Modeling the effects of dynamic group influence on shopper zone choice, purchase conversion, and spending

Abstract

In many retail contexts, social interaction plays an important role in the shopping process. We propose a three-stage dynamic linear model that captures the influence of group discussion on shopper behavior within a hierarchical Bayes framework. The model is tested using a video tracking and transaction dataset from a specialty apparel store. The research reveals that group conversations have a significant impact on the shopper’s department or “zone” choice, purchase likelihood, and spending over time. This group influence is magnified by the size of the group (particularly for zone penetration and purchase conversion), and is also moderated by group composition and cohesiveness. The conversations of mixed-age groups and groups who stay together while shopping have a significant influence on shopper behavior across all three stages, while discussions by adult groups exhibit a marginal carryover effect for purchase conversion. When shoppers have repeated discussions in a specific department, they are more likely to return to and buy from this department, while the cumulative number of discussions in the store drives higher spending levels. We also observe that group shoppers visit more departments than their solo counterparts; and mixed-age groups and solo shoppers are more likely to buy than adults-only or teen groups. This study has important implications for how retailers manage shopper engagement and group interaction in their stores.

Full text

Modeling the Effects of Dynamic Group Influence on
Shopper Zone Choice, Purchase Conversion, and Spending
Xiaoling Zhang*
Shibo Li
Raymond R. Burke
(Forthcoming in Journal of the Academy of Marketing Science)
*Xiaoling Zhang is Assistant Professor of Marketing, School of Management, Shanghai
University of International Business and Economics. Shibo Li is John R. Gibbs Professor and
Professor of Marketing, Kelley School of Business, Indiana University (shili@indiana.edu, 812-
855-9015). Raymond R. Burke is the E.W. Kelley Professor of Business Administration and
Director of the Customer Interface Laboratory, Kelley School of Business, Indiana University
(rayburke@indiana.edu, 812-855-1066). The mailing address for the Kelley School of Business
is 1309 East 10th Street, Bloomington, IN 47405.

2
Modeling the Effects of Dynamic Group Influence on
Shopper Zone Choice, Purchase Conversion, and Spending
Abstract
In many retail contexts, social interaction plays an important role in the shopping process. We
propose a three-stage dynamic linear model that captures the influence of group discussion on
shopper behavior within a hierarchical Bayes framework. The model is tested using a video
tracking and transaction dataset from a specialty apparel store. The research reveals that group
conversations have a significant impact on the shopper’s department or “zone” choice, purchase
likelihood, and spending over time. This group influence is magnified by the size of the group
(particularly for zone penetration and purchase conversion) and is also moderated by group
composition and cohesiveness. The conversations of mixed-age groups and groups who stay
together while shopping have a significant influence on shopper behavior across all three stages,
while discussions by adult groups exhibit a marginal carryover effect for purchase conversion.
When shoppers have repeated discussions in a specific department, they are more likely to return
to and buy from this department, while the cumulative number of discussions in the store drives
higher spending levels. We also observe that group shoppers visit more departments than their
solo counterparts; and mixed-age groups and solo shoppers are more likely to buy than adults-
only or teen groups. This study has important implications for how retailers manage shopper
engagement and group interaction in their stores.
Keywords: shopper marketing, social influence, shopping group, dynamic linear model,
preference revision, hierarchical Bayes model.

3
Introduction
People often work, shop, and consume products in groups, and these groups can have a powerful
influence on their decisions and behaviors (Forsyth 2006; Harmeling et al. 2017). Retail
shopping is generally a social behavior, frequently performed in the company of peers or family
members (Evans, Christiansen and Gill 1996). From the shopper’s perspective, the benefits of
shopping with others include obtaining opinions, having more fun and company, socializing,
staying focused, and relaxing (POPAI 2011). Stores that attract many families, friends, or
shopper groups often perform well, in part because shoppers in these groups tend to shop longer
and spend more money (Kahn and McAlister 1997; Underhill 1999; Woodside and Sims 1976).
Group shoppers engage in greater in-store decision making (Inman and Winer 1998) and may
dynamically revise and adapt their perceptions and attitudes as they hear their companions’
opinions and make concessions to achieve the group’s goals (Aribarg, Arora and Bodur 2002).
Unfortunately, marketers have a limited understanding of the dynamics of group
influence in retail settings. Past research using scanner panel data (e.g., Chiang 1991;
Manchanda et al. 1999) typically has focused on purchase outcomes (i.e., whether, what, when
and how much to buy), and not on the shopping process. Recently, several studies have
examined customers’ in-store shopping patterns and behaviors using RFID or video tracking data
(Hui et al. 2013; Hui, Bradlow and Fader 2009; Zhang et al. 2014). These studies have yielded
valuable insights about the influence of social factors like store crowding, salesperson
interaction, and group discussions, but they focus on the impact of these factors at specific points
in the shopping process, neglecting their potential dynamics over time. For example, groups may
have more or less influence as the customer moves from product search, to alternative
evaluation, to choice. Shoppers may initially consult with group members for their ideas and

4
opinions, but then focus on more personal considerations as the trip progresses. Alternatively,
shoppers may be more likely to trust their own feelings for simple and quick decisions but turn to
their companions when facing more complex and protracted decisions.
Although Zhang et al. (2014) have examined the impact of social influence on shoppers’
in-store touch and purchase behaviors, our work differs from it in a number of ways. First, the
focus of this research is on group shoppers and how they shop and buy. In contrast, Zhang et al.
essentially treat all customers as individual shoppers and examine the impact of social influence
on their touch and purchase behavior. Second, our study investigates the dynamic impact of
within-group conversations on shoppers’ zone choice and purchase behaviors over the course of
the store visit, which is ignored in Zhang et al. Third, the set of causal factors and behavioral
outcomes used is substantially different. While the purchase and path information from Zhang et
al. is incorporated into this study, we obtained and coded a great deal of additional information
on shopper group types and demographic and behavioral characteristics of the group (e.g.,
mixed-age groups, teen groups, group cohesiveness, cumulative discussion frequencies in the
same and different departments). These additional variables generate new insights as moderators
of dynamic group influence. Fourth, the model of the shopping process is also different. In this
study, we examine the dynamic impact of within-group conversations in a three-stage shopping
process: zone choice, purchase likelihood, and purchase amount, while Zhang et al. model touch
and purchase jointly. As a consequence, our research yields a number of new insights and
implications for marketing academics and practitioners.
To the best of our knowledge, there is no study in the marketing literature which
simultaneously models both customers’ within-store shopping process and the dynamic influence
of shopping companions. We propose an individual-level three-stage dynamic linear model of

5
group influence, which captures the impact of each successive conversation that a shopper has
with other members of the shopping party on his or her choice of which departments or “zones”
to visit, whether to make a purchase within each zone, and how much to spend. The proposed
model is implemented in a hierarchical Bayes framework which accounts for individual
heterogeneity. Dynamic group influence is modeled as a function of the shopper’s frequency of
talking with his or her companions during each department or zone visit, the group size, the
cumulative frequency of past discussions, and the interactions between group discussion and
salesperson contact as well as social crowding. Since shopper groups can have different
compositions, the model also allows for the specific type of group (e.g., mixed-age, adult, or
teen) to moderate the dynamics of group influence on the shopper’s baseline shopping/buying
tendency. In order to correct for self-selection sample bias (consumers choose whether to shop
together or alone), we use the Heckman correction method to address the issue. Further, we take
into account the potential endogeneity in shoppers’ navigation paths and in-store activities.
We estimate the model using a novel dataset that combines video tracking and point-of-
sale transaction data to capture the shopper’s shopping process and purchase outcomes. Unlike
traditional scanner panel and survey data, the video data capture the shopper’s navigation path
through the store, within-trip decisions (e.g., touching products, visiting a dressing room), and,
most importantly, group composition and shopper interactions. In contrast to RFID-based
customer tracking research (e.g., Hui, Bradlow and Fader 2009; Hui, Fader and Bradlow 2009a,
b), the video technology measures the actual path of each shopper who enters a store (not just
those who use a shopping cart or basket), shopper demographics (i.e., gender, age, and ethnicity),
within-group discussions, and salesperson contacts.
The findings indicate that group influence—measured by within-group talk frequency,

6
group size, and cumulative talk frequency within and across zones—has a significant impact on a
shopper’s baseline shopping/buying tendency, and hence the dynamics of a shopper’s zone
choice, purchase, and spending. This influence extends beyond the visit to a specific department
or product category. The dynamic group influence is greater in mixed-age groups across three
stages, while only marginally greater for adult groups in stage 2 (purchase conversion), and when
group members stay together for most of the store visit. Cumulative talk frequency in the same
department is positively related to zone penetration and purchase likelihood, whereas overall
cumulative talk frequency increases spending. We also observe that group shoppers walk slower
and visit more departments than their solo counterparts; mixed-age groups and solo shoppers are
more likely to buy and have larger baskets than adults-only or teen groups.
This study contributes to group influence research in the shopper marketing literature in
several ways. It is the first empirical study examining the dynamics of group influence on a
customer’s tendency to shop and buy, and it reveals that shoppers’ conversations have a
significant impact on their zone penetration, purchase likelihood, and purchase amount during
the course of the store visit. The dynamic influence is moderated by the group’s composition and
cohesiveness, and it has a differential impact on the three stages of the shopping process.
Furthermore, we find significant differences between various types of shopper groups, and
between group and solo shoppers, in their within-store shopping dynamics and the differential
impact of marketing, social, individual, and environmental factors. The findings can help
retailers to make more informed decisions in communicating with group shoppers and devising
tailored promotions to improve personal selling efficiency and store revenues.
The remainder of this paper is organized as follows: First, we review the literature and
propose a conceptual framework describing how group interaction influences a shopper’s

7
intrinsic shopping and buying tendency over time. Next, we develop a three-stage dynamic linear
model that captures the influence of group discussion on shopper behavior, followed by a
description of the data and variables, and the empirical results. Finally, we conclude with a
discussion of the key insights, managerial implications, and future research directions.
Group influence and decision making
Two research streams related to our study—group influence in a retail shopping context
and empirical models of group decision making—are summarized in Table 1. The first stream
includes several studies of how families shop, revealing the importance of conversations and
negotiation between family members in purchase decisions (e.g., Atkin 1978; Gram 2015; Rust
1993). Parents and children often interact positively during the shopping process, working
together in a collaborative manner to align their goals. Darian (1998) finds, for example, that a
parent and child are more likely to buy when a salesperson addresses both of their needs.
[Insert Table 1 about here]
There have also been many studies on the topic of “purchase pals,” defined as
“individuals who accompany buyers on their shopping trips in order to assist them with their on-
site purchase decisions” (Hartman and Kiecker 1991, p. 462). Research indicates that relatively
inexperienced consumers who lack confidence in their ability to evaluate products and brands are
more likely to shop with purchase pals (Bell 1967; Furse, Punj and Stewart 1984; Midgley 1983;
Mangleburg, Doney and Bristol 2004). Shopping with a pal provides social support and
guidance, reducing perceived risk (Kiecker and Hartman 1993) and increasing the buyer’s
confidence (Kiecker and Hartman 1994).
These studies reveal that group interaction can encourage buying, but it also has the
potential to reduce purchase likelihood. For example, Bell (1967) discovered that customers

