Download Learning from Your Friends` Check-Ins: An

Learning from Your Friends’ Check-Ins: An Empirical Study of
Location-Based Social Networks
Liangfei Qiu1, Zhan Shi2, Andrew Whinston3
Abstract
Recently, mobile applications have offered users the option to share their location information
with friends. Using data from a major location-based social networking application in China (a
Foursquare-like application), we estimate a structural model of restaurant discovery and
observational learning. The unique feature of repeated customer visits in the data allows us to
examine observational learning in both trial and repeat, and separate it from non-informational
confounding mechanisms, such as normative conformity and homophily, using a novel test based
on the structural model. The empirical evidence supports a strong observational learning effect and
insignificant non-informational mechanisms. We also find that the moderating role of
geographical locations on observational learning is critical in location-based social networks.
These findings suggest a nuanced view for local merchants to boost observational learning with
the advancement of location-based technology.
Keywords: Observational Learning, Location-Based Service, Social Networks, Homophily, Social
Ties.
1
2
3
Warrington College of Business Administration, University of Florida, [email protected]
W. P. Carey School of Business, Arizona State University, [email protected]
McCombs School of Business, The University of Texas at Austin, [email protected]
1
1. INTRODUCTION
The most famous example of observational learning is a sequential decision model in
Banerjee (1992) and Bikhchandani, Hirshleifer, and Welch (1992): People make their decisions on
whether or not to dine at a restaurant by observing how many (anonymous) consumers are already
in the restaurant. In current practice, however, more and more people tend to seek friends’
recommendation on location-based social networking applications for decision-making (e.g.,
Foursquare, Facebook Place, or Google+). These applications allow users to share their location
information, called check-ins, with friends through GPS-equipped mobile devices. The check-in
information generated by people’s social networks adds an important new dimension to prior
models on observational learning. People can observe the choices made by their Foursquare or
Facebook friends without having to physically visit the restaurants to observe behavior. As a result
of these new technologies, a striking difference has arisen: In Banerjee’s (1992) story, people
arrive at the restaurants in a sequence, and they can observe and rationally interpret all the choices
made by anonymous people before them; using location-based social networks, friends’ check-ins
are precisely recorded and “pushed” to users in real time. By knowing the identity and preference
of the people who made the visits, users can potentially derive more value from the information.
Observational learning is an informational explanation of the correlated behavior among
friends: An individual’s decision is affected by the observation of friends’ choices because of their
informational content (Cai, Chen, and Fang 2009). The effect of observational learning is
complicated by several plausible confounding mechanisms. The first mechanism is the normative
conformity effect. People may want to behave like their friends because they want to conform
2
(such as peer pressure). Asch’s (1951) classical conformity experiments show that an individual's
own opinions are influenced by those of a majority group. The second is the homophily-driven
diffusion process described by Aral, Muchnik, and Sundararajan (2009): inherent similarities in
friends’ personal characteristics can also cause correlated friends’ choices. Unlike observational
learning, these two confounding mechanisms are non-informational mechanisms.
In the present study, we aim to tease out the observational learning effect from
non-informational confounding mechanisms. We estimate a two-stage structural model of
location-based social networks. In our context, the first stage—awareness—means that friends’
check-ins lead some of the uninformed consumers to discover a new venue. The second
stage—observational learning—refers to the fact that check-ins made by friends help users learn
the quality of a venue. The intuition of identifying observational learning from other
non-informational mechanisms is as follows (Zentall and Galef 1988; Van den Bulte, and Valente
2011; Lee and Bell 2013; Iyengar, Van den Bulte, and Lee 2014): If we observe a sharp decline in
the clustering of check-in behaviors amongst peers as consumers proceed from trial to repeat, it
will be consistent with a significant observational learning effect in trial because personal dining
experience substitutes for observational learning from peers. Previous research has shown the
effect of observational learning or word of mouth on new product adoption/single purchase (Duan,
Gu, and Whinston 2009; Zhang 2010; Zhu and Zhang 2010). Our study contributes to the
literature by examining observational learning in both trial and repeat in a unified empirical
model.
After separating observational learning from other confounding mechanisms, we further
3
examine important factors that govern the efficacy of observational learning. Most of the prior
work has focused on the moderating role of social ties on the effectiveness of word of mouth or
observational learning: strong ties are more influential than weak ties (Bakshy et al. 2012; Shi and
Whinston 2013; Aral and Walker 2014). Besides confirming the moderating role of social ties, we
take a step further and add a new location dimension in our analysis. A unique feature that makes
location-based networks different from other types of social networks is the location information
shared by users. The check-in location indicates the current geographical status of a user in the
real world and reflects the user’s behavior more closely to the real world compared with other
online social networks (Gao and Liu 2014). Cairncross (2001) proposed the idea of “Death of
Distance”: the role of physical distance has been diminishing because of the communication
revolution in the Internet era. However, our empirical results suggest that the moderating role of
geographical locations on observational learning is critical in location-based social networks.
