You are on page 1of 11

AMAR CHEEMA and DILIP SOMAN*

The authors demonstrate that partitioning an aggregate quantity of a


resource (e.g., food, money) into smaller units reduces the consumed
quantity or the rate of consumption of that resource. Partitions draw
attention to the consumption decision by introducing a small transaction
cost; that is, they provide more decision-making opportunities so that
prudent consumers can control consumption. Thus, people are better
able to constrain consumption when resources associated with a
desirable activity (which they are trying to control) are partitioned rather
than when they are aggregated. This effect of partitioning is demon-
strated for the consumption of chocolates (Study 1) and gambles (Study
2). In Study 3, process measures reveal that partitioning increases
recall accuracy and decision times. Importantly, the effect of partitioning
diminishes when consumers are not trying to regulate consumption
(Studies 1 and 3). Finally, Study 4 explores how habituation may
decrease the amount of attention that partitions draw to consumption.
In this context, partitions control consumption to a greater extent when
the nature of partitions changes frequently.

Keywords: self-control, partitions, consumption, attention, guilt

The Effect of Partitions on Controlling


Consumption

Imagine that you are at the concession stand at a movie ation, the excessive degrees of these behaviors have
theater and order a bucket of popcorn. The server tells you resulted in crises in health (with obesity becoming an epi-
that there are no available buckets and instead hands you an demic) and on the economic front (with household savings
equivalent quantity of popcorn packed in three separate plummeting to record lows). The jury is still out on who is
bags. Would this change in packaging influence how much to blame for the epidemic proportions of eating and spend-
popcorn you eat and the pattern of consumption? ing, but most commentators point to the abundant supply of
Over the past few years, the popular press and academic these resources. In particular, access to large portions of
research alike have lamented that people are indulging in food is blamed for the obesity epidemic (The Economist
excessive consumption in terms of both eating and spend- 2006b; Spurlock 2005), and the easy availability of credit is
ing behaviors (Baumeister 2002; The Economist 2006a; typically blamed for the spending problem (CNN Money
Greenhouse 2006; Kyrios, Frost, and Steketee 2004). 2006; Manning 2000).
Although both eating and spending are necessary in moder- Our objective in this article is not to delve into the causes
of the consumption epidemic but rather to provide a pre-
scriptive recommendation. We study situations in which
*Amar Cheema is Assistant Professor of Marketing, Olin Business
School, Washington University in St. Louis (e-mail: cheema@wustl.edu). consumers aim to regulate consumption but are sometimes
Dilip Soman is Corus Chair in Communication Strategy and Professor of unable to do so (O’Donoghue and Rabin 1999). Further-
Marketing, Rotman School of Management, University of Toronto (e-mail: more, we study situations in which these consumers have
dilip.soman@rotman.utoronto.ca). The authors thank the review team; access to a certain quantity of a resource (a quantity of food
Xiuping Li; Anirban Mukhopadhyay; Vanessa Patrick; Yaacov Trope; Bob
Wyer; and seminar participants at the Chinese University of Hong Kong,
or a sum of money) and compare situations in which the
Hong Kong University of Science and Technology, National University of resource is available as an aggregate quantity with situa-
Singapore, New York University, University of Georgia, University of tions in which the resource is partitioned into smaller quan-
Maryland, and Yale University for comments. The authors also thank tities. The opening vignette represents an example of parti-
Prakash Bakshi, Emily Chow, Rachel Cohn, Vivian Lam, Frosti Lau, tioning (aggregate popcorn is partitioned into three bags).
Mandy Li, Jennifer Mach, Lisa Marchiondo, Dawn Moses, Ashley Rosen,
and Szeling Tam for excellent research assistance. Ravi Dhar served as Likewise, a cash payment of $500 might be presented in
associate editor for this article. one envelope or in five separate envelopes, each containing
$100, and 20 cookies might be packaged all together in a

© 2008, American Marketing Association Journal of Marketing Research


ISSN: 0022-2437 (print), 1547-7193 (electronic) 665 Vol. XLV (December 2008), 665–675
666 JOURNAL OF MARKETING RESEARCH, DECEMBER 2008

