Yuping Liu

The Long-Term Impact of Loyalty

Programs on Consumer Purchase
Behavior and Loyalty
Despite the prevalent use of loyalty programs, there is limited evidence on the long-term effects of such programs,
and their effectiveness is not well established. The current research examines the long-term impact of a loyalty
program on consumers’ usage levels and their exclusive loyalty to the firm. Using longitudinal data from a
convenience store franchise, the study shows that consumers who were heavy buyers at the beginning of a loyalty
program were most likely to claim their qualified rewards, but the program did not prompt them to change their
purchase behavior. In contrast, consumers whose initial patronage levels were low or moderate gradually
purchased more and became more loyal to the firm. For light buyers, the loyalty program broadened their
relationship with the firm into other business areas. The findings suggest a need to consider consumer
idiosyncrasies when studying loyalty programs and illustrate consumers’ cocreation of value in the marketing
process.

s an important component of firms’ customer rela- costly investments and require a firm’s long-term commit-

A tionship management (CRM) strategy, loyalty pro-

grams aim to increase customer loyalty by reward-
ing customers for doing business with the firm. Through
ment. It is vital for managers to know whether and how
these programs work before they take the plunge.
This research contributes to a better understanding of
these programs, firms can potentially gain more repeat busi- loyalty programs in three ways. First, although evidence
ness and, at the same time, obtain rich consumer data that about the effectiveness of loyalty programs has begun to
aid future CRM efforts. Since American Airlines launched accumulate recently, the field is still underdeveloped, and a
the first contemporary loyalty program in 1981, loyalty pro- clear picture has yet to emerge. Addressing this issue,
grams have blossomed and now span various industries, Bolton, Kannan, and Bramlett (2000, p. 28) suggest that “to
including retail, travel, and financial industries. It is esti- determine the long-term efficacy of a loyalty rewards pro-
mated that more than half of U.S. adults are enrolled in at gram, a company must quantify the program’s influence on
least one loyalty program (Kivetz and Simonson 2003). future purchase behavior (e.g., usage levels).” The current
Despite the prevalent use of loyalty programs, their research responds to the suggestion by quantifying on a
effectiveness is not well understood (Bolton, Kannan, and large scale the effectiveness of a loyalty program in the con-
Bramlett 2000). Some researchers question the value of loy- venience store industry. The key research question is
alty programs. For example, Dowling (2002) suggests that whether loyalty programs change consumers’ patronage
loyalty programs do not necessarily foster loyalty and are levels and exclusive loyalty to the firm. These outcomes are
not cost effective and that the proliferation of loyalty pro- important to study because they are directly related to con-
grams is a hype or a “me-too” scheme. Conversely, some sumer profitability and the financial success of a loyalty
recent studies show that loyalty programs have a positive program.
impact on consumers’ repatronage decisions and their share Second, this research examines consumers’ longitudinal
of wallet (e.g., Lewis 2004; Verhoef 2003). With limited behavior change after they join a loyalty program. Given
empirical validations, the debate on whether loyalty pro- the long-term orientation of loyalty programs and their
grams are truly effective continues. The divergent views transformation of single purchases into multiperiod deci-
suggest a need to understand these programs better. This is sions (Kopalle and Neslin 2003), it is natural that their
also of strategic importance because such programs are effectiveness should be examined longitudinally. Method-
ologically, studying loyalty programs over time alleviates
self-selection bias. Because loyalty program members may
Yuping Liu is Assistant Professor of Marketing, College of Business and already be frequent customers who are more likely to find
Public Administration, Old Dominion University (e-mail: YXXLiu@odu. the program attractive, simply comparing the behavior of
edu). The author thanks Ruth Bolton, William Boulding, Richard Staelin, loyalty program members with that of nonmembers cannot
and the four anonymous JM reviewers for their helpful suggestions on establish a conclusive causal relationship (Leenheer et al.
prior drafts of this article, as well as Old Dominion University for funding 2003). Thus, examining dynamic behavior change is more
the research. powerful than cross-sectional studies of behavior at a cer-
To read and contribute to reader and author dialogue on JM, visit tain point in time (Verhoef 2003). However, relatively few
http://www.marketingpower.com/jmblog. published empirical studies have examined longitudinal
loyalty program effects, especially from the perspective of

© 2007, American Marketing Association Journal of Marketing

ISSN: 0022-2429 (print), 1547-7185 (electronic) 19 Vol. 71 (October 2007), 19–35
continuous loyalty programs. This leaves a gap in the Loyalty programs provide value to consumers in two
understanding of the true effects of such programs. A recent stages. In the first stage, program points are issued to con-
study by Lewis (2004) has advanced this area by examining sumers at the time of purchase. Although these points have
dynamic postreward effects on consumer behavior in the no practical value until they are redeemed, recent studies
context of a continuous loyalty program. The current show that they have important psychological meaning to
research extends this work by studying the general effects consumers (Hsee et al. 2003; Van Osselaer, Alba, and Man-
of such a loyalty program on long-term purchase behavior chanda 2004). The psychological benefit increases the
change. transaction utility of a purchase (Thaler 1985) and, subse-
Third, this research studies how the idiosyncrasies of quently, the overall value perception of doing business with
individual consumers influence behavior changes that occur the firm. Because consumers can later redeem points for
after they join a loyalty program. Previous research has free rewards, point accumulation creates an anticipation of
shown that the idiosyncratic fit of an individual with a loy- positive future events, which increases consumers’ likeli-
alty program can influence his or her likelihood of joining hood of staying in the relationship (Lemon, White, and
the program (Kivetz and Simonson 2002). However, it is Winer 2002).
not clear whether such effects would carry over to how In the redemption stage, consumers receive both psy-
these consumers change their purchase patterns after pro- chological and economic benefits from a loyalty program.
gram enrollment. Adding to the limited set of individual The free reward functions as a positive reinforcement of
idiosyncrasies that have been examined in the literature, the consumers’ purchase behavior and conditions them to con-
current research explores how consumers with different ini- tinue doing business with the firm (Sheth and Parvatiyar
tial usage levels change their behavior to maximize the 1995). Psychologically, giving free rewards to customers
benefits they receive from a loyalty program. shows the firm’s appreciation and personal recognition of
its customers. This sense of being important can enhance
consumers’ overall sense of well-being and deepen their
Are Loyalty Programs Effective? relationship with the firm (Bitner 1995; Gwinner, Gremler,
Definition and Bitner 1998). Some researchers suggest that there are
Loyalty program. In this article, a “loyalty program” is other psychological benefits as well, such as the opportu-
defined as a program that allows consumers to accumulate nity to indulge in guilt-free luxuries (Kivetz and Simonson
free rewards when they make repeated purchases with a 2002) and a sense of participation (Dowling and Uncles
firm. Such a program rarely benefits consumers in one pur- 1997), which may be especially appropriate for brands that
chase but is intended to foster customer loyalty over time. do not carry this belongingness (Oliver 1999). All these
Thus, promotions that work as “one-shot deals,” such as psychological and economic benefits translate into an
instant scratch cards, are not considered loyalty programs attractive value proposition from the firm.
here. This exclusion is appropriate because these one-time
promotions do not create the same customer lock-in as true Loyalty Programs and Relationship Commitment
loyalty programs (Sharp and Sharp 1997). Beyond the need for superior value, a necessary condition
Consumer loyalty. This research adopts Oliver’s (1999, for any relationship to develop is the commitment of both
p. 34) definition of consumer loyalty as “a deeply held com- parties in the relationship (Morgan and Hunt 1994). Given a
mitment to rebuy or repatronize a preferred product/service wide variety of choices and a low switching barrier, it is
consistently in the future.” According to Oliver, consumer easy for today’s consumers to switch among different firms.
loyalty can occur at four different levels: cognitive, affec- This poses significant threats to customer relationships
tive, conative, and behavioral. Although all four facets of because consumers are not likely to commit to a single
consumer loyalty are important, the current research brand or firm. Loyalty programs can alleviate this lack of
focuses on behavioral loyalty. This aspect of consumer loy- commitment and reduce customer defection by raising
alty has not been thoroughly examined in previous research, switching costs. Because loyalty programs reward cus-
even though it has a direct impact on a firm’s bottom line tomers for their repeated patronage, consumers tend to
and facilitates the assessment of loyalty program profitabil- focus their purchases in one program to maximize the bene-
ity (and, relatedly, the decision to invest in such a program fits they receive (Sharp and Sharp 1997). Such vested inter-
or to expand or terminate an existing program). ests in a program make it difficult for competitors to entice
customers away from a firm. Using game-theoretic models,
Loyalty Programs and Value Enhancement Kim, Shi, and Srinivasan (2001) demonstrate that such a
Loyalty programs are often considered value-sharing instru- competitive barrier benefits the firm and results in higher
ments and can enhance consumers’ perceptions of what a prices in the marketplace. This is especially true for high-
firm has to offer (Bolton, Kannan, and Bramlett 2000; Yi variety-seeking products and services (Zhang, Krishna, and
and Jeon 2003). This value enhancement function is impor- Dhar 2000).
tant because the ability to provide superior value is instru- Loyalty programs not only help build customer commit-
mental to customer relationship initiation and retention ment but also demonstrate a firm’s commitment. It is often
(Sirdeshmukh, Singh, and Sabol 2002; Woodruff 1997). costly for firms to initiate and maintain a loyalty program. It
Indeed, enhanced value perception is considered a neces- requires extensive efforts to manage point records and
sary condition to a loyalty program’s success (O’Brien and reward issuance. After such a program is in place, it is usu-
Jones 1995). ally difficult to terminate it without risking the loss of con-

