Does it matter what people say about you: the impact of the content of buzz on firm performance

PDF

Download

Open document

Flip pages

Summarize

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

DOES IT MATTER WHAT PEOPLE SAY ABOUT YOU: THE IMPACT OF THE
CONTENT OF BUZZ ON FIRM PERFORMANCE

by
Seema Pai

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(BUSINESS ADMINISTRATION)

August 2008

Copyright 2008 Seema Pai
ii

DEDICATION

To the memory of
my brother,
R. Shivaramakrishnan.

iii

ACKNOWLEDGEMENTS

Though only my name appears on the cover of this dissertation, a great many people have
contributed to its production. I owe my gratitude to all those people who have made this
dissertation possible and because of whom my graduate experience has been one that I
will cherish forever.

My deepest gratitude is to my advisor, Dr. S. Siddarth. I have been amazingly fortunate
to have an advisor who gave me the freedom to explore on my own and at the same time
the guidance to recover when my steps faltered. Siddarth taught me how to question
thoughts and express ideas. His patience and support helped me overcome many crisis
situations and finish this dissertation. I hope one day to become as good an advisor to my
students as Siddarth has been to me.

My committee members, Dr. Gerry Tellis, Dr. Shantanu Dutta, Dr. Roger Moon and Dr.
Natalie Mizik have always been there to listen and give advice. I am deeply grateful to
them for always being generous with their time and for their constructive criticism at
various stages of this research.

I am also indebted to the following staff at the Marshall School of Business for their
various forms of support during my graduate study – Elizabeth Mathew, Yvonne King,
Robin Gaines and Michelle Silver Lee.
iv

I am also grateful to Young & Rubicam for providing me with the data for parts of this
dissertation.

Many friends have helped me stay sane through these difficult years. Their support and
care helped me overcome setbacks and stay focused on my graduate study. I greatly value
their friendship and I deeply appreciate their belief in me. I am particularly grateful to
Deepa Chandrasekaran and Ram Ranganathan as well as Hae Eun Chun and Hojoon Lee
for providing me with a home and family in Los Angeles when I needed it the most.

Most importantly, none of this would have been possible without the love and patience of
my family. I have to give a special mention for the support given by my parents and my
husband, Dilip, for always believing in me and helping in myriad ways, both tangible and
intangible.

v

TABLE OF CONTENTS
Dedication ii
Acknowledgements iii
List of Tables vi
List of Figures vii
Abstract viii
Chapter 1: The Impact of Word-Of-Mouth On Purchase Decisions: The
Case Of Motion Pictures

Chapter 1 Abstract 1
Chapter 2: The Impact of Mass Media on Corporate Reputation
Chapter 2 Introduction 46
Bibliography 90
Appendix : Checking the Reliability of Computerized Text-Analysis 100

vi

LIST OF TABLES

Table 1.1: Movie Characteristics Included in the Model 21
Table 1.2: Granger Causality Test Results 25
Table 1.3: Dickey-Fuller Test Results 26
Table 1.4: VAR Model Fit Comparisons 27
Table 1.5: Word-of-Mouth Effects on Box Office Revenues 28
Table 1.6: Short-Term versus Long-Term Elasticity of Revenues 35
Table 1.7: Comparison of Forecasting Models RAE 41
Table 2.1: Dickey-Fuller Test Results 76
Table 2.2: Granger Causality Test Results 77
Table 2.3: VAR Model Fit Results 78
Table 2.4: Elasticity of Reputation to PR & Independent Media Coverage 83

vii

LIST OF FIGURES

Figure 1.1: Distribution of Numerical Ratings on Yahoo! Movies &
IMDB

12
Figure 1.2a: IRF: Response of Revenues to Shock in Screens 36
Figure 1.2b: IRF: Response of Revenues to Shock in Advertising 37
Figure 1.2c: Response of Revenues to Shock in Buzz Volume 38
Figure 1.2d: Response of Revenues to Shock in Positive Buzz 38
Figure 1.2e: Response of Revenues to Shock in Negative Buzz 39
Figure 2.1a: Response of CR to Shock in PR Volume 85
Figure 2.1b: Response of CR to Shock in MC Activity 86

viii

ABSTRACT

Word-of-Mouth (WoM) has been recognized as one of the most influential sources of
information transmission. Recent advances in information technology have profoundly
changed the way in which WoM is transmitted leading to resurgence in interest in this
topic among practitioners and academics.

A major challenge facing researchers in the area of online WoM is to find a feasible way
to measure the content of buzz. In my first essay, I introduce a novel approach to
analyzing the content of buzz using a combination of human-coding, dictionary-
development and computerized sentiment mining. I explore the relationship between box-
office revenues and WoM measures for nearly 200 movies using a Vector Autoregression
(VAR) model.

Several interesting findings emerge from the empirical analysis. First, the results show
that the content of buzz has a significant impact on revenues. Second, the negative effects
of bad buzz are greater than the positive effects of favorable buzz. Third, the elasticity of
the content measures is comparable to that of advertising. This research also shows that
managers may be able to incorporate the content of pre-release buzz to improve opening
weekend box-office forecasts.

Firms use various elements of the communications including advertising, promotion,
personal selling, and public relations (PR) to drive various critical performance metrics
ix

including corporate reputation. While a majority of the marketing literature focuses on
the first three of these elements, there has been little to no research on the impact of PR.
In my second essay, I use a structural VAR model to examine the relation among and the
dynamic impact of the three key drivers of corporate reputation: independent mass media,
advertising, and PR.

My Corporate Esteem metric comes from Young & Rubicam’s Brand Asset Valuator
model. Data on advertising comes from the TNS AdSpender database. Data on mass
media coverage and PR come from LexisNexis and Factiva.

Model estimation yields several interesting findings. First, both the valence and the
prominence of the content of media articles are found to be important determinants of
reputation, with negative coverage having a greater and longer-lasting impact than
positive media coverage. Second, PR is found to have a significant impact on corporate
reputation and its elasticity is greater than that of advertising and of independent media-
reports. Finally, I find evidence that advertising and public relations tend to be used as
complementary communication tools.
1

CHAPTER 1: THE IMPACT OF WORD-OF-MOUTH ON PURCHASE DECISIONS:
THE CASE OF MOTION PICTURES

CHAPTER 1 ABSTRACT

The authors develop a new approach to process the content of both online and offline
sources of word-of-mouth or buzz using computerized text-analysis techniques and study
how movie buzz impacts box-office performance. The content of about 500,000 offline
and online sources of communication, including user and critic reviews, magazine
articles and blog postings, pertaining to nearly 200 movies released between January and
December, 2004 are analyzed to produce measures of both the volume and valence of the
source material. The relationship between box-office receipts and the word-of-mouth
measures is studied using a Vector Autoregression (VAR) model that accounts for the
potential endogeneity between box-office revenues, the volume and valence of buzz, the
number of screens allocated to the movie and the advertising expenditure.

Several interesting results emerge from the empirical analysis. First, in contrast to prior
empirical research in this area, the authors find that the buzz valence significantly
impacts box-office revenues. Further, consistent with prior theory, negative buzz is found
to have a greater impact on sales than positive buzz in both the short- and long-run.
Second, Granger Causality Tests demonstrate that box-office revenues, volume and
valence of buzz, advertising and the number of screens are endogenous suggesting that
models that ignore these feedback effects may produce biased estimates of the impact of
buzz. Third, the impact of the content of buzz is found to depend upon its source: user
2

content has a significant impact on revenues but critic comments do not. The authors
also show that incorporating pre-release buzz significantly improves the accuracy of
forecasts of opening week box-office performance.

Keywords: Buzz marketing; Word-of-mouth communication; Content Analysis; Movie
Forecasting.

3

1. INTRODUCTION

Word-of-mouth or “Buzz” is a phenomenon that has created several marketing legends.
The traffic-stopping retro Beetle, the addictive Pokémon, cuddly Beanie Babies, the hair-
raising Blair Witch Project are all examples of blockbuster commercial success driven by
customer hype. People like to share their experiences with one another – the restaurant
where they ate lunch, the movie they watched over the weekend, the computer they just
bought – and when those experiences are favorable, the recommendations can snowball,
resulting in runaway success.

Word of Mouth (WoM) has been recognized as one of the most influential sources of
information transmission since the beginning of society, especially for experience goods
(Godes & Mayzlin, 2004) and has been the subject of extensive research in marketing
(Anderson, 1998; Bansal & Voyer, 2000; Brown & Reingen, 1987). This prior literature
has collectively identified three key dimensions of WoM communication that determine
its effectiveness: who (Bone, 1995; Engel, Blackwell & Kegerreis, 1969), how much
(Arndt, 1967; Herr, Kardes & Kim, 1991) and what (Crane, 1989; Feldman & Spencer,
1965). The who dimension refers to the source of the WoM information, how much refers
to the volume or quantity of WoM and the what refers to the content or the body of
meaning conveyed by the symbolic elements such as the verbal, musical and pictorial
cues in the WoM communication.

4

Recent technological advances have enabled consumers to easily and freely access
information and exchange opinions on companies, products and services on an
unprecedented scale in real time. The emergence of online communities, discussion
groups, blogs and opinion websites has changed how consumers interact with one another
and provided new avenues for word-of-mouth communications. These changes have led
to a renewed interest in measuring buzz and understanding how it impacts firm
performance.

Most of the recent literature on the impact of word-of-mouth has focused on the ‘how
much’ aspect of WOM, i.e., on establishing that buzz volume significantly impacts
product sales (Chatterjee, 2001; Chen et al, 2005; Chevalier & Mayzlin, 2005; Dellarocas
et al, 2004; Godes & Mayzlin, 2004). However, this research has either ignored the
“what” aspect of WOM or has found that the content or valence of buzz does not impact
performance. This is puzzling given the large body of research in consumer behavior and
communications showing that content is an important driver of the effectiveness of word
of mouth.

A shared characteristic of the above studies is that buzz valence is operationalized using
easy-to-process summary measures, such as the star ratings that accompany a review,
while the actual textual content of the reviews is completely ignored. This approach
discards potentially valuable diagnostic information in the actual text and relies on
measures that may only be imperfectly correlated to the actual quality, which may
5

explain the null results for content. A key contribution of my research is to develop richer
measures of buzz valence based on a computerized text-analysis of various sources of
buzz and to show how buzz valence impacts product sales. Additionally, with regards to
the ‘who’ dimension, prior research has generally focused on a single source of buzz,
namely, user reviews about a product. However, because the source of the
communication itself may be an important indicator of its effectiveness (Fishbein &
Ajzen, 1975; Horland & Weiss, 1951; Sternthal, Dholakia & Leavitt, 1978), I study the
effects of different sources of communication, both online and offline.

The US motion picture industry serves as the empirical context for my study. This
industry is particularly suitable for my research because a) movies are classic experience
goods and buzz is likely to strongly influence consumer choice; and b) an extensive
literature on the motion picture industry in both marketing and economics provides
benchmark models of box-office performance against which my results may be compared
(DeVany & Walls, 1997, 1999 & 2004; Eliashberg & Shugan, 1997; Radas & Shugan,
1998; Elberse & Eliashberg, 2003; Swami, Eliashberg & Weinberg, 1999).

I collect box-office revenues and buzz data for nearly 200 movies released between
January and December, 2004. Because I treat buzz as any communication outside of the
firm’s own communication activities, I include user and critic reviews as well as articles
and postings from a wide variety of online and offline sources. Specifically, I include
user reviews from www.imdb.com and Yahoo! Movies, critic reviews from
6

www.rottentomatoes.com (a website where critic reviews from various sources are
aggregated) and also newspaper and magazine articles from a variety of leading
publications across the United States, which are obtained from sources including
www.lexis-nexis.com and www.factiva.org. I also collect blog postings about the movie
or the cast members or director from technorati.com. In all, my empirical analysis is
based on about 500,000 individual items about the movies in our sample.

A sub-sample of this material is first used to build a custom dictionary for the
computerized text-analysis, which then forms the basis for the detailed content analysis
of the remaining reviews. Recognizing the potential endogeneity of several variables
including the number of screens, advertising and the volume of buzz, I propose and
estimate a Vector Auto Regression (VAR) model of movie revenues that incorporates
these content measures, the volume of buzz, the number of screens allotted to a movie,
and several other control variables to obtain insights into how the volume and content of
buzz influences the box office performance of movies. Finally, I show how the volume
and valence of buzz helps to improve early forecasts of box-office revenues.

This research makes four key contributions. To my knowledge, it is the first to
demonstrate how computerized text-analysis can be used to develop valid measures of the
valence of user communication. The measurement of buzz has been identified as a major
problem hindering research on this phenomenon (Godes & Mayzlin, 2004; Liu, 2006)
and my approach provides a feasible and reliable approach to carry out this analysis. In
7

contrast to previous work based on human raters (Liu, 2006), the computerized approach
can more easily analyze much larger volumes of material. The proposed methodology
could easily be applied to other interesting marketing problems such as measuring the
impact of public relations activities and media stories that may appear about a firm or to
monitor product satisfaction in real-time based on consumer postings.

Second, to my knowledge, this research is the first to empirically validate the existence of
negativity bias in online buzz in a field setting. Several studies in marketing as well as
social psychology have shown that there is a general bias on both innate predispositions
and experience to give greater weight to negative entities. This is manifested in several
ways including negative potency, steeper negative gradients and negativity dominance
(Rozin & Royzman, 2001; Ito, Larsen, Smith & Cacioppo, 1998). Using my content
analysis approach, I am able to separately identify the positive and negative information
contained within each message. A noteworthy finding from the empirical analysis is that
negative sentiment not only has a bigger impact on sales than positive sentiment but that
its effect lasts for a longer time. As far as I know, this is the first study to highlight the
dynamic effects of negative information.

I include a wide range of buzz sources in order to compare their relative impact on box-
office performance. Extensive research has documented the role of expert or critic
reviews within the movie (Eliashberg & Shugan, 1997; Basuroy, Chatterjee & Ravid,
2003) and other contexts (Bansal & Voyer, 2000; Heath, Motta & Petre, 2006). I show
8

that consumer response depends upon whether the source is “official,” i.e. originating
from the movie studio, or from domain experts, i.e., critics, or from their peers
(Eliashberg, Elberse & Leenders, 2006). In contrast to some prior findings (Bone, 1995;
Bansal & Voyer, 2000; Reinstein & Snyder, 2005) on the role of expertise in driving the
effectiveness of WoM communication, I find that user reviews impact revenues more
than critic reviews. A possible explanation for these findings is that in artistic product
categories, consumers may pay less attention to expert opinions because they perceive
critics to use evaluation criteria different from “regular” users (Holbrook 1999).

Third, this research sheds light on how buzz influences a firm’s actions and how studios,
in turn, utilize the marketing mix to drive buzz for their movies. Specifically, I find that
rather than using buzz and advertising as substitutes, movie studios respond very
selectively to buzz by only increasing advertising spending when the level of unfavorable
buzz is high.

Finally, I demonstrate that incorporating the content of pre-release buzz significantly
improves the accuracy of predictions of opening weekend box-office performance, which
are typically the least accurate since managers have very little to no data to use to make
these forecasts. Interestingly, opening weekend forecasts are crucially important because
they typically represent the largest proportion of overall box-office revenues and because
these forecasts influence important decisions such as advertising and distribution. Thus,
9

our approach enables managers to monitor buzz in real time and use these buzz measures
to make important decisions about the marketing mix for the movie.

The remainder of this paper is organized as follows. Section 2 provides a detailed
discussion of how we use text analysis to measure the valence of communications about a
movie. Section 3 explains the model and the estimation procedure. Section 4 describes
the data and variables and Section 5 discusses the results from the empirical analysis.
Section 6 compares forecasts for opening-week box-office revenues from a model that
incorporates pre-release buzz to several benchmark models. Section 7 concludes by
discussing the key contributions of this work and potential directions for future research.

