How Does User Generated Content Influence Consumers’ New ... · NewRest ijt: dummy variable which equals 1 if consumer i hasn’t patronized restaurant j at time t. – X ijt: a

How Does User Generated Content Influence Consumers’ New Product Exploration? An Empirical Analysis of Consumer Search and Choice Behaviors

Wang QingliangGoh Khim Yong

Lu Xianghua

14-Jul-14 1 / 29

Presentation Agenda

• Introduction• Hypotheses• Data and Model• Estimation Results• Implications• Future Research

14-Jul-14 2 / 29

Introduction

• Social media platforms– Social network sites, micro-blogs, video sharing sites, product

review sites, discussion forums, chat rooms• E.g., Facebook, Twitter, LinkedIn, Youtube, Yelp

• User-Generated Content (UGC)– Reduce consumer uncertainty (Dellarocas 2003)

– Help consumers explore new product (Anderson 2006; Chen and Xie 2008)

Introduction Hypotheses Data and Model Estimation Results Implications Future Research 14-Jul-14 3 / 29

Research Questions

• How does consumers’ online UGC search influence new product exploration behaviors?

• How do consumers’ prior product consumption experiences affect their search or usage of online UGC to explore new products?

• To what extent does competition across online UGC of competing alternatives influence individual consumers’ purchase decision, especially when they explore new products?

14-Jul-14Introduction Hypotheses Data and Model Estimation Results Implications Future Research

4 / 29

Related Work

• Economic impact of online UGC– Impact of UGC on product sales (Chevalier and Mayzlin 2006; Chintagunta et al. 2010;

Goh, et al. 2013; Netzer et al. 2012) and stock returns (Luo 2009)

– Moderating factors on UGC impact (Forman et al. 2008; Zhu and Zhang 2010)

• Why consumers seek variety?– Satiation: internal or personal desire for variety (Givon 1984; McAlister 1982)

– External constraints • Such as multiple needs, multiple situations, multiple Uses (McAlister and

Pessemier 1982), price promotion (kahn and Louie 1990; Kahn and Raju 1991), retail environment (Menon and Kahn 1995)

– Preference uncertainty: within purchase occasion (Harlam and Lodish 1995).


5 / 29

Contributions

• Consumers’ search and usage online UGC• Consumers’ new product exploration• UGC may play different roles at different stages of

consumers’ decision process


6 / 29

Awareness Effect

• Online UGC as new information source– has become a pivotal source of product information to consumers

(ChannelAdvisor 2010; ComScore 2007)

– are helpful for consumers to identify the products that best match their idiosyncratic preferences (Anderson 2006; Chen and Xie 2008)

• H1: A consumer is more likely to choose a new product when he or she searches more new product alternatives from online UGC.


7 / 29

Experience Effect

• Prior experience and new product exploration– Consumers can easily evaluate the relative attractiveness of

products that they are unfamiliar with by comparing with the products they have purchased before.

• H2: A consumer is more likely to choose a new product when the new products he or she searches from online UGC can potentially provide higher utility than that of prior choice alternatives.


8 / 29

Competition Effect

• Consumers’ relative judgments (Laroche and Brisoux 1989; Laroche et al. 1994).

– Consumers’ choice for one product is not only influenced by online UGC of the focal product but also by those of competing products in a choice set (Li et al. 2011).

• H3: Information attributes from online UGC have a significant influence on a consumer’s choice decision among competing alternatives.


9 / 29

Competition Effect

• The moderating role of experience (Punj and Staelin 1983; Urbany et al. 1989).

– Consumers are likely to have more uncertainty on products which they are unfamiliar with or have not purchased before

• H4: Online UGC has a more significant influence when a consumer is choosing from a set of new products, compared to when he or she is choosing from a set of products with prior purchase experiences.


