Profiling Compulsive Buyers at Risk:
Demographics, Behavioral Buying
Characteristics and Buyer Satisfaction
Abstract
The current study adopts a consumer welfare perspective with the purpose of profiling and
identifying compulsive buyers before their behavior spins out of control. While building forward
on existing literature on demographical variables, buying behavior characteristics and buyer
satisfaction measures are introduced as potential new identifiers for compulsive buyers.
Demographical variables illustrate limited usefulness while both behavioral buying characteristics
and buyer satisfaction show potential. Compulsive buyers have a high frequency of buying
episodes, while the time and amount of money spend per episode seems irrelevant. Furthermore,
compulsive buyers have lower levels of satisfaction towards the good/service consumed and its
provider.
Lars de Goede
Introduction
“It’s almost like you’re on a drunk. You’re so intoxicated;… I got this great high. It was like you couldn’t have given me more of a rush”.
A response from one of O’Guinn and Faber’s (1989, p 153) interviewees on the seemingly innocent
topic of compulsive buying. However, one with a striking resemblance to the topic of substance
abuse, where the term innocence is be far from applicable. Some more of the experiences and
emotions from O’Guinn and Faber (1989, p. 154) interviewees’ illustrate the extent to which
compulsive buying manifests itself in consumers’ minds.
“I couldn’t tell you what I bought or where I bought it. It was like I was on automatic”. “I never bought one of anything, I always buy at least two. I still do. I can never go even
to the Jewel and buy one quart of milk. I’ve always got to buy two.”
The bottom line of these quotes is that compulsive consumer behaviors are more extreme forms of
regular consumer behavior, driven by more extreme and emotional motivations (Hirschman,
1992). They also illustrate that compulsivity in decision making goes way beyond the generally
assumed rational acting consumer, the consumer making decisions based on costs and benefits.
The question is when normal motivations and afterwards experienced positive feelings and
gratification become so intense that they can be labelled abnormal (Faber, O’Guinn & Krych,
1987). The following and final two quote’s shine a light into the topic of the potential practical
implications of compulsive buying (O’Guinn and Faber, 1989 p. 155).
“I didn’t have one person in the world I could talk to. I don’t drink. I don’t smoke. I don’t do dope. But I can’t stop. I can’t control it. I said I can’t go on like this…. My husband
hates me. My kids hate me. I’ve destroyed everything. I was ashamed and I just wanted to die.”
Although the first of these two quotes also has a laughable aspect, the second one assures the
jocular part of the story wears in a scenario where consumers start losing control and the severe
potential implications on an individual’s life become apparent. Looking at the opening quote in
hindsight, for consumers suffering from heavy compulsiveness their buying behavior indeed has
an abuse potential similar to substance addiction (Faber, O’Guinn & Krych, 1987). Koran et. al.
(2006, p. 1806) provides a brief summarizing list of possible consequences of compulsive buying: “guilt or remorse, excessive debt, bankruptcy, family conflict, divorce, illegal activities, such as
writing bad checks and embezzlement, and even suicide attempts”.
Ever since Faber and O’Guinn started to quantify the magnitude of compulsive consumption in
1989, considerable parts of studied populations are found to be represented by people that consume
compulsively in the way described before. At the time, they found the incidence of compulsive
buying to be 5.9%. Over twenty years of research afterwards lead to both more and less
conservative conclusions about the incidence, or prevalence as others name it. Summarizing the
results creates a range of a compulsive buying incidence between 1.8% and 16% (Dittmar, 2005; Faber & O’Guinn, 1992; Hassay & Smith, 1996; Koran et. al., 2006; Magee, 1994; Mueller et al.,
2010). The most moderate estimate translates into over three hundred thousand compulsive buyers
in The Netherlands and around 6 million in the United States. “Clearly, even the most restrictive
behavior” (Dittmar, 2005 p. 469). Thus, next to its severe consequences, compulsive buying is also
widespread, making the phenomenon far from innocent.
The paper by Kraepelin (1915) that is often seen as the first contribution to academic research on
the topic of irrational overconsumption dates from over 100 years ago, indicating the presence of
the topic of on scholars’ agendas for a considerable period of time. During the late 80’s begin 90’s,
when the stream of literature started to increase, scholars realized the implications of a considerable amount of compulsive buyers amongst us: “it is in the interest of both people who
suffer from compulsive spending and society in general to try to reduce the incidence of this problem” (Faber, O’Guinn & Krych, 1987 p. 135). However, a “chief unaddressed question” to
achieve this reduction, asked by Faber and O’guinn (1992 p. 460) nearly 25 years ago; “how one
can best identify those at risk for compulsive buying”, still has not been answered since literature
on this topic is limited to a range of papers looking into a restricted selection of aspects on which
to profile and identify compulsive buyers (e.g.: a selection of demographics). The bulk of literature
focusses on psychological aspects in order to unravel the black box of compulsive buying.
Although that aspect is important in creating a clear and complete image of what compulsive
buying entails and what goes on in compulsive buyers’ minds, psychological processes are not
readily observable and are often hard to grasp. They thereby do not practically assist in the
identification of those buyers that are at risk.
To address this unanswered question, this paper adopts a consumer welfare perspective, moving
the focus from the psychological to the identification part of compulsive buying research. Thereby
the timely identification of risk category buyers through observable aspects is of focal attention
here. “The dimensions which seem most important to include in building these typologies are
what types of things are bought, when they are bought and for whom they are purchased” (Faber,
O’Guinn & Krych, 1987 p. 134). So far no groundbreaking additions have been made to this list.
Therefore, the earlier mentioned demographics will be exploited further, but moreover, two novel
dimensions for profiling and identifying compulsive buyers are introduced. Multiple behavioral
buying characteristics as well as several buyer satisfaction measures will serve as new potential
anchors for the profiling and identification of those compulsive buyers at risk. Most importantly,
the exterior nature of these three dimensions of profiling provides the opportunity of an
identification process which is possible during buying episodes and inside buying environments,
something vital for consumer welfare in the topic of compulsive buying. Their potential thus
allows for timely identification and assistance of those buyers at risk before debt becomes too high,
social relationships become damaged beyond repair and physical or any other harm is done. Taking
this a step further, successful identification eventually allows for compulsive buyer assistance in
managing their behavior. Even more importantly, it provides clues for clear managerial
implications towards the creation of less provoking marketing designs (Prentice & Woodside,
2013) aimed at the prevention of compulsive buying escalation.
Lastly, in contrast to virtually all research concerning the topic of compulsive buying, the data
gathered for this research is gathered inside a retail environment. Seven of the largest casinos in
Macau, by far the largest gaming destination in the world, participated in the collection of data
from respondents in an actual consumption setting. In comparison, selft-reported mail surveys,
Literature
Discussing compulsivity in the marketplace in the light of consumer welfare implies discussing a
number of subtopics. Next to defining the phenomenon itself, its (partly) malignant nature, its
potential dangerous consequences, and its prevalence, the dimensions on which profiling
consumers at risk is done will be discussed. These topics will be discussed in order of appearance
in the following section.
