Astrachan2014 Important for SEM PLS & CB

7/26/2019 Astrachan2014 Important for SEM PLS & CB

1/13

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

in

family

firm

research

Claudia Binz Astrachan a, Vijay K. Patel b, Gabrielle Wanzenried c,*aWitten/Herdecke University, GermanybKennesaw State University, USAc Lucerne University of Applied Sciences and Arts, Switzerland

1. Structural equation modeling in a nutshell

Structural equation modeling (SEM) has seen a dramatic rise in

attention and utilization across a variety of scientific 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 reflect [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 themore widelyused 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

arehypothetical,ornotdirectly observable (Williams,Vandenberg,

& 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).WhileSEM

is similar to multiple regression in the sense that both techniques

test relationshipsbetweenvariables, SEM is able to simultaneously

examine multi-leveldependence 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 benefits and

limitations of SEM in general, and in family business research in

particular, by directly comparing two major approaches tostructural 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

twodifferentapproachesto thesameproblemnamely, theanalysis

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

Journal of Family Business Strategy 5 (2014) 116128

A R T I C L E I N F O

Keywords:

Structural equation modeling (SEM)

Covariance-based SEM

Partial least squares-SEM

Family firms

Organizational reputation

Organizational trustworthiness

A B S T R A C T

Structural equation modeling (SEM) has become the methodology of choice for many family businessresearchers 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 andintricaterelationships thatpreviously couldnot be easily untangledand examined. Inmany

cases,however, researchers struggle tomeet some of the challenging requirements of covariance-based

SEM(CB-SEM), themost commonly used approach to SEM, such as distribution assumptions or sample

size. In this article, we point out the benefits and disadvantages of CB-SEM, and present a comparison

with partial least squares-SEM (PLS-SEM) using an identical sample. We find 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.

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

E-mail address: [email protected] (G. Wanzenried).

Contents

lists

available

at

ScienceDirect

Journal of Family Business Strategy

journal homepage: www.elsev ier .co m/ locate / j fbs

1877-8585/$ see front matter 2014 Elsevier Ltd. All rights reserved.

http://dx.doi.org/10.1016/j.jfbs.2013.12.002
http://dx.doi.org/10.1016/j.jfbs.2013.12.002http://dx.doi.org/10.1016/j.jfbs.2013.12.002http://dx.doi.org/10.1016/j.jfbs.2013.12.002http://dx.doi.org/10.1016/j.jfbs.2013.12.002http://dx.doi.org/10.1016/j.jfbs.2013.12.002http://dx.doi.org/10.1016/j.jfbs.2013.12.002http://dx.doi.org/10.1016/j.jfbs.2013.12.002http://dx.doi.org/10.1016/j.jfbs.2013.12.002http://dx.doi.org/10.1016/j.jfbs.2013.12.002http://dx.doi.org/10.1016/j.jfbs.2013.12.002http://dx.doi.org/10.1016/j.jfbs.2013.12.002http://dx.doi.org/10.1016/j.jfbs.2013.12.002http://dx.doi.org/10.1016/j.jfbs.2013.12.002http://dx.doi.org/10.1016/j.jfbs.2013.12.002http://dx.doi.org/10.1016/j.jfbs.2013.12.002mailto:[email protected]:[email protected]://www.elsevier.com/locate/jfbshttp://www.elsevier.com/locate/jfbshttp://www.elsevier.com/locate/jfbshttp://www.elsevier.com/locate/jfbshttp://www.elsevier.com/locate/jfbshttp://www.elsevier.com/locate/jfbshttp://www.elsevier.com/locate/jfbshttp://dx.doi.org/10.1016/j.jfbs.2013.12.002http://dx.doi.org/10.1016/j.jfbs.2013.12.002http://www.elsevier.com/locate/jfbsmailto:[email protected]://dx.doi.org/10.1016/j.jfbs.2013.12.002http://crossmark.crossref.org/dialog/?doi=10.1016/j.jfbs.2013.12.002&domain=pdfhttp://crossmark.crossref.org/dialog/?doi=10.1016/j.jfbs.2013.12.002&domain=pdf


2/13

(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 fit (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 specified, and the

applicability and user-friendliness of available software, but also

the results and interpretations.

