You are on page 1of 16

Available online at www.sciencedirect.

com

Journal of Business Research 61 (2008) 1203 – 1218

Advancing formative measurement models


Adamantios Diamantopoulos ⁎, Petra Riefler 1 , Katharina P. Roth 2
Department of Business Administration, University of Vienna, Bruenner Strasse 72, A-1210 Vienna, Austria
Received 1 May 2007; received in revised form 1 November 2007; accepted 1 January 2008

Abstract

Formative measurement models were first introduced in the literature more than forty years ago and the discussion about their methodological
contribution has been increasing since the 1990s. However, the use of formative indicators for construct measurement in empirical studies is still
scarce. This paper seeks to encourage the thoughtful application of formative models by (a) highlighting the potential consequences of
measurement model misspecification, and (b) providing a state-of-the art review of key issues in the formative measurement literature. For the
former purpose, this paper summarizes findings of empirical studies investigating the effects of measurement misspecification. For the latter
purpose, the article merges contributions in the psychology, management, and marketing literatures to examine a variety of issues concerning the
conceptualization, estimation, and validation of formative measurement models. Finally, the article offers some suggestions for future research on
formative measurement.
© 2008 Elsevier Inc. All rights reserved.

Keywords: Formative index; Measurement model; Causal indicators

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1204
2. Reflective vs. formative measurement: first-order models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1204
3. Higher-order formative models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1205
4. Measurement model misspecification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1208
4.1. Parameter bias due to reversed causality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1208
4.2. Parameter bias due to incorrect item purification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1210
4.3. Effects on fit statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1210
5. The status quo of formative measures: issues and proposed remedies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1211
5.1. Conceptual issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1211
5.1.1. Error-free measures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1211
5.2. Interpretation of the error term . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1211
5.3. Estimation of formative models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1212
5.3.1. Multicollinearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1212
5.3.2. Exogenous variable intercorrelations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1212
5.3.3. Model identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1213

⁎ Corresponding author. Tel.: +43 1 4277 38031.


E-mail addresses: adamantios.diamantopoulos@univie.ac.at (A. Diamantopoulos), petra.riefler@univie.ac.at (P. Riefler), katharina.roth@univie.ac.at (K.P. Roth).
1
Tel.: +43 1 4277 38038.
2
Tel.: +43 1 4277 38040.

0148-2963/$ - see front matter © 2008 Elsevier Inc. All rights reserved.
doi:10.1016/j.jbusres.2008.01.009
1204 A. Diamantopoulos et al. / Journal of Business Research 61 (2008) 1203–1218

5.4. Reliability and validity assessment of formative models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1215


5.4.1. Reliability assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1215
5.4.2. Validity assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1215
6. Conclusion and future research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1216
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1216

1. Introduction The section that follows provides a brief conceptual dis-


cussion of reflective and formative measurement models. The
The literature in psychology, management, and marketing subsequent section covers the problem of measurement model
pays increasing attention to formative measurement models for misspecification, followed by a discussion of its consequences.
operationalizing latent variables (constructs). Researchers in Next, the article turns attention to a number of critical issues
various disciplines have undertaken considerable effort to (a) concerning the specification, estimation, and validation of for-
make the academic community aware of the existence of mative measures. Finally, the paper concludes by proposing
formative (cause, causal) indicators (e.g., Bollen and Lennox, some directions for future research.
1991), (b) demonstrate the potential appropriateness of formative
measurement models for a large number of latent constructs (e.g., 2. Reflective vs. formative measurement: first-order models
Diamantopoulos, 1999; Fassot and Eggert, 2005; Fassot, 2006;
Jarvis, MacKenzie and Podsakoff, 2003; Venaik, Midgley and The assessment of latent variables has a long tradition in
Devinney, 2004), (c) reveal consequences of measurement model social science (e.g., Churchill, 1979; Duncan, 1984; Nunally,
misspecification (e.g., Diamantopoulos and Siguaw, 2006; Law 1978). Latent variables are phenomena of theoretical interest
and Wong, 1999; MacKenzie, Podsakoff and Jarvis, 2005), and which cannot be directly observed and have to be assessed by
(d) develop practical guidelines for the construction of multi-item manifest measures which are observable. In this context, a
measures (indexes) comprising formative indicators (e.g., measurement model describes relationships between a construct
Diamantopoulos and Winklhofer, 2001; Eggert and Fassot, and its measures (items, indicators), while a structural model
2003; Giere, Wirtz and Schilke, 2006). Despite the growing specifies relationships between different constructs (Edwards
number of contributions on formative measurement, however, and Bagozzi, 2000; Scholderer and Balderjahn, 2006).
Bollen's (1989, p. 65) statement still holds true as even a cursory Anderson and Gerbing (1982, p. 453) note that “the reason
glance in the top management and marketing journals readily for drawing a distinction between the measurement model and
reveals, that is, “[M]ost researchers in the social sciences assume the structural model is that proper specification of the
that indicators are effect indicators. Cause indicators are measurement model is necessary before meaning can be
neglected despite their appropriateness in many instances”. assigned to the analysis of the structural model”. The
Two reasons help explain the prevalent lack of applications. measurement model (which is of focal interest in this paper)
On the one hand, a substantial number of researchers engaging in specifies the relationship between constructs and measures. In
measure development might still be unaware of the potential this respect, the direction of the relationship is either from the
appropriateness of formative indicators for operationalizing construct to the measures (reflective measurement) or from the
particular constructs (Hitt, Gimeno and Hoskisson, 1998; measures to the construct (formative measurement).
Podsakoff, Shen and Podsakoff, 2006); indeed “nearly all The first form of specification, that is, the reflective
measurement in psychology and the other social sciences assumes measurement model (see Fig. 1, Panel 1), has a long tradition
effect indicators” (Bollen, 2002, p. 616). On the other hand, in social sciences and is directly based on classical test theory
researchers might hesitate to specify formative measurement (Lord and Novick, 1968). According to this theory, measures
models because they “are often uncertain how to incorporate them denote effects (or manifestations) of an underlying latent
into structural equation models” (Bollen and Davis, 1994, p. 2). construct (Bollen and Lennox, 1991). Therefore, causality is
Indeed, there are a number of controversial and not fully resolved from the construct to the measures. Specifically, the latent
issues concerning the conceptualization, estimation and valida- variable η represents the common cause shared by all items xi
tion of formative measures (e.g., see Howell et al., 2007, 2008- reflecting the construct, with each item corresponding to a linear
this issue) including, among others, the treatment of indicator function of its underlying construct plus measurement error:
multicollinearity, the assessment of indicator validity, and the
interpretation of formatively-measured constructs. x i ¼ ki g þ e i ð1Þ
This article provides insights into the current state of literature
on formative measurement by merging major contributions in the where xi is the ith indicator of the latent variable η, εi is the
psychology, management and marketing literatures into an measurement error for the ith indicator, and λi is a coefficient
overall picture. The overall aim is to encourage the appropriate (loading) capturing the effect of η on xi. Measurement errors
use of formative indicators in empirical research while at the are assumed to be independent (i.e., cov(εi, εj) = 0, for i ≠ j)
same time highlighting potentially problematic issues and sug- and unrelated to the latent variable (i.e., cov(η, εi) = 0, for
gested remedies. all i).
A. Diamantopoulos et al. / Journal of Business Research 61 (2008) 1203–1218 1205

Fig. 1. Alternative measurement models.