Rational Consumer vs. Compulsive Consumer: the Reasonability of the Assumption of Rationality
As touched upon in the introduction, “Over the last 50 years, the theory of rational choice has
emerged as the dominant paradigm of quantitative research in the social and economic sciences.
The idea that individuals make choices by rationally weighing values and uncertainties (or that
they ought to, or at least act as if they do) is central to Bayesian methods of statistical inference
and decision analysis; the theory of games of strategy; theories of competitive markets, industrial
organization, and asset pricing; the theory of social choice; and a host of rational actor models in political science, sociology, law, philosophy, and management science” (Nau, 1999 p. 1). Rational
choice models are derived from utilitarian theory (Cook & Levi, 1990) and therefore assume the
weighing of costs and benefits of available alternatives (Hoch & Loewenstein, 1991). Consumers
evaluate and process the attributes of these alternatives, make trade-offs on their importance and
finally choose an alternative based on self-developed decision rules (Bettman, 1979).
Solely leaning on the philosophy of rationality neglects another important aspect in the human
decision-making process. As Hall (2002 p. 23-24) notes, “the weakness of these approaches is in
Or, as Poels and Dewitte (2006 p. 3) frame it, the study on emotions “has led to the general
conception that emotions are not a useless by-product but are essential for rational thinking and
behavior”. Hoch and Loewenstein (1991 p. 492) justly recognize that “a more complete
understanding of consumer behavior must recognize that people are influenced by both long-term
rational concerns and more short-term emotional factors.”
The above mentioned authors try to emphasize the existence of an interplay between rational and
emotional motives in consumer decision making, an interplay in which emotions possibly become
powerful enough to override cognitional reasoning in the decision making process (Zajonc, 1980).
Especially those emotional factors play an important role in the decision making process in the
mind of the compulsive consumer, illustrated by the irrationality of their decisions.
Via Emotions to Compulsivity
Defining compulsive buying is leaned on relatively recent work by Ridgway, Kukar-Kinney and
Monroe (2008) supported by a stream of roughly 15 years of research by Hollander and colleagues
(1993; 1993; 1995; 2005; 2006) and McElroy, Phillips and Keck (2004) on the
obsessive-compulsive spectrum. This stream of research incorporates aspects of both obsessive-obsessive-compulsive
(OCD) and impulse-control disorders (ICD) into the concept of compulsive buying. It is therefore formulated as “a consumers tendency to be preoccupied with buying that is revealed through
repetitive buying and lack of impulse control over buying” (Ridgway, Kukar-Kinney & Monroe,
2008, p. 622).
Consumers with compulsive inclinations feel emotionally compelled to act in a certain manner
(Robbins et. al. 2012). They experience obsessive thoughts caused by anxiety or distress
of the associated tension through short term pleasurable behavior (Black, Shaw & Blum, 2010)
that is sometimes experienced as a “high”, a term familiarized in the field of narcotics but also
found in testimonials on related phenomena like compulsive shopping (Dittmar & Drury, 2000).
Taking the concept of experiencing a “high” a step further, a share of respondents of 11% in
Schlosser et. al. (1994) experience shopping as sexually stimulating. Faber, O’Guinn and Krych,
(1987) and Hirschman (1992) both mention the effectiveness of these experienced positive
emotions as a self-treatment for unhappiness.
While compulsive behaviors are seemingly purposeful in providing relief, they are in fact highly
ego-dystonic (American Psychiatric Association, 1985 p. 234, 1994, 2000) since consumers
suffering from compulsions logically realize their behaviors or rituals are harmful but they still
feel emotionally compelled to pursue them (Robbins et. al. 2012). The numbers from Schlosser et.
al. (1994) underpin this, 72% of their respondents are concerned with their frequency of buying,
85% with the debt related to their compulsive buying behavior and 74% with the fact that they
experience a loss of control while shopping. Another point illustrating the awareness of the
negative effects of compulsive buying is the higher interest in store return policies and the higher
frequency and volume of returned products under compulsive buyers compared to non-compulsive
buyers as found by Hassay and Smith (1996).
Eventually compulsive consumers find themselves in a situation of inner conflicts between the
presence of a drive, an impulse or an urge, which create a (sudden) desire to act in a certain way
on the one side, and the willpower to override this desire on the other side. Typically, the
compulsive consumer falls short in employing self-control and starts the process of simultaneous
augmentation of the short term benefits and downplaying or denying the long term (harmful)
& Blum, 2010; Faber, O’Guinn & Krych, 1987; O’Guinn & Faber 1989; Hoch & Loewenstein 91;
Hirschman, 1992; Nataraajan & Goff, 1991). Often it is only the afterwards confrontation with the
harmful consequences of made compulsive decisions that feeds motivation to (temporary) stop
behaving compulsively (Black, Shaw & Blum, 2010).
Once behaving compulsively, it is perhaps the consciousness within the compulsive
decision-making process that ensures that consumers attempt to set up clear boundaries to limit their own
compulsive behavior (Hoch & Loewenstein, 1991), creating a process in which decisions are made ritualistically “performed according to certain rules or in a stereotyped fashion” (American
Psychiatric Association, 1985 p. 234, 1994, 2000). Experiencing the negative downside of
compulsivity however adds emotional distress to the already existing tension eventually causing
rules and limits to be broken and compulsive behavior to become repetitive and uncontrollable (Hirschman, 1992). “Thus, it appears that it is extremely difficult for compulsive consumers to
become noncompulsive” (Hirschman, 1992 p. 166). When eventually the outcome of the process
becomes chronically compulsive and consumers develop a physically or psychologically
dependency on the substance or activity relevant in their case, the compulsive character of the
individuals’ pattern of consumption fully manifests itself (Faber, O’Guinn & Krych, 1987;
Hirschman, 1992; O’Guinn & Faber, 1989).
Summing up, the difference between the normal and the compulsive consumer is far more than an
alarmingly higher frequency of consumption observed on the surface, but rather lies in divergent,
tension related motivations that go beyond the desire for material acquisition (O’Guinn & Faber,
Compulsive Buying, the Buying Part
This paper uses a wide perspective in discussing compulsive buying, a phenomenon occurring in
the goods as well as the services industry, and which can be triggered in, or by all stages of the
process that lead to and follow a transaction. In other words, the pre-buying phase: anticipation,
preparation and shopping (Black, 2007), the eventual spending/buying phase, and/or the
post-buying phase: owning, using or experiencing (O’Guinn & Faber, 1989) all potentially play a role
in triggering compulsivity.
How Bad Can it Get?