These findings suggest that in the presence of a location-based network, marketing strategies of
local businesses should be contingent on social ties as well as location factors. Our observational
learning interpretation provides a coherent explanation for the complex pattern of findings on the
moderating roles of social and location factors, whereas homophily and conformity behavior do
not.
2. LITERATURE REVIEW
A handful of empirical papers have examined the mechanism of observational learning. Duan,
Gu, and Whinston (2009) examined herd behavior and informational cascades in the context of
online software adoption. Zhang (2010) studied observational learning in the U.S. kidney market.
4
Chen, Wang, and Xie (2011) disentangled whether consumers' purchase decisions can be
influenced by others’ opinions (word of mouth) or actions (observational learning) using a natural
experiment from Amazon. Prior studies have also examined how product and consumer
characteristics moderate the efficacy of observational learning or word of mouth (Zhu and Zhang
2010; Tucker and Zhang 2011; Luo, Zhang, and Duan 2013; Lee and Raghu 2014).
The present study is closely related to Shi and Whinston (2013) because they also studied
observation learning in a context of location-based networks. Our analysis differs from Shi and
Whinston (2013) and other prior studies in three aspects. First, most of the previous studies
focused exclusively on single-purchase products (information goods), such as books, movies,
music, and video games, which are purchased only once. In our present study, we examine both
new product adoption and repeat purchases in the context of restaurant industry in a unified
structural model. Although Shi and Whinston (2013) examined observational learning in the same
industry as ours, they did not differentiate between trial and repeat. Second, Researchers often face
the challenge of identifying the causal effect of observational learning or social influence from
homophily due to the endogenous nature of social tie formation (Manski 1993). Some
identification strategies include the use of instrumental variables (Shriver et al. 2013), natural
experiments (Zhang and Wang 2012), matching methods (Aral, Muchnik, and Sundararajan 2009;
Wang, Zhang, and Hann 2014), controlled laboratory experiments (Qiu, Rui, and Whinston 2014),
and field experiments (Aral and Walker 2011). Shi and Whinston (2013) applied the machine
learning technique of nonnegative matrix factorization to uncover users’ latent features from the
network graph and identify the causal effect. In this study, we propose a different identification
5
method: a novel test based on the structural model of trial and repeat to separate observational
learning from other non-informational mechanisms, such as homophily and normative conformity.
As a robustness check, we also use the technique of instrument variables to account for correlated
unobserved heterogeneity and confirm the identification of the informational mechanism. Third, in
ordinary social networks, the efficacy of observational learning or word of mouth depends on
consumer and product characteristics as well as the strength of social ties. Our location-based
network offers a new location dimension: the effect of observational learning from friends’
check-ins depends crucially on whether the check-ins are in the focal users’ familiar regions. The
prior literature (Zhu and Zhang 2010; Wang, Zhang, and Hann 2014) focused on how consumer
characteristics (e.g. Internet experience, network size of users) and product characteristics (e.g.
video game popularity, old books) moderate the effect of word of mouth or peer influence. Unlike
Shi and Whinston (2013) focusing only on the moderating role of social ties, we examine how
social and location factors moderate the magnitude of observational learning together.
3. DATA
The dataset comes from a major location-based social networking application (a
Foursquare-like application) in China. Users can check in at a venue to say that they are currently
there (see Figure 1). It also lets them connect to their online friends; this function is equivalent to
the concept of “friends” on Facebook. Users can observe their network friends’ check-ins through
the mobile application (see Figure 2).
[Insert Figure 1 – 4 and Table 1 Here]
Our data includes restaurant check-in information and the users’ social network. The first part
6
of the data is consumer check-ins of 50 randomly selected restaurants in Shanghai, China. The
period of check-in history is from May, 2010 to Jan, 2013. We can observe when who checked in
and where. The total number of users is 34,207. They are randomly selected from users in
Shanghai by the application company. Figure 3 depicts the frequency histogram of the check-ins
of restaurants, and the frequency histogram of the unique customers of restaurants. The other part
of our data is the undirected social graph (see Figure 4). The social network is recorded as of
February 15, 2011. Table 1 summarizes the descriptive statistics of the location-based social
network by users. It shows that, on average, each user has approximately four direct friends (the
mean of the degree centrality is 4.375) and makes 36 check-ins during the sample period.
4. A STRUCTUAL MODEL OF LEARNING IN LOCATION-BASED NETWORKS
In this section, we develop and estimate a two-stage model of restaurant discovery and
quality learning. The notations for the parameters can be found in Table 2.
[Insert Table 2 Here]
4.1 A Two-Stage Structural Model of Restaurant Discovery and Quality Learning
Following Hendricks and Sorensen (2009), the probability that a consumer visits a venue is
the product of two probabilities: the probability that she likes the venue conditional on discovering
it and the probability that she discovers the venue. We outline the sequence of events in period t as
follows (the process proceeds in a similar manner in period t + 1).
