of expertise was slightly below
moderate, with an R2 = .42. The sizes of the R2 values for both
endogenous constructs were considered meaningful for interpretation purposes (Hair et al., 2014).
PLS-SEM includes an additional approach to evaluate structural
model predictive ability called blindfolding, which is not available
from CB-SEM analyses. Blindfolding was executed to evaluate the
predictive relevance of the endogenous latent construct indicators
(Hair et al., 2014). The blindfolding procedure produces the Q2,
which applies a sample re-use technique that omits part of the data
matrix and uses the model estimates to predict the omitted part.
For PLS-SEM models, a Q2 value larger than zero in the crossvalidated redundancy report indicates predictive relevance. As a
relative measure of predictive relevance, values of .02, .15 and .35
indicate that an exogenous construct has a small, medium, or large
predictive relevance for a selected endogenous construct (Hair
et al., 2014). For our path model the predictive relevance Q2 of Trust
was .496, the Q2 of Expertise was .264. The Q2 exceeded zero for
both endogenous constructs, indicating predictive relevance of the
construct indicators. Moreover, the predictive relevance for Trust
was large and for Expertise it was medium.
7. Discussion of CB-SEM and PLS-SEM results
Both analytical methods produced robust results and provided
conrmation that the Expertise construct acts as a meaningful
partial mediator. The results from both the CB-SEM and PLS-SEM
analyses show that a large amount of the variance in the
endogenous construct Trust is explained by the three constructs
of Social Expectations, Business Expectations, and Expertise,
namely, 89% (CB-SEM) and 69% (PLS-SEM), respectively. Both
models, therefore, clearly illustrate that the individuals social
expectations (How does a family rm do business?) inuence their
perception of organizational trustworthiness more than their
business-related expectations (What does a family rm do in order
to be successful?). In other words, being perceived as, for example,
an environmentally friendly company that supports good causes
contributes relatively more to organizational trustworthiness than
a companys growth perspectives or level of protability.

Table 6
Results of hypotheses tests based on PLS-SEM based model.
Hypothesis

Hypotheses paths

Path coefcients

T-values

Accept/reject signicance

H1
H2
H3
H4
H5

BE ! Trust
BE ! Expertise
SE ! Trust
SE ! Expertise
Expertise ! Trust

.13
.22
.40
.47
.41

1.999
2.314
7.135
5.515
7.669

Accept**
Accept**
Accept***
Accept***
Accept***

Critical t-values for a two-tailed test are: <1.96 (p > .05*), 1.96 (p = .05**), and 2.58 (p = .001***).