Eq. (1) is a simple regression equation where the observable the construct's domain (see also Jarvis et al., 2003; and Rossiter,
measure is the dependent variable and the latent construct is the 2002); indeed, omitting an indicator potentially alters the nature
explanatory variable. A fundamental characteristic of reflective of the construct (Bollen and Lennox, 1991). Second, there are no
models is that a change in the latent variable causes variation specific expectations about patterns or magnitude of intercorrela-
in all measures simultaneously; furthermore, all measures in a tions between the indicators; formative indicators might correlate
reflective measurement model must be positively intercorre- positively or negatively or lack any correlation (for a detailed
lated (for a proof, see Bollen, 1984). discussion see Bollen, 1984). Third, formative indicators have no
The second form of specification, that is, the formative individual measurement error terms, that is, they are assumed to
measurement model, was first proposed by Curtis and Jackson be error-free in a conventional sense (Edwards and Bagozzi,
(1962) who challenge the characteristic of positively correlated 2000). The error term (ζ) is specified at the construct level
measures as a necessary condition. They argue that in specific (MacCallum and Browne, 1993) and does not constitute
cases measures show negative or zero correlations despite measurement error (Diamantopoulos, 2006). Fourth, a formative
capturing the same concept. Blalock (1964, 1968, 1971) and measurement model, in isolation, is underidentified and, there-
Land (1970) subsequently discuss this alternative measurement fore, cannot be estimated (Bollen, 1989; Bollen and Davis,
perspective according to which measures are causes of the 1994). In contrast, reflective measurement models with three or
construct rather than its effects (see Fig. 1, Panel 2). In other more indicators are identified and can be estimated (e.g., see
words, the indicators determine the latent variable which Long, 1983). A later section of this paper addresses the
receives its meaning from the former. Some typical examples estimation of formative models.
are socio-economic status (Hauser and Goldberger, 1971;
Hauser, 1973), quality of life (e.g., Bollen and Ting, 2000, 3. Higher-order formative models
Fayers, Hand, Bjordal and Groenvold, 1997;), or career success
(e.g., Judge and Bretz, 1994); Table 1 provides further The formative model specified in Eq. (2) is a first-order
examples. measurement model (Edwards, 2001). However, constructs are
The formal specification of the formative measurement often conceptualized and subsequently operationalized as
model is: multidimensional entities (e.g., Brewer, 2007; Lin, Sher and
Shih, 2005; Venaik et al., 2004; Yi and Davis, 2003). From a
X
n
conceptual point of view, a construct is multidimensional “when
g¼ gi xi þ f ð2Þ
i¼1 it consists of a number of interrelated attributes or dimensions
and exists in multidimensional domains. In contrast to a set of
where γi is a coefficient capturing the effect of indicator xi on interrelated unidimensional constructs, the dimensions of a
the latent variable η, and ζ is a disturbance term. The latter multidimensional construct can be conceptualized under an
comprises all remaining causes of the construct which are not overall abstraction, and it is theoretically meaningful and
represented in the indicators and are not correlated to the latter; parsimonious to use this overall abstraction as a representation
thus following the assumption that cov(xi,ζ) = 0. of the dimensions” (Law, Wong and Mobley, 1998, p. 741).
Eq. (2) represents a multiple regression equation and in When dealing with multidimensional constructs, it is
contrast to Eq. (1), the latent variable is the dependent variable necessary to distinguish between (at least) two levels of
and the indicators are the explanatory variables. Diamantopoulos analysis, that is, one level relating manifest indicators to
and Winklhofer (2001) point out several characteristics of this (first-order) dimensions, and a second level relating the
model which make it sharply distinct from the reflective model. individual dimensions to the (second-order) latent construct
First, the indicators characterize a set of distinct causes which are (Jarvis et al., 2003; MacKenzie et al., 2005). Failing to carefully
not interchangeable as each indicator captures a specific aspect of specify the latter relationships, “one cannot derive the overall
1206 A. Diamantopoulos et al. / Journal of Business Research 61 (2008) 1203–1218

Table 1
Examples of formatively-measured constructs
Author(s) Journal Formative construct(s) Estimation method
Consumer behavior literature
Lin et al. (2005) International Journal of Service Customer perceived value SEM a
Industry Management
Hyman et al. (2002) Journal of Marketing Theory and Practice Household affluence MIMIC model
Sánchez-Pérez and Iniesta-Bonillo (2004) Journal of Business and Psychology Consumers' commitment towards retailers MIMIC model

Information technology literature


Brock and Zhou (2005) Internet Research Organizational internet use SEM (PLS) a
Pavlou and Gefen (2005) Information Systems Research Psychological contract violation SEM (PLS) a
Perceived effectiveness of
institutional structures
Santosa et al. (2005) European Journal of Information Systems Intrinsic motivators SEM (PLS) a
Situational motivators
Yi and Davis (2003) Information Systems Research Observational learning SEM (PLS) a

Management literature
Helm (2005) Corporate Reputation Review Firm reputation SEM (PLS) a
Venaik et al. (2005) Journal of International Business Studies Environmental controls: SEM (PLS) a
– Local government regulatory influence
– Quality of local business infrastructure
– Pressures of global competition
– Pressures from technological change
Witt and Rode (2005) Journal of Enterprising Culture Corporate identity SEM (PLS) a
Corporate culture
Dowling (2004) Corporate Reputation Review Corporate descriptors Regression model
Corporate reputation
Venaik et al. (2004) Management International Review Firm pressures SEM (PLS) a
Johansson and Yip (1994) Strategic Management Journal Industry drivers SEM (PLS) a
Organization structure
Management process
Global strategy

Marketing literature
Bruhn et al. (2008-this issue) In this Special Issue Customer equity management SEM (PLS, LISREL)
Cadogan et al. (2008-this issue) In this Special Issue Quality of market-oriented behaviors MIMIC model
(LISREL)
Brewer (2007) Journal of International Marketing Psychic distance n.a.
Collier and Bienstock (2006) Journal of Service Research e-service quality SEM (AMOS) a
Johnson et al. (2006) Journal of Advertising Perceived interactivity: SEM (EQS) a
– Reciprocity
– Responsiveness
– Nonverbal information
– Speed of response
Ulaga and Eggert (2006) Journal of Marketing Relationship value SEM with summated
dimension scores (PLS)
Reinartz et al. (2004) Journal of Marketing Research CRM process implementation MIMIC model
Arnett et al. (2003) Journal of Retailing Retailer equity MIMIC model
Homburg et al. (2002) Journal of Marketing Service orientation Composite score
Winklhofer and Diamantopoulos (2002) International Journal of Research in Marketing Sales forecasting effectiveness MIMIC model
Homburg et al. (1999) Journal of Marketing Marketing's influence SEM (LISREL) a
Market-related complexity
a
Identification achieved through linkage to two or more reflective constructs.

construct from its dimensions and can only conduct research at see Cohen et al., 1990; Edwards and Bagozzi, 2000; Giere et al.,
the dimensional level, even though these dimensions are 2006; Law et al., 1998; Law and Wong, 1999), (b) reflective
claimed theoretically to be under an overall construct“ (Law first-order and formative second-order, (c) formative first-order
et al., 1998, p. 741). Since for each level both formative or and reflective second-order, and (d) reflective first-order and
reflective specifications are applicable, Jarvis et al. (2003) reflective second-order models (synonyms for this type of
identify four different types of multidimensional constructs, model are “latent model”, “factor model”, “superordinate
namely, (a) formative first-order and formative second-order construct”, “indirect reflective model” and “second-order total
(synonyms for this model are “aggregate model”, “composite disaggregation model”; see Bagozzi and Heatherton, 1994;
model”, “emergent model” and “indirect formative model”; e.g., Edwards, 2001; Edwards and Bagozzi, 2000; Giere et al., 2006;
A. Diamantopoulos et al. / Journal of Business Research 61 (2008) 1203–1218 1207

Law et al., 1998). Since this review focuses on formative 2004; Venaik, Midgley and Devinney, 2005; Witt and Rode,
measurement, this section only briefly discusses the former 2005; Yi and Davis, 2003; see also Bruhn et al., 2008-this issue).
three types of multidimensional constructs (see Fig. 2). For example, Yi and Davis' (2003) construct of “obser-
The first model in Fig. 2 (Type I) conceptualizes the vational learning processes” comprises four formative first-
multidimensional construct as a composite of its dimensions order dimensions, namely “attention processes”, “retention
such that the arrows point from the dimensions to the construct processes”, “production processes”, and “motivation processes”.
(Williams, Edwards and Vandenberg, 2003). The dimensions are The second type of model shown in Fig. 2 (Type II) represents
thus analogous to formative measures; however, in contrast to a second-order construct with first-order formative dimensions
the traditional conceptualization of formative measures as which are themselves measured by several reflective manifest
observed variables (see Eq. (2)), the dimensions are themselves items. According to this conceptualization, the error term exists at
constructs and conceived as specific components of the second- two different levels, namely (a) at the level of the manifest
order construct (Edwards, 2001). In this type of model, the error indicators, where it represents measurement error, and (b) at the
term exists both at the level of the individual (first-order) level of the second-order construct, where it captures the amount
dimensions and at the overall construct level. Table 1 provides a of variance in the second-order construct which the first-order
number of empirical illustrations of Type I formative multi- dimensions do not account for. As Type II models have been
dimensional constructs (e.g., Arnett, Laverie, and Meiers, 2003; introduced rather recently, only most recent literature provides
Brewer, 2007; Reinartz, Krafft and Hoyer, 2004; Venaik et al., empirical examples of its use (e.g., Johnson, Bruner and Kumar

Fig. 2. Higher-order formative models (adapted from Jarvis et al., 2003, p. 205).
1208 A. Diamantopoulos et al. / Journal of Business Research 61 (2008) 1203–1218