box or in four subcontainers with 5 cookies in each. We ask decisions. The former process is typically assumed to occur
the following specific questions: outside of awareness, whereas the latter can be consciously
modified (Shiffrin and Dumain 1981; for a review, see
1. When we hold the quantity constant, does the pattern of con- Hastie 2001). Smith and DeCoster (2000) propose that the
sumption differ when the resource is aggregated versus parti- automatic (associative) system effortlessly processes salient
tioned? Across four studies involving food and money, we
cues. In contrast, the controlled (rule-based) system is con-
show that partitioning the resource results in reduced quan-
tity or rate of consumption.
scious and effortful. Rules control impulsive behavior (e.g.,
2. Why do partitions affect consumption? We suggest that when eating too much chocolate) by inflicting guilt, remorse, or a
consumers encounter a partition (e.g., finish one bag of pop- loss of faith in oneself when rules are violated (Benabou
corn), their decision making moves from being automatic to and Tirole 2004; Prelec and Herrnstein 1991; Thaler and
more deliberative. In other words, eating the next kernel of Shefrin 1981). Heckhausen and Gollwitzer (1987) also dis-
popcorn from an open bucket is typically a thoughtless cuss two stages of decision making: predecision delibera-
process, and consumers might not think about the decision. tion and postdecision implementation; people’s actions are
However, a small transaction cost (e.g., opening a fresh bag relatively more automatic in the latter (implemental) stage
of popcorn) draws attention to the decision, and the resultant than in the former (deliberative) stage (Gollwitzer 1999).
deliberation leads to consumers deciding to stop or postpone These streams of research suggest that as the amount of
consumption. attention consumers pay to the decision increases and as
consumers are provided with explicit decision-making
We test our predictions in a series of four studies. In
opportunities, the decision is more likely to be made
Study 1, we demonstrate the effect of partitioning on
because of rules rather than impulse.
chocolate consumption, and in Study 2, we replicate these
Thus, the provision of decision-making opportunities
effects in the domain of gambling (with tokens presented in
may have significant implications for consumption by
aggregate or partitioned form). In Study 3, we use process
drawing attention to the decision. On the one hand, con-
measures to illustrate the effect of partitioning on decision
sumers may impulsively consume hedonic products (e.g.,
making (in particular, more accurate recall and longer deci-
chocolates/gambles). On the other hand, consumers may
sion times). Importantly, the effect of partitioning weakens think deliberately, taking into account possible negative
when consumers are not trying to regulate their consump- consequences of breaking preset rules (weight gain/monetary
tion activity (Studies 1 and 3). In Study 4, we examine the loss), and thus restrain consumption. Such a conflict would
possibility that, over time, the partition could become part not arise if the deliberate decision were identical to the
of the automatic consumption ritual. We show that fre- impulsive one—that is, if consumers did not try to follow
quently changing the nature of partitions, rather than keep- rules that discouraged consumption. Indeed, Strack and
ing them unchanged, increases their effect on consumption. Deutsch (2004, p. 228) suggest that deliberate attention
These results suggest important implications for behav- associated with the conscious system “allows individuals to
iors that consumers should follow but find it difficult to do resist immediate rewards and strive for more valuable
so. Specifically, drawing attention to consumption through future outcomes.” In summary, consumers may be able to
partitions encourages control, especially by consumers who constrain consumption successfully if (1) this activity is
are trying to constrain consumption (for related work on something they are trying to control and (2) they pay delib-
food portions, see Delaney 2007). Although the domain of erate attention to the decision. Consequently, mechanisms
inquiry herein is restricted to physical partitioning, it repre- that increase attention may better enable consumers to con-
sents one form of intervention that draws attention to con- trol consumption.
sumption. Other interventions that could achieve the same How can decision-making opportunities that draw atten-
means include reminders, cognitive interventions, rhetorical tion to consumption be created? Partitions provide one such
questions, auditory signals, targets, or “progress markers.” intervention because consumers must physically negotiate
Such attentional tools are expected to have a similar effect them (e.g., tear, break, open, push aside) to continue con-
as that of partitions on consumption. In addition to decreas- sumption. Each such partition adds a small transaction cost
ing consumption of tempting items, drawing attention to that not only draws attention through the physical act of
decisions through such mechanisms may discourage pro- negotiating the partition but also adds a small temporal
crastination and encourage consumers to engage in behav- delay to consumption that can induce deliberation. Prior
iors that are necessities (e.g., dentist visits, medical exams). research has shown that physical proximity to tempting
options makes impulsive behavior more likely (e.g., Mis-
THEORETICAL BACKGROUND
chel, Shoda, and Rodriguez 1989), potentially creating
We propose that encountering a partition during con- “lapse districts” in which people cannot control consump-
sumption increases the amount of attention consumers pay tion (Ainslie 2000). Because the presence of partitions is
to the decision, giving them a “decision point” at which to likely to hamper proximity to tempting options, we expect
evaluate whether to continue, and thus shifts consumption that people with partitioned resources will be less likely to
decisions from an automatic mode to a deliberative mode. lapse and thus will consume less than people with aggre-
Consequently, people who are trying to control consump- gate resources.
tion can use this increased attention to exert self-control Importantly, another manifestation of successful control
and stop. In the following paragraphs, we review prior may be a slower rate of consumption over the entire set of
research and build our theoretical framework. resources, such that people who encounter partitions choose
Several streams of research in cognitive and social psy- to consume less in a given instance than people with aggre-
chology draw contrasts between automatic (implicit) and gated resources. Postponing consumption in a given
controlled (explicit or deliberative) processes of making instance will lead to a slower rate of consumption and, if
The Effect of Partitions on Controlling Consumption 667

this pattern is followed consistently by the consumer, to 11 = “very much”; M = 8.59) but acknowledged that eating
lower overall consumption. Thus: it was not always beneficial (“I believe that eating too much
chocolate is not good for me”: 1 = “disagree,” and 11 =
H1: People consume fewer resources associated with a desir-
able activity they are trying to constrain (or consume them “agree”; M = 8.41). We also measured the extent to which
more slowly) when the resources are partitioned than when overconsumption was aversive to them (average of two
the resources are aggregated. items: “If I eat too much chocolate, I feel guilty/I feel as if
I let myself down”: 1 = “disagree,” and 11 = “agree”). To
We emphasize that the proposed effect of partitions is not rule out alternative explanations, we measured participants’
expected in situations in which the consumption activity is (1) variety-seeking tendency (Donthu and Gilliland 1996)
not being controlled—that is, when the deliberative, con- and (2) belief that unwrapped chocolates spoil quickly (1 =
scious decision is consistent with the impulsive one. “disagree,” and 11 = “agree”).
Specifically: Thus, the study employed a 2 (partitioning: no, yes) × 3
H2: The effect of partitioning on consumption decreases when (chocolate type: white, milk, dark) mixed design, with par-
the consumption is not considered an undesirable activity. titioning as a between-subjects factor and chocolate type as
a within-subjects factor. The time at which participants ate
The efficacy of partitions in increasing attention may each piece (recorded on the response sheet) was the
also vary across consumption contexts. For example, parti- dependent measure. Participants also answered taste-related
tions are expected in some consumption domains and will questions for each piece (e.g., smoothness, creaminess,
not draw attention to the decision. Moreover, repetition sweetness) to maintain the pretense of a taste study.
may decrease attention paid to decisions that were once
deliberate because of adaptation and because consumers Results
could learn to break partitions habitually (e.g., Johnston
and Dark 1986). Thus, expected partitions (e.g., white parti- Effect of partitions. Of the 73 participants, 24 (33%)
tions between cookies) may not increase attention. In con- returned their response sheets within the seven-day period.
trast, unexpected partitions (e.g., partitions of varying col- Respondents and nonrespondents did not differ in terms of
ors) may draw greater attention to the decision and thus be liking of chocolate and overconsumption aversion. We
more effective in constraining consumption. A similar dis- excluded two incomplete responses; all analyses are for 22
tinction can be made between cash (bills) and gift cards of response sheets (11 wrapped, 11 unwrapped), each contain-
the same denomination. Thus: ing information on six pieces of chocolate. We conducted a
mixed-model analysis with chocolate type as a within-
H3: Partitions that change form over time continue to influence subject factor and two repeated observations for each
consumption over a longer duration than partitions that chocolate type (i.e., six observations per participant). Con-
remain unchanged. sistent with H1, participants with aggregate (versus parti-
tioned) chocolates took less time to eat each chocolate
STUDY 1: EFFECT OF PARTITIONS ON CHOCOLATE (Maggregate = 1.71 versus Mpartitioned = 2.50 days; F(1, 62) =
CONSUMPTION 6.70, p = .01, η2 = .07). Consumption speed did not vary
Participants, Method, and Design across chocolate types (F(2, 62) = 1.72, not significant
We recruited 73 undergraduate female students at a [n.s.]). The cumulative consumption of chocolate over
North American university to participate in a chocolate- seven days for the two packaging conditions appears in Fig-
tasting study in return for a token payment, along with an ure 1. Participants with aggregate (unwrapped) pieces con-
opportunity to taste different chocolates. Participants sumed most of the pieces in the first two days, whereas par-
received a sealed box containing six pieces of Godiva ticipants with partitioned (wrapped) pieces ate them more
chocolate (two pieces each of white, milk, and dark choco- slowly.
late). The box was accompanied by a response sheet con- Moderating role of aversion. We expected that extent of
taining pictures and titles of the different chocolates, along aversion to overconsumption would also deter chocolate
with questions that the participant would answer after eat- consumption (H2). Specifically, some participants may
ing each piece. Thus, all participants knew the contents of want to control consumption more than other participants.
the box. We asked the participants to consume these choco- Thus, the effectiveness of partitions in slowing consump-
lates over the following week and to return the response tion should be greater for the former versus the latter. A
sheet to the experimenter after finishing all the pieces. The mixed-model analysis revealed a significant partition ×
packaging was manipulated between subjects. Half the par- aversion interaction, indicating that the effect of partitions
ticipants received boxes that had unwrapped pieces inside was stronger for participants who had greater aversion to
the box (aggregate). Thus, opening the box allowed them overconsumption (F(1, 60) = 13.58, p < .001, η2 = .15), in
unhindered access to all six pieces. The remaining partici- support of H2. In addition, there was a negative main effect
pants received boxes that had the pieces individually of overconsumption aversion (F(1, 60) = 5.81, p < .05, η2 =
wrapped in foil (partitioned). These participants needed to .06) and a main effect of partitions (F(1, 60) = 4.58, p <
open the box and then open each piece to eat it. The .05, η2 = .05). These were qualified by the aforementioned
response sheet included color pictures of the wrapped or interaction.
unwrapped pieces, respectively.
Participants completed an in-class survey about their atti- Discussion
tudes toward chocolate that was matched with response The results from this experiment supported H1; specifi-
sheets. Overall, the participants liked chocolate (“How cally, resources were consumed at a slower rate when they
much do you like eating chocolates?” 1 = “not at all,” and were partitioned. In addition, partitioning was more effec-
668 JOURNAL OF MARKETING RESEARCH, DECEMBER 2008