C.B. Astrachan et al. / Journal of Family Business Strategy 5 (2014) 116128

In terms of the strength of the relationships, the CB-SEM model

revealed a weak and insignicant relationship between Business
Expectations and Trust (.12), and a much stronger relationship
between Social Expectations and Trust (.61). In PLS-SEM, the
relative strength of the relationships was similar (.13 and .40), but
both relationships were signicant. The strength of the relationship between Expertise and Trust was somewhat higher for PLSSEM (.41) than it was for CB-SEM (.32) The results of this research
therefore suggest that if family businesses want to gain trust from
their stakeholders, their focus should be on promoting characteristics such as their workplace or environmental policies rather than
their nancial goals or performance objectives. Finally, it should be
noted that responses of individuals from Switzerland or of family
businesses in Switzerland might differ from perceptions in other
countries, so one should be cautious in generalizing these ndings.
While the results were similar, some important distinctions were
evident. The constraints of t requirements in the CB-SEM model
resulted in the elimination of 15 of the original 25 latent variable
indicators. A particularly notable result of the t requirements was
that the Expertise construct was reduced to a single item construct.
The overall model t was satisfactory, but was achieved at a
signicant cost in construct validity and reliability. Generally, the
recommended minimum number of indicators to measure latent
constructs is 35 (Hair et al., 2010), thus the minima were met for
both SEM approaches, except in the case of Expertise with CB-SEM.
The deleted indicators had meaningful content and face validity, and
the number of indicators that had to be eliminated with CB-SEM was
much higher than with PLS-SEM. While the R2 values of .48 and .89
for Expertise and Trust for CB-SEM were strong, the loss of content at
the theory development stage is a difcult trade-off to accept. In fact,
several scholars recommend maximizing retention of measures
even at the cost of model t, if needed (Byrne, 2010; DeVellis, 2011;
Hair et al., 2010).
The PLS-SEM analysis, on the other hand, resulted in retention
of 17 indicators and more importantly, the expertise construct had
3 indicators after elimination of only one item. The comparative
retention of indicators in the PLS-SEM approach enhances the
validity and reliability of that model, and provides insights for
future research and managerial implications that otherwise would
be lost. For example, with the PLS-SEM approach the differential
measures retained reected perception of prospects for future
growth, competitive standing and high product quality and
services. Face validity and literature supports the importance of
these measures to the company prole in the minds of customers
(Fombrun et al., 2000; Newell & Goldsmith, 2001).
Another set of differences arose in the signicance of the
Business expectations to Trust path coefcient, which was not
signicant in CB-SEM. In contrast the PLS-SEM structural paths
exhibited signicant path coefcients of p < .05 in all cases. While
the differences were not major, the possibility of the CB-SEM
approach resulting in lower signicance and thus elimination of
entire relationships must be kept in mind.
Notably, the CB-SEM approach resulted in a signicantly higher
R2 of .89, as compared with .69 for PLS-SEM. The R2 values for both
models are in the substantial range. A deeper analysis, however,
reveals that the apparent difference can be misleading. Specically,
the AVE of the endogenous construct Trust in the CB-SEM approach
was lower than the AVE of Trust using the PLS-SEM approach. As a
result, the overall variance predicted in the Trust construct by both
methods was comparable (CB-SEM = 55%; PLS-SEM = 52%), even
though nominally appearing different.
7.1. Theory and hierarchical component models
The previous examples of SEM are based on rst order
hierarchical component models. In some research contexts,

125

however, higher order models are more appropriate and sometimes necessary. A hierarchical component model (HCM) is a
general construct that consists of several subdimensions, often
referred to as rst order components. There are four main types of
HCMs discussed in scholarly literature (Jarvis, MacKenzie, &
Posakoff, 2003; Ringle, Sarstedt, & Straub, 2012; Wetzels,
Odekerken-Schroder, & van Oppen, 2009). The HCM that was
modeled in this study was reectivereective, which means the
relationships between the higher order construct and the two rst
order constructs, as well as the relationships between the rst
order constructs and their associated indicators, were all reective.
Selection of this type was based on theoretical considerations that
provided the underpinnings of this research.
Among the most important reasons for establishing HCMs is to
reduce the number of relationships in the structural model, making
the path model more parsimonious and easier to understand (less
complex). A second situation where HCMs are useful is when rst
order constructs are highly correlated (Hair et al., 2014). If high
multicollinearity exists among the rst order constructs structural
model coefcients may be biased, the signs may actually change,
and discriminant validity may not be possible. In such situations,
HCMs (most often 2nd order constructs) can reduce collinearity
problems and possibly solve discriminant validity issues.
HCMs are possible with both CB-SEM and PLS-SEM, but there
are different considerations. For example, in general CB-SEM
requires a minimum of three rst order constructs in order to
overcome identication issues (Hair et al., 2010). The exception to
this requirement is if you set the variance of the second order
construct to 1, or x the values of the two relationships (paths)
between the second order and rst order constructs to the value of
the most reliable rst order loading, typically somewhere in the
range of .75.85 (note that some scholars believe that fooling the
mathematics in this manner to overcome identication problems
with CB-SEM is problematic). In contrast, PLS-SEM can easily be
executed with only two rst order constructs since identication is
not a consideration with this method.
The PLS-SEM HCM is shown in Fig. 6. Note that the repeated
indicators approach was used to execute this PLS-SEM HCM (Hair
et al., 2014). As with other PLS-SEM approaches, the rst step in
evaluating a model that includes an HOC (Higher Order Component) is to examine the outer model in an effort to validate the
measurement model. Since the constructs were the same as in the
previous model, with the exception of the HOC, the model again
exhibited reliability, convergent validity, and discriminant validity. Moreover, the HOC demonstrated convergent validity. Examination of the path coefcients and levels of signicance indicated
that all ve hypotheses were accepted when the HOC was included
in the structural model. Finally, the structural path coefcients and
explained variances in the endogenous constructs were comparable to the previous PLS-SEM results. The major difference was in
the meaningful contribution of the Business expectations construct to the HOC, and ultimately toward predicting the other
endogenous constructs. The path loadings of the rst order
constructs can be used to interpret the contribution of the LOC
(lower order constructs) to the HOC. Note that the loading of the BE
construct is .89 and of the SE construct is .95. In short, the
structural model with the HOC is better able to capture the
contribution of the BE construct than was the previous model.
While multicollinearity among the exogenous constructs was not
an issue in this SEM, if it had been the HCM would have
represented an excellent solution. In terms of theory development,
there is some theoretical support for the HCM so in the current
study it would be considered an acceptable alternative competing
model. Thus, both models can contribute toward better understanding of these types of relationships within a family business
context.