2. CONTENT ANALYSIS

As previously discussed, the internet has made it extremely easy for consumers to go
online and post their opinions on products and services. While this has made it possible
for marketers and researchers to observe WoM in a natural, real-world setting, the
exponential increase in the volume of material available has also made it more difficult to
process the content of the WoM as highlighted by Godes & Mayzlin (2004) and Liu
(2006). One important objective of this research is to apply innovations in the area of
natural language processing and text-analysis to process this large volume of material and
develop measures for its content, while avoiding the use of sampling (loss of information)
the use of human coders (expensive and time-consuming).
10

2.1 Shortcomings of Summary Measures

Prior research in the areas of word of mouth as well as customer satisfaction agree that
the valence of the communication is a basic measure of the content of WoM
communication (Calder, Phillips and Tybout, 1981; Wirtz and Chew, 2002) and plays a
crucial role in determining the effect of the communication on sales. Despite this, most of
the recent work on online buzz has mostly used relatively simple, summary, measures of
the valence of online reviews.

For instance, online reviewers typically provide numerical ratings of the product or
service along with a descriptive verbal review. Previous research by Chevalier &
Mayzlin, (2005), Dellarocas, Awad & Zhang (2005) and Hu, Pavlou & Zhang (2006) has
generally ignored the textual content in the review and used the numerical or “star”
ratings as a proxy for the content of the review, thus avoiding the problem of analyzing
the staggering number of reviews that are written, several hundred reviews for a single
movie on a single website being the norm.

However, there are several reasons why the overall numerical or star ratings may not be
the best measure of the valence of the user reviews. First, ignoring the actual comments
involves throwing away content that may impact a consumer reading the review. Second,
as shown by Hu, Pavlou & Zhang (2006), the average or mean of the online numerical
ratings (the measure used in most of the research on online WoM) may not reflect the
11

true quality of the product or service. They report that the underlying distribution of the
valence of the reviews for most products have a bimodal and non-normal distribution.
Consequently, using the mean of the ratings for forecasting or any other kind of
econometric analysis could be potentially misleading.

A third issue of concern is that the distribution of star ratings on most websites has been
found to be extremely positively skewed. For example, Resnick and Zeckhauser (2003)
find that of only 0.5% of the buyers on Ebay.com provide negative or neutral ratings
1
.
Godes & Mayzlin (2003) find that in their sub-sample (10% of the total sample of
conversations used in the analysis) of online conversations about TV shows, 22% of the
reviews were mixed in their valence. Additionally, of the conversations that were clearly
positive or negative, over 70% of the posts were in fact positive. Chevalier & Mayzlin
(2005) find in their study of book reviews on Amazon.com and Barnes&Noble.com that
over 72% of the reviews on Amazon and over 86% of the reviews on the Barnes&Nobles
website carry either a 4-star or 5-star rating. Thus, it seems that in most contexts, star
ratings do not exhibit much variance on the valence of the reviews. Within our own
sample of movies, as shown in Figure 1.1, we notice that over 75% of the reviews on
IMDB have ratings of seven or greater (on a scale of 1 to 10) while over 82% of movies
on Yahoo Movies had a rating of B or higher on a scale from F to A.

1
This is from the total number of buyers on Ebay.com and not just people who posted a
review.
12

Figure 1.1: Distribution of Numerical Ratings on Yahoo! Movies & IMDB

0
5
10
15
20
25
30
35
40
IMDB
Percentages
0
10
20
30
40
50
60
A B C D E F
Yahoo! Movies
Percentages
13

A fourth critical shortcoming of the numerical ratings is that they may not fully represent
the valence indicated by the textual content in the review. Even a cursory examination of
a random sample of reviews reveals that many reviews are in fact mixed in their valence,
even if the overall rating is either strongly positive or negative. Thus, reviewers may
articulate both the strengths and the weaknesses of the movie being reviewed, as
illustrated below by two sample reviews from our dataset, which have the same star
rating but convey very different sentiments. Whereas the first review is entirely positive
in its valence, the second review contains both positive as well as negative content and
seems more mixed in terms of its overall valence.

Review 1: I figured I'd like 50 First Dates because of the chemistry demonstrated by
Sandler and Barrymore in “The Wedding Singer”. Once again, Sandler and Barrymore
are naturals on screen. This is the best Adam Sandler movie I've ever seen, with great
acting on the lead's parts. I really enjoyed it, and I think it has a great ending.

Review 2: Sandler pulls it off but just barely. I have to say that this movie was very sweet
and the acting was very good. Unfortunately, the plot and direction were lacking.

In line with the example above, prior researchers (Liu, 2006) who used human judges to
code the valence of a subsample of reviews find that only about 1/3
rd
of the reviews
during the opening week are either clearly positive or negative in their valence while the
remaining 2/3rds are mixed in their valence. Thus, it seems that the real information
14

contained in a review (in terms of how informative it is to readers of the review) consists
of both the positive and negative aspects of the review.

2.2 Content Analysis Procedure:

In order to test the relative diagnosticity of these two different aspects of valence and to
provide a more nuanced and rich measure of the content of the reviews, I develop a
content-analysis tool that quantifies the favorability as well as the unfavorability of the
review. My approach builds on developments in the area of sentiment mining in a variety
of other disciplines and uses statistical and natural language processing techniques to
elicit emotive sentiment from a posted message. The steps involved in this process are
outlined below.

First, I use a “web-scraper” program to download different pieces of communication
from the internet. A multi-step approach is used to pre-process each of these elements in
order to make later analysis easier to carry out. Specifically, I first remove all HTML tags
from the body of the message as these are often concatenated with words that are
important determinants of valence. Second, I look for abbreviations and expand them to
their full form, making the representation of phrases with abbreviated words common
across the message. For example, the word “ain’t” is replaced with “are not”, “it’s” is
replaced with “it is”, etc. Finally, I handle negation words by detecting words such as
“not”, “never”, “no”, etc. and then tagging the rest of the words in the sentence after the
15

negation with markers, so as to ensure that the valence inferred from the sentence is
opposite to the one implied by the adjectives that are found in it. These steps deliver a
cleaner set of messages for valence inference.

This cleaned-up buzz data is then fed to a computer program that determines the level of
positive and negative content in each body of text. A dictionary database supports the
valence-classification algorithm. The first is an electronic “dictionary” that provides base
language data. This is useful in determining the nature of a word, i.e. noun, adjective,
adverbs etc. In addition, to exploit parts-of-speech usage in buzz messages, a second
dictionary was used to detect adjectives and adverbs. This dictionary is called
CUVOALD (Computer Usable Version of the Oxford Advanced Learner’s Dictionary). It
contains parts-of-speech tagging information and the computer program uses this
dictionary to analyze messages for grammatical information.

Implementing the procedure requires that each word in the text is first ‘stemmed’, i.e.
mapped to its root. For example, the root of ‘fascinating’ is fascinate. My goal is to use
the computer program to measure the degree of positively and negatively valenced
information in a particular buzz message. I use an algorithm that counts the number of
positive and negative words within the text and the program returns the count of positive
and negative phrases found in the body of each message. These message-specific counts
form the basis of the valence measures that are discussed in greater detail in Section 4.

16

Although the use of this type of automated content analysis makes it significantly easier
to analyze the content of word-of-mouth; it also brings up issues of reliability and
accuracy. I performed multiple reliability checks on the text-analysis software in order to
ensure that these measures were consistent with ratings provided by human coders.
Details of these tests are described in the Appendix.

3. MODEL

In order to measure the impact of the volume and the valence of buzz on box office
performance I turn to a time series regression approach. Dynamic effects are captured by
using lags of the marketing actions via an autoregressive-distributed lag model (e.g.
Hanssens et al. 2001). However, even this type of model does not capture the feedback
effects from box-office revenues to advertising, screens and the volume and valence of
the buzz. For instance, advertising may serve to directly increase revenues, engender
greater buzz (thereby indirectly increasing revenues), and also increase the number of
screens that are allocated to the movie. These types of interactions may either occur
immediately or play out dynamically over several days and can be tested using a VAR
model, which is commonly applied to quantify short- and long-run market response
(Dekimpe & Hanssens, 1999).

I note two features of this approach. First, the endogenous treatment of buzz implies that
it is explained both by its own past history and that of box-office revenues. In other
17

words, this dynamic model estimates the baseline of each endogenous variable and
forecasts its future values based on the dynamic interactions of all jointly endogenous
variables. Second, dynamic effects are not a priori restricted in terms of time, sign or
magnitude. The sign and magnitude of any dynamic effect need not follow any particular
pattern – such as the imposed exponential decay pattern from Koyck-type models.

Compared to alternate specifications, VAR models are especially well-suited to measure
dynamic interactions among performance and marketing mix variables. Recently, VAR
models have been used to analyze a wide variety of long-term marketing effects
including advertising, price promotions and new product introductions (e.g. Dekimpe &
Hanssens, 1999; Pauwels et al. 2002; Srinivasan et al. 2004) but have not, to my
knowledge, been applied in the movie context.

VAR Model Specification

I construct a VAR model where the volume of word-of-mouth, advertising and the
number of screens are all determined endogenously with movie sales. I draw upon
extensive prior research on movie box-office performance to determine the exact
variables to include in the VAR model (Sawhney & Eliashberg, 1996; Elberse &
Eliashberg, 2003; Dellarocas, Awad & Zhang, 2003; Liu, 2006). The data section
provides details on the complete set of variables included. The model is specified as:

18

!"#$

%&'##()

*++, -

. )*+

/#0*+

12$0

3

4
!"#$
4
%&'##()
4
*++, -
4
. )*+
4
/#0*+
4
12$0

5 6
3 7

8
99
8
9:
8
9;
8
9<
8
9=
8
9>
8
:9
8
::
8
99
8
99
8
99
8
99
8
;9
8
;:
8
99
8
99
8
99
8
99
8
<9
8
<:
8
99
8
99
8
99
8
99
8
=9
8
=:
8
99
8
99
8
99
8
99
8
>9
8
>:
8
99
8
99
8
99
8
99

?
@A9

B9

B9

B9

B9

B9

B9

3

D
99
0 0 0 0 0
0 D
::
0 0 0 0
0 0 D
99
0 0 0
0 0 0 D
99
0 0
0 0 0 0 D
99
0
0 0 0 0 0 D
99

F
!"#$
F
%&'##()
F
*++, -
F
. )*+
F
/#0*+
F
12$0

3

G
!"#$,
G
%&'##(),
G
*++, -,
G
. )*+,
G
/#0*+,
G
12$0,

where
BoxRev = Box office Revenues,
i = movie subscript,
t = day subscript,
α = constant term,

BuzzVol = Volume of buzz across different sources,
PosBuzz = Average % of positive content across buzz sources,
NegBuzz = Average % of negative content across buzz sources,
Screens = Number of screens allocated to the movie,
19

Advertising = Advertising expenditure in dollars,
T = deterministic time trend,
d
i
= dummies for days of the week (using Friday as the benchmark),
d
j
= dummies for month of the year (using December as the benchmark),
MovieChar = vector of movie characteristics,
J = number of lags of the dependent variable needed to ensure that the residuals, epsilon,
are white-noise errors (i.e. without any residual autocorrelation).

In the above model, X
BoxREV
, X
Screens
, X
BuzzVOL
, X
PosBuz,
X
NegBuz
and X
Advg
are the
exogenous variables associated with each of the endogenous variables. These consist
primarily of various movie characteristics that the prior literature has identified as having
an impact on box-office revenues, screens, word of mouth and advertising. A detailed
discussion of how the variables are operationalized is provided in the next section.

Estimation

The empirical analysis proceeds as follows. I begin by selecting between a single-
equation model, where box-office revenues are explained by movie characteristics, buzz
and other independent variables, and a full dynamic system, where multiple variables are
allowed to impact one another over time, by investigating the Granger causality between
variables (Granger, 1969, Hanssens et al. 2001). Granger causality implies that knowing
the history of a variable X helps explain a variable Y, over and above Y’s own history.
20

This type of ‘temporal causality’ is the closest proxy for causality that can be gained from
studying the variables’ time series.

I perform a series of Granger causality tests on each pair of key variables. I note that a
wrong choice for the number of lags in the test may erroneously conclude the absence of
Granger causality (e.g. Hanssens, 1980). Therefore, I run the causality tests for lags up to
15 time periods and report the results for the lag that has the highest significance for
Granger causality. If box-office revenues do Granger-cause some of the other key
variables such as the number of screens, buzz volume and so on, then I need to capture
these interactions in a full dynamic system.

Then, I test for evolution of all the variables in our study so that I can choose a Vector
Error Correction (VEC) model instead of a VAR if I find evidence of co-evolution. I take
logarithms on both sides while estimating the VAR model, which allows me to directly
interpret the short- and long-term performance impact estimates as elasticities (Nijs et al.
2001). Next, impulse response functions are derived from the VAR model. The impulse
response functions represent the over-time impact of a unit shock to any endogenous
variable on the other endogenous variables. Following Dekimpe & Hanssens (1999), I
use generalized impulse response functions (or simultaneous shocking) to ensure that the
ordering of the variables in the system does not affect the results. Following Nijs et al.
(2001), I determine the duration of the shock (maximum lag k) as the last period in which
the impulse response function has a t-statistic with an absolute value that is greater than
21

1. In the context of my research questions, I use impulse response functions to
disentangle the short and long-run effects of buzz and advertising on box-office revenues.

4. DATA, VARIABLES AND MEASURES

I use daily box-office performance for the top 25 movies playing at the box-office in any
week, for every week during the period January 2004 to December 2004 ( source:
www.boxofficemojo.com), which yields a total of 193 movies and 8256 data points. I
also collected information on other movie characteristics which I convert into variables.
A complete list of variables and their sources is described in the table below;

Table 1.1 Movie Characteristics Included in the Model
Characteristic Variable Name Source Timing
Number of Screens SCREENS Boxofficemojo.com Daily
Budget BUDGET Boxofficemojo.com Pre-release
Star Power STARPOW Hollywood Reporter Pre-release
Director Power DIRPOW Hollywood Reporter Pre-release
AdvertisingSpending ADVG TNS Stradegy Database Daily
Major Distributor
Dummy
DISTDUMM
Boxofficemojo.com Pre-release
Genre GENRE Boxofficemojo.com Pre-release
Competition for
Screens
SCRCOMP,
AVGAGE
Boxofficemojo.com Daily
Competition for
Revenues
REVCOMP
Boxofficemojo.com Daily
MPAA Rating MPAARAT Boxofficemojo.com Pre-release

I also include three dummy variables, one for the month, one for weekday vs. weekend
and the other for national holidays to account for seasonal fluctuations which are known
22

to occur in the movie industry (Moul, 2006). Most of these variables are standard and
self-explanatory but the following variables bear additional explanation. My model
includes two variables to represent the competition faced by a movie for the allocation of
screens. The first is, SCRCOMP, the number of new releases in the current week’s Top
25 and the second is the average age of the movies in the week’s Top 25 list, AVGAGE.
The competition for revenues is captured by the variable REVCOMP, which is calculated
as the number of movies in the current week’s Top 25 list that are of the same genre or
have the same MPAA rating as the movie under consideration, divided by the average
age of the competing movies. REVCOMP reflects the notion that movies from the same
genre are closer substitutes and, within this subset of similar movies, older movies are
less of a competitive threat than new movies. These measures of competitive effects are
consistent with prior literature on box-office performance (Elberse & Eliashberg, 2003).
Apart from this data on box-office performance and movie characteristics, I also
collected data on word-of-mouth or buzz related to the movie from a number of online as
well as offline sources. This yielded over half a million pieces of communication about
the movies in the sample, which I believe is the most comprehensive dataset on online
and off line word-of-mouth. The sources that I used were:
• User reviews on http://www.imdb.com and http://movies.yahoo.com
(about 443,000)
• Critic reviews on http://movies.yahoo.com and
http://www.rottentomatoes.com (about 10,000)
• Blog mentions from http://www.technorati.com (about 30,000)
23

• Magazine articles about the movie from http://www.lexis-nexis.com and
http://www.factiva.com (about 7,000)

Buzz Variables: I construct several variables based on this data. The variable VOLBUZZ
captures the volume of buzz about a movie and is computed as the sum, of the number of
reviews/articles about the movie at time t appearing in a particular source. Thus, I obtain
separate measures of the volume of buzz for each source namely, VOLUSER for user
reviews, VOLNEWSP for newspaper articles, VOLMAG for magazine articles,
VOLCRIT for critic reviews and VOLBLOG for Blog mentions.

