An Empirical Investigation of the Structural Antecedents of Perceived Value in a Heterogeneous Population. Wayne DeSarbo The Pennsylvania State University Kamel Jedidi Columbia University and Indrajit Sinha Temple University
ISBM Report 18-1998
Institute for the Study ofBusiness Markets The Pennsylvania State University 402 Business Administration Building University Park, PA 16802-3004 (814) 863-2782 or (814) 863-0413 Fax
This publication is available in alternative media on request. The Pennsylvania State University is committed to the policy that all persons shall have equal access to programs, facilities, admission, and employment without regard to personal characteristics not related to ability, performance, or qualffications as determined by University policy or by state or federal authorities. The Pennsylvania State University does not discriminate against any person because of age, ancestry, color, disability or handicap, national origin, race, religious creed, sex, sexual orientation, or veteran status. Direct all inquiries regarding the nondiscrimination policy to the Affirmative Action Director, the Pennsylvania State University, 201 Willard Building, University Park, PA 16802-2801; Tel. (814) 8654700/V; (814) 865-1150/ITY. U.Ed. BUS 98-026
An Empirical Investigation of the Structural Antecedents of Perceived Value in a Heterogeneous Population
September 1998
Wayne S. DeSarbo
*
Kamel Jedidi Indrajit Sinha *
The authors would like to thank the Institute for the Study ofBusiness Markets and Gary Lilien for their financial
support of this research.
AUTHOR INFORMATION
Wayne S. DeSarbo is the Smeal Distinguished Research Professor of Marketing at the Smeal College of Business of Pennsylvania State University in University Park, PA. Kamel Jedidi is Associate Professor of Marketing at the Graduate School of Business of Columbia University in New York, NY. Indrajit Sinha is Assistant Professor of Marketing at the School of Business and Management of Temple University in Philadelphia, PA.
Author Contact:
Wayne S. DeSarbo Marketing Department Smeal College of Business Pennsylvania State University University Park, PA 16802 Phone: 814-865-8150 Fax: 814-364-2573 e-mail:
[email protected]
An Empirical Investigation of the Structural Antecedents of Perceived Value in a Heterogeneous Population
ABSTRACT
In recent years, Customer Value Management (CVM) has become a major focus in both consumer and business-to-business marketing as a core strategy underlying all loyalty generating programs.
We develop a new approach for the structural analysis of the antecedent factors of
perceived value, (i.e., perceived quality and perceived price) through a recursive simultaneous equation model that is formulated to accommodate heterogeneity. In particular, the proposed latent structure methodology allows one to estimate the relative effects and integration rules of perceived value drivers at the market segment level, as well as to simultaneously determine the (unknown) segments themselves. We demonstrate the utility of the proposed methodology via an actual commercial application involving a CVM study for a large electric utility company. Finally, we discuss the contributions ofour research and how it may be extended in the future.
Key Words: Perceived Value Market Segmentation Latent Structure Analysis Simultaneous Recursive Equations Customer Value Management
INTRODUCTION In an increasingly competitive global, yet fragmented, marketplace, more and more firms have been focusing on customer value and loyalty as their principal salvation in an era of slow-
growth, pricing pressures, diminishing margins, and hard-won profits. Many companies frequently implement various customer-based surveys, relationship-marketing programs, and data-mining approaches to identify and keep their most loyal and profitable market segments and also to reduce their “chum rates.” For example, it has been reported that MBNA credit-card managers now listen in to customer complaint calls, State Farm ties its agent-compensation structure directly to customer retention, Sears encourages more frequent purchases by new and expectant parents at its stores through its nationwide “Kidvantage” program, and Home Depot estimates that while its “best” customer spends only $38 per visit, he/she comes in about 30 times a year, contributing $25,000 over his/her lifetime (Rakstis 1996; Reicbheld 1996). The basic principle underlying all such loyalty-based strategies is that businesses should deliver superior perceived customer value on a regular and consistent basis, and that managers may achieve this through instituting appropriate “Customer Value Management” (CVM) programs (Gale 1994; Kotler 1996; Naumann 1995; Reichheld 1996). Both perceived customer value and satisfaction have long been regarded as fundamental marketing concepts, and lately customer value has become a popular measurement tool (especially among marketing research practitioners) in representing the “voice of the customer” (see, e.g. Fredericks and Salter 1995; Gelb 1998; Jones and Sasser 1995; Reichheld 1996; Woodruff 1997). It has been reported that firms such as Fedex, AT&T, and Xerox have been able to better understand and predict customers’ loyalty and repeat purchase behavior by measuring their value perceptions (Gale 1994). Moreover, it has been widely reported in the business press how the American consumer is enamored with the concept of “value” (i.e. “getting more for the money”) and how companies (e.g. Toyota, Oldsmobile, Taco Bell, etc.) have been outdoing each other in emphasizing in their advertisements that only their products and services deliver value to their customers (Jacob 1993; Mason 1992; Sherman 1992). As Woodruff (1997) has noted, if current trends in manufacturing over-capacity as well as consumer resistance to price increases were to continue, customer value will become the “next source of competitive advantage” in the 2l~~ century.
