You are on page 1of 29

This article was downloaded by: [Universiti Sains Malaysia]

On: 01 December 2011, At: 17:00


Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954
Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH,
UK

The Journal of Psychology


Publication details, including instructions for
authors and subscription information:
http://www.tandfonline.com/loi/vjrl20

Formative Versus Reflective


Measurement: An Illustration
Using Work–Family Balance
a b
Thomas Ellwart & Udo Konradt
a
University of Trier
b
University of Kiel

Available online: 18 Jul 2011

To cite this article: Thomas Ellwart & Udo Konradt (2011): Formative Versus
Reflective Measurement: An Illustration Using Work–Family Balance, The Journal of
Psychology, 145:5, 391-417

To link to this article: http://dx.doi.org/10.1080/00223980.2011.580388

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-


and-conditions

This article may be used for research, teaching, and private study purposes.
Any substantial or systematic reproduction, redistribution, reselling, loan,
sub-licensing, systematic supply, or distribution in any form to anyone is
expressly forbidden.
The publisher does not give any warranty express or implied or make any
representation that the contents will be complete or accurate or up to
date. The accuracy of any instructions, formulae, and drug doses should be
independently verified with primary sources. The publisher shall not be liable
for any loss, actions, claims, proceedings, demand, or costs or damages
whatsoever or howsoever caused arising directly or indirectly in connection
with or arising out of the use of this material.
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011
The Journal of Psychology, 2011, 145(5), 391–417
Copyright 
C 2011 Taylor & Francis Group, LLC

Formative Versus Reflective


Measurement: An Illustration Using
Work–Family Balance
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

THOMAS ELLWART
University of Trier

UDO KONRADT
University of Kiel

ABSTRACT. The aim of this article is to propose the formative measurement approach
that can be used in various constructs of applied psychology. To illustrate this approach, the
authors will (a) discuss the distinction between commonly used principal-factor (reflective)
measures in comparison to the composite (formative) latent variable model, which is
often applied in other disciplines such as marketing or engineering, and (b) point out the
advantages and limitations of formative specifications using the example of the work–family
balance (WFB) construct. Data collected from 2 large cross-sectional field studies confirm
the reliability and validity of formative WFB measures as well as its predictive value
regarding criteria of WFB (i.e., job satisfaction, family satisfaction, and life satisfaction).
Last, the specific informational value of each formative indicator will be demonstrated and
discussed in terms of practical implications for the assessment in different psychological
fields.
Keywords: formative measurement model, reflective measurement model, satisfaction,
work–family balance

MOST VARIABLES IN PSYCHOLOGY are latent constructs, which are mea-


sured with scales founded on classical test theory. Classical test theory assumes
that each observable item can be viewed as a reflection of the underlying latent
construct (cf. Bollen & Lennox, 1991). The point is “that in this type of measure-
ment model, the latent construct is empirically defined in terms of the common
variance among the indicators” (MacKenzie, Podsakoff, & Jarvis, 2005, p. 710).
Because the latent construct explains the common variation in the indicators, one
speaks of reflective measurement, where meaning flows from the latent construct

Address correspondence to Thomas Ellwart, University of Trier, Department of Psychology,


D-54286 Trier, Germany; ellwart@uni-trier.de (e-mail).

391
392 The Journal of Psychology

to the indicators (cf. MacKenzie et al., 2005; Nunnally & Bernstein, 1994). From
this measurement perspective, almost all scales in psychological research are of
reflective nature with observable items as reflections of the latent variable. For
example, the construct of work–family balance (WFB) could be measured with
items that represent different but correlated reflections of a person’s ability to man-
age the work–family interface (e.g., overall, I’m satisfied with my WFB; if people
asked me, there are seldom conflicts between work and family demands; and my
family is happy how well job and family work together in our lives). However,
psychological research widely ignored a complementary alternative to the reflec-
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

tive specifications, namely formative indicator scales (Bollen & Lennox, 1991).
Not as in reflective scales (classical1 ), in the new formative indicator models the
items (i.e., indicators) are combinations that form the composite latent variable
(Jarvis, MacKenzie, & Podsakoff, 2003). In terms of the WFB example, a forma-
tive scale would capture observable indicators that are associated with work- or
family-related conflict or imbalance (e.g., job and family demands, time demands),
whereas reflective indicators capture the evaluation of the consequences. Forma-
tive scales are valuable to model and to quantify the impact of multiple dimensions
on a latent variable (Jarvis et al., 2003; MacKenzie et al., 2005), with applications
in various other disciplines such as marketing research (Coltman, Divinney, Midg-
ley, & Veniak, 2008; Fassott, 2006; Jarvis et al., 2003), engineering psychology
(Petter, Straub, & Rai, 2007), and training evaluation (Konradt, Christophersen,
& Ellwart, 2008). Several authors underline the importance of differentiating be-
tween formative (new) and reflective (classical) measurement models because
of statistical and conceptual reasons (for an overview, see MacKenzie et al.,
2005).
In the present article, the formative measurement approach is illustrated by
the example of the construct of WFB and by means of data collected from two
empirical studies. The construct of WFB is a useful example, because it represents
a typical psychological latent variable with multiple theoretical dimensions, differ-
ent observable indicators, and various antecedents. Therefore, the concept can be
easily transferred to other variables, for example, work satisfaction, identification,
motivation, and leadership. The objective is to underline the advantages of the new
formative scales for researchers and practitioners but also to discuss its limitations.
Our goal is not to rule out reflective conceptualization but rather to complement
the existing research on reflective measures by showing the potential advantages
of a formative approach. In the following, the concept of formative and reflective
indicator scales will be explained in more detail by means of the construct of
WFB. An exemplary formative WFB scale will then be developed, consisting of
indicators that influence rather than reflect the latent construct. With data collected
from two studies, the statistical and conceptual properties of formative measures
will be highlighted. Last, we will discuss practical advantages and limitations for
formative applications in applied psychology.
Ellwart & Konradt 393

The Reflective Measurement Models of WFB

Generally, WFB is a multifaceted concept that has been conceptualized in


many different ways (Eby, Casper, Lockwood, Bordeaux, & Brinley, 2005; Jones,
Burke, & Westman, 2006). Although research has not yet presented a consistent
definition of WFB, a common conception aims at delineating WFB as equilibrium
between job and family roles (Jones, Burke, et al., 2006) or as the lack of con-
flict between the two domains (cf. Barnett, 1998; Frone, 2003). Similar to other
constructs in psychological assessment, the WFB construct is mostly specified
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

as reflective (classical; Frone, 2003; Kossek & Ozeki, 1998; Tetrick & Buffardi,
2006). This means that the manifest variables (or the items measured) are consid-
ered to be a reflection or an effect of their latent variable as displayed in Figure 1 (cf.
Tenenhaus, Vinzi, Chatelin, & Lauro, 2004). As observable reflective indicators
are modeled as similar effects of latent constructs, items should be highly corre-
lated due to the fact that they all constitute manifestations of the same underlying
construct (i.e., internal consistency; Jarvis et al., 2003; MacKenzie et al., 2005).
This reflective specification may be appropriate in many instances, for example,
those when the interest is on a reflection or evaluation of WFB or work-family
imbalance (e.g., I’m satisfied with my personal WFB). For the construct of WFB
conflict, a reflective scale only specifies manifestations of one specific type of
conflict such as work-to-family conflicts (i.e., the demands of my work interfere
with my home and family life. My job produces strain that makes it difficult to ful-
fill family duties) or family-to-work conflicts (e.g., family-related strain interferes
with my abilities to perform job-related duties; the demands of my family interfere
with work related activities; cf. Netemeyer, Boles, & McMurrian, 1996).

(A) Formative latent variable (B) Reflective latent variable


model model

Y1 Y1

Work–family Work–family
Y2 Y2
balance balance

Y3 Y3

FIGURE 1. Latent construct models: (A) formative latent construct model and
(B) reflective latent construct model. The measurement error is not displayed.
394 The Journal of Psychology