Stage 1: Some uninformed consumers become aware of a restaurant
specify the probability that an uninformed consumer discovers venue
in period t. We
in period t, Pr
as follows: If the number of consumer ’s friends’ check-ins at restaurant j in period
7
=1 ,
− 1 is
zero, then the baseline awareness probability in period ,
Pr
A positive
=1 =
.
(1)
indicates that without a location-based social network, a consumer can still
discover a new restaurant, for example, by searching on Yelp, TripAdvisor, and other sources of
public information. If the number of consumer ’s friends’ check-ins at restaurant j in period
− 1 is greater than zero, then the awareness probability in period , Pr
Pr(
= 1) = 1 , then Pr(
= 1) = 1 , for
= 1 = 1. If
= + 1, . . . , . We model the awareness
process as a binary variable. As soon as a friend checks in, a notification is pushed and the focal
user becomes aware of the restaurant.
Stage 2: Conditional on their being aware of the restaurant, consumers make a decision on
whether to go to this restaurant. The utility function for consumer
who has not visited restaurant
j (her number of self check-ins at restaurant j up until period t – 1 is zero) conditional on having
learned the existence of venue
is
= (
where
,
,
,
,
+
,
+
~ (0,
,
is the total number of friends’ check-ins at venue
(2)
up until period t – 1, and
up until period t – 1. 4 The
is the total number of strangers’ check-ins at venue
unobserved latent quality of restaurant j is
),
. For simplicity, the conditional expected quality of
restaurant j for consumer i is given by a linear functional form:5
(
,
,
,
=
+
,
+
,
.
(3)
4
We also do a robustness check when
and
are normalized to be the number of friends’ or strangers’ check-ins per
,
,
month, and add centrality measures in the right hand side of equation (2). The basic results are similar.
5
Note that in equation (3), we do not directly model the process of Bayesian learning in networks because the decision rules used
in perfect Bayesian equilibria are complicated, and the analytic solution requires strong assumptions on network topology
(Acemoglu et al. 2011; Qiu and Whinston 2014).
8
This equation implies that when a consumer has never visited a restaurant before, we assume that
her expected quality of the restaurant is a function of the number of her friends’ check-ins and the
number of strangers’ check-ins. In other words, consumers interpret their friends’ repeated
check-ins and strangers’ check-ins as signals of the quality of restaurants. Following Banerjee
(1992) and Zhang and Liu (2012), the coefficient on
,
,
observational learning effect from strangers, and the coefficient on
, measures the classical
,
,
, measures the
effect of observational learning from friends. The effects of observational learning,
and
,
might be different across restaurants. For example, a restaurant in a downtown area might be more
affected by observational learning effects than a roadside restaurant on a highway. The parameter
represents the perceived quality of a restaurant (the observable characteristics of a restaurant)
before a customer actually visits it. For example, a consumer can know some of the characteristics
of a restaurant before she visits it, such as the restaurant type, the price range, the noise level, and
whether it has free Wi-Fi, from websites such as Yelp. Following Mayzlin, Dover, and Chevalier
(2014), we control for the density of restaurants (spatial competition). For each restaurant in our
data, we construct a variable,
The error term
observe her own
, that represents the number of neighbor restaurants within 0.5 km.
represents individual taste shock and is i.i.d. distributed. Each consumer can
and perceived quality
before making the decision, but the researcher
cannot.
The utility function for consumer
who has already visited restaurant j (her number of self
check-ins at restaurant j up until period t – 1 is at least one) conditional on having learned the
existence of venue
is given by
9
=
where the parameter
+
,
+
+
,
+
,
~ (0,
),
(4)
is the realized quality of the restaurant. It is worth noting the difference
and perceived quality
between realized quality
. A consumer can know
from public
information sources, such as Yelp, before she visits the restaurant. A consumer updates her belief
and knows the realized restaurant quality
realized quality
only after she visits the restaurant at least once. The
can be identified by the sensitivity of focal consumers’ visits to whether or not
they have visited this restaurant before. If a consumer’s self check-in can significantly increase her
future probability of visiting the restaurant, we would expect that the realized quality
is higher.
4.2 Identification of Observational Learning
We include
,
in equation (4) because we want to test an interesting and important
question: Does the parameter
mainly capture the observational learning effect instead of the
non-informational mechanisms, such as the conformity effect or the correlated personal tastes
(homophily)? We can separate observational learning from the non-informational mechanisms by
comparing parameters
in equation (3) and
in equation (4). If
mainly captures the
non-informational mechanisms (it implies the absence of observational learning), we would expect
to observe
=
because the effects of normative conformity or correlated unobserved tastes
(homophily) should remain unchanged regardless of whether consumer
However, if
has visited restaurant j.