1
-c
Consistent with the growing interest in this construct, recently several authors in marketing have published papers that have sought to explicate the nature of perceived customer value (Sinha and DeSarbo 1998; Zeithaml 1988) as well as the process of its formation in the minds of consumers (Bolton and Drew 1991; Chang and Wildt 1994; Dodds, Monroe, and Grewal 1991; Grewal, Monroe, and Krishnan 1998). One key insight gleaned from analyzing consumer “definitions” of perceived value is that value perceptions are idiosyncratically developed vis-A-vis both the consumer and the product/service category (Zeithaml 1988). In other words, corresponding to a particular product or service there are indeed heterogeneous interpretations of perceived value, and multiple consumer segments may assign differential importance weights to perceived quality and price as to “what constitutes value for them” in a specified category. As one article in the business press (Jacob 1993) has simply put it , “value can mean different things to different people”. Despite this, while earlier efforts at explicating the mechanism by which perceived value is formed (Dodds, Monroe, and Grewal 1991; Chang and Wildt 1994; Grewal, Monroe, and Krishnan 1998) have yielded important findings (discussed below), these papers do not explicitly accountfor consumer heterogeneity and effectively assume that consumers are all affected by the antecedent factors in the same manner. In fact, Grewal, Monroe, and Krishnan (1998) have mentioned that an important contribution would be to account for consumer differences (i.e. resulting from deal-proneness, price consciousness, attitude to risk, etc.) in the weighting of perceived value antecedents. From the manager’s perspective, the ability to ascertain the relative magnitude of perceived value “drivers” for different consumer segments should be particularly beneficial. Therefore, a key issue in customer value analysis is the accurate and reliable determination of the relative impact of these antecedent factors in perceived value formation. An aggregate analysis (ignoring heterogeneity) can yield misleading estimates of these impacts, and provide an erroneous view of the marketplace. Grewal, Monroe, and Krisbnan (1998) also highlight the question of the functional relationship of perceived quality and price (whether ratio or subtractive) as an unresolved issue in the literature, meriting further study. Additionally, we may note nearly all previous papers in this area have utilized convenience samples (comprised mainly of student subjects) for model calibration. Clearly, given the nature of the construct and potentially significant implications for practitioners, more meaningful and practicable insights can be gained from utilizing actual field data.
2
N.