Figure 1 Alternative explanations. We also ruled out two other


STUDY 1: PARTITIONING SLOWS RATE OF CHOCOLATE explanations. First, participants who seek more variety may
CONSUMPTION consume pieces faster than others. However, variety-
seeking tendency did not affect consumption (F(1, 61) =
.84, n.s.), and the effect of partitions remained significant
after we controlled for variety seeking (F(1, 61) = 7.34, p <
.01, η2 = .09). Second, the fear of the chocolate spoiling
could lead to the unwrapped (versus wrapped) pieces being
consumed faster. However, freshness did not affect con-
sumption (F(1, 61) = 1.39, n.s.), and the effect of parti-
tioning remained significant after we controlled for it
(F(1, 61) = 4.59, p < .05, η2 = .04).
Study 1 demonstrated the effect of partitions on con-
straining consumption of a hedonic product (chocolates).
We now extend this research to study spending decisions in
the domain of gambles. Our participant pool reported that
though gambling was enjoyable, it was an activity they did
not want to engage in too frequently.
STUDY 2: THE EFFECT OF PARTITIONS ON GAMBLES
Participants, Method, and Design
Fifty-five participants at an Asian university participated
in a risk study after finishing an unrelated experiment. Each
participant received 100 coupons with which they could
gamble. The game was similar to two rolls of dice: Each of
tive for participants who had greater aversion to over- two independently drawn cards could show any number
consumption (H2). from 1 to 6 with equal probability. It cost one coupon to
Individual-level analysis. As a robustness check, we ana- place a bet, and the participant won five coupons (placed in
lyzed the data using one observation per participant. We a winnings box that the participant could see but not draw
designated participants as being averse or not averse to from) if the sum of the two cards was 9 or higher. Partici-
overconsumption depending on whether they were above or pants bet repeatedly until they either exhausted the coupons
below the midpoint of the aversion scale, respectively. The or chose to stop. Coupons could be cashed at any point for
dependent measure was the amount of time participants US$.064 per coupon. The 10/36 winning probability
took to finish all six pieces of chocolate. Among partici- resulted in a positive expected value [.064 × 5 × (10/36) =
pants who were averse to overconsumption, those with US$.0889].
aggregate (versus partitioned) chocolates finished the We manipulated the number of envelopes containing the
six pieces significantly faster (Maggregate = 1.25 versus 100 coupons (1, 4, or 10) between subjects. Some partici-
Mpartitioned = 5.50 days; F(1, 18) = 10.05, p = .005, η2 = pants received one envelope with 100 coupons (aggregate
.31). In contrast, partitions did not affect speed of finishing resources), others received four envelopes with 25 coupons
the six pieces for those not averse to overconsumption in each, and the remaining participants received 10 sealed
(Maggregate = 3.00 versus Mpartitioned = 1.40; F(1, 18) = 1.74, envelopes with 10 coupons in each. We recorded the num-
n.s.). This led to a significant partition × aversion inter- ber of coupons used (i.e., the number of gambles) and the
action (F(1, 18) = 10.47, p < .005, η2 = .32), in support of percentage of coupons spent from the last open envelope.
H2. The small number of observations (n = 22) notwith-
standing, the pattern of results also suggests that aggregate Manipulation Check for Gambling Aversion
(versus partitioned) chocolates are consumed faster. Indeed, Although most people try to control spending on vices,
82% of the participants in the aggregate condition ate all such as gambling, we checked to determine whether this
their chocolates within the first two days, compared with was also true for the subject population. In a pretest with
45% who did so in the partitioned condition, a difference payoffs identical to the ones used here, 59 participants rated
that approaches significance (χ2(1) = 3.25, p = .07). how appropriate it was for them to gamble in the study (1 =
Consumption rate versus amount. All participants in “not at all appropriate,” and 7 = “very appropriate”). The
Study 1 were instructed to finish the task (i.e., eat all the responses indicated that, in general, participants believed
chocolates) and return the taste evaluations in a week. that it was inappropriate to gamble (M = 2.88). This evalua-
Consequently, overall consumption is unchanged across tion was significantly lower than the scale midpoint (of 4;
conditions (i.e., everyone consumed six pieces) but is F(1, 58) = 112.15, p < .0001), indicating an aversion to
slower in the presence (versus absence) of partitions. As gambling.
we discussed in the setup for H1, a slower rate of con-
sumption over time would also result in a smaller amount Results
of total consumption. Thus, the slower rate of consumption Effect of partitions on total consumption. Table 1 shows
in Study 1 acts as a proxy for reduced amount. Further- that participants with partitioned resources (four or ten
more, in Study 2, we study the effect of partitions on total envelopes) spent less than those with aggregate resources
spending. (one envelope). Participants with one envelope used more
The Effect of Partitions on Controlling Consumption 669