20 / Journal of Marketing, October 2007

sumers’ goodwill. Although a loyalty program brings real and Bramlett (2000) examine the effectiveness of loyalty
cost to the business, it also shows the firm’s commitment to programs in the financial industry. Verhoef finds that par-
establishing a long-term relationship with its customers. ticipation in an insurance firm’s loyalty program makes
Such a commitment and demonstration of goodwill can fur- consumers more likely to stay with the firm and encourages
ther deepen the relationship between the firm and its them to expand their business with the firm. Bolton, Kan-
customers. nan, and Bramlett offer a more in-depth examination of this
issue by studying the moderating effect of a credit card
Empirical Evidence of the Effectiveness of firm’s loyalty program on the relationship between con-
Loyalty Programs sumers’ service experiences and their subsequent behavior.
Lab and field studies have examined whether loyalty pro- They find that program members weigh negative experi-
grams indeed positively affect consumers. Although both ences less in their repatronage decisions than nonmembers.
types of research are important, this section focuses on This is consistent with the proposition that loyalty programs
empirical examinations of real loyalty programs based on form competitive barriers between firms and allow firms to
actual consumer behavior because they are most closely enjoy their customers more exclusively. Bolton, Kannan,
related to the current research. Existing studies in this area and Bramlett do not find a significant main effect of loyalty
can be classified into three categories depending on the program membership on customer retention, but their
comparison base on which conclusions are drawn. results show that loyalty program members used their credit
Comparison across competitors. The first group of stud- cards more than nonmembers.
ies quantifies the impact of loyalty programs by comparing As mentioned previously, studies comparing loyalty
them across multiple firms. The focal variables are usually program member and nonmember behavior are subject to
market share or share of wallet. Using consumer panel data self-selection bias. That is, the differences between mem-
of grocery purchases, both Mägi (2003) and Leenheer and bers and nonmembers may exist before the program rather
colleagues (2003) find mixed support for the positive than being a result of the program. This makes it difficult to
effects of loyalty programs on share of wallet. Leenheer and establish the direction of the causal relationship, prompting
colleagues’ study reveals increased share of wallet for four researchers to suggest the appropriateness of studying
of seven programs and offers support for the use of accu- dynamic behavior change instead (Lewis 2004; Verhoef
mulated rewards (as opposed to price discounts) in loyalty 2003).
programs. Mägi finds that loyalty program membership Comparison across time. The third category of studies
increases a consumer’s share of wallet and store visit and remedies self-selection bias by studying the same con-
decreases shares for competitors. However, this is supported sumers’ behavior across time. A majority of these studies
only at the chain level, not at the store level. Focusing on focus on short-term loyalty programs. A typical setting is an
the airline industry, both Kopalle and Neslin (2003) and N-week turkey/ham supermarket reward program in which
Nako (1992) conclude that frequent-flier programs enhance consumers need to spend over a set amount each week for
the value of an airline’s products and increase consumer N weeks to receive a free turkey or ham (e.g., Lal and Bell
demand for airlines that offer such programs. 2003; Taylor and Neslin 2005). Studies find general support
Within this same category of research but focusing on a for such programs in terms of increased spending levels.
different behavioral variable, Sharp and Sharp (1997) inves- Drèze and Hoch (1998) further conclude that a loyalty pro-
tigate the impact of Australia’s Fly Buys program by com- gram targeting a specific product category not only
paring observed purchase frequencies with the Dirichlet increases spending in the focal product category but also
baseline and find only a weak improvement in repeat- increases store traffic and overall spending in all categories.
purchase behavior for most stores. Using a similar When studying behavior changes over time, researchers
approach, Meyer-Waarden and Benavent (2006) find only suggest two types of effects: a short-term point pressure
mixed effects of the loyalty programs offered by several effect and a long-term rewarded behavior effect (Taylor and
French grocery retailers. Because members of the Fly Buys Neslin 2005). The point pressure effect represents a tempo-
program can earn loyalty points across stores, it can be rary shock in spending as consumers increase their purchase
argued that such a multistore loyalty program fosters loy- levels to qualify for a reward, analogous to the artificial
alty toward the program rather than toward any particular increase in sales during a promotion. Drawing on the goal-
store and may even encourage consumers to divide their gradient hypothesis, Kivetz, Urminsky, and Zheng (2006)
loyalty among multiple firms (Dowling and Uncles 1997). find that this point pressure effect increases as consumers
This limits the generalizability of the findings. Moreover, get closer to a reward, resulting in purchase acceleration.
rewards offered by the Fly Buys program (free air travel or However, they also find that after the reward is obtained, the
lodging) are unrelated to actions consumers need to take positive change in behavior dissipates, similar to the sales
(patronizing retail stores) to accumulate points. Recent dip after a promotion. In contrast, the rewarded behavior
research suggests that this type of program may elicit reac- effect refers to long-term sustained purchase increase,
tance from consumers and reduce their intrinsic motivation which can be a result of factors such as appreciation of the
to engage in the original purchase activities (Kivetz 2005). reward received and stronger loyalty toward the firm.
Comparison across consumers. The second type of These two effects manifest differently in short-term and
existing research compares the behavior of loyalty program continuous loyalty programs (e.g., frequent-flier programs).
members with that of nonmembers to identify the impact of Short-term programs are analogous to sales promotion and
loyalty programs. Both Verhoef (2003) and Bolton, Kannan, signify firms’ temporary commitment. They create a pri-

The Long-Term Impact of Loyalty Programs / 21

marily point pressure effect due to the programs’ novelty and Srinivasan 2001; Kivetz and Simonson 2003). These
and a clear time line and goal. Conversely, continuous loy- individual differences may have contributed to the mixed
alty programs represent firms’ long-term commitment and findings in the literature. A few studies reviewed have
are analogous to everyday low price. Although they may be begun to examine the moderating effects of individual char-
novel at first and consumers may experience point pressure acteristics (Lal and Bell 2003; Lewis 2004; Taylor and Nes-
each time they get close to a reward, the long-term effects lin 2005). However, the operationalization of some of these
of such programs are mainly the rewarded behavior type. variables limits their managerial relevance. The purchase-
Empirical findings on the effects of continuous loyalty level tiers in Lal and Bell’s (2003) study were based on the
programs across time are more sparse. Allaway and col- same time span (i.e., the entire analysis period) as the focal
leagues (2006) offer an indirect examination of longitudinal variables, and the reward redemption in Taylor and Neslin’s
effects by segmenting retail loyalty program members (2005) study occurred after the focal variables. Although
according to their behavior. They find that the program these variables are useful in assessing the success of a
positively affects only a small portion of consumers. Lewis short-term loyalty program after it is run, they provide lim-
(2004) examines the loyalty program of an online retailer ited foresight for continuous loyalty programs. For a contin-
and finds that the level of reward received in a prior period uous program, knowledge of preexisting differences among
positively affects the probability of making larger-sized consumers early on in the program will provide more mean-
transactions in the current period. Note that focusing on ingful managerial implications and better guidance on the
postreward effects presents two limitations. First, it creates design and management of such a program. Lewis (2004)
a recursive relationship because the level of reward a con- studied one such preexisting difference—namely, demo-
sumer receives in one period is itself contingent on the con- graphics. This should be extended to include other a priori
sumer’s behavior change. As a result, a higher-level pur- individual differences that may influence consumers’ reac-
chase in subsequent periods may simply be a continuation tions to a loyalty program.
of the previous positive reaction to the loyalty program
rather than a result of higher-level rewards received in prior
periods. Second, postreward effects capture only one type Research Hypotheses
of loyalty program effects. The point pressure research dis- Overview
cussed previously suggests that behavior changes can occur The current research aims to answer three questions: (1)
not only after but also before the receipt of a reward How do consumers change their usage levels after joining a
(Kivetz, Urminsky, and Zheng 2006; Taylor and Neslin loyalty program? (2) Does the program make consumers
2005). There are also subtler aspects of a loyalty program, more loyal over time? and (3) How do consumers with dif-
such as a sense of belongingness (Dowling and Uncle 1997) ferent initial spending levels respond differently to the pro-
and the perception of effort advantage (Kivetz and Simon- gram? These research questions specifically tackle the two
son 2003). Because consumers can earn free rewards only issues raised in the literature review. Because the purpose of
infrequently in most programs, their behavior changes are loyalty programs often is to increase the loyalty and value
typically driven by these subtler effects. contribution of consumers, it is important to know whether
this goal is fulfilled. This study examines how consumers
Summary gradually adapt their purchase behavior across an extended
The empirical studies reviewed here provide mixed support time span after their initial enrollment in a long-term loy-
for loyalty programs, and there is still much controversy alty program. As discussed previously, the implications of
over whether the loyalty program is an appealing marketing short-term versus long-term loyalty programs are different,
tool (Leenheer et al. 2003; Shugan 2005). The foregoing and so far, dynamic studies of the latter have been far more
analysis reveals the need to address two issues. First, given sparse and inconclusive. This research contributes to a bet-
the future orientation of loyalty programs, it is necessary to ter understanding in this area. By capturing program effects
integrate a long-term focus into the research of these pro- through the movement of time, this research extends
grams. This need is supported by the finding that CRM Lewis’s (2004) findings to include more general effects of a
efforts often do not produce obvious short-term gains but loyalty program and enables a firm to predict the path its
rather should be assessed in the long run (Anderson, For- consumers will take after joining such a program.
nell, and Lehmann 1994). Most of the longitudinal studies This research also examines how a loyalty program
reviewed focus on loyalty programs that ran only for a short affects consumers differently by examining the moderating
period. Lewis’s (2004) study is the only one that systemati- effect of consumers’ existing usage levels. It is of strategic
cally examines the dynamic effects of a continuous loyalty importance to examine the responses of consumers with dif-
program. However, as discussed previously, its focus on ferent initial usage levels because their value contributions
postreward effects represents an incomplete view of loyalty to a firm vary and can either increase or dilute the firm’s
program effects. Future longitudinal research should extend profitability (Dowling and Uncles 1997; Reinartz and
Lewis’s research to include more general effects of continu- Kumar 2000). Marketing researchers have suggested that
ous loyalty programs. usage level is an important consideration in loyalty program
Second, not all consumers respond to loyalty programs design and effects (Bolton, Kannan, and Bramlett 2000;
in the same way, because the appeal of a program can differ Dowling and Uncle 1997). However, this premise has not
among consumers, depending on factors such as current been subject to much empirical scrutiny. Although Lal and
usage levels and perception of effort advantage (Kim, Shi, Bell (2003) incorporate usage levels, they operationalize the