Although the reflective approach is successfully used in most studies, there


are limitations that hinder the application of these scales in some settings. We
will discuss these limitations by the example of the WFB conflict approach. WFB
conflict measures are very specific, consisting of several homogeneous reflective
subscales. The subscales are sensitive to the directions of the conflict, meaning that
they differentiate between conflicts as a result of the work domain’s interfering
with the private domain (work-to-family conflict) and vice versa (family-to-work
conflict, Netemeyer et al., 1996; see also Frone, Russell, & Cooper, 1992, 1997).
Other scholars specified WFB scales regarding the contents of the conflict, such
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

as conflicts based on time (i.e., my work keeps me away from my family activities
more than I would like), conflicts based on strain (stress; i.e., due to stress at
home, I’m often preoccupied with family matters at work), and conflicts based on
behavior (i.e., the behaviours that work for me at home do not seem to be effective
at work; cf. Stephens & Sommer, 1996). When researchers aim to measure WFB
in terms of conflict, the combination of both direction and content results in six
reflective conflict scores (Carlson, Kacmar, & Williams, 2000).
Thus, the first limitation concerns the problem of aggregating one single
conflict score. Although this intensive measurement approach is needed to prove
the theoretical model of WFB, it is limited when the interest turns to economic
field scales that provide one single WFB score for each subject. In aggregation
of each conflict score to one mean score of WFB conflict, the information about
the origin of the conflict or the pattern of conflict types will be lost. For example,
the two variables (a) limited time spent with family due to work demands and
(b) work conflicts due to family demands are indicators of the WFB conflict
construct. When researchers combine these into one single WFB score, both types
of conflict constitute independent occurrences that may not be expected to covary.
Consequently, it is not possible to explain the meaning of average or sum scores,
because large family-to-work conflicts have a different effect than large work-
to-family conflicts. Moreover, internal consistency will suffer when items from
different conflict domains are modeled in a single scale (Tetrick & Buffardi, 2006).
The reflective idea is that all items are from the same content domain (meaning
flows from the latent construct to the indicator), and therefore latent constructs are
empirically defined as reflective in terms of their total variance among indicators.
In turn, items with low item–scale correlations should be dropped from a single
scale to enhance internal consistency.
Closely related, the second limitation relates to the large number of consis-
tent subscales that are necessary to capture the different facets of second-order
constructs (Tetrick & Buffardi, 2006). Although a reflective approach might ade-
quately define the theoretical construct of WFB, its demand for the employment
of multiple scales renders the measurement of WFB inefficient and lengthy in the
field. However, many psychological constructs are multidimensional like WFB
conflicts with different facets or subtypes of conflict. This leads to the fact that sin-
gle reflective scales are only partially capable of reflecting the multidimensionality
Ellwart & Konradt 395

of constructs. For the purpose of achieving high internal consistency, scale devel-
opers need to eliminate indicators that are only weakly correlated with other items
of the same construct (Churchill, 1979). However, formative scales are able to
model different independent dimensions within one scale.
A third statistical limitation is given, if researchers integrate indicators that
define the construct as formative but still adopt a reflective measurement approach.
Besides possible low internal consistencies, measurement model misspecification
can sometimes bias estimates of the structural relationships between constructs
and potentially undermine statistical conclusion validity (Law & Wong, 1999;
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

MacKenzie et al., 2005). However, Bellavia and Frone (2004) point out from
a theoretical perspective that some WFB conflict measures are ambiguous and
do not conceptually distinguish between consequences and causes of the latent
construct WFB. Taken into a statistical perspective, this lack of clarity may lead to
a misspecification of the measurement model that has been shown to substantially
deteriorate parameter estimations (e.g., Jarvis et al., 2003; Law, Wong, & Mobley,
1998).
Taken together, in the field of WFB and other domains of applied psychology,
classical reflective measures are widespread and important, but they also face some
limitations. The present article proposes an alternative, the idea of a formative
specification. The formative argument is that items may not always reflect the
underlying latent construct but sometimes can be combined to form the latent
variable (cf. Law & Wong, 1999; MacCallum & Browne, 1993). This makes it
possible to integrate different influences on WFB, such as conflicts, strain, and
time, into one scale (see Figure 1).

The Formative Measurement Model in Psychological Research

Researchers have paid little attention to formative measurement models of


latent variables in applied psychology. As Petter et al. (2007) noted, “this lack of
concern regarding the measurement model has led some researchers to assume
that all constructs should be treated alike regardless of whether a given construct is
formative or reflective” (p. 624). However, a growing body of research addresses
formative modeling and shows its potential advantage (Bollen & Lennox, 1991;
Coltman et al., 2008; Diamantopoulos, Riefler, & Roth, 2008; Diamantopoulos &
Winklhofer, 2001; Edwards & Bagozzi, 2000; MacKenzie et al., 2005; Podsakoff,
MacKenzie, Lee, & Podsakoff, 2003). In a formative model, the measured items
“jointly influence the latent construct, and meaning emanates from the measures
to the construct in the sense that the full meaning of the composite latent construct
is derived from its measures” (MacKenzie et al., 2005, p. 713). Thus, formative
indicators represent a set of heterogeneous causes of the latent construct (Jarvis
et al., 2003). Reviews of marketing research (Fassott, 2006; Jarvis et al., 2003), en-
gineering psychology (Petter et al., 2007), and organizational research (MacKen-
zie et al., 2005) strongly suggest that measurement models often are misspecified,
396 The Journal of Psychology

which can lead to Type I and Type II errors, and put into question the validity of the
findings of studies that failed to correctly model measurement relations. However,
Jarvis et al. (2003); Howell, Breivik, and Wilcox (2007); and others argue that, in
many ways, constructs are not necessarily either reflective or formative in nature
but can be measured formatively or reflectively. Thus, the aim of our research is not
to rule out the reflective conceptualizations of constructs but rather to complement
the existing research on reflective measures by showing the potential advantage
of a formative approach.
In sum, there are four criteria describing the characteristics of classical reflec-
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

tive measurement models and their differences from new formative specifications
(cf. Bollen & Lennox, 1991; Diamantopoulos & Winklhofer, 2001): (a) In reflec-
tive measurement models, the latent construct influences its indicators, whereas
in formative measurement models, the direction points from observed measures
to the construct; (b) classical reflective indicators of one construct are replace-
able or homogeneous, whereas single formative indicators are rather unique and
not necessarily interchangeable; (c) there must be covariation among reflective
indicators, attesting to their high degree of internal consistency as expressed in co-
efficients such as Cronbach’s alpha, but this restriction does not apply to formative
indicator models; and (d) as reflective indicators are thought to be manifestations
of a single latent construct, they should all have the same antecedents and conse-
quences, whereas formative indicators are not expected to share antecedents and
consequences (see also Jarvis et al., 2003).
What are the advantages of such a formative approach? In the following sec-
tion, these advantages are explained with regard to the example of WFB but can
easily be transferred to other domains of psychological research. The formative
modeling of the construct of WFB defines indicators that are responsible for or
form specific types of work- or family-related conflict or imbalance. Due to the
changed direction of the relationship between indicators and latent variable in a
formative model, the observed variables do not need to be correlated (Bollen &
Lennox, 1991). They could even be entirely uncorrelated, if the formative con-
struct is represented by mutually exclusive indicators (Jarvis et al., 2003). For
example, different causes of WFB, such as job strains or time-related indicators,
can be specified as possible WFB dimensions within a formative scale, without
the need for high correlations between them (Tetrick & Buffardi, 2006). Thus, an
imbalanced work–family relation could be indicated if some people only experi-
ence high time-related demands, although strain-related demands are not present
at this time.
In addition to providing an overall balance score, a formative scale offers
the possibility to model each indicator of the WFB construct separately and thus
to identify its most influential facets. Consequently, formative scales allow for a
modeling of the multidimensionality of the WFB construct without using higher
levels of abstraction or second-order constructs. As covariation among indicators
is not necessary in formative indicator models (Bollen & Lennox, 1991), capturing
Ellwart & Konradt 397

different dimensions of WFB in a single formative scale will not create psycho-
metric problems as it does in reflective measures. Thus, the modeling of WFB as a
formative construct can overcome some of the problems of reflective approaches
already stated.
Because formative indicators can pertain to unrelated facets of WFB, individ-
uals do not have to exhibit similar ratings for all measured indicators. Low WFB
scores of some people may be due to high work demands but manageable family
demands, whereas other individuals’ imbalance may be caused by the opposite
pattern. Although participants with differing score constellations may share a sim-
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

ilar average WFB score, single indicators can be used to divide respondents into
subgroups. Such groups can be identified based on their specific configuration of
critical indicators, for example, a combination of high work demands and low
family demands versus low work demands and high family demands. Compar-
isons based on subgroup characteristics can improve our understanding of WFB
as a phenomenon and help to identify its antecedents and consequences more
accurately.

Application of a Formative Measure: Data Collected


From Two Studies on WFB

Following the theoretical introduction of the formative measurement ap-


proach, the next sections explain the logic of formative scale construction and pro-
vide empirical evidence from two studies. In the formative measurement model,
the observable indicators are represented as source of the latent construct. Al-
though many variables are likely to contribute to WFB, in this example only three
of them were selected, those that showed the strongest impact on WFB: WFB
Indicator 1, the ability to manage work demands; WFB Indicator 2, the ability
to manage family demands; and WFB Indicator 3, time for family activities. Al-
though these three indicators are chosen to exemplarily introduce the formative
approach, empirical evidence about their impact derives from a large number of
studies from different theoretical traditions (e.g., Frone, 2003; Rothbard & Du-
mas, 2006). For example, the strain-related Indicators 1 and 2 are causes of several
WFB constructs such as role conflict (Anderson, Coffey, & Byerly, 2002; Frone,
2003; Fu & Shaffer, 2001; for an overview, see O’Driscoll, Brough, & Kalliath,
2006) and health-related behaviors (Allen, Herst, Bruck, & Sutton, 2000; Jones,
Kinman, & Payne, 2006). The third indicator identifies time as another major
antecedent of a balanced work–family interface. Time demands refer to an indi-
vidual’s inability to perform activities at home as a result of temporal load due
to work responsibilities, and vice versa (Greenhaus & Beutell, 1985). Empirical
research has clearly demonstrated the influence of time demands on WFB-related
variables (e.g., Grzywacz & Marks, 2000; Major, Klein, & Ehrhart, 2002; Sparks,
Cooper, Fried, & Shirom, 1997).
398 The Journal of Psychology

If these three indicators of WFB are specified as items in a formative measure-


ment model, the formative scale (new) should be positively related to the reflective
WFB (classical), indicating content validity. However, it is important to note that
each single indicator will predict WFB independently (positive indicator weights),
but as the formative approach assumes, the single indicators may not be as highly
correlated as researchers know from reflective scales. Thus,

Hypothesis 1a: The formative WFB (new) construct is positively related to re-
flective WFB constructs (classical).
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

Hypothesis 1b: The three indicators of the formative WFB construct (new) dis-
play a positive influence (indicator weights) when predicting reflective WFB
(classical).