8
accompanied by friends or relatives are less likely to make an automobile purchase. Luo (2005)
reports that the presence of parents in a shopping situation can activate a consumer’s sense of
responsibility, decreasing impulse purchasing, whereas the presence of friends tends to
encourage spontaneity, increasing impulsive buying. Yim et al. (2014) find that co-shopping
with others in a superstore boosts purchases, especially when shoppers are impulsive, while
Inman, Winer and Ferraro (2009) and Page et al. (2018) do not find a significant relationship
between shopping with others and purchases. This research stream suggests that the influence of
groups depends on their demographic composition, the knowledge, confidence and impulsivity
of the buyer, and the type of product category (e.g., planned vs. impulse, hedonic vs. utilitarian).
However, these studies largely ignore the dynamics of group influence and how social
interactions over the course of a store visit affect shoppers’ navigation path and buying.
The second research stream focuses on developing empirical models of group decision
making (e.g., Corfman and Lehmann 1987; Chandrashekaran et al. 1996; Arora and Allenby
1999; Su et al. 2003; Yang, Narayan and Assael 2005; Yang et al. 2010). For example, Aribarg,
Arora and Bodur (2002) used a repeated conjoint-based choice task to measure the degree of
preference revision and concession in group decisions. They find that these responses varied
across product attributes, individuals, and product categories. Extending this research, Aribarg,
Arora and Kang (2010) developed a survey- and conjoint-based methodology to estimate parent–
teen joint preference for cell phones from individual data rather than actual group discussions.
Our study differs in several ways from research in this stream. First, we classify groups
based on observed in-store behavior, and we measure and model the influence of interpersonal
interactions on group-shopper zone choice and purchase activities. Past studies have typically
used experimental or survey data with predefined groups in a laboratory setting. Second, we

9
capture the dynamics of consumer preference updating over the course of a store trip across
multiple department visits, while prior research only allows for one-time consumer preference
revision. Third, our research looks at products (apparel and fashion accessories) that are
purchased and consumed by individuals under group influence, while prior studies have focused
on durable goods (e.g., PCs, cell phones), which are often jointly purchased and consumed.
Finally, past modeling research has focused on the joint group decision process and how
individual members’ preferences are revised to reach consensus, while we focus on the dynamic
influence of groups on individual shopper’s zone choice and buying decisions.
Conceptual framework
In order to investigate the dynamic influence of group discussions over the course of a
store visit, we divide the shopping process into three decision stages, where the shopper chooses:
(1) which department or zone to visit, (2) whether to buy from the zone, and (3) how much to
spend. This corresponds to three commonly used retail performance metrics: customer attraction
or traffic, purchase conversion rate, and sales per transaction (Lam et al. 2001; Hui, Bradlow,
and Fader 2009). Next, we discuss the dynamic influence of group interaction on the shopper’s
intrinsic preference revision, and the moderating roles of group composition and size, social
context, and shopper demographics. Then we examine other potential influences, including the
shopper’s in-store behavior, marketing variables, and product category characteristics. The
conceptual framework is presented in Figure 1.
[Insert Figure 1 about here]
Dynamics of group influence on shopper preferences and behavior
As suggested by the field theory of group dynamics (Lewin 1951), an individual’s
behavior is determined by the interaction of the person and the physical and social environment.

10
When a shopper walks into a retail store, he or she encounters a complex and dynamic setting
with a variety of merchandise, salespeople, displays, promotional activities, and other customers
(Underhill 1999). The shopper begins the trip with an initial intrinsic shopping preference and
tendency to buy (pre-discussion preference). As the shopper walks through the store and
examines the merchandise, he/she may interact with fellow shoppers who offer
recommendations, express opinions, or provide feedback, and this leads to an updated preference
that may positively influence his or her zone choice, buying probability, and purchase amount.
Previous research has shown that discussion is at the heart of group decision making,
where people seek out and process relevant information from others (Forsyth 2006; Kowert
2012). Discussions increase the amount of information exchanged, improve memory for the
shared information, and encourage more thorough processing of the information (Hinsz, Tindale
and Vollrath 1997; Larson and Christensen 1993; Propp 1999). Conversations can also make the
shopping group’s interpersonal relationships and group identity salient, exerting normative
influence on preferences (Harmeling et al. 2017).
During group interactions, shoppers may change their preferences or preference certainty
at any point in time (Stasser and Davis 1981). Discussions remind them of information that they
may have momentarily forgotten or give them new information, causing them to reevaluate their
preferences (Stasser 1988). This preference revision can take place repeatedly after each
conversation during the store visit, until a decision is reached (Aribarg, Arora and Bodur 2002;
Aribarg, Arora and Kang 2010). We conjecture that shoppers dynamically revise their intrinsic
zone choice and purchase tendencies after each zone visit occasion when there are group
interactions, and these group influences may have positive carryover effects on a shopper’s
subsequent zone choice, purchase likelihood, and spending. In this study, we did not have the

11
content information of the group discussions due to privacy concerns. Instead, we capture group
influence using the observed frequency of within-group interactions, which are a widely used
measure for tie strength (Granovetter 1973; Aral and Walker 2014). Formally, we propose:
H1: Within-group discussion frequency has a positive and dynamic impact on shoppers’ zone
choice, purchase likelihood, and purchase amount.
Moderators of dynamic group influence
The degree to which group members will influence a shopper’s intrinsic tendency is
likely to depend on several factors. The first important moderator is the shopping group size.
According to Social Impact Theory (SIT; Latané 1981), the degree of social impact is
determined by three factors: (1) the strength of the source of impact (how much influence, power
or intensity the target perceives the source to possess), (2) the immediacy of the event (how
recently the event occurred), and (3) the number of sources exerting an influence on the target.
There is more social impact when higher status individuals are the source, when the action is
more immediate, and when there are a greater number of sources. A larger shopping group will
have more sources of impact, and therefore a greater impact on a shopper’s dynamic zone choice
and purchase preference revision (Latané and Wolf 1981), so we predict:
H2: The dynamic impact of within-group discussion frequency on shoppers’ zone choice,
purchase likelihood, and purchase amount is magnified by the shopping group size.
The second is the demographic composition of the group. Demographics have been
shown to play an important moderating role in both the marketing and group decision-making
literatures (Childers and Rao 1992; Luo 2005; Seibold, Meyers and Sunwolf 1996). For example,
mixed-age (intergenerational) shopping groups appear to have a greater influence on decisions
about private necessities like toothpaste selection, while peer-based groups have more influence

12
on decisions about conspicuous luxuries (Childers and Rao 1992). Similarly, the presence of
parents can decrease impulse purchasing, while peers can stimulate impulsive buying (Luo
2005).
For this research, we classify shopper groups into mixed-age groups and peer groups,
with the latter further divided into adult groups and teen groups. Both mixed-age and peer groups
tend to have primary and strong relationships with strong ties according to Social Impact Theory
(Latané 1981). Strong ties constitute a base of trust, reducing resistance and providing comfort in
the face of uncertainty (Krackhardt 1992). Strong ties are associated with greater social influence
and cooperation (Coleman 1988). In a retail context, the mixed-age segment of shoppers will
largely consist of parent–child groups. In this case, there is likely to be a strong sense of group
identity and normative influence due to family bonds. Harmeling et al. (2017) report that, while a
group's informational influence diminishes over time, group-identity/normative influence grows.
Gender is also an important factor in group shopping. POPAI (2011) reports that
shoppers accompanied by female companions were three times as likely to spend more than
those accompanied by males. Wood (1987) suggests that female group interaction facilitates
performance on tasks requiring positive social activities (e.g., friendliness, agreement), which
would help to build consensus. We therefore predict that the presence of a female in a group will
strengthen the group’s dynamic influence on shopper behavior.
Shopper groups engage in “swarming” activities (Leykin and Tuceryan 2007). Unlike
solo shoppers, they usually arrive and leave together as a group, and connect with each other
periodically during the store visit. We use “group cohesiveness” to refer to whether members go
in separate directions or stick together after entering the store. When group members stay
together, communication is more likely, and social influence is likely to be higher due to spatial

13
immediacy and closeness according to Social Impact Theory (Latané 1981). It also signals that
they have common interests and goals or “homophily” (McPherson, Smith-Lovin, and Cook
2001), enhancing the dynamic impact of within-group discussion frequency. Therefore, we
predict:
H3: The dynamic influence of within-group discussion frequency is moderated by group
composition and cohesiveness. More specifically, the dynamic influence is greater in (a)
mixed-age groups, or (b) groups with females present, or (c) groups where the members stay
together most of the time during the trip.
The third important moderator is the social environment in the store. A common social
factor is the presence of other shoppers (i.e., crowding), which can lead to negative feelings and
coping strategies (Harrell, Hutt and Anderson1980; Arnold et al. 2005) and potentially
discourage shoppers from buying. Another is personal selling. The persuasive impact of the sales
associate may be enhanced or reduced by shopping companions. On the one hand, shoppers in
groups may be looking for advice and therefore be more susceptible to social influence from
salespeople (Bell 1967; Goff, Bellenger and Stojack 1994). On the other hand, group shoppers
have a certain degree of self-confidence (Bell 1967) and engage in relatively self-reliant
shopping activity. Hence, they may be less responsive to salesperson contact. Therefore, it is an
empirical question of the exact moderating impact of salesperson contact on the focal shopper’s
dynamic preference revision over the course of the store visit.
Cumulative discussion frequency
Group shoppers may communicate with each other on multiple occasions during their
store visit. We account for the frequency of discussions during the current zone visit as well as
for the cumulative frequency up until the last zone visited. We further divide a shopper’s overall