For every source of content about the movie (e.g User Reviews, Critics), two measures of
the valence of buzz are developed. The positive (negative) valence of an individual
message is captured by the proportion of positive (negative) words contained within that
message, which can readily be calculated from the word counts output by my text-
analysis procedure. The actual variables used in the model averages these proportions
over all messages from a particular source for a particular day yielding the measures
POSUSER, POSCRIT, POSNEWSP, POSMAG and NEGUSER, NEGCRIT,
NEGNEWSP and NEGMAG. For example, positive user sentiment at time t, can be
calculated as:

IJK

7 IJK
L
M
LA9
, N1P
24

where POSUSER
mt
is the proportion of positive content for message m posted at time t,
and the sum is over all M user messages that appear at time t. A similar summation over
the negative content of the M messages, and over the other sources, i.e., critics,
newspaper articles, and magazine articles generate the other variables in my model.

5. RESULTS

Endogeneity (Granger Causality Test Results)

I begin by examining the possible interdependence between revenues and other
marketing mix elements including advertising, buzz and screens using Granger Causality
Tests (Granger, 1969). The results from the Granger Causality tests are summarized in
Table 1.2. Each cell contains the minimum p-value obtained from the causality tests
conducted from one lag to 15 lags and the numbers in bold-face indicate that the variable
in the row Granger-causes the variable in the column. Thus, all of the variables Granger-
cause Revenues, while negative buzz is only impacted by Revenues, Screens and
Advertising. Overall, the results from the Granger Causality tests in Table 1.2 clearly
indicate that these variables simultaneously drive each other over time and confirm the
need to employ a full dynamic system, as in a VAR-model, to account for this type of
long-term feedback.

25

TABLE 1.2 : Results of the Granger-Causality Tests (Minimum p-
value across 15 lags)
DV is
Granger
Caused by:
Revenues Screens Advertising
Buzz
Volume
Positive
Buzz
Negative
Buzz
Revenues .00 .00 .00 .00 .00
Screens .02 .01 .04 .12 .17
Advertising .06 .14 .03 .06 .05
Buzz
Volume
.03 .01 .00 .26 .19
Positive
Buzz
.02 .04 .05 .07 .18
Negative
Buzz
.04 .01 .09 .11 .16
The bold faced entries denote pairs of variables where the variable in the column variable granger causes
the row variable

Next, I proceed by testing for the presence of unit roots in the data in order to select
between a Vector Autoregression Model and a Vector Error Correction Model for the
empirical analysis.

Dickey-Fuller Test Results

Augmented Dickey-Fuller tests recommended by Enders (1995) were used to test for the
presence of unit roots in the data. Results of both tests confirmed trend stationarity in all
series, i.e., all series were stationary after controlling for deterministic trend. Table 1.3
provides the details of the unit-root test results. I therefore proceed with a VAR (p) model
for the rest of the analysis with the order of the VAR model determined by the Akaike
26

Information Criterion (AIC) (Lutkepohl, 1993). Stationary variables are included in
levels while difference-stationary variables are included in differences.

TABLE 1.3: Results of Unit Root Tests
ADF Stat 5% Critical Value Unit Root?
Revenues −4.18 −3.44 No
Screens −5.17 −3.44 No
Advertising −3.99 −3.44 No
Buzz Volume −5.62 −3.44 No
Positive Buzz −4.07 −3.44 No
Negative Buzz −4.19 −3.44 No

VAR MODEL Estimation and Fit Statistics

Fit results for the estimated VAR models, with the appropriate lags determined by the
Akaike Information Criterion, are presented in Table 1.3 along with several benchmark
models. Model 1 employs the star ratings that accompanying a review to proxy for its
content while Model 2 incorporate the content of only a sub-sample of the reviews for
every time period. Specifically, the content measures for Model 2 were based on a
randomly selected sample of 5% of the reviews from every time period. However, instead
of using human coders, I applied the computerized content analysis approach described in
Section 2 to produce the valence measures. Model 3 is a simpler version of my full model
27

in which only buzz volume, but not the valence, is endogenous. Finally, Model 4 is my
proposed model in which the valence is based on a text-analysis of content and in which
both the volume and the valence of buzz are endogenous.

The estimation results allow me to draw the following conclusions. First, the inferior fit
of Models 1and 2 relative to Model 3, highlights the value of my computerized text-
analysis approach. Second, the better fit of Model 4 relative to Model 3, underscores the
importance of accounting for the simultaneity of buzz volume and revenues. Finally,
based on the AIC criterion, the proposed VAR model (Model 4), with endogenous buzz
volume, advertising and screens and that uses the valence measures from my content
analysis approach provides the best fit to the data.
TABLE 1.4: VAR Model Fit Results

Model
Specification
Log
Likelihood
AIC
1 Star Ratings and no Buzz
Valence
−5344.06 57.12
2 Subsample of Human-Coded
Buzz
−5346.18 57.13
3 Endogenous Buzz Volume &
Exogenous Buzz Valence
−5301.29 57.04
4 Endogenous Buzz Volume &
Buzz Valence
−5288.63 55.19

28

Parameter Estimates

Table 1.5 reports the parameter estimates obtained from the best-fitting Model 4. These
estimates are organized into groups of variables corresponding to the factors influencing
sales, namely Buzz Volume, Buzz Valence and movie characteristics and are discussed in
the same order.

TABLE 1.5: Word of Mouth Effects on Box Office Revenues

Revenues

Screens

Advertising
Elasticity t-stat Elasticity t-stat Elasticity t-stat
Volume Measures
Volume of user
word of mouth
(VOLUSER)
0.418 2.01** 0.306 1.99** 0.304 2.16**
# of Newspaper
articles
(VOLNEWSP)
0.153 2.01** 0.083 1.86** 0.02 0.95
# of magazine
articles
(VOLMAG)
0.105 1.92** 0.0 1.94** 0.18 2.31**
# of critic
reviews
(VOLCRIT)
0.166 1.88** 0.359 2.01** 0.03 0.99

29

TABLE 1.5, CONTINUED
# of Blog
Mentions
(VOLBLOG)
0.133 1.92** 0.16 0.75 0.19 1.01
Positive Valence Measures
Ratio of
positive words
in user reviews
(POSUSER)
0.087 1.88** 0.051 1.65** 0.04 1.21
Ratio of
positive words
in critic reviews
(POSCRIT)
0.045 1.80** 0.17 0.85 0.16 1.28
Ratio of
positive words
in newspaper
articles
(POSNEWSP)
0.031 3.18** 0.018 2.53** 0.16 1.13
Ratio of
positive words
in magazine
articles
(POSMAG)
0.049 1.53 0.05 1.67** 0.09 2.18**
Negative Valence Measures
Ratio of
negative words
in user reviews
(NEGUSER)
−0.107 −1.72** −0.0482 −1.75** −0.03 −1.19

30

TABLE 1.5, CONTINUED
Ratio of
negative words
in critic reviews
(NEGCRIT)
−0.018 −1.24 −0.16 −1.29 0.05 1.68**
Ratio of
negative words
in newspaper
articles
(NEGNEWSP)
−0.029 −2.64** −0.017 −1.24 −0.05 −0.85
Ratio of
negative words
in magazine
articles
(NEGMAG)
−0.018 −1.79** −0.10 −1.81** 0.27 2.25**

5.1 Impact of the Volume of Buzz

Consistent with the prior literature in this area (Chevalier & Mayzlin, 2006; Liu, 2006;
Godes & Mayzlin, 2004), I find the volume of buzz across sources to have a positive
impact on sales revenue reaffirming the popular adage; ‘Any publicity is good publicity’.
The short-term elasticity of the VOLUSER on revenues is positive and significant at
0.418. These results seem to support the studios’ strategy of having cast members appear
on talk shows, or the director doing radio interviews and trying to get people to talk about
the movies.

31

In addition, I also find positive and significant elasticity estimates for the volume of other
exogenous buzz measures such as the VOLNEWSP (0.153), VOLMAG (0.105),
VOLBLOG (0.130) and VOLCRIT (0.166) on movie revenues. However, the magnitude
of these effects is much smaller than the user-generated material. The website presence
dummy is not significant at the 95% level.

Turning to the results from the Screens regression, I find that VOLUSER not only
impacts box-office revenues directly, but also indirectly, by inducing exhibitors to
increase or sustain the distribution intensity of the movie. However, in contrast to the
revenues equation, VOLCRIT seems to have the single biggest impact on the screen-
allocation decision with an elasticity of 0.306. A potential explanation for this contrast is
the timing of the screen allocation decision. The contracts between exhibitors and studios
frequently call for a minimum run and a pre-committed screen allocation during that time
(DeVany and Walls, 1999). Since these contracts are signed prior to the wide release of
the movie, it is possible that the only reviews available at the time are critic reviews
making them a key driver of the screen allocation decision. The volume of buzz from
sources other than user and critic reviews is seen to have a smaller impact on the screen
allocation decision. VOLNEWSP (0.083) and VOLMAG (0.09) have a positive impact
on the number of screens allocated to the movie. However, the number of blog mentions
and the website dummy are insignificant at the 95% level.

32

5.2 Impact of the Valence of the Buzz

User reviews

The results from the Revenue model show that the affective content of the buzz
significantly impacts box-office revenues. The elasticity of POSUSER is 0.087 while the
elasticity of NEGUSER is −0.107. Thus, at least in the case of user reviews, it seems that
the negative content in user reviews has an asymmetrically bigger impact on box-office
revenues as compared to the positive content of user reviews. This is in line with research
in the consumer behavior literature which shows that negative information has greater
diagnosticity (Ito, Larsen, Smith and Cacioppo, 1998).

The impact of the valence of user reviews on the screen-allocation decision is also found
to be significant. The elasticity of POSUSER on the number of screens is 0.051 while the
elasticity of NEGUSER is −0.0482.

Newspaper and magazine articles

POSNEWSP has an elasticity of 0.031 while NEGNEWSP has an elasticity of −0.029.
POSMAG has an elasticity of 0.048 while NEGMAG is statistically insignificant.
POSCRIT has an elasticity of 0.071 while NEGCRIT has an elasticity of −0.095. Thus,
33

except for magazine articles, the negative information seems to have a bigger effect than
the positive content in the review. Overall, the valence of content other than user reviews
has a smaller impact on box-office performance than the content of user-reviews.

5.3 Interaction between Advertising & Buzz

I am interested in examining how advertising and consumer-generated buzz interact. I
find that advertising has a significant positive impact (elasticity = 0.304) on the volume
of consumer-generated buzz for a movie.

Another interesting question is whether advertising and buzz are complements or
substitutes. Complimentarity would imply that favorable buzz would induce firms to
reduce future advertising and negative buzz would spur firms to increase their advertising
expenditure. However, I find that no such effects in the data. One possible reason for this
is the difficulty associated with monitoring the consumer buzz, especially in terms of its
valence.

5.4 Critic versus User Reviews

Prior research (Holbrook, 1999) has shown that there are some fundamental differences
in the way experts and ordinary consumers evaluate motion pictures. While these two
sources of opinion are not negatively correlated, it appears that they do use different
34

criteria to form their judgments. In line with this literature, I find that user reviews
actually have a bigger impact on movie revenues as compared to critic reviews (elasticity
of number of user reviews = 0.279 as opposed to 0.418 for the volume of user reviews). I
find similar patterns for the valence measures as well. However, I find that the results are
reversed when it comes to the impact of reviews on screens. Here, the critic reviews play
a bigger role (elasticity of critic reviews = 0.359 vs. 0.306 for user reviews). Since
exhibitor contracts often guarantee a certain minimum run even before the movie is
released, these contracts are probably mostly influenced by pre-release buzz. Since most
consumers don’t have much information regarding the movie prior to its release, critic
reviews would intuitively seem to have a bigger effect on these contracts.

5.5 Long-Term Elasticity of Buzz Variables

To quantify the long-run elasticity of buzz volume (as well as buzz valence, advertising
and screens) on movie box office revenues, I calculate arc elasticities using the following
approach.

First, from the IRFs, I compute the total change in the revenues, ∆Y, in response to a one
standard deviation shock to Buzz Volume. Second, using the data, I calculate the standard
deviation for buzz volume (σ
X
) and mean values for revenues (Y ) and Buzz Volume ( X
). Finally, I use the following equation to calculate the arc elasticity,
arc
η :
35

.
arc
X
Y X
Y
η
σ
Δ
= ×

The above is a standard elasticity formula, except that
X
σ is substituted for ∆X. This
follows because
X
σ is the change in X that is used to generate the IRF. The results of the
elasticity computations are presented in Table 1.6. In the table, I present results at 1 day,
3 day, 5 day, 7 day and total long-term elasticity. Note that the 1-day elasticities here are
the short-term or immediate elasticities of the marketing variables.

TABLE 1.6: Short-Term vs. Long Term Elasticity of Revenues
1 Day 3 Days 5 Days 7 Days Long Term
Screens .106 .194 .205 .22 .278
Advertising .038 .085 0.133 .185 .217
Buzz Volume .075 .164 .241 .319 .531
Positive Buzz .052 .058 .063 .081 .104
Negative Buzz -.066 -.085 -.097 -.104 -.119

The results for the long-term elasticities indicate that both buzz volume and buzz valence
have a much slower decay rate than the effect of advertising and therefore yield higher
long-term elasticity estimates. In fact, the elasticity of buzz volume is about 12 times the
elasticity for advertising and the elasticity of positive buzz is about twice the advertising
elasticity. Also noteworthy is the fact that the elasticity of negative buzz is much higher
than the elasticity of favorable buzz.

36

As previously mentioned, the impulse response functions (IRFs) trace the incremental
effect of a one-standard deviation shock in Buzz Volume, Screens and Advertising on the
future value of box office revenues. These enable me to examine the dynamic effects of
each of these elements on box office revenues, fully accounting for the indirect effects of
these elements. Figures 1.2a, 1.2b, 1.2c, 1.2d and 1.2e plot these impulse response
functions.

Figure 1.2a: IRF: Response of Revenues to Shock in Screens

0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
1 2 3 4 5 6 7 8 9 10
Revenues
37

Figure 1.2b: IRF: Response of Revenues to Shock in Advertising
0
0.05
0.1
0.15
0.2
0.25
0.3
1 3 5 7 9 11 13 15 17
Revenues
Revenues
38

Figure 1.2c: IRF: Response of Revenues to Shock in Buzz Volume

Figure 1.2d: IRF: Response of Revenues to Shock in Positive Buzz
0
0.05
0.1
0.15
0.2
0.25
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
Revenues
Revenues
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Revenues
Revenues
39

Figure 1.2e: IRF: Response of Revenues to Shock in Negative Buzz

I look first at Figure 1.2a, the IRF for revenues with respect to a shock in Buzz Volume.
The graph shows that it takes approximately 16 days for revenues to stabilize after a one
standard deviation shock on Buzz Volume. After about 16 days, the IRF bands begin to
cross zero, indicating that further effects are not significant.