et al., 2006; Lin et al., 2005; see also Ruiz et al., 2008-this issue). fact that standardized development procedures for reflective
For example, Lin et al. (2005) conceptualize the construct of scales have been established over the years (e.g., see Churchill,
“customer perceived value” as a second-order factor which is 1979; DeVellis, 2003; Netemeyer, Bearden and Sharma, 2003;
formed by five reflectively specified first-order dimensions, Spector, 1992), whereas concrete guidelines for the construction
namely “monetary sacrifice”, “website design”, “fulfillment/ of formative indixes have been proposed very recently
reliability”, and “security/privacy”. (Diamantopoulos and Winklhofer, 2001; Eggert and Fassot,
The third model illustrated in Fig. 2 (Type III) has first-order 2003; Giere et al., 2006).
factors as reflective dimensions, but the first-order dimensions Jarvis et al. (2003) assess the degree of misspecification for
themselves have formative indicators. For this reason, the error studies published in four major marketing journals (Journal of
term exists at the level of the first-order dimensions only and Marketing, Journal of Marketing Research, Marketing Science,
represents both the variance not explained by the manifest and Journal of Consumer Research). Even though they apply a
indicators (due to the formative specification of the first-order conservative evaluation approach (i.e., classifying operationaliza-
dimensions) and variance not explained by the underlying tions as correct in case that either reflective or formative measures
(higher-order) construct. Although Jarvis et al. (2003) include could in general apply), they find about a third of all studies to be
this model in their typology, literature has not explicitly subject to measurement model misspecification. Fassot (2006)
recognized this kind of model, and empirical examples remain applies Jarvis et al.'s (2003) approach to three major German
virtually non-existent. The reasons for this are threefold (see management journals (Zeitschrift für Betriebswirtschaft, Zeitschrift
also Albers and Götz, 2006). First, as noted above, the nature of für betriebswirtschaftliche Forschung, Die Betriebswirtschaft) and
the error term is difficult to interpret due to the endogenous reports similar results (i.e., 35% of all investigated studies include
position of the formative first-order dimensions. Second, misspecified constructs). In a similar effort, Fassot and Eggert
formative indicators capture different facets of a construct and (2005) calculate a misspecification rate of some 80% for a major
are therefore not interchangeable (Diamantopoulos and Winkl- German marketing journal (Marketing ZFP).
hofer, 2001). These indicators give the first-order dimensions This problematic situation is not unique to marketing literature.
their meaning, which, by definition, has to be different for each In similar efforts, Podsakoff et al. (2006) reveal inappropriate
dimension because of the formative specification (Rossiter, modeling for 62% of constructs published in three major strategic
2002). Since a reflective specification at the second-order level management journals (Academy of Management Journal, Admin-
implies that the dimensions are manifestations of a second-order istrative Science Quarterly, Strategic Management Journal),
construct, it is unclear whether the meaning of the dimensions is while Podsakoff, MacKenzie, Podsakoff and Lee (2003) report a
attributable to the formative indicators or to the underlying misspecification rate of 47% for leadership research (including
common cause. Third, Type III models cannot be estimated publications in The Leadership Quarterly, Journal of Applied
using current procedures for achieving identification of Psychology, and again Academy of Management Journal).
formative constructs (see section on model identification later Given this documented existence of measurement model
in this paper). In short, Type III models do not represent an misspecification, the obvious question is to what extent
appealing option for specifying multidimensional constructs. misspecification does impact on model estimates and fit
statistics. This question is important because “any bias in the
4. Measurement model misspecification estimates […] could affect the conclusions about the theoretical
relationships among the constructs drawn from the research”
A number of researchers criticize the prevalent neglect of (Jarvis et al., 2003, p. 207).
explicit measurement model specification underlying scale The literature review identifies six studies empirically
construction efforts (Diamantopoulos and Winklhofer, 2001; investigating the consequences of measurement model mis-
Eberl, 2006; Fassot, 2006; Fassot and Eggert, 2005; Fornell and specification. Table 2 categorizes these studies along two
Bookstein, 1982; Jarvis et al., 2003; Podsakoff et al., 2006). characteristics. The first characteristic refers to the source of
Most researchers apply scale development procedures without bias investigated, which is either (a) the wrongly specified
even questioning their appropriateness for the specific construct direction of causality between a given set of indicators and a
at hand (see also Albers and Hildebrandt, 2006; Williams et al., construct, or (b) the application of an inappropriate item puri-
2004; for a noteworthy exception see Eberl and Schweiger, fication procedure (i.e., purifying formative indicators accord-
2005); indeed, Diamantopoulos and Winklhofer (2001, p. 274) ing to guidelines applicable for reflective indicators). The
speak of an “almost automatic acceptance of reflective second characteristic refers to the position of the focal mis-
indicators”. Consequently, misspecification commonly con- specified construct in the structural model, which is either
cerns the adoption of reflective indicators where formative exogenous or endogenous. A discussion of the findings of these
indicators (and thus index construction approaches) would be studies follows.
appropriate (which is a Type I error in Diamantopoulos and
Siguaw's (2006) terminology). The other case of misspecifica- 4.1. Parameter bias due to reversed causality
tion, that is, the incorrect adoption of a formative model where
indeed a reflective would be appropriate (Type II error), is rather Jarvis et al. (2003), Law and Wong (1999), and MacKenzie
negligible (Fassot, 2006; Jarvis et al., 2003). An explanation for et al. (2005) examine the impact of incorrect causal direction,
this difference in evidence of Type I and Type II errors is the that is, the specification of a reflective measurement model
Table 2
Empirical studies on consequences of measurement model misspecification
Focus (reason for estimation Data set Technique Structural parameter estimates a Model fit b Additional findings
bias)
Exogenous Endogenous Exogenous Endogenous
construct construct construct construct
misspecified misspecified misspecified misspecified

A. Diamantopoulos et al. / Journal of Business Research 61 (2008) 1203–1218


Law and Wong Reversed causality Survey data SEM Overestimation Not tested CFI ≈ Not tested Also biases in model relationships
(1999) (RAMONA) NFI ≈ which do not involve the misspecified
NNFI ≈ construct
IFI ≈
TLI ≈
χ2 / df ↑
Edwards (2001) Reversed causality Published SEM (LISREL, Underestimation c Over- and CFI ↓ Not comparable Concluded that both, the
covariance matrices RAMONA) underestimation of RMSEA ↑ (df = 0 and perfect multidimensional formative and
of survey data some parameters χ2 / df ↑ fit of formative reflective specification were inferior
models) to a multivariate structural model
Jarvis et al. (2003) Reversed causality Simulated data Monte Carlo Overestimation UnderestimationCFI ≈ Item correlation found to be negatively
simulation (335% to 555%) (88% to 93%) GFI ↑ related to magnitude of estimation bias
RMSEA ≈
SRMR ≈
χ2 / df ↑
MacKenzie et al. Reversed causality Simulated data Monte Carlo Overestimation Underestimation CFI ↓ CFI ↓ Type II error increases if endogenous
(2005) simulation (on average: 429%) (on average: 84%) GFI ≈ GFI ↓ or both constructs are misspecified
RMSEA ↑ RMSEA ↑
SRMR ≈ SRMR ↑
Albers and Reversed causality and Simulated data SEM (PLS, No bias Not tested Not given Not tested
Hildebrandt (2006) incorrect indicator purification LISREL) (stated that fit
indices were
similarly good)
Diamantopoulos and Incorrect indicator purification Survey data Regression Underestimation d Not tested CFI ≈ Not tested
Siguaw (2006) analysis GFI ≈
RMSEA ↑
NNFI ≈
χ2 / df ↑
a
Unstandardized parameter estimates.
b
≈ Goodness-of-fit index for reflective and formative model similar (difference +/−.05).
c
Edwards (2001) estimates several second-order models, this comparison concerns the Congeneric and Estimated Loadings Models.
d
R-squares compared.

1209
1210 A. Diamantopoulos et al. / Journal of Business Research 61 (2008) 1203–1218