126

C.B. Astrachan et al. / Journal of Family Business Strategy 5 (2014) 116128

Fig. 6. Structural model for PLS-SEM hierarchical component model.

8. Observations and conclusions

This study illustrates the use of two SEM approaches and
provides researchers with a format for CB-SEM and PLS-SEM
reporting. We explore the benets and applicability of SEM
methods by discussing the application of CB-SEM by means of
extant literature and the research study in this article.
Methodologically, our ndings highlight the differences regarding the requirements, application and analysis of the CB-SEM
and PLS-SEM approach. We found that the PLS-SEM approach at
the theory development stage enables retention of more indicator
variables and also conrmed the potential of a second-order
construct (Expectations). CB-SEM, while achieving a seemingly
better variance explained, on the other hand, resulted in a
signicant loss of indicator variables in order to nd an adequate
model t. Both approaches provided similar results, which were
robust in terms of R2. CB-SEM also requires the presumption of
normal distribution of data. In this case the data was not normally
distributed and the consequence was a most likely a higher
estimated R2. But the CB-SEM approach also requires a normal data
distribution. In this case the data was not normally distributed and
the consequences are lack of stability in the weights, and
potentially an inated R2. On balance, at the theory development
stage and low likelihood of normal distribution in general in the
social sciences, PLS-SEM seems to be more appropriate. CB-SEM is
more useful for later stage theory testing.
In terms of content, this study shows that social and business
expectations, two distinct dimensions of organizational reputation, are directly related to organizational trustworthiness. Our
ndings show that individuals social expectations are more
inuential on peoples perception of organizational expertise and
trustworthiness, as compared to business expectations. Moreover,
both social and business expectations are mediated by the
customers perception of expertise, which is enhanced by business
and more particularly social expectations. This level of insight in
the early stage of theory development as here can be useful for
further and managerial implications. We would like to note that
specic denitions of family or non-family (publicly-listed) rms

were not included in the questionnaire, as we wanted to assess the

respondents subjective attitudes toward these types of organizations. It is possible that providing an exact denition of both rm
types might have led to different results.
As SEM provides researchers with a comprehensive means for
assessing and modifying theoretical models (Anderson & Gerbing,
1988, p. 411), the use of PLS-SEM has grown in other disciplines
such strategic management or marketing (Hair et al., 2014). We
recommend the use of SEM approaches because it has been found
that contributions using SEM applications are more likely to be
recommended by reviewers, and hence more likely to published
than papers using other statistical procedures (Babin et al., 2008).
The rising popularity of second generation statistical approaches,
such as SEM methods, does not mean that traditional statistical
approaches such as multiple regression analysis are obsolete and
the results obtained using these methods are defective it simply
represents the evolution of statistical methods over time, and we
encourage family business scholars to consider using these new,
powerful and exible tools. Unlike in rst generation methods,
SEM approaches enable the researcher(s) to be more exible in
developing complex and realistic structural and measurement
models: we are now better able to develop new, or extend existing
theory and to evaluate the accuracy with which a theoretical
model reects reality (Kline, 2011). In fact PLS-SEM based
approaches can work with virtually any level of complexity in
constructs, observables and multi-level structural models (Hair
et al., 2014).
We recommend the use of SEM methods, and in particular, PLSSEM, because of its ability to handle small sample sizes, complex
models with numerous endogenous and exogenous constructs and
indicator variables, or non-normal data distributions many
situations commonly encountered in social sciences research (Hair
et al., 2014, p. xii) while still producing viable results. Lastly, we
recommend PLS-SEM because it is currently accessible to
everyone; the software can be downloaded at no cost, the
instructions are provided on the website, and the active PLSSEM community are extremely helpful in becoming familiar with
the software.