Turning to Figure 1.2b, I see that the long-term effect of an increase in advertising on
revenues lasts upto 9 days. Thus compared to advertising, Buzz Volume induces both a
larger short-term response as well as substantially larger carryover effect.

Turning to Figure 1.2c, which represents the effect of an increase in the number of
screens, I notice that while the short-term effect is significant, stabilization occurs within
just a few days.
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Revenues
Revenues
40

On the other hand, Figures 1.2d and 1.2e indicate that both positive and negative buzz
have significant long-term impact on revenues with the negative impact of bad buzz
lasting even longer than the positive impact of favorable buzz. These IRF results
highlight the need to employ models that can estimate longer-term effects for the effect of
buzz variables.

6. FORECASTING BOX-OFFICE REVENUES
One of the biggest challenges facing studios is forecasting box-office performance based
on the limited information available to them early in the life-cycle of the movie.
Consequently, we wanted to evaluate the feasibility of using pre-release buzz to predict
the opening week box-office performance of a movie. This forecast is especially
important because it is used by studio managers and exhibitors to make important
decisions regarding the number of screens to allocate to the movie, the advertising budget
for the movie and other marketing inputs (Houston, Hennig-Thurau, Spann and Skiera,
2008). Good forecasts help studios and theaters better plan screening capacity and
potentially optimize exhibition contracts. Prior research also finds that it is most difficult
to predict box-office performance in the early part of the movie’s life-cycle. I find that I
am able to significantly reduce the forecasting error by including the volume and content
of pre-release buzz as compared to prior models. Table 1.6 shows that the proposed
model provides a lower Root Absolute Error than several other benchmark models.

41

TABLE 1.7: Comparison of RMSE of Forecasting Models
Model Average RAE
Elberse & Eliashberg, 2003 59%
VAR without buzz variables 51%
VAR with buzz variables 41%

7. CONTRIBUTIONS, LIMITATIONS AND FUTURE RESEARCH

Contributions

This research is one of the first attempts to measure the valence of word of mouth using
text-analysis. I develop a computerized content-analysis approach that yields reliable
measures of the valence of buzz that help to predict box office performance more
accurately. I demonstrate how this approach overcomes the shortcomings of using the
numerical ratings (which are not always available) as well as other alternate approaches
used in prior research in this area. This technique is easily adaptable to a variety of
contexts including analyzing media stories, user reviews, and transcripts from focus
groups and so on by developing a lexicon that is suited to the context being studied.

42

I apply this approach to data from the US motion picture by investigating the impact of
the valence of buzz on movie box-office performance. I collect box-office revenues and
data on buzz for nearly 200 movies released between January and December, 2004. Buzz
sources include both user and critic reviews and also newspaper and magazine articles
from a variety of leading publications across the United States as well as blog postings
about the movie or the cast members or director. In all, the empirical analysis is based on
500,000 individual communications about the movies in the sample.

I estimate the parameters of a Vector Auto Regression (VAR) model of movie revenues
that incorporates several variables that capture the content and volume of buzz and
include other endogenous variables, such as the number of screens allotted to a movie
and advertising expenditures. Several other control variables identified in the vast
literature on movie sales forecasting are also included in the model.

Several interesting substantive findings about the impact of different aspects of buzz on
movie revenues emerge from the analysis. First, I show that not all sources are created
equal. Specifically, I find user reviews have the biggest impact, followed by critic
reviews and blog mentions, respectively. In other words, the opinion of peers or other
consumers exerts a larger influence on consumers than those of experts. It is interesting
to find, consistent with my results, an emerging trend among studios to move away from
the VIP treatment doled out to critics in the past and switching focus to consumers
instead (Forbes, 2006; Christian Science Monitor, 2006; New York Times, 2006). I also
43

conclude that consumers and exhibitors are influenced differently by different sources of
buzz. While critic reviews are relatively less effective at influencing sales directly as
compared to user reviews, they have a much bigger impact on the number of screens that
get allocated to a movie.

In contrast to most of the prior empirical research on online user reviews, I find that the
valence of the reviews is a key driver of box-office revenues. Both the positive and
negative content in the user reviews have a significant impact on revenues, not only in the
short run but also in the long run. Consistent with behavioral theories on word of mouth
and the diagnosticity of negative information, I find that the negative information in user
reviews has a disproportionately bigger effect on revenues. The negative impact of
negative information also lasts for a longer duration than the effect of positive
information in the reviews thereby highlighting the importance of not only driving
positive buzz for movies but also devising strategies to counter negative buzz.

My research also shows that the impact of both the volume and content of buzz exceed
the impact of advertising on box-office performance. Advertising is the biggest
component of the marketing budget for movies. Thus, this research highlights the
importance of word-of-mouth in case of experiential goods such as movies and books.

I also find that using the content of word of mouth for forecasting purposes improves the
accuracy of my forecasts. This is especially important in the context of movies and other
44

short life-cycle products where word of mouth provides early information on how the
product is being received by consumers. Noting that the most critical problem in
forecasting movie sales is obtaining accurate forecasts of the opening weekend revenues,
I focus my efforts on using the volume and valence of pre-release buzz to improve early
forecasts.

Limitations & Future Research

Although my research goes further than any previous research in terms of capturing the
content of word-of-mouth, I realize the need to develop richer measures of content that
go beyond volume and valence. While computerized text-analysis software is able to
measure valence with reasonable reliability, it is far more challenging to adapt these tools
to measure other dimensions of content such as source credibility and the focus of the
conversation which have been found to be important indicators of the effectiveness of
word-of-mouth communication. An important direction for future research is to develop
feasible methods to measure these other dimensions of content.

Another crucial gap in this area is to study the evolution of online word of mouth and the
drivers of online word of mouth. While this research shows that online word-of-mouth
affects box office performance, it is not quite clear how managers can influence the
volume and valence of online word-of-mouth. By gaining a better understanding of the
drivers of online word-of-mouth, managers can include word-of-mouth in their marketing
plans and develop tools to influence the same.
45

This research links buzz to box-office revenues. However, DVD sales are gradually
becoming a bigger part of overall movie revenues. An interesting area for future research
is to examine whether the buzz created during the movie’s stint at the box-office
continues to have a spillover effect on DVD sales. This may, in turn, have important
implications for studios to decide when they want to create the most buzz for a movie.
Finally, studios are looking to using the internet as a new distribution channel for movies
(Businessweek, 2007). The impact of online buzz on internet downloads of entertainment
products is another novel area for future research.

Finally, this research and the content-analysis techniques developed in this study can be
extended to study contexts outside of the motion picture industry. Websites such as
epinions.com and yelp.com offer rich data on user reviews in a variety of service
contexts. The present study focuses on sales or revenues. However, buzz is likely to
impact other key firm performance variables as well. For instance, corporate reputation
and brand equity are two key variables that are of interest to marketers that are likely to
be influenced by both the mass media and other sources of buzz. As the prevalence and
reach of buzz increases, it is also likely to impact firm financial performance as analysts
and investors use these online sources of buzz while making investment decisions.
Studying the impact of buzz and its content on these performance variables provides a
fertile area of future research.

46

CHAPTER 2: THE IMPACT OF MASS MEDIA ON CORPORATE
REPUTATION

CHAPTER 2 INTRODUCTION

Marketing communications are a critical component of a firm’s marketing mix and most
basic marketing texts list four major activities that comprise the communications mix,
namely, advertising, promotion, personal selling and public relations (Kotler &
Armstrong, 2006). Of these activities, the first three have received the majority of
attention in the academic literature.

For instance, a vast literature in marketing has examined how advertising works and how
it should be managed. This literature includes several published meta-analyses
documenting the magnitude of advertising elasticities (Assmus, Farley & Lehmann,
1984; Sethuraman & Tellis, 1991; Tellis et al, 2005) across a range of studies, that
examine advertising’s impact on not only sales but also a variety of intermediate metrics
including recall, attitude towards advertising, persuasiveness and memory (Heath &
Hyder, 2004). In the price and promotions area, Tellis (1988) and Bijmolt, Van Heerde
and Pieters (2005) meta-analyze several pricing studies and the availability of scanner
panel data has spawned a rich literature on the impact of promotions on sales (van
Heerde, Gupta, Wittink, 2003; Gupta, 1988, Pauwels, Hanssens and Siddarth 2002). The
marketing literature also features multiple studies that examine the effectiveness of the
salesforce and personal selling (Zoltners & Sinha, 1990 & 2001; Lodish, 1980).
47

In contrast, there has been very little empirical research on the impact of a firm’s public
relations efforts even though surveys (Dun & Bradstreet, 1999) indicate that firms spend
an average of about 10% of their marketing budget on public relations activities, with the
figure being as high as 25% for smaller firms. A search for the terms “public relations” in
the title or abstract of journal articles in the leading marketing journals turns up less than
10 articles over the past two decades highlighting the degree to which this element of the
communications mix has been neglected in the academic literature. Some of these studies
extend consumer behavior theories that explain advertising effectiveness to the domain of
public relations (Sheppard, Hartwick & Warshaw, 1988; Kallahan, 1999) while other
studies highlight the critical role of public relations in cause-related marketing
(Varadarajan & Menon, 1988; Ross, Patterson, Stutts, 1992). A final set of studies in the
area of integrated marketing communication (McArthur & Griffin, 1997; Gronstedt &
Thorson, 1996; Phelps & Johnson, 1996) underscore the importance of coordinating and
formulating a consistent message across the various communication tools adopted by the
firm. These studies primarily treat public relations as an extension of advertising and as a
tool to support the firm’s advertising investments. However, there has been no empirical
work to document the impact of public relations activities on sales or other metrics of
firm performance.

A review of the public relations literature suggests a number of definitions, metaphors
and approaches to the field. Edward Barnays (1955) defines public relations as “the
48

attempt, by information, persuasion, and adjustment to engineer public support for an
activity, cause, movement or institution”. In the seventies, Harlow (1975) constructed a
working definition of public relations condensed from 472 different definitions and with
input from 65 practitioners. According to his definition, “Public Relations is a distinct
management function which helps establish and maintain mutual lines of communication,
understanding, acceptance and co-operation between an organization and its publics;
involves the management of problems and issues; helps management to keep informed of
and responsive to public opinion and defines and emphasizes the responsibility of
management to serve the public interest”. A more practitioner oriented definition by
Denny Griswold, which first appeared in Public Relations News that is widely quoted
states, ‘Public Relations is the management function which evaluates public attitudes,
identifies the policies and procedures of an organization with the public interest, and
executes a program of action to earn public understanding and acceptance.’ A 2004
study by Impulse Research on Public Relations spending in the United States found that
Corporate Media Relations and Product Media Relations together made up over 55% of
total PR spending. This is widely considered to be the biggest PR survey in the country
and includes responses from 3500 PR firms across the United States.

Thus, the primary purpose of public relations, like all of the other communications
activities of the firm, is to influence and shape public perceptions about the firm. In this
sense, public relations can be compared to advertising, a much more widely-examined
component of the communications mix in terms of its primary goal being to shape
49

perceptions about the firm. A key distinction between advertising and public relations
however lies in the actual tools or tactics employed by public relations to achieve these
objectives. For example, some of the tools employed by PR professionals include
publicity, product placements, news releases, press conferences, speeches, websites,
publications and trade shows. Another key distinguishing factor between advertising and
PR is that while advertising is targeted directly at the end consumer, most PR
communication attempts to shape people’s perceptions indirectly through an intermediary
which is frequently the mass media. Thus, studying the impact of public relations
activities on firm performance also requires one to answer the more general question of
how the mass media affects firm performance.

A rich literature in the area of mass communications chronicles the role of the mass
media in shaping people’s perceptions about objects in the media. The prominence of
media attention to a particular topic and the ensuing salience within the minds of the
public has come to be known as the mass media’s agenda-setting function (McCombs &
Shaw, 1972; Benton & Frazier, 1976). The information reported in the media comes from
a variety of sources. Individuals write opinion pieces and letters to the editor. The
government and specialized rating agencies such as Moody’s evaluate firms and issue
their evaluations in press releases (Fombrun, 1996). A third source is media workers
(Shoemaker & Reese, 1991) such as reporters who write news and feature stories as well
as editors and columnists who write about firms. The specific stories that appear are
primarily based on the media agency’s judgments of importance and deviance from the
50

norm in both negative and positive directions (Shoemaker, 1996; Shoemaker, Danielian
& Brendlinger, 1992). From a public relations standpoint, company press releases are one
key source of information about the firm that provides a steady stream of information to
the media (Shoemaker & Reese, 1991).

A number of front-page corporate scandals that erupted in the U.S. economy beginning in
2001 – Enron, WorldCom, Tyco, Adelphia and others have brought the issue of corporate
reputation to the forefront both in the mass media as well as in the minds of executives.
In a study released at the 2004 World Economic Forum in Davos, Switzerland, 59% of
the CEOs responding and attending said that corporate reputation was a more important
measure of success than stock market performance, profitability and return on investment
(World Economic Forum, 1/22/04, press release). In addition, the business and popular
press such as Fortune, The Financial Times, and US News and World Report have fueled
this interest by publishing reputational rankings of businesses. Past research has also
found that the ratings obtained from Fortune magazine’s survey of America’s Most
Admired Corporations to measure reputation, had a positive effect on stock market and
accounting performance (McMillan & Joshi, 1997; Roberts & Dowling, 1997; Rupp &
Hamilton, 1996; Srivastava, McInish, Wood & Capraro, 1997; Vergin & Qoronfleh,
1998). Research also shows that corporate reputations have strategic value for firms
(Dierickx and Cool, 1989; Rumelt, 1987; Weigelt and Camerer, 1988). Organizational
scholars also recognize that reputation is valuable because it reduces the uncertainty
stakeholders face in evaluating firms as potential suppliers of products and services
51

(Weigelt and Camerer, 1988; Benjamin and Podolny, 1999). Due to all these reasons,
firms are increasingly starting to take reputation management more seriously and giving
it the attention it deserves. A recent article in Businessweek magazine (“What Price
Reputation”, Businessweek, July 9
th
2007) discusses the impact of public perceptions
regarding corporations on stock market performance and some of the measures taken by
firms to manage their reputation more proactively.

The concept of corporate reputation is also closely related to idea of customer-based
brand equity which has received a lot of attention in the marketing literature. This
research has identified several benefits to brand equity including higher sales, ability to
charge higher prices and increased cash flows (Keller, 1993; Erdem, Swait and Louviere,
2002; Ailawadi, Lehman and Neslin, 2003). A rich stream of work has also examined the
drivers of brand equity (Aaker, 1991; Keller, 1993) such as brand awareness and
associations. Firms follow several different branding strategies to manage the assortment
of brands they have in their portfolio. In case of single-product firms, they can choose to
have a single name for the product as well as the firm, a strategy that marketers refer to as
corporate branding (e.g. Disney). At the other end of the spectrum, they can choose to
have separate brand names for each product and have a totally different name for the firm
itself. In this way, we can see that the concept of corporate reputation or the awareness
and associations that consumers have towards the corporate brand are closely related to
the concept of customer-based brand equity. While there has been extensive research on
the drivers of brand equity and the role that advertising plays in the creation and
52

maintenance of brand equity (Aaker and Biel, 1993; Yoo, Donthu and Lee, 2000), there
has been relatively little research on what drives corporate reputation.