Individuals’ outcomes of work–family imbalance with regard to work, family,


and stress (Allen et al., 2000; Jones, Kinman, et al., 2006) have been related
to individuals’ satisfaction. Numerous studies and several meta-analyses have
made evident that WFB-related variables are associated with satisfaction across
life domains (e.g., Allen et al., 2000; Carlson & Kacmar, 2000; Kim & Ling,
2001; Kosek & Ozeki, 1998; Thomas & Ganster, 1995). To demonstrate the
content validity of a new formative WFB construct, it will be used as predictor of
satisfaction. We propose that the mean scores of the formative WFB construct are
positively related to work, family, and life satisfaction. Consequently, we postulate

Hypothesis 2: Formative (new) and reflective (classical) WFB constructs will be


positively related to job satisfaction, family satisfaction, and life satisfaction.

A benefit of formative measures is the sensitivity of each single WFB indicator


depending on the specific type of satisfaction predicted. For example, explaining
job satisfaction through the reflective and formative WFB constructs, both scales
are able to predict this type of satisfaction. However, using the formative approach
allows conclusions about the impact of each single indicator on the predicted vari-
able (different indicator weights). For example, work-related items may emerge
as the strongest predictor of job satisfaction and allow drawing conclusions about
the causes of job satisfaction more clearly (discriminant validity). Thus, we expect
the single weights of the new formative WFB indicators to vary depending on the
type of satisfaction they are predicting, advertising the indicator with the strongest
influence.
The formative WFB construct allows revealing the impact of specific indi-
cators on the latent construct. As mentioned before, as indicators of a formative
measurement model are not necessarily correlated, individuals with a similar mean
WFB score are not expected to show a similar scoring pattern across the WFB
indicators. Individuals may share similar scores on the average WFB scale, but
different groups are characterized by a specific score configuration of formative
Ellwart & Konradt 399

WFB indicators, such as high (positive) values in some facets but low (negative)
values in others. With reflective scales, individuals with similar mean scores are
treated equally, because all the items are interchangeable and homogeneous. For-
mative scales are sensitive to discriminant patterns and allow the detection of
qualitatively distinct WFB profiles. Based on exploratory analyses, we propose
that there are distinct WFB subgroups based on specific characteristics regarding
each formative WFB indicator.

Hypothesis 3a: The new formative indicators allow for a differentiation of extreme
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

subgroups with a distinct WFB indicator pattern.

Furthermore, we expect that subgroups with similar average reflective WFB but
a different scoring pattern on formative indicators exhibit a distinct relation to
outcomes such as life satisfaction, because the influences of formative WFB
indicators are domain-specific (see also Hypothesis 2).

Hypothesis 3b: Subgroups with similar average reflective WFB (classical) but
dissimilar formative indicator patterns will differ in life satisfaction.

For example, we expect that abilities to meet family demands have a strong
influence on WFB. Thus, subgroups may be similar with respect to classical
reflective WFB, but they will show less life satisfaction, when they score low on
the formative indicator “abilities to meet family demands.”

STUDY 1

Method

Participants
A total of 734 employees completed an online survey on stress and strain in
job and family, with a final sample of 698 in the analyses.2 Of respondents, 57%
were female, and 43% were male, with an overall mean age of 39 years (SD = 9.02
years). Of all, 73% were living with a partner or spouse, and 47% of all individuals
in the sample had children living at home. The average job tenure was 14.7 years
(SD = 10.1 years), and the average family income was about €39,000 per year. Of
respondents, 63% held a management position, 15% were technicians, 17% were
clerks, and 5% were temporary employees.

Procedure
Following printed and radio recruitment advertisements, respondents regis-
tered online for participation in a survey on work–life balance and completed
the online survey (questionnaire). On completion, participants received feedback
about their personal WFB together with suggestions for improvement.
400 The Journal of Psychology

Measures and Indicators


With the exception of the classical reflective single-item WFB scale, all items
were answered on five-point Likert-type scales ranging from totally disagree (−2)
to totally agree (+2). Scales that originally existed in English were translated
into German following the forward–back procedure proposed by Brislin (1980) to
arrive at conceptual equivalence.
The formative WFB measure was a 3-item, multidimensional scale that mea-
sured work-related demands (I can meet all my work-related demands), family-
related demands (I can meet all my private/family demands), and time-based
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

constraints (Besides work there is sufficient time for family/private activities) fol-
lowing definitions of work–life balance (cf. Jones, Burke, et al., 2006).
Reflective WFB constructs (classical) were measured with two different scales,
a five-item reflective scale and a single-item reflective scale. The five-item scale
assessed WFB outcomes related to affect, strain, and health and was from Brett
and Stroh (2003). Coefficient alpha was .85. The single-item WFB scale was taken
from Berg, Kalleberg, and Appelbaum (2003), consisting of a single global item
evaluating the fit between family and work domains (“All in all, I am satisfied
with the balance I have achieved between my work and family life”). Follow-
ing suggestions of Berg et al. (2003), a 7-point Kunin faces answer scale was
used (cf. Kunin, 1955). Both types of reflective scales were applied in Study 1
to estimate construct validity of the single-item scale. Although organizational
research showed single-item scales to provide reliable and valid measures (e.g.,
Nagy, 2002; Wanous, Reichers, & Hudy, 1997), the application of single-item
and 5-item scales in Study 1 supports the validity of the single-item scale in
Study 2.
Job satisfaction was measured with a three-item scale by Semmer (1984). A
sample item is “I am satisfied with my present work situation.” Coefficient alpha
was .83.
Family satisfaction was measured and coded by the same items as job satis-
faction, but the word “job” was replaced by “family” (e.g., I am satisfied with my
present family situation). Coefficient alpha was .87.
Life satisfaction was measured with three items from the Satisfaction with
Life Scale (SWLS; Diener, Emmons, Larsen, & Griffin, 1985; e.g., In most ways
my life is close to ideal). To recruit a large sample of participants, we aimed for
very short scales by shortening the original five-item scale. The original five-item
SWLS had an internal consistency of .87 (Diener et al., 1985), and the abbreviated
three-item version used in this study reached an alpha coefficient of .82.

Analysis
We analyzed the relationship between formative and reflective variables with
Partial Least Squares (PLS) path analysis, using the software SmartPLS 2.0
(Ringle, Wende, & Will, 2005). PLS was chosen because it allows the simul-
taneous inclusion of both reflective and formative measures (Barclay, Higgins, &
Ellwart & Konradt 401

Thompson, 1995; Tenenhaus et al., 2004). PLS does not make assumptions about
(a) data distributions to estimate model parameters, (b) observation independence,
or (c) variable metrics (Barclay et al., 1995). The analysis followed the two steps
suggested by Anderson and Gerbig (1988). In the first step, the factor weights,
composite reliabilities, and average variances extracted were estimated to achieve
convergent validity and discriminant validity of the measurement model. As sug-
gested by Fornell and Larcker (1981), the convergent and discriminant validities
of the reflective measures were calculated using the Average Variance Extracted
(i.e., the average variance shared between a construct and its measures), which
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

should be greater than the variance shared between the construct and other con-
structs. In the second step, the structural model was estimated to test the postulated
hypotheses. PLS generates estimates of standardized regression coefficients (beta
values) for the model’s paths, which are then used to measure the relationship be-
tween latent variables. The indicator weights determine the extent to which each
indicator contributes to the formation of the constructs. To test significance of
the model paths and indicator weights, t-statistics are estimated using a bootstrap
resampling procedure (Efron & Tibshirani, 1986).
Before analyses, data were screened for multivariate outliers, normality of the
distributions, and common method bias.
To differentiate subgroups with a distinct formative WFB indicator pattern,
participants were clustered according to ratings of the three formative items. Likert
ratings of “−1” and “−2” were classified as low, and Likert-ratings of “+1” and
“+2” were classified as high. Participants shared the same mean when two of the
three items had a negative rating (low) and one item had a positive rating (high)
(cf. Groups 1, 2, and 3). Similarly, participants with only one negative rating (low)
and two positive ratings (high) shared a similar WFB mean (cf. Groups 4, 5,
and 6).

