This perspec- tive specifies information, identification, and volition as antecedent processes of commitment that facilitate its root tendency, resistance to change.. With this in mind,
Trang 1Analyzing the Commitment-Loyalty
Link in Service Contexts
Mark P, Pritchard
Arizona State University
Mark E Havitz
University of Waterloo
Dennis R Howard
University of Oregon
This study addressed the ill-understood issue of how loy-
alty develops in service patrons Although many theorists
hold commitment to be an essential part of this process, the
link between commitment and loyalty has received little
empirical attention To address this void, the study first
portrayed commitment's root tendency to resist changing
preference as a function of three antecedent processes
Second, this portrayal formed the basis for developing a
psychometrically sound scale to measure the construct of
commitment Third, the scale was then used in a mediating
effects model (M-E-M) to test the commitment-loyalty link
Path analyses found this parsimonious structure to be a
significant improvement over rival direct effects models
(D-E-Ms) Results found the tendency to resist changing
preference to be a key precursor to loyalty, largely ex-
plained by a patron's willingness to identify with a brand
Implications of these findings for loyalty's development
and research are explored
Understanding how or why a sense of loyalty develops
in customers remains one of the crucial management
issues of our day While early interest in loyalty sought to
understand why consumers repeatedly preferred certain
brands of low-priced retail goods (e.g., Day 1969; Jacoby
Journal of the Academy of Marketing Science
Volume 27, No 3, pages 333-348
Copyright 9 1999 by Academy of Marketing Science
1971), attention has since shifted to look at this feature in service patrons (e.g., Pritchard and Howard 1997) In increasingly competitive markets, being able to build loy- alty in consumers is seen as the key factor in winning mar- ket share (Jarvis and Mayo 1986) and developing sustain- able competitive advantage (Kotler and Singh 1981) Yet, the "psychology" behind the development of customer loyalty is not well understood
Several consumer theorists have recognized this defi- ciency and called for studies to investigate the cognitive mechanisms that form the construct (Dick and Basu 1994; Jacoby and Chesnut 1978) Until recently, one of the more frequently offered explanations of the psychology at work
in a customer's loyalty was the confirmation/disconfirma- tion mechanism of satisfaction (e.g., Bitner 1990) How- ever, the sufficiency of this explanation is debatable given current discussion that contends that a satisfied customer
is not enough (Schulz 1998) and research that suggests that loyalty's antecedent relationships may be more com- plicated than we had thought (Oliva, Oliver, and MacMil- lan 1992) Indeed, Jacoby and Kyner (1973) argue that the thinking behind loyalty is much more complex, with sev- eral conditions or cognitions at work in the construct Another antecedent thought to provide a more accurate description of the sort of thinking that leads to customer loyalty is commitment (Day 1969) Some researchers hold that this construct could "provide the essential basis for distinguishing [and] assessing the relative degrees of brand loyalty" (Jacoby and Kyner 1973:3) Still, this view
is not universally held, as others suggest that the two
Trang 2334 JOURNAL OF THE ACADEMY OF MARKETING SCIENCE SUMMER 1999
constructs are either not related (Oliva et al 1992) or that
they are synonymous and represent each other (Assael
1987) An intermediate view on the matter asserts the con-
structs are related, yet by definition are distinct, with com-
mitment leading to loyalty (Beatty, Kahle, and Homer
1988)
On the surface, some definitions of loyalty and commit-
ment suggest that similar attitudinal biases are at work
Recent work by Dick and Basu (1994) offered a delinea-
tion between loyalty and close psychological relatives like
commitment Their model of effects inferred that commit-
ment serves as a precursor to loyal attitude and its
"appraisal [function]" of continued patronage Restated,
the formative relationship, in essence, casts commitment
as the "emotional or psychological attachment to a brand"
that develops before a customer would be able to deter-
mine that their repeat purchase behavior was derived from
a sense of loyalty (Beatty and Kahle 1988:4)
Many in the marketing field have defined loyalty as a
composite blend of brand attitude and behavior, with
indexes that measure the degree to which one favors and
buys a brand repeatedly (e.g., Day 1969; Pritchard and
Howard 1997) These loyalty indexes typically describe
what proportion of a patron's behavior was based on or
attributed to loyal attitude (i.e., P[behavior]/attitude)
Commitment differs from this composite definition as it is
usually considered in purely cognitive terms that measure
consumer attitudes of attachment to a brand Morgan and
Hunt (1994) endorse this distinction and describe commit-
ment as an enduring desire to continue an attachment [rela-
tionship] Kelley, Donnelly, and Skinner (1990:322) con-
tend that the attitudinal domain of this attachment is best
understood in symbolic terms (i.e., customer identifica-
tion), as committed patrons tend to identify strongly with
the goals and values of an organization
Reviews have tracked loyalty's development through
behavioral and attitudinal phases of measurement to the
current composite perspective (e.g., Jacoby and Chesnut
1978) Over time, the field's attention to loyalty's assess-
ment has been a dominant issue So much so, that few stud-
ies have investigated the nature of commitment's link and
role in explaining the construct Instead, the dated lament
that "no explanation of the phenomenon [loyalty], no indi-
cation of why?" continues to loom over the field (Jacoby
1971:26) The absence of work on the relationship is due,
perhaps, to the fact that some still view the two constructs
as one and the same Lack of attention may also be due to
the paucity of consumer research on commitment's defini-