10 / 29

Data: Overview

• Restaurant patronage choices– Experience good– Frequently purchased product or service

• Data sets– Restaurants

• Reviews from Dianping.com– Taste, ambience, service (0=very bad, 4=very good)– Average price per person, recommended dishes, review texts and comments

• Attributes information– Location, coupon promotion, search ads, cuisine type, price category

– Consumers• Dining transaction records

– Consumer ID, restaurant ID, expenditure, transaction date• Activities on Dianping.com

– Log-in, posting, browsing


Data: Restaurant Reviews

Number of ratings

Average ratings for taste, ambience, service

14-Jul-14

Overall ratingAverage price per person

Restaurant name

Introduction Hypotheses Data and Model Estimation Results Implications Future Research 12 / 29


14-Jul-14

Distribution of overall ratings



Overall ratingRatings for taste, ambience and service, and average price per person

Review texts and comments

Recommended dishes


14 / 29

Sampled Data

• Sample periods: from Dec 2007 to March 2008 • 798 consumers

– Whose browsing history is observable– At least 3 transactions in the sample period– At least 1 transaction before the sample periods

• 3335 dining records in total• 215 restaurants


15 / 29

Data


16 / 29

Data


17 / 29

Econometric Model

• Consumers’ two-stage choice decision process– At the first stage, a consumer decides whether to choose from a

set of new restaurants that he or she has no prior consumption experience or from a set of restaurants patronized before (whether to patronize a new restaurant).

– At the second stage, the consumer decides which specific restaurant to patronize.

• Notations– i: consumer– j: restaurant– t: time, or purchase occasion– c: a group of restaurant


18 / 29

Whether to Patronize a New Restaurant?

• First stage utility function:

– αi : consumer-specific fixed effect, – Zit : a vector of control variables

• InterPurchaseTimeit, NumOfPersonit, NewRestSearchPercentit, NewRestSearchPercent_sqit, and OldRestNumit

– : “inclusive value”, which measures the expected value of the maximum utility from a set of new/old alternatives

14-Jul-14

1,

1,

newit c new i it it it

oldit c old it it

U Z IV

U IV

α γ λ ε

λ ε=

=

= + + +

= +2~ N( , )i αα α σ

and new oldit itIV IV


Whether to Patronize a New Restaurant?

• First stage choice– Assuming error term follows an extreme value distribution,

consumer i’s choice probability for the new products set and the old products set at time t will be:

14-Jul-14

,

,

newi it it

new oldi it it it

oldit

new oldi it it it

Z IV

it c new Z IV IV

IV

it c old Z IV IV

ePe e

ePe e

α γ λ

α γ λ λ

λ

α γ λ λ

+ +

= + +

= + +

=+

=+


Which Restaurant to Patronize?

• Second stage utility function:

– Dj : the fixed effect of alternative j. • We use fixed effect of cuisine type in estimation

– Rijt : a vector of variables that measure the influence of UGC. • Volumeijt, QualityRatingijt, VarianceOfQualityRatingijt, Priceijt, and

VarianceOfPriceijt. • We assume to capture consumer heterogeneity

– NewRestijt : dummy variable which equals 1 if consumer i hasn’t patronized restaurant j at time t.

– Xijt : a vector of control variables• TagNumj, SearchNumijt, TripNumijt, Promotionijt, and

UserRestDistanceij.

14-Jul-14

2| *ijt c j i ijt ijt ijt ijt ijtU D R R NewRest Xϖ β δ ϕ ε= + + + +

2~ N( )iβ β σ,


Inclusive Value

• Based on consumer i’s second stage utility function, we define the first stage variable of “inclusive value” as

– Represents the expected maximum utility consumer i can get from category c (Ben-Akiva and Lerman 1985; Chintagunta 1999).

– Defines how consumers’ first stage choice depends on the expected utility from the second stage choice.

14-Jul-14

*=ln j i ijt ijt ijt ijtD R R NewRest Xcit

cIV eϖ β δ ϕ+ + +

∑


Second Stage Choice

• Conditional on consumer i’s first stage choice, if we assume error term follows an extreme value distribution, the probability of consumer i choosing alternative j at time t is:

• The unconditional probability of consumer i choosing alternative j at time t is

– Iit, j=new is a dummy variable which equals 1 if alternative j is a new product for consumer i at time t.