Table 1 Figure 2
STUDY 2: PARTITIONS DECREASE SPENDING STUDY 2: PARTITIONING DECREASES SPENDING ON
GAMBLES
Number of Envelopes
1 4 and 10 A: One Envelope (with 100 Coupons)
Aggregate 4 10 Partitioned
Experiment Design
Coupons in each envelope 100 25 10 —
Number of participants, n 21 18 16 34
Number of coupons gambled 42.62 25.50 16.44 21.24

Dependent Measures
% of last envelope used 64 90 88 89
% of participants in a condition
who did not open an envelope 33 11 13 12

coupons (M1envelope = 42.62) than those with four envelopes


(M4envelopes = 25.50; F(1, 52) = 4.23, p < .05, η2 = .06) or
those with ten envelopes (M10envelopes = 16.44; F(1, 52) =
9.27, p < .005, η2 = .14). The former (one envelope) used B: Four Envelopes (with 25 Coupons in Each)
more coupons (Maggregate = 42.62) than those with parti-
tioned resources (four or ten envelopes; Mpartitioned = 21.24;
F(1, 52) = 9.06, p < .005, η2 = .13), in support of H1. The
difference between four and ten envelopes, though in the
expected direction, was not significant (F(1, 52) = 1.04,
n.s.). The main effect of number of envelopes was signifi-
cant (F(2, 52) = 4.94, p < .05, η2 = .13).
Effect of breaking a partition on consumption. Figure 2
shows the impact of breaking a partition on subsequent
spending of resources. Each vertical bar shows the number
of coupons used by one participant. Among the participants
with one envelope (Panel A), seven did not open an enve-
lope; the eighth (Participant 10) used 14 coupons, and so
forth. The horizontal dashed line shows the number of
coupons in each envelope (the partition between enve-
C: Ten Envelopes (with 10 Coupons in Each)
lopes). Most participants with multiple envelopes emptied
the opened envelope and stopped short of opening the next
one. Among participants who opened at least one envelope,
those with partitioned resources (four and ten envelopes)
spent a greater proportion of the last opened envelope
(Mpartitioned = 89% versus Maggregate = 64%; F(1, 41) =
12.89, p < .005, η2 = .22).
Discussion
The effect of partitions on total spending was consistent
with H1. Drawing attention to the gambling decision
(breaking a partition) decreased spending. After a partition
was broken, however, participants appeared to spend read-
ily until they encountered the next partition. Because par-
ticipants who opened an envelope were more likely to
spend the coupons within, participants with fewer partitions
spent more than those with a larger number of partitions.
A greater proportion of participants with one envelope envelopes (100 versus 25 or 10 coupons). We leave this
did not use any resources (Maggregate = 33%), compared effect of resource size as an area for further research.
with those with four or ten envelopes (Mpartitioned = 12%; Study 2 demonstrated that partitions can constrain spend-
χ2(1) = 3.77, p = .05). We also contrasted the four- and ten- ing on gambles. Although pretests revealed that participants
envelope conditions because no participant with ten are averse to gambling in aggregate, we did not measure
envelopes opened more than four envelopes. Participants individual-level aversion to predict effectiveness of parti-
with four envelopes opened marginally fewer envelopes tions (as we did in Study 1). Study 3 extends the current
(M4envelopes = 1.11) than those with ten envelopes study by measuring individual-level aversion and also
(M10envelopes = 1.75; F(1, 32) = 3.84, p < .10, η2 = .08). recording two measures of participants’ attention: the
These results suggest that the effort associated with break- amount of time they take to place each bet and the accuracy
ing a partition was greater for larger (versus smaller) with which they can recall their spending after completing
670 JOURNAL OF MARKETING RESEARCH, DECEMBER 2008