14
cumulative discussion frequency into repeated discussions taking place in the same
department/zone versus those across different departments/zones. The former is more product
category specific and hence may result in positive effects on zone choice and purchase in the
zone, while the latter captures the overall level of group involvement, engagement, and interest
(Underhill 1999). The overall group involvement stimulates need recognition (Puccinelli et al.
2009), communication within the retail channel, and the time and amount of money spent (Flynn
and Goldsmith 1993). Therefore, we believe that zone-specific cumulative discussions will
encourage repeated visits and purchase conversion in a zone, while cumulative discussions in
different zones will drive consumer spending. Thus:
H4: Cumulative discussion frequencies have differential dynamic impact across three stages.
Specifically, cumulative discussions in the same zone are positively related to zone choice
and purchase probability, while cumulative discussions in different zones will have a more
positive impact on purchase amount.
Other influential factors
Prior research has shown that a shopper’s purchase likelihood during the course of a store
visit is affected by several other factors, including the shopper’s pattern of movement in the store
and interaction with products, marketing promotion, and product category characteristics (Zhang
et al. 2014). This is also consistent with the field theory of group dynamics (Lewin 1951;
Hackman and Morris 1975). Therefore, we incorporate these variables into the proposed model.
In addition, we control for the potential endogeneity in shopper self-selection and some of the
shopper behavior variables, as well as individual shopper heterogeneity. Table 2 summarizes the
explanation and operationalization of variables incorporated in the conceptual framework.
[Insert Table 2 about here]

15
Model setup
Shopping in a retail store is a dynamic process that occurs over time and space, where
customers’ interactions and experiences have a cumulative effect on their navigation path and
buying behavior. To account for the dynamics of shopper behavior over time, we propose a
three-stage dynamic linear model (DLM) to capture the dynamic influence of group
conversations on shopper zone choice, purchase conversion and purchase amount over the
course of the store visit. In the marketing literature, DLM has been successfully employed to
study the dynamic effects of the marketing mix on brand sales (Ataman, Mela and Van Heerde
2008; Ataman, Van Heerde and Mela 2010).
One important methodological issue that we need to address is shopper self-selection
into non-randomly selected samples, such that consumers who value group discussions are
more likely to go shopping with others. To control for this, we adopt the widely used Heckman
correction method (Heckman 1979), with details presented in Web Appendix 1.
Stage 1: zone choice
In the first stage of the store visit, a shopper selects a zone or department to visit. We use
a multinomial probit model to capture shopper i’s zone choice at zone visit occasion t, with the
utility of the jth zone (J = 7) determined by the types of products sold in that zone or department
(e.g., new arrivals, clearance), whether the items are for men or women, and the inverse mills
ratio based on the Heckman correction method.1
01 2 3 4
5
, 1, 2,...,6,()
jit ji t i jit i jit i jit i jit
i jit i jit it
j
U GenderApp NewArrival Clearance Accessories
Fema IMRleSide jZ
e
ββ β β β
βrs φ e
=++ +
+ +
+
+=
(1)
1 Although we have shopper path and social influence (e.g., crowding) information, they are not zone-specific and
hence will not impact the shopper’s zone choice (Cameron and Trivedi 2005). Therefore, we do not include these
variables in the zone choice utility function. However, we do incorporate them into the shopper’s purchase
conversion and amount equations.

16
For identification purposes, the least frequently visited zone (j = 7) is used as the base choice
with its utility normalized to zero, and the errors are assumed to be normally distributed such that
~ (0, ),
it
N
e
∑
where
16
[ ... ]
e ee
′
=
is a 6 x 1 vector.
As discussed in the conceptual framework, a group shopper’s intrinsic preference or
baseline utility of zone choice is dynamically revised in response to discussions with other group
members as he/she navigates through the store (Chandrashekaran et al. 1996). Applying a DLM,
we allow the shopper’s baseline utility
0ji t
β
to be dynamically revised after each zone visit as:
0 0 1 01 2 1 3 1
4 15 16 1 1
71 1
__
,
ji t ji ji ji t it it i
it it it it
it it jit
TalkFreq TalkFreq GroupSize
CumuTalk Same CumuTalk Diff TalkFreq SalesContact
TalkFreq Crowding
β a aβ a a
a aa
ad
−−−
− −− −
−−
=++ + ×
+ + +×
+ ×+
(2)
where
0ji
a
captures the shopper’s intrinsic preference for zone j.
1ji
a
denotes the carryover effect
of group influence from the last zone visit. Group interaction is measured by the number of times
shopper i talks with his/her companions in the same group at zone visit t (TalkFreqit), and the
parameter
2
a
estimates the impact of these discussions on the shopper’s revised intrinsic
preference (H1). The parameter
3
a
captures the moderating role of group size (H2). The
parameters
4
a
and
5
a
measure the cumulative effects of within-group conversations in the same
zone, and across all other zones, respectively (H4). The terms
it it
TalkFreq SalesContact×
and
it it
TalkFreq Crowding
×
capture the interaction between discussion frequency and two social
influence factors (salesperson contact, store crowding), respectively.
jit
d
is an error term such
that
2
(0, )
jit j
N
ds

.

17
As we discussed in the conceptual framework, we allow a group’s demographic
composition (age, gender, and ethnicity) and cohesiveness (
i
StayTogether
) to moderate the
dynamic influence of group discussion frequency (H3,
1
λ
-
6
λ
)2:
01 2 3 4 5
6
, where ~ 0,( ).
ji i i i i i
i ji i
Female Caucasian MixAge AdultGroup FemaleGroup
StayToge MVther N
a
a λλ λ λ λ λ
λ nn
=++ + + +
++ S
(3)
Note that there are spatial constraints for the shopper’s zone choice such that he/she can
only choose to enter those zones adjacent to the current zone for his/her next move. Therefore,
we take this constraint into account, and the zone choice of shopper i at time t conditional on the
choice set of adjacent zones is determined by:
1
7
argmax | ),
(
j jitijitt
Zo Cne U
=
=
(4)
where
jit
C
is the choice set of adjacent zones to the zone visited at time t -1.
Stage 2: purchase conversion
Conditional on the zone choice in Stage 1, the shopper decides whether to purchase an
item in a zone. The purchase probability is influenced by the shopper’s intrinsic purchase
preference, navigation path in the store, interactions with the merchandise, social influence from
in-store crowding, the retailer’s marketing activities, and the characteristics of the product
category (Lewin 1951; Zhang et al. 2014; see Figure 1). Let
it
U′
be shopper i’s purchase utility
at zone visit t, the utility function can be written as:
2 In order to avoid estimation of too many parameters in Eq. 2, we only allow the intercept and carryover effect
(
01
,
ji ji
aa
) to be zone- and individual-specific and moderated by group composition and cohesiveness. We also
tried the model with full zone and individual heterogeneity for all parameters in Eq. 2, but its performance was much
worse and unstable.

18
01 2 3 4 5
2
678 9
10 11 12
it i t i it i it i it i it i it
i it i it i it i it
i it i it i
U CumuSinu GenderApp FemaleSide Distance Speed
Crowding Crowding SalesContact SideSale
TouchFreq Dressroom NewArrival
ββ β β β β
βββ β
βββ
′′′ ′ ′ ′ ′
=+ + + ++
′′′ ′
++ + +
′′′
++ +
13 14 15 1 () ,
it
i it i it i it ti i
Clearance Accessories NumZones IMR Z
e
rββ β esφ
′
−
′′ ′ ′
++ + +
′
+
(5)
where
0
it
β
′
captures the shopper’s intrinsic preference for purchase which is time-specific and
affected by the dynamics of within-group interactions.
it
e
′
is the error term such that
2
1
~ (0, )
it
N
es
′
, and
2
1
s
is set to be 1 for identification purpose.
The first set of explanatory variables measure the shopper’s within-store shopping path:
(1) the cumulative sinuosity or curvature of the path, denoted by
it
CumuSinu
, (2) whether the
shopper stays on the side of the store matching his/her gender (
it
GenderApp
), (3) whether the
shopper shops on the female side of the store (
it
FemaleSide
), (4) the distance covered within a
zone (
it
Distance
) in feet, and (5) traveling speed within a zone (
it
Speed
) in feet per second. The
coefficients
1i
β
′
-
5i
β
′
estimate the impact of these shopping path variables on shopper i’s
purchase utility.
The second set of covariates measure social influence of in-store crowding and
salesperson contact.
it
Crowding
is measured by the number of people present in a zone when the
shopper enters the zone; with
6i
β
′
capturing shopper i’s responsiveness to the crowding
condition, and
7i
β
′
capturing the potential nonlinear effect of crowding (
2
it
Crowding
).
it
SalesContact
represents whether a salesperson talks to shopper i during zone visit t, and
8i
β
′
measures the shopper’s responsiveness to personal selling efforts.
The third and fourth sets of independent variables measure the retailer’s marketing
activities, and the shopper’s interaction with products during the store visit, respectively.

19
it
SideSale
is a dummy variable which denotes whether the retailer is holding a sidewalk sale.
10i
β
′
is the impact of the shopper’s product interactions or touch on the purchase decision.
11i
β
′
captures the impact of the shopper’s dressing room visit.
The last set of covariates are three product category variables:
it
NewArrival
,
it
Clearance
,
and
it
Accessories
, which denote whether the current product category being shopped by
customer i is new arrivals, clearance items, or accessories, respectively.
12
i
β
′
-
14
i
β
′
reflect the
product category effect on the purchase decision. Some categories have higher conversion rates
than others, such as fashion accessories, cosmetics, and jewelry, because the products are more
engaging and entertaining (Underhill 1999). Number of zones visited up to t-1 is included to
capture the state dependence. IMR is also included to correct for the shopper self-selection bias.
Similar to the first stage model, we allow the shopper’s intrinsic purchase preference
0it
β
to vary from one zone visit to the next and it can be written as a function of within-group talk
frequency and its interaction terms with group size, crowding and salesperson contact:
0 0 1 01 2 3 4 1
5 16 7
_
_,
i t i i i t i it i it i i it
i it i it it i it it it
TalkFreq TalkFreq GroupSize CumuTalk Same
CumuTalk Diff TalkFreq SalesContact TalkFreq Crowding
β a aβ a a a
aa a d
−−
−
′ ′ ′′ ′ ′ ′
=++ + × +
′′ ′ ′
+ + × + ×+
(6)
where
2
(0, )
it
N
d
ds

.
Further, since the impact of dynamic influence of within-group talk frequency (
i
a
′
) may be different for groups with different demographic composition (age, gender,
and ethnicity) and cohesiveness (
i
StayTogether
). We account for such a moderating
impact as follows:
01 2 3 4 5
6
,
i i ii i i
ii
Female Caucasian MixAge AdultGroup FemaleGroup
StayTogether
aλλ λ λ λ λ
λn
′′′ ′ ′ ′ ′
=++ + + +
′′
+ +
(7)