This paper represents an attempt to address many of the above-mentioned gaps in the literature. Consistent with theoretical suppositions and prior evidence, we accommodate customer heterogeneity by developing a new latent-structure simultaneous-equation modeling approach to investigate the effects of perceived value antecedents and their differential weighting by (unknown) consumer segments. Subsequently, we are able to profile the derived segments from individual descriptor variables. Our methodology also permits different functional forms of perceived value to be specified and tested, thereby allowing the investigation of the issue of “information integration” as discussed by Grewal, Monroe, and Krishnan (1998). Note that Sinha and DeSarbo (1998) have presented a new latent structure multidimensional scaling approach to simultaneously identify the dimensions of perceived value as well as map individual brands and segments, but these authors (by their own admission) do not model the structural process of perceived value formation. Furthermore, they do not make use of a “real-world” dataset, which we do in this paper. We devise a finite mixture-based, latent structure approach (e.g. Titterington, Smith, and Makov 1985) to explain perceived value formation in a heterogeneous population and calibrate the model on data obtained from customers of a major electric utility company. The net result is a better understanding of the mechanism by which consumers integrate different factors resulting in a perception ofvalue.
LITERATURE REVIEW
In the marketing literature, perceived value typically has been defined as a tradeoff of perceived quality and perceived price. Perceived quality, in turn, has been conceptualized as buyers’ ‘judgment about a product’s overall excellence or superiority” (Zeithaml 1988, p. 3), and perceived price is defined as the consumers’ subjective perception of the objective price of the product (Jacoby and Olson 1977). As Monroe (1990, p. 46) states, “buyers’ perceptions of value represents a tradeoff between the quality or benefits they perceive in the product relative to the sacrifice they perceive by paying the price.” A comparable view is taken in industry where perceived value has been variously defined as “quality at the right price” or as “affordable quality” (Progressive Grocer 1984). This may be confirmed further in Gale’s (1994) “customer
3
N.
value maps” where quality and price comprise the two axes, and in Naumann’s (1995) “customer value triad” where perceived product quality, perceived service quality, and perceived price form the three cardinal points. From an empirical perspective, the thesis that perceived value is a function of perceived quality and perceived price has received broad support. For instance, Chang and Wildt (1994) hypothesize that perceived value is positively related to quality but negatively to price, and find that both these effects are significant in a structural-equation framework. Similarly, in attempting to infer the dimensionality of perceived value from customer categorization ratings, Sinha and DeSarbo (1998) report the existence of only two latent factors or dimensions: one, a quality (or benefits) dimension and the other dimension loading mostly on cost/price. Comparable findings may be seen in the Grewal, Monroe, and Krishnan’s (1998) study. While Dodds, Monroe, and Grewal (1991) also have assumed that perceived value is a function of perceived quality and perceived price, they posit an interesting “twist” in their relationships. According to these authors, value should be related monotonically (and positively) with quality, but quadratically (inverted-U) with price. They argue that price is a signal of both quality and sacrifice to customers. Hence, at low prices the perception of poor quality may outweigh that of lower sacrifice causing perceived value to be less, whereas at high prices the perception of greater sacrifice may sweep that of good quality, resulting again in lower perceived value. Only at intermediate prices will the perception of value will be high, and hence the inverted-U relationship. Nonetheless, in their experimental results, these authors find that value is related linearly with both the factors, being positively related to quality but negatively with price. Consequently, in our model set-up (described below), we specify perceived value as a linear function of both of these variables. In subsequent sections of this manuscript, we discuss three salient aspects of the perceived value literature: (1) the dependence of perceived value on lower-level physical attributes; (2) consumer heterogeneity in the integration of perceived value antecedents; and, (3) the nature ofthe integration rules themselves, i.e., the functional form of perceived value.
4
Dependence on lower-level objective attributes While perceived value is an integration of such “higher level abstractions” as perceived quality and perceived price, perceived quality has been shown to be formed from the lower-level, tangible attributes of the product/service. As Zeithaml (1988) explains, in a means-end chain, quality is a consumer’s cognitive assessment of the excellence of a product that is developed as a second-order phenomenon from the “objective” (intrinsic and extrinsic) attributes ofthe product. To this extent, perceived quality may be considered a somewhat more complex, multidimensional, and abstract “B-attribute” (Olson and Reynold 1983). Chang and Wildt (1994, p. 18) also observe that product attribute information represent the “elemental or disaggregate form of information, whereas perceived quality can be considered a summarization of this information” (see also Olson and Jacoby 1972). In a similar vein, perceived price symbolizes the encoding or internalization of the objective price of the product/service (Jacoby and Olson 1977; Mazumdar and Monroe 1990). Based on their empirical findings, Chang and Wildt (1994) report that perceived quality is indeed significantly affected by objective product attributes, and perceived price is significantly related to actual price. Therefore, in our modeling formulation, we assume perceived quality to be a weighted combination of the customer perceived attribute performance/information. An important improvement of our approach over Chang and Wildt (1994) is that, given our latent structure framework, we are able to allow for dWferential weighting/importance of attributes by different (unknown) consumer segments. The theoretical rationale and empirical evidence for this supposition is detailed below.