the task. Greater attention may be reflected in the length of Figure 3


time participants take and their ability to recall spending STUDY 3: AVERSION MODERATES EFFECT OF PARTITIONS
more accurately. Furthermore, gambles in Study 3 had a ON SPENDING
neutral payoff, whereas those in Study 2 had a positive
expected value, removing a bias that may have caused
Study 2 participants to gamble too much.
STUDY 3: AVERSION MODERATES THE EFFECT OF
PARTITIONS ON GAMBLES
Participants, Method, and Design
We paid 91 participants at an Asian university to partici-
pate in a risk study. The game was similar to that in Study
2, with the exception of the payoff. Participants in Study 3
bet one coupon for each gamble and won six coupons if the
numbers on two independently drawn cards (which showed
a number between 1 and 6 with equal probability) totaled
10 or higher. This 6/36 probability of winning translated to
a neutral expected value (6 × 6/36) that was no different
than cashing in the coupon, which they could do at any
time. Thus, the gambling was driven by the participants’
affinity for or aversion to gambling. All participants
received 100 coupons that they could cash in or bet. We
manipulated the partitioning between subjects. Participants
in the aggregate condition received 100 coupons in a single
envelope, and the remaining participants (in the partitioned
condition) received 20 coupons each in five envelopes.
We used a proofreading task as a between-subjects
manipulation of gambling aversion before participants the high-aversion conditions, participants with partitioned
began the risk study. All participants read four paragraphs (versus aggregate) resources placed fewer bets
about gambling. Two of these paragraphs were common (Mpartitioned = 22.56 versus Maggregate = 52.70; F(1, 87) =
across conditions. In addition, participants in the low- 16.16, p < .0005, η2 = .14). However, this effect of parti-
aversion condition read two paragraphs about the positive tions was not significant in the low-aversion conditions
aspects associated with gambling (glamour, skill, and (Mpartitioned = 48.96 versus Maggregate = 56.10; F(1, 87) =
wealth), and those in the high-aversion condition read two .92, n.s.). The aversion × partition interaction (F(1, 87) =
paragraphs about the negative aspects of gambling (stress, 4.74, p < .05, η2 = .05) qualified the main effects of parti-
poor health, and debt). Manipulation checks (average of tions (Mpartitioned = 36.02 versus Maggregate = 54.40;
two items: “How guilty do you feel when you gamble?” 1 = F(1, 87) = 12.46, p < .001, η2 = .11) and aversion (Mhigh
“not at all guilty,” and 7 = “very guilty”; “How appropriate aversion = 35.96 versus Mlow aversion = 52.07; F(1, 87) = 7.97,
is it for you to gamble?” (reverse scaled): 1 = “not at all p < .01, η2 = .07).
appropriate,” and 7 = “very appropriate”; r = .48, p < The usage pattern appears in Figure 4. In the graph for a
.0001) revealed that guilt was higher in the high-aversion condition, each vertical bar shows the number of coupons
condition than in the low-aversion condition (Mhigh aversion = used by one participant in that condition. For example,
5.67 versus Mlow aversion = 2.90; F(1, 87) = 278.32, p < among participants in the high-aversion, aggregate
.0001). resources condition (Panel B), five participants did not
Thus, the study was a 2 (gambling aversion: low, high) × open the envelope, and the sixth and seventh participants
2 (partitioned: no, yes) full-factorial between-subjects each bet 40 coupons. The horizontal dashed lines in Panels
design with 20–25 participants randomly assigned to C and D denote the partition between one envelope and the
each condition. We recorded the number of gambles along next.
with how long participants took to gamble (i.e., the dura- Number of partitions opened. We find that in the high-
tion for which they played the game). After the participant aversion, partitioned condition (Figure 4, Panel D), most
decided to stop playing (or ran out of coupons to gamble participants opened two or fewer envelopes, in contrast to
with), the experimenter collected the remaining coupons those in the low-aversion, partitioned condition (Panel C).
and envelopes and the winnings. While the experimenter Thus, among those with partitioned resources, participants
calculated participants’ payoff at a separate table, partici- who were trying to control spending (i.e., those in the high-
pants completed the manipulation checks (described in the aversion condition) opened significantly fewer envelopes
preceding paragraph) and recalled how many coupons they than those in the low-aversion condition (Mhigh aversion =
had bet. Recall errors (indicated by the absolute difference 1.44 versus Mlow aversion = 2.73; F(1, 49) = 20.29, p < .0001,
between recalled and actual amounts) and time per bet were η2 = .28).
used as measures of attention. Recall errors. We expected participants who paid greater
attention to the gambling decision to exhibit fewer recall
Results errors (calculated as the absolute difference between
Number of gambles. Figure 3 reveals that gambling aver- recalled and actual number of bets for participants who bet
sion moderates the effect of partitions on gambles (H2). In at least one coupon). Consistent with the notion that parti-
The Effect of Partitions on Controlling Consumption 671

Figure 4
STUDY 3: PARTITIONS DECREASE SPENDING FOR HIGH-AVERSION CONSUMERS

A: Low Aversion, Aggregate (1 Envelope) B: High Aversion, Aggregate (1 Envelope)

C: Low Aversion, Partitioned (5 Envelopes) D: High Aversion, Partitioned (5 Envelopes)

tions draw attention to the decision, participants with parti- affect spending. Specifically, increased attention (reflected
tioned (versus aggregate) resources exhibited fewer recall by longer times per bet) diminished spending to a greater
errors (Mpartitioned = 2.83 versus Maggregate = 6.57; extent for participants who were averse to gambling, in sup-
F(1, 80) = 10.19, p < .005, η2 = .10). This increased atten- port of H2 (see Figure 5, Panel B). The effect of the aver-
tion (reflected by fewer errors) reduced spending only for sion × time interaction on spending was significant
participants who were averse to gambling, in support of H2 (F(1, 80) = 11.83, p < .001, η2 = .10). Log-transformed
(Figure 5, Panel A). Specifically, the aversion × error inter- time measures revealed similar results.
action significantly affects spending (F(1, 80) = 4.24, p <
.05, η2 = .04). Discussion
Decision time. Another measure of attention, the time In conditions in which participants were averse to gam-
taken per bet, shows a similar pattern of results. Partici- bling, the pattern of observed decisions was consistent with
pants with partitioned (versus aggregate) resources took the expected effect of partitions on spending. Betting from
longer to bet (Mpartitioned = 23.51 versus Maggregate = 15.67 partitioned resources drew greater attention to the gambling
seconds; F(1, 80) = 163.23, p < .0001, η2 = .67). Although decision than did using aggregate resources, as reflected by
the physical requirement of opening partitioned (versus fewer errors in recall and in longer decision times per bet.
aggregate) resources may increase decision time, the mod- Consequently, participants with partitioned resources spent
erating role of aversion suggests that attentional differences less than those with aggregate resources, in support of H1.
672 JOURNAL OF MARKETING RESEARCH, DECEMBER 2008