In concrete terms, chronically compulsive behavior leads to excessive acquisition and/or use of
goods or services. Depending on the nature of the good or service consumed, considerable negative
economic, psychological and societal effects (O’guinn & Faber, 1989) threaten and interrupt the
day to day life of the consumer (Faber, O’Guinn & Krych, 1987; Nataraajan & Goff, 1991). The
snowball effect starts when consumers spend beyond their disposable income and debt starts to
accumulate. Practices like overextending credit lines, an increasing number of credit cards used
and forced sales of property become necessary to free up financial space (Faber, O’Guinn & Krych, 1987; O’guinn & Faber, 1989). Simultaneously, feelings of guilt, remorse, anxiety, lower
self-esteem and the fear of justification by others increase psychological pressure (Faber, O’Guinn &
Krych, 1987; Hirschman, 1992; O’guinn & Faber, 1989). In a more extreme manner this leads to
the hiding of compulsive purchases or activities and criminal undertakings like theft and
embezzlement to be able to support compulsive behavior (Hirschman, 1992). The harmful
consequences extend beyond the consumers’ self when his or hers dysfunctioning economically
and socially puts pressure on an individual’s family, friends and occupational environment too, as
care system and crime control (Krych, 1989). In the worst case, compulsive behavior evolves into
situations described by either suicidal intentions due to feelings of loneliness and disintegration as
a consequence of social isolation, or heavy physical consequences of substance abuse, both
possibly causing death (Hirschman, 1992).
How Common is Compulsive Buying?
To illustrate the extent to which the phenomenon of compulsive buying spreads among society, a
brief chronological summary on findings about its prevalence is presented. Faber and O’Guinn
(1989; 1992) found the incidence of compulsive buying to be 5.9% and later between 1.8% and
8.1%. Magee (1994), using the same classification, came to the conclusion of a fraction of 15% to
16% of respondents that she labeled as compulsive consumers. Hassay and Smith (1996)
determined the incidence to be 12.2%. Dittmar (2005) draws a similar conclusion with a
percentage of 13.4 of individuals that can be labeled compulsive buyers. Koran and colleagues
(2006) were the first to employ a large sample of 2513 respondents. It lead them to the conclusion
of a prevalence of compulsive buying of 5.8%, more in line with the first insights provided by Faber and O’Guinn (1989; 1992). Most recently, Mueller et. al. (2010) joined this range of
incidences with their point prevalence of 6.9% within a German population.
Note that the epidemiological terms incidence and prevalence are used intertwined by the authors
referred to here, both simply denoting to the fraction of carriers within the population at the given
moment of measurement.
Profiling and Identifying the Compulsive Buyer
Since the theoretical purpose of this paper lies in profiling and identifying compulsive buyers in
demographics, behavioral buying characteristics and buyer satisfaction measures. The specific
aspects under each of these dimensions, as well as their hypothesized effect will be discussed in
the following section. With a consumer welfare point of view, the profiling aspects have been
chosen so that they share an exterior nature. They are publically observable to a large extend,
which makes identification possible in a stage early enough to ensure potential harmful damage to
the compulsive buyer in question can be prevented.
Demographics
Age
O’guinn and Faber (1989) are the first that quantitatively investigated the relationship between age
and compulsive buying. In comparing their general sample against their abnormal compulsive
group, age was found to be significantly lower in the latter, indicating a negative relationship. In
follow up research in the years after similar results are found, e.g., negative correlation between age and compulsive buying (d’Astous, 1990; d’Astous, Maltais & Roberge, 1990; Magee, 1994).
However, Sherhorn, Reisch and Raab (1990), failing to find a significant age difference in
comparing their compulsive with their normal sample, and Roberts (1998), producing a lack of
correlation between age and compulsive buying, couldn’t confirm these results within their
samples. A more recent stream of research joins the earliest and confirms a negative relationship
between age and compulsive buying, either by correlational analysis or comparing means (Dittmar,
2005; Koran et. al., 2006; Mueller et. al., 2010; Ridgway, Kukar-Kinney & Monroe, 2008).
Summarizing, the first hypothesis is proposed as follows:
Gender
One of the most vivid stereotypes about compulsive buying is that it’s a female matter in particular (Faber, O’Guinn & Krych, 1987). Except for a couple of authors concluding an insignificant
difference between men and women (Koran et. al., 2006; Magee, 1994; Mueller et. al., 2010), most
literature tends to support the basis for this stereotype in proving the higher compulsive buying
tendencies of females in contrast to males either by comparing groups or applying correlational
analysis (d’Astous, 1990; d’Astous, Maltais & Roberge, 1990; Christenson et. al., 1994; Dittmar, 2005; Faber & O’Guinn, 1992; O’Guinn & Faber, 1989; Ridgway, Kukar-Kinney & Monroe,
2008; Roberts, 1998; Scherhorn, Reisch & Raab, 1990).
H2: Females have higher compulsive buying tendencies than males.
An important side note considering gender is the fact that the data for analysis used in this paper
finds its origin in a casino environment. Within the gambling literature results in relation to gender
generally are the opposite as proposed above as it is said to be a male thing, although this
supposedly is a diminishing trend (Black, Shaw & Blum, 2010). Thus, this could influence results
towards the opposite than hypothesized direction.
Education and Income
Literature concerning these two demographics is relatively less abundant as for the preceding two
within the field of compulsive buying behavior. Generally, higher education ensures better job
opportunities, which in turn should go accompanied by a higher income. Thus, multicollinearity
in connecting them to the compulsive buying spectrum could arise. Different theories exist
however, in which they do not necessarily affect compulsive buying in the same direction. Higher
compulsive buying. Income, however, could positively drive compulsive buying as higher income
provides room to spend more without experiencing the negative downsides of it.
When it comes to education two contradictory results show up in existing (relatively recent)
scientific literature. Dittmar (2005) found an absence of a significant relationship between
education and compulsive buying. Ridgway, Kukar-Kinney and Monroe (2008), on the other hand,
confirm a negative relationship as sketched before. Following the latter, hypothesis three is
formulated as follows:
H3: Education drives compulsive buying negatively.
In contrast to education, income gets considerably more attention. Results are pointing in different
directions. Ironically, most research in the past decades agrees on the absence of a relationship
between income and compulsive buying (Christenson et. al., 1994; Dittmar, 2005; O’Guinn &
Faber, 1989; Ridgway, Kukar-Kinney & Monroe, 2008; Roberts, 1998; Scherhorn, Reisch & Raab,
1990). Contradictory results have been found however. Black et. al. (2001) note that greater
severity of compulsive buying is associated with lower income. Koran et. al. (2006), in the same
philosophy, claim that compulsive buyers have lower income compared to other respondents. In contrast to all former mentioned results, d’Astous (1990) found an inverted U-shape relationship
between income and compulsive buying, pointing at middle incomes to be driving compulsivity.
In formulating hypothesis four, the base is formed by the more recent results by Black et. al. (2001)
and Koran et. al. (2006) that indicate:
Behavioral Buying Characteristics
The idea behind the concept of behavioral buying characteristics and the connection to compulsive
buying lies in the higher buying frequency/severity observed for compulsive buyers. They are
argued to drive compulsive buying in the sense that for consumers that have experienced that their
buying behavior provides them a high (Dittmar and Drury, 2000) or emotional lift (O’guinn and
Faber, 1989, Christenson et. al., 1994), the fact that they more easily become loyalty/reward
program members or find themselves in buying situations more frequently, for longer periods and
spend an increasing amount of money, indicates that they are deteriorating into a compulsive
buying habit. The behavioral buying characteristics therefore serve as signals that consumers are
(becoming) compulsive buyers.