Results

By using Mahalanobis distance, 6 cases (about 0.9%) were identified as mul-


tivariate outliers and deleted from further analyses. Scores for all variables were
reasonably distributed with regard to skewness and kurtosis (skewness ranged from
−0.45 to 0.05, kurtosis varied between −0.83 and 0.23). Further analyses indi-
cated no influence of multicollinearity as variable inflation factors (VIF) remained
below 1.11, and condition indices remained below 13.60 (cf. Hair, Anderson,
Tatham, & Black, 1998).
To control for the potential impact of common method bias, recommenda-
tions by Podsakoff et al. (2003) were followed. The methodological separation of
predictor and criterion was accomplished by using different response formats for
formative and reflective scales (Likert) and the reflective single-item scale (Kunin).
However, as noted by Podsakoff et al. (2003), formative measures complicate the
402 The Journal of Psychology

statistical treatment of common method bias. Thus, procedures that obtain an es-
timate of the relationship between the constructs and measures partialling out the
effect of method bias cannot be transferred to formative models (for a detailed
explanation, see Bollen & Lennox, 1991; Podsakoff et al., 2003). Nevertheless,
as formative indicators are correlated in this sample, Harman’s single-factor test
(Andersson & Bateman, 1997) and partial correlation procedures (Dooley & Fryx-
ell, 1999) were applied and yielded no indication of common method bias.
Mean scores and standard deviation for all measures are reported in Table
1. Standard indexes for evaluating classical reflective measurement models with
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

PLS are composite reliability (C.Rel.), values for Cronbach’s alpha (CA), and
average variance extracted (AVE) (see Table 1). Results revealed adequate relia-
bility measures for the reflective scales (Nunnally & Bernstein, 1994). All items
showed higher loadings on their respective construct than on any of the other
reflective constructs. All loadings were found to exceed a threshold of 0.71 as
recommended by Barclay et al. (1995), which indicates adequate discriminant
validity (Chin, 1998). As traditional measures of validity are not appropriate for
formative constructs (Chin, 1998), the validity of the formative constructs was
estimated by an evaluation of the size and significance of their indicator weights
(cf. Tenenhaus et al., 2004).
Hypothesis 1a predicted that the new formative WFB is positively related
to reflective WFB constructs. The results of the bivariate correlations between
formative and reflective WFB (r ≥ .60, see Table 1) as well as the results of the
path analyses (Table 2) indicate content validity of the formative WFB construct.
As shown in Table 2, there are significant path coefficients between new forma-
tive WFB and both classical reflective WFB constructs. Moreover, determination
coefficient values (R2) indicate that the formative scale explains 42% and 37%
of the variance in the dependent variables. In support of Hypothesis 1b, all three
indicators of the formative WFB construct yielded significant and positive weights
(b ≥ .30) when predicting single-item and multi-item reflective WFB (Table 2).
This means that each formative indicator contributes to the formation of the con-
structs, but the impact of each indicator varies. The largest weights (impact) are
associated with the time-related formative indicator (b > .70). Thus, Indicator
3 (time for family) has the strongest influence on WFB compared to work and
family demands.
Hypothesis 2, which stated that the new formative WFB construct predicts
job, family, and life satisfaction, was strongly supported. Path coefficients range
from .42 to .45 and explain 18% to 21% of the variance in the satisfaction measures
(Table 2). Similar results were obtained from analyses of classical reflective WFB
constructs predicting job satisfaction (single-item: b = .50, p < .001, R2 = .25;
multi-item: b = .58, p < .001, R2 = .33) and life satisfaction (single-item: b =
.53, p < .001, R2 = .28; multi-item: b = .47, p < .001, R2 = 22). Predicting family
satisfaction, reflective measures displayed lower path coefficients and explained
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

TABLE 1. Descriptives, Correlations, and Reliabilities of Work–Family Balance Measures and Satisfaction in Study 1 and
Study 2
Study 1 Study 2
Measures M (SD) M (SD) AVE CA C.Rel. 1 2 3 4 5 6 7 8 9
∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗
1. Indicator 1 (work 0.56(0.78) 0.59(0.89) — .27 .27 .70 .36 .41 .06 .34∗∗
demands)
2. Indicator 2 (family 0.39(0.80) 0.49(0.85) .25∗ — .31∗∗ .69∗∗ .37∗∗ .12∗∗ .48∗∗ .42∗∗
demands)
3. Indicator 3 (time for −0.21(1.05) −0.08(1.09) .27∗∗ .21∗∗ — .78∗∗ .78∗∗ .53∗∗ .27∗∗ .32∗∗
family)
4. Formative WFB 0.25(0.62) 0.34(0.68) .68∗∗ .65∗∗ .77∗∗ — .59∗∗ .37∗∗ .30∗∗ .49
5. Reflective WFB 4.10(1.49) 4.10(1.44) .36∗∗ .35∗∗ .53∗∗ .60∗∗ — .50∗∗ .34∗∗ .61∗∗
(single item)
6. Reflective WFB 2.94(0.87) — .63/— .85/— .89/— .38∗∗ .34∗∗ .56∗∗ .62∗∗ .73∗∗ —
(multi-item)
∗∗ ∗∗ ∗∗ ∗∗ ∗∗
7. Job satisfaction 2.94(0.86) 2.89(0.90) .74/.71 .83/.80 .90/.88 .40 .15 .32 .41 .52 .59∗∗ — .10∗∗ .47∗∗
8. Family satisfaction 3.58(0.93) 3.58(0.90) .79/.78 .87/.85 .92/.91 .04 .41∗∗ .07† .23∗∗ .26∗∗ .19∗∗ .09∗ — .55∗∗
9. Life satisfaction 3.16(0.80) 3.14(0.86) .74/.74 .82/.82 .89/.89 .23∗∗ .36∗∗ .30∗∗ .42∗∗ .52∗∗ .46∗∗ .42∗∗ .55∗∗ —

Note. Correlations from Study 1 are presented below the diagonal. Correlations from Study 2 are presented above the diagonal.
Average variance explained (AVE), Cronbach’s alpha (CA), and composite reliability (C.Rel.) for reflective measures are presented for
Study 1/Study 2 (Study 1/Study 2). Formative indicator work demands = Ability to meet work demands;
Formative indicator family demands = Ability to meet family demands; Formative indicator time for family = Time for family/private activities.
Study 1 N = 698. Study 2 N = 2,210.
∗ p < .05. ∗∗ p < .01.
Ellwart & Konradt
403
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

404
TABLE 2. Formative Indicator Weights and Path Coefficients of PLS Analyses in Study 1 and Study 2
Study 1 Study 2

Path coefficients Formative indicator Path coefficients Formative indicator


Hypothesis and variable (t-values) R2 weights (t-values) (t-values) R2 weights (t-values)

Hypotheses 1a and 1b
Formative WFB → Reflective WFB (multi item) .65∗∗(28.24) .42
Indicator 1 (work) → Formative WFB .30∗∗ (5.88)
Indicator 2 (family) → Formative WFB .35∗∗ (7.19)
The Journal of Psychology

Indicator 3 (time) → Formative WFB .72∗∗ (17.78)


Formative WFB → Reflective WFB (single item) .61∗∗ (26.05) .37 .59∗∗ (34.25) .35
Indicator 1 (work) → Formative WFB .33∗∗ (5.49) .33∗∗ (8.02)
Indicator 2 (family) → Formative WFB .36∗∗ (6.84) .32∗∗ (8.02)
Indicator 3 (time) → Formative WFB .70∗∗ (14.78) .69∗∗ (20.96)
Hypothesis 2
Formative WFB → Job satisfaction .45∗∗ (14.54) .21 .44∗∗ (18.86) .19
Indicator 1 (work) → Formative WFB .73∗∗ (10.39) .81∗∗ (20.30)
Indicator 2 (family) → Formative WFB .02 (0.27) −.09† (1.52)
Indicator 3 (time) → Formative WFB .50∗∗ (6.15) .44∗∗ (7.60)
Formative WFB → Family satisfaction .42∗∗ (13.30) .18 .50∗∗ (24.58) .24
Indicator 1 (work) → Formative WFB −.14† (1.46) −.17∗∗ (3.67)
Indicator 2 (family) → Formative WFB 1.02∗∗ (52.49) 1.03∗∗ (62.96)
Indicator 3 (time) → Formative WFB .01 (0.16) .03 (0.56)
Formative WFB → Life satisfaction .45∗∗ (15.70) .20 .51∗∗ (25.26) .26
Indicator 1 (work) → Formative WFB .20∗∗ (2.73) .40∗∗ (9.01)
Indicator 2 (family) → Formative WFB .69∗∗ (10.36) .62∗∗ (14.93)
Indicator 3 (time) → Formative WFB .48∗∗ (6.58) .33∗∗ (7.32)