20
where the error term
i
n
′
is assumed to follow a multivariate normal distribution of MVN (0,
a
′
S
).
We denote the zone purchase decision by Bit, where Bit equals 1 if shopper i buys at zone
visit t, 0 otherwise. Given the above setup, shopper i’s purchase decision Bit is captured by a
binary probit model:
1, if 0
0,
it it
U
otherwis
Be
′
=>


(8)
Stage 3: purchase amount
In the third stage, we model a shopper’s purchase amount (in dollars) conditional on
zone purchase in Stage 2 using a log linear model. To capture the state dependence, we include
the number of zone purchases up to t-1 (
1
it
NumBuy
−
) and other independent variables are
similar to those in the purchase utility function in Stage 2. Thus, we have
01 2 3 4 5
2
6 7 8 9 10
11 12 13
it t it it it it it
it it it it it
it it it
LogAmount CumuSinu GenderApp Distance Speed Crowding
Crowding SalesContact SideSale TouchFreq Dressroom
NewArrival Clearance Accessories
ϕϕ ϕ ϕ ϕ ϕ
ϕ ϕ ϕϕ ϕ
ϕ ϕϕ
=+ + + ++
+ + ++ +
+ ++ +
14
15 1 3
() ,
it
it it
FemaleSide
NumBu IMR Zy
ζ
ϕ
ζrs φ
ϕ
−
+++
(9)
Given the limited number of observations at the third stage, we allow the intercept or
base amount (
0t
ϕ
) to be time-varying, but not individual specific. The base amount is updated
due to dynamic group influence as follows:
0 0 101 2 3
4 15 1
67
89
__
tt t t
tt
t t tt
tt
AvgTalkFreq AvgTalkFreq GroupSize
AvgCumuTalk Same AvgCumuTalk Diff
AvgTalkFreq SalesContact AvgTalkFreq Crowding
AvgTalkFreq Female AvgTalkFreq Caucasian
ϕ k kϕ k k
kk
kk
kk
−
−−
=++ + ×
++
+ ×+ ×
+ ×+ ×
+
10 11
12 13
tt
t tt
AvgTalkFreq MixAge AvgTalkFreq AdultGroup
AvgTalkFreq FemaleGroup AvgTalkFreq StayTogether
kk
k kx
×+ ×
+ ×+ ×+
(10)

21
where the explanatory variables are average terms pooled across individuals, and
2
(0, )
t
N
x
xs

.
We adopt a hierarchical Bayes framework to estimate the proposed model. The details on
shopper self-selection, heterogeneity, endogeneity and the MCMC algorithms are presented in
Web Appendices 1 and 2.
Data
Customer tracking data were collected by installing a six-lens panoramic video camera on
the ceiling of a specialty apparel store featuring men’s and women’s clothing and accessories.
The store was located in the Midwest region of the United States. We collected and analyzed the
video and point-of-sale transaction log data for three days in January 2006, after the conclusion
of the holiday shopping and returns period. Each dataset is approximately one hour, from 4:00
p.m. to 5:00 p.m., which is the peak period of customer traffic during the day.
The analysis of the video consists of four steps: First, a computer program is used to track
all visible customer activities including store entry, navigation through the store, and exit. Each
shopper is assigned a unique identification number. Second, a research assistant manually
connects the “broken tracks” that occur when the computer program is unable to track
individuals because they are (a) obscured by shelf fixtures or other people, (b) visiting the
dressing room or stock room, and/or (c) leaving and later reentering the store. Third, a researcher
distinguishes between shoppers and sales associates, and manually codes each shopper’s
demographics and interactions with other people (salespeople, group members) and products
(touching, trying on, purchasing). When there is an apparent pause between two conversations
(e.g., longer than 5 seconds), the count of customer discussions is increased by one. Fourth, the
researcher identifies members of each shopping group based on their common store entry and

22
exit times and their shared activities during the shopping process. Each group is assigned a
unique group ID number.
Figure 2A shows an example image from the panoramic video camera. Objects tracked
by the computer vision software are shown in blue, and the green ellipses represent projected
models of each human body. Each customer is labeled with a unique ID number, and customers
with the same colored label are from the same group. For example, customers #11 and #13
(green labels) and #29 and #30 (yellow labels) entered and exited the store together, and they are
tracked as two separate groups.
[Insert Figure 2A and 2B about here]
Looking from the store entrance, men’s products are positioned on the left side and
women’s on the right, with accessories and checkout areas in the middle of the store. New and
seasonal products are positioned on the front or “lead” fixtures, popular staple items (e.g., knit
shirts, graphic tees, khakis, and denim pants) reside in the middle, and sale and clearance items
are positioned in the back of the store. For purposes of analysis, the store is divided into eight
zones: women’s lead zone, men’s lead zone, women’s middle, men’s middle, accessories,
women’s sales, men’s sales, and checkout (excluded from analysis due to no products for sale).
Customers’ shopping paths and activities are tracked at the zone level. Overall and zone-specific
conversion rates are calculated by combining traffic data with transaction log data (Burke 2005).
Summary statistics for shoppers’ zone penetration, purchase conversion rate, and spending are
reported in Figure 2B.
The dataset consists of a total of 175 customers: 71 solo shoppers (41%), who visited 449
zones, and 104 group shoppers (59%), who visited 831 zones. Of the 46 tracked groups, one is
composed of five people, nine have three people, and 36 are two-person groups. There are 33

23
shoppers in mixed-age groups with 310 zone visits, 37 shoppers in adult groups with 223 zone
visits, and 34 shoppers in teen groups with 298 zone visits.
For model validation purposes, we use data from the first two days as the estimation
sample with 618 observations and keep the last day’s data as the holdout sample with 213
observations. To test for multicollinearity, we checked the correlations among the independent
variables (see Web Appendix 3) and the variance inflation factor. Distance, FemaleSide, and
Accessories are dropped in Eq. 12 due to high correlations. For the remaining variables, the
highest correlation is -.44, which is low, and the variance inflation factor is 2.89, which is much
less than 10. We therefore conclude that multicollinearity is not an issue (Belsley, Kuh, and
Welsch 1980).
Empirical results
Summary statistics for the major variables in the model are presented in Table 3, and
profiles of each of the demographically defined shopper segments are given in Table 4. Looking
across the four segments, we observe important differences in shopper behavior. The mixed-age
group shoppers are primarily adult women with children and represent 19% of customers, but
they visit the most departments, spend the most time shopping, and are the most likely to make a
purchase in the store (54.5%) and in each department they visit (9.42%), with an average basket
size of $24.35. This is consistent with POPAI’s (2011) finding that families tend to shop more of
the store.
[Insert Tables 3 and 4 about here]
Solo shoppers are typically adults and females (84.6% and 65.5%, respectively) and are
the largest proportion of customers (41%). They visit less than half as many departments as the

24
mixed-age groups and have a smaller average basket size ($12.82), but their average purchase
conversion rate for each of the zones visited (9.35%) is about the same as for families.
Adults-only groups are more likely to stay together while shopping, but they visit
relatively few departments and have the lowest store-level purchase conversion rate (24.3%).
Compared to the other groups, they are more likely to converse and touch the products in each of
the departments they visited.
Teen group shoppers are mostly male (63.2%) but are often accompanied by females
(76.1%). Teens appear to prefer shopping with their peers rather than alone, as more than 45%
of group shoppers are teenagers, but only 15.4% of solo shoppers are teens, consistent with the
findings of Tootelian and Gaedeke (1992). Teen groups walk the slowest and shop many
departments (especially new arrivals), but they spend the shortest amount of time in each
department (26 seconds on average). They don’t talk or stay together as often as other groups,
and they touch and buy the fewest items (with an average basket size of $5.57). These findings
are contrary to those of Luo (2005), who observed that peers had a greater urge to purchase when
they imagined peers being present rather than family members. The differences are due to lower
discretionary income for teenagers, and the fact that they usually come in bigger party size, and
average spending per person will therefore be smaller.
The proposed model is compared with a benchmark and a nested model in each stage,
and the statistics are presented in Web Appendix 4. The results demonstrate that the proposed
models have better model fit and prediction performance, and it is important to account for the
dynamics of group influence using DLM, spatial constraints in zone choice, and endogeneity in
the model. Therefore, we will focus on the estimation results from the proposed models, which
are shown in Tables 5 and 6.

25
[Insert Tables 5 and 6 about here]
Hypothesis testing
The dynamic group influence section of Table 5 shows that concurrent group discussions
(talk frequency) have a significant positive impact on zone choice (α2 = .194) and purchase
likelihood (
2
i
a
′
= .547) but not the amount spent, supporting H1 for the first two shopping stages
but not the third. The size of the group magnifies this discussion effect for both zone choice (
3
a
= .043, marginally significant) and purchase conversion (
3i
a
′
= .221), so H2 is likewise supported
in the first two stages. This is consistent with the findings of Narayan, Rao and Saunders (2011),
which suggest that more peers are associated with greater preference revision. We did not find
support for H1 and H2 in the Stage 3 model, perhaps because group influence tends to impact
shoppers’ decisions at earlier stages in the shopping process.
H3a predicts that the dynamic impact is greater in mixed-age groups, and Tables 6-1, 6-2,
and 6-3 show that it’s supported across all three stages (
3,1
λ
=.031;
3
λ
′
= .088, .083, .012, .046;
10
k
= .656), though marginally supported in Stage 1. Dynamic group influence is also higher for
adult groups (
4
λ
′
= .125), but the effect is only marginally significant, perhaps because adults are
more independent. H3b states that the dynamic influence is greater in groups with females, and
this is marginally supported in Stage 1 (
5
λ
= .031) and supported in Stage 2 (
5
λ
′
= .224), but not
in Stage 3. H3c predicts that the dynamic impact is enhanced by staying together, and it’s also
supported across all three stages, although only marginally supported in Stage 1 (
6,1
λ
= .032;
6
λ
′
=
.191, .123, .005, .120;
13
k
= .542). The group’s affinity or cohesiveness, as indicated by whether
shoppers stay together, also seems to strengthen the group’s influence (Dion 2000; Hogg 1992).