For the first aspect under this dimension, loyalty/reward program membership, there is a much
discussed endogeneity problem potentially leading to overestimation of its effect (Bolton, Lemon
& Verhoef, 2004). E.g.: compulsive buyers could be theorized to become loyalty/reward program
members more often or quicker than normal buyers because they are more interested in the benefits
the program offers. However, being a loyalty/reward program member, as proposed here, could
also drive compulsive buying in the sense that the benefits the program offers triggers buying
behavior. Theorizing similarly, the endogeneity problem could also hold for the other three aspects
under the topic of behavioral buying characteristics; duration, frequency and expenditure height.
Where being a compulsive buyer could drive buying more often, spending more money and shop
for longer periods, these three aspects could also cause consumers to lapse into a compulsive
buying habit. The existence of the endogeneity issue and the ongoing discussion is noted. Here it
Therefore, all four behavioral buying characteristics are treated as antecedents and modeled as
such.
Loyalty/Reward Program Membership
Loyalty as a concept has been virtually unanimously approached as a means aimed at generating
more profit from consumers. From the standpoint of consumer welfare, loyalty programs are
absent in literature to date. Theorizing therefore is inferred from other branches of research.
Loyalty or rewards programs are aimed at increasing the perceived value of the relationship for
the consumer though the provision of free benefits by the product or service provider (Bolton,
Lemon & Verhoef, 2004), often as a reward for their repeated purchases in the past and present
(Meyer-Waarden, 2008). It is this surplus that should ensure repeated visitations and eventual
loyalty. Indeed, authors find support for this theory. Positive connections have been found between
loyalty programs and consumers share of wallet/expenditures (Leenheer et. al., 2007;
Meyer-Waarden, 2007), between loyalty programs and customer retention/lifetime (Meyer-Meyer-Waarden,
2007; Verhoef, 2003), and between loyalty programs and purchase frequencies, basket value and
lower inter-purchase time (Meyer-Waarden, 2008). This teaches us that loyalty or reward program
membership could drive consumers to become heavier, potentially compulsive, buyers.
H5: Loyalty/reward program membership positively drives compulsive buying
Buying Severity: Duration, Frequency and Expenditure Height
The three concepts between brackets, although separately measured, are closely linked together
and therefore theoretically captured together under the buying severity flag. They describe the
amount of hours spend during a buying episode, the yearly amount of buying episodes and the
Defining compulsive buying before touched upon the more frequent or severe shape of
consumption of compulsive buyers compared to a more normal consumption pattern, when
profiling compulsive buyers this is a potential important factor in play. A key takeaway from
definitions found in literature is the repetitive nature of the consumers buying pattern (Hirschman,
1992; O’guinn & Faber, 1989; Ridgway, Kukar-Kinney & Monroe, 2008). Considering the level of expenditures, the repetitiveness aligns with findings by Faber, O’Guinn and Krych, (1987) and
O’guinn and Faber (1989) indicating overextended credit lines, a higher amount of credit cards
used and forced sales of property. Koran et. al., 2006 indeed indicated a positive relationship
between duration (the number of hours spend during a buying episode) and compulsive buying.
Supplementary, Ridgway, Kukar-Kinney and Monroe (2008) found a significant relationship
between frequency of buying (the yearly amount of buying episodes) and expenditure height
(amount of money spend per episode) on the one side and compulsive buying on the other.
Logically, the following hypotheses concerning buying severity are therefore proposed.
H6: Buying duration per episode positively drives compulsive buying
H7: Buying episode frequency positively drives compulsive buying
H8: Expenditure height per episode positively drives compulsive buying
Buyer Satisfaction
A third dimension of aspects on which profiling is based is the level of satisfaction with the
contextual service or good bought and the corresponding provider. The concept of buyer
satisfaction is divided into four aspects; overall service/product quality satisfaction, feelings of
service/product provider loyalty, propensity of positive word-of-mouth and the propensity to
In case of a normal consumer, satisfaction is driven by the price and general quality of the good or
service bought (Anderson, Fornell & Lehmann, 2004). High perceived quality and a good price
lead to higher satisfaction, which in turn is proposed to lead to a higher amount of consumption as
buyers consciously realize spending their money provides them with a satisfactory level of
benefits. The status quo describes a satisfied frequently visiting consumer spending a fair amount
of money.
For compulsive buyers however, the benefits of consumption have become outweighed by its
severe negative consequences mentioned earlier. As consumers either consciously or
unconsciously experience the downside of their compulsive buying behavior on their daily life,
their attitude towards the service or good compulsively consumed is proposed to become worse
and the satisfaction with the one providing it to them therefore is proposed to drop, as the
degeneration of their quality of life is attributed to both. Although their compulsivity drives them
to maintain a high level of consumption, their level of satisfaction is thus proposed to drop. The
status quo here describes a dissatisfied frequently visiting consumer spending a fair amount of
money.
Theorizing in this sense implies approaching buying satisfaction as an outcome of compulsive
buying. The corresponding hypotheses are therefore formulated as follows:
H9: compulsive buying drives overall service/product quality negatively
H10: compulsive buying drive feelings of service/product provider loyalty negatively
H11: compulsive buying drives the propensity of positive word-of-mouth negatively
See figure 1 for an overview of all variables used for profiling in the proposed hypotheses.
Figure 1: Hypothesized Profiling Antecedents and Outcomes
Methodology
This section sheds light into the methodological part. All topics concerning the data and its analysis
will be discussed here. First of all, the measurement, scaling, sampling survey and data collection
will be discussed, after which the method of analysis concludes this section. The data analyzed is
acquired through Professor Arch G. Woodside (Boston College) used in his coproduction with
Catherine Prentice (Swinburne University); Problem Gamblers’ Harsh Gaze on Casino Services
(2013). It finds its origin in the popular gaming destination of Macau. Although still underestimated by many, this Chinese gambling Valhalla established itself as the “heavyweight
At first glance there might seem to be a gap between the topic of compulsive buying and the
gambling related data used to examine it. However, as touched upon in the introduction, both
topics are share a great amount of commonality. As illustrated in figure 2, gambling and
compulsive buying, among other similar disorders, both fall on the obsessive-compulsive spectrum
(Hollander, 1999). The spectrum consists of a family of multiple disorders sharing symptoms,
cause, effect and other commonalities. The spectrum thus represents the similarities between the
compulsive and impulsive behaviors that belong to it.
Obsessive- Compulsive Disorder Impulse- Control Disorder
OCD AN Trich IIU PG
Binge Eating Klep
Compulsive Buying
Figure 2: Obsessive-Compulsive Spectrum Disorders
Note: Adapted from Hollander (1999, p.40). The figure is not scaled, however, the position on the scale indicates closer alignment with either OCD or ICD. OCD: obsessive-compulsive disorder; AN: anorexia; Trich: trichotillomania; Klep: kleptomania; IIU: impulsive internet usage; PG: pathological gambling.