Note. Formative Indicator 1 (work) = Ability to meet work demands; Formative Indicator 2 (family) = Ability to meet family demands; Formative
Indicator 3 (time) = Time for family/private activities; Values of significance: Study 1 N = 698. Study 2 N = 2210. R2 = explained variance in the
dependent variable (DV). WFB = work–family balance.
†p < .10. ∗ p < .05. ∗∗ p < .01.
Ellwart & Konradt 405

less variance (single-item: b = .27, p < .001, R2 = .07; multi-item: b = .22, p <
.001, R2 = .05).
Moreover, the varying impact of formative indicators on the relationship be-
tween WFB and different types of satisfaction can be estimated. Results show
that Indicator 1 (ability to meet work demands) emerged as the strongest predictor
of job satisfaction (b = .73) followed by Indicator 3 (time for family activity).
Indicator 2 (family demands) was not predictive for work satisfaction. However,
predicting family satisfaction, Indicator 2 (ability to meet family demands) pre-
dicted family satisfaction best (b = 1.02), whereas work demands and time for
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

family yielded small or no impact. Regarding life satisfaction, Indicator 2 (family


demands, b = .69) and Indicator 3 (time for family, b = .48) were found to be
the strongest indicators. Not only can a formative measurement approach (new)
predict job, family, and life satisfaction, but also the weights of each formative in-
dicator allow researchers to identify those variables that have the strongest impact
on the dependent variable.
Hypotheses 3a and 3b could not be tested because the subgroups did not
reach an adequate sample size for statistical testing and therefore will be tested in
Study 2.
Discussion

Results show that formative and reflective constructs are related and predic-
tive in terms of satisfaction. However, the formative approach allows modeling
different uncorrelated indicators as items in one scale. In turn, the new formative
measurement model can estimate the contribution of each single indicator on the
formative construct when predicting different types of satisfaction. The benefit of
this differentiated result compared to the global estimates from reflective measures
lies in the predictive potential. Researchers and practitioners are able to identify
highly potential variables to change WFB in order to improve dependent variables
such as satisfaction.
Although we find promising results using the formative measurement model,
our data do not illuminate whether WFB patterns will impact individuals’ well-
being differentially. For instance, people with high work demands might share the
same mean WFB score with people with high family demands but will nevertheless
differ regarding their well-being. Thus, to explore this proposed relationship and
to assess the generalizability of the results of Study 1, we conducted Study 2 using
a larger sample.

STUDY 2

To augment the validity of the results of Study 1 and increase their gener-
alizability, we conducted a replication study within a different sample. We used
the same scales and followed the statistical procedures as reported in Study 1.
406 The Journal of Psychology

However, divergent from Study 1 and for space reasons, the reflective multi-item
scale on WFB was omitted.

Method

Participants
Similar to the first study, data were collected from 2.378 employees who
completed an online questionnaire. The final number of participants included for
analysis consisted of 2,210 German participants.3 Participants were on average 39
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

years old (SD = 9.06 years), and with 73% of respondents being male, women were
underrepresented in the sample. In all, 80% were living with a partner or spouse,
and 47% of all participants had children living at home. Of the participants, 62%
had been holding the same job for 3–10 years, with an average family income of
about 53,000 Euro per year. Of respondents, 62% occupied a management position,
15% worked as technicians, 17% worked as clerks, and 6% were temporary
employees.

Procedures
Visitors of the corporate website of a business magazine were given the
opportunity to perform an anonymous online assessment of their individual WFB.
Immediately after completing the questionnaire, participants were provided with
feedback about their individual WFB and personal satisfaction, together with
suggestions for improvement.

Results

Preliminary Tests
Composite reliabilities, Cronbach’s alphas, and average variance extracted
(AVE) are reported in Table 1 and indicate sufficient discriminant validity, as
well as good reliability and validity statistics for the reflective scales. By using
Mahalanobis distance, 10 cases (about 0.5%) were identified as multivariate out-
liers and excluded from further analyses. Normality was evaluated by checking
the skewness and kurtosis coefficients, which indicated reasonable distributions of
the scales. There were no indications for multicollinearity, as VIF values remained
below 1.17, condition indices remained below 2.36 (cf. Belsley, Kuh, & Welsch,
1980; Hair et al., 1998), and no common method bias could be discovered.
Consistent with Hypothesis 1a, new formative and classical reflective WFB
constructs were found to be correlated (r = .58), constituting a replication of the
results of Study 1 (see Table 1 and Table 2). Consistent with this hypothesis, the
path coefficient for predicting the reflective single-item WFB equals .59 (Study 1:
b = .61). As proposed in Hypothesis 1b, the formative WFB scale yielded positive
weights for predicting reflective WFB, thus arguing for the predictive potential of
the three indicators. The formative indicator weights and path coefficients were
Ellwart & Konradt 407

found to be of size comparable to those in Study 1. Likewise, Indicators 1 and


2 (abilities to meet work and family-related demands) yielded weights similar in
the prediction of WFB to those of Study 1. The largest weights again derive from
Indicator 3 (time for family; b > .69).
In accordance with Hypothesis 2, the path coefficients indicate that the new
formative WFB construct predicts work, family, and life satisfaction, with coef-
ficients ranging from .44 to .51. Similar results were obtained from analyses of
the reflective WFB construct predicting job satisfaction (single-item: b = .51,
p < .001, R2 = .26) and life satisfaction (single-item: b = .61, p < .001, R2 = .37).
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

Again, when predicting family satisfaction, the reflective measure displayed lower
path coefficients and explained less variance (single item: b = .34, p < .001, R2 =
.12). The evaluation of individual indicator weights revealed results comparable
to those of Study 1, providing further support to Hypothesis 2. Depending on the
type of satisfaction predicted in the model, the differing indicator weights in Table
2 tell about the specific influence of each of the formative indicators.
As proposed in Hypothesis 3a, new formative constructs allow for a differ-
entiation between extreme subgroups with a distinct formative WFB indicator
pattern but similar average reflective WFB scores. Thus, distinct WFB subgroups
with specific formative indicator configurations of high (positive) or low (nega-
tive) values of each WFB dimension were selected (see Table 3). Most interesting
are two sets of subgroups, one comprising Subgroups 2, 3, and 4 (one high and
two low formative indicators), and the other one including Subgroups 5, 6, and
7 (one high and two low formative indicators). Sets of subgroups differ with
regard to their new formative indicator pattern, but are comparable in terms of
their average reflective and formative WFB scores. Congruent with Hypothesis
3a, subgroups with different formative indicator patterns exhibited similar aver-
age reflective WFB (classical). These findings applied to Subgroups 2, 3, and 4,
which are characterized by two negative/low WFB indicators—reflective WFB:
F(2, 155) = 1.63, ns; formative WFB: F(2, 18) = 2.01, ns—as well to Subgroups
5, 6, and 7, characterized by two positive/high WFB indicators—reflective WFB:
F(2, 258) = 2.14, ns; formative WFB: F(2, 317) = 2.12, ns. In summary, partici-
pants exhibiting a different formative indicator pattern, but a comparable number
of negative and positive WFB indicators, are similar with respect to their average
reflective and formative WFB score. Applying a reflective approach may lead to
the assumption that participants with similar mean scores share similar character-
istics in WFB and in the prediction of satisfaction. However, the new formative
approach would assume that participants with similar mean scores may not share
similar characteristics in WFB and differentiate in the prediction of satisfaction.
Hypothesis 3b examined whether groups with comparable reflective WFB
scores would differ in general life satisfaction. Separate ANOVAs were calculated
for Subgroup 2, 3, and 4 as well as for Subgroup 5, 6, and 7. Among groups with
two negative/low WFB dimension (Subgroups 2, 3, and 4), life satisfaction did
not significantly differ between the three subgroups, F(2, 176) = 1.58, ns. In these
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

408

TABLE 3. Work–Family Balance (WFB) Subgroups With Mean Scores and Standard Deviations From Study 2
The Journal of Psychology

Group 1: Group 2: Group 3: Group 4: Group 5: Group 6: Group 7: Group 8:


Low work Low work Low work High work Low work High work High work High work
low family low family high family low family high family low family high family high family
low time high time low time low time high time high time low time high time
N = 79 N = 22 N = 63 N = 92 N = 27 N = 42 N = 247 N = 507

Variable M SD M SD M SD M SD M SD M SD M SD M SD

Formative WFB −1.17 0.24 −0.33 0.00 −0.37 0.18 −0.44 0.20 0.47 0.23 0.40 0.13 0.37 0.21 1.14 0.25
Reflective WFB 2.44 1.04 3.28 0.83 2.91 1.20 3.22 1.02 4.19 1.44 3.87 1.32 3.70 1.32 5.26 1.11
Life Satisfaction 2.23 0.71 2.41 1.01 2.75 0.79 2.75 0.81 3.02 0.63 2.69 0.76 3.22 0.75 3.65 0.70

Note. Low = negative indicator score (−1 or −2 in the rating). High = positive indicator score (+1 or +2 in the rating). N = participants per subgroup.
Ellwart & Konradt 409

groups life satisfaction scores are rather low (Group 2 = 2.41, Group 3 = 2.75, and
Group 4 = 2.75) as one would expect if two WFB indicators have a low/negative
rating. Among groups with only one negative/low WFB dimension (Subgroups
5, 6, and 7), life satisfaction significantly differed between these subgroups, F(2,
315) = 9.44, p < .001. Comparing single group means, Group 6 (low Indica-
tor 2, ability to meet family demands) showed the lowest life satisfaction score
(M = 2.69, SD = 0.76). Although Group 6 shares a similar classical reflective
WFB, it differed significantly from Group 5, t(67) = 2.10, p < .05, as well as from
Group 7, t(287) = −4.22, p < .001, regarding life satisfaction. However, Group 5
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

shows a formative pattern with a low Indicator 1 (ability to meet work demands),
and Group 7 shows a formative pattern with a low Indicator 3 (time for family).
Taken together, although the groups have similar average WFB scores, they differ
significantly with regard to life satisfaction. For participants with a disability to
meet family demands (Group 6), life satisfaction was lower compared to groups
that are not able to manage work demands or have limited time. Thus, Hypothesis
3b was partially supported.