26
H4 proposes that zone choice and purchase likelihood are more positively related to the
frequency of discussions that occur within the same department/zone, which is supported in the
first two stages shown in the dynamic group influence section of Table 5 (
4
a
= .016, marginally
significant;
4i
a
′
= .278), and overall cumulative discussion frequency across different
department/zones has a positive impact on purchase amount, which is supported in the third
stage (
5
k
= .252). The results of hypothesis testing are summarized in Table 7.
[Insert Table 7 about here]
As shown in the dynamic group influence section of Table 5, we also observe a
significant, positive carryover effect of group influence across all three stages (
1
ji
a
= .017,
1i
a
′
=
.238,
1
k
= .042). The shopper’s prior discussions with companions contribute positively to his or
her visit and purchase utilities, and spending tendency during the current zone visit and this
carries over to subsequent zone visits. While customers are less likely to come back to a zone
where they had few prior discussions (
5
a
= -.046), store-wide discussions do increase their
spending. A shopper’s conversations with companions boost zone penetration when the store is
less crowded (
7
a
= -.229), but the impact of crowding is mitigated by group discussions in later
stages. This shows that while the presence of strangers reduces group influence initially, over
time group discussions shield members from unwanted social interaction from outside of the
group.
Tables 6-1, 6-2, and 6-3 summarize the heterogeneity estimation results for the dynamic
equations. The carryover effect of group influence in zone choice is marginally greater when the
shopper is female (λ1,1 = .030), Caucasian (λ2,1 = .031), and there are women present in the group
(λ5,1 = .031). In the purchase stage, these customers are also more susceptible to influence in the

27
carryover, discussion frequency, when they are in a larger shopping group, and when they return
to departments where they have had prior discussions. Shopper demographics and group
composition therefore appear to play an important role in dynamic group influence, consistent
with our conceptual framework.
Other estimation results
With respect to the impact of zone and category variables on zone choices, the upper
panel of Table 5 shows that shoppers are attracted to the clearance and promotional sections (
3i
β
= .318, marginally significant) and the men’s side of the store (
5i
β
= -.270, marginally
significant). While there were many female customers, mother/son and husband/wife groups
often shopped on the male side. Group shoppers are more likely to buy when they cover more
distance (
4i
β
′
= .005) and take their time in each zone (
5i
β
′
= -1.252); however, they spend more
if they walk faster (
4
ϕ
= .280) and follow a more circuitous path (
1
ϕ
= .133), probably visiting
more departments and making multiple purchases (
15
ϕ
= .598). High cumulative sinuosity may
reveal more general or diffuse goals and more browsing activity, which leads to greater
spending, while a straighter path may indicate more specific shopper requirements. Shoppers are
more likely to buy when they have sufficient personal space (
6
i
β
′
= -.359, marginally significant),
which provides a more comfortable shopping environment (Underhill 1999).
Both purchase and spending decisions are positively influenced by salesperson contact
(
8i
β
′
= .995, marginally significant;
7
ϕ
= .704), as salespeople cross-sell and up-sell merchandise
to increase basket size. Group shoppers are less interested in sidewalk sales (
9i
β
′
= -.697),
although the effect is only marginally significant. They are more likely to buy and spend more
when they pick up and examine the merchandise (
10i
β
′
= 1.535;
9
ϕ
= .179, marginally

28
significant) and try on items in the dressing room (
11i
β
′
= 2.056,
11
ϕ
= .915). However, product
interactions have a more significant effect on purchase conversion.
The coefficients on the IMR are significant across all three stages, showing the
importance of correcting for shopper self-selection bias. Finally, the more zones visited, the
greater the likelihood of making a purchase (
15i
β
′
= .117, marginally significant).
Discussion
Social interaction plays an important role in the retail shopping experience, and recent
innovations in video-based customer tracking allow marketers to quantify its impact on shopper
behavior. Toward this end, we developed an individual-level three-stage dynamic linear model of
group influence which estimates the impact of each successive conversation that a shopper has
with members of the shopping party on his or her zone choice, purchase conversion and purchase
amount during the course of the store visit. The proposed model, which allows for dynamic
preference revision and potential endogeneity, outperformed nested benchmark models at each
stage on both model fit and predictive performances in a holdout sample.
An analysis of the shopping patterns of 175 customers visiting 1,280 departments or
“zones” in the store revealed four distinct shopper segments: the fast and focused solo shoppers
(41%), the big trip, big basket mixed-age-group shoppers (19%), the connected and
conversational adult-group shoppers (21%), and the independent and frugal teen-group shoppers
(19%). These segments differ in how they shop the store, what they buy, and how they interact,
creating opportunities for targeted marketing activities. For example, the mixed-age-group
shoppers appear to be parents with children, who visit many departments and buy the most
merchandise (Tables 3 and 4). Family-friendly stores with inter-generational appeal could
potentially attract more of these high-value shoppers and increase store revenues. Teen groups

29
spend the least time in each department and touch the fewest items, so these shoppers could be
encouraged to slow down and interact with the merchandise, perhaps with hot fashions and/or
interactive technology, featured prominently during after-school hours and on weekends.
The key findings from our study are the following: (1) Conversations with companions
have a significant, positive impact on the customer’s likelihood of shopping and buying products
in the current zone, and this tendency carries over to affect store navigation, purchase
conversion, and spending over the entire store visit. (2) The impact of group discussions is
magnified by the group’s size, consistent with Social Impact Theory (Latané and Wolf 1981). (3)
The dynamic impact of shopper discussions is moderated by group composition and
cohesiveness. The conversations of mixed-age groups and those who stay together have a greater
influence across all three stages, while those of adult groups have a marginal influence on
purchase likelihood. These adult group members may be more independent in their decision
making. (4) Group discussions have a differential impact in the three stages. While shoppers are
more likely to come back and buy in the zone where they had prior discussions, the frequency of
conversations throughout the store is a better predictor of their spending. (5) Different behavioral
patterns and social influences are associated with each stage of the shopping process. For
example, people are more likely to buy when they cover more distance and slow down, but they
spend more when their path is faster and more circuitous. Salesperson contact has a positive
impact on purchase conversion and especially spending level, perhaps due to cross- and up-
selling. Trying on a product in the dressing room has a more significant impact on both purchase
likelihood and the amount spent than simply touching the item on the selling floor.
These findings have important implications for retail management. Groups can play a
powerful role in marketing products, but their influence depends on the extent and duration of

30
group members’ interactions (Harmeling 2017). Our research reveals that shoppers in groups
don’t necessarily buy or spend more than solo shoppers (see Table 4). Creating an atmosphere
that fosters discussion is key to driving sales.
Retailers have a number of creative options for encouraging shopper engagement and
interaction. These include designing multi-user interactive displays that stimulate conversations,
providing staged settings and backdrops that prompt in-store photography and sharing, providing
sitting areas and sufficient space for shoppers to browse and talk, and building dressing rooms
that are comfortable for family and friends. Activities that encourage interaction between mixed-
age groups (e.g., parent/child) would appear to be particularly effective.
Retail strategies should be tailored to the three stages of the shopping process. For
example, at the purchase conversion stage, customers are more likely to buy if they take their
time and touch and try on products. Once “converted,” salespeople can drive multi-item
purchases by encouraging customers to maintain their shopping momentum, visit additional
departments, and try on complimentary products.
The research also demonstrates that group conversations encourage buying over the
course of the shopping trip, especially when this interaction occurs in the same department and
the shopping group is large. To investigate how the shopper’s purchase likelihood is affected by
these discussions, we plot the relationship between zone visit number and the percentage of
shoppers who buy, with separate lines for individual shoppers, shopping groups who talk
infrequently, and those who talk frequently (Figure 3). The graph shows that solo shoppers and
interactive groups have consistently higher purchase conversion rates than groups who talk
infrequently. For solo shoppers, the longest trip includes 21 zone visits, with purchases dropping
off after the 17th visit. Shoppers who converse with companions visit more departments (with a

31
maximum of 27 zone visits), and continue buying through the 21st zone visit. Groups who don’t
converse have the lowest purchase rate. This underscores the importance of understanding the
dynamic impact of group influence, and it suggests the value of encouraging group interactions
early in the store trip.3
[Insert Figure 3 about here]
Although shoppers may see their companions as having a marginal impact on their
purchase decisions (POPAI 2011), we find that group interactions significantly influence a
shopper’s store navigation, purchase likelihood, and spending. One mechanism for this effect is
that shopping and talking with others reduces the risk associated with purchase, especially for
relatively expensive items like new arrivals (which are typically sold at full price). Woodside and
Sims (1976) report that the higher the price of a product, the greater the effect of purchase pals
on consumer purchasing behavior. Conversations may also boost buying because they facilitate
preference revision and goal alignment among group members (Aribarg, Arora and Bodur 2002).
Conclusion
When people shop with companions, it affects how they navigate through the store and
their decision making processes. Their intrinsic tendency to visit a department, make a purchase,
and spend money changes dynamically in response to each conversation that occurs during the
course of the store visit. However, every shopping group has a different story because the
composition of the group affects both the pattern of interactions and how its members respond to
social influence (Kiecker and Hartman 1993). By categorizing these groups based on their
3 To show that the increased purchases are due to group conversations and not shopper self-selection, we also
examined the effect of talk frequency by using propensity score matching on mixed-age groups (Rosenbaum and
Rubin 1983). Results show that group conversations (from zero to nonzero) significantly increase sales in a zone by
$4.92 on average from $0.12 to $5.04, confirming the causal effect of group conversations on sales (see details in
Web Appendix 7). The average treatment effect (causal effect) is $4.88 (Std. error 1.45).

32
demographic composition and monitoring their interactions, researchers can gain a better
understanding of group shopping dynamics and can develop strategies for improving marketing
efficiency and store performance.
This study examines how customers shop in groups—where they go, what they buy, and
how much they spend—and how their intrinsic shopping and purchase tendencies change
dynamically in response to their conversations with group members. We model shopper behavior
using a three-stage DLM model implemented in a hierarchical Bayes framework. The unique
video tracking data allow us to study shopper behavior over time, and measure the impact of
customer interactions on various decisions in the shopping process, which is not possible with
traditional scanner panel or survey data. The new insights garnered from this study have
important implications for improving the shopping experience, promotional efforts, sales
management, and retail performance.
Looking ahead, there are several opportunities to extend this research. First, we did not
observe the relationships between group members (e.g., family, friend, or acquaintance), and this
may affect how shoppers respond to conversations. In the future, video data can be combined
with exit interviews or field surveys to examine the impact of such relationships on the dynamics
of group influence. Second, the content of shoppers’ conversations was not captured due to
privacy concerns. If shoppers give their consent to allow their discussions to be recorded, then
the valence, specificity, and importance of individual comments can be incorporated into the
model. The research could also be extended to examine dynamic group influence on shoppers’
other decisions such as store patronage and post-purchase consumption. Finally, to improve the
generalizability of the findings, the research could move from specialty apparel to other retail
settings such as grocery stores, convenience stores, and mass retailers.