when a formative model is conceptually appropriate, for consequences of applying conventional scale development
exogenous latent variables. All three studies reveal an over- procedures on formative measures. Fassot (2006) provides an
estimation of structural parameters when the latent variable is example of a misspecified measure that leads to a neglect of a
affected by misspecification. In some cases, the (incorrect) key aspect of the focal construct. More specifically, “perceived
reflective specification even yields a significant parameter friendliness of the staff” is erroneously dropped from a measure
estimate, whereas the parameter estimate is not significant in the of hospital quality due to not meeting conventional standards
(correct) formative specification. Thus, the impact of the focal for reflective items (i.e., high item-total correlations), however,
latent variable on other constructs in the structural model tends despite being a key aspect of a hospital's quality assessment. In
to be overestimated. line with this example, Diamantopoulos and Siguaw (2006) find
Jarvis et al. (2003) and MacKenzie et al. (2005) additionally that the same initial item pool results in considerably differing
examine the impact of incorrect specifications of endogenous final item sets under reflective and formative purification
latent variables. In contrast to the exogenous case, both studies guidelines respectively. The former approach eliminates items
report an underestimation of the parameter estimate capturing with low inter-item correlations, whereas the latter drops items
the impact of antecedent variables on the focal construct. An with high inter-item correlations (thus causing problems of
explanation for these distinct findings of under- and over- multicollinearity). In Diamantopoulos and Siguaw's (2006)
estimation for endogenous and exogenous positions, respec- example, the resulting scale and index share not more than two
tively, is the difference in portions of variance accounted for by out of 30 initial items. Their study therefore demonstrates how
reflective and formative operationalizations. Specifically, a erroneous reflective scale purification processes can substan-
reflective treatment of a formative construct reduces the tially alter the meaning of formative constructs.
variance of the construct (see Fornell, Rhee and Yi, 1991 or Albers and Hildebrandt (2006) address the issue of
Namboodiri, Carter and Blalock, 1975) because the variance of parameter estimation bias due to incorrect indicator purification.
a reflectively-measured construct equals the common variance First, the authors compare parameter estimates of a reflectively
of its measures, whereas the variance of a formatively-measured and a formatively specified measurement model using the full
construct encompasses the total variance of its measures (Law item set prior to purification. Second, they compare two
and Wong, 1999). Consequently, if a misspecification reduces formatively specified models, once using the full item pool (i.e.,
the variance of the exogenous variable while the level of the in accordance with the requirement of a census of items) and
variance of the endogenous variable is maintained, the once using a reduced item set following purification guidelines
parameter estimate for their relationship increases. In contrast, for reflective scales. The latter comparison reveals an extensive
if a misspecification reduces the variance of the endogenous underestimation of structural parameters, while the former
variable while the variance of the exogenous variable is shows no significant differences. Therefore, in this example, it
unchanged, the relevant structural parameter estimate decreases. is the erroneous purification rather than the causal order
In any case, these analyses reveal that structural paths are misspecification that impacts on parameter bias.
either overestimated or underestimated as a result of measure-
ment model misspecification with undesirable effects on the 4.3. Effects on fit statistics
substantive interpretation of the structural model relationships.
The studies in Table 2 also examine the impact of
4.2. Parameter bias due to incorrect item purification misspecification on goodness-of-fit indices for the overall (i.e.,
measurement and structural) model. An intuitive expectation is
To fully capture the meaning of a formatively-measured that the consequences of misspecification in terms of changed
construct, a census of indicators is (ideally) required because construct meanings and biased parameter estimates would also
“[o]mitting an indicator is omitting a part of the construct” lead to poor model fit. However, the majority of models
(Bollen and Lennox, 1991, p. 308). Therefore, an omission of incorporating misspecified constructs show highly acceptable
indicators is equivalent to restricting the domain of the construct values for CFI, GFI, SRMR and RMSEA. Moreover, these
(MacKenzie et al., 2005). In the context of index construction, values are similar to the goodness-of-fit values obtained for the
this characteristic implies that the elimination of formative items corresponding correctly specified model. For example, Mac-
from the item pool has to be theoretically justified rather than Kenzie et al. (2005, p. 724) conclude from their study that “each
purely based on statistical properties (Diamantopoulos and of the four goodness-of-fit indices failed to detect the
Winklhofer, 2001; Diamantopoulos and Siguaw, 2006). Indeed, misspecification of the measurement model”. This equally
“internal-consistency checks on cause-indicators may lead applies to all other studies listed in Table 2. Only the chi-square
researchers to discard valid measures improperly” (Bollen, (per degree of freedom) statistic shows to be consistently higher
1984, p. 381) and “following standard scale development in the wrongly reflectively specified models throughout the
procedures – for example dropping items that possess low item- studies, thus providing some indication of the underlying
to-total correlations – will remove precisely those items that misspecification.
would most alter the empirical meaning of the construct” (Jarvis Summarizing, all studies empirically examining the con-
et al., 2003, p. 202; see also MacKenzie, 2003). sequences of measurement model misspecification on para-
In light of the extensive presence of measurement model meter estimates report serious under- or overestimation of
misspecification discussed earlier, recent studies examine the parameters as a consequence of misspecified causality, wrongly
A. Diamantopoulos et al. / Journal of Business Research 61 (2008) 1203–1218 1211

adopted purification procedures, or a combination of both. Such so, each formative indicator becomes a (single) reflective
biases may in turn lead to incorrect conclusions on tested measure of its respective latent variable ξi and consequently
relationships, thus putting many empirical results into question. comprises an error term; hence, the assumption of error-free
Especially alarming is the fact that a satisfactory overall model indicators is relaxed.
fit does not guarantee a correct specification and that Although this model has the substantial advantage of
misspecifications are not detected by poor fit index values. It incorporating measurement error, its conceptual justification is
is to hope that the empirical demonstration of the undesirable questionable for several reasons. First, the inclusion of the first-
consequences of measurement model misspecification will order constructs ξi introduces a “fictitious” level, which
yield more echo to Bagozzi's (1984) and Jarvis et al.'s (2003) adversely affects model parsimony and suggests that a latent
call to conceptually justify measurement relationships as variable can more or less automatically be specified for any
hypotheses and subsequently test them empirically. manifest variable. Second, given that the xi are not directly
linked to η, they cannot be legitimately considered to be
5. The status quo of formative measures: issues and indicators of η because indicators need to be linked by means of
proposed remedies a direct relationship to the construct they assess. Third, the
measures of the ξi in Fig. 3 are single-indicators with all
As the introductory section already outlines, literature has drawbacks such indicators entail (such as high specificity, and
only recently started to pay serious attention to formative low reliability). As a discussion of potential problems with
measurement models and empirical applications are still rare. single item measures is beyond the scope of this paper, the
As a result, experience with formative measures is limited and reader is referred to Gardner, Cummings, Dunham and Pierce
several conceptual and practical issues are not fully clarified yet. (1998) and Nunnally and Bernstein (1994) for further details.
The following sections discuss such issues and highlight
various (sometimes contradicting) views of proposed remedies. 5.2. Interpretation of the error term

5.1. Conceptual issues Eq. (2) and Fig. 1 (Panel 2) show that a formative measurement
model specification includes an error (disturbance) term at the
5.1.1. Error-free measures construct level. This error term represents the surplus meaning of
Formative measurement models incorporate the error term at the construct (Jarvis et al., 2003; Temme 2006) which is not
the construct level and specify individual indicators to be error- captured by the set of formative indicators included in the model
free (see Eq. (2) earlier). Some researchers find this non- specification. Diamantopoulos (2006, p. 7) points out that
existence of measurement error hard to accept. Edwards and “previous discussions of the error term are often problematic
Bagozzi (2000), for example, regard such an assumption as and fail to provide […] a clear interpretation of exactly what the
untenable in most situations. Addressing this objection, a model error term represents”. Jarvis et al. (2003), for example, describe
such as the one depicted in Fig. 3 is worth considering as one the error term as the collective (i.e., overall) random error of all
way of incorporating measurement error to formative measure- formative indicators taken as a group, while MacKenzie et al.
ment models. This model is similar to Edwards and Bagozzi's (2005, p. 712) interpret the error estimate as capturing “the
(2000) “spurious model” with multiple common causes, with invalidity of the set of measures — caused by measurement error,
the only difference that the latent variables are intentionally interactions among the measures, and/or aspects of the construct
introduced to enable the accommodation of measurement error domain not represented by the measures”. However, the first
in the indicators. source of error, that is, measurement error, is conceptually
This model inserts a latent variable ξi for each formative incorrect. Diamantopoulos (2006) demonstrates that the error
indicator xi so that the focal latent variable η is indirectly linked term does not represent measurement error because formative
to the indicators xi via the latent exogenous constructs ξi. Doing indicators are specified to be error-free and, therefore, measure-
ment error cannot be included in the error term at the construct
level. The second source, that is, measure interactions, is
statistically plausible but lacks substantive interpretation. Since
formative indicators determine the meaning of the latent variable,
it is not possible to separate the construct's meaning from the
indicators' content (Diamantopoulos, 2006). If two indicators
show interaction effects, these effects would also form the
construct's meaning as both indicators separately do. The third
source, is indeed the correct interpretation of the nature of the
error term, that is, aspects of the construct domain not represented
by the indicators. Specifically, “the error term in a formative
measurement model represents the impact of all remaining causes
other than those represented by the indicators included in the
model” (Diamantopoulos, 2006, p. 11). Formative latent variables
Fig. 3. Modified formative model with individual error terms. have a number of proximal causes which researchers try to
1212 A. Diamantopoulos et al. / Journal of Business Research 61 (2008) 1203–1218