Discussion

In sum, the results from Experiement 1 could be replicated in another sam-


ple. Again, classical reflective and new formative constructs were related and
were shown to predict satisfaction. The formative measurement model enables
researchers to differ between causes, because indicator weights differ depending
on the DV. For example, the ability to meet work and family demands seems to be
the most important indicator in predicting work and family satisfaction.
Since formative indicators do not need to be correlated, participants could
differ regarding their indicator pattern. As shown in Study 2, there are subgroups
who share a similar reflective WFB mean score with different indicators responsi-
ble for it. This formative analysis allows a deeper understanding about the causes
and processes behind latent variables. For example, for life satisfaction it seems
most relevant to manage the family demands and work demands. Even when hav-
ing little time to join family activities, life satisfaction will not suffer as much as
under conditions in which duties are not done. However, the formative approach
is not suited to pick out a solo indicator item as a stand-alone single-item measure
to estimate the “ability to meet work demands,” for example. The idea of the
formative approach is to estimate the different effects of numerous uncorrelated
indicators on the latent construct (MacKenzie et al., 2005; Podsakoff et al., 2003),
and therefore all indicators “form” the construct and cannot be separated.

GENERAL DISCUSSION

When building scales it is important to understand whether the latent construct


is measured reflectively or formatively. In most cases, both types of measurement
410 The Journal of Psychology

are possible, but formative measures showed several benefits. Accordingly, this
article illustrates the new formative measurement process in the field of applied
psychology. First, we established a formative measure for WFB that showed good
psychometric properties in Study 1. Next we identified different subpopulations
with a distinct indicator characteristic in Study 2 showing the potential advan-
tage of a formative approach. Last, we demonstrated invariance of results across
populations.
Previous research on WFB has exclusively focused on a reflective latent WFB
construct that affects manifest indicators like conflict or individuals’ satisfaction
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

(e.g., Eby et al., 2005; Jones et al., 2006). The aim of this research was to propose
a formative conceptualization of WFB that focuses on causes rather than on
consequences of balanced or imbalanced work–family domains and to demonstrate
psychometrical appropriateness, which is supposed to complement the existing
research on reflective measures.
We delineated WFB as a formative latent construct with three indicators,
including the ability to manage work demands, the ability to manage family
demands, and time for family activities as salient components of WFB (cf. Jones,
Burke, et al., 2006). Findings in two studies indicate that this formative measure
of WFB was positively related to reflective single-item and multi-item measures,
demonstrating the content validity of the formative WFB construct. Moreover,
and consistent with research on role conflict (O’Driscoll et al., 2006) and health-
related behaviors (Allen et al., 2000; Jones, Kinman, et al., 2006), each of the
three indicators explained a significant amount in three types of satisfaction.
These findings provide evidence of the robustness of this conceptualization across
different samples and contexts.
The formative measurement model allows estimating the specific influence
of single causes on the WFB variable. In both samples, the ability to meet work-
related demands (Indicator 1) and family demands (Indicator 2) yielded similar
weights. The largest weights derive from time for family (Indicator 3), which indi-
cates that sufficient time allocated to family activities and private roles contributes
most strongly to a balanced work–family interface. Together, findings in two large
and heterogeneous samples provide a robust test of the results.
Whereas formative measures predict the different types of satisfaction equally
well, reflective scales examined were less suitable for predicting family satisfac-
tion. This might be due to the fact that a major focus in reflective scales is on
work-related aspects of WFB, which leaves family-related causes underevaluated.
We expected to find discriminant validity, and even though the three path coeffi-
cients in the formative models were of similar size, each indicator contributed to
the prediction of job, family, and life satisfaction in distinguishable ways. Work-
related demands predicted job satisfaction best, whereas family demands emerged
as the most salient predictor of family satisfaction. Both, the ability to meet family
demands and the availability of time besides work contributed most strongly to
Ellwart & Konradt 411

life satisfaction. Its ability to provide such information about specific formative
indicators is one of the major benefits of the formative approach.
Consistent with our hypotheses, lastly, we identified subgroups of individuals
who show distinct WFB indicator patterns, as indicated by formative measures,
but similar average WFB scores using reflective measures. These findings sug-
gest that subgroups with similar means as indicated by reflective measures may
be very different in terms of their causes for WFB. Moreover, groups with sim-
ilar WFB means differ with regard to life satisfaction. Specifically, if there are
no abilities to meet family demands, life satisfaction is lower compared to in-
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

dividuals with the same average WFB score but deficits in another work or life
domain indicator. This indicates that using reflective measures might cover up
individual differences within the population of interest. Using formative measures
provides the opportunity to not only examine the overall level of WFB but also
distinguish individuals with a specific underlying indicator pattern within a given
level.

Limitations of the Study


As with any research, these studies have a number of limitations but also
suggest directions for future research. The first potential limitation of this research
is that using self-reports may have led to the inflation of our results due to common
method variance (Podsakoff et al., 2003). While common method variance cannot
be fully ruled out, research has shown that common methods bias does not auto-
matically invalidate theoretical interpretations and substantive conclusions (Doty
& Glick, 1998; Spector, 2006). Furthermore, the differential pattern of relation-
ship between our measures lends support to the assumption that common method
variance is not a major limitation of this study.
The second limitation of this research is that data were derived from Web-
based surveys. Hence, there might be a concern due to biased sample characteristics
and distorted measurement equivalence (Kraut et al., 2004; Simsek & Veiga, 2001;
Thompson, Surface, Martin, & Sanders 2003). Research showed that samples of
internet studies usually are demographically similar to samples in offline studies
(e.g., Azar, 2000; Buchanan, 2002) and are not biased toward maladjusted, socially
isolated, or depressed individuals (Kraut et al., 1998). Furthermore, correlations in
both studies were consistent with theoretically and empirically expected relations
found in paper-and-pencil studies (Eby et al., 2005; Frone, 2003; Jones, Kinman,
et al., 2006), thus, indicating internal validity of measures in both studies. How-
ever, there are also concerns about the representativeness of nonrandom samples
and the generalizability of the findings in other settings (Azar, 2000; Buchanan,
2000, 2002; Kraut et al., 2004). With regard to measurement equivalence, there is a
growing body of research that suggests that online instruments do not improperly
influence the psychometric properties of paper-and-pencil measures (Cole, Be-
deian, & Feild, 2006; Potosky & Bobko, 1997). Nevertheless, future research
412 The Journal of Psychology

should demonstrate the generalizability of our results using paper-and-pencil


methods.
Third, a potential limitation is that we used a cross-sectional design in both
studies. Although this is a common study design in measurement method research
(e.g., MacKenzie et al., 2005; Nunnally & Bernstein, 1994), we acknowledge that
it precludes causal inferences of the relations proposed by our model and should,
thus, also be tested in longitudinal field and experimental studies.
The fourth limitation of the research refers to the formative measurement
approach. The formative conceptualization of the construct we proposed does not
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

include an exhaustive set of indicators that cause WFB. By following the sugges-
tions of several researchers (e.g., Jarvis et al., 2003; MacKenzie et al., 2005), we
identified a salient, generalizable, and parsimonious group of indicators that have
been strongly associated with work–family issues by related research (e.g., Allen
et al., 2000; Kossek & Ozeki, 1998; Tetrick & Buffardi, 2006). However, drawing
on various other theories, models, and related research, further indicators remain
to be considered, including demands from internal and external sources (Carlson
& Frone, 2003), facilitative variables that positively influence WFB (Grzywacz
& Marks, 2000), and levels of involvement in family and career (Greenhaus,
Parasuraman, & Collins, 2001). Future research should continue to assess the
generalizability of the proposed measures across contexts, as well as the impact
of added components.