33
References
Aral, Sinan, Dylan Walker (2014), “Tie Strength, Embeddedness, and Social Influence: A Large-
Scale Networked Experiment,” Management Science, 60(6), 1352-1370.
Aribarg, Anocha, Neeraj Arora and H. Onur Bodur (2002), “Understanding the Role of
Preference Revision and Concession in Group Decisions,” Journal of Marketing
Research, 39 (3), 336-49.
------, Neeraj Arora and Moon Yong Kang (2010), “Predicting Joint Choice Using Individual
Data,” Marketing Science, 29 (1), 139-57.
Arnold, Mark J., Kristy E. Reynolds, Nichole Ponder, and Jason E. Lueg (2005), “Customer
Delight in A Retail Context: Investigating Delightful and Terrible Shopping
Experiences,” Journal of Business Research, 58 (8), 1132-45.
Arora, Neeraj and Greg Allenby (1999), “Measuring the Influence of Individual Preference
Structures in Group Decision Making,” Journal of Marketing Research, 36 (4), 476-87.
Ataman, M. Berk, Carl F. Mela, and Harald J. Van Heerde (2008), “Building Brands,” Marketing
Science, 27 (6), 1036-54.
------, M. Berk, Harald J. Van Heerde, and Carl F. Mela (2010), “The Long-Term Effect of
Marketing Strategy on Brand Sales,” Journal of Marketing Research, 47 (5), 866-82.
Atkin, C. K. (1978), “Observation of Parent-Child Interaction in Supermarket Decision Making”,
Journal of Marketing, 42 (4), 41-45.
Bell, Gerald (1967), “Self-Confidence and Persuasion in Car Buying,” Journal of Marketing
Research, 4 (1), 46-52.
Belsley, David A., Edwin Kuh, and Roy E. Welsch (1980), Regression Diagnostics: Identifying
Influential Data and Sources of Collinearity. New York: John Wiley & Sons.
Burke, Raymond R. (2005), “The Third Wave of Marketing Intelligence,” Manfred Krafft and
Murali Mantrala, eds., Retailing in the 21st Century: Current and Future Trends. New
York: Springer, 113-25.
Cameron, Colin and Pravin Trivedi (2005), Microeconometrics: Methods and Applications. New
York, USA: Cambridge University Press.
Chandrashekaran, Murali, Beth A. Walker, James C. Ward, and Peter H. Reingen (1996),
“Modeling Individual Preference Evolution and Choice in A Dynamic Group Setting,”
Journal of Marketing Research, 33 (2), 211-23.

34
Chiang, Jeongwen (1991), “A Simultaneous Approach to the Whether, What and How Much to
Buy Questions,” Marketing Science, 10 (4), 297-315.
Childers Terry L. and Akshay R. Rao (1992), “The Influence of Familial and Peer-Based
Reference Groups on Consumer Decisions,” Journal of Consumer Research, 19 (2), 198-
211.
Coleman, James S. (1988), "Free Riders and Zealots: The Role of Social
Networks," Sociological Theory 6 (1), 52-57.
Corfman, Kim P. and Donald R. Lehmann (1987), “Models of Cooperative Group Decision-
Making and Relative Influence: An Experimental Investigation of Family Purchase
Decisions,” Journal of Consumer Research, 14 (1), 1-13.
Darian, Jean C. (1998), “Parent-Child Decision Making in Children’s Clothing Stores,”
International Journal of Retail and Distribution, 26 (11), 421-31.
Dion, Kenneth L. (2000), “Group Cohesion: From ‘Field of Forces’ to Multidimensional
Construct,” Group Dynamics, 4, 7-26.
Evans, Kenneth R., Tim Christiansen, and James D. Gill (1996), “the Impact of Social Influence
and Role Expectations on Shopping Center Patronage Intentions,” Journal of the
Academy of Marketing Science, 24 (3), 208-18.
Flynn, Leisa Reinecke, and Ronald E. Goldsmith (1993), “Application of the Personal
Involvement Inventory in Marketing,” Psychology & Marketing, 10 (4), 357-366.
Forsyth, Donelson R. (2006), Group Dynamics. Belmont, CA: Thomson Wadsworth.
Furse, David H., Girish N. Punj, and David W. Stewart (1984), “A Typology of Individual
Search Strategies among Purchasers of New Automobiles,” Journal of Consumer
Research, 10, 417-31.
Goff, Brent G., Danny N. Bellenger, and Carrie Stojack (1994), “Cues to Consumer
Susceptibility to Salespeople Influence: Implications for Adaptive Retail Selling,”
Journal of Personal Selling & Sales Management, 14 (2), 25-39.
Gram, Malene (2015), “Buying Food for the Family: Negotiations in Parent/Child Supermarket
Shopping: An Observational Study from Denmark and the United States,” Journal of
Contemporary Ethnography, 44 (2), 169-195.
Granovetter, Mark S. (1973), “The strength of weak ties,” American Journal of Sociology, 78 (6),
1360–1380.

35
Hackman, J. Richard, and Charles G. Morris (1975), “Group Tasks, Group Interaction Process,
and Group Performance Effectiveness: A Review and Proposed Integration,” Advances in
Experimental Social Psychology, 8, 45-99.
Harmeling, Colleen M., Robert W. Palmatier, Eric Fang, and Dainwen Wang (2017), “Group
Marketing: Theory, Mechanisms, and Dynamics,” Journal of Marketing, 81(4), 1-24.
Harrell, Gilbert D., Michael D. Hutt, and James C. Anderson (1980), “Path Analysis of Buyer
Behavior under Conditions of Crowding,” Journal of Marketing Research, 17 (1), 45-51.
Hartman, Cathy L. and Pamela L. Kiecker (1991), “Marketplace Influencers at the Point of
Purchase: the Role of Purchase Pals in Consumer Decision Making,” 1991 AMA Summer
Educators’ Conference Proceedings. Chicago: American Marketing Association, 461-69.
Heckman, James J. (1979), “Sample Selection Bias as a Specification Error,” Econometrica, 47
(1), 153-161.
Hinsz, V. B., R. S. Tindale, and D. A. Vollrath (1997), “the Emerging Conceptualization of
Groups as Information Processors,” Psychological Bulletin, 121, 43-64.
Hogg, Michael A. (1992), the Social Psychology of Group Cohesiveness: From Attraction to
Social Identity. New York: New York University Press.
Hui, Sam, Eric Bradlow, and Peter Fader (2009), “Testing Behavioral Hypotheses using An
Integrated Model of Grocery Store Shopping Path and Purchase Behavior,” Journal of
Consumer Research, 36 (3), 478-93.
———, Peter Fader, and Eric Bradlow (2009a), “Path Data in Marketing: An Integrative
Framework and Prospectus for Model Building,” Marketing Science, 28 (2), 320-35.
———, ———, and ——— (2009b), “The Traveling Salesman Goes Shopping: The Systematic
Deviations of Grocery Paths from TSP-Optimality,” Marketing Science, 28 (3), 566-72.
Hui, Sam, Yanliu Huang, Jacob Suher, and Jeffrey Inman (2013), “Deconstructing the ‘First
Moment of Truth’: Understanding Unplanned Consideration and Purchase Conversion
Using In-Store Video Tracking,” Journal of Marketing Research, 50 (4), 445-462.
Inman, J. Jeffrey and Russell S. Winer (1998), “Where the Rubber Meets the Road: A Model of
In-Store Consumer Decision-Making,” Marketing Science Institute Report No. 98-122.
------, Russell Winer, and Rosellina Ferraro (2009), “The Interplay among Category
Characteristics, Customer Characteristics, and Customer Activities on In-Store Decision
Making,” Journal of Marketing, 73 (5), 19–29.

36
Kahn, Barbara E. and Leigh McAlister (1997), Grocery Revolution: The New Focus on the
Consumer. Reading, MA: Addison-Wesley.
Kiecker, Pamela and Cathy L. Hartman (1993), “Purchase Pal Use: Why Buyers Choose to Shop
with Others,” 1993 AMA Winter Educators’ conference Proceedings. Chicago: American
Marketing Association, 378-84.
------ (1994), “Predicting Buyers’ Selection of Interpersonal Sources: the Role of Strong and
Weak Ties,” Chris T. Allen and Deborah Roedder John, eds., Advances in Consumer
Research. Provo, UT: Association for Consumer Research, 464-69.
Kowert, P. A. (2012). Groupthink or Deadlock: When do Leaders Learn from Their Advisors?
Albany: State University of New York Press.
Krackhardt, David (1992), “The Strength of Strong Ties: The Importance of Philos in
Organizations,” in Networks and Organization: Structure, Form, and Action, Nitin
Nohria and Robert G. Eccles, eds. Boston: Harvard Business School Press, 216–39.
Lam, Shun Yin, Mark Vandenbosch, John Hulland, and Michael Pearce (2001), "Evaluating
Promotions in Shopping Environments: Decomposing Sales Response into Attraction,
Conversion, and Spending Effects," Marketing Science, 20 (2), 194-215.
Larson, J. R. and C. Christensen (1993), “Groups as Problem Solving Units: Toward A New
Meaning of Social Cognition,” British Journal of Social Psychology, 32, 5-30.
Latané, Bibb (1981), “The Psychology of Social Impact,” American Psychologist, 36 (4), 343-
56.
------, and Sharon Wolf (1981), “The Social Impact of Majorities and Minorities,” Psychological
Review, 88 (5), 438-53.
Lewin, Kurt (1951). Field Theory in Social Science. New York: Harper.
Leykin, Alex and Mihran Tuceryan (2007), “Detecting Shopper Groups in Video Sequences,”
Advanced Video and Signal based Surveillance (AVSS).
Luo, Xueming (2005), “How Does Shopping with Others Influence Impulsive Purchasing?”
Journal of Consumer Psychology, 15 (4), 288-94.
Manchanda, Puneet, Asim Ansari, and Sunil Gupta (1999), “The “Shopping Basket”: A Model
for Multi-category Purchase Incidence Decisions,” Marketing Science, 18 (2), 95-114.
Mangleburg, Tamara F., Patricia M. Doney, and Terry Bristol (2004), “Shopping with Friends
and Teens’ Susceptibility to Peer Influence,” Journal of Retailing, 80 (2), 101-16.