The remainder of this article is structured as follows: first, we

briefly highlight the most important benefits 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 findings of these studies. Third, the research

context of the example used in this study is briefly described, and

the hypotheses as well as an outline of the methodology are

presented. Fourth,wediscuss the results from theCB-SEM and PLS-SEM analyses. Finally, practical observations and conclusions are

provided, and limitations and suggestions for further research are

presented.

2. The benefits and limitations of SEM

2.1.

The

benefits

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 depen-

dent

variables,

latent

constructs

based

on

multi-item

indicator

variables, and multiple stages/levels of constructs in a structuralmodel. 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 influence 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

reflect

the

latent

variable

entirely accurately, which means the measurement will contain

error

as

will

the

results.

By

explicitly

assessing

error

in

thestructural

model,

SEM

provides

a

powerful

means

of

simultaneously

assessing

the

quality

of

measurement

and

examining

causal

relation-

ships 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

findings.

Additionally,

the

SEM approach is designed to consider interactive effects and

complex models to find an optimal model that reduces cross-

loadings

and

identifies

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 vari-

ables,

e.g.,

family

ownership

has

a

direct

positive

effect

on

firm

performance.

Indirect

effects

involve

relationships

between

independent and dependent variables that are mediated or

moderated by some other variable, e.g., the effect of family

ownership on firm 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 specifically 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 cohe-

sion 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 fit is based on accurately estimating the

observed covariance matrix, while with PLS-SEM fit is based upon

accounting for explained variance in the endogenous constructs

(Hair et al., 2014). As a result of model fit 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 theory-

based 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 hasmade 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 methodologi-

cal flaws 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

ofCB-SEM

in

particular,

since

it

is

a

confirmatory

approach,

the

method

requires

the

specification

of

the

full

theoretical

model

prior to data analysis. The researcher(s) must therefore define the

exact number of dependent (endogenous) and independent

(exogenous)

variables

used

in

the

theoretical

model,

the

relation-

ships

between

these

latent

variables,

the

type

of

measurement

model (formative or reflective), 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

specified

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

C.B. Astrachan et al./Journal of Family Business Strategy 5 (2014) 116128 117


3/13

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 find relationships, their directions and

strengths, as well as observable measures, PLS-SEM is appropriate.

The model fit constraints of CB-SEM are more appropriate for

established theory testing and confirmation, 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: Recommenda-

tions 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 first

on the method used. Specifically, CB-SEM requires larger samples

than PLS-SEM because relationships between all variables must be

assessed (i.e., a full information approach),whilewith PLS-SEM the

model is separated into different smaller components (a compo-

nent 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 efficiently

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). Thismakes PLS-SEM particularly suitablefor 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 offive times thenumber 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 maximumlikelihood 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 benefits, as

compared

with

first

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,

regressionanalysis

does

not

directly

permit

assessment

of

measurement

characteristics

so

latent

constructs

must

first

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 firm

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

publishedbetween1989 and 2013used SEMmethodologies, 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

wouldprovide 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 firms, 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 fit of

the overall measurement model as well as the strength andsignificance 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 relation-

ship between board monitoring and the board commitment to

strategic decisions was hypothesized to be positive the results did

not confirm the relationship. In sum, SEM can shed light on the

theoretical causality of relationships between latent and observ-

able 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 non-experimental data (Bagozzi, 1980). Astrachan, Klein, and Smyrnios

(2002; see also Klein, Astrachan, & Smyrnios, 2005) used CB-SEM

when

developing

their

continuous

(rather

than

dichotomous)

F-

PEC

scale

of

family

influence,

which

today

is

one

of

the

few

widely

accepted, measureable, and validated conceptualizations of fami-

ly-owned business (see also Holt, Rutherford, & Kuratko, 2010).