mainly captures the effect of observational learning, we would expect to observe
>
because when a consumer has visited a restaurant, she has a better idea about the realized
quality
, and she should rely less on her friends’ check-ins to infer the true restaurant quality
(
>
). It is worth noting that homophily may cause a potential endogeneity problem in the
10
estimation, and we will discuss this in the next section. Conditional on discovering the venue, if
the utility of visiting restaurant j in period t,
, is greater than the reservation utility, consumer i
will go to restaurant j in period t. The probability that consumer
conditional on discovering it is given by Pr
venue in period
≥
in period
. The probability that a consumer visits a
= 1 ⋅ Pr
is the product of two probabilities: Pr
of generality, the reservation utility
visits venue
≥
. Without loss
is normalized to zero. We construct the log likelihood
function to estimate the empirical model:
ln ( ) = ln
=
where
ln
Pr
= 1 ⋅ Pr
≥0
Pr
= 1 ⋅ Pr
≥0
1 − Pr
= 1 ⋅ Pr
1 − Pr
≥0
is an indicator for whether consumer
= 1 ⋅ Pr
+ ln 1 − Pr
visits venue
in period
In the estimation, we use one month as the time unit of analysis.
(T = 32), J is the number of venues (J = 50), and
Note that if
= 1, then Pr(
= 1) = 1, for
= 1 ⋅ Pr
≥0
≥0
,
from the real data.
is the number of time periods
is the number of consumers (N = 34,207).
= + 1, . . . , .6
Our estimates of the parameters are chosen to satisfy:
=
, , , ̂ ,
,
, ̂ ,
=
,
argmax
,
,
,
,
, ,
ln ( ).
(5)
5. EMPIRICAL RESULTS
5.1 Main Results from the Structural Model
In this section, we present the empirical results estimated from equation (5). Table 3 shows
the estimation results of a typical restaurant in our data. The basic results are robust for other
6
The restaurants a consumer has been to are automatically in her consideration set, i.e. the awareness probability is one.
11
restaurants (more estimation results for other restaurants are available upon request). In the main
model (column 1 of Table 3), we find that
is significantly larger than
, and
is not
statistically different from zero. This empirical evidence supports a strong observational learning
effect and insignificant non-informational mechanisms including normative conformity and
homophily. As argued in Cai, Chen, and Fang (2009), normative conformity is less severe in the
restaurant setting because, in contrast with popular cultural products, restaurant dining is a more
private experience. We also find that the coefficient on strangers’ check-ins,
, is not statistically
significant. Considering the size of the coefficients, the effect of a friend’s check-in is equivalent
to that of 476 check-ins made by strangers. The main takeaway of Banerjee’s (1992) observational
learning model is that check-ins made by strangers can convey quality. However, our empirical
results suggest that strangers’ check-ins are not as important as friends’ check-ins in determining
the expected quality of restaurants. We modify equation (2) and control for observable centrality
measures of users in the location-based social network as follows:
=Θ ∙
where
+ (
,
,
,
+
+
,
(6)
is a 4 × 1 vector of user i’s observable characteristics, including the degree centrality,
closeness centrality, betweenness centrality, and individual clustering coefficient summarized in
Table 1. The estimation results are shown in column 2 of Table 3.
[Insert Tables 3 and 4 Here]
5.2 Identification and Endogeneity
Identifying the causal observational learning effect from archival data is challenging. The
main confounding mechanism discussed in the literature is homophily because it highlights
12
whether a mechanism is causal (Jackson 2008; Aral, Muchnik, and Sundararajan 2009; Wang,
Zhang, and Hann 2014). In our specific context, the correlation between friends’ check-ins can be
driven by the correlated unobserved individual heterogeneity. Following Bramoullé, Djebbari, and
Fortin (2009), we use friends’ friends’ check-in behaviors as an instrumental variable to make
causal inference. In essence, we are using the fact that a user in our location-based social network
is not always friends with all of his friends’ friends. The intuition behind this instrument variable
is that the check-in actions of friends’ friends who are not the focal user's friends can only have an
impact on the focal user's future visit indirectly, by influencing the check-in actions of her friends.
For example, consider a three-person network, and our focal user is person A. Person B is a friend
of person A, and person C is a friend of person B, but person C and person A are not friends. We
use person C’s check-in behavior as an instrument variable for person B’s check-in behavior. Two
assumptions must hold for this instrument to be valid: (1) Person B and person C’s check-in
decisions correlate. (2) Person A’s check-in action is not influenced by her indirect friend, person
C, except through her direct friend, person B. Assumption (1) holds because person B and person
C are friends. More importantly, we argue that Assumption (2) is reasonable in our context.
Person C is not a friend of person A, and from our previous empirical results, we know that the
focal user’s check-in decision is not significantly influenced by strangers’ check-ins. In this case,
person C’s check-in behavior can be thought as an exogenous variation that facilitates our
identification. The estimation results are shown in column 1 of Table 4. Although the value of
has decreased, it is still statistically significant, which implies a robust effect of observational
learning.