22 / Journal of Marketing, October 2007

variable as total spending during the entire course of a allows for a more accurate assessment of these programs’
short-term loyalty program, which, as mentioned previ- effects.1 Methodwise, studying individual transaction size
ously, provides limited foresight for the management of a instead of total spending increases the power of the analysis
continuous loyalty program. This research remedies this by and reduces aggregation bias. It also allows for the assess-
using consumers’ spending levels during an initialization ment of exclusive loyalty using the relationship between
period. The following section considers how different tiers transaction size and interpurchase time, as is demonstrated
of consumers change their usage levels and exclusive loy- subsequently.
alty over time.
Moderating Effect of Initial Usage Level
Impact of Loyalty Programs on Usage Levels The extent to which consumers increase their spending
Most loyalty programs are designed to encourage increased because of the incentives can depend on their initial usage
usage of a firm’s products or services. In general, the more levels. In the only existing research that has studied the
a consumer buys, the more rewards he or she is likely to moderating effects of consumer usage levels on loyalty pro-
earn. Thus, loyalty programs create an expectancy of posi- gram impact, Lal and Bell (2003) find the least behavior
tive outcomes associated with making a purchase (Vroom change among the heaviest spenders and more significant
1964). When consumers realize that their purchase behavior changes among low and moderate spenders. The current
is instrumental in achieving a positive outcome, they will be study extends these findings to continuous loyalty pro-
more likely to engage in the behavior (Latham and Locke grams, in which sustained behavior change, rather than tem-
1991). From an operant conditioning point of view (Skinner porary shock, is more likely to be the goal. Furthermore,
1953), rewards from loyalty programs serve as the condi- unlike Lal and Bell, who segment consumers according to
tioning stimulus to sustain desired behavior. In support of their spending levels during the entire duration of a pro-
these views, empirical studies show that rewards can direct gram, this research uses consumers’ usage levels at the
behavior and increase task performance (Eisenberger and beginning of the program, thus increasing the predictive
Rhoades 2001; Strohmetz et al. 2002). Thus, it is expected usefulness of the findings.
that loyalty programs will positively affect consumers’ Two considerations underlie how different consumer
usage levels, which leads to the first hypothesis: segments may respond differently to the same loyalty pro-
gram. First, depending on consumers’ existing usage levels,
H1: Consumers gradually increase their usage level after join- a loyalty program may be appealing to various degrees. If a
ing a loyalty program.
consumer buys little from the firm, he or she will need to
In this research, consumers’ usage levels are captured wait a long time for a reward. Thus, the consumer may not
by two variables: purchase frequency and transaction size. consider the program relevant. In contrast, heavy and mod-
Purchase frequency is considered an important predictor of erate buyers have an effort advantage over light buyers
consumers’ status with a firm (Schmittlein, Morrison, and because they do not need to work that hard or to wait that
Colombo 1987; Venkatesan and Kumar 2003) and has been long for the rewards (Kivetz and Simonson 2003). This
used in the past as an indicator of loyalty program success effort advantage can enhance the perceived fit and attrac-
(Bolton, Kannan, and Bramlett 2000; Sharp and Sharp tiveness of the loyalty program to such consumers.
1997). However, transaction size has rarely been included Although the idiosyncratic fit theory has been confirmed on
in previous studies. Because consumers can maximize consumers’ intention to join a loyalty program, it may con-
rewards by increasing how much they spend in a transaction tinue to influence consumers after program enrollment,
and because this amount is an essential part of their value especially for moderate buyers. For heavy buyers, this influ-
contribution to the firm, it is important to include transac- ence of effort advantage on purchase behavior is more
tion size in the study of loyalty programs. ambiguous. The reward literature suggests that when
Although usage levels can be measured as the total rewards do not offer enough challenge to task performance,
amount a consumer spends, studying purchase frequency they lack the motivational effect to induce behavior change
and transaction size separately is deemed to be more appro- (Eisenberger and Rhoades 2001). Because heavy buyers
priate for two reasons. First, purchase frequency and trans- already obtain rewards easily at the current purchase level,
action size have different implications for a firm. With high they will have less incentive to increase their efforts.
fulfillment costs per order (e.g., due to payment processing The second consideration in studying initial usage level
or shipping costs), basket size is an important determinant as a moderator is the ultimate limit of a consumer’s demand
of the firm’s profit margin. In comparison, for a firm in for a product or service, such as how much he or she travels
which recency of purchase is an important predictor of (Dowling and Uncles 1997). Motivation notwithstanding,
future behavior, purchase frequency may be a more critical consumers will raise their usage level only if it is below
business factor. Studying purchase frequency and size sepa- their consumption limit. Consequently, this creates a ceil-
rately is more in line with such firms’ strategies. Second, ing effect that is most likely to affect frequent buyers
reflecting the strategic difference between purchase fre-
1Although the particular loyalty program studied in this article
quency and transaction size, some loyalty programs reward
issues reward points based on the amount consumers spend at the
consumers for frequency (e.g., rewarding a set number of store, studying purchase frequency and transaction size separately
points for each purchase), and others encourage larger pur- helps establish baseline effects of the program on these two
chases (e.g., by setting a minimum transaction size). Differ- variables, with which future studies of other types of loyalty pro-
entiating between purchase frequency and transaction size grams can be compared.

The Long-Term Impact of Loyalty Programs / 23

because they already consume heavily (Lal and Bell 2003). the name of the firm is not disclosed here. The loyalty pro-
Together, these considerations suggest that moderate buyers gram allows consumers to earn points for every dollar they
should experience the greatest change in usage levels spend at the store. Tiers of rewards, such as a bottle of soda,
because they perceive effort advantage and higher relevance are related to the total number of points accumulated. The
of the program but are not as likely to be subject to the ceil- program rewards consumers an average of $1 in free
ing effect as heavy buyers. This leads to the following products/services for every $100 spent (i.e., 1% reward
hypothesis: ratio), with higher-tier rewards carrying a higher reward
H2: Moderate buyers’ usage levels increase faster than those of ratio. Consumers need to enroll in the program to earn free
heavy and light buyers after loyalty program enrollment. rewards, but program membership is free.
A random sample of 1000 consumers was extracted
Impact of Loyalty Programs on Consumer Loyalty from the program using two criteria: (1) The consumer
joined the loyalty program in its first year of operation, and
A main advantage of loyalty programs is their ability to (2) the consumer made at least two purchases. The latter
increase switching cost (Kim, Shi, and Srinivasan 2001). constraint ensures that there are meaningful data for every
When consumers join a loyalty program, to accumulate consumer. The data cover purchases during the first two
rewards more quickly, they are likely to concentrate their years of the program, which started in March 2002. Alto-
purchases on one firm, such as booking all flights through gether, the sample made 42,788 purchases. The number of
one airline. Furthermore, because loyalty program members purchases made by a consumer ranged from 2 to 369, with
tend to overlook negative experiences with the firm and are a median of 25. The median transaction size was $13.75.
less likely to compare the firm with competitors (Bolton,
Kannan, and Bramlett 2000), they are more likely to buy Modeling Purchase Frequency
exclusively from the firm.
Consumers’ purchase frequency is modeled as Equation 1,
In the long run, the increase in switching costs has
where Frequencyim is the number of transactions by con-
important implications for customer loyalty. First, the
sumer i in month m and Monthim is the number of months
longer a consumer has been in a program, the more vested
consumer i has been in the loyalty program at month m.
interests he or she will have in the program, and the more
Logarithmic transformation of Monthim is used to accom-
the consumer will have at stake if he or she were to leave
modate the notion that purchase frequency is unlikely to
the firm. This creates a long-term customer lock-in (Sharp
increase forever and will gradually reach a maximum point
and Sharp 1997). Second, higher switching costs mean that
that reflects consumption limit.2 The model has a dummy
loyalty program members are less likely to have extended
variable LastMonthim, which is set to 1 if month m is the
experience with competitors, further reducing their ability
last month of transaction by consumer i and 0 if not. This
to weigh competitor comparison in their decisions (Bolton,
variable is included because relationship marketing litera-
Kannan, and Bramlett 2000). Consequently, consumers are
ture shows that consumers tend to reduce their purchase fre-
expected to become more loyal after joining a loyalty pro-
quency at the end of a relationship (Venkatesan and Kumar
gram, which leads to the next hypothesis:
2003). Thus, it is necessary to control for the effect of the
H3: A firm’s loyalty program members become more loyal to last month’s purchases when studying the trend in purchase
the firm over time. frequency. To allow for temporary inactivity, LastMonthim
Similar to the change in usage levels, a loyalty program is set to 1 only if a consumer’s previous purchase occurred
can influence different consumers’ loyalty levels differently. at least three months before the end of the analysis period.
At one end of the continuum, light buyers may not be moti- (1) Frequencyim = αi0 + αi1Log(Monthim)
vated to become more loyal to a firm, because the loyalty
program is not highly attractive to them. At the other end, + αi2LastMonthim + εim.
heavy buyers already can enjoy frequent rewards and thus Two-level hierarchical linear modeling (HLM) is used
do not have a strong incentive to change their behavior. to estimate the parameters in the model. Compared with tra-
Again, moderate buyers are the most attractive target. These
consumers perceive enough relevance and benefits from the
program to change their purchase behavior and shift their
2Although analysis of supermarket purchase data often uses
purchases to one firm. This leads to the following
hypotheses: week as the time unit of analysis, month is chosen as the indepen-
dent variable here for three reasons. First, convenience store pur-
H4: Moderate buyers’ loyalty levels increase faster than those chases occur less frequently than supermarket shopping; industry
of heavy and light buyers after loyalty program data show that 68% of consumers shop at convenience stores
enrollment. about or less than once a week (Chanil 2004). This causes the
number of data points in weekly intervals to be too small and pro-
duces a disproportionate number of zero frequencies, which can
Data and Methodology skew the results. Second, behavior changes due to a loyalty pro-
gram are not expected to happen overnight, and using month as the
The Data independent variable allows for a reasonable time interval to
observe visible behavior changes. Third, the convenience store
The data used in this study came from a convenience store chain regularly makes decisions on a monthly basis, which renders
chain’s loyalty program. For the purpose of confidentiality, monthly analysis meaningful from a practical point of view.