The

F-PEC

scale

is

an

index

of

family

influence,

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

confirm

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

(fit)with

all

the

relationships

(covariances)

in

the

data

set.

By

calculating

a

range

of

goodness-of-fit

statistics,

CB-SEM

can

assess

whether the theoretical model is confirmed. While scale develop-

ment is possible based on exploratory factor analysis (EFA), an

inherent

advantage

of

SEM

is

that

it

includes

Confirmatory

Factor

Analysis

(CFA),

which

is

considered

a

superior

approach

to

scale

development (Hair et al., 2010). SEM based modeling enablesmore

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) 116128118


4/13

(willingness to undertake high risks) and family firm image

(promotion of the firms family background) on the relationships

between the degree of family ownership and the owning familys

identification with the firm (independent variables) and firm

performance (dependent variable). Using a CB-SEM approach, and

comparing the fit indices of both a fully and partially mediated

model enabled the authors to show that a fully mediated model fit

our data best, showing, for example, that the possible relationship

between family expectations and family firm performance was fully

mediated

by

family

firm

image

and

risk

taking (Memili et al., 2010,

p. 206).

Besides the three examples presented above, other interesting

possible applications in familybusiness researchmight include the

examination of group differences such as differences between

family and non-family firms but also, and possibly more

important, within the family firm group cross-cultural compar-

isons (e.g., family firms 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 firms 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 specification and data collection con-straints, 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. Specifically, 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.

Further-

more, we test if perceived expertise acts as a mediating factor.Corporate credibility refers to the expertise and trustworthi-

ness a potential customer attributes to an organization, or in other

words,

the

extent

to

which

consumers

feel

that

the

firm

has

the

knowledge

or

ability

to

fulfill

its

claims

and

whether

the

firm

can

be

trusted to tell the truth (Newell & Goldsmith, 2001, p. 235).

Corporate credibility has been shown to influence customer

attitudes

and

ultimately

purchase

decisions

and

therefore

firm

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

firms

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

firm

reputation

and

branding,

and

findings

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 firm brand, which refers to the active

promotion

of

a

firms

family

background

(e.g.,

SC

Johnson:

A

family

company),

may

lead

to

superior

organizational

reputation,

and

that such distinct family firm reputation could be a unique asset

that family firms 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

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 firms 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 firms long-term existence implies. The fact

that a company has been around for decades, implying that

knowledgeand experiencehavebeen 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, family-

owned companies are assumed to be viewed as more trustworthy

than publicly-owned firms, 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 findings from a

previous study examining the effects of distinct family firm

reputation on customer preferences (Binz et al., 2013). In this

research, we use the adapted reputation construct with two

dimensions, which was identified in the previous study based on

exploratory factor analysis. The social expectations dimension

refers to how a company conducts its business, while the businessexpectations 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

satisfied with the quality of the products they purchased, the

company fulfilled 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 innovativeproducts and services) and the degree to which

they trust that an organization will fulfill 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.



5/13

H5.

Organizational

expertise

is

positively

related

to

organizational

trustworthiness.

5. Methodology

5.1.

Measures

To investigate the relationship between distinct family firmreputation and perceived trustworthiness of family businesses, a

standardized questionnaire was developed based on two estab-

lished scales. One scale was the Fombrun, Gardberg, and Sever

(2000) Reputation Quotient Scale, which consists of six dimen-

sions of corporate reputation, namely emotional appeal, products

and

services, vision

and leadership, workplace

environment, social

and environmental responsibility, and financial performance ,

measured with 7-point Likert scales. The original wording of

the scale items was adapted by replacing the term organization

with family firm in each question to fit 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

measurecorporate credibility or the amount of expertise and trustworthi-

ness

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

firm

(see Binz et al., 2013 for details).

5.2.