13
5.3 Important Factors Governing the Efficacy of Observational Learning
The Strength of Strong Ties The strand of research on social ties originates from the “strength of
weak ties” hypothesis proposed by Granovetter (1973). The gist of the hypothesis is that we
always get truly new information from acquaintances, rather than from our close friends. However,
when we consider observational learning in location-based social networks, our estimation shows
the strength of strong ties: Strong ties are more likely to be activated for observational learning.
Note that in our study, we focus on the effect of tie strength on observational learning instead of
on knowledge spillover. The strength of social ties between consumer i and her friend, consumer j,
is measured by the number of their common friends, adjusted by the number of consumers who
=
are friends of either consumer i or consumer j. More formally,
=
( )∩ ( )
( )∪ ( )
, where
()
represents the set of friends of consumer . We divide a consumer’s friends into two equally sized
groups, depending on the tie strength: The group of close friends includes consumer i’s friends
who have the highest 50% of the level of
. Those left are the group of ordinary friends.
[Insert Table 5 Here]
We modify the structural model to investigate the role that tie strength plays in the process of
observational learning. The awareness stage remains unchanged, and we focus on the
observational learning stage. The equation (3) is modified to the following linear function:
(
where
,
,
,
,
,
,
=
+
+
,
+
,
is the number of close friends’ total repeated check-ins at venue
period t - 1, and
,
,
(7)
up until time
is the number of ordinary friends’ total repeated check-ins at venue
until period t - 1. The parameter
up
measures the observational learning effect of strong ties, and
14
,
measures the observational learning effect of weak ties. Column 1 of Table 5 shows the
estimation results of
and
. We find that
is significantly greater than
, indicating that
strong ties can accelerate observational learning.
Location, Location, Location In our context, restaurants are horizontally differentiated by
geographical locations. From a consumer’s point of view, restaurants can be either in her familiar
region or in her unfamiliar region. We expect that the magnitude of observation learning for a
focal consumer is different when her friend checks in at a local restaurant in her familiar region or
a non-local restaurant in her unfamiliar region. In this study, we define consumer i’s familiar
region in period t as the zip code region in which this consumer has the largest number of
check-ins up until period t (Wang and Goh 2012). We modify equation (7) to add a location
dimension to our observational learning model of social ties as follows:
(
=
where
,
+(
+
,
,
)
,
,
,
,
+(
+
,
,
,
,
)
,
+
,
,
(8)
is a dummy variable that takes the value 1 if restaurant j is in consumer i’s familiar
region (i.e., local restaurant) in period t - 1, and 0 otherwise. Once again, in column 2 of Table 5,
we find that the parameter
parameter
is greater than
. More interestingly, we also find that the
is significantly less than 0, and the parameter
is not statistically different from
0. These estimation results show that the efficacy of observational learning crucially depends on
social ties as well as the location dimension (see Figure 5): (1) The magnitude of observational
learning from a close friends’ check-in at a focal user’s local restaurant is significantly less than
that from a close friends’ check-in at a focal user’s non-local restaurant; and (2) the magnitude of
15
observational learning from an ordinary friends’ check-in at a focal user’s local restaurant is
similar to that from an ordinary friends’ check-in at a focal user’s non-local restaurant.
A possible explanation for our finding (1) is that the focal user can more easily get
information of a local restaurant in her familiar region from offline word of mouth sources. Near
its own location, a restaurant can display signs or distribute flyers. Therefore, a consumer has less
quality uncertainty of local restaurants compared to that of non-local ones, and there is less of a
need for her to rely on observational learning from close friends.
6. CONCLUSIONS AND LIMITATIONS
The present study has several limitations. Like Hinz et al. (2011), we assumed that the
location-based social network remains fixed for the duration of our study. This assumption ignores
the effects of dynamic network formation in real-world social networks. Second, the business
model of location-based service relies on the active online sharing of check-ins. However, people
who highly value privacy can be less willing to share their check-ins when they visit venues.7
Studying the effect of privacy concerns on observational learning in social networks remains an
open question. Third, we are collecting customer reviews of restaurants over time from a different
online review website. As a future research direction, we will control for the effect of online word
of mouth in our estimation.
7
This might cause a selection bias problem (Heckman 1979). However, As Lindqvist et al. (2011) shows, privacy concerns have
not kept user from experimenting with and adopting location-based service. Restaurants and bars are fairly popular places to
check-in at. Therefore, we believe the selection bias is small.
16
FIGURES AND TABLES
Figure 1. A Screenshot of the Application Interface
Figure 2. A Location-Based Social Network
17
Figure 3. Histogram of the Check-ins and the Number of Customers at Restaurants
18
Figure 4. Snowball Sampling of 0.5% of Users in the Social Graph
Note: the node size represents the degree of a node, and the node color represents the number of
check-ins a user made in the sample period, Purple: 0; Blue: 1-10; Green: 11-100; orange: >100.
19
Figure 5: Moderating Role of Location and Social Ties on Observational Learning
Table 1. Summary Statistics of the Location-Based Service Users
8
Degree centrality
Closeness centrality
Betweenness centrality
Individual clustering coefficient
Number of check-ins
Number of unique restaurants visited
Mean
The Std. Dev.