tion and measurement, a concern voiced by several authors
in their call for that construct's development (e.g., Kelley
and Davis 1994; Morgan and Hunt 1994) Our article
attempts to address this void
Specifically, our investigation of the commitment-
loyalty relationship first asks, "What is the underlying
psychology [cognitive mechanism] that defines commit- ment?" A comprehensive review of the literature helps form a theoretical basis for clarifying what commitment means We then examine the issue of how commitment influences loyalty
CONCEPTUALIZATIONS OF COMMITMENT
Definitional work on the construct of commitment began in the sociology and psychology disciplines Early sociological perspectives reflected an interest in the socie- tal and social factors that constrained or committed indi- viduals to a consistent line of action (Becker 1960; Kanter 1968), whereas psychologists defined commitment in terms of decisions or cognitions that fix or bind an individ- ual to a behavioral disposition (Festinger 1957; Kiesler 1971) Conceptualizations of commitment as a relation- ship, in the context of marriage or work, have interpreted the construct within a social-psychological framework For example, research undertaken by organizational behavior theorists explained an employee's commitment
to a job as "the relative strength of an individual's identifi- cation with and involvement in a particular organization" (Mowday, Porter, and Steers 1982:27) Here, the construct was conceptually characterized by intent to remain, along with certain personal and environmental factors that underpin that intent In this sense, commitment was inferred not only from the employee's beliefs and opinions (a series of binding cognitions) but also by their level of intent to act in a particular way
Conceptual renderings in other disciplines characterize commitment as a multidimensional phenomenon com- posed of several cognitive features (e.g., Kiesler 1971) However, consumer research has seldom considered the complex nature of the construct For example, Kelley and Davis (1994) examined customer (service) commitment
as a general trait, adapting Mowday, Steers, and Porter's (1979) measure of organizational commitment Similarly, Morgan and Hunt's (1994) study on relationship market- ing used the same scale Viewing commitment as a single, general trait is somewhat problematic, given the multiple conceptual ingredients noted in the construct's definition and the fact that later work found some evidence of a more complex factor structure (Mowday et al 1982).1
Other consumer studies have used unidimensional measures to assess commitment (e.g., Beatty et al 1988), yet the epistemological depth and methodological sophis- tication of these instruments is questionable The primary criticism of such measures is the contention that any the- ory of commitment should move beyond a general expres- sion of attachment and incorporate an understanding of the psychology inherent in binding a person to that disposi- tion 2 Other constructs have benefited from this sort of specification Satisfaction, for example, is recognized as a
Trang 3level of emotional affect yet is more specifically described
by its formative process, as a function of confirmation/
disconfirmation (Oliver 1980)
Work by Crosby and Taylor (1983) provided a defini-
tion of commitment that articulated a more involved view
of the construct Based on experimental research suggest-
ing that commitment resists influence and change (e.g.,
Kiesler and Sakumura 1966), the authors described cus-
tomer commitment as a stable preference that was bound
by an attitude of resistance to change Crosby and Taylor
maintained that the "tendency to resist changing prefer-
ence" provided the principle evidence of commitment and
that this attitude was best explained by two antecedent
processes (p 414)
The first formative process dealt with cognitive struc-
ture and how people managed information about their
preference (informational processes) Crosby and Taylor
(1983) argued that the need to maintain a consistent infor-
mational structure (e.g., beliefs, reasons for pur-
chase/repurchase) helped maximize one's resistance to
change The second process dealt with personal attach-
ment and whether people identified with important values
and self-images linked to a preference (identification
processes) The more strongly consumers identified with
the values and images embodied by a particular brand, the
greater their sense of resistance to change that preference
would become
Earlier work by Salancik (1977) provides an interesting
contrast, as he felt people became committed when three
perceptual states revocability, publicness, and volition
were engaged Commitment was strengthened when peo-
ple sensed their decision was (1) not easily reversed, (2)
known to significant others, and (3) undertaken as an exer-
cise of free choice The first two states, revocability and
publicness, appear to be captured by the informational and
identification processes implied in Crosby and Taylor's
(1983) definition Revocability is determined by the psy-
chological cost entailed in rethinking and altering the
informational structure that supports one's commitment,
while publicness involves a willingness to be explicitly
identified with the images and values of a (brand) prefer-
ence by significant others (e.g., friends, family) Volition,
the third state mentioned by Salancik, refers to people's
perception that their preferences are "free" (i.e., not dic-
tated by any constraints) When people sense that their
choices are unhindered, the resulting commitment is likely
to be stronger and more deeply held
While Salancik (1977) used the three processes men-
tioned above to define the construct, our view is consistent
with Crosby and Taylor's (1983) definition This perspec-
tive specifies information, identification, and volition as
antecedent processes of commitment that facilitate its root
tendency, resistance to change Further review of these
psychological processes and their effect on resistance to
change follows
THE ANTECEDENT PROCESSES
OF COMMITMENT Informational Processes