37
McPherson, Miller, Lynn Smith-Lovin, and James M. Cook (2001), "Birds of A Feather:
Homophily in Social Networks," Annual Review of Sociology, 27(1), 415-444.
Midgley, David F. (1983), “Patterns of Interpersonal Information Seeking for the Purchase of a
Symbolic Product,” Journal of Marketing Research, 20 (1), 74-83.
Montgomery, A. L., Li, S., Srinivasan, K., Liechty, J. C. (2004), “Modeling online browsing and
path analysis using clickstream data,” Marketing Science, 23(4), 579-595.
Narayan, Vishal, Vithala R. Rao, and Carolyne Saunders (2011), “How Peer Influence Affects
Attribute Preferences: A Bayesian Updating Mechanism,” Marketing Science, 30 (2),
368-84.
Page, Bill, Anne Sharp, Larry Lockshin, and Herb Sorensen (2018), “Parents and Children in
Supermarkets: Incidence and Influence,” Journal of Retailing and Consumer Services,
40, 31-39.
POPAI (2011), Shopper Influence Study: The Global Association for Marketing at Retail.
Propp, K. M. (1999), “Collective Information Processing in Groups,” in L. Frey, D. S. Gouran,
and M. S. Poole (Eds.), the Handbook of Group Communication: Theory and Research
(pp. 225-50). Thousand Oaks, CA: Sage.
Puccinelli, Nancy M., Ronald C. Goodstein, Dhruv Grewal, Robert Price, Priya Raghubir, David
Stewart (2009), “Customer Experience Management in Retailing: Understanding the
Buying Process,” Journal of Retailing, 85 (1), 15–30.
Rosenbaum, Paul R., and Donald B. Rubin,(1983), "The Central Role of the Propensity Score in
Observational Studies for Causal Effects," Biometrika, 70 (1), 41–55.
Rust, Langbourne (1993), “Observations: Parents and Children Shopping Together: A New
Approach to the Qualitative analysis of Observational Data,” Journal of Advertising
Research, 33 (4), 65-70.
Seibold, David R., Renee A. Meyers, and Sunwolf (1996), “Communication and Influence in
Group Decision Making,” in Randy Y. Hirokawa and Marshall Scott Poole eds.,
Communication and Group Decision Making, 2nd ed. Thousand Oaks, California: SAGE
Publications.
Stasser, Garold (1988), “Computer Simulation as A Research Tool: the DISCUSS model of
Group Decision Making,” Journal of Experimental Social Psychology, 24 (5), 393-422.

38
------, and J. H. Davis (1981), “Group Decision Making and Social Influence: A Social
Interaction Sequence Model,” Psychological Review, 88, 523-51.
Su, Chenting, Edward F. Fern, Keying Ye (2003), “A Temporal Dynamic Model of Spousal
Family Purchase-Decision Behavior”. Journal of Marketing Research, 40 (3), 268-281.
Tootelian, Dennis H. and Ralph M. Gaedeke (1992), “The Teen Market: An Exploratory
Analysis of Income, Spending, and Shopping Patterns,” Journal of Consumer Marketing,
9 (4), 35-44.
Underhill, Paco (1999), Why We Buy: the Science of Shopping. New York: Simon & Schuster.
Wood, Wendy (1987), “Meta-Analytic Review of Sex Differences in Group Performance,”
Psychological Bulletin, 102 (1), 53-71.
Woodside, Arch and J. Taylor Sims (1976), “Retail Sales Transactions and Customer ‘Purchase
Pal’ Effects on Buying Behavior,” Journal of Retailing, 52 (3), 57-64.
Yang, Sha, Yi Zhao, Tülin Erdem, Ying Zhao (2010), “Modeling the Intrahousehold Behavioral
Interaction”. Journal of Marketing Research, 47( 3), 470-484.
------, Vishal Narayan and Henry Assael (2005), “Estimating the Interdependence of Television
Program Viewership Between Spouses: A Bayesian Simultaneous Equation Model,”
Marketing Science, 25(4), 336-349.
Yim, Mark Yi-Cheon, Seung-Chul Yoo, Paul L. Sauer, and Joo Hwan Seo (2014), “Hedonic
Shopping Motivation and Co-shopper Influence on Utilitarian Grocery Shopping in
Superstores,” Journal of the Academy of Marketing Science, 42, 528-544.
Zhang, Xiaoling, Shibo Li, Raymond R. Burke and Alex Leykin (2014), “An Examination of
Social Influence on Shopper Behavior Using Video Tracking Data,” Journal of
Marketing, 78 (5), 24-41.

TABLE 1: Selected Relevant Literature
Research
Study
Context
Group Type
Research Method
Measure of Group
Interaction
&Cohesiveness
Dynamics
of Group
Influence
Key Findings
Research Stream 1 – Group influence in a retail shopping context
Atkin (1978)
Supermarkets
Parent-child
Unobtrusive
observations
A verbatim description
of the sequence of
parent-child exchanges
No
The child plays the dominant role in family
cereal selection in the supermarket.
Rust (1993)
Supermarkets & toy
stores
Parents & children
Qualitative study based
on observational data
Records of what they
said and did.
No
Younger children were more prone to
pointing than older children and become
physically involved with product displays
or packaging.
Darian (1998)
Department and
specialty stores
Parents & children
Unobtrusively observe
and record the behavior
The tone of interaction
between the parent and
child
No
A purchase was more likely where both
parties were highly involved in the search,
and the interaction was collaborative.
Kiecker &
Hartman (1993)
Retail outlets
including Major
downtown stores,
Male or female
buyer with male or
female pal
Survey: quota sample of
shopping “teams”
No
No
Buyers’ perceived risk is a predominant
motivation for their use of purchase pals.
Luo (2005)
Department store
Peers or family
Lab experiment
High- vs. low-cohesive
group
No
The presence of peers increases the urge to
purchase, and the presence of family
members decreases it.
Inman, Winer &
Ferraro (2009)
Grocery stores
Accompanied by
others or not
In-store intercept
interviews
No
No
Shoppers accompanied by others were not
more likely to make unplanned purchases.
POPAI (2011)
Mass merchandise
store outlet
Family vs. friends
In-depth interviews
(entrance+exit)
No
No
Shoppers see companions as having
marginal impact on purchase decisions.
Yim et al.(2014)
Superstores
Friends, relatives,
parent-child
Field survey & video
ethnography
No
No
Co-shopping with others boosts purchases,
especially when shoppers are impulsive and
for longer store visits.
Page et al.(2018)
Supermarkets
Parents & children
Exit interviews +
density maps
No
No
Accompanied shoppers do not spend more
than unaccompanied shoppers, but shop
15% faster and avoid busy areas of the store
Research Stream 2 - Empirical models of group decision making
Aribarg, Arora
and Bodur (2002)
PC and sweet snacks
Parent-teen dyads
Conjoint analysis
No
Yes, one
preference
revision
Degree of preference revision and
concession varies across product attributes
and categories, and individuals.
Aribarg, Arora
and Kang (2010)
Cell phone
Parent-teen dyads
Online experiments
No
Yes, one
preference
revision
A new method to estimate joint preference
using only individual data is proposed.
Current study
Fashion specialty
store
All (mixed age,
adult, & teen
groups)
Video Tracking and
point of purchase
Talk frequency; staying
together most of the time
Yes, multiple
preference
revisions
Group conversations have a positive and
dynamic impact on 3 stages of decisions
moderated by group size, composition, and
cohesiveness.

TABLE 2
Operationalization of Variables
Notation
Variable
Operationalization
GenderApp
Gender appropriate
Dummy variable: 1 if a shopper is shopping at the
side of the store matching his/her sex, 0
otherwise.
NewArrival
New arrivals
Dummy variable: 1 if lead zones of the store, 0
otherwise.
Clearance
Clearance
Dummy variable: 1 if sales zone of the store, 0
otherwise.
Accessories
Accessories
Dummy variable: 1 if accessories zone of the
store, 0 otherwise.
FemaleSide
Female side
Dummy variable: 1 if female side of the store, 0
otherwise.
TalkFreq
Talk frequency
Frequency of within-group discussions during a
zone visit
GroupSize
Group size
Number of people in a shopping party
1
_
it
CumuTalk Same
−
Cumulative Talk
frequency-Same zone
Cumulative number of within-group discussions
in the same department/zone as the current, up to
visit t-1
1
_it
CumuTalk Diff −
Cumulative Talk
frequency-Different zones
Cumulative number of within-group discussions
in different zones from the current, up to visit t-1
SalesContact
Salesperson contact
Dummy variable: 1 if there is a shopper–
salesperson interaction during a zone visit, 0
otherwise
Crowding
Crowding
Number of people in a zone when the shopper
enters
MixAge
Mixed-age group
Dummy variable: 1 if a group is composed of
both adult and teen shoppers (intergenerational
group), 0 otherwise
AdultGroup
Adult group
Dummy variable: 1 if a shopper group is
composed of adults only, 0 otherwise
FemaleGroup
Female in group
Dummy variable: 1 if there are females present in
the shopping group, 0 otherwise
StayTogether
Stay together
Dummy variable: 1 if the shopping companions
stay together for more than half of the shopping
trip
it
CumuSinu
Cumulative sinuosity
The weighted average turning angles of the
shopper’s path
Distance
Distance covered
Distance the shopper walks, in feet
Speed
Speed
Traveling speed = Distance/duration
SideSale
Sidewalk sale
Dummy variable: 1 if sidewalk sale is on, 0
otherwise
TouchFreq
Touch frequency
Number of times handling the product
Dressroom
Dressing room visit
Dummy variable: 1 if shopper visits dressing
room, 0 otherwise
1it
NumZones
−
Number of zones visited
Total number of zones visited up to t-1
1it
NumBuy
−
Number of purchases
Number of purchases made up to t-1

41
TABLE 3
Summary Statistics – Mean (Std)
Variables
Mixed Age
Groups
Adult
Groups
Teen Groups
Solo Shoppers
Within-Store Behavior:
Shopping Path
Number of zones
visited*
14.604
(8.31)
8.649
(4.97)
12.389
(5.45)
8.719
(4.69)
Duration (in seconds)*
45.305
(101.32)
46.437
(97.18)
26.265
(38.95)
38.312
(73.47)
Cumulative sinuosity
4.410
(3.24)
4.287
(2.85)
4.780
(3.62)
4.749
(3.21)
Gender appropriate*
.607
(.49)
.640
(.48)
.631
(.48)
.724
(.45)
Distance covered
48.295
(55.45)
46.893
(56.11)
46.110
(53.36)
55.277
(72.58)
Speed+
3.770
(3.22)
3.531
(3.61)
3.469
(2.96)
4.176
(4.39)
Social Influence
Crowding*
.893
(1.01)
.950
(1.15)
1.108
(1.22)
.699
(1.09)
Salesperson contact
.052
(.22)
.041
(.20)
.003
(.06)
.047
(.21)
Talk frequency*
.136
(.45)
.203
(.47)
.111
(.37)
----
Group Size*
2.610
(.49)
2.045
(.21)
3.107
(1.24)
1
(0)
Cumulative Talk
frequency-Same zone*
.320
(1.11)
.130
(.41)
.022
(.17)
----
Cumulative Talk –
Different zones*
1.298
(1.95)
.590
(.88)
.181
(.53)
----
Stay together over half of
the time*
.490
(.50)
.673
(.47)
.507
(.50)
----
Within-Store Behavior:
Touch and Dressing
room visit
Touch frequency*
.519
(1.03)
.626
(1.22)
.416
(.68)
.661
(1.13)
Dressing room visit
.026
(.16)
.027
(.16)
.027
(.16)
.045
(.21)
Product Category
New arrivals*
.269
(.44)
.324
(.47)
.376
(.49)
.278
(.45)
Clearance
.302
(.46)
.248
(.43)
.242
(.43)
.267
(.44)
Accessories
.120
(.33)
.090
(.29)
.107
(.31)
.096
(.29)
Demographics
Teen*
.333
0
1
.154
(.47)
(0)
(0)
(.36)