Max
Min
Obs
4.375
4.57E-09
51,680.26
0.0243
35.920
3.257
9.578
1.63E-09
470,134.6
0.0474
8.149
1.559
566
5.92E-09
54,962,235
0.229
817
22
0
8.55E-10
0
0
0
0
34,207
34,207
34,207
34,207
34,207
34,207
8
The mathematical definitions of all centrality measures in Table 1 can be found in Jackson (2008).
20
Table 2. Summary of Notations
Notation
,
,
Description
The effect of friends’ check-ins before the focal user’s fist visit
The effect of friends’ check-ins after the focal user’s fist visit
The effect of strangers’ check-ins before the focal user’s fist visit
The effect of strangers’ check-ins after the focal user’s fist visit
The baseline awareness probability
The quality of the restaurant (cannot be observed by the focal user before the first visit)
Restaurant heterogeneity (can be observed by the focal user before the first visit)
The number of neighbor restaurants within 0.5km (restaurant density)
Centrality measures of a consumer
Whether a consumer is aware of the restaurant
The total number of friends’ check-ins at venue up until period t – 1
The total number of strangers’ check-ins at venue up until period t – 1
21
Table 3. Estimated Parameters of the Main Observational Learning Model
(1)
Main Model
(2)
Including Centrality
Measures
3.459***
[5.422]
0.227
[0.928]
0.00726
[0.724]
0.00655
[0.278]
3.226***
[4.572]
0.219
[0.845]
0.00677
[0.652]
0.00428
[0.215]
4.628***
4.532***
Centrality Measures
[7.094]
-1.011***
[-3.677]
0.000942*
[1.726]
0.594***
[3.941]
N
[6.411]
-1.008***
[-3.527]
0.000908*
[1.712]
0.496***
[3.108]
Y
Number of Users
34,207
34,207
z statistics in brackets, * p < 0.1, ** p < 0.05, *** p < 0.01
22
Table 4. Estimated Parameters using Instrument Variable
(1)
Instrument Variable
3.075***
[5.046]
0.204
[0.721]
0.00712
[0.409]
0.00672
[0.322]
4.675***
[5.403]
-1.117***
[-3.843]
0.000902*
[1.687]
0.741***
[3.422]
Number of Users
34,207
z statistics in brackets, * p < 0.1, ** p < 0.05, *** p < 0.01
23
Table 5. Important Factors Governing the Efficacy of Observational Learning
(1)
Social Ties
(2)
Familiar
Regions
0.401
[0.528]
0.00683
[0.421]
0.00502
[0.338]
0.292
[0.308]
0.00726
[0.655]
0.00614
[0.382]
3.257***
3.724***
[3.649]
-1.024***
[-3.012]
0.000823
[1.425]
0.573***
[3.309]
[4.023]
-1.205***
[-3.428]
0.000805
[1.296]
0.582***
[3.807]
4.302***
4.829***
[3.976]
[4.652]
1.825***
1.906***
[3.324]
[3.025]
-1.314***
[2.842]
-0.102
[0.264]
Number of
Users
34,207
34,207
z statistics in brackets, * p < 0.1, ** p < 0.05, *** p < 0.01
24
REFERENCES
1. Acemoglu, D., Dahleh, M. A., Lobel, I., and Ozdaglar, A. (2011). Bayesian Learning in Social
Networks. Review of Economic Studies, 78(4), 1201-1236.
2. Aral, S., Muchnik, L. and Sundararajan A. (2009). Distinguish Influence-Based Contagion
from Homophily-Driven Diffusion in Dynamic Networks. Proceedings of the National
Academy of Sciences, 106(51), 21544-21549.
3. Aral, S., and Walker, D. (2011). Creating Social Contagion through Viral Product Design: A
Randomized Trial of Peer Influence in Networks. Management Science, 57(9), 1623-1639.
4. Aral, S., and Walker, D. (2014). Tie Strength, Embeddedness, and Social Influence: A
Large-Scale Networked Experiment. Management Science, Forthcoming.
5. Asch, S.E. (1951). Effects of group pressure on the modification and distortion of judgments.
In H. Guetzkow (Ed.), Groups, leadership and men(pp. 177–190). Pittsburgh, PA: Carnegie
Press.
6. Bakshy, E., Rosenn, I., Marlow, C., and Adamic, L. (2012). The Role of Social Networks in
Information Diffusion. In Proceedings of the 21st international conference on World Wide
Web (pp. 519-528). ACM.
7. Banerjee, A. V. (1992). A Simple Model of Herd Behavior. Quarterly Journal of Economics,
107(3), 797-817.
8. Bapna, R., Qiu, L., and Rice, S. (2014). Repeated Interactions vs. Social Ties: Quantifying the
Economic Value of Trust, Forgiveness, and Reputation Using a Field Experiment, NET
Working paper.