C.B. Astrachan et al. / Journal of Family Business Strategy 5 (2014) 116128

Our goal was to compare CB-SEM and PLS-SEM, and to discuss

the possibilities these approaches offer to family business
researchers. While we nd PLS-SEM to be more adequate for
our research context, both SEM approaches are fully justied, and
it is the researchers responsibility to identity the method that best
suits their research objective, data characteristics, and model setup
(Hair et al., 2014, p. 18).
Appendix. Family rm reputation survey: scale items
How well do you think the statements below apply to a typical
family rm? (1 = I strongly disagree, 7 = I strongly agree)
Social expectations
Emotional appeal, workplace environment, social &
environmental responsibility
(SE1) I have a good feeling about family rms
(SE2) Are trustworthy
(SE3) I admire and respect family rms
(SE4) Have high standards with employees
(SE5) Stand behind their products and services
(SE6) Support good causes
(SE7) Are good companies to work for
(SE8) Are environmentally friendly
Business expectations
Vision & leadership, nancial performance, products & services
(BE1) Recognize and take advantage of market opp.
(BE2) Develop innovative products and services
(BE3) Have a clear vision for their future
(BE4) Strong prospects for future growth
(BE5) Have a strong record of protability
(BE6) Offer good value for money
(BE7) Tend to outperform their competitors
(BE8) Offer high quality products and services
(BE9) Look like they have good employees
How well do you think the statements below apply to a typical family rm? (1 = I
strongly disagree, 7 = I strongly agree)
Expertise
(EXP21) Have a great amount of experience
(EXP22) Are skilled in what they do
(EXP23) Have great expertise
(EXP24) Do not have much experience
Trust
(TRU25)
(TRU26)
(TRU27)
(TRU28)

I trust this company

This company makes truthful claims
This company is honest
I do not believe what this company tells me

References
Anderson, J. C., & Gerbing, D. W. (1988). Structural equation modeling in practice: A
review and recommended two-step approach. Psychological Bulletin, 3, 411423.
Armstrong, J. S., & Overton, T. S. (1977). Estimating nonresponse bias in mail surveys.
Journal of Marketing Research, 14(3), 396402.
Astrachan, J. H., Klein, S. B., & Smyrnios, K. X. (2002). The F-PEC scale of family
inuence: A proposal for solving the family business denition problem. Family
Business Review, 15(1), 4558.
Babin, B. J., Hair, J. F., & Boles, J. S. (2008). Publishing research in marketing journals
using structural equation modeling. Journal of Marketing Theory & Practice, 16(4),
279285.
Bagozzi, R. P. (1980). Causal models in marketing. New York: Wiley.
Binz, C., Hair, J. F., Pieper, T., & Baldauf, A. (2013). Exploring the effect of distinct family
rm reputation on consumers preferences. Journal of Family Business Strategy, 4(1),
311.
Byrne, B. M. (2010). Structural equation modeling with AMOS: Basic concepts, applications,
and programming (2nd ed.). New York: Routledge.
Carrigan, M., & Buckley, J. (2008). Whats so special about family business? An
exploratory study of UK and Irish consumer experiences of family businesses.
International Journal of Consumer Studies, 32(6), 656666.
Chin, W. W., Peterson, R. A., & Brown, S. P. (2008). Structural equation modeling in
marketing: Some practical reminders. Journal of Marketing Theory & Practice, 16(4),
287289.
Craig, J. B., Dibrell, C., & Davis, P. S. (2008). Leveraging family-based brand identity to
enhance rm competitiveness and performance in family businesses. Journal of
Small Business Management, 46(3), 351371.
Debicki, B. J., Matherne, C. F., Kellermanns, F. W., & Chrisman, J. J. (2009). Family
business research in the new millennium: An overview of the who, the where, the
what, and the why. Family Business Review, 22(2), 151166.