Benefits of Formative Measures for Research and Application


Research will benefit greatly from guidelines for developing, modeling, and
evaluating constructs with formative indicators (for details, see Jarvis et al., 2003;
MacKenzie et al., 2005). However, the formative approach cannot replace reflective
models. In particular, a formative model is evaluated by estimating its relationship
with reflective constructs. Thus, to achieve identification for formative measures,
appropriate models need to include paths between the latent formative construct
(new) and valid reflective constructs (classical).
The proposed formative measurement approach contributes to psychological
assessment by offering valuable research implications in many fields of applied
psychology. In cases where researchers and practitioners want to know more
about influences that cause the characteristics of a variable, formative measures
are useful alternatives. For example, applications of formative scales are of great
benefit in measures of work satisfaction, absenteeism, leadership styles, learning
approaches, and group behaviors. In all these cases, the construct may be modeled
by indicators (items), which are not necessarily correlated but which are compos-
ites of the construct. Let us take work satisfaction, for example. The prominent
job descriptive index by Smith, Kendall, and Hulin (1969) distinguishes between
five different facets of work satisfaction (e.g., satisfaction with the leader, with
colleagues, with the task, with work conditions). Besides reflective scales for each
facet of satisfaction, there are also validated single-item measures (Smith et al.,
Ellwart & Konradt 413

1969). However, the different facets cannot be included into one single scale of
general satisfaction in classical reflective models. This is because they may not be
correlated (indicated by low Cronbach’s alpha) due to the fact that some workers
may be satisfied with their bosses but not with their colleagues. In many cases,
the multiple measurement of each facet may be useful, but a formative approach
could be beneficial by adding new information for researchers and practitioners.
The formative modeling of the five single items will yield information about the
relative impact of each single indicator in explaining overall work satisfaction.
Moreover, the integration of single items into a formative model avoids some of
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

the problems with reflective single-item approaches (e.g., Nagy, 2002). This will
lead to a better understanding of the development of work satisfaction as well as
telling (informing) practitioners about the variables, which should be changed to
achieve the highest impact on individual satisfaction.
Other fields are, for example, training and learning. In order to extract the
influence of different (uncorrelated) learning strategies on learning transfer, Kon-
radt et al. (2008) used a formative measurement model. Overall, they identified
specific cognitive and metacognitive learning strategies that predicted learning
transfer that can be used to derive specific training and interventions. Moreover,
by means of the indicator weights that reflect the relative amount of impact on
criteria, the significance of interventions can be deduced and prioritized.

Conclusion

As suggested by Law and Wong (1999), MacKenzie et al. (2005), and other
scholars, a formative conceptualization is appropriate for many constructs in orga-
nizational and behavioral research. We propose a formative measurement model
and provide theoretical and empirical support for the conceptualization from two
large samples. A formative approach allows for modeling various factors that are
associated with balance and imbalance, which preserves the multidimensional
character of the pertinent construct. In addition, a formative approach enables re-
searchers to identify indicators as sources for possible imbalances. Moreover, these
indicators allow examining whether distinct combinations of critical antecedents
will lead to particular problems and imbalances. Thus, formative measures could
complement existing reflective measures by integrating antecedents and conse-
quences into a single model.

NOTES
1. Throughout the text, we refer to the reflective approach as “(classical)” and to the
formative approach as “(new)” to provide the reader a mnemonic device helping to connect
the term and the method.
2. Among 734 registered German participants who completed an online survey, 36
participants were excluded because either 90% of their scores consisted of extreme values
and/or their questionnaires were incomplete with more than 30% of responses missing.
414 The Journal of Psychology

3. Among 2,378 registered German participants who completed an online survey, 168
participants were excluded because either 90% of their scores consisted of extreme values
and/or their questionnaires were incomplete with more than 30% of responses missing.

AUTHOR NOTES
Thomas Ellwart is a professor at the University of Trier, Germany. His current
research interests are work-group diversity, team cognition, and work–life balance. Udo
Konradt is a professor at the University of Kiel, Germany. His research interests are virtual
teams, performance and satisfaction, and work–life balance.
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

REFERENCES
Allen, T. D., Herst, D. E. L., Bruck, C. S., & Sutton, M. (2000). Consequences associ-
ated with work-to-family conflict: A review and agenda for future research. Journal of
Occupational Health Psychology, 5, 278–308.
Anderson, J. C., & Gerbing, D. W. (1988). Structural equation modelling in practice: A
review and recommended two-step approach. Psychological Bulletin, 103, 411–423.
Andersson, L. M., & Bateman, T. S. (1997). Cynicism in the workplace: Some causes and
effects. Journal of Organizational Behavior, 18, 449–469.
Anderson, S. E., Coffey, B. S., & Byerly, R. T. (2002). Formal organizational initiatives and
informal workplace practices: Links to work-family conflict and job-relates outcomes.
Journal of Management, 28, 787–810.
Azar, B. (2000). A web of research: They’re fun, they’re fast, and they save money, but do
Web experiments yield quality results? Monitor on Psychology, 31, 42–47.
Barclay, D., Higgins, C., & Thompson, R. (1995). The partial least squares (PLS) approach
to causal modeling: Personal computer adoption and use as an illustration. Technology
Studies, 2, 285–309.
Barnett, R. C. (1998). Toward a review and reconceptualization of the work/family literature.
Genetic, Social and General Psychology Monographs, 124, 125–182.
Bellavia, G., & Frone, M. R. (2004). Work-family conflict. In J. Barling, E. K. Kelloway, &
M. R. Frone (Eds.), Handbook of work stress (pp. 113–148). Thousand Oaks, CA: Sage.
Belsley, D. A., Kuh, E., & Welsch, R. E. (1980). Regression diagnostics: Identifying
influential data and sources of collinearity. New York, NY: Wiley.
Berg, P., Kalleberg, A. L., & Appelbaum, E. (2003). Balancing work and family: The role
of high-commitment environments. Industrial Relations, 42, 168–188.
Bollen, K., & Lennox, R. (1991). Conventional wisdom on measurement: A structural
equation perspective. Psychological Bulletin, 110, 305–314.
Brett, J. M., & Stroh, L. K. (2003). Working 61 plus hours per week: Why do managers do
it? Journal of Applied Psychology, 88, 67–78.
Brislin, R. W. (1980). Translation and content analysis of oral and written materials. In H.
C. Triandis & W. Lonner (Eds.), Handbook of cross-cultural psychology: Methodology
(pp. 389–444). Boston, MA: Allyn & Bacon.
Buchanan, T. (2000). Potential of the internet for personality research. In M. H. Birn-
baum (Ed.), Psychological experiments on the Internet (pp. 121–140). San Diego, CA:
Academic Press.
Buchanan, T. (2002). Online assessment: Desirable or dangerous? Professional Psychology:
Research and Practice, 33, 148–154.
Carlson, D. S., & Frone, M. R. (2003). Relation of behavioral and psychological involve-
ment to a new four-factor conceptualization of work-family interferences. Journal of
Business and Psychology, 17, 515–535.
Ellwart & Konradt 415

Carlson, D. S., & Kacmar, K. M. (2000). Work-family conflict in the organization: Do life
role values make a difference? Journal of Management, 26, 1031–1054.
Carlson, D. S., Kacmar, K. M., & Williams, L. J. (2000). Construction and initial validation
of a multidimensional measure of work-family conflict. Journal of Vocational Behavior,
56, 249–276.
Churchill, G. A., Jr. (1979). A paradigm for developing better measures of marketing
constructs. Journal of Marketing Research, 16, 64–73.
Cole, M. S., Bedeian, A. G., & Feild, H. S. (2006). The measurement equivalence of Web-
based and paper-and-pencil measures of transformational leadership. Organizational
Research Methods, 9, 339–368.
Coltman, T., Devinney, T. M., Midgley, D. F., & Veniak, S. (2008). Formative versus
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

reflective measurement models: Two applications of formative measurement. Journal of