Informational complexity One factor that contributes
to the attitudinal stability of commitment uses the process- ing of information to form complex cognitive structures Deviation from an attitude (sense of resistance) that is sup- ported by a complex cognitive structure is said to involve a high psychological cost to an individual Salancik (1977) discussed commitment's revocability in terms of the psy- chological cost involved in the cognitive reordering and rethinking of what was known (about the product) When complex informational schema gird a person's commit- ment, changing your mind becomes more difficult, as ac- commodating disparate cognitions requires even greater change (Millar and Tessar 1986) Much of a person's resis- tance to change may be driven by a desire to avoid the cost
of dissonance and the disruption it brings to one's organi- zation of salient cognitions (Festinger 1957) For the highly committed, these costs are more pronounced than those incurred when change is contemplated in the simple structure of the less committed (Robertson 1976)
Cognitive consistency In addition to complexity, con- sistency is also part of commitment's informational pro- cesses While complex informational structures form a detailed array of cognitions that support commitment, consistency works by defending those structures and one's commitment when facing conflicting information In this sense, the informational processes of commitment not only serve as a cognitive blueprint to process and accumu- late consistent information but also as a defense mecha- nism that reinterprets, suppresses, or loses information that is inconsistent (Tessar and Leone 1977) When this sort of "information processing parsimony" is at work, re- sistance to change is maximized, as people may seldom be challenged by conflicting information (Holbrook 1978) Based on Rosenberg's (1960) work, Crosby and Taylor (1983) described this operant condition as the need to maintain consistency in what one thinks and feels about a preferred brand Dick and Basu (1994) argued that when these "antecedents are consistently favorable for a b r a n d the degree of differentiation in relative attitude [such as re- sistance] increases" (p 105) Just as consistency within each antecedent has an effect, congruency between all of them creates a further "psycho-logical" reason for com- mitment (Crosby and Taylor 1983:414)
Confidence Another element at work in commitment's informational processes is confidence Berger and Mitchell (1989) noted that consistent information from re- peated exposure can provide greater brand-relevant cogni- tive elaboration (i.e., informational complexity), and enhance a consumer's confidence in a resulting attitude
Trang 4336 JOURNAL OF THE ACADEMY OF MARKETING SCIENCE SUMMER 1999
Attitudinal confidence has been described as an evaluative
mechanism where consumers assess whether brand beliefs
are accurate and their attitude warranted (Dick and Basu
1994) Day (1970) argued that when involved consumers
are confident about their judgements, it stabilizes their
brand attitude Burton and Netemeyer's (1992) causal
model supported this notion, as voter confidence main-
tained election preference over time Whereas a lack of
confidence, when reflected as uncertainty or ambiguity
about the information to hand, increased the potential for
attitude change
Identification Processes
Position involvement Some distinction should be made
between product involvement (Zaichkowsky 1985) and
the identification processes that operate in position in-
volvement Freedman (1964) distinguished the two forms
of involvement by noting that product involvement results
when important values are made salient by a certain deci-
sion situation (e.g., the need to buy a hotel room or pur-
chase an air ticket), whereas position involvement is
evident when those values or self-images are identified
with a particular stand or brand choice Crosby and Taylor
(1983) suggest that this link between a preference and
one's personal values and self-images strengthens resis-
tance to change
The process of identification can also be considered in
terms of consistency People can evaluate their position
involvement to determine whether their public association
with the brand in question is consistent with certain values
and self-images In this context, the values and self-images
perceived in any public association with a brand (i.e.,
social self) would be personally evaluated to see if they are
truly consistent with the consumer's internal views (i.e.,
personal self) Salancik (1977) takes a slightly different
stance Rather than internal views determining consis-
tency via an assessment of shared values (with a brand/
company), Salancik argued that commitment was maxi-
mized when that sense of consistency was driven by the
public persona (publicness) Here, the more public or well
known our association is, the greater one's desire is to
remain consistent and resist changing that relationship
Public connections or associations with certain brand
images are often used for the purpose of self-presentation,
with some being more prone than others to this sort of
desire (Bearden, Netemeyer, and Teel 1989)
The desire to appear consistent publicly is believed to
exert considerable influence over a wide range of human
actions Assael (1987) contends that purchasing or repur-
chasing in consumer behavior contexts is frequently based
on symbolic rather than utilitarian value In this sense,
symbolic purchasing indicates that the product is pur-
chased not for what it is but for what it means as a
subjective symbol (Solomon 1983) Some researchers consider that the highest form of commitment is driven by
a need for social (symbolic) representation and self- identity (Buchanan 1985) Theories about the impact of desired values and self-images on commitment have been tested and collectively suggest that the degree to which consumers publicly and personally identify with a brand directly effects the resistance to change that preference (e.g., Beatty et al 1988)
Volitional Processes
Volitional choice Several theorists hold that consumer