9. Bikhchandani, S., D. Hirshleifer, and I. Welch. (1992). A Theory of Fads, Fashion, Custom,
and Cultural Change in Informational Cascades. Journal of Political Economy, 100(5),
992-1026.
10. Bramoullé, Y., Djebbari, H., and Fortin, B. (2009). Identification of Peer Effects through
Social Networks. Journal of Econometrics, 150(1), 41-55.
11. Cai, H., Chen, Y., and Fang, H. (2009). Observational Learning: Evidence from a Randomized
Natural Field Experiment. American Economic Review, 99(3), 864-882.
12. Cairncross, F. (2001). The Death of Distance: How the Communications Revolution is
Changing Our Lives. Harvard Business Press.
13. Chen, Y., and Xie, J. (2005). Third-Party Product Review and Firm Marketing
Strategy. Marketing Science, 24(2), 218-240.
14. Chen, Y., Wang, Q., and Xie, J. (2011). Online social interactions: A natural experiment on
word of mouth versus observational learning. Journal of Marketing Research, 48(2), 238-254.
15. Chatterjee, P. (2001), Online Reviews: Do Consumers Use Them? in Advances in Consumer
Research, Mary C. Gilly and Joan Myers-Levy, eds. Provo, UT: Association for Consumer
Research, 129–34.
16. Dhar, R. (1997). Consumer Preference for a No-Choice Option. Journal of Consumer
Research, 24(2), 215-231.
17. Dou, Y., Niculescu, M. F., and Wu, D. J. (2013). Engineering Optimal Network Effects via
Social Media Features and Seeding in Markets for Digital Goods and Services. Information
25
Systems Research, 24(1), 164-185.
18. Duan, W., B. Gu, and A. B. Whinston. (2009). Informational Cascades and Software Adoption
on the Internet: an empirical investigation. MIS Quarterly. 33(1), 23-48.
19. Erdem, T., Keane, M. P., and Sun, B. (2008). The Impact of Advertising on Consumer Price
Sensitivity in Experience Goods Markets. Quantitative Marketing and Economics, 6(2),
139-176.
20. Fong, N. M., Fang, Z., and Luo, X. (2015). GEO-Conquesting: Competitive Locational
Targeting of Mobile Promotions. Journal of Marketing Research, forthcoming.
21. Forman, C., Ghose, A., and Wiesenfeld, B. (2008). Examining the Relationship between
Reviews and Sales: The Role of Reviewer Identity Disclosure in Electronic
Markets. Information Systems Research, 19(3), 291-313.
22. Gao, H., and Liu, H. (2014). Data Analysis on Location-Based Social Networks. In Mobile
Social Networking (pp. 165-194). Springer New York.
23. Ghose, A., Ipeirotis, P. G., and Li, B. (2012). Designing Ranking Systems for Hotels on Travel
Search Engines by Mining User-Generated and Crowdsourced Content. Marketing
Science, 31(3), 493-520.
24. Ghose, A., Goldfarb, A., and Han, S. P. (2012). How is the mobile Internet different? Search
costs and local activities. Information Systems Research, 24(3), 613-631.
25. Granovetter, M. S. (1973). The Strength of Weak Ties. American Journal of Sociology, 78(6),
1360-1380.
26. Heckman, J. J. (1979). Sample Selection Bias as a Specification Error. Econometrica, 47(1),
153-161.
27. Hendricks, K., and Sorensen, A. (2009). Information and the Skewness of Music
Sales. Journal of Political Economy, 117(2), 324-369.
28. Hinz, O., Skiera, B., Barrot, C., and Becker, J. U. (2011). Seeding Strategies for Viral
Marketing: An Empirical Comparison. Journal of Marketing, 75(6), 55-71.
29. Hoyer, W. D. (1984). An Examination of Consumer Decision Making for a Common Repeat
Purchase Product. Journal of Consumer Research, 11(3), 822-29.
30. Iyengar, R., Van den Bulte, C., and Valente, T. W. (2011). Opinion Leadership and Social
Contagion in New Product Diffusion. Marketing Science, 30(2), 195-212.
31. Iyengar, R., Van den Bulte, C., and Lee, J. Y. (2014). Social Contagion in New Product Trial
and Repeat. Marketing Science, forthcoming.
32. Jackson, M.O. Social and Economic Networks, Princeton University Press, Princeton, NJ,
2008.
33. Lee, J. Y., and Bell, D. R. (2013). Neighborhood Social Capital and Social Learning for
Experience Attributes of Products. Marketing Science, 32(6), 960-976.
34. Lee, G., and Raghu, T. S. (2014). Determinants of Mobile Apps Success: Evidence from App
Store Market. Journal of Management Information Systems, 31(2), 133-170.
35. Lee, S., Qiu, L., and Whinston, A.B. (2014). Sentiment Manipulation in Online Platforms
and Opinion Forums: A Regression Discontinuity Approach. Working paper, University of
Texas at Austin.