Sample

profile

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

LucerneUniversity

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,

repre-

senting an overall response rate of36.25%. The sample size exceeds

the minimum required for the application of either CB-SEM or PLS-

SEM

(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 significant 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 con-

structs. When the CFA did not achieve acceptable fit, and thus the

data

did

not

reflect

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

of

questions).

The

SE

and

BE

constructs

along

with

the

Expertise

and Trust constructs were then used to run the CB-SEM and PLS-SEM 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 theoreticalmodel

and

hypotheses.

We

present

an

overview

of

our

approach

and

findings

as

well

as

comparative

results.

We

also

discuss

the

specific

findings 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 character-

istics.

This

step

is

particularly

important

for

CB-SEM

since

itassumes

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

Confirmatory

factor

analysis

(CFA)

was

undertaken

to

further

assess

the

factor

structure

and

validate

the

scales

(Hair

et

al.,

2010;

Table 1

Exploratory factor analysis.

Variables Factor 1:

Social

expectations

Factor 2:

Business

expectations

I have a good feeling about family firms .881

I trust family firms .884

I admire and respect family firms .883

Family firms stand behind their products

and

services

.747

Family firms look like good companies

to work for

.645

Family firms support good causes .705

Family firms are environmentally friendly .658

Family firms have high standards with

employees

.812

Family firms develop innovative products

and services

.759

Family firms offer high quality products

and services

.557

Family firms offer good value for money .563

Family firms have a clear vision for

their future

.741

Family firms recognize and take advantage

of market opp.

.838

Family firms look like they have good

employees

.549

Family firms have a strong record

of profitability

.663

Family firms tend to outperform

their competitors

.580

Family firms have strong prospects

for future growth

.638

Eigenvalue 6.88 2.65

% of variance 40.4 15.6

Total variance explained 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).

C.B. Astrachan et al./Journal of Family Business Strategy 5 (2014) 116128120


6/13

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

congeneric model was examined for model fit, 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; Exper-

tise (EXP) = 4 indicators; and Trust (TRU) = 4 indicators, as shown

in Fig. 2.

The results of the initial CFA revealed a lack of fit (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

modelfit itwasnecessary toeliminate 16of theoriginal 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 .40and an AVE >.50 could not be achieved.

The chi-square for the final 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, itwasnot included in the

final CFA). The comparative fit index (CFI) was .973 and RMSEA was

.071.Acceptable ranges for CFI are .9orhigher,and forRMSEA .08or

less.Theoverall modelfit for themeasurement modelwas 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, confirming convergent

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. Discrimi-

nant validity is satisfactory for all constructs except the relation-

ship between Trust and SE. The result was not unexpected,

Fig.

2.

CB-SEM

CFA

model

with

25

indicators.

Table 2

CB-SEM convergent validity, reliability, and discriminant validity.

Variables Business

expectations

Social

expectations

Trust Item

reliabilities

BE_1 .72 .53

BE_2 .72 .52

BE_9 .75 .57

SE_1 .81 .66

SE_2 .87 .75

SE_4

.69

.48TRU25 .87 .76

TRU26 .70 .49

TRU27 .77 .60

Average variance extracted .54 .63 .62

Composite reliability .78 .84 .83

Cronbach alpha .78 .83 .83

FornellLarcker criterion* BE SE Trust

BE .537

SE .425 .63

Trust .482 .819 .617



7/13

however,

since

it

represents

a

relationship

between

an

exogenous

construct and an endogenous construct. Examination of theindicators 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

coefficients

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

fit.

The

CFI

is

.969

(.9

or

greater

recommended) and RMSEA is .075 (.08 or less recommended).

These

fit

measures

are

comparable

to

those

obtained

with

the

CFA.The

path

coefficients

are

all

significant

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 coefficients 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 forthe 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 five

hypotheses

are

supported

at

a

significance

level

of

p

Documents

Astrachan2014 Important for SEM PLS & CB