14-Jul-14

*

| *

j i ijt ijt ijt ijt

j i ijt ijt ijt ijt

D R R NewRest X

ijt c D R R NewRest X

c

ePe

ϖ β δ ϕ

ϖ β δ ϕ

+ + +

+ + +=∑

, | , |(1 )ijt it j new ijt new new it j new ijt old oldP I P P I P P= == + −


Identification

• Endogeneity of consumer search– Consumers who have the intention to seek variety are more likely

to search for new products from online UGC– Control function approach (Petrin and Train 2010)

• Instruments:– Consumers’ prior contributions to the UGC site

– Instrument for square term (Wooldridge 2002)

14-Jul-14

3it it itNewRestSearchPercent ContriNumκ ε= +


Results from Two-Stage Choice Model

14-Jul-14

H1: Awareness Effect

H2: Experience Effect


Results from Two-Stage Choice Model

14-Jul-14

Mean for H3

H4


Results Summary

• Two stage decision model – First stage: whether to explore new product

• Awareness effect: consumers are more likely to explore a new product when they are exposed to online UGC

• Experience effect: the influence of online UGC on consumers’ new product exploration behaviors depends on consumers’ prior consumption experiences

– Second stage: which product to choose• Competition effect: online UGC plays a important role on consumer

product choice, especially for new products .


27 / 29

Theoretical Contributions

• Consumer search and online UGC– How researchers can make use of consumers’ search data to

explicitly model consumers’ decision process in the light of online UGC.

• UGC and consumers’ new product exploration– How individual consumers perceive and use UGC information to

guide their new product exploration and purchase decisions.

• Variety seeking literature – Online UGC as an external trigger of consumer variety seeking

behaviors.


28 / 29

Q & A

Thank you!

14-Jul-14 29 / 29

Results: Reduced-Form Evidence

14-Jul-14

instrument quality

Experience Effect

Awareness Effect


Alternative Model

• One stage choice– Consumers simultaneously evaluate all (both “new” and “old”)

choice alternatives and choose the alternative which yields the maximum utility.

• Utility function:

• Assuming error term follows an extreme value distribution, the probability of consumer i choosing alternative j at time t will be

14-Jul-14

,( )ijt j i ijt i it ijt it j new ijt ijtU D R Z R I Xϖ β α γ δ ϕ ε== + + + + ∗ + +

,

,

( )*

( )*

j i ijt i it ijt it j new ijt

j i ijt i it ijt it j new ijtijt

D R Z R I X

D R Z R I XeP

e

ϖ β α γ δ ϕ

ϖ β α γ δ ϕ

=

=

+ + + + +

+ + + + +=∑


Model Comparison


32 / 29

Results from One-Stage Choice Model

14-Jul-14

Comparable to the first stage of two stage choice model


Results from One-Stage Choice Model

14-Jul-14

The influence of UGC


Estimation

• Simulated maximum likelihood – 10 Halton draws for random parameters– Tolerance for convergence: 1.e-4

• The model parameters for two stages are simultaneouslyestimated– The interdependence of parameters, and – The interdependence of the two stage decisions,


35 / 29

Practical Implications

• For firm managers– First, it is beneficial to stimulate consumers to generate more

word of mouth information for new products.– Second, it is necessary and important to take consumers’ prior

consumption experiences into consideration when influencing consume choice via UGC

– Third, positive online word of mouth not only increases a firm’s customer base but can also mitigate against customer defections.

• For designers of product recommendation systems– Take individual consumers’ consumption experience into

consideration when designing recommendation systems.– Show how this product is relatively rated in the market


36 / 29

Limitations and Future Research

• Modeling consumers’ decision of information search• Incorporating consumers’ product quality learning

– We have a separate paper to study this question

• The influence of review texts and comments • The influence of alternative unobserved sources of

information, such as information from other UGC platforms and offline word of mouth


Data


38 / 29

Documents

How Does User Generated Content Influence Consumers’ New ... · NewRest ijt: dummy variable which equals 1 if consumer i hasn’t patronized restaurant j at time t. – X ijt: a