identify when conceptually specifying the construct. However, in consequently combined into an index, what exactly does this
many cases researchers will be unable to detect all possible causes index capture? Second, having included this index into Eq. (2),
as there may be some which have neither been discussed in prior what kind of information does its corresponding regression
literature nor revealed by exploratory research. The construct- parameter estimate provide? Is it the impact of a joint unit
level error term represents these missing causes. This means that change in both, the income and the age?
the more comprehensive the set of formative indicators specified
for the construct, the smaller the influence of the error term. 5.3.2. Exogenous variable intercorrelations
Williams et al. (2003, p. 908) note in this context that “as the One general issue when specifying measurement models is
variance of the residual increases, the meaning of the construct the specification of inter-indicator correlations. In reflective
becomes progressively ambiguous”. models, a common approach is to free all covariances among
exogenous variables allowing for intercorrelations. In formative
5.3. Estimation of formative models models, following this strategy leads to a large number of
additional parameters, namely correlation estimates of covar-
5.3.1. Multicollinearity iances between (a) formative indicators within a construct, (b)
Multicollinearity is an undesirable property in formative formative indicators between constructs, and (c) exogenous
models as it causes estimation difficulties (Albers and Hildeb- latent constructs (Bollen and Lennox, 1991; MacCallum and
randt, 2006; Diamantopoulos and Winklhofer, 2001). These Browne, 1993).
estimation problems arise because a multiple regression links the Bollen and Lennox (1991) recommend allowing for inter-
formative indicators to the construct (see Eq. (2)). Substantial correlation of formative measures which relate to the same
correlations among formative indicators result in unstable construct (without, however, expecting any specific pattern).
estimates for the indicator coefficients γi and it becomes difficult Furthermore, they argue that for both reflectively and
to separate the distinct influence of individual indicators on the formatively-measured constructs it is likely, but though not
latent variable η. Diamantopoulos and Winklhofer (2001) further necessary, that item correlations within constructs exceed item
note that multicollinearity leads to difficulties in assessing correlations between constructs. Based on this argument,
indicator validity on the basis of the magnitude of the parameters MacCallum and Browne (1993) consider two possible
γi (Bollen, 1984; MacKenzie et al., 2005). approaches of specifying correlations.
Literature proposes different approaches for dealing with The first approach specifies formative indicators of the same
multicollinearity. Bollen and Lennox (1991) argue that indica- construct to be correlated with each other but uncorrelated with
tors which highly intercorrelate are almost perfect linear indicators of other constructs. The obvious advantage of this
combinations and thus quite likely contain redundant informa- procedure is that it retains model parsimony as no non-
tion. Based on this view, several authors (e.g., Diamantopoulos hypothesized paths are added. The obtained goodness-of-fit
and Winklhofer, 2001; Götz and Liehr-Gobbers, 2004) suggest indices are hence solely based on the hypothesized relation-
indicator elimination based on the variance inflation factor ships, that is, the relationships of interest. The shortcoming of
(VIF), which assesses the degree of multicollinearity. Some this approach, however, is that the fixing of covariances to zero
empirical studies on formative measure development (e.g., leads to blocks of zeros in the implied covariance matrix. These
Diamantopoulos and Siguaw, 2006; Helm, 2005; Sánchez-Pérez zero covariances assume that the corresponding indicators and/
and Iniesta-Bonillo, 2004; Witt and Rode, 2005) follow this or latent variables are perfectly uncorrelated. MacCallum and
advice usually by applying the commonly accepted cut-off value Browne (1993) note that this assumption implies substantive
of VIF N 10 or its tolerance equivalent (see Giere et al., 2006; meaning for the model which requires theoretical justification.
Hair, Anderson, Tatham and Black, 1998; Kennedy, 2003). They therefore refrain from recommending this approach. Jarvis
However, considering that this multicollinearity check leads to et al. (2003) further argue that any common cause of the
indicator elimination on purely statistical grounds and given the concerned variables that is not incorporated in the model
danger of altering the meaning of the construct by excluding contributes to a lack of model fit. Consequently, they also
indicators (Bollen and Lennox, 1991), “[i]ndicator elimination – conclude that fixing covariances to zero is an inappropriate
by whatever means – should not be divorced from conceptual method.
considerations when a formative measurement model is The second approach specifies formative indicators to be
involved” (Diamantopoulos and Winklhofer, 2001, p. 273). correlated with each other as well as with indicators of other
Albers and Hildebrandt (2006) put forward a different constructs or exogenous variables. The major advantage of this
approach for overcoming multicollinearity by combining method is that all variables are allowed to covary instead of
formative indicators into an index (using either an arithmetic assuming complete independence which is theoretically not
or geometric mean) and using the latter as a single-item justifiable. This approach, however, also raises a number of
construct in the subsequent analysis. However, although problematic issues. First, the number of parameters to be
intuitively appealing, this suggestion raises two important estimated increases, thereby decreasing the number of degrees
questions. First, what is the interpretation of the joint index of of freedom. Second, MacCallum and Browne (1993) empiri-
two indicators in terms of its substantial meaning? If, for cally show that the additional parameters provide little ex-
example, income and age show a high intercorrelation (which planatory value. Consequently, models lack parsimony without
appears to be a likely assumption) and their measures are providing substantive meaning in explaining inter-measure
A. Diamantopoulos et al. / Journal of Business Research 61 (2008) 1203–1218 1213

relationships. Furthermore, the estimates for unhypothesized ture of these two approaches, that is, adding a single reflective
parameters influence the overall model fit, even though they are indicator and a reflectively-measured construct as an outcome
not of interest. Despite these shortcomings, MacCallum and variable.
Browne (1993) recommend this method compared to the option
of having zero blocks in the implied covariance matrix. Jarvis et 5.3.3.1. Adding two reflective indicators. The first option is
al. (2003) agree on that fact but stress that locating the impact of adding two reflective measures to the set of formative indicators
the non-hypothesized parameter estimates on the model fit is (see Fig. 4). Jarvis et al. (2003) and MacKenzie et al. (2005)
necessary. They suggest estimating a series of nested models, advise this method based on the key arguments that (a) this
that is, freeing parameters step by step and comparing the approach does not require adding constructs to the model solely
overall model fit across steps. for identification purposes (which contributes to model
parsimony), and (b) measurement parameters are stable and
5.3.3. Model identification less sensitive to changes in structural parameters.
A major concern of formative measurement models is how to However, this model allows for different conceptual
establish statistical identification to enable their estimation. In interpretations (Jarvis et al., 2003), namely (a) a MIMIC
isolation, formatively-measured constructs as defined by Eq. (2) model (Jöreskog and Goldberger, 1975), (b) an endogenous
are underidentified (Bollen and Lennox, 1991; MacCallum and construct with two reflective indicators that is influenced by
Browne, 1993; Temme, 2006) and, thus, cannot be estimated. exogenous observed variables, or (c) a formatively-measured
This inability to estimate formative measurement models construct which influences indicators of another construct.
without the introduction of additional information (see below) MacKenzie et al. (2005) argue that the constellation resulting
has resulted in criticisms of the value of formative measurement from adding two reflective measures to the formative speci-
in general (see Howell et al., 2007, 2008-this issue). fication should not be interpreted as a MIMIC model but as a
As with reflective measurement models, two necessary but yet latent variable having a mixture of formative and reflective
not sufficient conditions have to be met for identifying models indicators (since both types of indicators belong to the same
including formatively-measured constructs (Bollen, 1989; Bollen concept domain and are content-valid operationalizations of the
and Davis, 1994; Cantallupi, 2002a,b; Edwards, 2001; Temme, same construct). In contrast, MacCallum and Browne (1993),
2006). First, the number of non-redundant elements in the Scholderer and Balderjahn (2006) and Temme (2006) explicitly
covariance-matrix of the observed variables needs to be greater equate models with mixed indicators and MIMIC models. It is
or equal the number of unknown parameter in the model (t-rule). outside the scope of this paper to discuss these interpretations in
Second, the latent construct needs to be scaled (scaling rule). For detail, but it should be stressed that despite the different possible
the latter condition, three main options are available (Bollen and interpretations at a conceptual level, there are no differences
Davis, 1994; MacCallum and Browne, 1993), namely (a) fixing a at the empirical level (the models yield the same parameter
path from a formative indicator to the construct, (b) fixing a path estimates).
from the formatively-measured construct to a reflectively-
measured endogenous latent variable, or (c) standardizing the
formatively-measured construct by fixing its variance to unity.
Edwards (2001) advises the last of the three options because
fixing path parameters precludes estimating standard errors of
theoretically interesting relationships. Note that the choice of
scaling method can affect substantive conclusions as the
significance of different relationships in the model with a
formatively-measured construct may vary depending of how the
scale of the latter is set (see Franke et al., 2008-this issue).
The t-rule and scaling rule are, however, not sufficient
conditions for identifying formative measurement models. In
this context, Bollen (1989) draws attention to the fact that the
formative measurement model needs to be placed within a
larger model that incorporates consequences (i.e., effects) of the
latent variable in question to enable its estimation. Specifically,
for identifying the disturbance term ζ at the construct level, the
formative latent variable needs to emit at least two paths to other
(reflective) constructs or indicators (MacCallum and Browne,
1993); literature also refers to this condition as 2+ emitted paths
rule (Bollen and Davis, 1994).
Literature discusses three approaches for applying the 2+
emitted paths rule, which are (a) adding two reflective indicators
to the formatively-measured construct, (b) adding two reflec-
tively-measured constructs as outcome variables, and (c) a mix- Fig. 4. Identification using a MIMIC model.
1214 A. Diamantopoulos et al. / Journal of Business Research 61 (2008) 1203–1218

Fig. 5. Identification with two reflectively-measured constructs.

5.3.3.2. Adding two reflective constructs. According to Bollen issue). Indeed, as Bagozzi (2007, p. 236) observes, “the param-
and Davis (1994), another option of establishing model eters regarding the observed variables to their purported
identification is the specification of two structural relations from formative latent variable are functions of the number and nature
the formative latent variable to two reflectively-measured of endogenous latent variables and their measures”.
constructs (Fig. 5). While these two reflectively-measured
constructs need to be unrelated in models which only comprise 5.3.3.3. Adding one reflective indicator and one reflective
the focal formatively-measured construct and the two reflectively- construct. This model is a mixture of the two previous
measured constructs (as in Fig. 5), the reflective constructs may be procedures and involves adding one reflective indicator to the
causally related in larger models (Temme, 2006). latent construct and linking the latter to a reflectively-measured
This model is justifiable in case that two reflectively- latent variable (Fig. 6). This mixed approach is applicable if the
measured constructs can be included in the nomological theoretical model includes only one structural relationship of
network based on theoretical considerations. However, includ- the formatively-measured latent variable to a reflectively-
ing any reflectively-measured outcome variables purely for measured latent variable. In this case, including a reflective
identification reasons puts the theoretical model specification indicator such as a global measure helps to overcome under-
into question if these outcomes are not of theoretical interest. identification and might simultaneously be used for validation
Note also that the choice of outcome constructs potentially purposes (see Diamantopoulos and Winklhofer, 2001).
affects the interpretation of the formatively-measured construct Temme (2006) demonstrates that the 2+ emitted paths rule is
itself by influencing the estimates of γ-parameters (see Heise, a necessary but yet not necessarily sufficient condition for
1972; Howell et al., 2007; and also Franke et al., 2008-this identification when the two reflectively-measured outcome