127

DeVellis, R. F. (2011). Scale development: Theory and applications. Thousand Oaks:

Sage.
Dyer, W. G., & Dyer, W. J. (2009). Putting the family into family business research.
Family Business Review, 22(3), 216219.
Dyer, W. G., & Whetten, D. A. (2006). Family rms and social responsibility:
Preliminary evidence from the S&P 500. Entrepreneurship Theory and Practice,
30(6), 785802.
Fombrun, C. J. (1996). Reputation: Realizing value from the corporate image. Boston, MA:
Harvard Business School Press.
Fombrun, C. J., Gardberg, N. A., & Sever, J. M. (2000). The reputation quotient: A
multistakeholder measure of corporate reputation. Journal of Brand Management,
7(4), 241255.
Fornell, C., & Larcker, D. F. (1998). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(1),
3950.
Hair, J. F., Black, W. C., Babin, B. J., & Anderson, R. E. (2010). Multivariate data analysis
(7th ed.). NJ: Prentice Hall.
Hair, J. F., Celsi, M., Money, A., Samouel, P., & Page, M. (2011). Essentials of business
research methods (2nd ed.). Armonk, NY: ME Sharpe.
Hair, J. F., Hult, G. T. M., Ringle, C. M., & Sarstedt, M. (2014). A primer on partial least
squares structural equation modeling (PLS-SEM). Thousand Oaks: Sage.
Hair, J. F., Ringle, C. M., & Sarstedt, M. (2011a). PLS-SEM: Indeed a silver bullet. Journal of
Marketing Theory and Practice, 18(2), 139152.
Hair, J. F., Ringle, C. M., & Sarstedt, M. (2011b). The use of partial least squares (PLS) to
address marketing management topics: From the special issue guest editors.
Journal of Marketing Theory and Practice, 18(2), 135138.
Hair, J. F., Ringle, C. M., & Sarstedt, M. (2013). Partial least squares structural equation
modeling: Rigorous applications, better results and higher acceptance. Long Range
Planning, 46(1-2), 112.
Hair, J. F., Sarstedt, M., Ringle, C. M., & Mena, J. A. (2012). An assessment of the use of
partial least squares structural equation modeling in marketing research. Journal of
the Academy of Marketing Science, 40(3), 414433.
Hinkin, T. R. (1998). A brief tutorial on the development of measures for use in survey
questionnaires. Organizational Research Methods, 1(1), 104121.
Holt, D. T., Rutherford, M. W., & Kuratko, D. F. (2010). Advancing the eld of family
business research: Further testing the measurement properties of the F-PEC.
Family Business Review, 23(1), 7688.
Jarvis, C. B., MacKenzie, S. B., & Posakoff, P. M. (2003). A critical review of construct
indicators and measurement model specication in marketing and consumer
research. Journal of Consumer Research, 30, 199218.
Joreskog, K. G., Sorbom, D., du Toit, S., & du Toit, M. (1999). LISREL 8: New statistical
features. Chicago, IL: Scientic Software.
Kashmiri, S., & Mahajan, V. (2010). Whats in a name? An analysis of the strategic
behavior of family rms. International Journal of Research in Marketing, 27(3), 271
280.
Klein, S. B., Astrachan, J. H., & Smyrnios, K. X. (2005). The F-PEC Scale of family
inuence: Construction, validation, and further implication for theory. Entrepreneurship Theory and Practice, 29(3), 321339.
Kline, R. B. (2005). Principles and practice of structural equation modeling (2nd ed.). New
York: Guilford Press.
Kline, R. B. (2011). Principles and practice of structural equation modeling (3rd ed.). New
York: The Guilford Press.
Lei, P. W., & Wu, Q. (2007). Introduction to structural equation modeling: Issues and
practical considerations. Educational Measurement: Issues and Practices, 26(3),
3343.
MacCallum, R. C., & Austin, J. T. (2000). Applications of structural equation modeling in
psychological research. Annual Review of Psychology, 51, 201226.
Memili, E., Eddleston, K. A., Kellermanns, F. W., Zellweger, T. M., & Barnett, T. (2010).
The critical path to family rm success through entrepreneurial risk taking and
image. Journal of Family Business Strategy, 1(4), 200209.
Miller, D., & Le Breton-Miller, I. (2005). Management insights from great and struggling
family businesses. Long Range Planning, 38(6), 517530.
Miller, N. J., McLeod, H., & Young Ob, K. (2001). Managing family businesses in small
communities. Journal of Small Business Management, 39(1), 7387.
Mustakallio, M., Autio, E., & Zahra, S. A. (2002). Relational and contractual governance
in family rms: Effects on strategic decision making. Family Business Review, 15(3),
205222.
Newell, S., & Goldsmith, R. (2001). The development of a scale to measure perceived
corporate credibility. Journal of Business Research, 52(3), 235247.
ODonnell, A., Gilmore, A., Carson, D., & Cummins, D. (2002). Competitive advantage
in small to medium-sized enterprises. Journal of Strategic Marketing, 10(3),
205223.
Orth, U. R., & Green, M. T. (2009). Consumer loyalty to family versus non-family
business: The roles of store image, trust and satisfaction. Journal of Retailing and
Consumer Services, 16(4), 248259.
Rigdon, E. E. (2012). Rethinking partial least squares path modeling: In praise of simple
methods. Journal of Long Range Planning, 45, 341358.
Ringle, C., Sarstedt, M., & Hair, J. F. (2013). Partial least squares structural equation
modeling: Rigorous applications, better results and higher acceptance. Journal of
Long Range Planning, 46(1), 112.
Ringle, C., Sarstedt, M., Hair, J. F., & Pieper, T. (2012). The use of partial least squares
structural equation modeling in strategic management research: A review of past
practices and recommendations for future applications. Journal of Long Range
Planning, 45(6), 320340.
Ringle, C., Sarstedt, M., & Straub, D. (2012). A critical look at the use of PLS-SEM in the
MIS quarterly. MIS Quarterly, 36, iiixiv.