Business Research, 61, 1250–1262.
Diamantopoulos, A., Riefler, P., & Roth, K. P. (2008). Advancing formative measurement
models. Journal of Business Research, 61, 1203–1218.
Diamantopoulos, A., & Winklhofer, H. M. (2001). Index construction with formative
indicators: An alternative to scale development. Journal of Marketing Research, 38,
269–277.
Diener, E., Emmons, R. A., Larson, R. J., & Griffin, S. (1985). The satisfaction with life
scale. Journal of Personality Assessment, 49, 71–75.
Dooley, R. S., & Fryxell, G. E. (1999). Attaining decision quality and commitment from
dissent: The moderating effects of loyalty and competence competence in strategic
decision-making teams. Academy of Management Journal, 42, 389–402.
Doty, D. H., & Glick, W. H. (1998). Common methods bias: Does common methods
variance really bias results? Organizational Research Methods, 1, 374–406.
Eby, L. T., Casper, W. J., Lockwood, A., Bordeaux, C., & Brinley, A. (2005). Work-family
research in IO/OB: Content analysis and review on literature (1980–2002). Journal of
Vocational Behavior, 66, 124–197.
Edwards, J. R, & Bagozzi, R. (2000). On the nature and direction of relationships between
constructs and measures. Psychological Methods, 5, 155–174.
Efron, B., & Tibshirani, R. J. (1986). Bootstrap methods for standard errors, confi-
dence intervals, and other measures of statistical accuracy. Statistical Science, 1, 54–
77.
Fassott, G. (2006). Operationalisierung latenter variablen in strukturgleichungsmodellen:
Eine standortbestimmung [Operationalization of latent variables in structural equation
models: An assessment of the current situation]. Zeitschrift für betriebswirtschaftliche
Forschung, 58, 67–88.
Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobserv-
able variables and measurement error. Journal of Marketing Research, 18, 39–50.
Frone, M. R. (2003). Work-family balance. In J. C. Quick & L. Tetrick (Eds.), Handbook
of occupational health psychology (pp. 143–162). Washington, DC: American Psycho-
logical Association.
Frone, M. R., Russell, M., & Cooper, M. L. (1992). Antecedents and outcomes of work fam-
ily conflict: Testing a model of the work-family interface. Journal of Applied Psychology,
77, 65–78.
Frone, M. R., Russell, M., & Cooper, M. L. (1997). Relation of work-family conflict
to health outcomes: A four-year longitudinal study of employed parents. Journal of
Occupational and Organizational Psychology, 70, 325–355.
Fu, C. K., & Shaffer, M. A. (2001). The tug of work and family: Direct and indi-
rect domain-specific determinants of work-family conflict. Personnel Review, 30, 502–
522.
416 The Journal of Psychology

Greenhaus, J. H., & Beutell, N. J. (1985). Sources of conflict between work and family
roles. Academy of Management Review, 10, 76–88.
Greenhaus, J. H., Parasuraman, S., & Collins, K. M. (2001). Career involvement and
family involvement as moderators of the relationship between work-family conflict and
withdrawal from profession. Journal of Occupational Health Psychology, 6, 91–100.
Grzywacz, J. G., & Marks, N. F. (2000). Reconceptualizing the work-family interface: An
ecological perspective on the correlates of positive and negative spillover between work
and family. Journal of Occupational Health Psychology, 5, 111–126.
Hair, J. F. J., Anderson, R. E., Tatham, R. L., & Black, W. C. (1998). Multivariate data
analysis with readings (5th ed.). Englewood Cliffs, NJ: Prentice Hall.
Howell, R. D., Breivik, E., & Wilcox, J. B. (2007). Reconsidering formative measurement.
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

Psychological Methods, 12, 205–218.


Jarvis, C. B., MacKenzie, S. B., & Podsakoff, P. M. (2003). A critical review of construct in-
dicators and measurement model misspecifications in marketing and consumer research.
Journal of Consumer Research, 30, 199–218.
Jones, F., Burke, R. J., & Westman, M. (2006). Work-life-balance: A psychological per-
spective. New York, NY: Psychological Press.
Jones, F., Kinman, G., & Payne, N. (2006). Work stress and health behaviors: A work-life
balance issue. In F. Jones, R. J. Burke, & M. Westman (Eds.), Work-life balance: A
psychological perspective (pp. 185–221). New York, NY: Psychology Press.
Kim, J. L. S., & Ling, C. S. (2001). Work-family conflict of women entrepreneurs in
Singapore. Women in Management Review, 16, 204–221.
Konradt, U., Christophersen, T., & Ellwart, T. (2008). Erfolgsfaktoren des lerntransfers
unter computergestütztem lernen: Der einfluss von lernstrategien, lernmotivation und
lernorganisation [Factors of successful learning transfer in computer-assisted learning:
The impact of learning strategies, motivation, and setting]. Zeitschrift für Personalpsy-
chologie, 7, 90–103.
Kossek, E. E., & Ozeki, C. (1998). Work-family conflict, policies, and the job-life satisfac-
tion relationship: A review and directions for organizational behavior-human resources
research. Journal of Applied Psychology, 83, 139–149.
Kraut, R., Olson, J., Banaji, M., Bruckman, A., Cohen, J., & Couper, M. (2004). Psycholog-
ical research online: Report of board of scientific affairs’ advisory group on the conduct
of research on the Internet. American Psychologist, 59, 105–117.
Kraut, R., Patterson, V., Lundmark, M., Kielser, S., Mukophadhyay, T., & Scherlis, W.
(1998). Internet paradox: A social technology that reduces social involvement and psy-
chological well-being? American Psychologist, 53, 1017–1031.
Kunin, T. (1955). The construction of a new type of attitude measure. Personnel Psychology,
8, 65–78.
Law, K. S., & Wong, C. S. (1999). Multidimensional constructs in structural equation
analysis: An illustration using the job perception and job satisfaction constructs. Journal
of Management, 25, 143–160.
Law, K. S., Wong, C. S., & Mobley, W. H. (1998). Towards a taxonomy of multidimensional
constructs. Academy of Management Review, 23, 741–755.
MacKenzie, S. B., Podsakoff, P. M., & Jarvis, C. B. (2005). The problem of measure-
ment model misspecification in behavioral and organizational research and some recom-
mended solutions. Journal of Applied Psychology, 90, 710–730.
Major, V. S., Klein, K. J., & Ehrhart, M. G. (2002). Work time, work interference with
family, and psychological distress. Journal of Applied Psychology, 87, 427–436.
Nagy, M. S. (2002). Using a single-item approach to measure facet job satisfaction. Journal
of Occupational and Organizational Psychology, 75, 77–86.
Ellwart & Konradt 417

Nunnally, J. C., & Bernstein, I. H. (1994). Psychometric theory (3rd ed.). New York, NY:
McGraw-Hill.
O’Driscoll, M., Brough, P., & Kalliath, T. (2006). Work-family conflict and facilitation.
In F. Jones, R. J. Burke, and M. Westman (Eds.), Work-life balance: A psychological
perspective (pp. 117–142). New York, NY: Psychology Press.
Petter, S., Straub, D., & Rai, A. (2007). Specifying formative constructs in IS research. MIS
Quarterly, 31, 623–656.
Podsakoff, P. M., MacKenzie, S. B., Lee, J.-Y., & Podsakoff, N. P. (2003). Common method
bias in behavioral research: A critical review of the literature and recommended remedies.
Journal of Applied Psychology, 88, 879–903.
Potosky, D., & Bobko, P. (1997). Computer versus paper-and-pencil administration mode
Downloaded by [Universiti Sains Malaysia] at 17:00 01 December 2011

and response distortion in noncognitive selection tests. Journal of Applied Psychology,


82, 293–299.
Ringle, C. M., Wende, S., & Will, A. (2005). SmartPLS (Version 2.0; beta) [Computer
software]. Hamburg, Germany: University of Hamburg.
Rothbard, N. P., & Dumas, T. L. (2006). Research perspectives: Managing the work-
hone interface. In F. Jones, R. J. Burke, & M. Westman (Eds.), Work-life balance: A
psychological perspective (pp. 71–89). New York, NY: Psychology Press.
Semmer, N. (1984). Stressbezogene tätigkeitsanalyse [Stress-based task analysis]. Wein-
heim, Germany: Beltz.
Simsek, Z., & Veiga, J. F. (2001). A primer on Internet organizational surveys. Organiza-
tional Research Methods, 4, 218–235.
Smith, P. C., Kendall, L. M., & Hulin, C. L. (1969). The measurement of satisfaction in
work and retirement. Chicago, IL: Rand McNally.
Sparks, K., Cooper, C., Fried, Y., & Shirom, A. (1997). The effect of hours of work on
health: A meta-analytic review. Journal of Occupational and Organizational Psychology,
70, 391–408.
Spector, P. E. (2006). Method variance in organizational research: Truth or urban legend?
Organizational Research Methods, 9, 221–232.
Stephens, C. K., & Sommer, S. M. (1996). The measurement of work to family conflict.
Educational and Psychological Measurement, 56, 475–486.
Tenenhaus, M., Vinzi, V. E., Chatelin, Y.-M., & Lauro, C. (2004). PLS path modeling.
Computational Statistics & Data Analysis, 48, 159–205.
Tetrick, L. E., & Buffardi, L. C. (2006). Measurement issues in research on the work-
home interface. In F. Jones, R. J. Burke, & M. Westman (Eds.), Work-life balance: A
psychological perspective (pp. 90–114). New York, NY: Psychology Press.
Thomas, L. T., & Ganster, D. C. (1995). Impact of family-supportive work variables on
work-family conflict and strain: A control perspective. Journal of Applied Psychology,
80, 6–15.
Thompson, L. F., Surface, E. A., Martin, D. L., & Sanders, M. G. (2003). From paper to
pixels: Moving personnel surveys to the Web. Personnel Psychology, 56, 197–227.
Wanous, J., Reichers, A. E., & Hudy, M. (1997). Overall job satisfaction: How good are
single-item measures? Journal of Applied Psychology, 82, 247–252.
Wilcox, R. R. (1996). Statistics for the social sciences. San Diego, CA: Academic Press.

Original manuscript received May 25, 2010


Final version accepted April 5, 2011

You might also like