Fig. 6. Identification with one reflective measure and one reflectively-measured construct.
A. Diamantopoulos et al. / Journal of Business Research 61 (2008) 1203–1218 1215

constructs are either directionally related (i.e., one directly indicators and an alternative measure assessing the focal
impacts on the other), or their disturbance terms are correlated. construct. What needs to be clarified, however, is how such a
These models require imposing further restrictions in order to correlation should be interpreted. Would a non-significant
establish full model identification (such as fixing the covariance correlation unambiguously mean that the focal measure lacks
of the disturbance terms to zero, or using a partially reduced reliability? What if the alternative measure is itself unreliable?
form model; for details, see Bollen and Davis, 1994; Does this approach actually test the reliability of the focal
Cantaluppi, 2002a,b; and Temme, 2006). measure or is it rather a test of convergent validity?
Finally, models which violate the 2+ emitted paths rule due to
containing formatively-measured constructs that emit only one 5.4.2. Validity assessment
path can be identified by fixing the variance of the disturbance One of the most controversial issues in formative measure-
term to zero (MacCallum and Browne, 1993). MacCallum and ment literature is validity assessment. Some researchers argue
Browne (1993) alert to applying this approach with caution as it that no quantitative quality checks are usable for assessing the
implies the theoretical assumption that the formative indicators appropriateness of formative indices (e.g., Homburg and
completely capture the construct. In other words, this approach Klarmann, 2006). Others note that the applicability of statistical
assumes that a census of indicators of the latent variable is procedures is limited as the choice of formative indicators
undertaken at the item generation stage, and, hence, no determines the conceptual meaning of the construct (Albers and
unexplained variance exists. By fixing the disturbance term to Hildebrandt, 2006). Rossiter (2002, p. 315) dismisses any
zero the formative construct becomes a weighted linear combina- validity assessment for formative indicators claiming that “all
tion of its indicators without any surplus meaning (Diamantopou- that is needed is a set of distinct components as decided by
los, 2006; MacKenzie et al., 2005). Although there are examples expert judgment”. However, most researchers do not share the
of constructs for which all possible indicators could be above views. Edwards and Bagozzi (2000, p. 171), for example,
conceivably specified (Diamantopoulos, 2006), in most cases stress that “if measures are specified as formative, their validity
this assumption is not reasonable (Bollen and Davis, 1994) and must still be established. It is bad practice to […] claim that
therefore setting the error term to zero is not justifiable. one's measures are formative, and do nothing more”.
Finally, like all formative measurement models, also the
three higher-order models in Fig. 2 are in isolation statistically 5.4.2.1. Individual indicator validity. Bollen (1989) argues
underidentified and cannot be estimated. Since a discussion of that the γ-parameters, which reflect the impact of the formative
necessary conditions for identifying higher-order models is indicators on the latent construct (see Eq. (2)), indicate indicator
beyond the scope of this review, the reader is referred to Albers validity. The γ-parameters capture the contribution of the
and Götz (2006), Cantaluppi (2002a,b), Edwards (2001), Giere individual indicator to the construct, therefore items with non-
et al. (2006), Jarvis et al. (2003), Temme (2006), and Williams significant γ-parameters should be considered for elimination as
et al. (2003). they cannot represent valid indicators of the construct
(assuming that multicollinearity is not an issue). Diamantopou-
5.4. Reliability and validity assessment of formative models los and Winklhofer (2001) build upon Bollen's (1989) argument
and recommend using a MIMIC model due to simultaneously
5.4.1. Reliability assessment allowing for the estimation of γ-parameters and for the
As the correlations between formative indicators may be provision of an overall model fit (which is indicative of the
positive, negative or zero (Bollen, 1984; Diamantopoulos and validity of formative indicators as a set).
Winklhofer 2001), reliability in an internal consistency sense is An alternative (or additional) approach is assessing indicator
not meaningful for formative indicators (Bagozzi, 1994; Hulland, validity by estimating the indicators' correlations with an external
1999). As Nunally and Bernstein (1994) put it, “internal variable. For example, Diamantopoulos and Winklhofer (2001)
consistency is of minimal importance because two variables suggest including a global measure summarizing the essence of
that might even be negatively correlated can both serve as the construct (see also Fayers et al., 1997). Assuming that the
meaningful indicators of a construct”. Similarly, Bollen and overall measure is a valid criterion, the relationship between a
Lennox (1991) explicitly alert researchers not to rely on formative indicator and the overall measure indicates indicator
correlation matrices for indicator selection as this might lead to validity (Eggert and Fassot, 2003; MacKenzie et al., 2005).
eliminating valid measures. Following this approach, indicators correlating highly with the
While Rossiter (2002, p. 388) condemns all sorts of external variable are retained whereas those showing low or non-
reliability assessments claiming that “for a formed attribute, significant relationships are candidates for elimination.
there is […] no question of unreliability” and several other Lastly, a formative measurement model specification implies
authors skip the issue of reliability assessment when discussing that the latent variable completely mediates the effects of its
formative measure development (e.g., Diamantopoulos and indicators on other (outcome) variables (see Figs. 4 and 5). This
Winklhofer, 2001; Eggert and Fassot, 2003), Bagozzi (1994) implies certain proportionality constraints on the model
and Diamantopoulos (2005) recommend reliability assessment coefficients (Bollen and Davis, 1994; Hauser, 1973). If such
for formative indicators in form of test-retest reliability (see e.g., proportionality constraints do not hold for a particular indicator,
DeVellis, 2003; Spector, 1992). MacKenzie et al. (2005) the validity of the latter is questionable (see Franke et al., 2008-
additionally propose using the correlation between formative this issue).
1216 A. Diamantopoulos et al. / Journal of Business Research 61 (2008) 1203–1218

5.4.2.2. Construct validity. After examining validity at the models, this article hopefully lends a helping hand to researchers
individual indicator level, the next step involves assessing considering the adoption of formative measurement in their
validity at the overall construct level. An important point in this empirical efforts, while, at the same time, encouraging a critical
regard is that “causal indicators are not invalidated by low perspective in the application of formative indicators.
internal consistency, so to assess validity we need to examine Concerning future research, one major issue concerns the
other variables that are effects of the latent variable” (Bollen conceptual plausibility of formatively-measured constructs
and Lennox, 1991, p. 312, emphasis added). One common occupying endogenous positions in structural models. While a
approach is focusing on nomological (Jarvis et al., 2003; number of studies incorporate formative latent variables in such
MacKenzie et al., 2005; Reinartz et al., 2004) and criterion- positions (e.g., Edwards, 2001; Jarvis et al., 2003; MacKenzie
related (Diamantopoulos and Siguaw, 2006; Edwards, 2001; et al., 2005), Wiley (2005, p. 124, emphasis in original) notes
Jarvis et al., 2003) validity. that there is “no mechanism by which an antecedent variable
MacKenzie et al. (2005) suggest proceeding as with reflective can influence a formative index”. Since the set of causal
scales, that is, estimating hypothesized relationships of the focal indicators and the disturbance term jointly account for the total
construct with theoretically related constructs. These estimated variation of a formatively-measured construct, the specification
relationships should be consistent with the expected direction and of an additional source of variation (i.e., an antecedent con-
be significantly different from zero. Diamantopoulos and struct) is conceptually questionable. Given the conceptual and
Winklhofer (2001) also underline the importance of nomological practical importance of this issue, a debate on the use of
validation particularly in cases where indicators have been formatively-measured constructs as endogenous variables is
purified. Rossiter (2002, p. 327) challenges the approach of urgently required.
evaluating the validity of a formative index by relating it to other Another issue for future research concerns modeling forma-
constructs by arguing that “[a] scale's validity should be tively-measured constructs as moderator variables in structural
established independently for the construct”. In response, models. Although literature provides empirical examples of
Diamantopoulos (2005) points out that, by definition, all forms employing formatively specified moderators (e.g., Reinartz et al.,
of validity — with the exception of face and content validity — 2004), more research using formatively-measured constructs
are defined in terms of relationships with other measures (see when forming interaction terms is needed.
Carmines and Zeller, 1979; Zeller and Carmines, 1980). Finally, there is a debate on whether formative measurement
Concerning other types of validity assessments, Bagozzi is really necessary, that is, whether it should be used in the first
(1994, p. 338) states that “construct validity in terms of place. Bagozzi (2007, p. 236), for example, states that, for-
convergent and discriminant validity [is] not meaningful when mative measurement can be done but only for a limited range of
indexes are formed as linear sums of measurement”. In contrast, cases and under restrictive assumptions”, while Howell et al.
MacKenzie et al. (2005) suggest that standard procedures for (2007, p. 216; see also Howell et al., 2008-this issue) argue that
assessing discriminant validity are equally applicable to “formative measurement is not an equally attractive alternative
formative indexes, which include testing (a) whether the focal [to reflective measurement]”. Although there are those (includ-
construct less than perfectly correlates with related constructs, ing the authors and Bollen, 2007) who feel that, despite its
and/or (b) whether it shares less than half of its variance with various shortcomings, formative measurement is indeed a
some other construct, that is, construct intercorrelation is less viable alternative to reflective measurement based on con-
than .71 (Fornell and Larcker, 1981). ceptual grounds, further theoretical and methodological
Diamantopoulos (2006) proposes using the variance of the research is necessary to finally settle this debate. Time will tell.
error term as an indication of construct validity. Since the error
term captures aspects of the construct's domain that the set of References
indicators neglect, the lower the variance of the error term, the
more valid the construct (see also Williams et al., 2003). If the Albers S, Götz O. Messmodelle mit Konstrukten zweiter Ordnung in der
set of indicators is comprehensive in the sense that it includes all betriebswirtschaftlichen Forschung. Betriebswirtschaft 2006;66(6):669–77.
important construct facets, the construct meaning is validly Albers S, Hildebrandt L. Methodische Probleme bei der Erfolgsfaktoren-
forschung — Messfehler, formative versus reflective Indikatoren und die
captured; accordingly, the residual variance is likely to be small. Wahl des Strukturgleichungs-Modells. Zfbf 2006;58:2–33.
Finally, confirmatory tetrad analysis (CTA) (Bollen and Anderson J, Gerbing D. Some methods for respecifying measurement models to
Ting, 1993, 2000; Eberl, 2006, Gudergan et al., 2008-this issue) obtain unidimensional construct measurement. J Mark Res 1982;19(4):
offers a basic test of construct validity. Although Bollen and 453–60.
Ting (2000, p. 4) originally propose CTA as “an empirical test Arnett DB, Laverie DA, Meiers A. Developing parsimonious retailer equity
indexes using partial least squares analysis: a method and applications.
of whether a causal or effect indicator specification is J Retail 2003;79:161–70.
appropriate”, interpreting evidence supporting the latter as Bagozzi RP. A prospectus for theory construction in marketing. J Mark
also supporting the construct's validity is reasonable. 1984;48:11–29.
Bagozzi RP. Structural equation models in marketing research: basic principles.
6. Conclusion and future research In: Bagozzi RP, editor. Principles of marketing research. Oxford: Blackwell;
1994. p. 317–85.
Bagozzi RP. On the meaning of formative measurement and how it differs from
Building on a review of literature relating to the specifica- reflective measurement: comment on Howell, Breivik, and Wilcox. Psychol
tion, estimation, and validation of formative measurement Methods 2007;12(2):229–37.
A. Diamantopoulos et al. / Journal of Business Research 61 (2008) 1203–1218 1217