128

C.B. Astrachan et al. / Journal of Family Business Strategy 5 (2014) 116128

Sarstedt, M., Ringle, C., Smith, D., Reams, R., & Hair, J. (2014). Partial least squares
structural equation modeling (PLS-SEM): A useful tool for family business
researchers. Journal of Family Business Strategy5(1).
Sharma, P. N., & Kim, K. H. (2013). A comparison of PLS and ML bootstrapping
techniques in SEM. In H. Abdi et al. (Eds.), New perspectives in partial least squares
and related methods. Heidelberg: Springer Verlag.
Shook, C. L., Ketchen, D. J., Cycyota, C. S., & Crockett, D. (2003). Data analytic trends in
strategic management research. Strategic Management Journal, 24(12), 12311237.
Shook, C. L., Ketchen, D. J., Hult, G. T. M., & Kacmar, K. M. (2004). An assessment of the
use of structural equation modeling in strategic management research. Strategic
Management Journal, 4, 397404.
Tagiuri, R., & Davis, J. (1996). Bivalent attributes of the family rm. Family Business
Review, 9(2), 199208.
Wang, J., & Wang, X. (2012). Structural equation modeling: Applications using Mplus.
Chicester, UK: John Wiley & Sons Ltd.
Ward, J. L. (1997). Growing the family business: Special challenges and best practices.
Family Business Review, 10(4), 323337.
Westhead, P., & Howorth, C. (2006). Ownership and management issues associated
with family rm performance and company objectives. Family Business Review,
19(4), 301316.

Wetzels, M., Odekerken-Schroder, G., & van Oppen, C. (2009). Using PLS path modeling
for assessing hierarchical construct models: Guidelines and empirical illustration.
MIS Quarterly, 33, 177195.
Williams, L. J., Vandenberg, R. J., & Edwards, J. R. (2009). Structural equation modeling
in management research: A guide for improved analysis. Academy of Management
Annals, 3(1), 543604.
Wilson, S. R., Whitmoyer, J. G., Pieper, T. M., Astrachan, J. H., Hair, J. F., & Sarstedt, M.
(2014). Method trends and method needs: Examining methods needed for accelerating the eld. Journal of Family Business Strategy. (forthcoming).
Zahra, S. A., Hayton, J. C., Neubaum, D. O., Dibrell, C., & Craig, J. B. (2008). Culture of
family commitment and strategic exibility: The moderating effect of stewardship.
Entrepreneurship Theory and Practice, 32(6), 10351054.
Zellweger, T. M., Kellermanns, F. W., Eddleston, K. A., & Memili, E. (2012). Building a
family rm image: How family rms capitalize on their family ties. Journal of
Family Business Strategy, 3(4), 239250.
Zellweger, T. M., Nason, R. S., & Nordqvist, M. (2012). From longevity of rms to
transgenerational entrepreneurship of families introducing family entrepreneurial
orientation. Family Business Review, 25(2), 136155.