perceptions of volition also play an important role in any theory of commitment (e.g., Bagozzi 1993) Salancik (1977) argued that when perceived volition is high, a per- son should feel more personally responsible for their deci- sions than when perceived volition is low Kiesler (1971) also found that perceptions of free choice and self- responsibility operate at the very core of one's commit- ment Indeed, free choice is seen by some as a required condition before attitudes of internal commitment can de- velop from behavior (Shamir 1988) This is because the freedom to choose greatly influences the internal organi- zation of an action's meaning and hence the degree of commitment
Volitional choice can best be described as a process that involves both a freedom from constraints and a freedom to
choose Freedom from refers to the notion that brand
choice is elicited freely and not constrained by external considerations that might limit one's sense of personal ownership in that decision As Shamir (1988) noted, we are less likely to feel personally committed (to our choice
of product or service) if that decision had to first meet the
requirements of certain external actors The freedom to
component of volitional choice refers to the potential for choice to reflect meaningful action or effort (Bagozzi 1993) Research suggests that when people sense they are acting freely in choosing an object, they attribute attitude (meaning) toward that object (Bem 1967) For instance, when consumers choose Rainforest Crunch or another ice cream from a wide range of 31 flavors, they usually infer greater meaning in that selection (e.g., "It's my favorite or the best alternative")
Cialdini, Cacioppo, Bassett, and Miller's (1978) inves- tigation of the "low-ball phenomenon" noted that percep- tions of free choice can form the basis for a customer's cognitive commitment to an outcome Their research argued that once volitional processes were engaged, con- sumers developed a sense of resistance to change in which they were prepared to continue with an initial decision even though certain sales incentives were withdrawn (e.g.,
"Those options are extra" or "The boss won't agree to the sale at this price") An experiment undertaken by
Trang 5Freedman and Steinbruner (1964) also strongly supported
the fact that high choice in the initial decision greatly
increased one's resistance to change These findings and
the previous discussion purport that volitional processes
(i.e., perceptions of freedom) can influence the sense of
self-responsibility and meaning choice reflects and
enhance the tendency to resist changing preference
In summary, the literature reviewed here suggests that
psychological commitment is best defined by a tendency
to resist change and that three formative processes activate
this tendency Contrary to past consumer research that
viewed commitment simplistically as a general attitude of
attachment (e.g., Beatty and Kahle 1988), our study pro-
poses that a more complex network is needed to assess the
construct
Resistance to Change:
The Key Mediating Variable?
Previous studies of purchase behavior (Beatty and
Kahle 1988), consumer expectations (Kelley and Davis
1994), and advertising effectiveness (Robertson 1976) all
attest to commitment's ability to affect a variety of out-
comes Our conceptualization of commitment holds that
informational, identification, and volitional processes are
active precursors of the construct in that they maximize the
tendency to resist changing preference As the principle
evidence of commitment, resistance to change is central to
a variety of outcomes, the foremost of which is loyalty
(Jacoby and Kyner 1973)
Path analytic work has shown that in some relation-
ships, commitment is best portrayed as a key mediating
variable (Kelley and Davis 1994; Morgan and Hunt 1994)
In a similar vein, we postulate that resistance to change, as
the primary evidence of commitment, will act as a media-
tor between the construct's antecedent processes and loy-
alty According to Baron and Kenny (1986), a variable is
"said to function as a mediator to the extent that it accounts
for the relation between the predictors [i.e., antecedents]
and the criterion [i.e., outcomes]" (p 1176) With this in
mind, our proposed mediating-effects-model (M-E-M)
describes the relationship between resistance to change
and loyalty as substantive and direct, while the informa-
tional, identification, and volitional processes, although
related to loyalty, will have a significant yet indirect effect
(via resistance to change) on that outcome In other words,
commitment's antecedent processes will first foster a
sense of resistance to change, which in turn will mediate
the effect of these processes on loyalty (see Figure 1)
Rival direct effects models A current practice in struc-
tural equation modeling calls for researchers to move be-
yond simply testing proposed models by comparing their
performance with rival structures (Bagozzi and Yi 1988)
FIGURE 1 Competing Models of Commitment and Its Link With Loyalty
Antecedent Processes Pnnc:pal Ewdence Latent Outcome(s)
of Commitment of Commttment of Commitment
~ r e c t Effects Model ] (D-E-M |)
Direct Effects Model II (D-E-M If) Jl~-~ Processes J ~
Morgan and Hunt (1994) examined commitment's role as
a mediator in this manner, comparing proposed and rival models This work looked at whether outcomes (such as loyalty) were best explained by a key mediating variable model of direct and indirect effects or by a competing model of direct effects Their findings supported media- tion by noting it offered the most parsimonious explana- tion of commitment's relationships Although our proposed model is a M-E-M that is consistent with Crosby and Taylor's (1983) definition, comparing its performance with direct-effects-models (D-E-Ms) will provide a way to (1) test this specification and (2) clarify whether mediation
is in fact the most accurate way to describe the construct's link with loyalty
In essence, the D-E-Ms shown in Figure 1 can be used