Bagozzi RP, Heatherton TF. A general approach to representing multifaceted Diamantopoulos A, Siguaw J. Formative versus reflective indicators in
personality constructs: application to state self-esteem. Struct Equ Modeling organizational measure development: a comparison and empirical illustra-
1994;1(1):35–67. tion. Br J Manage 2006;17(4):263–82.
Blalock HM. Causal inferences in nonexperimental research. Chapel Hill: Dowling G. Journalists' evaluation of corporate reputations. Corp Reputation
University of North Carolina Press; 1964. Rev 2004;7(2):196–205.
Blalock HM. Theory building and causal inferences. In: Blalock HM, Blalock A, Duncan OD. Notes on social measurement: historical and critical. New York:
editors. Methodology in social research. New York: McGraw-Hil; 1968. Russell Sage; 1984.
p. 155–98. Eberl M. Formative und reflektive Konstrukte und die Wahl des Strukturgle-
Blalock HM. Causal models involving unobserved variables in stimulus- ichungsverfahrens. Betriebswirtschaft 2006;66(6):651–68.
response situations. In: Blalock HM, editor. Causal models in the social Eberl M, Schwaiger M. Corporate reputation: disentangling the effects on
sciences. Chicago: Aldine; 1971. p. 335–47. financial performance. Eur J Mark 2005;39:838–54.
Bollen K. Multiple indicators: internal consistency or no necessary relationship? Edwards JR. Multidimensional constructs in organizational behavior research:
Qual Quant 1984;18:377–85. an integrative analytical framework. Organ Res Methods 2001;4(2):
Bollen K. Structural equations with latent variables. New York: Wiley; 1989. 144–92.
Bollen K, Lennox R. Conventional wisdom on measurement: a structural Edwards JR, Bagozzi R. On the nature and direction of relationships between
equation perspective. Psychol Bull 1991;110(2):305–14. constructs and measures. Psychol Methods 2000;5(2):155–74.
Bollen K. Latent variables in psychology and the social sciences. Annu Rev Eggert A, Fassot G. Zur Verwendung formativer and reflektiver Indikatoren in
Psychol 2002;53:605–34. Strukturgleichungsmodellen. Kaiserslaut Schr reihe Mark 2003;20:1–18.
Bollen K. Interpretational confounding is due to misspecification, not to type of Fassot G. Operationalisierung latenter Variablen in Strukturgleichungsmodellen:
indicator: comment on Howell, Breivik, and Wilcox. Psychol Methods Eine Standortbestimmung. Zfbf 2006;58:67–88.
2007;12(2):219–28. Fassot A, Eggert G. Zur Verwendung formativer und reflektiver Indikatoren in
Bollen K, Ting K. Confirmatory tetrad analysis. In: Marsden PV, editor. Strukturgleichungsmodellen: Bestandaufnahme und Anwendungsempfeh-
Sociological methodology. Washington, D.C: American Sociological lung. In: Bliemel FW, Eggert A, Fassot G, Henseler J, editors. Handbuch
Association; 1993. p. 147–75. PLS-Modellierung, Methode, Anwendung, Praxisbeispiele. Stuttgart:
Bollen K, Davis W. Causal indicator models: identification, estimation, and Schaeffer-Poeschel; 2005. p. 31–47.
testing. Paper presented at the American Sociological Association Conven- Fayers PM, Hand DJ, Bjordal K, Groenvold M. Causal indicators in quality of
tion, Miami; 1994. life research. Qual Life Res 1997;6:393–406.
Bollen K, Ting K. A tetrad test for causal indicators. Psychol Methods 2000;5(1): Fornell C, Larcker DF. Evaluating structural equation models with unobservable
3–22. variables and measurement error. J Mark Res 1981;18:39–50.
Brewer P. Operationalizing psychic distance: a revised approach. J Int Mark Fornell C, Bookstein FL. A comparative analysis of two structural equation
2007;15(1):44–66. models: LISREL and PLS applied to market data. In: Fornell C, editor. A
Brock JK, Zhou Y. Organizational use of the internet — scale development and second generation of multivariate analysis, vol. 1. New York: Praeger; 1982.
validation. Internet Res 2005;15(1):67–87. p. 289–324.
Bruhn M, Georgi D, Hadwich K. Customer equity management as a formative Fornell C, Rhee BD, Yi Y. Direct regression, reverse regression, and covariance
second-order construct. Journal of Business Research 2008;61:1292–301 structure analysis. Mark Lett 1991;2(3):309–20.
(this issue). doi:10.1016/j.jbusres.2008.01.016. Franke G, Preacher C, Rigdon E. The proportional structural effects of formative
Cadogan JW, Souchon AL, Procter DB. The quality of market-oriented indicators. Journal of Business Research 2008;61:1229–37 (this issue).
behaviors: formative index construction and validation. Journal of Business doi:10.1016/j.jbusres.2008.01.011.
Research 2008;61:1263–77 (this issue). doi:10.1016/j.jbusres.2008.01.014. Gardner D, Cummings L, Dunham R, Pierce J. Single-item versus multiple-
Cantaluppi G. Some further remarks on parameter identification of structural equation item measurement scales: an empirical comparison. Educ Psychol Meas
models with both formative and reflexive relationships. In: A.A., V.V., editors. 1998;58(6):898–915.
Studi in onore di Angelo Zanella. Milano: Vita e Pensiero; 2002a. p. 89–104. Giere J, Wirtz B, Schilke O. Mehrdimensionale Konstrukte: Konzeptionelle
Cantaluppi G. The problem of parameter identification of structural equation Grundlagen und Möglichkeiten ihrer Analyse mithilfe von Strukturgle-
models with both formative and reflexive relationships: some theoretical ichungsmodellen. Betriebswirtschaft 2006;66(6):678–95.
results. Serie Edizioni Provvisorie 2002b; No. 108, Istituto di Statistica, Götz O, Liehr-Gobbers K. Analyse von Strukturgleichungsmodellen mit Hilfe
Università Cattolica del S. Cuore, Milano:1–19. der Partial-Least-Squares(PLS)-Methode. Betriebswirtschaft 2004;64(6):
Carmines EG, Zeller RA. Reliability and validity assessment. In: Sullivan JL, 714–38.
editor. Quantitative applications in the social sciences. Beverly Hills: Sage; Gudergan SP, Ringle CM, Wende S, Will A. Confirmatory tetrad analysis for
1979. evaluating the mode of measurement models in PLS path modeling. Journal
Churchill GA. A paradigm for developing better measures of marketing of Business Research 2008;61:1238–49 (this issue).
constructs. J Mark Res 1979;16:64–73. Hair JF, Anderson RE, Tatham RL, Black WC. Multivariate data analysis. New
Cohen P, Cohen J, Teresi J, Marchi M, Velez CN. Problems in the measurement Jersey: Prentice Hall; 1998.
of latent variables in structural equations causal models. Appl Psychol Meas Hauser RM. Diaggregating a social-psychological model of educational
1990;14:183–96. attainment. In: Goldberger AS, Duncan OD, editors. Structural equation
Collier JE, Bienstock CC. Measuring service quality in e-retailing. J Serv Res models in the social sciences. San Diego: Academic Press; 1973. p. 255–89.
2006;8(3):260–75. Hauser RM, Goldberger AS. The treatment of unobservable variables in path
Curtis RF, Jackson EF. Multiple indicators in survey research. Am J Sociol analysis. Sociol Method 1971:81–117.
1962;68:195–204. Heise DR. Employing nominal variables, induced variables, and block variables
DeVellis Robert F. Scale development — theories and applications. Applied in path analysis 1972; 1:147–173.
social research methods series. 2nd edition. Sage Publications; 2003. Helm S. Designing a formative measure for corporate reputation. Corp
Diamantopoulos A. Viewpoint: export performance measurement: reflective Reputation Rev 2005;8(2):95–109.
versus formative indicators. Int Mark Rev 1999;16(6):444–57. Hitt MA, Gimeno J, Hoskisson RE. Current and future research methods in
Diamantopoulos A. The C-OAR-SE procedure for scale development in strategic management. Organ Res Methods 1998;1:6–44.
marketing: a comment. Int J Res Mark 2005;22:1–9. Homburg C, Klarmann M. Die Kausalanalyse in der empirischen betriebs-
Diamantopoulos A. The error term in formative measurement models: inter- wirtschaftlichen Forschung – Problemfelder und Anwendungsempfehlun-
pretations and modelling implications. J Modell Manage 2006;1(1):7–17. gen. Betriebswirtschaft 2006;66(6):727–48.
Diamantopoulos A, Winklhofer H. Index construction with formative indicators: Homburg C, Workman JP, Krohmer H. Marketing's influence within the firm.
an alternative to scale development. J Mark Res 2001;38(2):269–77. J Mark 1999;63(2):1–17.
1218 A. Diamantopoulos et al. / Journal of Business Research 61 (2008) 1203–1218