In the current research, I use the Corporate Esteem metric, which forms one of the five
pillars of the Young & Rubicam Brand Asset Valuator Model, to measure Corporate
Reputation. The Brand Asset Valuator initiative undertakes large scale surveys of
consumers regarding perceptions of brands on a host of different brand metrics. The
annual esteem values for a panel of 175 firms between 2001 and 2005 forms the
dependent variable in my model. The print media coverage data comes from
www.nexis.com and www.factiva.com and combines over 1,000 news sources in the
United States comprising of both newspapers and magazines. I distinguish between
articles that cite newswires and press releases as their source and use these to measure the
firm’s public relations efforts. The data on advertising spending comes from the TNS
Media AdSpender database. Using a combination of human raters as well as a
computerized text-analysis methodology, I develop measures of the volume and content
of over 200,000 news stories for the empirical analysis.

I use a Structural Vector Autoregression Model to examine the relative influence of mass
media and public relations on Corporate Reputation over time. I also study the impact of
public relations on the amount and type of mass media coverage that the firm enjoys over
time. Finally, I compare the use of advertising and public relations as alternate marketing
mix variables used to enhance corporate reputation.
53

I find that public-relations activities have the biggest impact on corporate reputation. The
elasticity of public-relations activities is bigger than the elasticity of advertising as well
as independent media-reports not originating out of PR activities engaged in by the firm
itself. Additionally, I also find significant impact of the content of mass media coverage.
Both the valence and the prominence are important determinants of reputation and
negative coverage has a longer-lasting impact than positive media coverage. Finally, I
find that advertising and public relations seem to be used as complementary
communication media. Advertising drives future mass media coverage and media
coverage also drives future advertising. This is in line with prior literature which suggests
that public relations and word-of-mouth have a bigger impact when undertaken in
conjunction with advertising.

In the next section I review the literature from various disciplines including mass
communications, sociology and economics on the impact of mass media on people’s
perceptions. Given that our focus is on mass media coverage of firms and the impact on
corporate reputation, I also review the literature on corporate reputation especially in the
marketing literature. I then go on to describe the data used in the empirical analysis and
the computerized text-analysis methodology used to analyze the valence of the news
stories. The subsequent section describes the model and the estimation procedure. I
conclude by discussing the results and the implications for theory and practice and
directions for future research.
54

LITERATURE REVIEW

Effect of Mass Media

The earliest conceptualization of mass media’s effects on society believed that the media
have direct and powerful effects. Labels such as the ‘bullet theory’ (Schramm, 1954) and
the ‘hypodermic-needle theory’ (Berlo, 1960) or the ‘stimulus-response theory’ (DeFleur
and Ball-Rokeach, 1982) have all been used to describe this point of view. As a
consequence of these theories, the prevailing notion at the time was that “mass
communications inject ideas, attitudes, and dispositions toward behavior into passive,
atomized, extremely vulnerable individuals” (DeFleur & Ball-Rokeach, 1982) or that
audience members are very vulnerable to messages from the media that are specifically
crafted to strike a particular ‘target’, or to produce some specific, intended result.

In contrast, Paul Lazarsfeld’s work attempted to displace this guiding notion and search
for specific, measurable, short-term, individual attitudinal and behavioral effects of media
content. Lazarsfeld and his team at the Bureau of Applied Social Research at Columbia
University studied voters during the 1940 and 1948 elections. Both studies were sample
surveys involving the same set of respondents interviewed several times and concluded
that the media played a weak role in election decisions, as the respondents voted the same
way as members of their primary political affiliations. Lazarsfeld introduced several
55

intervening variables that exist between the media content and the audiences that receive
it. These include standard demographic variables such as age, education and income and
also interpersonal variables such as ‘gregariousness’, the ‘degree and extent of contact
the respondent has with significant other’, and even the degree of ‘reputation’ or ‘status’
of the media. Out of this work came the concept of ‘two-step flow of communication’
(Katz & Lazarsfeld, 1955), whose central tenet was that ‘opinion leaders’ actually served
as a ‘mediating variable’ in shaping what the masses believed, rather than the mass media
themselves. Lazarsfeld’s conclusion was that the media were “not very important in the
formation of public opinion” (Gitlin, 1981).

Another paradigm holds that the media’s power to influence is not limited or direct, but
indirect instead. With the growth in popularity of cognitive psychology as a rival
perspective to behavioral psychology, Severin and Tankard (2001) noted that researchers
had begun to re-evaluate their understanding of media effects on audiences, and thus
begun to shift their focus from changing attitudes to changing perceptions. This view of
the world grounded in cognitive psychology held that people were active decision-makers
and information processors, rather than objects being acted upon, conditioned and
manipulated. This viewpoint gradually led to the emergence of Agenda-Setting Theory
(McCombs & Shaw, 1972) which has been the dominant view on media effects over the
past few decades.

56

Agenda-Setting is the process by which the news media create public awareness and
concern for certain issues (McCombs & Shaw, 1972; McCombs & Shaw, 1993). In this
process, a wide range of news issues are filtered and reduced to a few that can be
presented to the public, leading to a concentration upon certain issues or subjects under
discussion with the result that the public perceives those to be more important than
others.

The agenda of a news organization is to be found in its pattern of coverage on public
issues over a period of time. Over this period, certain issues are emphasized, some
receive light coverage and some are seldom or never mentioned. The media ‘agenda’ is
simply the totality of news discussion emerging from decisions made by journalists and
editors regarding the news at any given point in time. The public agenda, i.e. the focus of
public attention is commonly assessed by public opinion polls that question people about
issues in the news that they consider important (McCombs & Ghanem, 2001).

While there has been a lot of research on the agenda-setting effects of mass media, the
results are mixed depending on the issues studied and the research design employed. The
work of McCombs & Shaw provided the basic empirical foundation for this theory
(Lowery & DeFleur, 1987). McCombs & Shaw (1972) postulated that the media sets the
agenda for each political campaign, influencing the salience of attitudes toward political
issues. This landmark study examined agenda-setting effects during the 1968 presidential
election by comparing issues voters felt to be important with the media messages to
57

which these voters were exposed. A random sample was used to select 100 registered
voters in Chapel Hill, North Carolina from five demographically representative precincts.
Among the undecided voters, high correlations were found between what voters cited as
major issues and the attention of the mass media to those issues. The respondents’
choices of important election issues mirrored issues covered in the media. In addition,
while the three different presidential candidates placed widely divergent emphases upon
these different issues, the judgments of those voters seemed to reflect the composite of
the mass media coverage.

Agenda setting theory provides a framework for investigating the comprehensive
communication process through which media professionals, the public and media objects
compete for the limited resources of media publicity and favorability. Media salience is
the primary variable in agenda-setting theory, and communication scholars have
recognized that it is a multi-dimensional construct. Three distinct dimensions of media
salience have emerged in the literature, namely, attention, prominence and valence
(Kiousis, 2004). ‘Attention’ is the amount of media awareness that an object receives,
usually indicated by the sheer volume of stories or space devoted to a particular media
object (Benton & Frazier, 1976; Dearing & Rogers, 1996; Golan & Wanta, 2001).
Another dimension of media salience is prominence, which has been characterized in a
variety of ways. Stories in the media indicate their importance to the readers by their
placement, length and treatment. Second, stories within elite newspapers such as The
New York Times could be characterized as prominence (Reese & Danielian, 1989). The
58

third dimension of media salience is valence (Kiousis, 2004; Manheim, 1986). Drawing
on the theories described above, I develop measures of media salience to empirically test
agenda-setting theory in a business context.

Agenda-setting theory concerns media objects of news topics that receive coverage in
media outlets. Media objects may be any aspects of central importance discussed in a
particular news story. Issues such as healthcare, corporate governance or education are all
common media objects. In fact, agenda-setting has mostly been concerned with the social
and political realm (Funkhouser, 1973; Palmgreen & Clarke, 1977; Smith, 1987; Winter
& Eyal, 1981) and have aimed to establish links between media coverage and public
attention on such issues. Prior research has demonstrated empirically that such links do
exist. For example, examinations of civil rights issues from the 1950s to the 1970s by
Winter and Eyal (1981) found a correlation of 0.71 between the level of public concern
(as measured by the Gallup Poll for public concern for civil rights issues) and the
prominence of news coverage in the weeks immediately preceding. Eaton (1989) found
the same pattern over a 41-month period while investigating eleven different issues. In a
meta-analysis of 90 empirical studies reporting correlations coefficients for the agenda-
setting effect, Wanta & Ghanem (2000) found a mean of 0.53. This suggests that news
sources strongly influence public opinion. Additionally, cross-lagged correlations show
stronger influence between public opinion polls and prior media coverage, rather than
later news coverage suggesting that it is indeed the media that influences public
perception rather than the other way around.
59

Assessments of the affective dimension of media coverage recognize that news coverage
public survey responses carry not only descriptions of objects but also convey feeling and
tone about the object described. The feeling, or tone, expressed for a given object is
typically a measurement of how favorably that object is portrayed, and may be articulated
as positive, negative, or neutral (Lopez-Escobar, Llamas, McCombs & Lennon, 1998;
McCombs, Llamas et al., 1997; McCombs, Lopez-Escobar & Llamas, 2000). Researchers
using the agenda-setting framework also determine to what degree the tone of media
coverage for certain objects affects the sentiments expressed in public perceptions
regarding these objects.

A generally accepted consequence of media effects within the agenda-setting framework
is ‘priming’ (Iyengar & Kinder, 1987; McCombs, 1993; McCombs & Reynolds, 2002).
Certain media objects begin to occupy a place within the public’s frame of concern, and
become part of the public’s ‘top-of-mind’ awareness related to decision making. When
faced with multiple issues and events that are complex in nature, people do not
thoroughly review their memory entirely in order to respond, but instead, rely upon
information that is currently accessible, or most easily retrieved (Iyengar & Kinder,
1987). Greater degrees of emphases on specific issues in newspapers, television or other
media enhance the issues’ accessibility to readers.

60

The agenda-setting framework has received substantial empirical support over the past
three decades for the media’s effects on public perceptions. Dearing & Rogers (1996)
note that since the McCombs & Shaw (1976) initial study defining agenda-setting as a
paradigm for conceptualizing media effects, there have been over 350 studies between
1972 and 1994. Over this time period, empirical support for agenda setting has been
found in relation to the media’s portrayals of public issues and political candidates in
numerous ways across geographical regions (Weaver, Graber, McCombs & Eyal, 1981),
time (Winter & Eyal, 1981), time and space (Brosius & Kepplinger, 1990; Takeshita &
Mikami, 1995; Weaver, 1996), and through various forms of media, including
newspapers, TV, radio, and the internet. However, this theory hasn’t yet been tested in
the business domain. In fact, a review of research in journalism and mass communication
shows a paucity of research on the business press (Carroll, 2002). Fürsich (2002), in a
review of over 25 years of journalism research, calls it one of the areas least investigated
by communication scholars. Thus, the present research extends prior empirical research
on agenda setting as well as research on the role of the business press in shaping public
perception about corporations.

Corporate Reputation

The growing interest and attention to corporate reputation in the popular press can be
traced to Fortune magazine’s “Most Admired Corporations” special issue 20 years ago.
Since then, a number of media publications have developed their own rankings including
61

the Wall Street Journal and the Financial Times. In a study of world CEOs, the Financial
Times (2004) reported that 60% of their CEO respondents said that corporate reputation
was more important now than it was five years ago. In 2003, Edelman (2003) interviewed
850 opinion leaders (400 in the United States and 450 in Europe) on behalf of the World
Economic Forum. Nine out of ten U.S. respondents said that a corporation’s reputation
played a large role in forming opinions of products and services; 80% said they were
more willing to pay for goods and services from a company with a well-regarded labor
and environmental record.

Literature in the areas of marketing, mass communications and strategic management is
replete with definitions of ‘corporate reputation’. Reputation is said to be synonymous
with esteem and renown and is said to refer to “….what is generally said or believed
about the abilities, qualities etc. of somebody or something” (Bromley, 1993). A more
firmly established definition of corporate reputation is “a perceptual representation of a
company’s past actions and future prospects that describes the firm’s overall appeal to all
of its key constituents when compared with other leading rivals” (Fombrun, 1996).
Wartick (1992) calls corporate reputation “the aggregation of a single stakeholder’s
perceptions of how well organizational stakeholders”. Deephouse (2000) adds that
reputation is the evaluation of a firm by its stakeholders in terms of their affect, esteem
and knowledge. Waddock (2000) defines reputation as “essentially the external
assessment of a company or any other organization held by external stakeholders”.
Reputation includes several dimensions, including the firm’s perceived capacity to meet
62

stakeholder’s expectations, the attachments that a stakeholder forms with an organization
and the overall ‘net image’ that the stakeholders have of the organization. “Corporate
Reputation is the consensus of perceptions about how well a firm will behave in a given
situation, based on what people know about it… But corporate reputation is not about
likeability; it is about the predictability of behavior and the likelihood that a company
will meet expectations” (Whetten & Mackey, 2002). While reputation is generally
described in relation to a particular type of stakeholder (e.g. customers, employees,
community, investors) with regards to a reputation for something (e.g. quality of products
and services, workplace environment, social responsibility or financial performance),
most stakeholder groups reference positions taken by stakeholders outside of their own
focus and thus there seems to exists a global perception held by stakeholders in general.

A number of academic disciplines including sociology, psychology, marketing,
organizational theory, strategic management, accounting and economics have all studied
‘corporate reputation’. Several theories from economics offer insight into the construct of
corporate reputation including ‘agency theory’, and ‘signaling theory’. Agency Theory
suggests that relationships are a series of implicit and explicit contracts that bring with
them certain associated rights (Jensen & Meckling, 1976). The theory assumes that
corporate managers need to be watched because they are more likely to act out of self-
interest rather than in the interests of the firms. Investors and regulatory agencies who are
cautious about their interests being held secondary may rely on the executive’s reputation
63

which can serve as a substitute for a more binding contract, as a method of ensuring trust
or control that their interests are being protected.

Signaling theory suggests that observers of a company are faced with incomplete
information regarding its actions and must therefore rely on interpretations from signals
that firms send out. Signals are “alterable observable attributes” (Spence, 1974). They
must also reply on the evaluative signals send by key intermediaries such as market
analysts, professional investors, rating agencies and reporters. According to signaling
theory, the presence of information asymmetry forces stakeholders to rely on proxies in
making assessments about the firm. Consumers rely on firms’ reputations because they
have less information than managers do about firms’ commitments to delivering quality
and reliability. When the quality of a company’s products and services is not directly
observable, high-quality producers are expected to invest in reputation building in order
to signal their quality.

In the strategic management literature, corporate reputation has been viewed as both an
asset and a mobility barrier. Established corporate reputations have the capacity of
blocking the entry of new firms because of the high costs and the long duration of time
involved in building reputation. From this viewpoint, incumbent firms establish
reputations that are difficult for newcomers to imitate or duplicate. This viewpoint also
suggests that firms must allocate resources over time to build reputational barriers to
prevent rivals from entering the field (Barney, 1991). Conversely, reputation may impede
64

market acceptance of new products and services perceived by consumers as having a
“low fit” with the firm’s traditional area as subjectively inferred from their reputation.

Another theory from the sociology/organization theory area that offers insights into
corporate reputation is institutional theory. This theory suggests that firms are less
rational or efficient than they appear. From this perspective, firms turn away from more
essential (e.g. economic or efficient) activities in order to pursue those that may not be as
economic, rational or efficient (Zajac & Westphal, 1995), but may increase their
reputational standing (Staw and Epstein, 2000). In this literature, reputation is viewed as
an indication of legitimacy. For example, firms may adopt practices or project the image
of adopting practices of other firms with stronger reputational signals. This process of
adopting appearances can overcome the negative economic consequences through the
enhancement of legitimacy which leads to other benefits for the firm.

The accounting literature is concerned with the insufficiencies of financial reporting
standards in documenting the value of intangible assets held by the firm. These
researchers seek to develop better measures for investments in branding, training and
research and put a financial value to reputation as an asset.