24 / Journal of Marketing, October 2007

ditional linear regression, HLM has two advantages. First, it The terms β4 and β5 represent the differential effects of
does not require independent observations, as is often the loyalty program on light buyers and heavy buyers in
assumed in traditional regression. This accommodation of comparison with moderate buyers. According to H2, both
nonindependent observations is important here because the light and heavy buyers are not expected to increase their
purchase frequencies for a consumer are likely to be corre- usage levels as quickly. This implies a negative value for
lated across time. Second, HLM allows the model coeffi- both β4 and β5. The coefficient for the LastMonth dummy
cients at lower levels to be randomly distributed, thus variable, β6, indicates how much purchase frequency
accommodating individual heterogeneity. Explanatory decreases during a consumer’s last month in the relation-
variables can be included at higher levels to explain such ship. The μi1Log(Monthim) and μi2LastMonthim terms rep-
heterogeneity. resent the unexplained individual heterogeneity in purchase
In the purchase frequency model, the regression coeffi- frequency change over time and during the consumer’s last
cients in Equation 1 for an individual consumer are month as a customer.
assumed to be normally distributed across the sample, and
the expected parameter values for each individual consumer Modeling Transaction Size and Exclusive Loyalty
depend on whether the consumer is a light, moderate, or Traditionally, researchers have evaluated consumers’ loyalty
heavy buyer. Equations 2–4 show the second-level equa- to a brand or store through their brand-switching behavior.
tions used: In reality, however, a firm often does not have information
(2) αi0 = β0 + β1LightBuyeri + β2HeavyBuyeri + μi0, on its customers’ purchases other than those related to its
own products. To solve this problem, this research adopts an
(3) αi1 = β3 + β4LightBuyeri + β5HeavyBuyeri + μi1, and approach similar to that of Boatwright, Borle, and Kadane
(2003) that examines the relationship between interpur-
(4) αi2 = β6 + β7LightBuyeri + β8HeavyBuyeri + μi2, chase time and transaction size. The basic premise for this
approach is that interpurchase time and transaction size
where LightBuyeri and HeavyBuyeri are two dummy should have a proportional relationship for regularly pur-
variables indicating light and heavy buyers, respectively. chased products. For the same consumer, the longer he or
Appendix A explains in detail how these and other variables she waits before making a purchase, the more he or she will
were operationalized. need to buy in that shopping trip. For example, if a con-
To interpret the coefficients in the equations, Equations sumer who normally buys groceries once a week must wait
2–4 can be substituted into Equation 1, which produces the for two weeks instead, he or she will likely need to buy
following: twice the amount. If the consumer is loyal to one store, this
proportional relationship should be strong. However, if the
(5) Frequency im = β 0 + β1LightBuyeri + β 2 HeavyBuyeri
consumer frequents multiple stores, the observed purchases
+ β3Log( Month im ) from a single store are not likely to reflect such a systematic
relationship.
+ β 4 Log( Month im ) × LightBuyeri Mathematically, a three-level HLM (see Appendix B) is
+ β5Log( Month im ) × HeavyBuyeri used to model loyalty and transaction size. First, the amount
consumer i spent in the kth transaction in quarter j (Sizeijk)
+ β6 LastMonth im is modeled as a function of the elapsed time since the previ-
+ β 7 LastMonth im × LightBuyeri ous transaction, or interpurchase time (IPTijk). As discussed
previously, with exclusive loyalty, the expected value of ρij1
+ β8 LastMonth im × HeavyBuyeri should be close to 1.3 If little loyalty exists, ρij1 should be
+ μ i1Log( Month im ) + μ i 2 LastMonth im + υ im ,
3Assuming constant total demand from the consumer and per-
where υim = εim + μi0 represents random error. The intercept fect loyalty to the store, there should be a proportional relationship
β0 shows the expected purchase frequency of a moderate between interpurchase time and transaction size. In other words, if
buyer during the first month of the program (when interpurchase time is doubled and two shopping trips are com-
Log[Monthim] = 0), the coefficients for the LightBuyeri and bined into one, the amount spent in the trip should also double.
HeavyBuyeri dummy variables (β1 and β2) represent the Mathematically, according to Equation B1, if the interpurchase
time changes by a factor of α, the new transaction size
differences in these consumers’ initial purchase frequency (NewSizeijk) will be exp[ρij0+ ρij1log(α × IPTijk) + ςijk]. Thus,
compared with that of moderate buyers, and β3 is a key NewSizeijk/Sizeijk = αρij1. In a perfectly proportional relationship,
parameter that corresponds to the longitudinal effect of the NewSizeijk/Sizeijk should be equal to α, which means that ρij1
program on moderate buyers’ purchase frequency. The should be equal to 1. If a consumer is totally disloyal or if total
change in behavior due to Monthim is β3 × log(Monthim/ demand is unstable, the change in interpurchase time may have no
Monthim – 1). Because consumers’ purchase frequencies are impact on transaction size at all, and ρij1 will be equal to 0. This
hypothesized to increase as a result of the program, β3 does not mean that ρij1 is bound between 0 and 1. Mathematically,
ρij1 can be larger than 1 or smaller than 0. If ρij1 is larger than 1—
should be positive. The rate of purchase frequency growth for example, if it equals 2—doubling interpurchase time will result
gradually slows down as Monthim increases and in the quadrupling of transaction size. Although such situations
log(Monthim/Monthim – 1) becomes smaller and eventually may exist, logically, it is more likely to be the exception rather
approaches zero. than the norm. It is also possible for ρij1 to be negative, which sug-

The Long-Term Impact of Loyalty Programs / 25

close to 0. At the second level, the Level 1 intercept and the main models, one with homogeneous variance and the
slope are assumed to depend on the quarter in which the other with heterogeneous variance. Given the previous dis-
transaction occurred (Quarterij). Here, the unit of quarter cussion on the effects of initial usage levels, the two dummy
instead of month is used to accommodate consumers with variables for consumer tiers (HeavyBuyeri and LightBuyeri)
lower purchase frequencies and to ensure more accurate were used to explain Level 1 error variance. The results
estimates based on a stable trend. At the third level, con- show that the heterogeneous variance specification outper-
sumer heterogeneity is captured by allowing Level 2 formed the homogeneous variance specification for both
parameters to depend on a consumer’s initial usage tier. models, suggesting the existence of heterogeneous error
Equation B8 in Appendix B combines all three levels of variance at Level 1. However, the parameter estimates from
equations. For transaction size change due to the loyalty the two specifications did not differ substantively from each
program, the relevant parameters are λ0–λ5. The parameters other. The results are based on heterogeneous error variance
λ0, λ1, and λ2 are related to the intercept term in Equation specification.
B8, which reveals consumers’ baseline transaction sizes at
the beginning of the program, assuming daily transactions
(i.e., IPT = 1), and the parameters λ3–λ5 have to do with the Results
changes in transaction size over time for different segments
Change in Purchase Frequency
of consumers. Specifically, λ3 refers to the changes in mod-
erate buyers’ transaction size because of the loyalty pro- The maximum likelihood estimates of the purchase fre-
gram, and λ4 and λ5 indicate the moderating effect of initial quency model and its goodness of fit appear in Table 1. The
usage levels on the change in transaction size. HLM does not produce an R-square as in traditional regres-
For consumer loyalty, the parameters of central interest sion. However, it yields a deviance statistic, which equals
are λ6–λ11. The parameters λ6–λ8 show the loyalty of the –2 times the value of the log-likelihood function and can be
three consumer segments during the first quarter in the pro- used to evaluate alternative models (Raudenbush and Bryk
gram. The parameters λ9–λ11 reflect the longitudinal 2002). As Table 1 shows, compared with an unconditional
change in consumer loyalty. As H3 predicts, consumer loy- model that does not have any explanatory variables, the pro-
alty should rise after program enrollment. Thus, λ9 should posed purchase frequency model shows a significantly bet-
be positive, reflecting the increase in loyalty for moderate ter fit (χ2 = 2170.58, p < .001).
buyers. In contrast, λ10 and λ11 should be negative, indicat- Recall that the model intercept represents expected pur-
ing the slower increase in loyalty for light and heavy buy- chase frequency of moderate buyers during the first month,
ers. Note that though convenience store purchases occur which was 2.59 times (p < .001). The coefficients for the
less regularly than supermarket purchases, industry statis- LightBuyer and HeavyBuyer dummy variables (β1 and β2)
tics show that many consumers shop at convenience stores indicate the differences in these consumers’ initial purchase
consistently and extensively (General Electric Capital Fran- frequencies compared with moderate buyers. Thus, they
chise Finance Corporation 2001). In addition, the store serve as a check of the correct classification of consumers.
chain from which the data were obtained sells both fuel and Consistent with their segmentation, light buyers’ initial pur-
convenience store items. The regular nature of fuel purchase chase frequency was .93 times lower than that of moderate
adds regularity to the sample’s purchases. Mathematically, buyers (p < .001), and the average initial frequency for
the model is estimated with entire quarters of transactions, heavy buyers was 3.00 times higher (p < .001).
which helps smooth out the randomness in purchase pat- H1 and H2 predict that consumers’ purchase frequencies
terns and makes the comparison across time meaningful. will gradually increase and that this increase will be fastest
for moderate buyers. The results show a positive coefficient
Data Analysis Overview for Log(Month) (β3 = .26, p = .002), suggesting that moder-
ate buyers purchased more frequently over time. At the end
Both the purchase frequency model and the loyalty/trans-
of the two years, the average purchase frequency for moder-
action size model were fitted using HLM6. The standard
ate buyers was 4.42 times, nearly doubling their initial fre-
HLM assumes that higher-level units (in this case, con-
quency. Consistent with H2, the increase in purchase fre-
sumers) are drawn from the same population, which implies
quency for moderate buyers was significantly higher than
homogeneous error variance at Level 1. This assumption
that for heavy buyers (β5 = –.32, p = .008). The combined
may not be realistic here, because different segments of
effect of Log(Month) for heavy buyers was nonsignificant
consumers may behave differently. To accommodate such
(β3 + β5 = –.05, p = .48), suggesting an unchanged purchase
considerations, the assumption can be relaxed to allow for
frequency for these consumers. This is further confirmed by
heterogeneous Level 1 error variance and to include predic-
the finding that heavy buyers’ average frequency at the end
tor variables to explain the heterogeneity (Raudenbush and
of the period (5.68 times) was virtually unchanged from the
Bryk 2002). Thus, two specifications were fit for each of
initial frequency. The combined effect of Log(Month) for
light buyers was .34. Contrary to H2, the Log(Month) ×
LightBuyer interaction was nonsignificant, indicating that
gests that an increase in interpurchase time results in a decrease in light buyers showed the same level of increase in purchase
transaction size. Again, this is likely to be an exception to what is frequency as moderate buyers. Their purchase frequency
usually observed in the marketplace, given that purchases occur more than doubled to 3.73 times at the end of the period.
relatively often. These increases in purchase frequency by light and moder-