Furthermore, individuals exhibiting serial and simultaneous compulsive behaviors are far from an
exceptions to the rule; “… drug addiction was part of a larger pattern of compulsive behavior that
was manifested in multiple ways in the consumer's life. For example, I did not just take stimulants
consistently. I also drank alcohol almost daily, exercised ritualistically, worked in binges, ate
certain foods repeatedly, and would go on shopping "splurges" during which I would buy large quantities of one item” (Hirschman, 1992 p. 163). An overview on numerical findings on
the perspective that compulsive buying condition extends beyond retail shopping for products” and adopting the idea that “problem gambling is regarded as a subcategory of compulsive buying”
(Prentice & Woodside, 2013 p. 1110), advocates the suitability of the use of gambling data as a
case-setting within the broader framework of compulsive buying, a setting not limited to any
product or service.
Measurement and Scaling
This section describes the measurement and scaling of compulsive buying itself and the aspects
under each of the three dimension linked to compulsive buying (demographics, behavioral buying
characteristics and buyer satisfaction measures), which will be discussed in order of appearance.
Compulsive Buying
When reviewing the available definitions in literature, true compulsive behavior is described by
not in the least vague terms like: abnormal (Edwards, 1993; Nataraajan & Goff, 1991), beyond the
bounds of normalcy (Hirschman, 1992), quantitatively and qualitatively different from societal
norms, inappropriate, excessive and disruptive (Faber, O’Guinn & Krych, 1987), placing any compulsive consumer far outside the “normal” world. However, the question remains when the
point is reached at which one can be labeled as an abnormal and inappropriate compulsive
consumer. As with many conditions, severity differs highly across carriers, in this case from
occasionally to excessively compulsive. Also, the boundary between normal and abnormal
behavior is culture as well as context specific (Hassay & Smith, 1996). Therefore, instead of using
a dichotomous philosophy, compulsivity is approached as a continuum or spectrum (Hassay &
Smith, 1996; Nataraajan & Goff, 1991; Prentice & Woodside, 2013). On the one side of the
spectrum lies normal (occasionally compulsive) behavior, on the other, abnormal (excessive
Table 1: PGSI Items
I have… …wagered larger amounts to get the same feeling of excitement.
…tried to win back losses.
…borrowed money or sold something to get money for gambling. …felt a gambling problem existed.
…been criticized for betting or told a gambling problem exists. …felt guilty about gambling
…bet more than could be lost.
Has… …gambling caused health problems including stress and anxiety.
…gambling caused financial problems.
The gambling specific measure used as a proxy for compulsive buying here is the Problem
Gambling Severity Index (PGSI), also known as the Canadian Problem Gambling Index (Ferris &
Wynne, 2001). The construct is build up out of nine items (see table 1) on which a 4-point scale answer is requested: “1 = never, 2 = sometimes, 3 = most of the time and 4 = almost always”,
resulting in a summed PGSI rating ranging from 9 (normal) to 36 (abnormal).
Table 2: Item-to-Item/Item-to-Total Correlations for PGSI Scale
1 1 1 2 2 0,556** 1 3 3 0,263** 0,734** 1 4 4 0,424** 0,307** 0,280** 1 5 5 0,502** 0,595** 0,495** 0,472** 1 6 6 0,539** 0,605** 0,588** 0,432** 0,595** 1 7 7 0,429** 0,578** 0,627** 0,443** 0,555** 0,671** 1 8 8 0,493** 0,407** 0,342** 0,659** 0,565** 0,539** 0,538** 1 9 9 0,398** 0,560** 0,584** 0,387** 0,537** 0,566** 0,589** 0,358** 1 Total Total 0,650** 0,830** 0,790** 0,592** 0,780** 0,822** 0,815** 0,676** 0,744** 1 ** P < 0,001
The Cronbach’s Alpha for construct reliability on the gathered data for the nine items is 0.894,
meaning good reliability, nearly excellent. All item-to-item and item-to-total correlations are
Demographics
The four demographic variables are all scaled categorical. Gender is measured in two, both age
and education in three, and income in four categories. For an overview of the specific categories
per variable and the affix for the various variable category dummies see table 3. For age, education
and income categories are divided; low, middle and high, except for income which has a fourth
category that allows for very high income levels. Junior high/high school is regarded as the
breakpoint between low and middle education because of the intermediate nature of junior high
and the relatively higher equality with elementary school versus high school.
Table 3: Demographic Variable Scales
Scale Age Gender Education Income
0 18 to 35 Female Junior high or below $0 to $6000
1 36 to 55 Male High school $6001 to $15000
2 56+ Diploma $15001 to $25000
3 $25001 +
Behavioral Buying Characteristics
Of the four behavioral buying characteristic measures, two are categorical while the other two are
continuous scales. Loyalty/reward program membership is a yes/no question and both duration
(hours) and frequency (number of buying episodes) are open questions and continuous scaled.
Finally, expenditure height is measured in four categories. For the specifics see table 4 below.
Table 4: Demographic Variable Scales
Scale Loyalty
membership Duration Frequency Expenditure height
Buyer Satisfaction
To measure the four aspects of buyer satisfaction (overall service/product quality satisfaction,
feelings of service/product provider loyalty, propensity of positive word-of-mouth and the
propensity to switch to another service/product provider), seven items were used (see table 5), each
to be answered on a 7-point scale (from low/disagree to high/agree). Service/product provider
loyalty was measured as a single item construct, the other three are were measured on two items.
Table 5: Buyer Satisfaction Items
1. I think the overall service quality of the casino is…
2. Overall, how satisfied are you with the service provided by the casino? 3. I always consider this casino as my first choice to visit.
4. I say positive things about the casino to other people.
5. I recommend the casino to other players who haven’t been here. 6. I will switch to another casino that offers better services and deals.
7. I will switch to another casino if I experience a problem with the casino service.
Results for a factor analysis (n = 411) over all seven items and six out of seven (without item 3)
are provided in table 6 and 7. The corresponding Kaiser-Meyer-Olkin measures for sampling
adequacy are 0.679 and 0,604, combined with the result of Bartlett’s test of sphericity, which is
highly significant in both cases (P < 0,001), this indicates appropriateness of the use of factor
analysis. Theoretically, a four factor solution is advocated, support for this approach is found in
the rule of thumb concerning the percentage of variance explained per factor (> 5%). The four
factor solutions on all items show mixed results, depending on the type of factor analysis (see table
6). Principal axis factoring includes the separate feelings of service/product provider loyalty item
within the word-of-mouth aspect of buyer satisfaction. The principal components analysis however
Table 6: Factor Analysis on All Buyer Satisfaction Items
Principal Axis Factoring – Varimax rotation Principal Components Analysis – Oblimin
rotation
Rotated matrix Pattern matrix
Item 1 2 3 4 1 2 3 4 1 0,214 0,159 0,809 0,088 -0,087 0,034 -0,912 0,129 2 0,206 0,119 0,805 -0,070 0,107 -0,019 -0,920 -0,111 3 0,692 0,042 0,170 0,203 0,068 -0,003 -0,015 0,946 4 0,911 0,073 0,238 -0,106 0,963 0,009 -0,040 -0,001 5 0,953 0,057 0,167 -0,077 0,957 0,001 0,029 0,051 6 0,089 0,870 0,140 0,062 -0,014 0,935 -0,009 0,059 7 0,025 0,872 0,121 -0,055 0,020 0,948 0,007 -0,063
When the service/product provider loyalty is isolated from the analysis, its one item nature makes
exclusion theoretically possible and statistically just, the factor analysis results concerning the
other three aspects show great similarity (see table 7). Using theoretical background and the
eigenvalue (>1) and percentage of variance explained per factor (>5%) rules, a three factor solution
is proposed.
Table 7: Factor Analysis on Six Buyer Satisfaction Items
Principal Axis Factoring – Varimax rotation
Principal Components Analysis – Oblimin rotation
Rotated matrix Pattern matrix
Item 1 2 3 1 2 3 1 0,193 0,160 0,807 -0,003 0,027 -0,917 2 0,200 0,117 0,806 0,004 -0,025 -0,931 4 0,928 0,070 0,249 0,964 0,005 -0,036 5 0,937 0,055 0,187 0,994 -0,003 0,032 6 0,081 0,869 0,146 0,029 0,935 -0,008 7 0,031 0,868 0,121 -0,027 0,949 0,007
Combining the results, the items are joined as theoretically proposed: item 1 and 2, item 3, item 4
and 5, and item 6 and 7. Scales represent the average of the individual items (1 to 7). Item-to-item
and item-to-total correlations are shown in table 8, the corresponding Cronbach’s Alhpa values for
product/service quality, word-of-mouth propensity and switching propensity are: 0.828, 0,958 and
Table 8: Item-to-Item/Item-to-Total Correlations for Buyer Satisfaction Scales
Overall quality satisfaction Word-of-mouth propensity Switch propensity
1 2 Total 4 5 Total 6 7 Total
1 1 4 1 6 1
2 0,708** 1 5 0,920** 1 7 0,775** 1
Total 0,929** 0,919** 1 Total 0,979** 0,981** 1 Total 0,941** 0,943** 1
** P < 0,001
Sampling, Survey and Data Collection Procedures
The participants of this study are Chinese residents on a holiday trip to one of the seven largest casino’s (in terms of total annual revenues) in the popular Chinese gaming destination of Macau.
With support of the Macau Secretariat for Economy and Finance, the seven casinos were
approached for data collection, after which all seven agreed.
Data was gathered using a complete survey written in the Chinese language. Before final data
gathering in situ, the survey was piloted two times in order to improve clarity. Firstly eight
questions, secondly two were revised to come to the final instrument that required four minutes of
time to complete.
Final data collection was done with assistance of three master-level students. They approached the
casino visitors on their way departing the partaking casinos’ table gaming area to request
participation in the survey, using a seated area near these gaming areas. Participating visitors were
all confirmed to be gambling. This procedure was repeated three days a week, alternately between
14:00 and 17:00, and 19:00 and 22:00 for the month of February 2013. Upon agreement to
participate they were offered a food voucher worth USD 25 or an umbrella. Eventually, 50 percent
of the visitors approached in this way participated in the survey. The percentage ranged from 40
Analysis - Ordinary Least Squares and Seemingly Unrelated Regression
This section describes the methods of estimation used to test the hypotheses. The analysis is spread
into two parts to account for the antecedent and outcome approach to the three dimensions on
which compulsive buying profiling is based. The first part of the analysis relies on ordinary least
squares regression (OLS), where the antecedents are modelled into one regression model with
compulsive buying as the dependent variable. The second part uses seemingly unrelated regression
(SUR) to account for the expected correlation of the error terms for the different one on one models
between the four satisfaction measures (dependent) and compulsive buying. Both analyses will be
preceded by pooling-tests to confirm the (in-) appropriateness of pooling the data from the
different casinos. Rosenthal’s (1991) meta-analytic method of added Z’s will be used to generalize
over the per casino models in case pooling is not allowed.
Results
This section will discuss the results of all analyses in order to answer to the proposed hypotheses.
The final results section is divided into two parts, one for each of the methodologies described
earlier. A brief data description will precede the final results to shed light into the structure of the
data.
Data description
Table 9 shows the total and per casino amount of observations. Descriptive statistics concerning
all the variables used in the analyses are provided in table 10. For the categorical variables the
minimum and maximum values are shown to provide insight into the structure of the responses.
The number of missing values is shown in the last column. Next
to four variables with a small number, two variables have a more
serious number of missing values (income with 44 and frequency
with 63). Missing values have been remedied in three ways. For
the demographical variables, observations for age and education
have been removed prior to analysis since saving two degrees of
freedom that “category unknown” dummies would have cost outweighs the small sacrifice of
losing ten observations here. Since income has 44 missing values, those have been replaced with
a dummy variable for category unknown. For the behavioral buying characteristics, missing values
for the categorical variable loyalty/reward program membership are replaced by a missing
category dummy variable while the continuous nature of duration and frequency allowed for
replacement of the missing values by mean imputation.
Table 10: Descriptive Statistics of Full Dataset
Variable Average (µ) / frequency St. Dev. (σ) Min Max # Missing values
Compulsive buying 11,09 3,14 9 24 - Demographics Age 0=151, 1=212, 2=41 7 Gender 0=161, 1=250 - Education 0=161, 1=163, 2=84 3 Income 0=282, 1=58, 2=16, 3=11 44
Behavioral buying characteristics
Loyalty progr. memb. 0=171, 1=234 6
Duration 3,03 1,54 1 10 5
Frequency 2,66 1,79 1 7 63
Expenditure height 0=112, 1=214, 2=68, 3= 17 -
Buyer satisfaction
Overall quality satisfaction 4,83 0,78 2,5 7 -
Analysis – OLS and SUR
The first part in this section will discuss pooling issues raised by the various sources used to gather
the data. The second and main section will discuss the results from both the models that link the
demographical variables and behavioral buying characteristics (treated as antecedents) to
compulsive buying, and the models that link compulsive buying to the buying satisfaction
measures (treated as outcomes). The latter is done in separate models as buyer satisfaction is
theoretically proposed as an outcome of compulsive buying and thus requires compulsive buying
to become the independent instead of dependent variable.
Pooling
Since the data is gathered in seven different casinos, regular (OLS) and separate intercept
(OLSDV) pooling tests verify the (in-) appropriateness of pooling all data together. Table 11 shows
the results for the antecedent models, table 12 for the outcome models. As can be seen from the
p-values in the lower row of table 11, both single and separate intercept pooling of the data is not
allowed (p<0,05). Therefore, models will be estimated per-casino and results will be generalized through Rosenthal’s (1991) meta-analytic method of added Z’s.
Table 11: Pooling Tests for Antecedent Models
OLS OLSDV dfpooled 384 278 dfunpooled 286 286 SSRpooled 2571,93 2445,33 SSRunpooled 1543,93 1543,93 F value 1,94 1,82 P value 0,000 0,000
The p-values for the outcome models in table 12 indicate separate intersect pooling of the data is
propensity models (p>0,05). The SUR models will therefore be estimated in a pooled manner, for
matters of uniformity all models will be estimated with per casino intercepts.