Figure 5 resources placed significantly fewer bets than those with


STUDY 3: AVERSION MODERATES THE EFFECT OF aggregate resources (Mpartitioned = 21.76 versus Maggregate =
ATTENTION ON SPENDING 56.20; F(1, 87) = 21.53, p < .0001, η2 = .18). In contrast,
partitions did not affect participants who were not averse to
A: Effect of Recall Errors on Spending gambling (Mpartitioned = 49.73 versus Maggregate = 52.60;
F(1, 87) = .15, n.s.). The main effects of partitions
(F(1, 87) = 12.74, p < .001, η2 = .18) and aversion
(Maverse = 37.07 versus Mnot averse = 50.98; F(1, 87) = 5.44,
p < .05, η2 = .04) were qualified by the aforementioned
interaction.
In Study 4, we explored another potential moderator of
partition effectiveness—the nature of the partition. Some
partitions may be expected, and over time people may
begin to remove partitions habitually. In such situations,
partitions that differ from expectations (versus those that
are expected) may constrain consumption to a greater
extent. Specifically, we study the effect of separators in a
pack of cookies. Consumers often expect white separators
between cookies, and these partitions may be removed
automatically as consumers become habituated to them.
Thus, an acceleration in consumption over time (with the
decreasing effectiveness of white partitions) may be
observed. In contrast, separators of varying colors are more
B: Effect of Decision Time on Spending likely to attract attention, and consumers may be less likely
to remove such partitions automatically. Thus, partitions of
varying colors may constrain consumption more than white
partitions.
STUDY 4: EFFECT OF DIFFERENT TYPES OF
PARTITIONS ON CONSUMPTION
Participants, Design, and Procedure
Fifty-four students at a North American university who
were participating in a series of unrelated research studies
were given 20 cookie crisps in a container as refreshments
that they could eat while completing the studies. The pack-
aging of the cookies was manipulated between subjects.
One-third of the participants received cookies in an aggre-
gate form (no partitions). Another one-third of the partici-
pants received white (wax paper) partitions. The last one-
third of participants received partitions of varying colors
between cookies. Therefore, the study design was a single-
factor 3 (partitions: none, white, color) between-subjects
design with 18 participants randomly assigned to each of
the three experimental conditions. The entire session of
However, when participants were not averse to gambling, studies took approximately 140 minutes to complete. We
partitioning did not affect gambles, in support of H2. This discreetly monitored participants’ cookie consumption in
interaction of aversion with attention measures (recall (14) 10-minute intervals during this session. The primary
errors and decision time) provides additional support for dependent measure was how long each participant took to
the proposed process. finish the cookies (range = 1–14). We conservatively used
In Study 3, we also asked each participant to report the 15 intervals as the duration for participants who did not fin-
extent to which he or she was averse to gambling (reported ish the 20 cookies in the 14 intervals. To explore the pat-
aversion: average of two items: “How guilty do you feel tern, we also split the 14 intervals into three periods and
when you gamble?” 1 = “not at all guilty,” and 7 = “very contrasted consumption across conditions.
guilty”; “How appropriate is it for you to gamble?” (reverse
scaled): 1 = “not at all appropriate,” and 7 = “very appropri- Results
ate”; r = .48, p < .0001). As for Study 1, we used scores Rate of consumption. We expected that white partitions
that were higher or lower than the midpoint to categorize would slow consumption to a lesser extent than color parti-
participants as being averse to gambles or not, respectively. tions. Although participants with no partitions finished the
Analyses of the number of bets with this individual-level cookies as quickly as participants with white partitions
measure reveal a significant partition × aversion interaction (Mnone = 9.39 versus Mwhite = 9.94; F(1, 51) = .90, n.s.),
(F(1, 87) = 9.12, p < .005, η2 = .07). Among participants participants with color partitions took significantly longer
who were averse to gambling, those with partitioned (i.e., consumed more slowly; Mcolor = 14.39) than those
The Effect of Partitions on Controlling Consumption 673

with no partitions (F(1, 51) = 72.93, p < .0001, η2 = .58) or titions consumed most of the cookies in Period 1 (Intervals
those with white partitions (F(1, 51) = 57.46, p < .0001, 1–5: Mnone, Period 1 = 13.61) and consumed the remaining
η2 = .52). Thus, participants with white partitions appeared cookies in Period 2 (Intervals 6–10: Mnone, Period 2 = 5.83).
to be less affected by the partitions, eating cookies in a A mixed model analysis revealed that more cookies were
manner consistent with having no partitions. However, consumed in Period 1 than in Period 2 (F(1, 102) = 107.95,
those with color partitions consumed at a significantly p < .0001, η2 = .16). For participants with white partitions,
slower rate. however, cookie consumption accelerated from Period 1
Consumption pattern. Next, we studied the dynamic pat- (Mwhite, Period 1 = 7.28) to Period 2 (Mwhite, Period 2 = 11.61;
tern of consumption by dividing the 14 intervals into three F(1, 102) = 33.52, p < .0001, η2 = .04). In contrast, this
periods. This enabled us to explore whether participants acceleration was not significant for participants with color
paid less attention to partitions (especially white ones) as partitions (Mcolor, Period 1 = 5.33 versus Mcolor, Period 2 = 6.56;
time progressed and partitions were repeatedly removed. F(1, 102) = 2.66, n.s.). This interaction between partitions
The pattern of cumulative cookie consumption appears in (white versus color) and period (first versus second) was
Figure 6. Participants with no partitions consumed cookies significant (F(1, 34) = 5.92, p < .05, η2 = .04). The overall
the fastest, followed by those with white partitions. partition × period interaction was also significant
Participants with white partitions appeared to accelerate (F(4, 102) = 58.88, p < .0001, η2 = .35).
their consumption as time progressed in relation to partici- Discussion
pants who had colored partitions. Table 2 summarizes con-
sumption across the three periods. Participants with no par- The results support H3; moreover, consumption differ-
ences between white versus color partitions rule out physi-
Figure 6 cal effort as the primary driver of the effect. Specifically,
STUDY 4: PARTITIONS WITH CHANGING COLORS ARE MORE acceleration in consumption from Period 1 to Period 2 for
SUCCESSFUL IN SLOWING COOKIE CONSUMPTION participants who had white (versus color) partitions indi-
cated that with habituation, these participants decreased the
amount of attention paid to partitions. The proportion of
participants finishing the 20 cookies was also consistent
with the consumption rate. Significantly fewer participants
with color partitions finished their cookies (4/18 = 22%)
than participants with white partitions (17/18 = 94%) or
participants with no partitions (17/18 = 94%; χ2(2) = 19.62,
p < .0001).
GENERAL DISCUSSION AND CONCLUSIONS
Summary of Research
We demonstrate the effect of partitions on consumption
and explore the boundaries of this effect across four stud-
ies. Consistent with H1, we find that partitions slow the
consumption of chocolates (Study 1) and cookies (Study 4)
and decrease the amount of gambles (Studies 2 and 3).
These results are consistent with the notion that partitions
draw attention to the activity that consumers are trying to
control, providing consumers with a deliberate decision-
making opportunity. Process measures in Study 3 (errors in
recall and decision times) support this attention-drawing
role of partitions. Importantly, the success of partitions in
constraining consumption is attenuated when consumers’
desire to control consumption is low (when the consumers
are not averse to eating chocolate in Study 1 or to gambling
in Study 3), in support of H2. In Study 4, we also demon-
strate that when people become habituated to a (white) par-
tition and thus pay less attention to it, they consume at a
faster rate. In contrast, partitions (of varying colors) that
Table 2 draw more attention are relatively more effective in con-
STUDY 4: PARTITIONS WITH CHANGING COLORS ARE MORE straining consumption, consistent with H3.
SUCCESSFUL IN SLOWING COOKIE CONSUMPTION Viewed together, these studies reveal that people with
partitioned (versus aggregate) resources are more likely to
Period 1 Period 2 Period 3 deliberate on the consumption decision. This increased
Interval range 1–5 6–10 11–14
attention decreases consumption for people who are trying
to control it.
Average Number of Cookies Consumed
No partition (aggregate) 13.61 5.83 .50 Implications
White partition 7.28 11.61 .83 The demonstrated effect of partitions suggests applica-
Colored partition 5.33 6.56 4.72
tions in myriad domains in which people try to control con-
674 JOURNAL OF MARKETING RESEARCH, DECEMBER 2008