to represent a competing theory of commitment, which defines the construct as the sum of its informational, iden- tification, and volitional processes (Salancik 1977) Resis- tance to change and loyalty are alike in these models in that they are both outcomes of commitment Given commit- ment's close link to loyalty, a simple test of the two com- peting definitions could be undertaken by noting which model (M-E-M or D-E-M I) offers the most effective explanation of the criterion, loyalty Another contrast with D-E-M I can determine whether the antecedent processes
Trang 6of commitment are discrete in maximizing the principle
evidence of the construct or whether they should also be
considered as formative agents of loyalty This tests the
M-E-M network's discriminant validity by determining
whether the processes distinguish between resistance to
change and loyalty and explain more of the construct's
root tendency (Burnkrant and Page 1982)
D-E-M theses that direct links between commit-
ment's processes and loyalty exist -do have some sup-
port Empirical studies have linked informational (i.e., that
loyal users agglomerate more information about important
brand decisions), identification (i.e., that the loyal are
more interested than the nonloyal in status/value associa-
tions), and choice processes (i.e., that loyalty is a function
of choice from a set of alternatives) as important correlates
of brand loyalty (Carmen 1970; Jacoby, Chesnut, and
Fisher 1978; Jacoby and Kyner 1973) Although this
research does not directly support causality, recent con-
ceptual work on loyalty's antecedents does propose sev-
eral links that rival our view of commitment's processes
Dick and Basu (1994) argued that loyalty is directly
caused by attitude accessibility (i.e., the ease of retrieving
an attitude from the informational structure), confidence,
cognitive consistency, and centrality (i.e., the degree to
which brand attitude is related to an individual's values)
Empirically testing some of these proposed links (in the
D-E-Ms) should clarify if such relationships actually exist
or whether resistance to change intervenes and mediates
these effects on loyalty
Specific analyses can also be used to test whether resis-
tance to change functions as a mediator These tests will
address certain questions or "conditions" for mediation
(Baron and Kenny 1986:1176) The first two questions are
examined by the M-E-M They ask, (a) Do commitment's
three antecedent processes have a significant effect on
resistance to change? and (b) Does resistance to change
have a significant effect on loyalty? A D-E-M (i.e., D-E-M
I in Figure 1) will then examine condition (c) and address
the question, Do the antecedent processes also have a sig-
nificant direct effect on loyalty? D-E-M II will then answer
a final question, (d) Do the previously significant effects of
the antecedent processes on loyalty (noted in c) become
nonsignificant when the path between resistance to change
and loyalty is opened? These structural comparisons
should corroborate the nature of commitment and the role
of resistance to change in explaining customer loyalty
METHODOLOGY
The primary intent of this study was to examine the
commitment-loyalty relationship However, prior to this
investigation, a valid and reliable measure of commitment
needed to be developed The following discussion outlines
the methods involved in achieving these objectives
Overview of Scale Development
The commitment items The development of a psycho-
logical commitment instrument followed scale construc- tion procedures recommended by Churchill (1979) It began with a literature review that generated an item pool designed to measure psychological commitment's hy- pothesized components Sixty-five items were initially generated to reflect psychological commitment's ten- dency to resist change and the five components involved in that phenomenon's antecedent processes (informational complexity, cognitive consistency, confidence, position involvement, and volitional choice) A panel of judges ex- amined the content validity of the item pool Three faculty members and three doctoral students were given the con- ceptual definitions of each component and asked to cate- gorize each item by its theme Further panel evaluation of the face validity for each item was undertaken by rating its appropriateness and clarity (wording) Consensus on cate- gorization, fit, and clarity ratings of 4 and above (on a scale
of 1 to 5) admitted items to the final pool The rigor of re- peatedly examining content validity during item genera- tion suggested by Churchill resulted in the retention of 51 items (cf Bearden et al 1989; Zaichkowsky 1985) A 7-
point numeric bipolar scale ranging from strongly dis- agree (1) to strongly agree (7) was attached to each state-
ment (e.g., "I would resist changing my preference to use brand X.") Wording for just less than half of the items was randomly reversed to avoid response bias
A self-administered questionnaire was given to an ini- tial sample of 391 airline and hotel patrons to determine the dimensionality and reliability of the scale items Com- mitment items that remained following an iterative sequence used in scale purification (i.e., internal reliability and exploratory factor analysis: Zaichkowsky 1985) were then examined with a confirmatory factor analysis (CFA) that included measures of the loyalty construct (cf Bear- den et al 1989; Ruekert and Churchill 1984) CFA again tested the validity of this initial measurement model by reexamining it with data from a second sample of 290 patrons Following this, the two samples were combined (N = 681) for a final psychometric assessment
The Loyalty Index Day (1969:30) first suggested using
a simultaneous consideration of loyal attitude and behav- ior to generate a composite index of the construct He ar- gued that the most effective way to view the behavioral aspect of loyalty was with a proportion of purchase measure that concentrated on a specific brand However, as Day noted, consistency in a person's purchase behavior did not necessarily mean that he or she was brand loyal That be- havior might be spuriously driven by price and not the steadfast allegiance (attitude) attributed to loyalty Thus, brand attitude had to be assessed as well as brand behavior While Day had used a single statement to assess the attitu-
Trang 7dinal component, subsequent studies developed multiitem
scales for this (Muncy 1983; Selin, Howard, Udd, and Ca-
ble 1988) By and large, these scales reflected work by
Cunningham (1967) and asked questions about brand pref-
erence (e.g., was it their favorite or the best brand?) and the
consumer's willingness to switch in given situations (e.g.,
if the brand was overpriced or not available or in stock)
Loyal attitude items were recoded so that 1 indicated
(1969) equation, the ratio proportion of purchase P[B] was
then divided by mean attitude scores [A] In this sense,
high proportion of purchase when divided by strong atti-
tude created index scores that were closer to 1 (true loy-
alty), whereas low proportion of purchase when divided by
weak attitude created scores that were closer to 0 (low loy-
alty) 3 Our study used two previously developed attitude
scales to create two composite (i.e., P[B]/A) indicators of
the construct (see the appendix) One of the advantages of
assessing loyalty in this manner is that it "allows investiga-
tion of the phenomenon from a causal perspective" where
questions about "how underlying processes influence loy-
alty" can be addressed (Dick and Basu 1994:102) Prior to
addressing the issue of causality, measurement validity
was examined with AMOS 3.61, a structural equations
program (Arbuckle 1994, 1997)
Study Samples
A convenience sample of 421 consumers was asked to
provide data for the initial stages of scale purification
Missing data reduced this sample to 391 The age of the
subjects, 46 percent of whom were female, ranged from 17
to 76 years, with amean age of 42 years Respondents were
intercepted at a city airport and asked to complete the ques-
tionnaire in one of four contexts (see Figure 2); the
consumer's favorite or most recently used brand of airline
service or their favorite or most recently used brand of hotel
service The use of favorite or most recent brand ensured
response variability (i.e., different levels of attachment)
A second convenience sample (n = 290) of customers
was surveyed for two specific brands Specific brands of
service were designated in the airline and the hotel indus-
try to further confirm the commitment and loyalty measures
developed with the first sample Convenience samples
were collected on site at each of two designated services
(United Airline patrons: N1 =148; Hilton Hotel guests:
N2 =142) Sixty percent of this sample was male Respon-
dents ranged from 17 to 80 years, with an average age of
46.6 years A primary screening question was used in both
samples to ensure that respondents had used the respective
brand of service more than once in the past year The two
convenience samples were then combined as one data set
(N = 681) to provide overall measurement and structural
assessments
FIGURE 2 Sample Frame and Content
Inmal Data Collectson (n=391 )
Favorite Brand of AJrhne Service (n t =99)
Most Recenl Brand of Airhne Service (n =101)
Second Data Colleetton (n=290)
I Airline Servtce (n5=148l
Airline Patrons (N =348)
Favorite Brand of Hotel Serwce (n2=95)
Most Recent Brand of Hotel Service (n4=96)
Hotel Service (n6=142)
Hotel Patrons
(N =333)
RESULTS
Developing and Evaluating the Measures
content validity process were subjected to scale purifica- tion procedures using responses from the initial sample (n = 391) Survey responses were randomly split into two halves so as to cross validate the decisions made during the item reduction phase Cross validation is recommended to minimize error probability and capitalization on chance Scale purification began with the computation of alpha co- efficients for resistance to change and the five components involved in that phenomenon's antecedent processes (in- formational complexity, cognitive consistency, confi- dence, position involvement, and volitional choice) The decision criteria for item deletion involved cross valida- tion between the split samples, with item elimination im- proving corresponding alpha values to the point at which all items retained had corrected item-to-total correlations greater than 0.4 (cf Zaichkowsky 1985) The 36 items common to both split samples were then subjected to ex- ploratory factor analysis (principal axis factoring with an oblique rotation) Cattel's scree test indicated that a four- factor solution was appropriate in both cases The general structure of all four factors in both split samples was con- sistent, although the reduction of the hypothesized compo- nents to that of a four-factor solution did initially lead to some mixed item themes Further screening of items to en-
Trang 8340 JOURNAL OF THE ACADEMY OF MARKETING SCIENCE SUMMER 1999
TABLE 1 Developing the Psychological Commitment Instrument:
Principal Axis Factoring Pattern Matrix for 16 Items (n = 391)
NOTE: Item/factor loadings are italicized
sure consistency and significance of factor loadings across
samples resulted in a final four-factor structure with a
common core of 24 items
The split samples were then combined to analyze this
common core (n = 391) Again, internal homogeneity was
examined, and eight items were eliminated (item-to-total
correlations < 0.4) The remaining 16 items were then sub-
jected to a second principal axis factoring, with the solu-
tion restricted to four factors This structure was again sup-
ported, with eigenvalues greater than 1.0 for all four
factors (see Table 1) Three additional items were elimi-
nated Two had low communality estimates (< 0.4) in the
volitional choice and information complexity factors
(Tinsley and Tinsley 1987), and the third had a significant
cross loading in the position involvement factor (see Table 1)
The remaining 13 items provided a "simple structure"
(Bearden et al 1989:475), with four thematically consis-
tent factors providing an assessment of commitment's
three antecedent processes and its central tendency, resis-
tance to change (see the appendix)
These findings represent a parsimonious explanation of
the data in which cognitive consistency and confidence did
not converge as discrete components While the scale
methodology employed here sought to achieve parsimony
by determining the principal components of the construct,
the external validity of this four-factor solution requires
further substantiation with confirmatory analyses For
ease of the discussion, this measure of commitment is
referred to hereafter as the psychological commitment
instrument (PCI)
PCI was subsequently examined using alpha coefficients and CFA A CFA was conducted with the PCI's four fac- tors and a two-item measure of brand loyalty adapted from previous studies (Muncy 1983; Selin et al 1988) The first analysis of the 15-item, five-factor measurement model was undertaken with the initial sample (n = 3 91) This pro- duced a chi-square statistic of 124.1 (df= 80, p < 01), with goodness-of-fit (GFI) and adjusted goodness-of-fit (AGFI) indices of 96 and 94, respectively Each indica- tor loaded significantly on its designated factor (p < 01) Although the overall chi-square statistic is significant, the model's fit represents a substantial improvement over the chi-square values obtained from one-factor (X 2 = 1495.0; df= 90,p < 01) and null models (Z 2 = 2696.4; df=
105, p < 01)
The sensitivity of the X2 test does pose some problems for large samples as "even minute differences tend to be detectable" (Hayduk 1987:167) Several researchers have suggested that sample sizes should range from 100 to 200
to avoid these problems of misspecification Hoelter (1983) addressed the issue of sample size sensitivity by providing a formula for what he termed the "critical-N" This CN is "the size that a sample must reach in order to accept the fit of a given model on a statistical basis" (p 330) Using p > 05 as the criterion for acceptable fit, a
CN of 326 was calculated for the current measurement model This exceeded Hoelter's rule of thumb, which argued that models with a CN greater than 200 were admissible, as problems with Z 2 significance tended to
Trang 9reflect a trivial misspecification that was more a feature of
sample size than design
Given this initial support, the five-factor correlated
model was again tested with the second population of hotel
and airline patrons (n = 290) Here, the CFA generated a
chi-square statistic of 205.3 (df= 80, p < 01), with a GFI
and AGFI of 92 and 87, respectively Although not as
strong as in the initial sample, these findings do help to
confirm the solution's fit For a final assessment of the
measurement model's performance, we combined both
samples and reexamined the data (N = 681) The results
from this CFA are shown in Figure 3 The GFI and AGFI
both corroborate the solution's adequacy (i.e., > 9), and
Bentler's (1990) comparative fit index (CFI) and Hoelter's
(1983) CN size provide further evidence of an appropriate
fit between the model and this larger data set
In Figure 3, the loadings on the right of each of the 15
items are standardized regression weights On the left of
each item are squared multiple correlations that reflect
how much of each item's variance was explained by its
respective factor Construct reliability estimates for the
four PCI components and the loyalty measure were tabu-
lated from these results (Bagozzi and Yi 1988) All of the
scales demonstrated adequate internal consistency (con-
struct estimates: Volitional Choice, 80; Position Involve-
ment, 84; Information Complexity, 83; Resistance to
Change, 81; and Loyalty, 91) The more conservative
variance extracted estimates were also calculated and pro-
vided further support for the scales' soundness (Volitional
Choice, 57; Position Involvement, 63; Information Com-
plexity, 62; Resistance to Change, 53; and Loyalty, 84)
Correlations between the five factors are also shown in
Figure 3 The PCI's four components are all positively cor-
related, which provides some evidence of convergent
validity Although the strength of these associations vary,
the measure's discriminant validity is supported by the fact
that the three process-based components (volitional
choice, position involvement, and informational complex-
ity) have stronger relationships with commitment's prin-
ciple tendency (resistance to change) than with a closely
related outcome (loyalty) (cf Ruekert and Churchill
1984) Closer examination of the high correlation between
position involvement and resistance to change does, how-
ever, raise another question about the PCI's discriminant
validity Fornell and Larker (1981) recommended that the
square of the correlations between constructs should not
exceed their explained variance estimates While this con-
dition was satisfied in all the other factors, it suggests that
resistance and position involvement may not be distinct
This particular dilemma was raised by Crosby and Taylor
(1983) who noticed that "separating the influence of
involvement from other commitment factors is problem-
atic because of a tendency for these phenomena to coexist
in the consumer's psyche" (p 415)
FIGURE 3 Confirmatory Factor Analysis (N = 681)
48 69
55 74
70 84
\
,,/ / /
NOTE: Chi-square = 232.17 (df 80); p < 01; goodness-of-fit index (GFI)/adjusted goodness-of-fit index (AGFI) 96/.94; comparative fit index (CFI)/parsimonious normed fit index (PNFI) = 97/.73; CN = 299.00
Burnkrant and Page (1982) recommended a series of tests to help clarify whether such factors should stand alone or be combined In our case, this procedure com- pared the existing model (Z 2 = 232.1, df= 80, p < 01), in which loyalty and the four separate PCI factors were allowed to correlate, to one in which resistance to change and position involvement were hypothesized to have a unity correlation that depicted them as unidimensional (Z 2 = 240.1, df = 81, p < 01) A chi-square difference test
between the two structures supported Figure 3 and estab- lished that the two should be considered as discrete factors (Ax 2 = 8.0, dr= 1,p < 01), a finding consistent with theory and our earlier factor analyses (see Table 1) The measure- ment work done to this point has used multiple samples to confirm the internal consistency and integrity of the factor structure proposed by our initial scale development proce- dures These findings hold that commitment is a multidi- mensional construct that is related to, yet distinct from, customer loyalty
Additional evidence of validity Before testing the
structural relationship between commitment and loyalty, the measurement model's validity was reexamined in dif- ferent contexts and situations AMOS (3.6) was again used
to review the validity of the model in two different con-
Trang 10342 JOURNAL OF THE ACADEMY OF MARKETING SCIENCE SUMMER 1999
texts A simultaneous CFA that fitted the model to data
from both service populations was conducted (Arbuckle
1997:441) While the same model structure was specified
for the airline (n = 348) and hotel patrons (n = 333), the re-
gression weights and the unique and common variances
were allowed to differ in each group Similar to Figure 3,
these results substantiated the validity of the measurement
model in both cases (Z 2 = 295.9, dr= 160, p < 01, GFI =
.95, AGFI =.92, CFI = 97, CN = 439) Subsequent alpha
reliability estimates in the aidine and hotel samples noted
that all of the model's factors were internally consistent
(i.e., > 80)
The integrity of the model was again reviewed to see if
its specifications held true in different situations Essen-
tially, the favorite and most recent brand groups that gener-
ated the initial sample (see Figure 2) provided a variety of
consumer-brand responses to test this feature Generally,
most recent brand responses are thought to be character-
ized by lower levels of attachment due to the presence of
habitual or happenstance purchasers (Beatty and Kahle
1988), whereas favorite brand responses tend to reflect
higher levels of consumer attachment (Jacoby and
Chesnut 1978) To examine the model's stability in both
situations, we again used a simultaneous CFA with the two
populations Regardless of whether low or high levels of
attachment were being assessed, the parameters speci-
fied by the model demonstrated an acceptable fit (X2 =
237.7, df= 160, p < 01, GFI = 93, AGFI =.89, CFI = 97,
C N = 313)
Corroborating the model's fit in this manner encour-
aged a specific test of the PCI's nomological validity to
determine whether the scale was not simply consistent in
but also sensitive to different situations (Ruekert and
Churchill 1984) Zaichkowsky (1985) tested her mea-
sure's validity by proving its ability to detect low and high
involvement decision situations In a similar vein, a valid
measure of commitment should also be able to differenti-
ate between favorite and most recent brand responses
CFA was again employed to test whether the favorite and
most recent groups had different factor means (cf
Arbuckle 1997:467) Analyses were run simultaneously
for both groups The first analysis tested the significance
of different factor means This imposed group-invariant
factor loadings and intercept patterns on both groups and
fixed the most recent brand group's factor means to zero
(~2 = 269.8, dr= 180,p < 01) A second CFA tested the null
hypothesis that the factor means were the same in both
groups Group-invariant factor loadings and intercept pat-
terns were again imposed; however, this time the favorite
brand group's means were also set to zero (X~ = 287.2, df=
185, p < 01) The chi-square difference between the first
and second analysis (Ax 2 = 17.4, df= 5, p < 01) rejects the
notion that the two groups had similar means and supports
the contention that the measures were sensitive to different
decision situations (Note that mean contrasts between the two groups consistently characterized the favorite brand group with stronger attitudes of commitment.)
The Structure of the Commitment-Loyalty Relationship
Testing the M-E-M The procedures used thus far have
found the PCI to be a valid and reliable measure of com- mitment that is related to, yet distinct from, a composite in- dice of loyalty (Day 1969) This factor analytic work found three components (information complexity, posi- tion involvement, and volitional choice) to represent the effect of commitment's antecedent processes on the con- struct's principle tendency, resistance to change The M- E-M holds that a single causal path from resistance to change provides the best explanation of why loyalty devel- ops (see Figure 1) This model restricts the effects of com- mitment's three antecedent processes so that they only have indirect relationships with loyalty via resistance, the key mediating variable
Three path analyses were conducted to examine the fit
of the M-E-M: one with the total sample (N= 681) and the other two with the airline (n = 348) and hotel (n = 333) subsampIes Analysis of the combined data set noted a sig- nificant chi-square statistic (df= 83, p < 01), with a GFI and AGFI of.96 and 94 The strength of the latter fit indi- ces (> 9) and the CN of 296 supported accepting the model In fact, indicators from all three analyses suggested
a good fit between this model and the data (see results in Table 2)
With the exception of the informational complexity resistance to change link in the hotel sample, 11 out of the
12 direct pathways specified in the three sample analyses had significant structural coefficients (p < 01) In each case, paths from the antecedent processes were able to explain more than 60 percent of the variance in resistance
to change (simultaneous multiple correlations [SMCs] = .63, 67, 60) Identification, represented by position involvement, consistently made the strongest contribution
to a consumer's sense of resistance (t3s = 66, 70, 64) The resistance to change -> loyalty link was also significant in each sample (SMCs > 30) This link was particularly strong in the airline sample, with more than half of the variance in customer loyalty accounted for by this path (SMC = 52) The first two conditions for mediation were supported by these results, namely, that the antecedent processes had an effect on resistance to change, and resis- tance to change had an effect on loyalty
direct links between commitment's processes and loyalty were also tested (see Figure 1) Like the M-E-M, analyses
of D-E-M I with the airline, hotel, and combined samples noted significant chi-square statistics (df = 81, p < 01)