Table 12: Pooling Tests for Outcome Models
Overall quality
satisfaction Loyalty
Word-of-mouth
propensity Switch propensity
OLS OLSDV OLS OLSDV OLS OLSDV OLS OLSDV
dfpooled 399 393 399 393 399 393 399 393 dfunpooled 387 387 387 387 387 387 387 387 SSRpooled 235,63 220,25 681,66 659,90 583,18 559,07 243,16 337,99 SSRunpooled 215,07 215,07 653,41 653,41 551,88 551,88 331,85 331,85 F value 3,083 1,553 1,395 0,444 1,829 0,840 1,099 1,195 P value 0.000 0.159 0,166 0,949 0,042 0,609 0,359 0,284
Findings – part one: demographics and behavioral buying characteristics
Table 13 provides an overview of the results for the separate OLS models for the demographics
and behavioral buying characteristics. Even though the limited number of per casino observations,
with 38 for casino six as the lowest, six out of seven models show to be significant (p<0,05) with
adjusted R2 values ranging from 0,091 to 0,715. Explanatory power thus highly variates, not
surprisingly the number of significant (p<0,05) variables also highly variates from zero to six over
the different casinos. Since casino thee is insignificant, it will be excluded from the meta-analysis.
For matters of completeness, pooling tests have been rerun to look into the consequences of
exclusion of casino three, results indicate no change in conclusions with p-values of 0,000 pooling
still not allowed. Another quick glance at table 13 shows the absence of beta’s for three variables
in three casinos (after exclusion of casino three) due to a lack of presence of observations for those
categories, those will thus not be available for meta-analysis which for those three variables will
Table 13: Results for OLS Models per Casino
Model Cas 1*** Cas 2*** Cas 3 Cas 4*** Cas 5*** Cas 6** Cas 7**
R2 0,435 0,566 0,326 0.671 0,795 0.636 0,471 Adj. R2 0,264 0,436 0,091 0,529 0,715 0,388 0,269 Intercept 9,670*** 8,638*** 7,725*** 7,843*** 8,322*** 5,848* 6,263*** Unstandardized β Age1 1,266* 0,619 1,136 1,082 1,117 0,151 1,303 Age2 -0,405 0,03 2,122 -0,201 1,574 1,951 1,896 Gender 1,241* 0,805 0,376 2,503*** 0,307 -2,466 0,721 Education1 0,26 -0,292 -0,109 -1,082 -0,135 -1,293 1,076 Education2 -0,81 0,124 0,554 -1,288 -0,373 -0,789 1,09 Income1 -0,209 -0,849 0,469 0,285 1,135 -1,607 -0,316 Income2 4,074** -0,506 2,447 -5,224** 3,412 -0,819 Income3 8,068*** 2,563 3,311 0,276 5,77* 4,24 Income Unknown -0,282 1,165 -0,702 2,076 -2,607** -0,722 -2,066
Loyalty progr. memb. 0,564 -0,201 1,846* 2,463** 4,87*** 2,675* 2,065*
Loyalty progr. memb. Unknown 0,595 3,592** -0,498 3,685 8,014*** 1,369
Duration -0,219 0,326* 0,332 -0,515 -0,578** 0,515 -0,083 Frequency -0,017 0,093 -0,074 0,697** 1,045*** 1,527*** 0,895*** Expenditure height1 -0,091 -1,002** -0,036 1,238 -0,749 0,21 -0,273 Expenditure height2 0,368 1,003 -0,676 -1,253 -2,607** -3,248 0,263 Expenditure height3 -2,774* 0,254 0,015 -2,427 1,22 -1,497 * p < 0,10 ** p < 0,05 *** p < 0,01
Durbin Watson values for the 6 remaining models are in-between 1,584 and 1,975, indicating no
reason to suspect autocorrelation (dL, < d < dU at p=0,05). P-values for the Sharpiro-Wilk test of
normality ranging from 0,058 to 0,979 indicate normal distributions (p>0,05), except for casino
five which has a p-value of 0,049, considering its minimal insignificance it is not assumed to cause
any severe misestimation. Finally, multicollinearity is excluded as a potential issue since the
maximum VIF value per casino model range between 2,833 and 4,426 (VIF<10).
As mentioned earlier, Rosenthal’s (1991) meta-analytic method of added Z’s will be used in order
Heerde et. al. (2013), to arrive at the effect sizes of the meta-analytic beta’s, a weighted mean of separate casino beta’s is calculated in which the weight is based on the inverse of the beta’s
corresponding standard errors, normalized to one.
Table 14: Results Added-Z Meta-Analysis
Variable Weighted mean β Meta-Analytic Z Meta-Analytic P Age1 0,940*** 2,724 0,005 Age2 0,504 0,852 0,139 Gender 0,832** 2,294 0,014 Education1 -0,226 -0,641 0,162 Education2 -0,352 -0,768 0,148 Income1 -0,168 -0,389 0,185 Income2 0,476 0,506 0,175 Income3 3,901*** 3,358 0,001 Income Unknown -0,313 -0,616 0,165
Loyalty progr. memb. 1,732*** 4,519 0,000
Loyalty progr. memb. Unknown 3,492*** 3,210 0,001
Duration -0,097 -0,817 0,143 Frequency 0,539*** 5,191 0,000 Expenditure height1 -0,273 -0,851 0,139 Expenditure height2 -0,530 -1,058 0,114 Expenditure height3 -1,161* -1,232 0,093 * p < 0.10 ** p < 0.05 *** p < 0.01
Looking at table 14 that provides an overview of the meta-analysis results, six variables show to
be significant at a 0,05 cutoff level, one more at a cutoff level of 0,1. Each of them will be discussed
in relation to the proposed hypotheses.
Hypothesis one, proposing age to negatively drive compulsive buying is rejected since the middle
age category (36 to 55) scores significantly higher (p<0,01) than the lowest category (reference).
results are in support of an inverted U-shape relationship between age and compulsive buying,
with middle incomes positively driving compulsive buying.
As already pointed out, hypothesis two is particularly vulnerable to the casino origin of the data.
Indeed this hypothesis needs to be rejected considering the positive significant result (p<0,05)
indicating males are more prone to be compulsive buyers (or gamblers) than females. Results
clearly indicate the caution that has to be taken when generalizing over different types of
compulsive behaviors.
Concluding the demographical variables, hypotheses three and four also need to be rejected.
Hypothesis three due to a complete lack of significance, education does not influence compulsive
buying. Hypothesis four due to insignificance of the middle two income categories and a
significant high loading for the highest category beta (p<0,01) in contrast to the reference level,
indicating high incomes strongly positively drive compulsive buying.