42
Female*
Caucasian
.702
(.46)
.877
(.33)
.664
(.47)
.843
(.36)
.309
(.46)
.859
(.35)
.655
(.48)
.846
(.36)
Female in group*
1
(0)
.906
(.29)
.738
(.44)
----
Note: *indicates one or more group means are significantly different at p < 0.05.
+indicates one or more group means are significantly different at p < 0.10.
TABLE 4
Shopper Segment Comparison
Shopper
Type
Percent of
Total
Shoppers
Trip
Purchase
Rate
Zone
Purchase
Rate
Average
Basket
Size
Shopper Profile
Mixed-
age
Group
19%
54.5%
9.42%
$24.35
Often a mother and child.
This group spends the most time in store
and shops the most departments.
Highest cumulative talk and touch
frequency.
Solo
Shopper
41%
36.6%
9.35%
$12.82
Solo shoppers are mostly female (66%).
Walk fastest and cover the most distance,
but shop relatively few departments
(matching their gender).
Highly engaged with touching and trying
on items.
Adult
Group
21%
24.3%
5.41%
$11.61
Group members walk relatively slowly
and stay together while shopping.
Visit relatively few departments, but
often talk and touch items within these
departments.
Teen
Group
19%
26.4%
3.69%
$5.57
Mostly male shoppers (63%), but often
accompanied by female (76%).
Walk slowest, shop many departments
(especially new arrivals), but for the
shortest duration.
Do not talk or stay together as often as
other groups. Touch and buy the fewest
items.

43
Table 5 Estimation Results
Equation
Variables
Coefficients
in the 3
Stages
Stage 1-
Zone
Choice
Stage 2-
Purchase
Conversion
Stage 3-
Purchase
Amount
Zone
characteristic
Female side
5i
β
,
3
i
β
′
,
14
ϕ
-.270+
(.16)
.129
(.92)
---a
Gender
appropriate
1i
β
,
2
i
β
′
,
2
ϕ
.127
(.31)
.318
(.42)
.294
(.36)
Product
Category
New arrival
2i
β
,
12i
β
′
,
11
ϕ
.121
(.18)
-.503
(.51)
.297
(.51)
Clearance
3i
β
,
13i
β
′
,
12
ϕ
.318+
(.22)
.244
(.59)
.177
(.33)
Accessories
4
i
β
,
14
i
β
′
,
13
ϕ
-.208
(.18)
.204
(.71)
---a
Shopping
Path
Cumulative
sinuosity
1i
β
′
,
1
ϕ
---
-.168
(.29)
.133*
(.08)
Distance covered
4
i
β
′
,
3
ϕ
---
.005*
(.002)
---a
Speed
5i
β
′
,
4
ϕ
---
-1.252*
(.18)
.280*
(.16)
Utility
function/
Log of
Amount
Social
Influence
Crowding
6i
β
′
,
5
ϕ
---
-.359+
(.34)
-.208
(.32)
Crowding x
Crowding
7i
β
′
,
6
ϕ
---
.022
(.14)
.056
(.08)
Salesperson
contact
8i
β
′
,
7
ϕ
---
.995+
(.52)
.704*
(.28)
Marketing
Promotions
Sidewalk sales
9
i
β
′
,
8
ϕ
---
-.697+
(.44)
.314
(.34)
Within-store
Activity
Touch frequency
10i
β
′
,
9
ϕ
---
1.535*
(.31)
.179+
(.14)
Dressing room
visit
11i
β
′
,
10
ϕ
---
2.056*
(.42)
.915*
(.37)
Inverse Mills
Ratio
j
r
,
i
r
′
,
3
r
1.760*
(.56)
b
-.993*
(.53)
1.237*
(.58)
Number of zones
visited
15i
β
′
---
.117+
(.08)
---a
Number of
purchases (
15
ϕ
)
15
ϕ
--- ---
.598*
(.45)
Intercept
0ji
a
,
0i
a
′
,
0
k
-1.648*
(.34)
b
-1.989*
(.05)
.104+
(.16)
Lag term
1ji
a
,
1i
a
′
,
1
k
.017*
(.01)
b
.238*
(.02)
.042*
(.02)
Dynamic
Group
Influence
Talk frequency
2
a
,
2i
a
′
,
2
k
.194*
(.05)
.547*
(.21)
.006
(.03)
Talk frequency x
Group size
3
a
,
3i
a
′
,
3
k
.043+
(.03)
.221*
(.09)
.012
(.04)
Cumulative talk
frequency-Same
zone
4
a
,
4i
a
′
,
4
k
.016+
(.01) .278*
(.11)
.026
(.06)

44
Cumulative talk-
Different zones
5
a
,
5i
a
′
,
5
k
-.046*
(.01) -.012
(.05)
.252*
(.10)
Talk frequency x
Sales contact
6
a
,
6i
a
′
,
6
k
-.127
(.21)
---c
.001
(.03)
Talk frequency x
Crowding
7
a
,
7
i
a
′
,
7
k
-.229*
(.03)
.046
(.07)
.003
(.03)
Variance
2
s
,
2
d
s
,
2
x
s
.011*
(.04)
b
.001*
(.0003)
.014*
(.01)
Note: * indicates significant estimates with the 95% posterior probability interval does not cross
zero.
+ indicates marginally significant estimates with the 90% posterior probability interval
does not cross zero.
a variables dropped due to collinearity.
b please see the table below for zone specific values.
c variable dropped because its non-zero value predicts success perfectly.
Table 6 Moderators: Group Composition and Group Cohesiveness
6-1: Stage 1 – Zone Choice
Variable
Intercept
(
0
λ
)
Female
(
1
λ
)
Caucasian
(
2
λ
)
Mix Age
(
3
λ
)
Adult Group
(
4
λ
)
Female in
Group (
5
λ
)
Stay together
(
6
λ
)
Lag term
.031+
(.03)
.030+
(.03)
.031+
(.03)
.031+
(.03)
-.00004
(.03)
.031+
(.03)
.032+
(.03)
6-2: Stage 2 – Purchase Conversion
Intercept
(
0
λ
′
)
Female
(
1
λ
′
)
Caucasian
(
2
λ
′
)
Mixed
Age
(
3
λ
′
)
Adult
Group
(
4
λ
′
)
Female in
Group
(
5
λ
′
)
Stay
Together
(
6
λ
′
)
Lag term
.242*
(.14)
.172*
(.11)
.226*
(.13)
.088+
(.07)
.125+
(.09)
.224*
(.13)
.191*
(.12)
Talk frequency
.166+
(.20)
.131+
(.16)
.170+
(.21)
.083+
(.08)
.095
(.13)
.163+
(.20)
.123+
(.16)
Talk frequency
x Group size
.002
(.01)
.015*
(.004)
.009
(.02)
.012*
(.005)
.005
(.16)
.008+
(.005)
.005*
(.002)
Cumulative talk
-Same zone
.147+
(.16)
.106+
(.12)
.138+
(.16)
.046+
(.06)
.082
(.09)
.139+
(.16)
.120+
(.14)
Cumulative talk
-Different zones
.088
(.24)
.095
(.22)
.092
(.25)
.046
(.11)
.071
(.14)
.089
(.23)
.097
(.21)
Talk frequency
x Crowding
.005
(.01)
.003
(.01)
.011+
(.01)
.004
(.003)
.007
(.01)
.010
(.008)
.001
(.01)

45
6-3: Stage 3 – Purchase Amount
Talk
Frequency x
Female
(
8
k
)
Caucasian
(
9
k
)
Mix Age
(
10
k
)
Adult Group
(
11
k
)
Female in Group
(
12
k
)
Stay together
(
13
k
)
Mean
(Std.)
.002
(.03)
.003
(.03)
.656*
(.27)
.001
(.03)
.004
(.03)
.542*
(.21)
Table 7
Summary of the Hypothesis Testing Results
Hypothesis
Stage 1 -
Zone Choice
Stage 2 - Purchase
Likelihood
Stage 3 -
Spending
H1: Talk frequency
√
√
✗
H2: Group size
√ +
√
✗
H3a: Mixed-age
√ +
√
√
H3b: Female in group
√ +
√
✗
H3c: Stay together
√ +
√
√
H4: Cum. talk frequency
√ +
√
√
Note: + indicates marginally significant support.
FIGURE 1
Conceptual Framework
FIGURE 2
Zone Purchase
• Group size,
composition, &
cohesiveness
• Social Influence
• Shopper within-
store behavior
(path and touch)
• Social influence
• Marketing
• Product category
• Controls:
- Heterogeneity
- Endogeneity
• Within-group
talk frequency
Purchase
Amount
Zone Choice
Dynamic
shopping/
buying
tendency

46
FIGURE 2
A: Tracking Individual and Group Shoppers
B: Store Floor Map
#1: Men’s lead
Penetration: 15.2%
Conversion: 2.1%
Spending: $23.50
#3: Men’s middle
Penetration: 18.8%
Conversion: 4.3%
Spending: $37.20
#4: Accessor ies
Penetration: 11.2%
Conversion: 13.0%
Spending: $13.81
#2: Women’s lead
Penetration: 12.3%
Conversion: 1.3%
Spending: $19.50
#5: Women’s middle
Penetration: 16.8%
Conversion: 2.9%
Spending: $21.65
#8: Women’s sale
Penetration: 11.0%
Conversion: 10.3%
Spending: $13.75
Store
entrances
#7: Men’s sale
Penetration: 14.7%
Conversion: 12.1%
Spending: $22.59
#6: Checkout

47
FIGURE 3
Purchase Conversion Rate as a Function of Shopper Type,
Talk Frequency, and Number of Departments Visited
0.00
0.05
0.10
0.15
0.20
1 to 5 6 to 10 11 to 15 16 plus
% of Individual
Shoppers Who Buy
% of Group Shoppers
(Not Talking
Frequently)
% of Group Shoppers
Talking Frequently
Zones Visited