sumption. We first discuss the two domains that were high- that could achieve the same means include reminders, cog-
lighted at the outset, namely, food and money. We then dis- nitive interventions, rhetorical questions, auditory signals,
cuss wider implications of partitions and conclude with a targets, or even progress markers. Research conducted by
discussion of other attentional mechanisms that may have one of the authors shows that an auditory beep and an auto-
similar effects. matic shutdown at certain intervals of leaving an air condi-
Food consumption. People trying to exert self-control tioning unit on forces the consumer to actively decide
(and eat less) could divide food into smaller portions, thus whether to reactivate the unit for another period and that
forcing themselves to consciously consider the decision of giving people soda in large sizes, but in cups marked by
breaking a partition to consume the next portion. In con- volume (much like a measuring cup), decreases the volume
trast, having one large portion may lead the person to con- they consume. Other contextual or individual-level con-
sume food automatically until the serving is finished. structs that increase deliberative (versus automatic) evalua-
Wertenbroch (1998) studies a similar strategy of “ration- tion of decisions would also be successful in constraining
ing” vices by buying only small quantities. However, that consumption in the contexts studied herein.
strategy appears to be based on limited availability of the
resource. The current research suggests that even if addi- Limitations and Further Research
tional units of the resource are available in other partitions, Slowing versus reducing consumption. With limited
paying more attention to the consumption decision (need- resources and sufficient time, participants consumed most
ing to break more partitions) leads to greater self-control of the food resources in Studies 1 and 4. However, the rate
success. Frequent variation in partition type would also of consumption was slower with partitioned (versus aggre-
enhance partition effectiveness. Conversely, from the per- gate) resources. We expect that, over time, a slower rate of
spective of a seller, larger packages with no partitions consumption will result in lesser total consumption. We
would increase consumption (Ailawadi and Neslin 1998). also show that partitions decrease consumption in the con-
Financial decisions. Partitions could also help people text of gambles (Studies 2 and 3), and fewer participants
constrain spending and save more. Partitions of money may with colored (versus white or no) partitions finished all the
take the form of physical demarcations, in which money is cookies in Study 4. The relationship between rate of con-
put aside for a specific purpose (e.g., in one or more sumption and overall consumption can also be studied in
envelopes or in bank accounts). The current research sug- greater detail in further research.
gests that people will be less likely to use that money for a Demonstrating dual process. Attentional effects of parti-
purpose other than the one it was earmarked for if it exists tions are demonstrated in process measures from Study 3
in several partitions than when it exists in one aggregate (recall errors and decision times) and in effects of colored
amount. Three studies we conducted in the domain of (versus white) partitions in Study 4. However, demonstrat-
spending on unplanned desirable purchases show that ing the existence of automatic versus deliberate processes
money earmarked for a different expense was less likely to remains a fruitful avenue of further research. Such research
be spent when the person needed to break multiple parti- could focus on the effects of cognitive load. The automatic
tions. These partitions were implemented in the forms of behavior of consuming a proximal tempting option should
(multiple) gift cards, bank accounts, and certificates of not change as a function of whether the person is placed
deposit. It could be argued that currency denominations are under cognitive load. In contrast, deliberative decision mak-
also partitions of an aggregate amount (e.g., $100 may be ing (when a person encounters a partition) would be ham-
partitioned into ten bills of $10 each). However, we expect pered by cognitive load. Thus, cognitive load may attenuate
that these partitions would draw less attention to spending the effect of partitions on consumption.
than other physical partitions because people are relatively Individual and contextual differences in attention to con-
habituated to currency denominations. sumption. In addition to interventions that increase atten-
These findings have important implications for savings tion to consumption, further research could examine
rates, in particular among consumers who do not have individual-level differences that affect decision delibera-
access to banking and guaranteed monthly salaries. We tion. For example, people who construe activities at a high
tested our basic partitioning effects in a field study in rural level (why they should consume a resource) may exert bet-
China and India with construction workers who get paid ter self-control than those who construe activities at a low
daily wages in cash. Some of the workers were given sev- level (how they should consume it), consistent with the
eral currency bills as one bundle, and others were given the process that Fujita and colleagues (2006) suggest. In con-
same set of bills but divided equally across four sealed trast, factors that decrease deliberation may increase impul-
envelopes. Over a three-month window, the workers who sive spending. The shopping momentum effect (Dhar,
were given wages in sealed envelopes reported saving more Huber, and Khan 2007) details how the momentum from
of their earnings than the others. Although these findings one purchase makes a second, subsequent purchase more
are preliminary and data need to be collected over a signifi- likely because people who start purchasing are in an imple-
cantly longer time horizon, the results suggest the power of mental (versus a deliberative) mind-set. On the basis of the
the partitioning manipulation in influencing behavior in the current research, we would expect such an implemental
real world. mind-set to persist until the person encounters a partition
Other attentional mechanisms. We emphasize that that draws deliberate attention to the decision.
though physical partitions are the domain of inquiry herein, Effect of partition size on decision to consume. We also
they represent one operationalization of the underlying con- find some evidence that the size of the resource may affect
struct of an intervention that moves consumption decision people’s decision to break a partition. Participants in Study
making from automatic to deliberative. Other interventions 2 were marginally less likely to break a large partition than
The Effect of Partitions on Controlling Consumption 675