Considering the behavioral buying characteristics, support for hypotheses five and seven (p<0,01)
and limited significant (p<0,1) support for hypothesis eight is found, while hypothesis six needs
to be rejected. Both loyalty/reward program membership and the frequency of buying episodes
indeed positively drive people towards the compulsive buying category. The time spend on these
buying episodes seems to be irrelevant, just as the amount of money spend per episode, as there is
only limited evidence that solely extreme high levels of expenditure positively drive compulsive
buying while the middle two categories do not significantly differ from the reference category.
Findings – part two: buyer satisfaction
For matters of uniformity the seemingly unrelated regression models are also run without casino
model allowing for single intercept pooling, the only uniform manner of analysis remains separate
intercept pooling.
Table 15 shows the overview of the outcomes from the analysis for the separate intercept system
of SUR models linking compulsive buying to the satisfaction measures. Although effect sizes and
adjusted R2 values are considerably low, three out of four satisfaction measures are significant
(p<0,01). This signifies that hypotheses nine, ten and eleven find support indicating overall
satisfaction, feelings of loyalty and propensity of word-of-mouth are negatively driven by
compulsive buying. However, this does not mean compulsive buying positively drives the
propensity to switch good/service provider, rejecting hypothesis twelve.
Robustness: Fussy-Set Qualitative Comparative Analysis
Methodologically, another novel and potentially interesting aspect will be displayed here by means
of a robustness check of the OLS analysis. In the current paradigm of correlational analysis,
variability within data is not always exploited to a full extend. Especially in the focal case, with
multiple categorical variables (14 dummies) there is quite some potential for implicit interactions
impossible to estimate in OLS for example. Fuzzy-set qualitative comparative analysis (fsQCA)
provides an algorithmic solution easily allowing for any of those potential interactions. It provides
set-theoretic configurations of causes and conditions that affect the outcome (Ragin, 2008).
Namely, in configurational analysis the influence of the antecedent on the outcome condition is
Table 15: Results for separate intercept SUR Models
either negative, positive or absent depending on the context of the other antecedents. E.g., where
X1, X..., Xetc lead to higher or lower Y in correlational analyses, it is assumed that there are multiple
configurations, or causal recipes (Ragin, 2008) of high and/or low values of X1, X..., Xetc that lead
to high or low Y values in configurational analysis. For a complete background on fsQCA see
Ragin (2008).
Since algorithmic analysis implies combining antecedents to come to a certain outcome, and
interactions in one-on-on modes are impossible by definition, fsQCA will only be applied on the
demographical variables and the behavioral buying characteristics. Before analysis, two topics
need to be discussed for clarification of the methodology and interpretation of the results; the
calibration of the data and the measures of fit (consistency and coverage).
Calibration
Before analyzing the data a procedure similar to z-score transformation has to be performed, in
this case called calibration. Values are converted to a 0.00 to 1.00 scale indicating the degree of
membership on a certain antecedent condition, from full non-membership (0.00 to 0.05) to full
membership (0.95 to 1.00). Maximum ambiguity lies at 0.50. The philosophy behind this lies in
the irrelevancy of reaching scores on a ratio scale that lie beyond the principle of full (non-)
membership. Therefore, the foundation on which calibration is based lies in face validity more
than on sample descriptive statistics (mean/spread), and thus leans on general judgments, standards
or guidelines and researchers’ interpretation. More information can be found in Ragin (2008).
Table 16 provides an overview of the characteristics of the calibration of the original Macau dataset
with the first value being the 0.05 threshold of full non-membership, the middle being the 0.50
Table 16: Data Calibration For fsQCA Analysis
Variable Calibration score
(0.05, 0.50, 0.95) Compulsive buying (9, 11, 16) Age1 (0, 0.5, 1) Age2 (0, 0.5, 1) Gender (0, 0.5, 1) Education1 (0, 0.5, 1) Education2 (0, 0.5, 1) Income1 (0, 0.5, 1) Income2 (0, 0.5, 1) Income3 (0, 0.5, 1) Income Unknown (0, 0.5, 1)
Loyalty progr. memb. (0, 0.5, 1)
Loyalty progr. memb. Unknown (0, 0.5, 1)
Duration (1, 3, 5)
Frequency (1, 2, 3)
Expenditure height1 (0, 0.5, 1)
Expenditure height2 (0, 0.5, 1)
Expenditure height3 (0, 0.5, 1)
Consistency & Coverage
The measures of model fit/usefulness relied on in fsQCA analysis are consistency and coverage,
both sharing similarities with measures from more common statistical analyses. The first,
analogous to correlation, is a measure of the degree to which the data displays a set of common
antecedent conditions associated with an outcome condition. The second, describing the degree to
which this common set of antecedent conditions accounts for the instances of this outcome
condition, is analogous to the coefficient of determination (R2). Both measures range from zero to
one, the closer to one the better. For a full explanation and numerical examples see Ragin (2008).
Findings
Table 17 shows the fsQCA models investigating the relationship between the demographical
high score, empty dots represent a low score. All three models show very high consistency and
moderate coverage, indicating the strength of the models as such (nearly all respondents with this
configuration achieve this outcome) but their moderate explanatory power (not too many of these
configurations exist within the data).
Table 17: fsQCA Models for High Compulsive Buying
Model 1 2 3 Coverage (raw/unique) 0,223/0,084 0,155/0,016 0,172/0,033 Consistency 0,921 0,939 0,901 Age1 ● ● ● Age2 Gender ● ● ● Education1 ● Education2 Income1 ● ● Income2 Income3 Income Unknown
Loyalty progr, memb, ● ● ●
Loyalty progr, memb, Unknown
Duration ● ● ●
Frequency ● ● ●
Expenditure height1 ●
Expenditure height2 ●
Expenditure height3 ●
Fully in line with the results from the meta-analysis, all three algorithms consist of middle aged
males. Looking at education and income however, low and middle categories are present in the
algorithms while the role of education is insignificant and the role of income purely plays a role
when high in the meta-analysis. Both results are thus contesting the meta-analytical results in the
sense that low and middle categories of education and income can drive compulsive buying in the
Looking at the behavioral buying characteristics, results for loyalty/reward program membership
(being a member positively drives compulsive buying), frequency (buying episode frequency
positively drives compulsive buying) and expenditure height (the amount spend during a buying
episode is largely insignificant) are in line with the meta-analysis. In case of expenditure since all
categories pop up in the algorithms. Duration however also plays an important role in all the
algorithms, indicating a high amount of time spend during a buying episode positively drives
compulsive buying while it is insignificant in the meta-analysis.
Concluding, results from the fsQCA analysis are largely corresponding with the meta-analysis.
The allowance for implicit interactions however indicates that configurations with low or middle
levels of education and income and a high amount of time spend per buying episode positively
drive compulsive buying.
Discussion, Recommendations & Limitations
DiscussionThe results for the demographical variables show mixed results in contrast to former literature.
The negative relationship between age and compulsive buying supported by the mass of the
authors only finds limited support here as the high age category does not significantly differ from
the low category benchmark but the middle category does score significantly higher. The negative
relationship thus only holds for middle to high age groups, while between low to middle age groups
age seems to drive compulsive buying positively. Perhaps this contrast arose due to the linear nature of some of the authors’ analysis in contrast to the nonlinearity that dummy coding here