Homburg C, Hoyer W, Fassnacht M. Service orientation of a retailer's business Podsakoff NP, Shen W, Podsakoff PM. The role of formative measurement
strategy: dimensions, antecedents, and performance outcomes. J Mark models in strategic management research: review, critique, and implications
2002;66(4):86–101. for future research. Res Methodol Strat Manag 2006;3:197–252.
Howell RD, Breivik E, Wilcox JB. Reconsidering formative measurement. Reinartz W, Krafft M, Hoyer WD. The customer relationship management pro-
Psychol Methods 2007;12(2):205–18. cess: Ist measurement and impact on performance. J Mark Res 2004;41(3):
Howell RD, Breivik K, Wilcox JB. Questions about formative measurement. 293–305.
Journal of Business Research 2008;61:1219–28 (this issue). doi:10.1016/j. Rossiter J. The C-OAR-SE procedure for scale development in marketing. Int J
jbusres.2008.01.010. Res Mark 2002;19:305–35.
Hulland J. Use of partial least squares (PLS) in strategic management research: a Ruiz DM, Gremler DD, Washburn JH, Capeda-Carrion G. Service value revisited:
review of four recent studies. Strateg Manage J 1999;20:195–204. specifying a higher-order, formative measure. Journal of Business Research
Hyman M, Ganesh G, McQuitty S. Augmenting the household influence 2008;61:1278–91 (this issue). doi:10.1016/j.jbusres.2008.01.015.
construct. J Mark Theory Pract 2002;10(3):13–31. Sànchez-Pérez M, Iniesta-Bonillo M. Consumers felt commitment towards
Jarvis C, MacKenzie S, Podsakoff PA. Critical review of construct indicators retailers: index development and validation. J Bus Psychol 2004;19(2):
and measurement model misspecification in marketing and consumer 141–59.
research. J Consum Res 2003;30(2):199–218. Santosa PI, Wei KK, Chan HC. User involvement and user satisfaction with
Johansson JK, Yip GS. Exploiting globalization potential: U.S. and Japanese information-seeking activity. European Journal of Information Systems:
strategies. Strateg Manage J 1994;15(8):579–601. including a special section on the Pacific Asia conference, vol. 14(4). ; 2005.
Johnson GJ, Bruner II GC, Kumar A. Interactivity and its facets revisited. J Advert p. 361–70.
2006;35(4):35–52. Scholderer J, Balderjahn I. Was unterscheidet harte und weiche Strukturgle-
Jöreskog K, Goldberger A. Estimation of a model with multiple indicators and ichungsmodelle nun wirklich? Mark ZFP 2006;28(1):57–70.
multiple causes of a single latent variable. J Am Stat Assoc 1975;10:631–9. Spector PE. Summated rating scale construction: an introduction. Series:
Judge TA, Bretz RD. Person-organization fit and the theory of work adjustment: quantitative applications in the social sciences. CA: Sage Publications; 1992.
implications for satisfaction, tenure, and career success. J Vocat Behav Temme D. Die Spezifikation und Identifikation formativer Messmodelle der
1994;44(1):32–54. Marketingforschung in Kovarianzstrukturanalysen. Mark ZFP 2006;28(3):
Kennedy PA. Guide to econometrics. 5th edition. Boston: MIT Press; 2003. 183–209.
Land K. On estimation of path coefficients for unmeasured variables from Ulaga W, Eggert A. Value-based differentiation in business relationships:
correlations among observed variables. Soc Forces 1970;48:506–11. gaining and sustaining key supplier status. J Mark 2006;70(1):119–36.
Law K, Wong C. Multidimensional constructs in structural equation analysis: an Venaik S, Midgley DF, Devinney TM. A new perspective on the integration-
illustration using the job perception and job satisfaction constructs. J Manage responsiveness pressures confronting multinational firms. Manag Int Rev
1999;25(2):143–60. 2004;44(Special Issue 2004/1):15–48.
Law KS, Wong CS, Mobley WH. Toward a taxonomy of multidimensional Venaik S, Midgley DF, Devinney TM. Dual paths to performance: the impact of
constructs. Acad Manage Rev 1998;23(4):741–55. global pressures on MNC subsidiary conduct and performance. J Int Bus
Lin CH, Sher PJ, Shih HY. Past progress and future directions in conceptualizing Stud 2005;36(6):655–75.
customer perceived value. Int J Serv Ind Manag 2005;16(4):318–36. Wiley J. Reflections on formative measures: conceptualization and implication
Long JS. Confirmatory factor analysis: a preface to LISREL. Bloomington, IN: for use. ANZMAC Conference, Perth; December 5–7; 2005.
Sage Publications; 1983. Williams LJ, Edwards JR, Vandenberg RJ. Recent advances in causal modeling
Lord FM, Novick MR. Statistical theories of mental test scores. Reading, MA: methods for organizational and management research. Journal of Manage-
Addison-Wesely; 1968. ment 2003;29(6):903–36.
MacCallum R, Browne M. The use of causal indicators in covariance structure Williams LJ, Gavin MB, Hartman NS. In: Ketchen DJ, Berg DD, editors. Structural
models: some practical issues. Psychol Bull 1993;114(3):533–41. equation modeling methods in strategy research: applications and issues.
MacKenzie SB. The danger of poor construct conceptualization. J Acad Mark Research Methodology in Strategy ManagementBoston, MA: Elsevier; 2004.
Sci 2003;31(3):323–6. p. 303–46.
MacKenzie S, Podsakoff P, Jarvis C. The problem of measurement model Winklhofer H, Diamantopoulos A. Managerial evaluation of sales forecasting
misspecification in behavioural and organizational research and some effectiveness: a MIMIC model approach. Int J Res Mark 2002;19:151–66.
recommended solutions. J Appl Psychol 2005;90(4):710–30. Witt P, Rode V. Corporate brand building in start-ups. J Enterp Cult 2005;13(3):
Namboodiri NK, Carter LF, Blalock HM. Applied multivariate analysis and 273–94.
experimental designs. New York: McGraw-Hill; 1975. Yi MY, Davis FD. Developing and validating an observational learning model of
Netemeyer RG, Bearden WO, Sharma S. Scaling procedures. CA: Sage; 2003. computer software training and skill acquisition. Inf Syst Res 2003;14(2):
Nunnally JC. Psychometric theory. 2nd edition. New York: McGraw-Hill; 1978. 146–69.
Nunnally JC, Bernstein IH. Psychometric theory. 3rd edition. New York: Zeller RA, Carmines EG. Measurement in the social sciences — the link
McGraw-Hill; 1994. between theory and data. CA: Cambridge University Press; 1980.
Pavlou P, Gefen D. Psychological contract violation in online marketplaces:
antecedents, consequences, and moderating role. Inf Syst Res 2005;16(4):
372–99.
Podsakoff PM, MacKenzie SB, Podsakoff NP, Lee JY. The mismeasure of man
(agement) and its implications for leadership research. Leadersh Q
2003;14:615–56.

You might also like