26 / Journal of Marketing, October 2007

 TABLE 1
Model Results

Purchase Loyalty/Transaction Reward Claim

Frequency Model Size Model Model

Intercept 2.59*** (.13) 1.43*** (.04) .32*** (.03)

LightBuyer –.93*** (.18) –.48*** (.09) –.17*** (.04)
HeavyBuyer 3.00*** (.20) .38*** (.07) .25*** (.04)
Log(Month) .26*** (.08) — —
Log(Month) × LightBuyer .08n.s. (.11) — —
Log(Month) × HeavyBuyer –.32*** (.12) — —
LastMonth –1.21*** (.19) — —
LastMonth × LightBuyer .46* (.26) — —
LastMonth × HeavyBuyer –1.12*** (.36) — —
Log(Quarter) — .16** (.07) .13*** (.02)
Log(Quarter) × LightBuyer — .07** (.03) .01n.s. (.03)
Log(Quarter) × HeavyBuyer — –.14*** (.05) –.06** (.03)
Log(IPT) — .15*** (.03) —
Log(IPT) × LightBuyer — –.05** (.02) —
Log(IPT) × HeavyBuyer — .21*** (.08) —
Log(IPT) × Log(Quarter) — .10*** (.04) —
Log(IPT) × Log(Quarter) × LightBuyer — .03* (.02) —
Log(IPT) × Log(Quarter) × HeavyBuyer — –.09*** (.02) —

Deviance (–2 log-likelihood) 50,393.27 110,393.08 9684.30

χ2 2170.58*** 711.67*** 475.60***
d.f. 16 22 10
*p ≤ .10.
**p ≤ .05.
***p ≤ .01.
Notes: The numbers in parentheses are standard errors of the estimates. The chi-square statistic compares the deviance of the estimated
model with an unconditional model that does not contain predictor variables at any of the levels. n.s. = not statistically significant.

ate buyers are especially impressive in light of industry sta- size by moderate buyers should be eλ0, and the baseline size
tistics showing that more than two-thirds of shoppers fre- for light and heavy buyers should be exp(λ0 + λ1) and
quent convenient stores about or less than once a week exp(λ0 + λ2), respectively. The results show a baseline
(Chanil 2004). In other words, after two years, most of the transaction size of $4.18 for moderate buyers (λ0 = 1.43,
sample became the top third of all convenience store shop- p < .001). Consistent with their classifications, heavy buy-
pers in terms of purchase frequency. ers initially spent $1.93 more in a transaction than moderate
Figure 1, Panel A, displays the observed average pur- buyers (λ1 = .38, p < .001), and light buyers’ initial daily
chase frequencies of the three consumer segments for each transaction size was $1.59 less than moderate buyers (λ2 =
month. For both light and moderate buyers, the most visible –.48, p < .001). The observed average transaction sizes for
jump in purchase frequency occurred within three months these consumers appear in Figure 1, Panel B.
of joining the loyalty program. These higher frequencies Consistent with H1, moderate buyers spent more in a
sustained and steadily increased at a slower pace after the transaction over time (λ3 = .16, p = .03) and had an average
first three months. In contrast, heavy buyers’ purchase fre- observed transaction size of $20.11 at the end of the analy-
quency remained mostly flat during the analysis period. sis period. Again, heavy buyers’ transaction size did not
Paired comparison tests suggest that at the end of the two increase as fast as moderate buyers’ (λ5 = –.14, p = .006),
years, there was still a significant difference in observed and the combined coefficient for Log(Quarter) was nonsig-
purchase frequencies between light buyers and heavy buy- nificant (λ3 + λ5 = .02, p = .76), suggesting that heavy buy-
ers (t = –1.99, p = .05). However, there was no significant ers did not spend more in a purchase after they joined the
difference between light and moderate buyers or between loyalty program. Surprisingly, light buyers showed a faster
moderate and heavy buyers. Overall, H1 and H2 are partially increase in transaction size than moderate buyers (λ4 = .07,
supported for purchase frequency. p = .03). Their average observed transaction size increased
to $11.29 at the end. This contradicts H2, which predicts a
Change in Transaction Size slower increase for light buyers. For all three segments, the
The loyalty/transaction size model showed a significantly average transaction size far exceeded the industry average
better fit than an unconditional model (χ2 = 711.67, p < of $7.60 per transaction (Chanil 2004).
.001). The model estimates appear in Table 1. Recall that To explore further the source of light buyers’ transaction
the logarithmic transformation of transaction size is used in size increase, an additional analysis was performed on their
the model. Thus, the expected beginning daily transaction shopping basket composition. The convenience store chain

The Long-Term Impact of Loyalty Programs / 27

 FIGURE 1
Observed Changes in Consumer Purchases

A: Observed Average Monthly Purchase Frequencies

B: Observed Average Transaction Sizes

sells products in two major categories: fuel and store mer- dise also increased from 20% to 40%. This inclusion of
chandise. During their first quarter in the program, only more product categories in one purchase explains why light
26% of light buyers bought both fuel and store merchan- buyers spent more in a transaction. Although these con-
dise, and the other 74% bought only either fuel or store sumers may not initially experience a strong incentive in the
merchandise. By the end of the two years, the percentage of loyalty program, they were able to diversify their purchases
double-category buyers increased to 58%. The percentage into more categories and thus make the program more
of transactions that included both fuel and store merchan- attractive.

28 / Journal of Marketing, October 2007

Exclusive Consumer Loyalty (Log[Quarteri]). Here, the unit of quarter instead of month
Estimates related to consumer loyalty appear in Table 1. is used because of the relatively infrequent reward issuance.
Recall that the more loyal a consumer is to the store, the Again, the consumers’ initial usage levels were entered as
closer the coefficient of Log(IPTijk) (hereinafter, “the loy- explanatory variables at the second level. Equation 9 shows
alty parameter”) should be to 1. At the beginning, the loy- the combined model:
alty parameter for moderate buyers was .15 (p < .001), indi- (6) RCRate iq = γ i 0 + γ i1Log(Quarteriq ) + π ij ,
cating relatively low loyalty. Light buyers were even less
loyal, with the loyalty parameter lower by .05 (p = .035). (7) γ i 0 = θ0 + θ1LightBuyeri + θ2 HeavyBuyeri + τ i 0 ,
Heavy buyers were most loyal among the three groups, with
(8) γ i1 = θ3 + θ4 LightBuyeri + θ5HeavyBuyeri
the loyalty parameter higher than that of moderate buyers
by .21 (p = .007). + τ i1 , and
The dynamic change in the loyalty parameter is of cen-
tral interest. As hypothesized, moderate buyers’ loyalty (99) RCRate iq = θ0 + θ1LightBuyeri + θ2 HeavyBuyeri
increased significantly (λ9 = .10, p = .005). Consistent with + θ3Log(Quarteriq )
H4, the increase in loyalty among heavy buyers was slower
than that for moderate buyers (λ11 = –.09, p < .001). A non- + θ4 Log(Quarteriq ) × LightBuyeri
significant combined coefficient for heavy buyers (λ9 +
λ11 = .01, p = .99) suggests that these consumers’ loyalty + θ5Log(Quarteriq ) × HeavyBuyeri
levels did not change during the analysis period. Con- + τ i1Log(Quarteriq ) + τ i1 + π iq .
versely, light buyers experienced a larger increase in exclu-
sive loyalty than moderate buyers (λ10 = .03, p = .07). This This two-level HLM model was estimated with hetero-
contradicts H4, which predicts that moderate buyers will geneous error variance at Level 1, and the results appear in
experience a faster increase in loyalty than light buyers.
Table 1. The deviance of the model was 9684.30, showing a
Overall, H4 is only partially supported.
significantly better fit than an unconditional model (χ2 =
475.60, p < .001). The intercept estimate indicates that
Alternative Explanations moderate buyers had an initial reward claim rate of 32.22%
(p < .001). Consistent with Lal and Bell (2003), heavy buy-
Learning Effect ers initially claimed 25.42% more of their rewards than
An alternative explanation for the findings is that light buy- moderate buyers (p < .001), and light buyers’ initial reward
ers may be new customers who learn to spend more over claim rate was 16.71% lower than that of moderate buyers
time as they become more familiar with the store. In con- (p < .001). The results on the longitudinal change in con-
trast, heavy buyers may already be long-term loyal cus- sumers’ reward claim behavior revealed notable patterns. In
tomers and do not experience the same incremental learn- two years, moderate buyers’ reward claim rate increased to
ing. Thus, the differences among the consumer segments 59.79% in the eighth quarter (θ3 = .13, p < .001). In con-
may be due to a learning effect rather than different reac- trast, heavy buyers’ reward claim rate did not increase as
tions to the loyalty program. To rule this out, consumers’ fast (θ4 = –.06, p = .029). Light buyers showed the same
reward redemption was examined. For this program, to level of increase in reward claim rate as moderate buyers
receive a reward, consumers needed to request a certificate (θ5 = .01, p = .563).
from the store, which could then be redeemed for the Because claiming a reward is an extra step that con-
reward. In reality, only some consumers will make the effort sumers need to take to reap the benefits of the loyalty pro-
to request their reward certificates (Lal and Bell 2003). gram, their decision to do so reveals their level of interest in
Because all three consumer segments were exposed to the the loyalty program. Given their high spending levels and,
loyalty program at the same time, there is no learning thus, the ability to obtain many rewards, it is not surprising
advantage on reward claim for any segment, and thus any that heavy buyers had the highest interest in rewards. In the
systematic differences should truly reflect their diverse meantime, the faster increase in light and moderate buyers’
responses to the program. Existing research shows that reward claim rates shows a rising interest in the loyalty pro-
reward redemption tends to be the highest among heavy gram from these consumers. As these consumers change
buyers (Lal and Bell 2003). The following analysis retests their purchase behavior to make the loyalty program more
such findings and examines longitudinal change in reward “profitable” for them, they gradually become more invested
redemption behavior of loyalty program members, which and interested in the program. Higher reward claim rates
has yet to be answered by existing research. further allow them to benefit more from the program, form-
Similar to purchase frequency, consumers’ reward claim ing a virtuous cycle that provides more incentive for these
behavior was modeled with a two-level HLM model, as consumers to become better customers. Because reward
shown in Equations 6–8. The main dependent variable is claim behavior is not subject to the same learning differ-
RCRateiq, which is the number of reward certificates ences among consumer segments as purchase behavior,
requested as a percentage of the rewards consumer i quali- these findings support the conclusion that the change in
fied for in the qth quarter. It is modeled as a function of a behavior of the three segments was indeed a result of the
logarithm transformation of the corresponding quarter loyalty program.