Consumer heterogeneity in the integration ofperceived quality andperceivedprice A major omission in previous papers examining the process of perceived value formation has been their authors’ failure to incorporate consumer’s heterogeneity in the integration of the underlying dimension of value. In other words, they have implicitly assumed the effects of the antecedents of perceived value to be the same for all consumers, whereas actual field evidence has demonstrated otherwise. For instance, based on a comprehensive exploratory study comprising focus-groups and in-depth interviews of adult consumers, Zeithaml (1988) reports considerable heterogeneity among subjects in their perceived value interpretations. While one segment perceived the value of a product only from its quality, another segment judged it only from price, a third from both quality and price, and so on. Sinha and DeSarbo (1998), who
5
N
develop a new, latent class MDS-based perceived value-mapping methodology (as an improvement over Gale 1994), infer the presence of two distinct segments in their data. Grewal, Monroe, and Krishnan (1998) also state that consumer differences resulting from such factors as price consciousness, deal proneness, etc. should affect their perceptions. Even in industry, this view that customer value is heterogeneous is prevalent. Bishop (1984) remarks that different consumers place differing emphasis on the underlying constituents of perceived value. As one insightful IBM executive quipped recently, “It (value) is not as simple as price, or we’d all be driving Hyundais” (Sherman 1992, p. 91). There are two notable consequences from this omission. First, from a purely statistical point of view, the modeling of a diverse consumer base (consisting of a mixture of different distributions) as one homogeneous population necessarily will yield inconsistent and unreliable estimates (see Titterington, Smith, and Makov 1985). Second, from the practitioner’s perspective, the failure to represent consumer differences in perceived value interpretations will present an inaccurate and misleading view of the market.
The diagnostic value of such a
simplified model in terms of managerial actionability vis-~-vis the drivers of perceived value will be severely limited unless consumer heterogeneity is explicitly accounted for. We accomplish this via a latent structure methodology that allows for simultaneous estimation of effects of antecedent factors of perceived value and consumer differences in how these factors are integrated. Unlike Sinha and DeSarbo (1998), we start with the assumption that perceived value is a function of quality and price and investigate its formation process across different (unknown) market segments.
Functionalform ofperceived value Finally, an interesting empirical question that has not been fully resolved concerns the question of the functional form of perceived value: how the perceived value antecedents, i.e. quality and price, are integrated in the minds of consumers. Monroe (1990) suggests a ratio or proportional specification, i.e. v
=
q p
—,
implying the perceived value is judged to be quality at
unit price in consumers’ minds. Other authors (e.g. Hagerty 1978, Levin and Johnson 1984) argue for a subtractive formation, i.e. v
=
q p. The latter form implies a linear, compensatory —
rule in which consumers integrate price and quality in an subtractive manner, and that they
6
:
subtract perceptions of higher price from those of higher quality. Between the two functional forms, the difference or subtractive rule seems to have received broader support. For instance, Grewal, Monroe, and Krishnan (1998) test both forms and report that the difference form provided a better fit to their data. However, they call for more systematic exploration of this issue in future research. Our proposed latent structure model framework permits for the testing of alternative specifications ofperceived value to explicitly address this issue.