In the marketing literature, reputation is often used interchangeably with ‘brand image’
and a lot of the marketing literature on reputation is focused on the processing of
information to produce these images of external objects (Lippman, 1922). The foundation
65

of this literature lies in the ‘elaboration likelihood model’ of Petty and Cacioppo (1986)
designating three layers of elaboration: high, medium and low. A high degree of
elaboration of information about an object results in a complex network of meanings. A
low degree of elaboration results in simple descriptions such as ‘good’ or ‘bad’,
‘attractive’ or ‘unattractive’. The degree of elaboration is found to depend on various
factors including the degree of familiarity between the individual and the object, the level
of involvement of the subject with the object and the intensity and integrated nature of
the marketing communications.

Another theory from the consumer behavior literature that is relevant to the concept of
corporate reputation is Attribution Theory. Attribution Theory (Kelley, 1971) suggests
that people seek to understand what causes events that they experience in their life,
motivated by the need to establish a sense of control over their environment. For instance,
consumers are known to make attributions about why a particular product failed, why
they switched brands or why a celebrity endorses a certain product (Folkes, 1988).
Accordingly, they form a relationship between attributions and subsequent attitudes and
behaviors (Kelly and Michela, 1980). Thus, if consumers form attributions about
corporate motives for the activities the firm undertakes (as reported in the mass media for
instance), these attributions should exert some influence on subsequent attitudes and
behavior.

66

A major gap in the literature on corporate reputation seems to be an empirical
examination of the impact of the firm’s own communication mix on corporate reputation.
I seek to fill this gap by empirically investigating the impact of public relations vs.
independent media coverage on corporate reputation. Additionally, I also examine the
relative impact of various types of media stories based on the topic and the valence of the
story. This offers important insights to managers on where to focus their public relations
efforts. Finally, my framework enables me to examine these effects over time thereby
allowing me to shed some light on the process through which firms build as well as
maintain their reputations.

Data & Content Analysis Methodology

This study considers the impact on corporate reputation of the persistent appearance of
new stories related to specific firms in the newspaper media. The importance of printed
media over radio and television has been demonstrated by several studies that indicate the
public’s collective preference for newspapers in getting business news, as well as the
higher rate of recall shared by newspaper readers. For instance, concerning the former,
Epstein (1978) found the newspaper to be a predominant choice for the public,
outweighing television and radio. More recently, in a similar study, Stempel (1991) found
that roughly two-thirds of those polled claim to get their business news from newspaper.
Furthermore, those reading newspapers have been shown to possess a greater capacity for
67

recall than those listening to radio or watching television (DeFleur, Favenport, Cronin &
DeFleur, 1992).

For my corporate reputation measure, I use the Corporate Esteem metric that forms one
of the five pillars of the Young & Rubicam Brand Asset Valuator Model. The Brand
Asset Valuator initiative undertakes large scale surveys of consumers regarding
perceptions of brands on a host of different brand metrics. The survey (along with a
monetary incentive) is mailed to members of a large rotating consumer panel, which is
balanced based on age, gender and religion. On average about ten thousand surveys are
sent in each US data collection wave and around 66% are completed and returned.
Approximately 2,400 unique brands are included in each data collection wave and every
respondent evaluates a subset of around 120 brands. Esteem reflects the level of
deference, respect, and regard a consumer holds for a particular brand/firm. The most
current Y&R measure for Corporate Esteem comprises of four components, namely, i.)
the percentage of respondents that indicated they felt the brand was of high quality, ii.)
the proportion of respondents that indicated they felt the brand was a leader, iii.) the
proportion of respondents that indicated they felt the brand was reliable, iv.) a 7-point
scale indicated the person regard the respondent had for the brand. A composite Esteem
measure is calculated by first computing z-scores for these items across all brands and
then averaging the four z-standardized measures. I restricted my sample to mono-brand
publicly-traded firms, i.e. firms where a single brand represents the bulk of the firm’s
68

business. The annual esteem values for a panel of 175 firms between 2001 and 2005
forms the dependent variable in our model.

The print media coverage data comes from www.nexis.com and www.factiva.com and
combines over 1,000 news sources in the United States comprising of both newspapers
and magazines. I distinguish between articles that cite newswires and press releases as
their source and use these to measure the firm’s public relations efforts. The data on
advertising spending comes from the TNS Media AdSpender database.

My objective is to link the volume and content of the mass media coverage about a firm
and the public relations efforts that the firm undertakes to its reputation, as measured by
the Y&R Corporate Esteem measure. I measure the volume of mass media coverage
directly by counting the number of media stories that appear about a firm at any given
point in time. Based on the prior empirical literature on agenda-setting theory, the content
of mass media coverage is assessed using three different measures, namely, the
prominence, topicality and valence of the media stories.

Specifically, lexis-nexis and factiva specify whether the articles appear on the front page,
the front page of the business section, the ‘Company News’ section, letters to the editor,
editorials or the op-ed pages. Thus, I have dummy variables for 5 of the 6 possible
locations within the newspaper where the article can possibly appear and these variables
capture the prominence of the story within the media publication.
69

In order to identify stories prominently featuring a particular discussion topic,
bibliographers from Lexis-Nexis generate and catalogue topical index terms for subjects,
public persons, publicly-traded companies, and organizations. These terms are attached to
news stories on Lexis-Nexis. A complete list of the topical index terms is available online
at http://www.lexis-nexis.com/infopro/products/index. These terms provide a more
sophisticated way of searching for relevant content by specifying only the related key
words. The terms also allow end users to see other keywords used to identify the content.
These topical index terms were then classified by independent human coders into five
attribute dimensions, namely, (i) Products & Services, (ii) Executive Performance, (iii)
Workplace & Employees, (iv) Corporate Social Responsibility and (v) Financial
Performance.

In order to test the relative diagnosticity of these two different aspects of valence as well
as to provide a more nuanced and rich measure of the content of the reviews, we develop
a content-analysis tool that quantifies the favorability as well as the unfavorability of the
review. Our approach builds that on developments in the area of sentiment mining in a
variety of other disciplines and uses statistical and natural language processing
techniques to elicit emotive sentiment from a posted message. The steps involved in this
process are outlined below.

70

Finally, the media data is then fed into a computer program that determines the valence or
the level of positive and negative content in each body of text. A database supports the
valence-classification algorithm:

An electronic English “dictionary”, provides base language data. This comes in handy
when determining the nature of a word, i.e. noun, adjective, adverbs etc. In addition, to
exploit parts-of-speech usage in buzz messages, a dictionary was used to detect adjectives
and adverbs. This dictionary is called CUVOALD (Computer Usable Version of the
Oxford Advanced Learner’s Dictionary). It contains parts-of-speech tagging information
and the computer program uses this dictionary to analyze messages for grammatical
information.

Secondly, each word in the text is ‘stemmed’, i.e. every word is mapped to its root. For
example, the root of ‘fascinating’ is fascinate and so on. We use an algorithm that counts
the number of positive and negative words within the text. Our goal is to use the
computer program to measure the degree of positively and negatively valenced
information in a particular media story. The program returns the count of positive and
negative phrases found in the body of each news story.

Model
In order to identify the impact of the volume and the content of media coverage that
stems from a firm’s own public relations efforts separately from the impact of
71

independent media coverage, I turn to a time series regression approach. My key
objective is to capture the feedback effects from corporate reputation to the volume and
content of media stories and public relations efforts. For instance, public relations may
serve to directly enhance corporate reputation and also simultaneously engender more
independent media coverage, thereby indirectly improving corporate reputation. These
types of interactions may either occur immediately or play out dynamically over time,
which can be well represented by a VAR model which enables me to allow for co-
dependence between independent media coverage, public relations and corporate
reputation over time.

VAR models are commonly used to quantify short- and long-run market response
(Dekimpe & Hanssens, 1999). I note two features of this approach. First, the endogenous
treatment of mass media coverage implies that it is explained both by its own past history
and that of corporate reputation. In other words, this dynamic model estimates the
baseline of each endogenous variable and forecasts its future values based on the
dynamic interactions of all jointly endogenous variables. Secondly, dynamic effects are
not a priori restricted in terms of time, sign or magnitude. The sign and magnitude of any
dynamic effect need not follow any particular pattern – such as the imposed exponential
decay pattern from Koyck-type models.

Compared to alternate specifications, VAR models are especially well-suited to measure
dynamic interactions among performance and marketing mix variables. Recently, VAR
72

models have been used to analyze a wide variety of long-term marketing effects
including advertising, price promotions and new product introductions (e.g. Dekimpe &
Hanssens, 1999; Pauwels et al. 2002 & 2004).

VAR Model Specification

Recognizing that there exists an endogenous relationship between corporate reputation
and the media coverage received by the firm, I incorporate the drivers of news media
coverage into the model (Whitney & Becker, 1982). Similarly, I also account for the fact
that not only does news media coverage affect box-office corporate reputation but the
firm’s historical reputation also affects the volume of and content of news media
coverage in the future. Consequently, I also include the drivers of news media coverage
in the model.

I construct a VAR model where the volume of news media coverage, the volume of
public relations activities and the level of advertising are all endogenous to the model.
The order of the VAR model is based on the Schwarz’s Akaike Information Criterion.
The model can be written out as follows:

73

Q

,

,

,
RQ

,

S
T
U"
T
."
T
12$0
T
MU
V 3 S
T
U"
T
."
T
12$0
T
MU
V 3

4
U"
4
."
4
12$0
4
MU

F 6
3 7 S
W
99
W
9:
W
9;
W
9<
W
:9
W
::
W
:;
W
:<
W
;9
W
;:
W
;;
W
;<
W
<9
W
<:
W
<;
W
<<
V F

Q

,B@

,B@

,B@
RQ

,B@

3
?
@A9

S
D
99
0 0 0
0 D
::
0 0
0 0 D
;;
0
0 0 0 D
<<
V F

F

,U"
F

,."
F

,12$0
F

,MU

3 S
G
U",
,
G
.",
,
G
12$0,
,
G
MU,
,
V
where
CR = Corporate Esteem Measure,
PR = Volume of public relations effort (i.e. Number of Press Releases),
Advg = Advertising Expenditure,
MC = Volume of independent media coverage (i.e. Number of independent media
stories),
i = firm subscript,
t = day subscript,
α = constant terms,
T = Time Trend Variable
J = number of lags of the dependent variable needed to ensure that the residuals are
white-noise errors (i.e. without any residual autocorrelation).

74

In the above model, X
CR
, X
PR
, X
Advg
and X
MC
are the exogenous variables associated with
each of the endogenous variables.

Estimation

The empirical analysis proceeds as follows. I start by investigating which variables
Granger cause other variables (Granger, 1969, Hanssens et al. 2001). Granger causality
implies that knowing the history of a variable X helps explain a variable Y, over and
above Y’s own history. This type of ‘temporal causality’ is the closest proxy for causality
that can be gained from studying the variables’ time series. The only other way to
examine causality would be through controlled laboratory experiments.

I perform a series of Granger causality tests on each pair of key variables. I note that a
wrong choice for the number of lags in the test may erroneously conclude the absence of
Granger causality (e.g. Hanssens, 1980). Therefore, I run the causality tests for lags up to
15 time periods and report the results for the lag that has the highest significance for
Granger causality. If corporate reputation does Granger-cause some of the other key
variables such as public relations, independent mass media coverage and advertising,
then I need to capture these interactions in a full dynamic system. Then, I test for
evolution of all the variables in our study. If I find evidence of co-evolution then I need to
choose a Vector Error Correction (VEC) model instead of a VAR.

75

Since I don’t find any evidence of co-evolution in our data, I specify a VAR model. I take
logarithms on both sides while estimating the VAR model. This allows me to directly
interpret the short- and long-term performance impact estimates as elasticities (Nijs et al.
2001). Next, impulse response functions are derived from the VAR model. The impulse
response functions represent the over-time impact of a unit shock to any endogenous
variable on the other endogenous variables. Following Dekimpe & Hanssens (1999), I
use generalized impulse response functions (or simultaneous shocking) to ensure that the
ordering of the variables in the system does not affect the results. Following Nijs et al.
(2001), I determine the duration of the shock (maximum lag k) as the last period in which
the impulse response function has a t-statistic with an absolute value that is greater than
1. In the context of our research questions, I use impulse response functions to
disentangle the short and long-run effects of public relations and independent mass media
coverage on corporate reputation.

As described earlier, I start by testing for the presence of unit roots in the data in order to
select between a Vector Autoregression Model and a Vector Error Correction Model for
the empirical analysis.

Dickey-Fuller Test Results

Augmented Dickey-Fuller tests recommended by Enders (1995) were used to test for the
presence of unit roots in the data. Results of both tests confirmed trend stationarity in all
76

series (i.e. all series appeared stationary after controlling for deterministic trend). Table
2.1 provides the details of the unit-root test results. I can therefore proceed with a VAR
(p) model for the rest of the analysis. The order of the VAR model is based on statistical
criteria such as the Akaike Information Criterion (AIC) (Lutkepohl, 1993). Stationary
variables are included in levels while difference-stationary variables are included in
differences.

TABLE 2.1: Results of Unit Root Tests
ADF Stat 5% Critical Value Unit Root?
CR −4.18 −3.44 No
PR −5.17 −3.44 No
Advg −3.99 −3.44 No
MC −5.62 −3.44 No

Having determined trend stationarity in the data, I proceed to examine possible
interdependence between reputation, advertising, public relations and other mass media
coverage using Granger Causality Tests (Granger, 1969). This interdependence may
occur immediately but play out dynamically over a period of time.

77

Endogeneity (Granger Causality Test Results)

I summarize the results from the Granger Causality tests in Table 2. Each cell contains
the minimum p-value obtained from the causality tests conducted from one lag to 15 lags
with the numbers in bold-face indicating that the variable in the row Granger causes the
variable in the column. Thus, advertising, PR & MC all Granger cause CR, while PR is
only impacted by Reputation. The results from the Granger Causality tests in Table 2.2
clearly indicate that these variables simultaneously drive each other over time and the
need to employ a full dynamic system, as in a VAR-model, to account for this type of
long-term feedback.

TABLE 2.2: Granger Causality Test Results (Minimum p-values across 10 lags)

DV Is Granger
Caused by:
CR Advg PR MC
CR .01 .00 .00
Adv .00 .03 .13
PR .00 .04 .09
MC .03 .31 .02
The bold faced entries denote pairs of variables where the
variable in the column variable granger causes the row variable

78

VAR MODEL Estimation and Fit

Fit results for the estimated VAR model, with the appropriate lags determined by the
Schwarz Bayesian Information Criterion, are presented in Table 2.3 along with the
parameter estimates. I also note that the AIC criterion selects 2 months as the optimal lag
length.

TABLE 2.3: Effect of Media Coverage on Corporate Reputation

Corporate Reputation
Elasticity t-stat
Volume Measures
PR 0.729 2.19**
MC 0.614 2.05**

Prominence Measures
PR_Frontpage 0.192 1.99**
PR_FrontpageBus 0.117 2.03**
PR_CompanyNews 0.083 1.97**
MC_Frontpage 0.185 2.18**
MC_FrontpageBus 0.064 2.93**
MC_CompanyNews 0.017 3.05**
79

TABLE 2.3, CONTINUED
MC_LetEd 0.016 1.77
MC_Editorial 0.009 1.92*
Subject Matter / Topic
PR_Products&Services 0.299 2.17**
PR_Vision&Leadership 0.176 3.08**
PR_FinPerf 0.258 3.13**
PR_CSR 0.042 1.99**
MC_Products&Services 0.372 2.53**
MC_Vision&Leadership 0.208 3.16**
MC_FinPerf 0.416 2.95**
MC_CSR 0.341 2.07**

Valence
PR_Pos 0.285 3.22**
PR_Neg -0.492 -2.63**
MC_Pos 0.187 3.06**
MC_Neg -0.419 2.98**

Advertising 0.083 2.07**

80

Impact of Public Relations vs. Independent Media Coverage

I find that the volume of public relations activity has the single biggest impact on
corporate reputation. The elasticity of the volume of PR activity is over 1.5 times the
elasticity of advertising on corporate reputation. This further reinforces the importance of
public relations as an element of the marketing mix. Additionally, the content variables,
namely, prominence, subject and valence, all show significant effects for PR stories.
Negative content in PR stories seems to have nearly twice the impact of negative content
in independent news stories. On the other hand, positive content in PR stories although
significant, shows lower elasticity than positive content in independent media stories.
Amongst the various subject classifications, stories about products and services have the
biggest impact followed by financial performance and social responsibility. Executive
Performance and Workplace and Employees seem to have no significant impact.