36. Lindqvist, J., Cranshaw, J., Wiese, J., Hong, J., and Zimmerman, J. (2011). I'm the mayor of
26
my house: examining why people use foursquare-a social-driven location sharing application.
In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (pp.
2409-2418). ACM.
37. Luo, X., Zhang, J., and Duan, W. (2013). Social Media and Firm Equity Value. Information
Systems Research, 24(1), 146-163.
38. Luo, X., Andrews, M., Fang, Z., and Phang, C. W. (2013). Mobile targeting. Management
Science, 60(7), 1738-1756.
39. Ma, H., Yang, H., Lyu, M. R., & King, I. (2008). Sorec: Social Recommendation Using
Probabilistic Matrix Factorization. In Proceedings of the 17th ACM conference on Information
and knowledge management (pp. 931-940). ACM.
40. Mann, L (1969). Social Psychology. New York: Wiley.
41. Manski, C.F. (1993). Identification of Endogenous Social Effects: The Reflection Problem.
Review of Economic Studies, 60(3), 531-542.
42. Mayzlin, D., Dover, Y., and Chevalier, J. (2014). Promotional Reviews: An Empirical
Investigation of Online Review Manipulation. American Economic Review, forthcoming.
43. Nevo, A., and Whinston, M. D. (2010). Taking the Dogma out of Econometrics: Structural
Modeling and Credible Inference. Journal of Economic Perspectives, 24(2), 69-82.
44. Pavlou, P. A., and Fygenson, M. (2006). Understanding and Predicting Electronic Commerce
Adoption: An Extension of the Theory of Planned Behavior. MIS Quarterly, 115-143.
45. Qiu, L., Rui, H., & Whinston, A. B. (2014). Effects of Social Networks on Prediction Markets:
Examination in a Controlled Experiment. Journal of Management Information Systems, 30(4),
235-268.
46. Qiu, L., and Whinston, A. B. (2014). Naive Herding in Location-Based Networks: Induced
Social Learning and Optimal Dynamic Pricing. Available at SSRN 2090093.
47. Shi, Z., and Whinston, A. B. (2013). Network Structure and Observational Learning: Evidence
from a Location-Based Social Network. Journal of Management Information Systems, 30(2),
185-212.
48. Shriver, S. K., Nair, H. S., and Hofstetter, R. (2013). Social Ties and User-Generated Content:
Evidence from an Online Social Network. Management Science, 59(6), 1425-1443.
49. Tolbert, P. S., and Zucker, L. G. (1983). Institutional Sources of Change in the Formal
Structure of Organizations: The Diffusion of Civil Service Reform, 1880-1935. Administrative
Science Quarterly, 22-39.
50. Tucker, C. (2008). Identifying Formal and Informal Influence in Technology Adoption with
Network Externalities. Management Science, 54(12), 2024-2038.
51. Tucker, C., and Zhang, J. (2011). How does popularity information affect choices? A field
experiment. Management Science, 57(5), 828-842.
52. Vogel, J. (2008). Spatial Competition with Heterogeneous Firms. Journal of Political
Economy, 116(3), 423-466.
53. Wang, A., Zhang, M., and Hann, I. H. (2014). Socially Nudged: A Quasi-Experimental Study
of Friends' Social Influence in Online Product Ratings. Information Systems Research,
forthcoming.
54. Wang, Q., and Goh, K. Y. (2012). Consumer Segmentation and the Information Rule of
27
Online Reviews in Horizontally Differentiated Product Markets. In E-Life: Web-Enabled
Convergence of Commerce, Work, and Social Life (pp. 225-233). Springer Berlin Heidelberg.
55. Xu, H., Teo, H.H., Tan, B.C.Y., and Agarwal, R.. (2012). Effects of Individual Self-Protection,
Industry Self-Regulation, and Government Regulation on Privacy Concerns: A Study of
Location-Based Services. Information Systems Research, 23(4), 1342-1363.
56. Zentall TR, and Galef BG, Jr. (eds) (1988) Social Learning: Psychological and Biological
Perspectives. Hillsdale NY: Erlbaum.
57. Zhang, J. (2010). The Sound of Silence: Observational Learning in the US Kidney
Market. Marketing Science, 29(2), 315-335.
58. Zhang, J., and Liu, P. (2012). Rational Herding in Microloan Markets. Management
Science, 58(5), 892-912.
59. Zhang, J., Liu, Y., and Chen, Y. (2014). Social Learning in Networks of Friends versus
Strangers. Marketing Science, Forthcoming.
60. Zhang, X., and Wang, C. (2012). Network Positions and Contributions to Online Public Goods:
The Case of Chinese Wikipedia. Journal of Management Information Systems, 29(2), 11-40.
61. Zhu, F., and Zhang, X. (2010). Impact of Online Consumer Reviews on Sales: The Moderating
Role of Product and Consumer Characteristics. Journal of Marketing, 74(2), 133-148.
28