AN EMPIRICAL INVESTIGATION AN ELECTRIC UTILITY COMPANY Customer satisfaction and perceived value measurement studies in a business-to-business context often involve multiple research objectives. Managers of a company seek to measure and track performance on key business dimensions in an attempt to integrate customer requirements throughout their organization. Additionally, there is a desire to identify aspects of the business that have the most impact on overall measures or indices of customer opinion or perceptions. The impact of these elements often are identified through the application of statistical procedures which model an overall performance measure as a function of more specific attribute measurements. For example, in many CVM studies, while overall value perceptions are modeled as a function ofperceived quality and perceived price, perceived quality is typically modeled as a function of customer perceived performance or satisfaction on specified product and service attributes. (In fact, this is the recursive framework to be employed with respect to the electric utility company as to be described). However, there are concerns associated with this aggregate analysis strategy. The aggregate analysis may inappropriately pool members from heterogeneous subpopulations resulting in parameter estimates that are inconsistent. Further, analysis by a-priori segments/groups (e.g., formed on the basis of firmographics) provides no assurance that an optimal segmentation is in place with respect to the CVM model of interest. Segments may be easy to understand and reach, but may not produce differentiated CVM models with the best possible explanatory power. That is to say, these a-priori segments may be firmographically
7
different but may not behave differently with respect to the drivers or the manner in which they form their value perceptions. It is from this perspective that our proposed method may provide maximum benefit.
The CVM Study The CVM data for our application was obtained from a survey of commercial and industrial customers of a large electric utility company (Bolton and Drew 1991; DeSarbo and Grissaffe 1998). Customers of this company received a notification letter one week prior to the interviewing to alert them that a call would be coming shortly concerning their customer opinions.
Following this mailing, customers were contacted and a survey (entailing
approximately 20 minutes per respondent) was administered via telephone interview.
The
survey and data collection procedure was conducted over a 15-month period. A total of 1509 cases were thus obtained for modeling purposes. We formulate a recursive simultaneous equation model employing two equations. The first equation specifies perceived value (VALUE) as a function of perceived price (PRICE) and perceived quality (QUAL). The second equation models perceived quality as a function of ratings of power reliability (PROD), preventative maintenance (MAINT), repair service (REPAIR), account representative (ACCT), technical support (TECH), customer service (SVC), record keeping (REC), and billing (BIL). A variety of firmographic and demographic variables were also collected including measures of sales region (REGl-REG6),
account type
(ACCTYPE1-ACCTYPE5), business type
(BUSTYPEl-
BUSTYPE4), respondent job type (PROF 1-PROF4), presence of relations with other suppliers (RELATION), number of employees (NUMEMP), number of years as a customer (NUMYBARS), and a standardized measure of revenue coming from the particular account (REVENUES). We refrain from providing a more detailed description of these firmographic variables given the confidentiality ofthe data.
The Finite-Mixture Simultaneous Equation Model We employ a two equation, latent structure simultaneous equation approach to model the process of perceived value formation. As mentioned above, the first equation relates perceived value to perceived price and perceived quality; the second equation relates perceived quality to
8
perceived service attributes. We capture customer heterogeneity in this process through a finite-
mixture approach (Jedidi, Ramaswamy, DeSarbo, and Wedel, 1996) which derives segments of customers who are homogeneous in terms of their response parameters. Let i index customers (i=1,
...
,
N) and g denote membership in a (a-priori unknown) segment (g=1,
...
,
G).
Conditional on membership in segment g, we postulate the following recursive simultaneous equation model: VALUE~ Ig
QUAL~
Ig
y~+/3f2QUAL,+y~PRICE, +e~ 72%
=
+7~ PROD, ~72%MAINI +72% REPAIR, ~72~ ACCT +72% TECH,
+ 72% SVC, +y2% REC,
+72%
BIL1 +e
(1)
~,
where yjg0 and y~ are intercept parameters, y~is a parameter that measures the impact of price
on value,
fi~
is a parameter that captures the simultaneity between perceived value and
perceived quality and
72%
,. • .,
y~8 are parameters that reflect the effects of product and service
attributes on perceived quality. e ~ and s
are error terms that jointly follow a bivariate normal
distribution with zero meanvector and covariance matrix:
g
[~ ~j.