a smaller one. However, this effect was smaller than the Delaney, Anna (2007), “Eating Behavior and Portion Control,”
constraining effect of smaller partitions (i.e., participants (accessed August 7, 2008), [available at http://www.calorie
still spent more from one large partition than from several king.com/library/article.php?content_id=3&path=13,66&art_id
small partitions). The reticence to break large partitions =843].
Dhar, Ravi, Joel Huber, and Uzma Khan (2007), “The Shopping
may be based in part on the notion that breaking the parti-
Momentum Effect,” Journal of Marketing Research, 44
tion signals the breaking of a rule, with the size of the parti- (August), 370–78.
tion acting as an indicator of the magnitude of failure. Donthu, Naveen and David Gilliland (1996), “The Infomercial
Thus, people who break a large partition may feel worse Shopper,” Journal of Advertising Research, 36 (March–April),
than those who break a small partition. Indeed, in studies in 69–76.
which people could consume by breaking one partition, we The Economist (2006a), “Battling the Flab,” (August 18),
find that they are directionally more likely to consume (accessed August 20, 2008), [available at http://www.economist.
when they can do so by breaking a small (versus large) par- com/agenda/displaystory.cfm?story_id=E1_SRTQDTV].
tition. This may be because a person who breaks one small ——— (2006b), “Food Firms and Fat-Fighters,” (February 9),
partition is still able to conserve other small partitions, fol- (accessed August 20, 2008), [available at http://www.economist.
com/agenda/displaystory.cfm?story_id=E1_VVDNNTV].
lowing the rule of not opening those. This remains a fruitful
Fujita, Kentaro, Yaacov Trope, Nira Liberman, and Maya Levin-
area for further investigation. Sagi (2006), “Construal Levels and Self-Control,” Journal of
Ambiguous versus clearly defined partitions. Partitioned Personality and Social Psychology, 90 (3), 351–67.
(versus aggregate) resources may also enable consumers to Gollwitzer, Peter M. (1999), “Implementation Intentions: Strong
define clearer rules. Clear rules are easier to follow (e.g., Effects of Simple Plans,” American Psychologist, 54 (July),
Ainslie and Haslam 1992) and may be a factor that affects 493–503.
successful self-control in our results. However, this would Greenhouse, Steven (2006), “Borrowers We Be,” The New York
not explain attentional differences, as demonstrated by Times, (September 3), 12.
recall errors and response times in Study 3. Moreover, even Hastie, Reid (2001), “Problems for Judgment and Decision Mak-
with clearly defined (white versus colored) partitions in ing,” Annual Review of Psychology, 52 (February), 653–83.
Heckhausen, Heinz and Peter M. Gollwitzer (1987), “Thought
Study 4, we observed differences in consumption as a
Contents and Cognitive Functioning in Motivational Versus
consequence of consumers’ habituation to such partitions, Volitional States of Mind,” Motivation & Emotion, 11 (June),
implicating attentional differences as the driver of that 101–120.
effect. Johnston, William A. and Veronica J. Dark (1986), “Selective
This research used clearly defined partitions in all the Attention,” Annual Review of Psychology, 37 (January), 43–75.
studies. In the presence of ambiguity, however, consumers Kyrios, M., R.O. Frost, and G. Steketee (2004), “Cognitions in
may be motivated to redefine the partitions rather than Compulsive Buying and Acquisition,” Cognitive Therapy and
break them (e.g., Cheema and Soman 2006) or to reallocate Research, 28 (April), 241–58.
resources across partitions after a transgression to “make Manning, Robert D. (2000), Credit Card Nation: The Conse-
things right.” Further research could also study how quences of America’s Addiction to Credit. New York: Basic
Books.
consumers may reexert control after breaking a partition.
Mischel, Walter, Yuichi Shoda, and Monica L. Rodriguez (1989),
This may involve repartitioning the available resources to “Delay of Gratification in Children,” Science, 244 (4907),
increase deliberation of future consumption. Such research 933–38.
on reexerting self-control would also have significant impli- O’Donoghue, Ted and Matthew Rabin (1999), “Doing It Now or
cations for controlling consumption. Later,” American Economic Review, 89 (March), 103–124.
Prelec, Drazen and Richard J. Herrnstein (1991), “Preferences or
REFERENCES Principles: Alternative Guidelines for Choice,” in Strategy and
Ailawadi, Kusum L. and Scott A. Neslin (1998), “The Effect of Choice, R.J. Zeckhauser, ed. Cambridge, MA: MIT Press.
Promotion on Consumption: Buying More and Consuming It Shiffrin, Richard and Susan T. Dumain (1981), “The Development
Faster,” Journal of Marketing Research, 35 (July), 390–98. of Automatism,” in Cognitive Skills and Their Acquisition, J.R.
Ainslie, George (2000), “A Research-Based Theory of Addictive Anderson, ed. Hillsdale, NJ: Lawrence Erlbaum Associates,
Motivation,” Law and Philosophy, 19 (January), 77–115. 110–40.
——— and Nick Haslam (1992), “Self-Control,” in Choice Over Smith, Eliot R. and Jamie DeCoster (2000), “Dual-Process Mod-
Time, G. Loewenstein and J. Elster, eds. New York: Sage Publi- els in Social and Cognitive Psychology: Conceptual Integration
cations, 177–209. and Links to Underlying Memory Systems,” Personality and
Baumeister, Roy F. (2002), “Yielding to Temptation: Self-Control Social Psychology Review, 4 (2), 108–131.
Failure, Impulsive Purchasing, and Consumer Behavior,” Jour- Spurlock, Morgan (2005), Don’t Eat This Book: Fast Food and the
nal of Consumer Research, 28 (March), 670–76. Supersizing of America. New York: Putnam.
Benabou, Ronald and Jean Tirole (2004), “Willpower and Per- Strack, Fritz and Roland Deutsch (2004), “Reflective and Impul-
sonal Rules,” Journal of Political Economy, 112 (4), 848–86. sive Determinants of Social Behavior,” Personality and Social
Cheema, Amar and Dilip Soman (2006), “Malleable Mental Psychology Review, 8 (3), 220–47.
Accounting: The Effect of Flexibility on the Justification of Thaler, Richard H. and Hersh M. Shefrin (1981), “An Economic
Attractive Spending and Consumption Decisions,” Journal of Theory of Self-Control,” Journal of Political Economy, 89
Consumer Psychology, 16 (1), 33–44. (April), 392–406.
CNN Money (2006), “Consumer Groups Blast Fed,” (July 10), Wertenbroch, Klaus (1998), “Consumption Self-Control by
(accessed November 9, 2006), [available at http://money. Rationing Purchase Quantities of Virtue and Vice,” Marketing
cnn.com/2006/07/10/pf/creditcards/index.htm]. Science, 17 (4), 317–37.

You might also like