One interesting finding is that for both press releases as well as independent media
coverage, the volume and valence of the stories have the biggest impact while the subject
matter of the story seems to play a secondary role. Additionally, consistent with prior
theory on negativity bias (Rozin & Royzman, 2001; Ito, Larsen, Smith & Cacioppo,
1998), negative news stories seem to have a disproportionately bigger impact on
consumer perceptions about corporate reputation. This indicates that consumers seem to
consider negative stories to be far more diagnostic than positive stories regardless of the
source of the information.
81

Drivers of PR Activity & Independent News Media Coverage

One interesting finding is that PR activity is driven by prior reputation as well as the
valence and subject matter of media coverage. Both positive and negative news stories
have a positive impact on the number of press releases issued by the firm. On the other
hand, advertising seems to have no impact on PR. Thus, firms seem to treat advertising
and public relations as independent elements of the marketing mix.

Independent media coverage seems to be driven primarily by Public Relations. Both
volume and the subject matter of the press release have significant positive effects on the
number of independent news stories that appear about the firm. Here again, press releases
about products and services and financial performance seem to generate the most
independent media coverage.

Drivers of Advertising Expenditure

In general, advertising expenditure seems to be independent of the volume of both Public
Relations and independent media coverage. Advertising does not seem to be affected by
either the valence or subject matter of these news stories or press releases. Thus, contrary
to the notion that firms use advertising to counter the negative effect of unfavorable news
82

stories (Argenti & Druckenmiller, 2004), I actually find that advertising seems to operate
independent of media coverage and public relations.

Long-Term Elasticity of Buzz Variables

To quantify the long-run elasticity of public relations and media coverage on corporate
reputation, I calculate arc elasticities using the following approach. First, from the IRFs, I
compute the total change in the reputation, ∆Y, in response to a one standard deviation
shock to PR or MC. Second, using our data, I calculate the standard deviation for PR (σ
X
)
and mean values for reputation (Y ) and PR ( X ). Finally, I use the following equation to
calculate the arc elasticity,
arc
η :

.
arc
X
Y X
Y
η
σ
Δ
= ×
The above is a standard elasticity formula, except that
X
σ is substituted for ∆X. This
follows because
X
σ is the change in X that is used to generate the IRF. The results of the
elasticity computations are presented in Table 4. In the table, I present results at the 1
year, 2 year, 3 year and total long-term elasticity. Note that the 3-month elasticities here
are the short-term or immediate elasticities of the marketing variables.

83

Table 2.4: Short-Term vs. Long Term Elasticity of Corporate Reputation
1 Year 2 Year 3 Year Long Term
PR .164 .188 .216 .238
Advg .031 .047 .049 .050
MC .158 .183 .194 .216

The results for the long-term elasticities indicate that both public relations and
independent media coverage have a much slower decay rate than the effect of advertising
on corporate reputation and therefore yield higher long-term elasticity estimates. In fact,
the elasticity of PR is about 5 times the elasticity for advertising and the elasticity of
independent media coverage is about 4 times the advertising elasticity.

As previously mentioned, the impulse response functions (IRFs) trace the incremental
effect of a one-standard deviation shock in PR and Independent Media Coverage on the
future value of Corporate Reputation. These enable me to examine the dynamic effects of
each of these elements on box office revenues, fully accounting for the indirect effects of
these elements. Figures 1a and 1b plot these impulse response functions.

I look first at Figure 1a, the IRF for reputation with respect to a shock in PR volume. The
graph shows that the effect of a shock to PR lasts for as long as 5 years. Thus, the total
84

impact of public relations expenditure can only be examined by looking at the long-term
as well as immediate effect of PR on corporate reputation.

Turning to Figure 1b, I see that the long-term effect of an increase in independent media
coverage lasts approximately the same duration as an increase in PR but the magnitude of
total elasticity of independent media coverage is smaller than the impact of PR activity.

85

Figure 2.1 a: Response of CR to Shock in PR Volume

0
0.5
1
1.5
2
2.5
1 2 3 4 5
Response to Shock in PR
86

Figure 2.1 b: Response of CR to Shock in MC Activity

CONTRIBUTIONS, LIMITATIONS AND FUTURE RESEARCH
Contributions
I collect data on nearly all the newspaper media coverage received by a panel of firms
over a period of five years and link this media coverage to the reputation of the firm. I
examine the impact of different dimensions of the media coverage including prominence,
valence and subject matter. Additionally, I distinguish between public relations activities
87

and independent media coverage thereby providing insights on the extent to which firms
are able to use public relations to influence their own reputation. This research represents
the first attempt to quantify the impact of a specific public relation activity, specifically
the issuing of press releases by a firm, via its impact on an important aspect of firm
performance, namely the reputation of the firm. The approach yields estimates of both the
short- and long-run impact of this activity via the elasticities.

Measuring the valence of the large body of newspaper stories is a difficult problem. My
data consisted of a total of over 25,000 separate news stories that human coders would
find very time-consuming and tedious to rate. Therefore I develop a computerized
content-analysis approach that yields reliable measures of the valence of these news
stories that can then be linked to our corporate reputation measures. Another feature of
this technique is that it can be easily adapted to a variety of contexts including analyzing
consumer reviews, transcripts from focus groups and so on by developing a lexicon that
is suited to the context being studied.

Several interesting substantive findings about the differential impact of public relations
versus independent news stories emerge from the analysis. First, I find that positive
stories coming from the independent media have a much larger impact on corporate
reputation as compared to positive news stories that originate from the firm through press
releases. This is consistent with theories on source credibility being a key driver of the
persuasiveness of a message (Curtin, 1999; Friestad & Wright, 1994). On the other hand,
88

I find that negative stories coming from the firm itself have a significantly higher impact
than negative stories appearing independently in the media. I also find that the volume of
public relations activities is a key driver of the amount of independent media coverage
received by the firm. Coupled with the finding that in general, the volume of media
coverage has a positive impact on corporate reputation (“All Publicity is Good
Publicity”), this implies that firms can use public relations to build and maintain a strong
positive reputation.

Limitations & Future Research
Although this research goes further than any previous research in terms of capturing the
elasticity of public relations, I realize the need to link it to other measures of firm
performance such as sales and firm financial performance. Additionally, corporate
reputation may also provide other benefits to the firm such as the ability to hire and retain
better talent, hinder the entry of competitor firms and charge higher prices for products
and services. However, the direct role of public relations activities in providing these
advantages remains to be studied.

Another crucial gap in this area is to study the effect of other public relations activities
(besides press releases) on corporate reputation and other firm performance metrics.
Recently, there has been an increasing focus on the use of company blogs and websites as
a public relations tool. A key gap in this literature is to develop a comprehensive
89

typology of the various types of public relations activities undertaken by firms and
studying their relative impact on firm performance.
90

BIBLIOGRAPHY

Aaker, DA (1991), Managing Brand Equity, Free Press, New York.

Aaker, DA and Biel, AL (1993), Brand Equity and Advertising: Advertising’s Role in
Building Strong Brands, Lawrence Erlbaum Associates, Hillsdale, NJ.

Ailawadi, KL, Lehman, DR and Neslin, SA (2003), “Revenue Premium as an Outcome
Measure of Brand Equity,” Journal of Marketing, Vol. 67 (October), 1-17.

Anderson, EW (1998), “Customer Satisfaction and Word of Mouth,” Journal of Service
Research, Vol.1(1), 5-17.

Arndt, J (1967), “Role of Product-Related Conversations in the Diffusion of a New
Product,” Journal of Marketing Research, Vol.4 (3), 291-295.

Bansal, HS and PA Voyer (2000), “Word of Mouth Processes Within A Services
Purchase Decision Context,” Journal of Service Research, Vol. 3(2), 166-177.

Barney, JB (1991), “Firm Resources and Sustained Competitive Advantage,” Journal of
Management, Vol.17, 99-120.

Basuroy, S, Chatterjee, S and Ravid, SS (2003), “How Critical Are Critical Reviews? The
Box Office Effects of Film Critics, Star Power, and Budgets,” Journal of Marketing, Vol.
67, 103-117.

Basuroy, S and A Ravid (2008), “In Word of Mouth We Trust? Fresh Insights From the
Motion Picture Industry,” Working Paper, Carl De Santis Center for Motion Picture
Studies.

Benjamin, BA and Podolny, JM (1999), “Status, Quality and Social Order in the
California Wine Industry,” Administrative Science Quarterly, Vol. 44.

Benton, M and Frazier, PJ (1976), “The Agenda-Setting Function of the Mass Media at
Three Levels of Information-Holding,” Communication Research, Vol.3, 261-274.

Berlo, DK (1960), The Process of Communication, New York: Holt, Rinehart, Winston.

Bijmolt, Tammo H.A., Harald J. Van Heerde, Rik G.M. Pieters (2005), “New Empirical
Generalizations on the Determinants of Price Elasticity,” Journal of Marketing Research,
42 (May), 141-156.

Bone, PF (1995), "Word of mouth effects on short-term and long-term product
judgments", Journal of Business Research, Vol. 32 No.3, pp.213-23.

91

Box, GEP and GM Jenkins (1970), “Time Series Analysis: Forecasting and Control,” San
Francisco:Holden-Day.

Bromley, DB (1993), Reputation, Image and Impression Management, London: Wiley.

Brosius, HB and Kepplinger, HM (1990), “The Agenda-Setting Function of Television
News: Static and Dynamic Views,” Communication Research, Vol.17, 183-211.

Brown, JJ and PH Reingen (1987), “Social Ties and Word of Mouth Referral Behavior,”
Journal of Consumer Research, Vol.14(3), 350-362.

Calder, BJ, LW Phillips, and AM Tybout (1981), "Designing Research for Application",
Journal of Consumer Research, 8 (September), 197-207.

Carroll, CE (2002), “Corporate Reputation as a New Media Agenda Item,” Paper
presented at the Association of Educators in Journalism and Mass Communication,
Miami, FL.

Chatterjee, P (2001), "Online Reviews - Do Consumers Use Them?" ACR 2001
Proceedings, eds. M. C. Gilly and J. Myers-Levy, Provo, UT: Association for Consumer
Research, 129-134.

Chen, Y and J Xie (2005), “Third-Party Product Review and Firm Marketing Strategy,”
Marketing Science, 24(2), 218-240.

Chevalier, J and D Mayzlin (2005), “The Effect of Word of Mouth on Sales: Online
Book Reviews,” Journal of Marketing Research, forthcoming.

Christian Science Monitor (2006), “Film Critics Getting Left In the Dark,” Stephen
Humphries.

Crane, FG (1989), "Choice Criteria and Cue Usage in Selecting Lawyers," Journal of
Professional Services Marketing, Vol. 5 (1), 113-121

Dearing, JW and Rogers, EM (1996), Agenda-Setting, Thousand Oaks, CA: Sage.

Deephouse, DL (2000), “Media Reputation as a Strategic Resource: An Integration of
Mass Communication and Resource-Based Theories,” Journal of Management, Vol.26,
No.6, 1091-1112.

DeFleur, M and Ball-Rokeach, SJ (1982), Theories of Mass Communication, 4
th
Ed.,
New York: McKay.

92

DeFleur, ML, Davenport, L, Cronin, M and DeFleur, M (1992), “Audience Recall of
News Stories Presented by Newspaper, Computer, Television and Radio,” Journalism
Quarterly, Vol.69, 1010-1022.

Dekimpe, MG and DM Hanssens (1999), “Sustained Spending and Persistent Response:
A New Look at Long Term Marketing Profitability,” Journal of Marketing Research, 23,
397-412.

Dellarocas, C, NF Awad and X Zhang (2005), “Exploring the Value of Online Product
Ratings in Revenue Forecasting: The Case of Motion Pictures,” Working Paper.

De Vany, A and WD Walls (1997), “The Market for Motion Pictures: Rank, Revenue and
Survival,” Economic Inquiry.

De Vany, A and WD Walls (1999), “Uncertainty in the Movie Industry: Does Star Power
Reduce the Terror of the Box Office,” Journal of Cultural Economics, Vol. 23(4).

De Vany, A and WD Walls (2002), “Does Hollywood Make Too Many R-rated Movies?
Risk, Stochastic Dominance and the Illusion of Expectation,” Journal of Business,
Vol.75.

Dierickx, I and Cool, K (1989), “Asset stock accumulation and sustainability of
competitive advantage,” Management Science, Vol. 35, 1504–1511.

Eaton, H Jr. (1989), “Agenda-Setting with Bi-Weekly Data on Content of Three National
Media,” Journalism Quarterly, Vol.66, 942-948.

Elberse, A, J Eliashberg (2003), “Demand and Supply Dynamics for Sequentially
Released Products in International Markets: The Case of Motion Pictures,” Marketing
Science, 22(3), 329-354.

Eliashberg, J, SM Shugan (1997), “Film Critics: Influencers or Predictors,” Journal of
Marketing, 61(2), 68-78.

Eliashberg, J, A Elberse and M Leenders (2006), “The Motion Picture Industry: Critical
issues in Practice, Current Research and New Research,” Marketing Science, 25(6), 638-
661.

Enders, W (1995), Applied Econometric Time Series. New York: John Wiley.

Engel JF, RD Blackwell and RJ Kegerreis (1969), “How Information is Used to Adopt an
Innovation,” Journal of Advertising Research, Vol.9(4), 3-8.

Epstein, L (1978), Women and the News, New York: Hastings.
93

Erdem, T, Swait, J and Louviere, J (2002), “The Impact of Brand Credibility on
Consumer Price Sensitivity,” International Journal of Research in Marketing, Vol. 19,
No.1: 1-19.

Feldman, SP and MC Spencer (1965), "The Effect of Personal Influence in the Selection
of Consumer Services," in Marketing and Economic Development, Proceedings of the
Fall Conference of the American Marketing Association, Peter D. Bennett, ed. Chicago:
American Marketing Association, 440-452.

Fishbein, M and Ajzen, I (1975), “Belief, Attitude, Intention and Behavior: An
Introduction to Theory and Research,” Addison-Wesley, Reading, MA.

Folkes, VS (1988), “Recent Attribution Research in Consumer Behavior: A Review and
New Direction,” Journal of Consumer Behavior, Vol.14, 548-565.

Fombrun, CJ (1996), Reputation: Realizing Value From the Corporate Image. Boston:
Harvard Business School Press.

Forbes (2006), “Freezing Out The Critics,” Babej and Pollak.

Funkhouser, GR (1973), “The Issues of the Sixties: An Exploratory Study in the
Dynamics of Public Opinion,” Public Opinion Quarterly, Vol.37, 62-75.

Fürisch, E (2002), “Nation, Capitalism, Myth: Covering News of Economic
Globalization,” Journalism and Mass Communication Quarterly, Vol.79, No.2, 353-373.

Gitlin, T (1981), “Media Sociology: The Dominant Paradigm,” in G.C. Wilhoit and H.
DeBock (Eds.), Mass Communication Review Yearbook (Vol.2, 73-122). Thousand Oaks,
CA: Sage Publications.