The Long-Term Impact of Loyalty Programs / 29

Consumer Attrition non–loyalty program members as a control group and com-
Another alternative explanation for the current findings is pare their behavior changes over the same period. Unfortu-
consumer attrition. It is natural that some consumers would nately, the firm does not track individual purchases of its
drop out during the two years. Because high-value con- non–loyalty program members, making data about such a
sumers who consider the program attractive are more likely control group unavailable. However, it records company-
to stay, this creates a self-selection effect. Consequently, the level total sales and overall number of transactions. This
trends found may be a result of a sample composition allows for the derivation of the total spending and number
change toward a denser concentration of high-value con- of transactions made by all non–loyalty program members
sumers rather than a result of the program. To rule out this as a whole. Although comparison of individual consumer
alternative account, the models were reestimated with only behavior is still impossible with such data, it is possible to
data from consumers who were still with the firm at the end study the trends in all loyalty program members’ versus
of the two years. The model estimates appear in Table 2. nonmembers’ purchase behavior as two aggregated units.
The patterns of findings from this active consumer In the spirit of prior analysis of purchase frequency and
group are similar to those from the entire sample. That is, transaction size, the trends in total number of transactions
light and moderate buyers exhibited positive change in pur- per month (TotalTransactionsi) and average transaction size
chase behavior/loyalty, whereas heavy buyers did not. Not by members versus nonmembers (AvgSizei) are examined
surprisingly, the extent of change by light and moderate with the following regression equations:
buyers was even more prominent in this active consumer (10) TotalTransactions i = a 0 + a1Loyalty
group because they were likely to perceive the program as
more valuable. In contrast, the heavy buyers in this group + a 2 Log( Month i ) + a 3Log( Month i ) × Loyalty + ζ, and
maintained the same “no-change” pattern, as the estimates
(11) AvgSize i = b0 + b1Loyalty + b2 Log( Month i )
from the entire sample suggest. Overall, after sample com-
position change was controlled for, the substantial findings + b3Log( Month i ) × Loyalty + ξ,
still remain the same, suggesting that consumer attribution
is not the reason for the findings. where Loyalty is a dummy variable that equals 1 for loyalty
program member group and 0 for the nonmember group,
Store-Level Trends and ζ and ξ are errors. Although these equations are not in
The trends discovered in this research may also be attrib- hierarchical forms, they resemble prior models in that the
uted to other factors concurrent with but unrelated to the dependent variables are also a function of the number of
loyalty program, such as sales promotion. The most desir- months since the start of the program (i.e., Log[Monthi]). If
able way to rule out such extraneous factors is to include loyalty program members’ behavior change was indeed due

TABLE 2
Model Estimates with Active Consumers Only

Loyalty/Transaction
Frequency Model Size Model

Intercept 2.69*** (.22) 2.23*** (.08)

LightBuyer –.95*** (.34) –.62*** (.14)
HeavyBuyer 3.22*** (.30) .32*** (.10)
Log(Month) .55*** (.13) —
Log(Month) × LightBuyer .03n.s. (.20) —
Log(Month) × HeavyBuyer –.54*** (.18) —
Log(Quarter) — .27*** (.04)
Log(Quarter) × LightBuyer — .05*** (.02)
Log(Quarter) × HeavyBuyer — –.26*** (.05)
Log(IPT) — .22*** (.03)
Log(IPT) × LightBuyer — –.11*** (.05)
Log(IPT) × HeavyBuyer — .14*** (.04)
Log(IPT) × Log(Quarter) — .17*** (.05)
Log(IPT) × Log(Quarter) × LightBuyer — .05*** (.02)
Log(IPT) × Log(Quarter) × HeavyBuyer — –.15*** (.03)

Deviance (–2 log-likelihood) 31,261.72 80,786.44

χ2 937.28*** 425.33***
d.f. 9 22
*p ≤ .10.
**p ≤ .05.
***p ≤ .01.
Notes: The numbers in parentheses are standard errors of the estimates. The chi-square statistic compares the deviance of the estimated
model with an unconditional model that does not contain predictor variables at any of the levels. n.s. = not statistically significant.

30 / Journal of Marketing, October 2007

to the program, there should be a larger increase in these programs are one-to-one programs. How much a consumer
consumers’ purchases than in nonmembers’ purchases. That can benefit from such a program depends on his or her
is, the coefficients for the Loyalty × Log(Monthi) inter- “investment” in the relationship with the firm. However,
action term should be positive and significant. This was this one-to-one nature of loyalty programs has not been
confirmed for both the number of transactions per month thoroughly examined in existing research. A surprising
(a3 = 1.14, p < .001) and the average transaction size (b3 = finding from the current research is that consumers who
.83, p < .001).4 started with low usage levels changed their behavior as
Consistent with these results, the number of purchases much as or more than moderate and heavy buyers. This
that loyalty program members made increased from 4.98% contradicts the commonly held belief that light buyers are
to 8.11% of total transactions by the end of the two-year less-than-ideal targets for loyalty programs and that they
period. When this ratio is calculated for dollar sales, loyalty will not perceive much value in the program (Dowling and
program members accounted for 73.66% of total sales at the Uncles 1997; O’Brien and Jones 1995). In the current case,
beginning of the program, which increased to 88.91% after the loyalty program did not initially appear very attractive
two years. In contrast with the transaction ratio, the dollar to light buyers. However, these consumers diversified their
amount ratio suggests that these consumers spent much purchases and branched into the firm’s other service areas.
more in each transaction than nonmembers. Regression By claiming a higher portion of rewards, they also gradually
analysis with these two ratios as the dependent variables invested more efforts into the program. Through these mea-
and Log(Monthi) as the independent variable showed sig- sures, the opportunity for these consumers to benefit from
nificant, positive trends over time. Together, these findings the loyalty program increased, further motivating them to
suggest that loyalty program members exhibited more posi- spend more and patronize the store more exclusively.
tive trends than nonmembers, providing further support that
the program caused the purchase increase beyond other fac- Limitations and Further Research
tors that may exist in the environment. This study has several limitations that need to be addressed
in further research. For example, this study examined loy-
Conclusions alty program members’ behavior without a control group of
consumers who did not enroll in the program. Although the
This research examines the impact of a loyalty program on use of longitudinal data reduces the self-selection bias that
consumers’ purchase behavior over a two-year period. It often complicates cross-sectional analysis of loyalty pro-
extends prior studies by explicitly modeling the dynamic grams (Leenheer et al. 2003), the lack of control leaves the
change in consumers’ spending levels and their behavioral possibility that extraneous factors produced the trends
loyalty to the store. The results suggest that depending on rather than the loyalty program, such as concurrent market-
consumers’ initial usage levels, the loyalty program had dif- ing activities or environmental factors. Although measures
ferent effects on their behavior. Consumers who were heavy were taken to address several alternative explanations, they
buyers at the beginning of the program were most likely to do not cover the full range of issues. A complete test of loy-
claim the rewards they earned and thus benefited the most alty program effects is needed in the future, which should
from the program. However, their spending levels and use longitudinal data from both loyalty program members
exclusive loyalty to the store did not increase over time. In and nonmembers. Such a comparison of trends between the
contrast, the loyalty program had positive effects on both two groups will reveal more precise loyalty program effects.
light and moderate buyers’ purchase frequencies and trans- More comprehensive tests of loyalty program effects should
action sizes, and it made these consumers more loyal to the also go beyond spending levels and purchase timing to
store. The most visible change for these two segments include brand choice (Gupta 1988) and attitudinal loyalty
occurred within three months of joining the program, and (Oliver 1999).
the growth continued at a steady but slower pace in the fol- When interpreting the current findings, it is necessary to
lowing months. At the end of the analysis period, these con- keep in mind that the results are bound by the context and
sumers’ average purchase frequencies were not statistically structure of the program studied and thus may not general-
different from that of an adjacent tier. This supports the ize to other programs. In reality, the performance of differ-
argument that loyalty programs can accelerate consumers’ ent loyalty programs varies. It is important to understand
loyalty life cycle and make them more profitable customers why some programs achieve their goals whereas others fail
(O’Brien and Jones 1995). to do so (Bolton, Kannan, and Bramlett 2000). Several fac-
The diverse responses from consumers suggest a need tors are proposed in the literature, such as the effort
to consider consumer idiosyncrasies when assessing the required to earn rewards (Kivetz and Simonson 2002) and
impact of loyalty programs. By their very nature, loyalty the convenience of participating in the program (O’Brien
and Jones 1995). Further research should test how these and
other factors can affect a program’s effectiveness. Within a
4Note that the sample size for the regressions is small (24).
loyalty program, the effects of consumer self-segmentation
Thus, the power of the analysis is limited. Furthermore, because of also need to be examined. A particularly worthwhile topic is
the lack of individual data for nonloyalty consumers, it is not pos- how reward redemption behavior interacts with purchase
sible to derive the loyalty parameter or to account for possible behavior to moderate the influence of a loyalty program.
changes in sample composition over time for these consumers. Existing studies have begun to examine this in the context

The Long-Term Impact of Loyalty Programs / 31

of short-term loyalty programs (Lal and Bell 2003; Taylor that light and moderate customers enrolled in the loyalty
and Neslin 2005). However, more research is needed to program increased their value contribution and accelerated
study this issue in the context of continuous loyalty pro- their relationship life cycle with the firm, turning the pro-
grams. At the firm level, the performance of a loyalty pro- gram into much more than a passive loss-prevention instru-
gram may depend on the firm’s other marketing activities, ment. These findings suggest a need for managers to expand
such as sales promotion. Future studies should examine the their mentality toward loyalty programs beyond mere reac-
interaction between loyalty programs and other CRM tech- tive tactics.
niques and marketing activities in enhancing consumers’ The additional sales from the light and moderate buyers
relationships with a firm. Research in this area will provide in this study came from two sources: (1) concentration of
theoretical and managerial guidance on formulating the purchases originally scattered at other firms and (2) expan-
most effective CRM strategy. sion of the relationship with the firm into other business
Another limitation of this research is the lack of areas. These findings suggest a few prerequisites for the
competitive information. As a result, it relied on the propor- success of a loyalty program as a more active marketing
tional relationship between transaction size and interpur- tool. First, the lower spending among light and moderate
chase time to infer consumer loyalty. Although this is a use- users should be mainly due to polygamous loyalty (i.e., fly-
ful way to assess loyalty when data are limited, it can be ing on multiple airlines) rather than insufficient need for the
subject to other influences, such as stockpiling and cherry- product/service category (i.e., infrequent need to travel). If
picking behavior. The more irregular nature of convenience most consumers are not spending much because of low
store purchases can also make the relationship more tenu- absolute demand, a loyalty program is unlikely to have a
ous than it would be in more regular purchase scenarios. significant impact. Second, the loyalty program structure
Thus, when competitive information is available, share of should be set in such a way that it creates enough incentive
wallet is still a better way to assess consumer loyalty. Relat- for light and moderate buyers to strive for the rewards. In
edly, further research should go beyond a single program to other words, the possibility of obtaining a reward should
examine the market dynamism of loyalty programs because not be so remote that these consumers simply give up and
multiple firms often compete with one another through their dismiss the program as irrelevant to them. From this per-
loyalty programs. How does the introduction of new loyalty spective, smaller but easier-to-achieve rewards are likely to
programs influence the effectiveness of existing programs? be more effective than larger rewards that require a signifi-
Does order of entry affect loyalty programs’ performances? cant amount of effort. Finally, gaining additional sales
These are all important questions for further research. through a loyalty program is more likely when a firm has
Finally, this research used HLM to accommodate corre- multiple business areas that it can cross-sell to consumers,
lated observations within each consumer and estimated two such as in the case of retail and financial industries. The
separate models for transaction size and purchase fre- availability of such additional venues for consumers to
quency. However, because consumers may make purchase accumulate program points can make better use of con-
timing and quantity decisions simultaneously, the two mod- sumers’ creative minds and involve them in deeper, more
els may actually be related models. As a result, modeling extensive relationships with the firm.
the two decisions separately can lead to inefficient and Across consumer segments, a loyalty program redistrib-
biased model estimates (Krishnamurthi and Raj 1988). The utes revenues and costs among consumers. The eventual
amount of bias depends on the degree of simultaneity benefit of the program depends on the trade-off between the
between the two decisions and the level of correlation costs of rewards (a majority of which will be incurred as a
between the two models’ error terms (Leeflang et al. 2000). result of heavy buyers, who will enjoy the free rewards
This issue can be remedied in future studies through alter- without changing their behavior) and the increased profits
native modeling techniques, such as multivariate Tobit II from moderate and light buyers as they purchase more and
models in a simultaneous equations approach (Kamakura become more loyal over time. In highly competitive mar-
and Wedel 2001; Leenheer et al. 2003). kets in which loyalty programs are widely used (i.e., the air-
line industry), consumers may come to expect loyalty pro-
Managerial Implications grams as a standard offering from each firm. When this
The loyalty program is an important form of CRM strategy. happens, the cost of a loyalty program can become an
It is costly to initiate and maintain and often requires a essential cost of business, creating a competitive equilib-
firm’s long-term commitment. For many firms, a loyalty rium in which consumers are redistributed among compet-
program is considered a defensive marketing mechanism, ing firms. The ultimate benefit for the firm is the ability to
used to keep a core group of best customers from defecting. attract repeat business and to have more profitable loyal
This is especially the case when competing loyalty pro- customers.
grams are offered in the same market. Because the best cus- From an analytical standpoint, a loyalty program can
tomers often are already heavy buyers of a firm’s products produce rich data about customers, which should be used to
and services, the possibility of obtaining additional reve- enhance a firm’s relationship marketing efforts. The varia-
nues from these consumers is low. Thus, when treated as a tion in different consumer segments’ responses to the loy-
defensive strategy, loyalty programs are almost purely cost alty program found here suggests a need to assess the
items used to prevent potential sales loss. In contrast to this effects of loyalty programs beyond overall sales impact.
traditional view, the results from the current research show This research demonstrates a simple way of evaluating loy-

32 / Journal of Marketing, October 2007

alty program effects using individual purchase data. The Appendix B
HLM method better accommodates heterogeneity and cor-
related observations than traditional regression, without
Three-Level HLMs of Exclusive
incurring substantial execution costs. The models can also Consumer Loyalty
be easily adapted by adding other consumer variables, such Level 1
as demographics, into the consumer-level equations.
As firms often have only internal transaction data, the ( B1) Log(Size ijk ) = ρij0 + ρij1Log( IPTijk ) + ς ijk .
exclusive loyalty model presented here can be useful in
assessing consumers’ behavioral loyalty using limited data. Level 2
It is appropriate for situations in which consumer purchases
( B2) ρij0 = φi 0 + φi1Log(Quarterij ) + ϖij0 , and
occur frequently and only internal data are available, such
as consumers’ patronage choices at a supermarket. For such ( B3) ρij1 = φi 2 + φi 3Log(Quarterij ) + ϖij1.
high-frequency purchases, retrospective self-report data
tend to be highly inaccurate (Schacter 1999). The current Level 3
model is easy to compute and does not require knowledge
of where and to what extent consumers buy outside the ( B4) φi 0 = λ 0 + λ1LightBuyeri + λ 2 HeavyBuyeri + πi0 ,
firm, thus making it practical with minimal data or compu- ( B5) φi1 = λ 3 + λ 4 LightBuyeri + λ 5HeavyBuyeri + π i1 ,
tational requirements. However, a precaution is that the loy-
alty parameter can be affected by extraneous factors, such ( B6) φi 2 = λ 6 + λ 7 LightBuyeri + λ8 HeavyBuyeri + π i 2 , and
as stockpiling and subsequent consumption rate change ( B7) φi 3 = λ 9 + λ10 LightBuyeri + λ11HeavyBuyeri + π i3.
(Sun 2005). Thus, it should be used for product categories
with relatively stable consumption. It is not suitable when Interpretive Equation (Combining All Levels into a
the sample size is relatively small or the data time duration Single Equation)
is relatively short.
( B8) Log(Size ijk ) = λ 0 + λ1LightBuyeri + λ 2 HeavyBuyeri

Appendix A + λ 3Log(Quarterij ) + λ 6 Log( IPTijk )

Variable Operationalization + λ 4 LightBuyeri × Log(Quarterij )
•Purchase frequency (Frequencyim): Total number of pur-
+ λ5HeavyBuyeri × Log(Quarterij )
chases made by a consumer in a certain month. If consumer i
did not purchase anything during month m but that month is + λ9 Log(Quarterij ) × Log( IPTijk )
not the last month of the relationship (see the subsequent
explanation for the LastMonthim variable), Frequencyim is set + λ 7 LightBuyeri × Log( IPTijk )
to 0 to indicate zero purchase frequency.
•Transaction size (Sizeijk): Total dollar amount spent in a + λ8 HeavyBuyeri × Log( IPTijk )
transaction.
•Month (Monthij, Monthim): Number of months since joining + λ10 LightBuyeri × Log(Quarterij )
the loyalty program.
× Log( IPTijk ) + λ11HeavyBuyeri
•Quarter (Quarterij, Quarteriq): Number of quarters since
joining the loyalty program. × Log(Quarterij ) × Log( IPTijk )
•Consumer tiers (HeavyBuyeri, LightBuyeri): Classification of
a consumer based on his or her total spending during the first + [π i 0 + π i1Log(Quarterij )
month of the program. Consumers in the top, middle, and
bottom thirds were classified as heavy, moderate, and light + π i 2 Log( IPTijk )
buyers, respectively.
+ π i 3Log(Quarterij ) × Log( IPTijk )
•Interpurchase time (IPTijk): Number of days between the
prior purchase and the current purchase. + ϖij1Log( IPTijk )] + (ϖij0 + ς ijk ),
•Last month in the program (LastMonthim): A dummy variable
indicating the last month of transactions a consumer made.
This variable is set to 1 if (1) the previous transaction con- where Sizeijk is the dollar amount consumer i spent in the
ducted by consumer i occurred in month m and (2) month m kth transaction during the jth quarter in the program, IPTijk
was at least three months before the end of the analysis is the number of days since the previous transaction, Light-
period, offering a three-month lapse window before labeling Buyeri and HeavyBuyeri are dummy variables that indicate
a consumer as “dead.” If consumer i made any purchase after whether consumer i is a light buyer or a heavy buyer,
month m or if month m is within three months of the end of Log(Quarterij) is the logarithm transformation of the jth
the analysis period (to allow temporary inactivity), the Last-
quarter, πi0 + πi1Log(Quarterij) + πi2Log(IPTijk) +
Monthim variable is set to 0.
πi3Log(Quarterij) × Log(IPTijk) + ϖij1Log(IPTijk) indicate
•Reward claim rate (RCRateiq): The percentage of rewards
that consumer i claimed in the qth quarter; this equals the
the random effects of these variables that are not systemati-
ratio of the number of reward certificates requested to the cally accounted for by the fixed coefficients, and ϖij0 and
total number of rewards qualified for in that quarter. ςijk are error terms.

The Long-Term Impact of Loyalty Programs / 33

REFERENCES
Allaway, Arthur W., Richard M. Gooner, David Berkowitz, and sionary Goal Progress, and Customer Rentention,” Journal of
Lenita Davis (2006), “Deriving and Exploring Behavior Seg- Marketing Research, 43 (February), 39–58.
ments Within a Retail Loyalty Card Program,” European Jour- Kopalle, Praveen K. and Scott A. Neslin (2003), “The Economic
nal of Marketing, 40 (11–12), 1317–39. Viability of Frequent Reward Programs in a Strategic Competi-
Anderson, Eugene W., Claes Fornell, and Donald R. Lehmann tive Environment,” Review of Marketing Science, 1, 1–39.
(1994), “Customer Satisfaction, Market Share, and Profitabil- Krishnamurthi, Lakshman and S.P. Raj (1988), “A Model of Brand
ity: Findings from Sweden,” Journal of Marketing, 58 (July), Choice and Purchase Quantity Price Sensitivities,” Marketing
53–66. Science, 7 (1), 1–20.
Bitner, Mary Jo (1995), “Building Service Relationships: It’s All Lal, Rajiv and David E. Bell (2003), “The Impact of Frequent
About Promises,” Journal of the Academy of Marketing Sci- Shopper Programs in Grocery Retailing,” Quantitative Market-
ence, 23 (Fall), 246–51. ing and Economics, 1 (2), 179–202.
Boatwright, Peter, Sharad Borle, and Joseph B. Kadane (2003), “A Latham, Garry P. and Edwin A. Locke (1991), “Self-Regulation
Model of the Joint Distribution of Purchase Quantity and Tim- Through Goal Setting,” Organizational Behavior and Human
ing,” Journal of the American Statistical Association, 98 (Sep- Decision Processes, 50 (2), 212–47.
tember), 564–72. Leeflang, Peter S.H., Dick R. Wittink, Michel Wedel, and Philippe
Bolton, Ruth N., P.K. Kannan, and Matthew D. Bramlett (2000), A. Naert (2000), Building Models for Marketing Decisions.
“Implications of Loyalty Program Membership and Service Boston: Kluwer Academic.
Experiences for Customer Retention and Value,” Journal of the Leenheer, Jorna, Tammo H.A. Bijmolt, Harald J. van Heerde, and
Academy of Marketing Science, 28 (Winter), 95–108. Ale Smidts (2003), “Do Loyalty Programs Enhance Behavioral
Chanil, Debra (2004), “Profile of the C-Store Shopper,” Conve- Loyalty? A Market-Wide Analysis Accounting for Endogene-
nience Store News, (February 9), 1–7. ity,” working paper, Department of Marketing, Tilburg
Dowling, Grahame R. (2002), “Customer Relationship Manage- University.
ment: In B2C Markets, Often Less Is More,” California Man- Lemon, Katherine N., Tiffany Barnett White, and Russell S. Winer
agement Review, 44 (Spring), 87–104. (2002), “Dynamic Customer Relationship Management: Incor-
——— and Mark Uncles (1997), “Do Customer Loyalty Pro- porating Future Considerations into the Service Retention
grams Really Work?” Sloan Management Review, 38 (Sum- Decision,” Journal of Marketing, 66 (January), 1–14.
mer), 71–82. Lewis, Michael (2004), “The Influence of Loyalty Programs and
Drèze, Xavier and Stephen J. Hoch (1998), “Exploiting the Short-Term Promotions on Customer Retention,” Journal of
Installed Base Using Cross-Merchandising and Category Des- Marketing Research, 41 (August), 281–92.
tination Programs,” International Journal of Research in Mar- Mägi, Anne W. (2003), “Share of Wallet in Retailing: The Effects
keting, 15 (December), 459–71.
of Customer Satisfaction, Loyalty Cards, and Shopper Charac-
Eisenberger, Robert and Linda Rhoades (2001), “Incremental
teristics,” Journal of Retailing, 79 (2), 97–106.
Effects of Reward on Creativity,” Journal of Personality and
Meyer-Waarden, Lars and Christophe Benavent (2006), “The
Social Psychology, 81 (October), 728–41.
Impact of Loyalty Programmes on Repeat Purchase Behav-
General Electric Capital Franchise Finance Corporation (2001),
iour,” Journal of Marketing Management, 22 (February),
2001 Convenience Store and Petroleum Marketing Industry
61–88.
Review & Outlook. Scottsdale, AZ: General Electric Capital
Morgan, Robert M. and Shelby D. Hunt (1994), “The
Franchise Finance Corporation.
Gupta, Sunil (1988), “Impact of Sales Promotions on When, What, Commitment–Trust Theory of Relationship Marketing,” Jour-
and How Much to Buy,” Journal of Marketing Research, 25 nal of Marketing, 58 (July), 20–38.
(November), 342–55. Nako, Steven (1992), “Frequent Flyer Programs and Business
Gwinner, Kevin P., Dwayne D. Gremler, and Mary Jo Bitner Travellers: An Empirical Investigation,” Logistics and Trans-
(1998), “Relational Benefits in Services Industries: The Cus- portation Review, 28 (December), 395–414.
tomer’s Perspective,” Journal of the Academy of Marketing Sci- O’Brien, Louise and Charles Jones (1995), “Do Rewards Really
ence, 26 (Spring), 101–114. Create Loyalty?” Harvard Business Review, 73 (May–June),
Hsee, Christopher K., Fang Yu, Jiao Zhang, and Yan Zhang (2003), 75–82.
“Medium Maximization,” Journal of Consumer Research, 30 Oliver, Richard L. (1999), “Whence Consumer Loyalty?” Journal
(June), 1–14. of Marketing, 63 (Special Issue), 33–44.
Kamakura, Wagner A. and Michel Wedel (2001), “Exploratory Raudenbush, Stephen W. and Anthony S. Bryk (2002), Hierarchi-
Tobit Factor Analysis for Multivariate Censored Data,” Multi- cal Linear Models: Application and Data Analysis Methods.
variate Behavioral Research, 36 (1), 53–82. Thousand Oaks, CA: Sage Publications.
Kim, Byung-Do, Mengze Shi, and Kannan Srinivasan (2001), Reinartz, Werner J. and V. Kumar (2000), “On the Profitability of
“Reward Programs and Tacit Collusion,” Marketing Science, Long-Life Customers in a Noncontractual Setting: An Empiri-
20 (Spring), 99–120. cal Investigation and Implications for Marketing,” Journal of
Kivetz, Ran (2005), “Promotion Reactance: The Role of Effort- Marketing, 64 (October), 17–35.
Reward Congruity,” Journal of Consumer Research, 31 Schacter, Daniel L. (1999), “The Seven Sins of Memory: Insights
(March), 725–36. from Psychology and Cognitive Neuroscience,” American Psy-
——— and Itamar Simonson (2002), “Earning the Right to chologist, 54 (March), 182–203.
Indulge: Effort as a Determinant of Customer Preferences Schmittlein, David C., Donald G. Morrison, and Richard Colombo
Toward Frequency Program Rewards,” Journal of Marketing (1987), “Counting Your Customers: Who Are They and What
Research, 39 (May), 155–70. Will They Do Next?” Management Science, 33 (January),
——— and ——— (2003), “The Idiosyncratic Fit Heuristic: 1–24.
Effort Advantage as a Determinant of Consumer Response to Sharp, Byron and Anne Sharp (1997), “Loyalty Programs and
Loyalty Programs,” Journal of Marketing Research, 40 Their Impact on Repeat-Purchase Loyalty Patterns,” Inter-
(November), 454–67. national Journal of Research in Marketing, 14 (5), 473–86.
———, Oleg Urminsky, and Yuhuang Zheng (2006), “The Goal- Sheth, Jagdish N. and Atul Parvatiyar (1995), “Relationship Mar-
Gradient Hypothesis Resurrected: Purchase Acceleration, Illu- keting in Consumer Markets: Antecedents and Consequences,”

34 / Journal of Marketing, October 2007

 Journal of the Academy of Marketing Science, 23 (Fall), Venkatesan, Rajkumar and V. Kumar (2003), “Using Customer
255–71. Lifetime Value in Customer Selection and Resource Alloca-
Shugan, Steven M. (2005), “Brand Loyalty Program: Are They tion,” Working Paper No. 03-112, Marketing Science Institute
Shams?” Marketing Science, 24 (Spring), 185–93. Paper Series, Cambridge, MA.
Sirdeshmukh, Deepak, Jagdip Singh, and Barry Sabol (2002), Verhoef, Peter C. (2003), “Understanding the Effect of Customer
“Consumer Trust, Value, and Loyalty in Relational Relationship Management Efforts on Customer Retention and
Exchanges,” Journal of Marketing, 66 (January), 15–37. Customer Share Development,” Journal of Marketing, 67
Skinner, B.F. (1953), Science and Human Behavior. New York: (October), 30–45.
The Free Press. Vroom, V. (1964), Work and Motivation. New York: John Wiley &
Strohmetz, David B., Bruce Rind, Reed Fisher, and Michael Lynn Sons.
(2002), “Sweetening the Till: The Use of Candy to Increase Woodruff, Robert B. (1997), “Customer Value: The Next Source
Restaurant Tipping,” Journal of Applied Social Psychology, 32 for Competitive Advantage,” Journal of the Academy of Mar-
(February), 300–309. keting Science, 25 (Spring), 139–53.
Sun, Baohong (2005), “Promotion Effect on Endogenous Con- Yi, Youjae and Hoseong Jeon (2003), “Effects of Loyalty Pro-
sumption,” Marketing Science, 24 (3), 430–43. grams on Value Perception, Program Loyalty, and Brand Loy-
Taylor, Gail Ayala and Scott A. Neslin (2005), “The Current and alty,” Journal of the Academy of Marketing Science, 31 (Sum-
Future Sales Impact of a Retail Frequency Reward Program,” mer), 229–40.
Journal of Retailing, 81 (4), 293–305. Zhang, Z. John, Aradhna Krishna, and Sanjay K. Dhar (2000),
Thaler, Richard (1985), “Mental Accounting and Consumer “The Optimal Choice of Promotional Vehicles: Front-Loaded
Choice,” Marketing Science, 4 (Summer), 199–214. or Rear-Loaded Incentives?” Management Science, 46
Van Osselaer, Stijn M.J., Joseph W. Alba, and Puneet Manchanda (March), 348–62.
(2004), “Irrelevant Information and Mediated Intertemporal
Choice,” Journal of Consumer Psychology, 14 (3), 257–70.

The Long-Term Impact of Loyalty Programs / 35