(2)
Before proceeding to model estimation, it is necessary to establish that this specified latent structure simultaneous equation model is identified. Jedidi, Jagpal, and DeSarbo (1997a, b) have demonstrated that finite mixture of structural equation models are identified provided that the structural equation model is identified for known segments and the data (within segment) follow multivariate normal distributions.
As we indicate below, our model satisfies both
conditions and is therefore identified. First, this model is identified for known segments based
[~ i,’]
on the sufficient rank condition (see Bollen 1989 p. 98). assumption that the error vector
=
Second, we explicitly make the
follows a bivariate normal distribution.
9
Estimation
J
Let yig.~{VALUElI~ denote the joint vector of endogenous variables conditional on membership in segment g. Then its conditional mean vector is defined by:
~ig
L
~ +~ 72%
QUAL1
+7~ PRD,
+72%
SVC,
+
y~ PRICE1
+72%
+72%
A/MINT, +y~ REPAIR, +y~ ACCT,
REC,
72%
+72%
TECH]
(3)
BIL1
and its conditional covariance matrix is ‘I’s,. By assumption y, Ig follows a conditional bivariate normal distribution. Hence the unconditional distribution of the observed vector y g is a finite mixture of these distributions. That is:
where w = (w,
,...,
w0)’ is the vector of the G mixing proportions such that wg
>
0 and
G
XWg g=I
=1, and fg(.) is the bivariate normal density function. The likelihood function for a
sample (y,
One ‘Pg~ Fg
now
...~
YN)’
ofN randomly drawn observations is:
maximizes L
=(y~o,Y~1,Y’~o,...,Y~s)’,
(or
ln
L)
with
respect
to
and w given the sample data
account the constraints on wand ‘Pg >0 for all g.
10
the
free
(yl...,YN)’
parameters
13%,
while taking into
V
We use a modified EM algorithm (see Jedidi et al. 1996 for details) to maximize L. This method is suited for models that deal with unobserved data (e.g., segment membership). The BM method maximizes the complete data likelihood function (formed by assuming that segment membership is assumed to be known) by iterating between an Expectation step and a Maximization step until convergence. In the E-step, the segment memberships are estimated by their expected values given provisional estimates for ‘Pg~ Fg~ provisional estimates for
‘Pg
,
Fg~
13~’
and w. In the M-step, the
13~, and w are updated in light of the newly estimated values
of segment memberships. Within any iterate, one can use the estimates for ‘Ps, Fg, j3~2,and w to compute the posterior probability of membership:
(6)
to assign each of the individual observations in each of the G segments.
We compute the
asymptotic standard errors using the inverse of the empirical information matrix (see Meilij son 1989). We infer the appropriate number of segments G by running the estimation procedure for varying number of segments. We choose the solution that corresponds to the minimum value of Bozdogan’s (1987) Consistent AIC criterion:
CAIC~
-2LnLG+MGLn(N+ 1),
(7)
where Mu is the number of free parameters in a G-segment model. We use an entropy-based measure E~ to assess the quality of separation among the segments (see Ramaswamy, DeSarbo, Reibstein, and Robinson 1992). Specifically, E~ is defined by:
FN
G
11
This measure is bounded between 0 and 1. A value close to 0 indicates that the posterior probabilities are not well separated (i.e., it is difficult to classify observation accurately into distinct segments).
Results Table 1 presents the various summary statistics and goodness of fit indices for g= 1,2,3,4 market segments. According to the CAIC criterion, G=3 segments appears to be the most parsimonious solution. Notice how the G1 aggregate solution is rejected in this analysis suggesting sufficient sample heterogeneity to warrant further disaggregation. Table 2 depicts the one
-
(aggregate solution) and three segment solutions in terms of coefficients, standard errors -
(in parentheses), and mixing proportions (size of the segments). Note that the signs of the significant parameters are all in the expected direction in that perceived value is related positively to perceived quality (and all service attribute variables), but is associated negatively with perceived price. For the aggregate solution, virtually all coefficients are significant (except customer service) at p