Godes, D and D Mayzlin (2004), “Using Online Conversations to Study Word of Mouth
Communication,” Marketing Science, 23(4), 545-560.

Golan, G and Wanta, W (2001), “Second-Level Agenda Setting in the New Hampshire
Primary: A Comparison of Coverage in Three Newspapers and Public Perceptions of
Candidates,” Journalism & Mass Communication Quarterly, Vol.78, 247-259.

Granger, CWJ (1969) Prediction with a generalized cost of error function. Operational
Research Quarterly 20, 199-207.

94

Greenleaf, EA (1992), “Measuring Extreme Response Style,” The Public Opinion
Quarterly, 56(3), 328-351.

Hanssens, Dominique M. (1980), "Market Response, Competitive Behavior, and Time
Series Analysis," Journal of Marketing Research, 17 (November), 470-85.

Hanssens, Dominique M., Leonard J. Parsons, and Randall L. Schultz (2001), Market
Response Models, 2nd ed. Boston, Mass.: Kluwer Academic Publishers.

Heath, T, Motta, E and Petre, M (2006), “Person to Person Trust in Word of Mouth
Recommendation,” Proceedings of the Workshop on Reinventing Trust.

Herr, PM, FR Kardes, J Kim (1991), “Effects of Word-of-mouth and Product Attribute
Information on Persuasion: An Accessibility-Diagnosticity Perspective,” Journal of
Consumer Research, Vol. 17(4).

Holbrook, MB (1999), “Popular Appeal versus Expert Judgments of Motion Pictures,”
Journal of Consumer Research, Vol. 26 (September), 144-155.

Horland, CI and Weiss, W (1951), “The Importance of Source Credibility on
Communication Effectiveness,” Public Opinion Quarterly, Vol.15, 635-650.

Houston, M, T Hennig-Thurau, M Spann and B Skiera (2008), “Consumer Anticipation
of New Products: Conceptualization and Empirical Evidence Regarding Pre-Release
Buzz,” 2008 Summer Educator’s Conference, American Marketing Association.

Hu, N, PA Pavlou and J Zhang (2006), “Can Online Reviews Reveal a Product’s True
Quality? Empirical Findings and Analytical Modeling of Online Word-of-Mouth
Communication,” Proceedings of the 7th ACM conference on Electronic commerce.

Ito, TA, JT Larsen, NK Smith and JT Cacioppo (1998), “Negative information weighs
more heavily on the brain: The negativity bias in evaluative categorizations,” Journal of
Personality and Social Psychology, 75, 887–900.

Iyengar, S and Kinder, DR (1987), News That Matters: Television and American
Opinion, Chicago: University of Chicago Press.

Jensen, MC and Meckling, WH (1976), “Theory of the Firm: Managerial Behavior,
Agency Costs, and Ownership Structure,” Journal of Financial Economics, Vol.3, 305-
360.

Katz, E and Lazarsfeld, PF (1955), Personal Influence, New York: Free Press

95

Keller, KL (1993), “Conceptualizing, Measuring and Managing Customer-Based Brand
Equity,” Journal of Marketing, Vol. 57, No. 1: 1-22.

Kelly, HH (1971), Attribution in Social Interaction, New York: General Learning.

Kelly, HH and Michela, JL (1980), “Attribution Theory and Research,” Annual Review
of Psychology, Vol.31, 457-501.

Kiousis, S (2004), “Explicating Media Salience: A Factor Analysis of New York Times
Issue Coverage During the 2000 US Presidential Election,” Journal of Communication,
Vol.54, No.1, 71-87.

Lippman, W (1922), Public Opinion, New York: Harcourt Brace.

Liu, Y (2006), “Word-of-Mouth for Movies: Its Dynamics and Impact on Box Office
Revenue,” Journal of Marketing, 70(3), 74-89.

Lopez-Escobar, E, Llamas, JP, McCombs, M and Lennon, FR (1998), “Two Levels of
Agenda-Setting among Advertising and News in the 1995 Spanish Elections,” Political
Communication, Vol.15, No.2, 225-238.

Lowery, S and DeFleur, ML (1987), Milestones in Mass Communication Research:
Media Effects. New York: Longman.

Lutkepohl, H (1993) Introduction to Multiple Time Series, 2nd edn, Springer-Verlag,
Berlin.

Manheim, JB (1986), “A Model of Agenda Dynamics,” in ML McLaughlin (Ed.),
Communications Yearbook 10, Newbury Park, CA: Sage Publications.

McCombs, ME (1993), “The Evolution of Agenda-Setting Research: Twenty-Five Years
in the Marketplace of Ideas,” Journal of Communication, Vol.43, No.2, 58-67.

McCombs, M, Llamas, JP, Lopez-Escobar, E and Rey, F (1997), “Candidate Images in
Spanish Elections: Second Level Agenda-Setting Effects,” Journalism & Mass
Communication Quarterly, Vol.74, 703-713.

McCombs, M, Lopez-Escobar, E and Llamas JP (2000), “Setting the Agenda of
Attributes in the 1996 Spanish General Election,” Journal of Communication, Vol.50, 77-
92.

McCombs, ME and Reynolds, A (2002), “News Influence on Our Pictures of the World,”
in J Bryant & D Zillman (Eds.), Media Effects: Advances in Theory and Research, 1-18,
NJ: Lawrence Erlbaum Associates, Publishers.
96

McCombs, ME and Shaw, DL (1976), “Structuring the Unseen Environment,” Journal of
Communication, Vol.26, No.2, 18-22.

McMillan, G. S., & Joshi, M. P. (1997), “Sustainable competitive advantage and firm
performance: The role of intangible resources,” Corporate Reputation Review, Vol.1, 81–
85.

Moul, CC (2006) “Measuring Word of Mouth’s Impact on Theatrical Movie
Admissions,” mimeo, Washington University (St. Louis).

New York Times (2006), “Studios Turn Thumbs Down on Film Critics,” David Carr.

Nijs,VR, MG Dekimpe, JB Steenkamp and DM Hanssens (2001), “The Category-
Demand Effects of Price Promotions,” Marketing Science, 20(1), 1-22.

Palmgreen, P and Clarke, P (1977), “Agenda Setting with Local and National Issues,”
Communication Research, Vol.4, 435-452.

Pauwels, K, DM Hanssens and S Siddarth (2002), “The long-term Effects of Price
Promotion on Category Incidence, Brand Choice and Purchase Quantity,” Journal of
Marketing Research, 34, 421-439.

Pauwels, K, J Silva-Risso, S Srinivasan and DM Hanssens, (2004), “The Long-Term
Impact of New-Product Introductions and Promotions on Financial Performance and
Firm Value,” Journal of Marketing, 68(4), 142-156.

Petty, RE and Cacioppo, JT (1986), Communication and Persuasion: Central and
Peripheral Routes to Attitude Change. New York: Springer-Verlag.

Radas, S and S Shugan (1998), “Seasonal marketing and timing new product
introductions,” Journal of Marketing Research, Vol. 35(3), pp. 296-315.

Reese, SD and Danielian, LH (1989), “Intermedia Influence and the Drug Issue:
Converging on Cocaine,” in PJ Shoemaker (Ed.), Communication Campaigns About
Drugs, 29-46, Hillsdale, NJ: Lawrence-Erlbaum.

Resnick, P and R Zeckhauser (2003), “Trust Among Strangers in Internet Transactions:
Empirical Analysis of eBay’s Reputation System,” Working Paper, University of
Michigan, Ann Arbor.

Roberts, PW and Dowling, GR, (1997), “The value of a firm’s corporate reputation: How
reputation helps attain and sustain superior profitability,” Paper presented at the
97

Conference on Corporate Reputation, Image, & Competitiveness, New York University,
New York.

Rozin, P and Royzman, E (2001), “Negativity Bias, Negativity Dominance and
Contagion,” Personality and Social Psychology Review, Vol.5, 296-320.

Rumelt, RP (1987), “Theory, strategy, and entrepreneurship,” inn D. J. Teece (Ed.), The
competitive challenge: 137–158. Cambridge, MA: Ballinger.

Rupp, WT and Hamilton, RH (1996), “Reputation as innovation in international firms:
An empirical analysis,” Paper presented at the annual meeting of the Academy of
Management, Cincinnati.

Sawhney, M and J Eliashberg (1996), “A parsimonious model for forecasting gross box-
office revenues of motion pictures,” Marketing Science, 15(2), 113-131.

Schramm, W (1954), How Communication Works. In W.Schramm (Ed.), The Process
and Effects of Mass Communication, Urbana: University of Illinois Press, 3-26.

Severin, WJ and Tankard, JW Jr. (2001), Communication Theories. Origins, Methods
and Uses in the Mass Media. New York, NY: Addison Wesley Longman.

Shoemaker, PJ, Danielian, LH and Brendlinger, N (1992), “Deviant Acts, Risky Business
and US Interests: The Newsworthiness of World Events,” Journalism Quarterly, Vol. 68,
No. 4, 781-795.

Shoemaker, PJ and Reese, SD (1996), Mediating the Message: Theories of Influences on
Mass Media Content, White Plains, NY: Longman.

Smith, KA (1987), “Newspaper Coverage and Public Concern About Community
Issues,” Journalism Monographs, Vol.101, 1-32.

Spence, AM (1974), Market Signaling: Information Transfer in Hiring and Related
Screening Processes, Cambridge, MA: Harvard University Press.

Srivastava, RK, McInish, TH, Wood, RA and Capraro, AJ (1997), “The value of
corporate reputation: Evidence from the equity markets,” Corporate Reputation Review,
Vol. 1, 62–68.

Stempel, GH. III (1991), “Where People Really Get Most of Their News,” Newspaper
Research Journal, Vol.12, No.4, 2-9.

98

Sternthal, B, Dholakia, RR and Levitt, C (1978), “The Persuasive Effect of Source
Credibility: Test of Cognitive Response,” Journal of Consumer Research, Vol.4, 242-
250.

Swami, S, J Eliashberg and CB Weinberg (1999), “SilverScreener: A Modeling
Approach to Movie Screens Management,” Marketing Science, Vol.18(3)

Takeshita, T and Mikami, S (1995), “How Did Mass Media Influence the Voter’s Choice
in the 1993 General Election in Japan?,” Keio Communication Review, Vol.17, 27-41.

Tellis, Gerard J. (1988), “The Price Sensitivity of Selective Demand: A Meta-Analysis of
Econometric Models of Sales,” ,” Journal of Marketing Research, 25 (November), 391-
404.

Vergin, RC and Qoronfleh, MW (1998), “Corporate reputation and the stock market,”
Business Horizons, 41 (1): 19–26.

Waddock, S (2000), “The Multiple Bottom Lines of Corporate Citizenship: Social
Investing, Reputation and Responsibility Audits,” Business and Society Review, Vol.105,
No.3, 323-345.

Wartick, SL (1992), “The Relationship Between Intense Media Exposure and Change in
Corporate Reputation,” Business & Society, Vol.31, 33-49.

Weaver, DH (1996), “What Voters Learn From the Media,” Annals of the American
Academy of Political and Social Science, Vol.546, 34-47.

Weaver, DH, Graber, DA, McCombs, ME and Eyal, C (1981), Media Agenda Setting in a
Presidential Election: Issues, Images and Interest. New York, NY: Praeger.

Weigelt, K and Camerer, C (1988), “Reputation and corporate strategy: A review of
recent theory and applications,” Strategic Management Journal, 9: 443–454.

Whetten, DA and Mackey, A (2002), “A Social Actor Conception of Organizational
Identity and Its Implications for the Study of Organizational Reputation,” Business &
Society, Vol.41, No.4, 393-414.

Whitney, DC and Becker, LB (1982), “Keeping the Gates for Gatekeepers: The Effects of
Wire News?” Journalism Quarterly, Vol.59, 60-65.

Winter, JP and Eyal, CH (1981), “Agenda-Setting for the Civil Rights Issue,” Public
Opinion Quarterly, Vol.45, No.3, 376-383.

99

Wirtz, J and P Chew (2002), “The Effects of Incentives, Deal Proneness, Satisfaction
and Tie Strength on Word-of-Mouth Behavior,” International Journal of Service Industry
Management, 13(2), 141-162.

Yoo, B, Donthu, N and Lee, S (2000),”An Examination of Selected Marketing Mix
Elements and Brand Equity,” Journal of the Academy of Marketing Science, Vol. 28,
No.2, 195-211.

Zajac, EJ and Westphal, JD (1995), “Accounting for the Explanations of CEO
Compensation: Substance and Symbolism,” Administrative Science Quarterly, Vol.40,
283-308.
100

Appendix : Checking the Reliability of Computerized Text-Analysis

In order to test the reliability of the measures generated through this automated
process, I used three independent human raters to count the number of positive and
negative words/phrases in 5,000 randomly sampled reviews. First, I tested for inter-rater
correlation and found a minimum correlation of 0.92 between any two raters. I then tested
the correlation between the measures generated by the software for each of these reviews
with the average rating given by the human raters. I found a correlation of 0.87 between
the measures generated by Diction and the average of the human raters.
I also performed an additional reliability check based on another subsample of
reviews. I randomly sampled 5,000 reviews and instructed two human judges to
“carefully read the following reviews and rate them on a scale of 1 to 5 where 1 = “The
review is completely negative”, 2 = “The review is mostly negative with some positive
content”, 3 = “The review has almost equal positive and negative content”, 4 = “The
review is mostly positive with some negative content” and 5 = “The review is completely
positive”
2
. I then used the scores from the computerized text analysis to compute an
overall net valence score for each buzz message measured as the count of positive
words/phrases minus the count of negative words/phrases found in the message. I
subsequently plotted the empirical distribution of these net valence scores and divided the
distribution into quintiles. I then assigned any review whose score fell in the first quintile

2
We removed the star ratings accompanying these reviews so that the ratings given by the human judges
would be independent on the star ratings that were provided by the original review writers. The idea here is
that the textual content does not correspond very well to the star ratings provided and that these human
judges provide a rating that is more reflective of the textual content of the review.
101

as a 1 on the scale of 1 to 5 and so on until the 5
th
quintile. I calculated the Spearman rank
correlation between the two judges and found a correlation of 0.92. I then examined the
Spearman rank correlation between the average rating of the two judges and the ratings
assigned using the valence scores from the computer program and found a correlation
coefficient of 0.85 indicating that the scores from the computer program are reliable
indicators of the true valence of the buzz messages.

Abstract (if available)

Abstract Word-of-Mouth (WoM) has been recognized as one of the most influential sources of information transmission. Recent advances in information technology have profoundly changed the way in which WoM is transmitted leading to resurgence in interest in this topic among practitioners and academics.

Linked assets

University of Southern California Dissertations and Theses

Asset Metadata

Creator Pai, Seema (author)

Core Title Does it matter what people say about you: the impact of the content of buzz on firm performance

School Marshall School of Business

Degree Doctor of Philosophy

Degree Program Business Administration

Publication Date 08/07/2008

Defense Date 05/05/2008

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag buzz marketing,corporate reputation,Forecasting,mass media,Motion picture industry,OAI-PMH Harvest,Public Relations,time series analysis,word of mouth

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Siddarth, Sivaramakrishnan (committee chair), Dutta, Shantanu (committee member), Mizik, Natalie (committee member), Moon, Hyungsik Roger (committee member), Tellis, Gerard (committee member)

Creator Email seema.pai@gmail.com,seemapai@bu.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m1556

Unique identifier UC1214387

Identifier etd-Pai-2312 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-105483 (legacy record id),usctheses-m1556 (legacy record id)

Legacy Identifier etd-Pai-2312.pdf

Dmrecord 105483

Document Type Dissertation

Rights Pai, Seema

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu