University of Southern California Dissertations and Theses

Essays on the turning points of the product life cycle

(USC Thesis Other)

Essays on the turning points of the product life cycle

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ESSAYS ON THE TURNING POINTS OF
THE PRODUCT LIFE CYCLE

by
Deepa Chandrasekaran

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)

May 2007

Copyright 2007 Deepa Chandrasekaran
ACKNOWLEDGEMENTS
I owe a debt of gratitude to my advisor Gerard Tellis, who has motivated me
in my work with his incredible energy, absolute commitment to research, patient
mentoring and incessant support of all my endeavors.
I am very grateful to S. Siddarth, Gareth James and Lan Luo who were on my
dissertation committee. They have always had an ‘Open door’ policy and been
available to answer any questions, no matter how trivial. I thank the faculty of the
Marketing department, particularly Fred Zufryden, Allen Weiss, Debbie McInnis and
Rakesh Niraj for their constant support.
I thank my friends and colleagues Maria Ogneva, Seema Pai, Srabana
Dasgupta, Nilesh Saraf, Anil Srinivasan, Vanessa Patrick, Eden Yin, Shashi Matta,
Ramkumar Janakiraman, Ashish Sood, Tae-Kyun Kim and Hae Eun Chun, for their
support and encouragement throughout these years.
I thank my parents for laying the foundations of my academic career by being
inspiring teachers and guides to their many students, and for being there for me every
step of the way. I thank my in-laws and grandparents for taking pride in my
accomplishments and my brother Bharath and sister-in-law Sirsha for providing the
laughs at the end of a tiring day. I dedicate this dissertation to my husband Ram, my
best friend and most loyal supporter through the turning points of my years as a
doctoral student. It has been the best of times. There is light at the end of the thesis.
ii
TABLE OF CONTENTS
ACKNOWLEDGEMENTS .........................................................................................ii

LIST OF TABLES .......................................................................................................v

LIST OF FIGURES ...................................................................................................vii

ABSTRACT................................................................................................................ix

CHAPTER 1: INTRODUCTION AND OUTLINE ...................................................1
1.1 MOTIVATION ..................................................................................................1
1.2 OVERVIEW OF ESSAY 1................................................................................3
1.3 OVERVIEW OF ESSAY 2................................................................................3
1.4 OVERVIEW OF ESSAY 3................................................................................7

CHAPTER 2: A CRITICAL REVIEW OF MARKETING RESEARCH ON
DIFFUSION OF NEW PRODUCTS.........................................................................10
2.1 INTRODUCTION ...........................................................................................10
2.2 POTENTIAL GENERALIZATIONS..............................................................13
2.3 FUTURE RESEARCH ....................................................................................23
2.4 BASS MODEL OF DIFFUSION ....................................................................27
2.5 IMPROVEMENTS IN SPECIFICATION OF BASS MODEL ......................33
2.6 IMPROVEMENTS IN ESTIMATION............................................................49
2.7 ALTERNATE MODELS OF DIFFUSION.....................................................55
2.8 MODELING DIFFUSION ACROSS SPACE ................................................67
2.9 MODELING DIFFUSION OF ENTERTAINMENT PRODUCTS................70
2.10 MODELING TURNING POINTS IN DIFFUSION .....................................73
2.11 CONCLUSION..............................................................................................88

CHAPTER 3: GLOBAL TAKEOFF OF NEW PRODUCTS: CULTURE’S
CONSEQUENCES OR WEALTH OF NATIONS? .................................................90
3.1 INTRODUCTION ...........................................................................................90
3.2 THEORY AND HYPOTHESES .....................................................................93
3.3 METHOD.......................................................................................................103
3.4 MODEL .........................................................................................................114
3.5 RESULTS ......................................................................................................115
3.6 CONVERGENCE IN TIME-TO-TAKEOFF................................................130
3.7 TESTS OF ROBUSTNESS ...........................................................................134
3.8 DISCUSSION ................................................................................................136

iii

CHAPTER 4: GETTING A GRIP ON THE SADDLE: CYCLES, CHASMS,
OR CASCADES? ....................................................................................................143
4.1 INTRODUCTION .........................................................................................143
4.2 THEORY AND HYPOTHESES ...................................................................147
4.3 METHOD.......................................................................................................154
4.4 RESULTS ......................................................................................................166
4.5 ROBUSTNESS ANALYSIS .........................................................................177
4.6 SUMMARY OF FINDINGS .........................................................................182
4.7 CONTRIBUTIONS .......................................................................................183
4.8 MANAGERIAL IMPLICATIONS................................................................183
4.9 FUTURE DIRECTIONS ...............................................................................184
4.10 LIMITATIONS............................................................................................185

REFERENCES.........................................................................................................186

iv
LIST OF TABLES
Table 1: Studies Included for Assessing Potential Generalizations...........................13
Table 2: Factor Analysis of Economic Variables ....................................................116
Table 3: Time-to-takeoff across Countries .............................................................118
Table 4: Estimates from the Hazard Model .............................................................121
Table 5: Time-to-Takeoff by Product Class and Economic Development..............122
Table 6: Cultural Clusters in Europe and America ..................................................124
Table 7: Cultural Clusters in Latin America and Europe.........................................125
Table 8: Cultural Clusters in Asia and Africa..........................................................126
Table 9: Time-to-takeoff across Cultural Clusters...................................................127
Table 10: Hazard Model with Cultural Clusters ......................................................128
Table 11: Hazard Model for Fun versus Work Products .........................................130
Table 12: Comparison of 2% Rule and Threshold Rule of Takeoff ........................135
Table 13: Saddle Occurrence by Products ...............................................................167
Table 14: Comparisons across Countries.................................................................169
Table 15: Saddle by New versus Old Categories.....................................................170
Table 16: Hazard Model Results..............................................................................172
Table 17: Hazard Model with Technological Change .............................................175
v
LIST OF TABLES

Table 18: Hazard Model with Repurchase Lag........................................................176
Table 19: Hazard Model with Magnitude of Economic Expansion.........................179
Table 20: Hazard Model with Leading Economy Indicator.....................................181
vi
LIST OF FIGURES
Figure 1: New Product Diffusion Curves ..................................................................15
Figure 2: Takeoff of Microwave Oven Sales in Europe ............................................74
Figure 3: Slowdown of Dishwasher Sales in Europe.................................................82
Figure 4: Slowdown of Computer Sales in Europe....................................................82
Figure 5: Time-to-Takeoff over Calendar Time .....................................................120
Figure 6: Time Spread by Year of First Takeoff......................................................133
Figure 7: Time Spread by Year of First Commercialization....................................133
Figure 8: Saddle in New Product Sales....................................................................145
Figure 9: Saddle in Microwave Oven Sales in UK..................................................145
Figure 10: Saddle in PC sales in USA .....................................................................146
Figure 11: Chasms in Adopter Segments.................................................................148
Figure 12: Saddle due to Chasms.............................................................................149
Figure 13: Information Cascades in Consumer Markets..........................................150
Figure 14: Saddle Due to Negative Cascades ..........................................................151
Figure 15: Saddle and Recessions............................................................................152
Figure 16: Threshold Rule for Takeoff ....................................................................156
Figure 17: Identification of Year of Saddle .............................................................157
vii
LIST OF FIGURES

Figure 18: Cyclical Component of GDP Series for Austria.....................................158
Figure 19: Repurchase Lags in Product Sales..........................................................164
Figure 20: Time from Takeoff to Saddle by Product Class .....................................167
Figure 21: Depth of Saddle by Product Class ..........................................................168
Figure 22: Duration of Saddle by Product Class......................................................168
Figure 23: Takeoff and Saddle by Expansion and Recession..................................173
Figure 24: Takeoff and Saddle by Market Penetration ............................................174
Figure 25: Washing Machine Sales .........................................................................182

viii
ABSTRACT
This dissertation examines how and why diffusion of new products varies
across products, countries and time. The word product refers to a product category
and not the brand.
Chapter 1 gives an overview of the three essays that form part of the
dissertation.
The first essay in Chapter 2 is a critical review of the literature on models of
the diffusion of new products and turning points of the diffusion curve. It critically
examines the characteristics, models and drivers of new product growth, delineates
turning points of the diffusion curve- takeoff and slowdown, and discovers important
potential generalizations to describe empirical findings with substantial support.
The second essay in Chapter 3 examines how and why time-to-takeoff vary
across major economies, whether these differences are constant or varying over time,
and, if varying, is time-to-takeoff converging or diverging across countries? These
issues are examined using a heterogeneous sample of 31 countries from different
cultural clusters across 16 categories over time. On this metric of time-to-takeoff,
Japan, is the most innovative country, followed by Nordic and Anglo American
countries. While takeoff has been held previously to be a purely cultural
phenomenon, this essay finds that both economics and culture are dominant drivers
of time-to-takeoff. Most importantly, time-to-takeoff is shortening over time and
converging across countries.

ix

The third essay in Chapter 4 examines the second turning point in product
sales- Slowdown, which leads to a trough in sales, termed as the ‘Saddle’. This essay
integrates and distinguishes between rival explanations for the Saddle- Business
cycles, Chasms in adopter segments, and Information Cascades. A Saddle in sales
occurs in 109 of 160 product-country combinations from 12 new products in 18
countries. The drivers of the occurrence of the Saddle are tested via a discrete-time-
hazard model. The results find support for recessions, negative cascades and
important technological changes but not chasms in adopter segments as drivers of the
Saddle.
x

1
CHAPTER 1: INTRODUCTION AND OUTLINE
1.1 MOTIVATION
In this dissertation, the word product refers broadly to cover any good or
service at the category level, not the brand level. Diffusion is defined as the spread of
a new product across markets over time. Researchers commonly measure diffusion
using the sales and especially the market penetration of a new product during the
early stages of its life cycle. To characterize this phenomenon carefully, the author
adopts and revises the definitions of the stages and turning points of the product’s
life cycle by Golder and Tellis (2004):
Commercialization is the date a new product is first sold to a mass market.
Takeoff is the first dramatic and sustained increase in a new product’s sales.
Introduction is the period from a new product’s commercialization until its takeoff.
Slowdown is the beginning of a period of temporarily decreasing product sales after
takeoff, leading to the Saddle.
Saddle is the first trough in new product sales following the post takeoff peak.
Growth is the period from a new product’s takeoff until its slowdown.
Maturity is the period from a product’s slowdown until sales begin a steady decline.
In recent years, two main research approaches have addressed the
international diffusion of new products. One approach is the continued application of
the Bass diffusion model to understand the multi-market diffusion of new products.
These studies predominantly address the growth stage of the product life cycle. The

2
studies in this tradition either determine the role of new covariates in diffusion
(Takada and Jain 1991, Helsen, Jedidi and Desarbo 1993, Ganesh and Kumar 1996,
Van den Bulte and Stremersch 2004) or develop better models to estimate
parameters of the Bass model (Gatignon, Eliashberg and Robertson 1989, Talukdar,
Sudhir and Ainslie 2002). Despite their valuable contributions, most of these studies
focus on a limited number of countries, geographic regions and categories, biasing
the sample towards developed nations and established categories, and do not address
the early phase of the product life cycle.
A second approach is an analysis of the time-to-takeoff of a new product
across different countries of the world (Tellis, Stremersch, and Yin 2003). The
authors find that the time-to-takeoff varies dramatically across countries ranging
from an average time of 4 years in Scandinavian countries to 7 years in
Mediterranean countries. While clearly describing country differences, the study has
two limitations. First, it considers only takeoff, which occurs at around 2% of the
market penetration of new products. Thus, it ignores differences in the remaining 3%
to 98% of market penetration of new products across countries. Second, it focuses
only on western European countries.
This dissertation hopes to address these limitations by offering a
comprehensive view of how and why diffusion of new products varies across
products, countries and time. It covers developing and developed countries,
established and new products, and examines the patterns in both the early and later
stages of the product life cycle.

3
1.2 OVERVIEW OF ESSAY 1
Because new products affect every aspect of the life of individuals,
communities, countries, and economies, the study of the diffusion of new products is
of vital importance. The first essay critically reviews the research on the diffusion
of new products primarily in the marketing literature
There are three unique contributions in this review. It critically examines the
characteristics, models and drivers of the S-curve of new product growth, and also
delineates the specific turning points- takeoff and slowdown. It considers the vast
tradition of research in marketing built on the Bass model, and also other models of
diffusion, and drivers of new product diffusion other than communication. It
discovers important potential generalizations or regularities to describe empirical
findings with substantial support (from reviews or meta-analyses of the literature or
individual studies with a large sample of over ten categories or ten countries).
1.3 OVERVIEW OF ESSAY 2
1.3.1 Introduction
Markets are seeing faster introductions of new products and more intense
global competition than ever before. In this environment, firms need to know how
new products diffuse across countries, which markets are most innovative, and in
which markets they should first introduce new products.
Recently, studies have introduced and validated a new metric to measure how
quickly a market adopts a new product: the takeoff of new products (e.g., Agarwal

4
and Bayus 2002, 2004; Golder and Tellis 1997; Tellis, Stremersch and Yin 2003).
Takeoff marks the turning point between introduction and growth stages of the
product life cycle, when the sales of a new product enters a phase of rapid growth.
When used consistently across countries, this metric provides a valid means by
which to compare and analyze the innovativeness of countries. However, the existing
research suffers from: One, a limited focus on industrialized countries; Two, a lack
of agreement on the role of culture versus economics in driving the takeoff of new
products; Three, a lack of agreement on which countries are the most innovative and
hence the best launch pads for new products.
1.3.2 Research Questions
This essay examines the following questions to fill this gap in the research.
First, how does time-to-takeoff vary across the major economies of Asia, Europe,
North America, South America, and Africa? Second, what drives the variation in
time-to-takeoff across countries? Third, are these differences constant or varying
over time? Fourth, if varying, is time-to-takeoff converging or diverging across
countries?
1.3.3 Data, Measures and Method
The author collects market penetration data on 16 categories in 40 countries
across using a combination of syndicated sources, archival search and publicly
available data.

5
The year of takeoff is the first year the market penetration of a product
crosses 2%. Takeoff is modeled as a time dependent binary event. A parametric
hazard function is used to test the effects of the independent variables on the time to
takeoff.
The independent variables are time-varying economic variables such as
wealth (factor of economic development, mobility and exposure to information),
income disparity and economic openness (factor of trade and foreign investment);
and time-invariant cultural variables such as in-group collectivism, religiosity,
performance orientation, uncertainty avoidance and cultural clusters of countries.
1.3.4 Key Findings
The study leads to several new findings:
• Time-to-takeoff is getting shorter over calendar time. In addition, there is
strong convergence in time-to-takeoff over calendar time among developed
countries.
• Despite these two effects, differences across countries are quite strong.
o Products takeoff fastest in Japan and Norway, followed by other
Nordic countries, US and some countries of Mid-western Europe.
o Newly developed countries of Asia (e.g., South Korea) see faster
times-to-takeoff of products than established, major European
countries (e.g., France, Italy) with centuries of industrialization.

6
o Latin countries across Europe and South America have similar times
to takeoff.
o Despite the recent and rapid increases in the GDP of emerging
markets such as China, India, Philippines, these countries still
substantially lag other countries in time-to-takeoff of new products.
• Takeoff is not a purely cultural phenomenon. Differences in both Economics
(Wealth) and Culture (In-group collectivism) account for differences in time-
to-takeoff across countries and regions.
• The mean time-to-takeoff varies considerably between developing countries
(11 years) and developed countries (7 years). The mean time-to-takeoff varies
between 6 and 12 years across cultural clusters.
• Time-to-takeoff varies considerably between fun products (7 years) and work
products (12 years).
o Fun products takeoff substantially faster than work products within
each cultural cluster.
o Takeoff of fun products also shows smaller differences across cultural
clusters than work products do.
o Takeoff of fun products is driven more by dynamic economic
variables and is be converging faster over time than work products.

7
1.4 OVERVIEW OF ESSAY 3
1.4.1 Introduction
In this essay, the author examines the patterns of and drivers of the Saddle in
new product sales across 12 products in 18 countries. The Saddle is the first trough
in product sales after the post takeoff peak. The pattern of a sharp decline averaging
nearly 20% in sales and lasting on an average more than 7 years has grave
consequences for new product marketers. Managers may, on seeing such a sharp
deviation in trend from the expected bell-shaped curve, pull the plug on new
products they think are performing poorly. Or their processes may be inadequately
geared to meeting a sudden slowdown in sales, and to the subsequent growth in sales.
1.4.2 Drivers of the Slowdown in Sales
This essay integrates competing explanations for the Saddle: Business cycles,
Chasms in adopter segments and Information Cascades across.
Recent literature has examined the relationship between product life cycles
and business cycles. It is thought that consumers are more likely to cut back on
discretionary expenditure such as spending on consumer durables during times of
economic contraction (Deleersnyder et al 2004, Golder and Tellis 1998, 2004).
Goldenberg, Libai and Muller (2002) argue that there is weak communication
across adopter segments, leading to dual markets- an early market and a late market.
They argue that a Saddle is observed if these markets adopt products at differing
rates. A conflicting explanation relies on the notion of Information Cascades: the

8
tendency of markets to snowball onto a popular behavior. Acceleration in sales
triggered by a positive cascade accompanied by a deceleration in sales immediately
afterward triggered by a negative cascade, may lead to a Saddle.
The impact of technological changes and repurchase lags are also considered
as control variables.
1.4.3 Data, Measures and Method
The sales history of 12 new products across 18 countries over time is
analyzed over time. A discrete time hazard model is used to determine the drivers of
the Saddle. The year of Saddle is the first year after takeoff when the sales declines
by 10 % or more from the initial peak and takes more than 2 years to cross the initial
peak.
Recessions are measured using the cyclical components of the GDP series in
each country using a filtering approach. Chasm in Adopter Segments is measured
using a measure of dispersion from the mean penetration at Saddle. Negative
cascades are measured at an aggregate cross-country level as the number of Saddles
that occur in the same category, in the same year or the year before in other countries.
1.4.4 Key Findings
110 products (> 60% of all products that takeoff) experience a Saddle. More
than 90% of all established products experience a Saddle. The Saddle occurs on
average 8 years from the time of takeoff. It leads to a sales decline by 18% on

9
average in the year of Saddle and an average depth of 27% from the year of the
Saddle to the year of Recovery. The average time to recovery is 7 years.
Recessions, Negative Cascades,Time since Takeoff are found to be dominant
drivers of the Saddle and robust to specifications. Technological changes, as
measured by Patent Importance have a positive effect on the hazard of a Saddle, but
there is no significant impact of Replacement Cycles.

10

CHAPTER 2: A CRITICAL REVIEW OF MARKETING
RESEARCH ON DIFFUSION OF NEW PRODUCTS
1
2.1 INTRODUCTION
Because new products affect every aspect of the life of individuals,
communities, countries, and economies, the study of the diffusion of innovations is
of vital importance. Researchers have studied this topic in various disciplines,
including marketing, economics, medicine, agriculture, sociology, anthropology,
geography, and technology management. This is a critical review of research on the
diffusion of new products primarily in the marketing literature, but also in the
economics and geography literature. The word product is used to broadly to cover
any good, service, idea, or person. This is distinguished from the broader term
innovation, which refers to new product and new method, practice, institution, or
social entity. The marketing literature on this topic is vast, dating back at least as
early as the publication by Fourt and Woodlock (1960).
The term diffusion has been used differently in two groups of literatures.
Within economics and most non-marketing disciplines, diffusion is defined as the
spread of an innovation across social groups over time (Brown 1981; Stoneman

1
One paper coauthored with Gerard J. Tellis was published based on this essay. The
details are: Deepa Chandrasekaran and Gerard J. Tellis (2007), “A Critical Review
of Marketing Research on Diffusion of New Products” Review of Marketing
Research, Vol 3, (In Press).

11
2002). As such, the phenomenon is separate from the drivers, which can be
consumer income, the product’s price, word-of-mouth communication, and so on. In
marketing and communication, diffusion typically has come to mean the
communication of an innovation through the population (Golder and Tellis 1998;
Mahajan, Muller, and Wind 2000a; Mahajan, Muller, and Bass 1990; Rogers 1995).
In this sense, the phenomenon (spread of a product) is synonymous with its
underlying driver (communication). The Webster (2004) definition of the noun
“diffusion” is “the spread of a cultural or technological practice or innovation from
one region or people to another, as by trade or conquest” and the verb “diffusing” is
“pour, spread out or disperse in every direction; spread or scatter widely.” This latter
interpretation is synonymous with the term’s use in economics and most other
disciplines. In addition, some researchers in marketing have subscribed to the
definition used in economics (Bemmaor 1994; Dekimpe, Parker, and Sarvary 2000a;
Van den Bulte and Stremersch 2004). Hence, in this review, diffusion is defined as
the spread of an innovation across markets over time.
Researchers commonly measure diffusion using the sales and especially the
market penetration of a new product during the early stages of its life cycle. To
characterize this phenomenon carefully, this review adopts and revises the
definitions of the stages and turning points of the product’s life cycle by Golder and
Tellis (2004):
Commercialization is the date a new product is first sold to a mass market.
Takeoff is the first dramatic and sustained increase in a new product’s sales.

12
Introduction is the period from a new product’s commercialization until its takeoff.
Slowdown is the beginning of a period of temporarily decreasing product sales after
takeoff, leading to the Saddle.
Saddle is the first trough in new product sales following the post takeoff peak.
Growth is the period from a new product’s takeoff until its slowdown.
Maturity is the period from a product’s slowdown until sales begin a steady decline.
Hence, there are two key turning points in the diffusion curve: takeoff and
slowdown.
Prior reviews address various aspects of the marketing literature on the
diffusion of new products. For example, Mahajan, Muller, and Bass (1990) provide
an excellent overview of the Bass model, its extensions, and some directions for
further research. Parker (1994) provides an overview of the Bass model and
evaluates the various estimation techniques, forecasting abilities, and specification
improvements of the model. Mahajan, Muller, and Bass (1995) summarize the
generalizations from applications of the Bass model. An edited volume by Mahajan,
Muller, and Wind (2000b) covers in depth various topics in diffusion models, such as
specification, estimation, and applications. Sultan, Farley, and Lehmann (1990) and
Van den Bulte and Stremersch (2004) meta-analyze the diffusion parameters of the
Bass model.
The current review differs from prior reviews in two important aspects. First,
the prior reviews mostly cover growth. This review focuses on phenomena besides
the S-curve, such as takeoff and slowdown. Second, the above reviews focus mainly

13
on the Bass model. This review considers the Bass model as well as other models of
diffusion and drivers of new product diffusion other than communication.
The next section summarizes potential generalizations from prior research. In
the third section, the limitations of past research and directions for future research
are pointed out. In the fourth section, key models and drivers of the diffusion curve
are evaluated. In the fifth section, models of the key turning points in diffusion:
takeoff and slowdown are evaluated.
2.2 POTENTIAL GENERALIZATIONS
The term potential generalization is used to describe empirical findings with
substantial support. This means that the support comes from reviews or meta-
analyses of the literature or individual studies with a large sample of over ten
categories or ten countries. Table 1 lists the studies on which the potential
generalizations are based.
Table 1: Studies Included for Assessing Potential Generalizations
Authors Categories Countries
Gatignon, Eliashberg and
Robertson (1989)
6 consumer
durables
14 European
countries
Mahajan, Muller and Bass (1990) Numerous
studies

Sultan, Farley and
Lehmann (1990)
213
applications
US, European
countries
Helsen, Jedidi and
DeSarbo (1993)
3 consumer
durables
11 European
countries
and US
Ganesh and Kumar (1996) 1 industrial
product
10 European
countries, US,
Japan

14
Table 1(continued): Studies Included for Assessing Potential Generalizations
Authors Categories Countries
Ganesh, Kumar,
Subramaniam (1997)
4 consumer
durables
16 European countries
Golder and Tellis (1997) 31 consumer
durables
USA
Putsis et al (1997) 4 consumer
durables
10 European
countries
Dekimpe, Parker and Sarvary
(1998)
1 service 74 countries
Kumar, Ganesh and
Echambadi (1998)
5 consumer
durables
14 European
countries
Golder and Tellis (1998) 10 consumer
durables
USA
Kohli, Lehmann and
Pae (1999)
32 appliances,
house wares
and electronics
USA
Dekimpe, Parker and
Sarvary (2000)
1 innovation More than
160 countries
Mahajan, Muller
and Wind (2000)
Numerous
studies
-
Van den Bulte (2000) 31 consumer
durables
USA
Talukdar, Sudhir,
Ainslie (2002)
6 consumer
durables
31 countries
Agarwal and Bayus (2002) 30 innovations USA
Goldenberg, Libai and
Muller (2002)
32 innovations USA
Tellis, Stremersch and
Yin (2003)
10 consumer
durables
16 European
countries
Golder and Tellis (2004) 30 consumer
durables
USA
Stremersch and Tellis (2004) 10 consumer
durables
16 European
countries
Van den Bulte and
Stremersch (2004)
293 applications 28 countries

15
2.2.1 Shape of the Diffusion Curve
The most important and most widely reported finding about new product
diffusion relates to the shape of the diffusion curve (see Figure 1). Numerous studies
in a variety of disciplines suggest that (with the exception of entertainment products)
the plot of sales of new products against time forms a Bell-shaped curve while the
plot of the cumulative sales of new products against time is an S-shaped curve (e.g.,
Mahajan, Muller, and Bass 1990; Mahajan, Muller, and Wind 2000a).
Figure 1: New Product Diffusion Curves

2.2.2 Parameters of the Bass Model
Most of the marketing studies use the Bass diffusion model to capture the
new products sales curve (see later section for explanation). This model has three
key parameters: the coefficient of innovation or external influence (p), the coefficient
of imitation or internal influence (q), and the market potential ( α or m).

16
Coefficient of Innovation
• The mean value of the coefficient of innovation for a new product lies
between 0.0007 and .03 (Sultan, Farley, and Lehmann 1990; Talukdar,
Sudhir, and Ainslie 2002; Van den Bulte and Stremersch 2004).
• The mean value of the coefficient of innovation for a new product is 0.001
for developed countries and 0.0003 for developing countries (Talukdar,
Sudhir, and Ainslie 2002).
• The coefficient of innovation is higher for European countries than for the
United States (Sultan, Farley, and Lehmann 1990).
Coefficient of Imitation
• The mean value of the coefficient of imitation for a new product lies between
0.38 and 0.53 (Sultan, Farley, and Lehmann 1990; Talukdar, Sudhir, and
Ainslie 2002; Van den Bulte and Stremersch 2004).
• Industrial/medical innovations have a higher coefficient of imitation than
consumer durables and other innovations (Sultan, Farley, and Lehmann 1990).
• The mean value of the coefficient of imitation for a new product is 0.51 for
developed countries and 0.56 for developing countries (Talukdar, Sudhir, and
Ainslie 2002).

17
Market Potential
The average market penetration potential ceiling of a new product is 0.52 for
developed countries and 0.17 for developing countries (Talukdar, Sudhir and Ainslie
2002).
Time to Peak Sales
It takes about nineteen years on average for a new product to reach peak sales
in developing countries, which is 18 percent longer than the average of sixteen years
for developed countries (Talukdar, Sudhir, and Ainslie, 2002).
Biases in Parameter Estimation
The nonlinear estimation of static models such as the Bass model leads to
downward biases in parameter values of market potential and the coefficient of
innovation and an upward bias in the coefficient of imitation (Van den Bulte and
Lilien 1997). The market potential can be underestimated by 20 percent, the
coefficient of innovation can be underestimated by 20 percent, and the coefficient of
imitation can be overestimated by 30 percent (Van den Bulte and Lilien 1997). Using
longer time series and using data with higher frequency is associated with lower
estimated q / p values (Van den Bulte and Stremersch 2004).

18
2.2.3 Drivers of Diffusion
• There is mostly indirect and some direct support for drivers of diffusion. Key
drivers in order of support are word-of-mouth communication, economics,
marketing mix variables such as prices, consumer heterogeneity, and
consumer learning (Dekimpe, Parker and Sarvary 1998, 2000a; Ganesh,
Kumar, Subramaniam 1997; Kumar, Ganesh and Echambadi 1998; Gatignon,
Eliashberg, and Robertson 1989; Mahajan, Muller, and Bass 1990; Mahajan,
Muller, and Wind 2000; Putsis et al. 1997; Stremersch and Tellis 2004;
Talukdar, Sudhir, and Ainslie 2002; Van den Bulte and Stremersch 2004).
• A 1 percent change in purchasing power parity adjusted per capita income is
likely to change the market penetration potential by about 0.3 percent
(Talukdar, Sudhir, and Ainslie 2002).
• A 1 percent change in international trade or urbanization is likely to change
the market penetration potential by about 0.5 percent and 0.2 percent
respectively (Talukdar, Sudhir, and Ainslie 2002).

19
2.2.4 Turning Points of the Diffusion Curve
This section describes findings about the regularities in takeoff, and
slowdown, the two turning points of the diffusion curve.
Patterns of Takeoff
Estimates of the average time to takeoff range are from six to ten years
(Agarwal and Bayus 2002; Golder and Tellis 1997; Kohli, Lehmann, and Pae 1999).
However, the average time to takeoff varies across products, countries, and time
(Tellis, Stremersch, and Yin 2003).
• Brown goods (entertainment and information products) take off faster, with
an average of two years, than white goods (kitchen and laundry appliances),
with an average of eight years (Tellis, Stremersch, and Yin 2003).
• The average time to takeoff of new products in Scandinavian countries is four
years, in mid-European countries the average is six years, and in
Mediterranean countries, it is eight years (Tellis, Stremersch, and Yin 2003).
• The average time to takeoff is eighteen years for categories introduced before
World War II (Golder and Tellis 1997), but only six to ten years for
categories introduced after World War II in the United States, as mentioned
above.

20
Drivers of Takeoff
• Every 1 percent decrease in price leads to a 4.2 percent increase in the
probability of takeoff (Golder and Tellis 1997).
• Takeoff in the number of firms in the market precedes product takeoff by at
least three years (Agarwal and Bayus 2002).
• The average penetration at takeoff is 1.7 percent (Golder and Tellis 1997).
This finding is consistent with Rogers’s (1995) estimate that innovators make
up 2.5 percent of the population and Mahajan, Muller, and Srivastava’s
(1990) upper bound of 2.8 percent for innovators.
Patterns of Slowdown
• Sales drop at slowdown in 50–96 percent of categories (Goldenberg, Libai,
and Muller 2002; Golder and Tellis 2004).
• Sales decline by an average 15–32 percent during these drops after slowdown
(Goldenberg, Libai, and Muller 2002; Golder and Tellis 2004)
Drivers of Slowdown
Price declines, market penetration, wealth, and information cascades seem to
influence the probability of slowdown (Golder and Tellis 2004). In particular,
• Every 1 percent increase in price is associated with a 5 percent increase in the
probability of slowdown.
• Slowdown occurs on average at 34 percent penetration.

21
• Every 1 percent increase in penetration is associated with a 3.6 percent
increase in the probability of slowdown.
• Every 1 percent decrease in total gross national product is associated with a
17 percent increase in the probability of slowdown.
• There is indirect evidence for information cascades driving sales increases
and declines in the early stage of the life cycle. Products that tend to have
large increases during takeoff seem to have large declines at slowdown.
2.2.5 Findings across Stages
This section compares the key findings on the duration, growth rates, and
price declines in the various stages and transition points of the product life cycle.
Duration
• On average, the duration of the introduction stage is six to ten years, of the
growth stage is eight to ten years, and of the early maturity stage is five years
(Agarwal and Bayus 2002; Goldenberg, Libai, and Muller 2002; Golder and
Tellis 2004; Golder and Tellis 1997; Stremersch and Tellis 2004; Tellis,
Stremersch, and Yin 2003).
• Timesaving products are associated with longer growth stages than non-
timesaving products (Golder and Tellis 2004).
• Leisure-enhancing products are associated with shorter growth stages than
non-leisure-enhancing products (Golder and Tellis 2004).

22
• The duration of the introduction and early maturity stages is getting shorter
over time, but not the duration of the growth stage (Golder and Tellis 2004).
• Overall, a new product reaching 5 percent household penetration in 1946 in
the United States took about fourteen years to go from 10 percent to 90
percent of its estimated maximum adoption ceiling. In 1980, that time has
dropped to about half, at seven years (Van den Bulte 2000).
Price
Price reductions are larger in recent periods for both the introduction and the
growth stages. The price at takeoff is 80 percent of the price at commercialization for
pre–World War II products and 63 percent for post–World War II products. The
price at slowdown is 56 percent of the price at commercialization for pre–World War
II products and 30 percent for post–World War II products (Golder and Tellis 2004).
Growth Rates
• The mean growth rate is 31 percent during introduction, 428 percent during
takeoff, 45 percent during growth, –15 percent during slowdown, –25 percent
during early maturity, and 3.7 percent during late maturity (Golder and Tellis
2004).
• The mean economic growth rate is 1 percent during introduction, 4.3 percent
during takeoff, 3.1 percent during growth, 0.86 percent during slowdown, 2.4
percent during early maturity, and 3.1 percent during late maturity of new
products (Golder and Tellis 2004).

23
• Timesaving products tend to have lower growth rates in the growth stage than
non-time-saving products (Golder and Tellis 2004).
• Leisure-enhancing products tend to have higher growth rates in the growth
stage than non-leisure-enhancing products (Golder and Tellis 2004).
• The average growth rate during the growth stage is 45 percent per year in the
United States, 46 percent for the Nordic countries, 41 percent for Mid-
European countries, and 36 percent for Mediterranean countries (Golder and
Tellis 2004; Stremersch and Tellis 2004).
2.3 FUTURE RESEARCH
Despite decades of research and a large body of potential generalizations in
diffusion, many problems remain unaddressed. This situation provides exciting
opportunities for future research. These are divided into four sections: measurement,
theories, models, and findings.
2.3.1 Measurement
The literature in this area has mostly ignored the problem of measurement.
Yet, measurement plays a critical role in documenting the phenomena under study.
Measurement is also an important prerequisite for modeling. For example, no clear
rules are available for the measurement of the start of the product life cycle or the
year of introduction of a new product. Most researchers consider the date from which
data become available as the date for the introduction of the new product. However,
syndicated data sources that track sales of new products tend to do so only when a

24
product has become popular and shows promise of becoming a mass-market product.
Using the date of availability of sales as a surrogate for the start date may grossly
underestimate the duration of the introductory period and the time for takeoff. In
addition, models such as the Bass model, which are highly sensitive to the number of
observations, can yield biased estimates and predictions due to erroneous start dates.
Researchers can correct for this by using model specifications that give statistically
valid estimates of the launch date.
In addition, most researchers use sales as the dependent variable. As such,
sales should consist only of first purchases of a new product. However, in effect,
most databases do not discriminate between first purchases and repurchases when
describing sales. In addition, the data measured as sales often represent “shipments,”
which captures supply of products rather than demand.
Further, researchers do not define a clear stopping rule for the period of the
study. The period modeled should end when the entire market has made first
purchases or at least when adoptions have peaked. Often researchers use the data
available until the first peak in sales.
The literature contains several competing measures for takeoff. Measures for
slowdown and the Saddle or a trough in sales are still tentative and have little
validation. Although under researched, measures for some of the key phenomena are
very important and play a critical role in the validity and interpretation of the
parameters of models. Perhaps this is the most important area for future research.

25
2.3.2 Theories
Researchers have identified various drivers for the diffusion of innovations.
However, no researcher has developed an integrated theory that either incorporates
or differentiates among all these drivers. This issue is important because theory
constitutes the key explanation for a phenomenon and informs good models and
managerial practice.
2.3.3 Models
In the area of modeling, there are five pressing issues.
First, most models have focused on modeling diffusion from slightly before
the takeoff to approximately the slowdown, while a few models have focused on
only takeoff and slowdown. Research needs to develop an integrated model of sales
from commercialization to takeoff, during growth, and after slowdown.
Second, the marketing literature has focused extensively on consumer
durables and a little on movies. Research needs to consider other categories such as
services, software, agricultural products, and medical products.
Third, research needs to include diffusion of products using new media such
as the Internet, where the process can be quite different from the traditional brick and
mortar medium.
Fourth, researchers are realizing that network effects can play a key
moderating role in the takeoff or success of a new product. Thus, research needs to

26
incorporate the role of network effects and technological choices of the suppliers on
product diffusion.
Fifth, the Bass model has long been the platform of diffusion research in
marketing because of its simplicity and good predictive ability. Researchers can
explore other platforms for research on diffusion.
2.3.4 Findings
While research in this area has led to some potential generalizations, further
research can help to ascertain the extent to which these generalizations either are
universal or vary by context. In particular, research could address the following three
issues.
First, the bulk of research has focused extensively on identifying patterns of
growth across countries and over time. There is also a need to identify subgroups or
regions within such populations where we are likely to see varying rates of diffusion.
Second, all research has focused on successful products. Future research
needs to study failed products to understand what aspects of their diffusion led to
failure.
Third, studies of diffusion speed have been largely limited to the United
States. Future research should consider the facets of cross-national speed of diffusion
together with how technology and entry strategy affect the speed of diffusion

27
2.4 BASS MODEL OF DIFFUSION
Much of the literature follows an early model by Bass (1969). The Bass
model is similar to epidemiological or contagion models, which describe the spread
of a disease through the population due to contact with infected persons (see Bailey
1957, 1975).
This section discusses the specification of the Bass model, evaluates the
model’s strengths and weaknesses, and discusses improvements in specification and
estimation.
2.4.1 Specification
The basic assumption in the Bass model is that the adoption of a new product
spreads though a population primarily due to contact with prior adopters. Hence, the
probability that an individual purchases at time T, given that the individual has not
purchased before, is a linear function of the number of previous buyers, thus
P(t) = f(t) / (1 – F(t)) = p + q / m Y(t) (1)
where P(t) is a hazard rate, which depicts the conditional probability of a
purchase in a (very small) time interval (t, t + ∆), if the purchase has not occurred
before t. Y(t) refers to the cumulative number of adopters up to time t; m is the total
number of initial purchases for the time interval for which replacement purchases are
excluded. F(t) denotes the cumulative fraction of adopters at time t and f(t) is the
likelihood of purchase at time t. The next equation is obtained by rearranging
equation 1.

28
f(t) = (p + qF(t))[1 – F(t)] (2)
Since Y(0) = 0, p represents the probability of an initial purchase at time 0
and its magnitude reflects the importance of innovators, the product q / mY(t) reflects
the pressure of prior adopters on imitators.
The number of adoptions at time t, S(t), is derived by multiplying f(t) in
equation (2) with m, the market size, thus:
S(t)= mf(t) = pm + (q – p) Y(t) – q / m Y
2
(t) (3)
Since f(t) = dF(t) / dt = (p + qF(t))[1 – F(t)] (4)
By rewriting this equation, Bass solves the following differential equation:
dt = dF / (p + (q – p)F – qF
2
) (5)
to obtain
F(t) = (1– e
– (p + q)t
)/ (1 + (q / p)e
– (p + q)t
) (6)
Hence, the cumulative adoptions are
Y (t) = m[(1 – e
– (p + q)t
) / (1 + (q / p)e
– (p + q)t
)] (7)
Bass rewrites equation (3) in a discrete form to obtain an equation for sales in
only three unknown parameters, which he estimates by simple regression, thus:
S
t
= a + bY
t – 1
+ cY
2
t – 1
, t = 2, 3 . . . (8)
where S
t
refers to sales at time t, Y
t – 1
refers to cumulative

sales through
period t – 1 and
a = p* m, (9)
b = q – p, (10)
c

=

–q / m (11)

29
Hence, he derives the values of p, q, and m from the estimated a, b, and c as
follows:
p = a

/ m (12)
q = –cm (13)
m = (–b ± (b
2
–

4ac)
1/2
) / 2c (14)
2.4.2 Evaluation
This section describes the strengths and limitations of the Bass model and
relates it to other models in the literature.
2.4.3 Strengths
The derived and testable function of the Bass Model (1969), equation (8), has
several excellent properties. First, because sales is a quadratic function of prior
cumulative sales, the model provides a good fit to the S-shaped curve that is typical
of the sales of most new products. Indeed, decades of subsequent research have
shown that the simple Bass model fits sales almost as well as much more complex
models that sought to correct its limitations (Bass, Krishnan, and Jain 1994).
Second, the model has two very appealing behavioral interpretations. Bass
interprets the coefficient p as the coefficient of innovation because it reflects the
spontaneous rate of adoption in the population. He interprets q as the coefficient of
imitation because it reflects the effect of prior cumulative adopters on adoption.
Other researchers conservatively interpret p as the external influence referring to the
influence of mass-media communications and q as internal influence referring to the

30
influence of interpersonal communication from prior adopters (Mahajan, Muller, and
Srivastava 1990).
Third, the model enables the researcher to resolve an important concern of
managers of new products, that is, to determine the time to, and magnitude of, peak
sales (t
*
and S(t)
*
), respectively. Bass shows that the time to peak sales and the
magnitude are, respectively:
t
*
= (1 / (p + q))* ln (q / p) (15)
S(t)
*
= m*(p + q)
2
/ 4q (16)
Fourth, the model encompasses two well-known earlier models in the
literature. If p = zero, the Bass model reduces to a logistic diffusion function,
assumed to be driven by only imitative processes (Fisher and Pry 1971; Mansfield,
1961; Van den Bulte 2000). If q = zero, the Bass model reduces to an exponential
function assumed to be driven only by innovative processes (Bernhardt and
Mackenzie 1972; Fourt and Woodlock 1960)
1
. Hence, the Bass model makes fewer
assumptions and is more general than these two models.
These four strengths of the Bass model account for its great appeal,
popularity, and longevity in the marketing discipline. Indeed, it has spawned a
paradigm of research in marketing, which remains unrivalled by any other model or
theory.

31
2.4.4 Limitations
Despite its strengths and strong appeal, the Bass model (1969) suffers from
several limitations. Subsequent research has sought to address these problems with
varying degrees of success. The next section describes these efforts.
First, any individual fit of the Bass model has poor predictive ability. The
model needs data at both turning points (takeoff prior to growth and slowdown prior
to maturity) to provide stable estimates and meaningful sensible forecasts. However,
by the time those events occur, the predictive value of the Bass model is limited. In
other words, the Bass model requires as inputs two of the most important events that
managers would like to predict: takeoff and slowdown.
Second, the model’s parameters are unstable and fluctuate with the addition
of new observations (Bemmaor and Lee 2002; Golder and Tellis 1998; Heeler and
Hustad 1980; Mahajan, Muller, and Bass 1990; Van den Bulte and Lilien 1997). This
variation in estimates for small changes in observations leads one to question
whether the parameters really capture the underlying behavior (internal and external
influences). Indeed, researchers question the basic assumption that product growth is
driven only by communication (Golder and Tellis 1998; Van den Bulte and Lilien
2001; Van den Bulte and Stremersch 2004). One of the strengths of the model may
account for the instability in parameters. The quadratic function fits the sales curve
so well that it sacrifices estimating the true underlying behaviors (Golder and Tellis
1998).

32
Third, the Bass model does not include the direct influence of any marketing
variable such as price or advertising. This is a serious problem because most
managers want to influence sales with these two variables. The model assumes,
however, that the coefficients m and p capture the effect of such external influences.
Fourth, the product definition in the Bass model is static, that is, it assumes
that the product itself does not change over time. However, there may be several
technological changes within a product category itself, before a dominant design
emerges (Srinivasan, Lilien, and Rangaswamy, 2006), and this variation is not
allowed for in the Bass model.
Fifth, Bass used OLS regression in the model to estimate the values of p, q,
and m. However, this method suffers from three shortcomings (Mahajan, Muller, and
Bass 1990). (1) There is likely to be multicollinearity between Y
t – 1
and Y
2
t – 1
making
the parameter estimates unstable. (2) The procedure does not provide standard errors
for the estimated parameters p, q, and m, and hence it is not possible to assess the
statistical significance of these estimates. (3) There is a time interval bias because the
model uses discrete time series data to estimate a continuous model.
Sixth, this tradition of research entails several problems in measuring the
dependent variable (sales) and determining the starting and ending points of the time
interval sampled. (1) Most researchers use sales as the dependent variable. As such,
sales should consist of only first adoptions of the new product. However, in effect,
most databases do not discriminate between first purchase and repurchases when
describing sales. (2) Sales should be from the very first year of commercialization of

33
the new product. However, in effect, the models only use published sales figures,
which often report sales when a product has already been selling well, if not after
takeoff of the product. (3) Researchers do not define a clear stopping rule for the
time interval. The period modeled should end when the entire market has made first
purchases or at least when adoptions have peaked.
The following sections describe how researchers correct for some of these
weaknesses by improving the estimation techniques, predictive ability, and model
specification.
2.5 IMPROVEMENTS IN SPECIFICATION OF BASS MODEL
The specification of the Bass model is very simple, as it contains no
deterministic explanatory variables. Over the past thirty-five years, a vast body of
literature has sought to enrich the model by including marketing variables, supply
restrictions, and multi-product interactions (such as the presence of competitive
products, complementary products, and newer technological generations),
incorporating time-varying parameters, replacement purchases, multiple purchases,
and trial and repeat purchases, and by analyzing cross-country diffusion patterns.
The subsections evaluate the literature concerning each of these improvements
concluding with an overall evaluation of this stream of literature.

34
2.5.1 Allowing Marketing Variables
Many authors consider the impact of marketing variables on new product
diffusion (Bhargava, Bhargava, and Jain 1991; Bass 1980; Bass, Krishnan and Jain
1994; Danaher, Hardie, and Putsis 2001; Jain and Rao 1990; Jones and Ritz 1991;
Kalish 1985; Kamakura and Balasubramanian 1988; Krishnan, Bass, and Jain 1999;
Horsky 1990; Horsky and Simon 1983; Robinson and Lakhani 1975).
A decline in price adds households whose reservation price structure
accommodates the new prices. Thus, price declines could affect the ultimate market
potential. Price declines could also stimulate the flow of households from being
potential adopters to adopters by increasing the probability of adoption. Kamakura
and Balasubramanian (1988) find that price seems to influence only the probability
of adoption and only for relatively high-price goods. Hence, the role of price seems
to be heterogeneous across products.
Other models incorporate the effects of advertising on diffusion (Horsky and
Simon 1983; Simon and Sebastian 1987). For instance, Horsky and Simon (1983)
include the level of the producer’s expenditures on advertising at time t directly into
the Bass model.
Researchers also consider the influence of the distribution process in
influencing diffusion. Jones and Ritz (1991) assume that there are two adoption
processes occurring for any new product, one for the retailers and one for the
consumers. Moreover, the number of retailers who have adopted the product

35
determines the size of the consumer’s potential market. The authors show that even
if the consumer adoption curve is exponential, when the initial level of distribution is
limited, the pattern of consumer adoptions takes an S-shaped curve similar to that
obtained from a Bass model.
Research on channels of distribution has focused typically on traditional
brick-and-mortar channels. Rangaswamy and Gupta (2000) discuss the application of
the Bass model to digital environments. They posit that the market potential for an
innovation, the coefficient of imitation, and the coefficient of innovation will be
larger, leading to increased sales and speed of adoption through online channels.
They also expect that in the digital environment, good products, with positive word-
of-mouth will succeed faster, whereas bad products, with negative word-of-mouth,
will fail faster.
Bass, Krishnan, and Jain (1994) include both price and advertising to give
what they call, the Generalized Bass model, wherein:
f(t) / [1 – F(t)] = [p + qF(t)]x (t) (17)
where x(t) is the current marketing effort that reflects the impact of price and
advertising on the conditional probability of product adoption at time t, such that
x(t) = 1 + β
1
∆Pr(t) / Pr(t – 1) + β
2
∆A(t) / A(t – 1) (18)
where ∆Pr(t) refers to Pr(t) – Pr (t – 1) and ∆A(t) refers to A(t) – A(t - 1).
Both these variables refer to the rates of changes in prices and advertising. The
model reduces to the Bass model when price and advertising remain the same from
one period to the next. Hence, the authors find that when percentage changes in the

36
decision variables are constant the Generalized Bass model provides no better fit
than the Bass model. Because the Bass model is quadratic in prior period’s
cumulative sales, it fits the S-shaped curve very well even when researchers omit
marketing variables. However, when the coefficients for the decision variables are
statistically significant, the Generalized Bass model provides a better fit than the
Bass model.
No study has empirically tested for the effect of all the marketing variables
simultaneously. The limitation of the empirical application by Bass, Krishnan, and
Jain (1994) is that they consider the effects of changes in only price and advertising
and not other marketing variables. However, the Generalized Bass model can
potentially include all relevant marketing variables and hence is managerially
relevant. The limitation of the model is that it considers only the effect of changes
and not the absolute levels of these variables. It also does not allow for the influence
of other important non-marketing factors that influence product growth such as
income changes.
2.5.2 Allowing Supply Restrictions
Jain, Mahajan, and Muller (1991) model the impact of restrictions on the
production capacity or the distribution system on the diffusion process. They model
the customer flow from being potential adopters to waiting applicants and from
waiting applicants to adopters, as follows:

37

dA(t) / dt = (p + (q
1
/ m)A(t) + (q
2
/ m)N(t)) (m – A(t) – N(t)) – c(t)A(t) (19)
and dN(t) / dt = c(t)A(t) (20)
In equation (19), d(A) / dt reflects the rate of changes of waiting applicants.
This is increased by the new applicants (first term on the right-hand side) generated
by the influence of both waiting population A(t) and adopters N(t) on the potential
applicants, but is decreased by the conversion rate of waiting applicants to adopters
(second term on the right-hand side) where c(t) is the supply coefficient. Equation
(20) captures the impact of supply restrictions on adoption rate at time t. The growth
process of the total number of new applicants is given by
dZ(t) / dt = dA(t) / dt + dN(t) / dt =
(p + (q
1
/ m)A(t) + (q
2
/ m)N(t)) (m – A(t) – N(t)) (21)
Though this model demonstrates a way to incorporate the effect of supply
restrictions, the authors assume that the level of capacity grows with the number of
back orders. However, in practice, this assumption may not hold. In addition,
dissatisfied consumers might cancel orders or negative word-of-mouth might
discourage others from ordering. Ho, Savin, and Terwiesch (2002) allow some
waiting applicants to abandon their adoption decisions after a point in time in their
theoretical model incorporating both demand and supply dynamics. Their results
suggest that when faced with the choice between selling an available unit
immediately versus delaying the sale to reduce the degree of future shortages, the

38
firm should always favor an immediate sale. The authors thus show that the time
benefit of immediate cash flows outweighs the limitation of demand acceleration.
Both these studies show sensitivity to distribution issues and offer an
opportunity to blend operations planning and marketing research. Such a confluence
helps managers to deal with the dilemma of keeping inventory low while making
products available to consumers (Cohen, Ho, and Matsuo 2000). Nevertheless, a still
greater challenge is the tackling of competitive effects.
2.5.3 Allowing Competitive Effects
While most models typically aggregate across individual diffusion processes
by studying the product class, asymmetries may exist in diffusion across brands
within a category.
Researchers consider the impact of competitive entry on the diffusion of
other brands. A new brand may have two effects: (1) it could increase the entire
market potential for the category due to increased promotion or product variety; and
(2) it could compete for the same market potential and hence slow down the
diffusion of the existing brands.
For instance, in an empirical application of the model to the instant-camera
market, Mahajan, Sharma, and Buzzell (1993) find that Kodak drew more than 30
percent of its sales from potential buyers of the pioneer brand, Polaroid. However, at
the same time, its entry also led to an expansion of the market. Krishnan, Bass, and
Kumar (2000) study the impact of a late entrant on the diffusion of a new product.

39
Using brand level sales data from the cellular telephone industry, they find that the
impact of entry of a new brand varies from market to market, increasing the market
potential of the category in some, hastening or slowing the diffusion process of other
brands in others. Parker and Gatignon (1994) find that in the category of hair-styling
mousses, for the pioneer, there seem to be strong brand identification effects and the
diffusion is independent of competitive effects. For the second brand and other
generic followers, prior adopters of the product class as a whole negatively influence
their trials. The sensitivity of the diffusion of these brands to marketing variables
also varies with the entry of competing brands.
Hence, research on competitive effects indicates that the diffusion process
may differ depending on the order of a new brand’s entry and the competition it
faces. However, while the models help determine the direction of the impact, they do
not clearly identify what causes these differential impacts across brands and markets.
2.5.4 Allowing Complementary Effects
Researchers have sought to account for the fact that the adoption of an
innovation is dependent on the presence of related innovations (e.g., Rogers 1995).
Bayus (1987) incorporates this notion in forecasting the sales of new contingent
products, that is, where the purchase of a product is contingent on the purchase of a
primary product. In an empirical application to the CD-player market, the author
demonstrates that the hardware sales can be modeled using a standard diffusion

40
framework and the software sales can be forecasted by calculating the sum of current
and future software purchase streams of first time hardware owners.
In markets with such indirect network externalities, the sales of software
could affect hardware sales as well. Subsequent papers have accounted for two-way
interactions in diffusion processes. Bucklin and Sengupta (1993) develop a model to
examine the co-diffusion (both one-way and two-way interactions) of two
complementary products, universal product codes (UPCs) and scanners. From their
analysis of the two categories, the authors find that co-diffusion does exist and may
be asymmetric in that one product has a stronger influence on the other product’s
diffusion than vice versa.
Gupta, Jain, and Sawhney (1999) incorporate the effect of indirect network
externalities from suppliers of digital programming in modeling the evolution of
digital TV sets. The authors use a combination of a latent class probit model of
consumer demand and complementor response models. Consumer demand for digital
TV is dependent on the hardware attributes and the software attributes of the set of
competing products. Complementor (suppliers of digital programming) response is
modeled as a function of the consumer demand for digital TV and exogenous
variables such as regulatory scenarios.
Lehmann and Weinberg (2000) focus on sequentially released products: new
products that are released sequentially across channels (for instance, movie releases
via movie theaters and then video rentals). A crucial question in the distribution of
these products is the optimal timing of release across the channels in the face of

41
cannibalization. Waiting too long to release the videos may reduce the marketing
impact from the theater release. The authors determine that the sales of the initial
product (theater attendance) can help forecast the sales of the sequential product
(videotape rentals), and also that the optimal time to release the video is sooner than
what is being done in practice.
These models reflect growing efforts to understand strategic
interdependencies among complementary and competing products. It would be
useful to model the effects of supplier actions/reactions, apart from consumer
response, on complementor response. It would also be useful to trace these effects
when a new market of an initially complementary product grows to the extent that it
becomes a competitive product. For example, mobile phones have become
competitive with landlines (Shocker, Bayus, and Kim 2004). A related issue is
modeling the evolution of successive generations of products.
2.5.5 Allowing Technological Generations
Norton and Bass (1987) assess the market penetration for successive
generations of a high-technology product. The diffusion equation for the first-
generation product when r
2
is the time of introduction of the second-generation
product is
S
1
(t) = m
1
F
1
(t) – m
1
F
1
(t) F
2
(t – r
2
) (22)
The diffusion equation for the second-generation product is
S
2
(t) = F
2
(t – r
2
)[m
2
+ F
1
(t) m
1
] (23)

42
where S
i
(t) refers to the sales of generation i in time period t, F
i
(t) refers to
the fraction of adoption for each generation, where i =1,2; m
1
refers to the potential
for the first generation, and m
2
refers to the potential for the second generation.
Hence, this simultaneous model captures both adoption and substitution effects. The
authors empirically test the model in the semiconductor industry. Norton and Bass
(1992) extend this model to cover the electronics, pharmaceutical, consumer, and
industrial goods sectors.
Mahajan and Muller (1996) account for the fact that users may skip a
generation and buy a later generation (leapfrogging behavior) in a model that also
captures both adoption and substitution patterns for each successive generation of a
durable technological good. They propose a “now or at maturity” rule for new
product introduction, that is, they determine that the optimal rule for a firm to use in
the decision to introduce a new generation of a technological durable good is either
to introduce it as soon as possible or to delay its introduction until the maturity stage
in the life cycle of the first generation.
Kim, Chang, and Shocker (2000) try to capture not only the substitution
effects between successive generations within a product category, but also
complementary and competitive effects among product categories in a single model.
Hence, the market potential of a generation of a product category is affected not only
by technological substitution from another generation within the category but also by
the sales of other categories. The authors illustrate the model by capturing the growth
dynamics among pagers, analog and digital cellular phones, and the cordless

43
telephone 2 in the wireless telecommunications market in Hong Kong. Their results
indicate that the category of pagers that was introduced earliest seems to have a
positive impact on the cellular phone’s market potential while the cellular phone
appears to have a negative impact on the pager’s market potential. The cordless
telephone 2, however, has a positive impact on both pager and digital cellular phone,
possibly because it serves as a complement.
Danaher, Hardie, and Putsis (2001) capture the role of interdependencies in
marketing-mix variables in the diffusion of successive generations of technology and
show that there are substantial price response interactions across two generations of
technology in the cellular telephone industry in Europe.
2.5.6 Allowing Time-Varying Parameters
The parameters of the Bass model can change over time due to several factors
such as the changing characteristics of the population, products, or economy.
Researchers have looked for ways to incorporate this dynamic specification into the
Bass model (Bass, Krishnan, and Jain 1994; Bretschneider and Mahajan 1980;
Bretschneider and Bozeman 1986; Horsky 1990; Lavaraj and Gore 1990; Mahajan
and Peterson 1978; Sharma and Bhargava 1994; Xie et al. 1997).
Mahajan and Peterson (1978) model the market potential as a function of
time-varying exogenous and endogenous factors such as socioeconomic conditions,
population changes, and government or marketing actions. Easingwood, Mahajan,

44
and Muller (1983) develop a nonuniform influence model that allows the coefficient
of imitation to be time varying. They use the specification
dF(t) / dt = [p + qF(t)
δ
][1 – F(t)] (24)
where δ is called the nonuniform influence factor. If the value of δ equals 1, it
indicates that diffusion takes place with uniform influence, similar to the Bass model.
Values of δ between 0 and 1 cause an acceleration of influence leading to an earlier
and higher peak. This leads to a high initial coefficient of imitation, which declines
with penetration. Values of δ greater than 1 cause delay in influence leading to a
lower and later peak. This indicates that the coefficient of imitation increases with
penetration. Indeed, Easingwood (1987) demonstrates that nine classes of diffusion
shapes can be determined by examining different values of the coefficient of
imitation and the nonuniform influence parameter. For instance, a product with low
values of both parameters has a brief initial period where influence is relatively high,
leading to a steep start to the diffusion process. Subsequently, adoption is constant
and low as influence becomes low.
Sharma and Bhargava (1994) question the assumption that all prior adopters
are equally influential. They propose an extension of the nonuniform influence
model where not only is the influence of previous adopters considered nonuniform,
but also adopters who have adopted in the recent past are considered more influential
than those who did so much earlier.
Several researchers propose alternate functional forms capable of allowing
for dynamic formulation of the parameters. Hjorth (1980) proposes the term “IDB”

45
to denote the distribution that can describe increasing (I), decreasing (D), constant
and bathtub (B) shaped failure rates. Lavaraj and Gore (1990) demonstrate the use of
this distribution to model an adoption function flexible enough to incorporate
increasing, decreasing, constant or bathtub shapes, and nonuniform parameters.
Bretschneider and Mahajan (1980), Bretschneider and Bozeman (1986), and Xie et al.
(1997) demonstrate the use of feedback estimation approaches to estimate dynamic
parameter paths.
The advantage of such dynamic specifications is that they provide a realistic
interpretation of the diffusion process. They not only improve the estimation results
but also help to examine the causes of accelerating or decelerating influences over
time. However, the gain of accuracy and insights from the model comes with a loss
of parsimony.
2.5.7 Allowing Replacement and Multi-Unit Purchases
Though the Bass model covers only first purchases of a durable good,
typically the sales comprise both replacement and multiple purchases. Several papers
in the diffusion literature cover these phenomena (Bayus, Hong, and Labe 1989;
Kamakura and Balasubramanian 1987; Olson and Choi 1985; Steffens 2002).

46
Kamakura and Balasubramanian (1987) incorporate the role of replacement
purchases in the following model:
y(t) = [a + bX(t)] [ αPop(t) Pr
β
(t) – X(t)] + r(t) + e(t) (25)
where y(t) is the sales of a product at year t, Pr(t) is the price index, Pop(t) is
the population of electrified homes, X(t) is the total number of units in use at the
beginning of year t assuming that all dead units are replaced immediately, and r(t) is
the number of units that have died or need replacement at year t. The parameters a
and b denote the coefficients of innovation and imitation, β denotes the impact of
price changes on ultimate penetration, and α refers to the ultimate penetration, if
price was kept at its original level. The researchers demonstrate the incorporation of
replacement purchases into a diffusion setting even when replacement data are not
specifically available.
A related problem is the purchase of multiple units by one household.
Steffens (2002) develops and tests a model for multiple unit adoptions of durable
goods. He models first-unit ownership using a Bass diffusion model with a dynamic
population potential. External influences and earlier adopters of multiple units drive
a proportion ∏
1
of first unit adopters to making multiple purchases giving the model
for multiple unit adopters M(t) as
dM(t) / dt = ( ∏
1
N(t) – M(t)) (a
1
+ b
1
M(t)) (26)
where N(t) refers to the number of cumulative adopters at time t, a
1,
and b
1
are
parameters representing external and word of mouth influences on the first multiple
unit adoption. There are people who adopt more than two units. The upper potential

47
of subsequent multiple unit adoptions is modeled as a fixed proportion ∏
2
of multiple
unit adopters M(t). The model for subsequent multiple unit adoptions Q(t) is
dQ(t) / dt = ( ∏
2
M(t) – Q(t)) (a
2
+ b
2
M(t)) (27)
where a
2
and b
2
are parameters representing external influences and word of
mouth influences on subsequent multiple unit adoptions.
While these models throw light on how to capture replacement demand and
multiple purchases, they do not give insights on what drives these processes. For
instance, Olson and Choi (1985) assume that the life of a product ends due to wear-
out failure only and hence product age and wear-out drive replacement demand.
Other factors such as ability to pay could also determine replacement demand (Bayus
and Gupta, 1992).
2.5.8 Allowing Trial-Repeat Purchases
Markets grow not only through acquiring new trials (first purchases) but also
through repeat purchases by the original buyers. While some researchers look at
trial-repeat purchase behavior in the context of packaged goods industries (Blattberg
and Golanty 1978; Fourt and Woodlock 1960), other researchers examine trial-repeat
purchase in the context of the pharmaceutical goods industries (Hahn et al. 1994;
Lilien, Rao, and Kalish 1981).
Hahn et al. (1994) develop a four-segment trial-repeat purchase model in
which the four segments comprise non-triers, triers, post-trial non-repeaters, and
post-trial repeaters. They find that while word-of-mouth from prior adopters and

48
marketing efforts influence trial, product quality, marketing activity, and market
familiarity influence the repeat rate.
2.5.9 Allowing Variations across Countries
The initial application of the Bass model was limited to the study of diffusion
of new products within the United States. Researchers have since examined the role
of wealth, social system heterogeneity, cosmopolitanism, activity of women,
mobility, mass media availability, culture, and learning, in inducing variations in
diffusion parameters across countries (Dekimpe, Parker and Sarvary, 1998, 2000a
2000b; Ganesh and Kumar 1996; Ganesh, Kumar, and Subramaniam 1997; Gatignon,
Eliashberg, and Robertson 1989; Helsen, Jedidi, and DeSarbo 1993; Kumar and
Krishnan 2002; Kumar, Ganesh, and Echambadi 1998; Putsis et al. 1997; Talukdar,
Sudhir, and Ainslie 2002; Takada and Jain 1991; Van den Bulte and Stremersch
2004).
2.5.10 Evaluation
These improvements have individually addressed various limitations of the
Bass diffusion model. While a single model, which incorporates all these
improvements would enable a rich and comprehensive analysis, this benefit would
likely come at the loss of parsimony. As a result, the contributions remain separate.
In the meantime, managers and analysts can use any one of these models that
addresses the most salient limitation for the product and category they are modeling.
In addition, many of these models assume that the underlying behavior driving the

49
process is knowledge dispersion through communication across consumers. This is,
however, only one of the many processes driving growth.
2.6 IMPROVEMENTS IN ESTIMATION
Since the Bass (1969) model, many articles have attempted to better estimate
the parameters of these models (Lenk and Rao 1990; Schmittlein and Mahajan 1982;
Srinivasan and Mason 1986; Venkatesan, Krishnan, and Kumar 2004; Xie et al.
1997). Schmittlein and Mahajan (1982) propose a maximum likelihood estimation
to estimate the parameters of the Bass model from the expression of the cumulative
fraction of adopters F(t) derived in the Bass model. Though the maximum likelihood
approach eliminates the time-interval bias, Srinivasan and Mason (1986) suggest that
the approach underestimates the standard errors of the parameter estimates as it
focuses only on sampling errors and ignores other forms of errors. They propose an
alternative estimation technique termed the nonlinear least squares approach.
Subsequent improvements belong to one of four approaches: nonlinear least squares,
hierarchical Bayesian methods, adaptive techniques, and genetic algorithms.
2.6.1 Nonlinear Least Squares
Srinivasan and Mason (1986) propose the following nonlinear least squares
approach:
S(i) = m[F(t
i
) – F (t
i – 1
)] + u
i
(28)
where m is the number of eventual adopters, and S(i) is the sales in the
interval (t
i – 1
, t
i
)

50
S(i) =
m[(1 – e
–(p + q)ti
) / (1+ (q / p) e
–(p + q)ti
) – (1 – e
–(p + q)t i – 1
) / (1 + (q / p) e
–(p + q)t i – 1
)] + u
i
where i =1, 2,…T (29)
Jain and Rao (1990) also propose a similar nonlinear approach. These models
can be easily estimated using standard software packages such as SAS. The
nonlinear approach provides the following advantages over the OLS approach. First,
the model is not constrained to be linear in the parameters. Second, the model
overcomes the time-interval bias of the OLS estimation. Third, the model provides
valid estimated standard errors and T-ratios.
However, researchers have determined that the nonlinear technique suffers
from a few limitations. The estimates can be poor and noisy when obtained from data
sets with too few observations. Van den Bulte and Lilien (1997) point at a downward
bias in the estimates of m and p and an upward bias in the estimates of q. Using
longer time series and using data with higher frequency is associated with lower
estimated q / p values (Van den Bulte and Stremersch 2004). These biases may cause
managers to under invest in advertising and external media and overestimate the
impact of the social contagion.
One reason for the biases could be the omission of time-varying parameters.
For instance, as price falls, lower income households may be better able to afford the
new products, increasing the market potential, while the nonlinear least squares
estimation would provide a downward-biased estimate of m. However, Van den

51
Bulte and Lilien (1997) show and Bemmaor and Lee (2002) corroborate that a bias
exists even if the model is correctly specified.
In addition, the model proposed by Srinivasan and Mason (1986) does not
allow for parameter updating and hence does not have good predictive ability for
forecasting sales of very new products. Parameter updating is necessary to improve
the stability of new product market forecasts. The next section examines attempts by
researchers to incorporate Bayesian updating procedures with the nonlinear least
squares estimation method.
2.6.2 Hierarchical Bayesian Methods
To estimate the Bass model reliably and make accurate predictions,
researchers need data beyond the two inflexion points: takeoff and slowdown. Some
researchers propose using expert judgments coupled with industry surveys or
purchase intention questionnaires (Infosino 1986) or information acceleration
techniques (Urban, Weinberg, and Hauser 1996) to develop pre-launch estimates.
2

Other researchers suggest using data for similar products, termed as analogies, for
this purpose (Easingwood 1989). However, to do so, we need to answer two
questions: (1) How can products be classified as similar/dissimilar? (2) What
happens when products are dissimilar? Bayus (1993) proposes a solution to the first
question by developing a product segmentation scheme using demand parameters,
marketing and manufacturing related variables. He demonstrates its application to
generate pre-launch forecasts for the high-definition TV.

52
As a solution to the second question, that is, when data of only dissimilar
products are available, researchers propose the use of hierarchical Bayesian methods
to model new product sales more accurately (Lee, Boatwright, and Kamakura 2003;
Lenk and Rao 1990; Neelamegham and Chintagunta 1999; Talukdar, Sudhir, and
Ainslie 2002). Here, the forecaster can obtain information from different products
that share some common structures, even when no sales data for the focal product is
available. Researchers then develop pre-launch forecasts for the focal product,
updating them when sales information about the focal product does become available
(Putsis and Srinivasan, 2000). The approach helps to produce more stable forecasts
(Lenk and Rao 1990; Neelamegham and Chintagunta 1999; Talukdar, Sudhir, and
Ainslie 2002).
Talukdar, Sudhir, and Ainslie (2002) demonstrate an application of the
hierarchical Bayesian technique to the international diffusion context, pooling
information across multiple products and countries. They use the nonlinear Bass
diffusion model proposed by Srinivasan and Mason (1986), while incorporating two
changes: (1) they model the error term in a multiplicative fashion to reduce the
effects of heteroscedasticity, and (2) they model auto correlated errors to allow for
the possibility of serial correlation.
They model the evolution of a cumulative fraction of adopters over time as
F
pr,c
(t)= {1-exp[-(p
pr,c
+ q
pr,c
) t]}/ {1 + (q
pr,c
/p
pr,c
)exp[-(p
pr,c
+ q
pr,c
) t]} (30)
where the subscripts pr and c refer to the product and country, respectively,
and t refers to the time. The subscripts denote the fact that the authors allow for

53
heterogeneity in the values across both countries and products. They find that their
procedure yields lower mean-squared errors when compared either to models that
estimate the parameters of the Bass model for one product across many countries
(Gatignon, Eliashberg, and Robertson, 1989) or to models that estimate the
parameters across multiple products for one country (Lenk and Rao 1990). However,
the limitation of this model is that the parameters are not allowed to vary over time.
2.6.3 Adaptive Techniques
Other researchers use stochastic techniques that allow parameters to vary
over time to model new product growth. These techniques use feedback filters and
Bayesian techniques to update the parameters over time (Bretschneider and Bozeman
1986; Bretschneider and Mahajan 1980; Xie et al 1997).
Xie et al. (1997) propose the use of the augmented Kalman filter (AKF) to
update parameter estimates as new data become available. The estimation technique
uses continuous and discrete observations (AKF (C – D)) thus:
dn / dt = f
n
[n(t), u(t), β, t] + w
n
(31)
dβ / dt = f
β
[ β, n(t), t] + w
β
(32)
z
k
= n
k
+ v
k
(33)
where n is the cumulative number of adopters, u is the marketing mix
variable vector, β is the unknown parameter vector, w
n
and w
β
are the process noise,
n
k
and z
k
are the actual and observed cumulative number of adopters at time t
k
, and v
k
is the observation noise.

54
Equation (31) is the systems equation that characterizes the diffusion rate at
time t (the evolution of the cumulative adopters) as a function of the current adopters
(n), the marketing mix variables (u), the diffusion parameters β, time t, and random
noise w
n
. Equation (32) specifies the time varying behavior of the parameters while
equation (33) is the measurement equation that specifies the errors in measuring the
number of adopters. At time 0, based on prior information, the best prior estimates of
the parameter distributions are developed. At a specific time, the diffusion model
predicts the sales and parameter values for the next period, using a time updating
process given the current observations. There is also a measurement update as new
information arrives, using the forecast error between the predicted and observed
number of adopters.
The authors show that the augmented Kalman filter estimates the parameters
directly, avoids time interval bias, forecasts more accurately than other techniques
such as the nonlinear least squares and the OLS, and can estimate time-varying
parameters. This technique is however not as easy to use as the nonlinear regression.
2.6.4 Genetic Algorithms
Venkatesan, Krishnan, and Kumar (2004) propose the use of genetic
algorithms to estimate the Bass model. They find that since this technique combines
the advantages of both systematic search and random search, it has a better chance of
reaching the global optimum as compared with sequential-search-based nonlinear
least squares. In simulations, the authors find that unlike the nonlinear least squares

55
method, this technique does not suffer from bias and systematic change in parameter
values as more observations are added. The authors also find that the mean of the
absolute deviations in forecasting for the genetic algorithms is significantly lower
than the augmented Kaman filter estimation technique. However, the technique does
not allow for the fact that the parameters can vary over time.
2.6.5 Evaluation
This body of research indicates that improved estimation techniques,
combined with product classification schemes such as that developed by Bayus
(1993) can lead to increased accuracy in the forecasts of peak sales and the sales
evolution from takeoff to peak during the growth stage. However, the models, which
focus on the general diffusion curve, have paid scant attention to the turning points in
sales, such as slowdown and especially takeoff. For these critical events, researchers
have proposed entirely new models, which will be described below in the section on
modeling the turning points in diffusion.
2.7 ALTERNATE MODELS OF DIFFUSION
Due to the many limitations of the Bass model, especially its reliance on only
a process of communication, several researchers have departed from the framework
and proposed entirely new models. Three of these relate to alternate drivers:
affordability, heterogeneity, and strategy; and two relate to alternate phenomena,
spatial diffusion, and diffusion of entertainment products

56
2.7.1 Affordability
The assumption that underpins the Bass model is that the market consists of a
homogenous population of adopters, all of whom can afford the product equally well.
Their different times of adoption occur because they hear of the product, either from
the firm or from other adopters, at different times. I review models that question this
assumption.
Golder and Tellis (1998) propose an alternate model based on the idea of
affordability. They argue that most consumers know about new products long before
purchasing them. They hold back from purchasing these products due to high prices.
New products are expensive when they first appear on the market, and become
attractive to the mass market only when their price drops sufficiently. Consumers
delay their purchases until prices decline or incomes rise sufficiently for them to
afford the new product. Hence, affordability is a key driver of new product growth.
The authors wish to model product sales as a function of price, income, consumer
sentiment, and market presence, in a parsimonious manner. Hence, they use the
Cobb-Douglas model, which is:
S = P
β1
* I
β 2
* CS
β 3
* MP
β4
* e
є
(34)
where S denotes sales, P denotes price, I denotes income, CS denotes
consumer sentiment, and MP denotes market presence. While this model does not fit
the data as well as the Bass model, the estimates of the coefficients and price
response seem more stable with the addition of observations to the data series and
the model seems to yield better year-ahead forecasts.

57
Horsky (1990) develops a model that incorporates the role of price and
income (affordability) in addition to the word-of-mouth effect in aiding sales growth.
He assumes distributions for both wages and prices, and considers that only a
proportion of the population will purchase the product.
He models sales as:
S(t) = [ θM(t) / (1 + e
–(K + ẃ(t) – k p(t)) / δ (t)
) – Q(t)] [ α + βQ(t)] (35)
where M(t) refers to the number of households in the population, with an
average wage ẃ(t), its dispersion being δ(t); p(t) refers to the average price of the
durable; θ refers to the fraction of the population who will buy the product; and Q(t)
is the number of eligible individuals who have purchased before time t. The term [ α+
βQ(t)] depicts how an eligible individual may become aware of a product due to
word-of-mouth information from those who have already purchased the product. If
the size of the population, the income distribution, and price remain constant, the
equation reduces to the more familiar
S(t) = [N – Q(t)] ([a + βQ(t)] (36)
where N equals the number of people eligible to purchase. In an empirical
application of the performance of the model, the author determines that in categories
where the word-of-mouth effects are weak, the model fits the data better than the
Bass model. The author also derives the policy implication that a price skimming
strategy is appropriate for a monopolist when weak word-of-mouth effects exist and
a price penetration strategy is appropriate when word-of-mouth effects are strong.

58
2.7.2 Evaluation
These models have the advantages of specifically accounting for the role of
price, income and product benefits in the adoption process, hence providing a richer
interpretation. However, this richness comes at the cost of parsimony, ease of
interpretation, and predictive ability, which are the key benefits of the Bass model.
2.7.3 Heterogeneity
Some researchers have looked at the adoption problem as a decision problem
under conditions of belief updating and heterogeneity among consumers (Roberts
and Lattin 2000). The models that fall under this classification have typically been
termed “disaggregate level” diffusion models as they do not assume an aggregate
homogenous population. Individual level models first originated in the economics
literature (Feder and O’Mara 1982; Hiebert 1974; Stoneman 1981). Here eleven
models are reviewed, the first seven predominantly from marketing and the next four
from economics.
Roberts and Urban (1988) assume that individual consumers choose the
brands that provide them with the highest expected risk-adjusted utility and update
their prior beliefs about the brand in a Bayesian fashion with the arrival of new
information. This updating occurs in two ways. (1) Word-of-mouth communications
(positive or negative reviews) may change the estimated mean attribute levels of the
brand. (2) Uncertainty may decline due to the availability of new information. The
authors derive the individual hazard of purchase as a multinomial logit model. The

59
authors apply the model to the pre-launch planning of a new automobile where they
collect measures of mean values, perceived attribute levels, uncertainty, and
purchase probabilities from respondents, and aggregate the probabilities of purchase
over consumers to get the expected market share.
Oren and Schwartz (1988) study the choice between an innovative new
product with uncertain performance and a currently available product with certain
performance. Uncertainty leads risk-averse consumers to delay adoption until they
get more evidence on the performance. Early adopters are those who are less averse
to risk while later adopters are imitators who delay purchase until they get enough
information from the market to overcome their initial uncertainty. The authors derive
an aggregate-level logistic market growth model for market share.
Chatterjee and Eliashberg (1990) develop a model where consumers are risk
averse and adopt a product only if their expectations of its performance exceed a
“risk hurdle” and a “price hurdle.” The consumers update their expectations of
performance based on the information (positive or negative) they receive. Consumers
are hence heterogeneous in the cumulative information they need for adoption. The
authors derive a diffusion curve by aggregating the predicted individual adoption
behavior over the population. The authors show conditions in which their model can
reproduce the Bass (1969) and Fourt and Woodlock (1960) models. The authors
obtain individual level parameters for price, risk, and uncertainty by means of a
survey of respondents.

60
Bemmaor (1994) demonstrates that an aggregate level diffusion model can be
derived from individual level heterogeneity assumptions in the gamma/shifted
Gompertz model (G/SG). Bemmaor and Lee (2002) demonstrate the superiority of
this model to the Bass model in terms of forecasting ability. In this model, individual
level adoption timing is randomly distributed according to a two-parameter shifted
Gompertz distribution whose cumulative distribution function.
The model is as follows:
F(t / ŋ, b) = (1 – e
–bt
) exp(– ŋ e
–bt
), t > 0 (37)
where b is a scale parameter constant across all consumers, and ŋ captures an
individual’s propensity to buy, which varies across consumers according to a gamma
distribution, with a shape parameter α, and a scale parameter β. Here, small values of
α indicate greater heterogeneity.
The authors derive an aggregate level distribution of adoption times given by
F(t) = (1 – e
–bt
) / (1 + β e
–bt
)
α
(38)
Here, if α = 1, b = p + q and β = q / p, equation (38) reduces to the Bass
model, and if α = 0, equation (38) reduces to the exponential model. The authors test
the model by forecasting the sales of twelve new products and find that the G/SG
model provides better forecasts than the Bass model. However, they show that with
the addition of more observations, there are systematic changes in the market
potential and imitation coefficients. Hence, the more complex G/SG model shows
greater parameter instability than the Bass model.

61
Song and Chintagunta (2003) develop a model in which they account for both
heterogeneity and forward-looking behavior by consumers in the adoption of new
high-tech durables products. They use aggregate sales data, rather than intent
measures obtained from surveys, to estimate the model. In the model, consumers
have expectations of the future states of prices and quality levels, both of which
change over time, leading to a probability distribution on the transition of future
states of these variables conditional on current states. A consumer can choose either
to buy or not to buy a product in each period, selecting the alternative that maximizes
the discounted sum of expected utility. The authors aggregate these individual level
adoption decisions to obtain an aggregate diffusion curve, and use the more easily
available aggregate level data to estimate the individual level decision parameters.
Sinha and Chandrasekaran (1992) demonstrate the application of a split
hazard model to analyze the probability of adoption and adoption timing of an
individual firm. By splitting the population into eventual adopters and non-adopters,
and modeling both the probability and the timing of adoption as a function of
individual level variables, they capture heterogeneity at the individual level. They
test their model in the context of the adoption of automated teller machines in a
sample of individual banking firms.
Chandrashekaran and Sinha (1995) account for variation in the volume of
adoption as well as the timing of adoption by applying a split-population Tobit
duration model in examining the adoption of personal computers by a sample of
firms.

62
Karshenas and Stoneman (1993) and Stoneman (2002) describe what they
term “rank,” “stock” or “order” effects. In models considering “rank” effects, actors
adopt as soon as the utility of the innovation exceeds some critical level or threshold.
If the utility increases systematically over time and the thresholds follow some bell-
shaped distribution, then the cumulative number of adopters, that is, the diffusion
curve, will be S-shaped. In the consumer marketing literature, income distribution
within a population can determine reservation prices, and hence pose one such
threshold (Van den Bulte and Stremersch 2004). In models considering “stock”
effects, the assumption is that the marginal benefit from adoption decreases with the
number of prior adopters (Karshenas and Stoneman 1993; Stoneman 2002). Over
time, the cost of acquisition falls leading to an increase in the number of adopters. As
more firms adopt the new technology, costs of production fall, increasing output. As
a result, the industry price falls and adoption is unprofitable beyond a certain point.
In the economics literature, such models typically follow a game-theoretic approach
(Reinganum 1981). In models incorporating the “order” effects, the assumption is
that there are first-mover advantages in using a new technology. The returns to the
firm from the new technology depend on its position, with higher-order firms getting
more returns than lower-order firms do. Each firm, considering how moving down
the order affects its return, generates the diffusion path. For any given costs of
acquisition, only some firms will find it profitable to adopt at a given point in the
order, and only these numbers adopt. As costs of acquisition fall, more firms adopt.
Fudenberg and Tirole (1985) develop a game theoretic model where they argue that

63
earlier adopters get the highest return and hence there will be a race to be an early
adopter, and the decisions of higher-order firms can then influence the decision of
lower-order firms.
Karshenas and Stoneman (1993) determine the effect of rank, stock, order,
and epidemic effects on the diffusion of CNC machine tools in the U.K. engineering
industry. They estimate a hazard model of the form
h(t / X, β) = h
0
(t)* exp(X´ β) (39)
where X incorporates acquisition costs, cumulative number of adopters at
time t (stock), firm characteristics (rank), expected change in the number of
cumulative adopters in the time interval (t, t + 1) (order), price, and expected change
in price, and the baseline hazard denotes the epidemic effects. They find that rank
and endogenous learning effects play an important role in the diffusion process, but
find little support for the stock and order effects prescribed by game theoretic models,
lending support for the interest paid by the marketing literature to the communication
process in adoption.
2.7.4 Evaluation
Following the Bass model, the vast tradition of diffusion research in
marketing has focused on communication among potential adopters and prior
adopters as the main driver of diffusion. In contrast, the models discussed in this
section indicate alternate reasons as to why individual consumers adopt new products
and change their judgments over time.

64
However, these models, which focus extensively on individual level adoption
decisions, have some limitations. First, most individual models lack the parsimony
and ease of understanding that are the strengths of aggregate level models. Second,
when individual level models use aggregate level data, it is difficult to identify the
precise drivers of the adoption process.
2.7.5 Strategy
Strategy implies the explicit modeling of a firm or a central decision maker’s
choices such as market entry, marketing mix efforts and location. In this section,
three such models are considered. While some extensions of the Bass model do
consider the marketing mix, as seen in a previous section (Bass, Krishnan, and Jain
1994), such extensions are subservient to the model structure and lead to potentially
understated effects for marketing variables.
DeKimpe, Parker, and Sarvary (2000a) consider two stages in the
technological adoption of digital communication switches: (1) the time between the
first availability of an innovation in the world and its introduction in a country (the
implementation stage), and (2) the time between the introduction of an innovation
into a country and its full adoption (the confirmation stage). They examine the
impact of economic, socio-demographic factors, installed base, and the international
experience of the innovation on the transition times from one stage to another, using
the coupled hazard approach. The authors point out that for telecommunications
innovations, the local government or a central communications unit often acts as a

65
key decision maker in setting standards and regulations. This may affect the
product’s diffusion path. For instance, in some small countries, the central decision-
making unit may decide to replace the old technology fully with the new technology,
and hence these countries may reach full penetration immediately on adoption
whereas other countries may exhibit the more gradual S-shaped diffusion path.
Van den Bulte and Lilien (2001) reexamine the medical innovation study
(Coleman, Katz, and Menzel 1966). This study examines the role of social networks
in the diffusion of the broad-spectrum antibiotic tetracycline among 125 physicians
in the United States in the 1950s. Van den Bulte and Lilien (2001) use a discrete time
hazard modeling approach to examine the role of both social influence and
marketing efforts by drug companies in influencing the hazard of adoption by a
physician. They find that marketing efforts, rather than contagion seem to influence
the diffusion process, and indicate that the medical innovation study might have
confounded social contagion with marketing effects.
Bronnenberg and Mela (2004) study the spatial and temporal introduction of
two brands in the frozen pizza category in the United States. The process begins with
manufacturers deciding which markets to enter. Subsequently, in the markets that
they enter, manufacturers offer the product along with incentives to retail chains. The
retail chain decides whether to approve the brand for distribution on its entire trade
area. Individual stores from this chain can carry the brand once it becomes locally
available and is approved for adoption. The authors model the manufacturer’s timing
of local market entry and the retailer’s timing of adoption of the brand, conditional

66
on entry, using a discrete time hazard modeling approach. They determine that
manufacturers sequentially enter markets based on the spatial proximity to markets
already entered, and based on whether the chains in these markets have previously
adopted the product elsewhere. The retail chains adopt the product, based on whether
competing chains have adopted the product, and the manufacturers push into the
trade area of the retailer. The study highlights the importance of taking into account
the marketing actions (launch strategy) of manufacturers, without which the effect of
local competitive contagion may be overstated. The study also points out the
importance of understanding how products diffuse over space, which is elaborated
upon in the next section.
2.7.6 Evaluation
Researchers who consider strategic factors, such as marketing variables, or
entry decisions, find that these factors often dominate the role of communication in
driving diffusion (Bronnenberg and Mela 2004; Sultan, Farley, and Lehmann 1990;
Van den Bulte and Lilien 2001). This finding highlights the need to consider such
variables in order to avoid spurious results.

67
2.8 MODELING DIFFUSION ACROSS SPACE
2.8.1 Theories of Spatial Diffusion
Spatial diffusion models address the way products diffuse over space rather
than over time as the prior models do. Though not considered explicitly in the field
of marketing, spatial diffusion has had a long tradition of research in the fields of
geography and agricultural history, originating in the seminal work of Hagerstrand
(1953).
3
There are various types of spatial diffusion (Morrill, Gaile, and Thrall 1988).
Contagious diffusion occurs when the distance or adjacency is the controlling factor,
for instance, the spread of infectious diseases. Expansion diffusion describes a
process similar to that of a wildfire, when there is a source and the diffusion occurs
outward from the source. Hierarchical diffusion occurs when diffusion progresses
through an ordered series of classes, such as a phenomenon first being observed in
the largest city, then jumping to the next largest, and so on. Relocation diffusion
occurs when the number of agents with the diffusion characteristics does not change.
The agents merely change spatial location or as the trait passes on to additional
agents, it is lost in the original agents. Here I consider some aspects of the seminal
work by Hagerstrand (1953) as well as four models in marketing that examine
explicitly the notion of diffusion across space (Bronnenberg and Mela 2004; Garber
et al. 2004; Mahajan and Peterson 1979; Redmond 1994).
Hagerstrand (1953) conducts a detailed mapping of the geographic spread of
agricultural indicators such as state-subsidized pastures and of general indicators

68
such as postal checking services, automobiles, and telephones. He observes that a
synoptic growth curve could conceal a large number of individual events that occur
simultaneously in different parts of the observed area. Typically, diffusion seems to
have the following spatial regularities: at first, there is a local concentration of initial
acceptance followed by a radial dissemination outward while the original core of
acceptance continues to become denser. Finally, growth ceases, as there is saturation.
For agricultural indicators, the initial acceptance groups are clear and radial
dissemination proceeds along clear-cut lines. For instance, the acceptance of state-
subsidized pastures spreads from the west to the eastern part of the area. In contrast,
for general indicators, the initial acceptance is more dispersed and the subsequent
dissemination less orderly. Much of Hagerstrand’s work is relevant to marketing. For
instance, he introduces the notion of a “mean information field” where the frequency
of contacts in a social network is assumed to diminish with distance. He also argues
that potential adopters may vary in their “resistance” to the innovation, leading to a
longer period of incipient growth and a greater degree of spatial concentration that is
evident in the diffusion of some products.
Mahajan and Peterson (1979) introduce the notion of the “neighborhood
effect” in technological substitution models in the marketing literature, that is, the
further a region is from the “innovative region,” the later substitution will occur.
Redmond (1994) argues that diffusion models typically assume spatial
homogeneity by examining the process at a national level, and this ignores variations
within a country. In an application of the Bass model to the diffusion of two

69
consumer durables across nine regions within the United States, he determines that
differing local conditions and demographics across regions lead to differing diffusion
rates within a country.
Garber et al. (2004) argue that it is possible to predict the success of new
products by looking at spatial patterns of diffusion by means of complex systems
analysis. In such an analysis, the market is a matrix in which the discrete cells
represent adoption by individuals. Each cell interacts with the other cells, the
interactions not being limited to strictly neighboring cells (in what is termed a
“small-world” framework). The value “0” represents non-adopters and “1”
represents adopters; “p” represents the probability that an individual will be affected
by external factors, and “q” the probability that an individual is affected by an
interaction with a single other individual who has adopted the product.
The probability that an individual adopts at time t given that the individual
has not yet adopted is:
Prob (t) = 1 – (1 – p) (1 – q)
v(t) + r(t)
(40)
where v(t) represents the number of neighboring previous adopters with
whom the individual maintains contact and r(t) is the number of previous adopters
who are weak-tie contacts. The authors argue that a spatial analysis of diffusion data
can help in the early prediction of new product success. They state that for a well-
received product, word-of-mouth and imitation will feed the flow of internal
influence, leading to the formation of clusters. However, if the product is a failure,
then internal effects activity will be minimal, diffusion will be mainly due to external

70
effects, and adopters will hence be randomly distributed. Thus, the distribution in the
case of a failure would be closer to a uniform distribution. Therefore, the authors
argue that it is possible to predict the success of a new product within a few periods
of its introduction by comparing the spatial distribution of the product with respect to
a uniform distribution using a measure of divergence known as cross-entropy. They
expect successful products to have a declining cross-entropy measure while failures
will have a consistently low cross-entropy measure
2.8.2 Evaluation
The use of techniques such as complex systems analysis helps to provide a
micro view of the patterns of interaction among individuals and an understanding of
how this influences the diffusion of new products. However, these models seem to
follow the Bass model tradition of viewing new product diffusion entirely through a
process of “communication”, ignoring alternate explanations such as those described
in previous sections.
2.9 MODELING DIFFUSION OF ENTERTAINMENT PRODUCTS
The sales of entertainment and information products, especially release of
movies to theaters, typically follow a pattern of exponential decay rather than the
bell-shaped pattern of durable goods sales. A vast stream of marketing research has
focused on forecasting sales in the movie industry and sales of other entertainment
products. This section reviews some of the important models in this area.

71
Eliashberg and Sawhney (1994) develop a model to predict individual
differences in movie enjoyment. Sawhney and Eliashberg (1996) model the total
time to adopt (see) a movie by an individual as the sum of the total time to decide,
which is related to information intensity and the total time to act, which is in turn
related to distribution intensity. Both these processes are assumed to be
exponentially distributed with the stationary parameters λ and γ. The authors find
that their model can determine three classes of adoption patterns that can represent
all box-office patterns. The authors hence develop a simple model, based on just two
parameters, which needs less data than the Bass model to forecast effectively.
However, when the authors extend their analysis in an attempt to model with little or
no revenue data, they find that while their model does well in predicting the ultimate
cumulative box-office potential, it does not help capture the shape parameters λ and γ
and hence provides little insight regarding how the box-office performance is spread
over time.
Subsequent researchers of entertainment products show how to develop better
pre-launch forecasts. For instance, Eliashberg et al. (2000) assume that initially all
consumers are in an “undecided” state and are exposed to both media advertising and
word-of-mouth (positive or negative). Depending on the impact of advertising and
word-of-mouth effects, there is a behavioral transition from the “undecided” to the
“considerer” (one who eventually sees the movie) or “rejecter.” The considerer
becomes either a positive or a negative spreader. The authors model the state
transitions via an interactive Markov chain model. The parameters of the model:

72
word-of-mouth frequency, duration of spread, consideration duration, and
distribution delay, are determined via prerelease experiments. This model is intuitive
and appealing as it reflects the actual behavioral states and transitions of a movie
consumer.
Elberse and Eliashberg (2003) examine movie forecasting in a cross-cultural
context and determine how the performance of a movie in a domestic market
influences its performance in a subsequent international launch. Researchers have
also examined the impact of advertising (Zufryden 1996), movie critics (Eliashberg
and Shugan 1997), and movie Web site promotion (Zufryden 2000) in forecasting
box-office performance. Shugan (2000) and Shugan and Swait (n.d.) demonstrate
how researchers can utilize consumer intent-to-see measures in developing
prerelease forecasts.
A number of other models examine various aspects related to the sales
evolution of entertainment products. For instance, Moe and Fader (2002)
demonstrate the use of the hierarchical Bayesian technique to develop pre-launch
forecasts of new product sales of entertainment goods such as music CDs, based on
patterns of advance purchase orders. Lee, Boatwright, and Kamakura (2003)
elaborate a hierarchical Bayesian model to develop pre-launch forecasts of recorded
music.

73
2.9.1 Evaluation
These models show in general that alternate models help capture the growth
of entertainment products better than the Bass model in terms of insights, fit, and
pre-launch predictions of sales. The question is whether these different models can
be generalized beyond the specific product modeled to all entertainment products.
They are unlikely to be suitable to non-entertainment products. In contrast, the
strength of the Bass model is that it can be generalized beyond the durable goods
setting.
2.10 MODELING TURNING POINTS IN DIFFUSION
This section examines the definition, measurement, drivers, and models of
the specific turning points of the general diffusion, that is, takeoff and slowdown.
2.10.1 Theory of Takeoff
A key characteristic of new products is that not all consumers accept them
instantaneously at the time of introduction. The Bass model assumes the presence of
a certain number of consumers (p * m) before “takeoff” (Golder and Tellis 1997;
Mahajan, Muller, and Bass 1990, p. 21). Researchers using the Bass model also
frequently use data from the point of takeoff or slightly before (Golder and Tellis
1997). However, most new products experience a long period when sales are low. At
some point, a sudden spurt in sales is followed by a period of rapid growth. When
viewed graphically this trend appears as a sharp bend in the curve or a “takeoff.”
Figure 2 compares the takeoff patterns of a white good (microwave oven) across

74
various West European countries. The sharp bend in the curves of the graphs signals
takeoff.
Figure 2: Takeoff of Microwave Oven Sales in Europe

0
500
1000
1500
2000
2500
1970 1972 1974 1976 1978 1980 1982 1984 1986 1988
Year
S a le s (In '000 s)
Finland
France
Germany

Prior to 1997, academic literature and the trade press have often referred to
the takeoff of new products, without any formal definition or measure of the
phenomenon. However, a few articles discuss the phenomenon from select angles.
For instance, Gort and Klepper (1982) define the diffusion of product
innovations as the spread in the number of producers engaged in manufacturing a
new product. They define the takeoff as the second stage in this evolution, involving
a sharp increase in the number of producers. However, though they are able to
demonstrate these distinct stages of market entry, they do not relate it to the adoption
of the new products by consumers. Thus, the takeoff in number of producers may not
coincide with takeoff in sales.
Kohli, Lehmann, and Pae (1999) define a concept termed “incubation time”
as the time between the completion of product development and the beginning of
substantial sales of the product. They find that the length of the incubation time

75
affects parameters of the Bass diffusion model. The beginning of “substantial sales”
of the product can be analogous to takeoff. However, their definition of “substantial”
and the measurement of when substantial sales begin, and hence of incubation time
is imprecise.
Golder and Tellis (1997) define takeoff in sales of a new product as the point
of transition from the introduction stage to the growth stage of the product life cycle.
They also provide the first formal and precise measure of takeoff. This measure is
described later in the context of other measures for takeoff.
Why is takeoff important? A sudden and sharp increase in sales requires
enormous resources in terms of manufacturing, inventory, distribution, and support.
Hence, knowing when it occurs and what causes it is critical for managers in
handling the sales and success of a new product. Most important, takeoff represents a
difficult-to-predict turning point in a new product’s life. It might well be a sign to the
managers that the product has become desirable to the mass market. It might also be
an early sign of the future success of the new product.
2.10.2 Measuring Takeoff
The literature describes many different measures of takeoff.
Golder and Tellis (1997) provide a simple measure for this phenomenon that
they find to work quite well in an extensive study of new consumer durables in the
United States. The authors find that when the base level of sales is small, a relatively
large increase in sales can occur without signaling takeoff. Alternatively, when the

76
base sales are large, a relatively small increase in sales can signal takeoff. Hence,
they develop a threshold of takeoff, which is a plot of percentage sales growth
relative to a base level of sales, common across all categories. The authors measure
takeoff as the first year in which an individual category’s growth rate relative to the
base sales crosses this threshold. They find that this heuristic measure of takeoff
successfully fits a visual inspection for 90 percent of the categories in their sample.
Golder and Tellis (1997) also compare this rule to measure takeoff with two
alternatives: a logistic curve rule and a maximum growth rule. The logistic curve rule
involves finding the first turning point of a logistic curve fitted to each sales series.
This involves determining the maximum of the second derivative of the logistic
curve since this captures the largest increase in sales growth. The maximum growth
rule uses the largest sales increase within three years of takeoff as determined by the
logistic curve rule. However, the authors identify problems with the latter two rules.
Researchers can apply the logistic curve rule only in hindsight, as it requires sales
beyond takeoff. The logistic curve rule is also a continuous rule to measure what is
essentially a discontinuity. The maximum growth rule has three limitations. First, the
largest sales growth sometimes occurs after takeoff has already occurred and sales
are clearly in the growth stage. Second, large percentage increases can occur even
with small base level sales. Third, the researcher can apply this rule only in hindsight.
Agarwal and Bayus (2004, 2002) propose a fourth measure of takeoff. They
distinguish between any two consecutive intervals by examining the data on annual
percentage change in sales (for the sales takeoff) and annual net entry rates (for firm

77
takeoff) for each product. To determine the takeoff year for a product, first they
partition the appropriate series into three categories. Here, the first and third
categories contain the years where the percentage change in sales or net entry rate
reflect the pre- and post-takeoff periods, respectively. They classify the in-between
years based on mean values. This is a method similar to that used by Gort and
Klepper (1982) to identify firm takeoff.
Stremersch and Tellis (2004) and Tellis, Stremersch, and Yin (2003) use a
fifth measure of takeoff to suit an international sample of countries. It is similar in
spirit to the threshold rule proposed by Golder and Tellis (1997). The authors define
the threshold as a standard plot of growth in sales for various levels of market
penetration to provide for a more standard comparison across several countries.
Takeoff is the first year in which an individual category’s growth rate relative to the
base sales crosses this threshold
Garber et al (2004) and Goldenberg, Libai, and Muller (2001a) use a measure
that takeoff occurs when 16 percent of the population adopts. This is similar to
Rogers’ (1995) argument that the S-shaped curve of diffusion “takes off” at around
10–20 percent adoption.
So far, no study has compared these six different measures of takeoff to
assess their simplicity, domain of relevance, validity, and predictive accuracy.

78
2.10.3 Explaining Takeoff
The literature on takeoff itself is in the introductory and pre-takeoff stage of
its life cycle. This search revealed only a few studies on this topic, three of which
deal specifically with the determinants of takeoff. These three studies examine three
different drivers of takeoff: affordability, infrastructure factors, and heterogeneity,
and they reach substantially different conclusions.
Golder and Tellis (1997) propose that price declines are a principal driver of
takeoff. At some point in the price decline, the new product crosses a critical point of
affordability, leading to a takeoff. They find that economic characteristics such as
GNP, consumer sentiment, or number of households do not affect the probability of
takeoff, arguing that this may be because when the primary condition for takeoff
(consumer affordability) is satisfied, even a weak economy cannot forestall takeoff.
Agarwal and Bayus (2002) argue that an increase in firm entry leads to
increased consumer awareness due to an increase in the number and quality of
product offerings, marketing infrastructural facilities, and promotions. The authors
examine both product takeoff and firm takeoff and find that both firm entry and price
declines are related to product takeoff times. Moreover, they find that firm entry
dominates price declines in explaining takeoff times.
Tellis, Stremersch, and Yin (2003) examine the relative impact of country,
product, and time characteristics on the takeoff of new products across categories
and countries. They determine that a “venturesome” culture seems to affect takeoff,

79
and similar to the results in Golder and Tellis (1997), they find that economic wealth
and economic progressiveness do not seem to affect takeoff.
2.10.4 Modeling Takeoff
Researchers typically use a hazard function to model takeoff. Both Agarwal
and Bayus (2002) and Golder and Tellis (1997) model the rate at which takeoff
occurs as a function of a baseline hazard function that captures the effect of time
since introduction, and independent variables. Hence, they model time to takeoff
using the following proportional hazards specification:
h
i
(t) = h
0
(t)e(z
it
β) (41)
where h
0
(t) is an unspecified baseline hazard, z
it
is the vector of independent
variables for the ith category and β is the vector of unknown parameters.
The advantage of using this specific formulation is that it does not constrain
the baseline hazard to be of any specific functional form, such as monotonically
increasing or decreasing. Cox’s partial likelihood estimator provides a method for
estimating β without requiring estimation of the baseline hazard. Positive beta
coefficients increase the hazard of takeoff, negative beta coefficients decrease the
hazard of takeoff, and the effect of an increase by one unit of any independent
variable on the hazard of takeoff is captured by the magnitude 100 * (e
β
– 1). In a
similar vein, Tellis, Stremersch, and Yin (2003) use the parametric log-logistic
hazard approach to model time to takeoff.

80
2.10.5 Evaluation
The literature on takeoff is small but critical to managers and researchers for
several reasons. First, it identifies an important phenomenon and shows that it can be
scientifically modeled. Second, the models are somewhat successful in identifying
explanatory variables and predicting the phenomenon. Third, managers have already
applied the models in practice and for formulating strategy (e.g., Foster, Golder, and
Tellis 2004).
At the same time, the literature has some important limitations. First, it
considers only successful innovations. As such, its implications are good for
predicting when a takeoff might occur. It cannot tell whether a takeoff might occur or
predict the success or failure of a new product. Second, the empirical applications of
takeoff have involved only a limited geographic domain (only the United States and
Western Europe). Third, models of takeoff focus only on the growth of the product
until takeoff, which on average occurs at 2 percent penetration of the market. The
models give no insights about the sales pattern after takeoff. So far, no published
study has tried to integrate the modeling of these two phenomena.
2.10.6 Theories of Slowdown
The most common conception of a product life cycle portrays the sales
history of a product as following a smooth bell-shaped curve, with just four stages—
introduction, growth, maturity, and decline. Some researchers have noted, however,
that the classic bell shape might not be quite so smooth. Cox (1967) documented

81
evidence for a scalloped product life cycle. Wasson (1978) argued that there is a
period of slowdown in sales, or “competitive turbulence,” which follows the period
of rapid growth. In his review of the literature on product life cycles, Day (1981)
remarked that while interesting, this pattern had virtually no empirical evidence to
support it. Nearly twenty years later, three papers (Stremersch and Tellis 2004;
Golder and Tellis 2004; Goldenberg, Libai, and Muller 2002) find empirical
evidence of a sudden decline in sales following the growth stage.
Golder and Tellis (2004) define slowdown to be the point of transition from
the growth stage to the maturity stage of the product life cycle. Hence, early maturity
begins with the year sales slow down and continues until sales grow to the previous
local peak. This is similar in spirit to the concept of the “Saddle” proposed by
Goldenberg, Libai, and Muller (2002).
Figure 3 shows the typical pattern of a slowdown in sales in the case of
dishwashers in Europe. After takeoff, the sales of the products reach an initial peak,
followed by a sharp and deep decline, and seem to take some time before regaining
the initial peak. Figure 4 shows similar patterns for the newer electronic goods
category of computers.

82

Figure 3: Slowdown of Dishwasher Sales in Europe
0
20
40
60
80
100
120
140
160
180
200
1958 1961 1964 1967 1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003
Year
Sales (In '000s)
Finland Belgium

Figure 4: Slowdown of Computer Sales in Europe
0
50
100
150
200
250
300
350
1981 1984 1987 1990 1993 1996 1999 2002
Year
Sales (In '000s)
Austria Belgium

83
2.10.7 Measuring Slowdown
Early maturity begins with the year sales slow down and continues until sales
grow to the previous local peak (Golder and Tellis 2004).
Late maturity begins with the first year sales being higher than the local peak
and continues until a product’s sales begin to fall steadily during the decline stage
(Golder and Tellis 2004).
Goldenberg, Libai, and Muller (2002) define and measure the Saddle as a
trough following an initial peak in sales, reaching a depth of at least 20 percent of the
peak, lasting at least two years, followed by sales that ultimately exceed the initial
peak. Golder and Tellis (2004), and Stremersch and Tellis (2004) operationalize
slowdown, or the end of growth, as the first year, of two consecutive years after
takeoff, in which sales are lower than the highest previous sales.
2.10.8 Explaining Slowdown
What are the reasons for the sudden decline in sales following slowdown?
Recent literature in marketing proposes three key processes driving slowdown of
new products: dual-market phenomenon, informational cascades, and affordability.
Dual Market Phenomenon
Goldenberg, Libai, and Muller (2002) argue that the initial product offered to
consumers is different from that offered in a later phase, and the consumers in two
stages of the product life cycle differ in a meaningful way. Hence, the early market

84
and the late market adopt in different ways, and the social contagion process is
broken at the point of transition from the early market to the late market. Both
demand-side and supply-side factors seem to be at work here.
This theory builds on work by Moore (1991), who argues that a chasm exists
between the early adopters and early majority. He posits that in the case of
technological products, early adopters are looking to buy a change agent and expect
to get a jump on competition. They expect some radical discontinuity between the
old and new ways and are prepared to champion the cause. The early majority on the
other hand, wants to buy a product improvement for existing operations. They are
looking to minimize discontinuity with old ways and want technology that enhances,
not overthrows established ways of doing business. This lack of communication
between the two segments can create a difference in the adoption rates of both
segments, leading to the slowdown in sales.
Informational Cascades
Golder and Tellis (2004) posit an alternative explanation based on the theory
of informational cascades (Bikhchandani, Hirshleifer, and Welch 1992). Cascades
occur when many consumers base their choice on the behavior of a few other
consumers rather than on their own private assessments of the utility of alternatives.
Some consumers first decide to buy a new product on its merits. A few other
consumers note their behavior and follow suit, causing an increase in sales. The
increase triggers still more consumers to buy the new products, leading to much

85
bigger increases. The process cascades into the takeoff and rapid growth of the new
product. Due to the cascade, during the growth stage, sales increase far more than
they would have based on consumers’ private assessment of the utility of the new
product to them.
Such cascades are fragile. Some small doubt or turbulence in the market can
cause a slowdown in sales and hence trigger a negative cascade. Such behavior can
account for the common drop in sales of a new product after slowdown, and the
pickup of sales after the turbulence.
Affordability
Golder and Tellis (2004) posit a third explanation for slowdown based on the
notion of affordability. An economic contraction can trigger a corresponding decline
in the disposable income of consumers. As a result, consumers cut down on
discretionary expenditures, such as purchases of new products, which have typically
not yet become essential (Deleersnyder et al. 2004). If the economic decline is
substantial, it can lead to the slowdown and even subsequent drop in sales that I
observe at the end of the growth stage of a new product life cycle.
2.10.9 Modeling Slowdown
The two studies of slowdown offer conflicting explanations of what
determines slowdown and they use different models to test their hypotheses.
Goldenberg, Libai, and Muller (2002) use cellular automata to describe the
process by which internal communication breaks down between the early adopters

86
and early majority. As mentioned earlier in the review, cellular automata models are
simulations that reveal aggregate patterns based on local interactions between cells.
This technique has three benefits. First, researchers often find it difficult to obtain
data at the individual level. Second, aggregate level models sometimes do not
provide insight about individual level phenomena. Third, there is the persistent
difficulty of determining how aggregate phenomena evolve from changes in
individual actions. The use of cellular automata helps to circumvent this problem.
These models can help validate the assumptions made in aggregate level models
(Goldenberg, Libai, and Muller 2001a, 2001b). However, the cellular automata
models consider adoptions only in a binary state (0 or 1). There do not seem to be
ways of obtaining socioeconomic characteristics of these adopters or any such
information that aids the modeling of diffusion processes.
Golder and Tellis (2004) use hazard models to determine the impact of
explanatory variables such as price declines, income declines, and market
penetration on the time to slowdown. They find that every 1 percent decrease in total
GNP is associated with a 17 percent increase in the probability of slowdown,
indicating that economic factors affect slowdown in a substantial manner (though
Golder and Tellis (1997) find no effect of economics on takeoff). In addition, they
find that categories with large sales increases at takeoff will also have large sales
declines at slowdown, giving some support to the notion of informational cascades.
They find that every 1 percent higher price is associated with a 4.7 percent increase
in the probability of slowdown, indicating that price declines can extend the duration

87
of the growth stage. They also find that every 1 percent increase in penetration is
associated with a 3.6 percent increase in the probability of slowdown, indicating that
the probability of slowdown increases with a depleting pool of adopters.
Support for economic variables leading to a slowdown in sales is also found
to some extent in Deleersnyder et al. (2004). These authors find that consumer
durables are highly sensitive to business-cycle fluctuations. In addition, they find that
every percentage decrease in the cyclical component of GNP translates to a drop in
the cyclical component of durable sales by, on average, more than 2 percent.
2.10.10 Evaluation
Research on the slowdown in new product growth is new. There is still no
consensus on whether and to what extent the phenomenon is pervasive, how to
define and model it, and what factors drive it. If the pattern proves to be regular, it
represents a challenge for researchers to model it and integrate it within any of the
prior models. New research in this area can also make a substantive contribution by
developing one integrated model to investigate the impact of the different drivers of
slowdown.

88
2.11 CONCLUSION
This comprehensive review of the marketing literature on the diffusion of
new products provides the following benefits to the reader. First, the review
delineates key phenomena associated with the diffusion of innovations such as the
shape, turning points, and stages of diffusion. Second, the review identifies the
variety of drivers of diffusion and explains how they have been either modeled or
ignored in various research traditions. Third, the review provides a critical evaluation
of the models. This evaluation gives readers a simple synopsis of the models with
their strengths and weaknesses. Fourth, the review identifies a large number of
regularities or potential generalizations in the areas of shape of the diffusion curve,
the turning points, and the early stages of the new product’s life cycle.
While extensive, the review is still incomplete in one important respect. It
does not cover the literature in many related fields such as medicine, agriculture,
sociology, anthropology, and technology management. It also covers only very
limited aspects of the economics and geography literatures. While the models,
drivers, and potential generalizations identified in marketing can be extended to
these other fields, this is a topic for further research.

89
End Notes
1. New product growth can follow alternate growth patterns. A shape of growth
that has not been captured by the logistic or the exponential growth curves is
seen when the period of rapidly increasing sales is shorter than the period in
which sales converge to a certain saturation level. Frances (1994), in an
illustration of the Dutch new car market, and Chow (1967), in the rental of
electronic computers in the United States, capture these growth processes
using a Gompertz curve. Bemmaor (1994) develops a gamma/shifted
Gompertz model, which will be discussed later in this chapter.
2. Urban, Weinberg, and Hauser (1996) suggest a technique known as
“information acceleration” to forecast consumer reactions to radically new
products such as electric vehicles. Here, researchers utilize a multimedia
computer to create a virtual buying environment and accelerate information
to a consumer so that he/she can react as if they were in the future. The
authors develop market forecasts using combinations of stated intent
measures, conjoint analysis, and diffusion models. See Urban et al. (1997) for
further applications of this technique.
3. See Morrill, Gaile, and Thrall (1988) for a review of more recent approaches
to model spatial diffusion, in the geography literature tradition, examining
both spatial diffusion and the incorporation of time and space in diffusion.

90

CHAPTER 3: GLOBAL TAKEOFF OF NEW PRODUCTS:
CULTURE’S CONSEQUENCES OR WEALTH OF NATIONS?
2
3.1 INTRODUCTION
Markets are seeing faster introductions of new products and more intense
global competition than ever before. In this environment, firms need to know how
new products diffuse across countries, which markets are most innovative, and in
which markets they should first introduce new products.
Recently, studies have introduced and validated a new metric to measure how
quickly a market adopts a new product: the takeoff of new products (e.g., Agarwal
and Bayus 2002; Golder and Tellis 1997; Tellis, Stremersch and Yin 2003). Takeoff
marks the turning point between introduction and growth stages of the product life
cycle. When used consistently across countries, this metric provides a valid means
by which to compare and analyze the innovativeness of countries. However, the
existing literature on takeoff suffers from the following limitations.
First, prior studies analyze takeoff of new products primarily in the U.S. and
Western Europe. Hence, they exclude some of the largest economies of the world
(Japan, China, and India) and many of the fastest growing economies of the world
(China, India, Korea, Brazil, and Venezuela). In fact, this limited focus on

2
The study is supported by grants from the Marketing Science Institute, the Center
for Global Innovation at USC and from the Center for International Business
Research Education and Research at USC.

91
industrialized countries is seen as symptomatic of much of the prior research on
product diffusion with several calls for broader sampling for new insights into the
phenomenon (Dekimpe, Parker and Sarvary 2000)
Second, takeoff has been portrayed to be a cultural phenomenon (Tellis,
Stremersch and Yin 2003). Yet, other studies cite economics as the primary driver of
new product takeoff or diffusion (Dekimpe, Parker and Sarvary 2000, Stremersch
and Tellis 2004). Talukdar, Sudhir and Ainslie (2002) find that wealth affects the
estimated market potential of new products across both developed and developing
countries. Golder and Tellis (1997) find that price is a key driver of takeoff, implying
that higher disposable income would promote faster takeoffs.
Third, researchers have debated about which countries have the most
innovative consumer markets and are thus the best launch pads for a new product.
The international strategy literature has long held that the US is the pre-eminent
origin for new products and fads (Chandy and Tellis 2000; Wells 1968). Within
Europe, Tellis, Stremersch and Yin (2003) find that Scandinavian countries are the
most innovative. In contrast, Putsis et al (1997) point out that Latin European
countries such as France, Italy and Spain may be the most innovative because they
are the most gregarious. Lynn and Gelb (1996) argue that Mid-European countries
are most innovative in Western Europe. Note that most empirical studies focus on
European countries.
Fourth, researchers in marketing have debated about whether diffusion speed
is accelerating over time. While Bayus (1992) found no systematic evidence of

92
accelerating diffusion rates over time, Van den Bulte (2000) finds evidence for
accelerating diffusion. Golder and Tellis (1997) find time-to-takeoff to be declining
for post War categories as compared to pre-War categories but neither Golder and
Tellis (1997) not Tellis, Stremersch and Yin (2003) find a significant effect for the
year of introduction in hazard models.
Fifth, while there has been a lot of debate in other disciplines whether
countries are converging in terms of economic development (Barro and Sala-i-
Martin 1992, Sala-i-Martin 1996) or culture (Dorfman and House 2004), there has
been no effort made in marketing to determine whether there is convergence or
divergence across countries in their ability to adopt new products.
This paper seeks to address these issues. In particular, it seeks answers to four
specific questions: First, how does time-to-takeoff vary across the major economies
of Asia, Europe, North America, South America, and Africa? Second, what drives
the variation in time-to-takeoff across countries? Third, are these differences
constant or varying over time? Fourth, if varying, is time-to-takeoff converging or
diverging across countries? These issues are examined by studying a heterogeneous
sample of 31 countries across 16 categories, including products and services.
The subsequent sections of the paper describe the theory, method, results,
implications, and limitations of the study.

93
3.2 THEORY AND HYPOTHESES
This section explores why time-to-takeoff of new products may vary across
countries. The word product broadly refers to both goods and services. Time-to-
takeoff can differ across countries due to one of two broad factors: economics or
culture. Economics can be thought of as differences in opportunities and wealth,
which limit consumers’ ability to purchase new products. Culture can be thought of
as differences in attitudes or inclinations, which accelerate or slow consumers’
acceptance of new products. I explore the role of various dimensions of these two
factors.
3.2.1 Culture’s Consequences
What is culture? Triandis (1995) defines subjective culture as shared beliefs,
attitudes, norms, roles and values found among speakers of a particular language
who live during the same historical period in a specified geographical region. Major
changes in climate and ecology, historical events, cultural diffusion (migration or
exposure to products from other countries) may drastically affect culture (Triandis
1995) but national cultures are generally thought to be stable over time (Dorfman
and House 2004; Hofstede 2001; Yeniyurt and Townsend 2003).
How does culture influence behavior? When people are immersed in a certain
culture, they develop a common pattern of thinking. This pattern of thinking
influences the degree to which the behavior of individuals, groups, and institutions

94
are viewed as legitimate, acceptable, and effective (Dickson, BeShears and Gupta
2004; House and Javidan 2004).
Cross-cultural researchers have documented several different dimensions of
national culture. I discuss five of the dimensions that are likely to affect the time-to-
takeoff of new products: In-Group Collectivism, Power distance, Performance
orientation, Religiosity and Uncertainty avoidance. The specific roles of in-group
collectivism, performance orientation and religiosity have not been addressed in the
prior literature on takeoff or diffusion.
3.2.2 In-Group Collectivism
Gelfand et al (2004) define In-group collectivism (subsequently referred to
simply as Collectivism) as the degree to which individuals express pride, loyalty and
cohesiveness in their organizations or families. Collectivist societies consist of
closely linked individuals who are primarily motivated by the norms of and duties
imposed by the collectives, are willing to give priority to the goals of these
collectives over their own personal goals, and emphasize their connectedness to the
members of these collectives (Triandis 1995). This attribute is in contrast to low
Collectivism (or Individualism), which refers to a society of loosely linked
individuals who view themselves as independent of collectives. These individuals are
primarily motivated by their own preferences, needs, rights and the contracts they
have established with others. They give priority to their personal goals over goals of

95
others and emphasize rational analyses of the advantages and disadvantages to
associating with others (Triandis 1995).
What is the effect of collectivism on time-to-takeoff? Collectivism is thought
of as making for adults that are compliant but not innovative (Triandis 1995;
Yeniyurt and Townsend 2003). In countries that have high levels of collectivism,
there may be less independence and freedom. As such, individuals may be less likely
to sample or try new products, adhering instead to the norms and traditions of their
society. Hence, I hypothesize
H1: New products may takeoff slower in countries that are high on
collectivism than in countries that are low on collectivism.
3.2.3 Performance Orientation
Performance orientation is the extent to which a community encourages and
rewards innovation and improvement in performance. Though originally thought of
as a byproduct of the “Protestant ethic’, researchers argue that this dimension is not
related to religion and that societies with high performance orientation could consist
of diverse religions (McClelland 1976, Javidan 2004). High performance-oriented
societies are characterized by a thirst for knowledge and improvement. This
dimension is also related to a high need for achievement (McClelland 1976), a trait
that may lead to greater willingness to experiment with new products.
What is the effect of performance orientation on time-to-takeoff? Because
performance orientation is associated with self-improvement, achievement, and a

96
thirst for new ideas, the speed of new product takeoff may be positively associated
with performance orientation. Hence, I hypothesize
H2: New products may takeoff faster in countries that are high on
performance orientation than in countries that are low on performance orientation.
3.2.4 Power Distance
Power distance is the extent to which the less powerful members of
organizations and institutions accept and expect that power is distributed unequally
(Hofstede 2001). Power distance is related to the concept of social inequality and
may be fostered by an emphasis on experience, tradition, heredity, class roles, and
spiritual leadership (Carl, Gupta and Javidan 2004).
What is the effect of power distance on time-to-takeoff? Countries with low
power distance typically have low inequalities of power and wealth. This situation
may ease communication between different sections of the population, leading to
faster diffusion of new products. When power distance is low, the underprivileged
have fewer barriers to the acquisition of new skills and are thus more likely to also
acquire new products (Carl, Gupta and Javidan 2004). Hence, I hypothesize
H3: New products may takeoff faster in countries that are low on power
distance than in countries that are high on power distance.

97
3.2.5 Religiosity
A superficial reading of religion has led some to assert that the “protestant
ethic” is different from Catholicism and other religions in that it encourages work,
industriousness, and innovation (e.g., McClelland 1976; Stulz and Willamson 2003).
Tellis, Stremersch and Yin (2003) found that the percentage of Protestants in a
country was positively related to speed of takeoff of new products. However, a focus
on the Protestant-Catholic distribution may confuse a superficial distinction between
some religions with a more basic underlying trait of religiosity. For example, many
researchers argue that certain core religious values, such as form or frequency of
worship, religious practices, and religious laws that are common across different
religions, constitute deep cultural traits (Lindridge 2005). These traits may drive
overt behavior associated with consumption and innovation (McClelland 1976). For
instance, Hossain and Onyango (2004) argue that an opposition to biotechnology is
not specific to any one religion but reflects conflict between mainstream religious
beliefs and the acceptance of scientific principles, experimentation, and learning. I
refer to this underlying trait as religiosity and formally define religiosity as the extent
to which individuals rely on a faith-based, non-scientific body of knowledge to
govern their daily lifestyle and practices.
What is the effect of religiosity on time-to-takeoff? Countries with a high
level of religiosity are likely to emphasize spiritual benefits and de-emphasize
material possessions and progress. In these countries, people are likely to be less

98
interested in or even skeptical of innovations. They are likely to adopt new products
only when they become mass products and are considered part of routine possessions
of individuals or households. For instance, Miller and Hoffman (1995) find a
negative correlation between religiosity and attitude toward risk. Hence, I
hypothesize
H4: New products may takeoff slower in countries that are high on religiosity
than in countries that are low on religiosity.
3.2.6 Uncertainty Avoidance
Uncertainty Avoidance deals fundamentally with the level of anxiety about
the future and the consequent need to protect society through traditions, rules, and
rituals (Sully de Luque and Javidan 2004). Cultures with high Uncertainty
Avoidance are characterized by a tendency towards orderliness, structured lifestyles,
clear specifications of social expectations, and rules to regulate uncertain situations.
What is the effect of Uncertainty Avoidance on time-to-takeoff? A society
characterized by low Uncertainty Avoidance may create an environment more
encouraging toward the adoption of innovations. In such societies, people are more
open to change, new ideas, risks, and diversity (Nakata and Sivakumar 1996;
Steenkamp, Hofstede and Wedel 1999; Yeniyurt and Townsend 2003). Hence,
societies with low levels of Uncertainty Avoidance may see a faster time-to-takeoff.
However, authors have also noted that societies with high levels of Uncertainty
Avoidance look toward technology to ward off uncertainty (Sully de Luque and

99
Javidan 2004). This might create an environment that encourages the adoption of
new high technology products. Hence, I do not hypothesize a specific effect of
Uncertainty Avoidance on time-to-takeoff.
3.2.7 Wealth of Nations
I examine the impact of economic development, information access, and
trade openness on time-to-takeoff of new products.
3.2.8 Economic Development
I consider two aspects of economic development: the absolute level of
economic development within a country as well as economic differences within a
country. Receptivity to new products is likely to be higher in countries that have a
higher level of economic development than in countries with a lower level of
economic development for at least two reasons: First, consumers in richer countries
are better able to afford new products early on when the prices are relatively high
(Golder and Tellis 1998). Indeed, innovators in general, have higher incomes
(Rogers 1995). Second, new products have uncertain performances. Richer
consumers can better afford the risks of buying such products.
Economic disparity in a country refers to the extent that a country’s wealth is
concentrated in the hands of a few. As a result, some consumers are wealthy while
others are poor. This can prevent the latter group from being able to afford new
products (Tellis, Stremersch and Yin 2003, Talukdar, Sudhir and Ainslie 2002, Van
den Bulte and Stremersch 2004). Hence, I hypothesize

100
H5a: New products takeoff faster in countries that have a higher absolute
level of economic development than in countries with a lower level of economic
development.
H5b: New products takeoff slower in countries that have a higher level of
economic disparity than in countries with a lower level of economic disparity.
3.2.9 Information Access
I consider two aspects of information access: mass media availability and
mobility. I expect products to takeoff faster in countries that have a higher level of
information access than countries that with a lower level of information access for
two reasons. One, information access is enhanced by a greater level of mass media
availability, which can greatly aid manufacturers in spreading knowledge about new
products via advertising. Availability of mass media can also aid consumers in
seeking out knowledge about new products or make them aware of the availability
and spread of new products (Gatignon and Robertson 1985; Horsky and Simon 1983;
Talukdar, Sudhir and Ainslie 2002). Two, prior research suggests that the presence
of better infrastructural facilities enhances mobility and promotes interpersonal
communication and the spread of information within local systems (Gatignon,
Eliashberg and Robertson 1989; Tellis, Stremersch and Yin 2003). Hence, I expect
H6: New products to takeoff faster in countries that have a higher level of
mobility than countries with a lower level of mobility.

101
3.2.10 Trade Openness
Trade openness reflects economic linkages across countries and may
influence both the demand and supply of new products. Advocates of neo-liberal
reform argue that greater openness encourages efficiency, productivity, and
competitiveness (Perkins and Neumayer 2004). Trade, foreign direct investment and
technology flows help in knowledge spill-over across countries leading to awareness
about new products and greater availability, which in turn lead to faster takeoff
(Perkins and Neumayer 2004; Talukdar, Sudhir and Ainslie 2002; Tellis, Stremersch
and Yin 2003). Hence, I expect
H7: New products take off faster in countries that have a higher level of trade
openness than countries with lower levels of trade openness.
3.2.11 Control Variables
I control for three important variables: product class, product vintage and
prior takeoffs. In addition, I also examine the impact of population density as another
control variable. Here, I discuss the rationale for the first three variables.
Product Class
The time-to-takeoff may vary depending on the product class. I use the term
work products for products that help consumers work efficiently, such as microwave
ovens, washing machines, and dryers. I use the term entertainment products for those
products that provide consumers with information, pleasure, or enjoyment, such as
personal computers and DVD players. The former groups have also been called time-

102
saving and the latter time-consuming products respectively. Fun products may
takeoff faster than work products because they appeal to more individuals in the
household, are on public display, are often discussed in social circles, and provide
instant gratification and stimulation (Bowden and offer 1994; Horsky 1990; Tellis,
Stremersch and Yin 2003). Hence consumers are willing to adopt the products soon
after introduction even if the prices are relatively high.
Product Vintage
Bayus (1992) argues that product diffusion rates are not accelerating over
time. If that is true, then the vintage of the product would not have any impact on its
time-to-takeoff. However, Van den Bulte (2000) argues that diffusion speed has been
accelerating over time due to systematic increase in purchasing power, demographic
changes, and changing nature of products. I believe that products maybe taking off
faster now than in prior periods for several reasons. First, recent decades have
witnessed a liberalization of markets within countries, which have stimulated growth,
openness, and investment rates within countries (Wacziarg and Welch 2003). This
situation is likely to lead to increased manufacture, supply, and trade of higher
quality products. Second, several media, especially TV, telephony, and internet are
rapidly increasing in penetration all over the world allowing faster diffusion by news,
advertising, and word-of-mouth. Third, technology is evolving faster in recent
periods than in prior periods (Sood and Tellis 2006). All these factors can shorten the

103
time-to-takeoff. Thus, I expect product vintage to have a negative effect on time-to-
takeoff.
Prior Takeoffs
Researchers have argued that consumers learn about the product from its
prior diffusion (Ganesh, Kumar and Subramaniam 1997; Kumar, Ganesh and
Echambadi 1998). Also, imports from, travel to, and news reports from a country
where a new product has already taken off may facilitate its acceptance in a
neighboring country where it has not. As a result, takeoff in a lagging country may
occur faster than that in a leading country proximate to it. We expect new products to
takeoff faster when there are a greater number of prior takeoffs in neighboring
countries
3.3 METHOD
This section describes the sampling, sources, measures, and model for the
analysis.
3.3.1 Sample
Two criteria guide the selection of product categories. One, they should
include a mix of both work and fun products. Two, they should include a mix of
categories studied in prior research and others not studied before. Based on these
criteria, and data availability, I collect market penetration data across 16 products
and services. Work products are microwave oven, dishwasher, freezer, tumble dryer

104
and washing machine. Fun products are CD player, cellular phone, personal
computer, video camera, video tape recorder, MP3 player, DVD player, digital
camera, hand-held computer, Broadband and Internet. Broadband, DVD player,
digital camera, hand-held computer, and MP3 player, commercialized after 1990,
may be considered contemporary new products while the others may be considered
to be established categories. Cellular phone, Internet, and Broadband may be
considered as three generations of telecommunication services.
Two criteria guide the selection of the sample of countries. First, the sample
should be representative of major cultures and populations of the world. Second, the
sample should include major economies of the world. Using these criteria, I obtain
data on 40 countries of the world. Since, for some countries, I had very little data, to
avoid data-specific biases, I retain countries where I have data for at least 10
categories. As a result, I had to drop Argentina, Australia, Colombia, Hong Kong,
Malaysia, New Zealand, Singapore, South Africa and Turkey.
In total there are 430 categories x countries, which comprises 90% of the
possible combinations. On each category x country I have time series data ranging
from 4 to 55 years. This is probably the largest data set assembled for the study of
the diffusion of new products across countries.

105
3.3.2 Sources
I collect this data from a variety of sources: subscription-based sources
(Euromonitor Global Marketing Information Database, World Development
Indicators Online, Fast Facts Database), archival search through freely available
secondary sources (Historical Statistics of Japan, Historical Statistics of Canada,
Electrical Merchandising, Merchandising, Merchandising Week and Dealerscope
journals for US, OECD statistics), and proprietary data (Database of Tellis,
Stremersch and Yin 2003), over several hundreds of research hours. Our final sample
of 429 observations represents around 90% of the possible category x country
combinations. This is probably the largest data set assembled for the study of the
diffusion of new products across countries. I collect this data from a variety of
sources: subscription-based sources (Euromonitor Global Marketing Information
Database, World Development Indicators Online, Fast Facts Database), archival
search through freely available secondary sources (Historical Statistics of Japan,
Historical Statistics of Canada, Electrical Merchandising, Merchandising,
Merchandising Week and Dealerscope journals for US, OECD statistics), and
proprietary data (Database of Tellis, Stremersch and Yin (2003), over several
hundreds of research hours.

106
3.3.3 Measures
This section describes the measures for Market Penetration, Year of
Commercialization, Year of Takeoff, independent variables and control variables.
Market Penetration
For market penetration, I use the measure, where available, of possession of
durables per 100 households. For four categories (DVD player, digital camera, MP3
player and hand-held computer) where only sales data is available for most countries,
we used the following formula to obtain Market Penetration
100 * / ) (
1 t r t t t t
useholds NumberofHo Sales Sales n Penetratio n Penetratio
− −
− + = (1)
Where ‘r’ is the average replacement time for the category. I use an average
replacement cycle of 4 years for DVD player, MP3 player, and hand held computer
and 5 years for digital camera. I checked robustness of these assumptions by varying
r by plus or minus 1 year. The year of takeoff varies insignificantly with the changes.
Year of Commercialization
There are two inherent problems in identifying the exact year of introduction
of products in countries. One, this date is not explicitly published in journal articles
while various data sources provide conflicting dates. Two, most databases include a
product only when it has achieved non-trivial sales. Hence, there is an inherent
survivor bias. Following Agarwal and Bayus (2002), we use the word
commercialization to reflect the fact that data bases seem to include a product only

107

when it has become available to the mass market or achieved some minimal level of
sales or penetration.
We use a combination of rules to obtain reasonable estimates of the
approximate year of commercialization that best reflects individual categories. For
work products, we look for the earliest year of commercialization for each country
from the data published in the various sources viz. Euromonitor Inc. journals and
databases, various issues of Merchandising, Merchandising Week, and Dealerscope,
published dates in Agarwal and Bayus (2002), Golder and Tellis (2004, 1997),
Talukdar, Sudhir and Ainslie (2002), and by examining our own data.
In the case of telecommunication products (Cellular phone, Internet and
Broadband), the year of commercialization is dependent on the national regulatory
policies and hence we use varying dates made available from reliable secondary
sources. For Cellular phone, we use the date of first adoption of Cellular
technologies reported in Gruber (2005) and reports in the OECD web-site
(www.oecd.org) for the EU countries and secondary reports by market research firms
on the ISI Emerging Markets Database for emerging markets. For Internet, we use
the date of the initial NSFNET connection by OECD countries as obtained from
OECD reports
3
and dates of first internet services launch for emerging markets from
the ITU database and by market research firms on the ISI Emerging Markets
Database. For Broadband, we look for the earliest commercial launch of either the

3
Information Infrastructure Convergence and Pricing: The Internet, Organization
For Economic Co-Operation and Development, Committee For Information,
Computer and Communications Policy, Paris 1996

108

Cable or the DSL service in each country, as reported in the reports in the OECD
web-site
4
and the ISI Emerging Markets Database.
For four fun products (personal computer, CD player, VCR and Video
Camera), the data as well reports and published dates in secondary sources reflect a
common date for North America, Europe, Japan and South Korea. We use the
earliest year of commercialization based on our data and published sources (Talukdar,
Sudhir and Ainslie 2002) for each remaining individual country. For products
introduced after 1990, i.e., DVD players, digital camera, MP3 player, hand-held
computers, where validation from secondary reports is not as yet available, and the
data-derived years of commercialization seem similar across countries, we use a
common year of commercialization across all countries.
We further validate each of these dates by checking that penetration in the
year of commercialization has not exceeded 0.25%, which is a stricter rule than the
0.5% rule recommended by Tellis, Stremersch and Yin (2003).
Year of Takeoff
The literature contains many measures of takeoff. Agarwal and Bayus (2002)
define takeoff as the central partition between a pre-takeoff and post-takeoff period,
determined by a percentage change in sales. Garber et al (2004) and Goldenberg,
Libai and Muller (2001) define takeoff at the point when market penetration is 16%.

4
The Development of Broadband Access in OECD Countries, Directorate For
Science, Technology and Industry Committee For Information, Computer and
Communications Policy, 2001

109
Golder and Tellis (1997) define takeoff as the first year in which a new product’s
sales growth rate relative to the prior year’s sales crosses a threshold based on sales
levels. Tellis, Stremersch and Yin (2003) define takeoff as the first year a new
product’s sales growth rate relative to the prior year’s sales crosses a threshold based
on a penetration levels.
For a cross-country study such as ours, the measure of takeoff proposed by
Tellis, Stremersch and Yin (2003) while appropriate, is also very demanding, as it
requires both sales and market penetration data. Rather than sacrifice the breadth of
categories and countries for which I had market penetration, I use a measure of
takeoff that is similar in form to one of Garber et al (2004) and Goldenberg, Libai
and Muller (2001) but in substance to that of Tellis, Stremersch, and Yin (2003).
However, Golder and Tellis (2004, 1997) find that the average penetration at takeoff
is 1.7%. Interestingly, this latter finding is similar to Roger’s (1995) estimate that
innovators make up 2.5% of the population and Mahajan, Muller and Srivatsava’s
(1990) upper bound of 2.8% for innovators. So I use the heuristic that the year of
takeoff is the first year the market penetration reaches 2%. The key issue for
subsequent analysis is that I use the same rule consistently across countries. In
essence, our measure of takeoff reduces our definition of takeoff to an instrumental
one. Thus, an alternate interpretation of all our results is how quickly and why do
new products reach a 2% market penetration in various countries of the world.
I define time-to-takeoff as the difference between the year of takeoff and the
year of commercialization in a country.

110
Independent Variables
One measure for Economic Development is the Real GDP per capita
(Laspeyres) measured in $ terms from the Penn World Tables (Heston, Summers and
Aten 2002). This is obtained by adding up consumption, investment, government and
exports, and subtracting imports in any given year. It is a fixed base index where the
reference year is 1996. Since this data is available only up to 2000, I calculate GDP
per capita for the years 2001 to 2004 using average growth rate figures from the
UNDP Human Development report. A related measure for Economic Development
is the electric power consumption in kwh per capita (production of power plants and
combined heat and power plants, less distribution losses, and own use by heat and
power plant). The measures for Information Access include radio receivers in use for
broadcasts to the general public per 1000 people, television sets per 1000 people,
telephone mainlines (lines connecting a customer's equipment to the public switched
telephone network) per 1000 people, and vehicles (including cars, buses, and freight
vehicles but not two-wheelers) per 1000 people.
Multiple items measure the extent of Trade Openness- trade (the sum of
exports and imports of goods and services) as % of GDP, trade in goods (the sum of
merchandise exports and imports) as % of GDP, gross foreign direct investment (the
sum of the absolute values of inflows and outflows of foreign direct investment
recorded in the balance of payments financial account) recorded as a % of GDP and
gross private capital flows ( sum of the absolute values of direct, portfolio, and other
investment inflows and outflows recorded in the balance of payments financial

111
account) recorded as a % of GDP. I derive all these measures from World
Development Indicators Online, a database provided on subscription basis by the
World Bank.
Gini Index is a measure of Economic Disparity that exists in the population. I
derive this from the Deninger and Squire (1996) database. This database gives
multiple Gini coefficients, and hence I consider only those coefficients that are
considered ‘acceptable’, and are measured at the national level. For some countries
(Austria, Egypt and Morocco), where acceptable estimates are not obtainable from
the database, I use measures derived from the CIA World Factbook (2003). I use
People per Square Kilometer as a measure for Population Density, from the World
Population Prospects: The 2000 Revision, United Nations Population
Division/Department of Economic and Social Affairs.
I measure dimensions of culture (Collectivism, Power Distance, Performance
Orientation and Uncertainty Avoidance) using the societal practices scores reported
in the Global Leadership and Organizational Behavior Effectiveness (hereby referred
to as GLOBE) research program (House et al 2004). This is a long-term program
designed to conceptualize, operationalize, test, and validate a cross-level integrated
theory of the relationship between culture and societal, organizational, and
leadership effectiveness. The cultural dimensions proposed in this project are similar
in spirit but vary operationally from the traditional indices used in cross-cultural
research such as Hofstede’s indices (Hofstede 2001). The GLOBE dimensions are
better defined and suffer less from confounds in meaning and interpretation than the

112
Hofstede measures (House and Javidan 2004). The GLOBE dimensions are
constructed based on responses to questionnaires by 17000 managers in 62 cultures
to two types of questions- managerial reports of actual practices in their societies or
their organizations and managerial reports of what should be the practices and/or
values in their societies or organizations. The values are expressed in response to
questionnaire items in the form of judgments of ‘What Should Be’. I however use
actual practices as measured by indicators assessing ‘What Is, or What Are’, common
behaviors, institutional practices, proscriptions and prescriptions. House et al (2004)
note that the practices approach to the assessment of culture grows out of a
psychological/behavioral tradition, in which it is assumed that shared values are
enacted in behaviors, policies, and practices. Hence, I believe that actual practices
reflect the behavior of the people and are more useful in explaining time-to-takeoff
than the values measures.
Religiosity or religiousness has been measured in prior literature through the
use of variables such as church

attendance, frequency of prayer, belief

in God, belief
in

the authority of the

Bible, and self-appraised level

of religiousness (Hossain and
Onyango 2004; Lindridge 2005; Wilkes, Burnett and Howell 1986). Since I require a
measure that is suitable across countries, some of whom have many different
religions, I construct a unified measure of Religiosity using two items which I obtain
from the World Values Survey from the site http://www.worldvaluessurvey.org/.
This survey is a large investigation of socio-cultural and political change carried out
by an international network of social scientists in several waves since 1981. For the

113

first measure, I use the responses to the question ‘How often do you attend religious
service?’ in the World Values Survey. The responses can range from ‘> 1 week’ to
‘Never’. In some religions, such as Hinduism, worship can be done within the home
and attendance in religious services may not be necessary (Lindridge 2005). Hence, I
also consider a second item from the World Values Survey involving a response to
the question, ‘How important is God to your life’. The responses can range from
‘Not at all’ to ‘Very’. I take the average of One, the percentage of respondents in the
sample answering either ‘> 1 week’ or ‘Weekly’ to the first question on the
attendance of religious service and Two, the percentage of respondents in the sample
answering either ‘Very’ or ‘9’ to the second question on the importance of God to
construct a unified measure of Religiosity5.
Control Variables
I use the year of first ever commercialization of the product category in any
country as a measure of Product Vintage. I measure Prior Takeoffs as the number of
takeoffs in the prior or same year in countries in the same region as a target country.

5
For Thailand, the World Values Survey does not give measures that can be used to
construct Religiosity. We have taken the corresponding measures for Vietnam as a
surrogate for Thailand, as it has geographical and religious proximity.

114
3.4 MODEL
I model takeoff as a time dependent binary event. I face two issues with our
data. One, there are a number of censored observations. Two, the probability of
takeoff may increase with the length of time a product has not taken off. Hence, I use
a hazard function to model takeoff. The time to the takeoff from the time of
commercialization of a product in a country T is a random variable with a probability
density f(t) and a cumulative density F(t). The likelihood that a product takes off,
given that it has not taken off in the interval [0, T] is
)) ( 1 /( ) ( ) ( t F t f t h − = (2)
I can use either a non parametric method to model the effects of covariates on
the hazard or use parametric methods, like the accelerated failure time approach, to
model the effects of independent variables on time to event i.e., takeoff. In the
accelerated failure time approach, the hazard of takeoff is of the form
) (exp exp ) , (
0
t h X t h
i i
aX aX
i i
= (3)
i.e., the impact of independent variables on the hazard for the i
th
observation
(each category*country combination) is to accelerate or decelerate time-to-takeoff as
compared to the baseline hazard (see Srinivasan, Lilien and Rangaswamy (2004) for
a detailed description of this approach).

115
An easier way of estimating this model is to write it as follows
σ ε β + = X Y (4)
Where Y is the vector of the log of time-to-takeoff, X is the matrix of
covariates, ß is a vector of unknown regression parameters, σ is an unknown scale
parameter and є is a vector of errors, assumed to come from a known distribution
such as normal, log-gamma, logistic or extreme value forms, leading to the log-
normal, gamma, log-logistic, or the Weibull/exponential distributions for T
respectively. I use PROC LIFEREG in SAS to estimate this model assuming a two
parameter weibull distribution for the error term which leads to a Weibull
distribution for T (Allison 1995). The estimation is done via maximum likelihood.
3.5 RESULTS
The first section presents the results of a factor analysis of some of the
independent measures carried out to achieve parsimony in the data. The second
section presents some descriptive statistic for initial insight into the phenomenon.
The third section presents results of the hazard model. The fourth section examines
differences in time-to-takeoff across economic and cultural clusters. The fifth section
examines whether there is convergence in takeoff. The sixth section presents results
of robustness checks.

116
3.5.1 Factor Analysis of Economic Variables
Economic variables are highly correlated, suggesting the presence of
underlying factors. In particular, Dekimpe, Parker and Sarvary (2000) note in their
review of global diffusion that constructs like mobility, operationalized by the
ownership of automobiles, are often considered distinct from wealth but are actually
highly related to wealth and are also used in some studies as describing the wealth of
a country (Helsen, Jedidi and DeSarbo 1993, Ganesh, Kumar and Subramaniam
1997). I test this point of view with a factor analysis of all the economic measures,
including those for information access. I run an exploratory factor analysis of our
economic measures using data from 1950 to 2004, using the principal components
approach and Varimax rotation of these dimensions. I obtain a two factor solution
from the exploratory factor analysis (See Table 2). Based on the loading of items, I
term these factors: Wealth and Openness, and these factors are used in the hazard
model as the two dimensions related to economics.
Table 2: Factor Analysis of Economic Variables
WealthOpenness
Television sets per 1000 people 0.93 0.25
GDP per capita 0.90 0.33
Vehicles per 1000 people 0.90 0.00
Telephone mainlines per 1000 people 0.89 0.33
Electricity consumption per capita 0.86 0.23
Radios per 1000 people 0.85 0.22
Trade (% of GDP) 0.11 0.91
Trade in goods (% of GDP) 0.09 0.90
Gross private capital flows (% of GDP) 0.34 0.74
Gross foreign domestic investment (%
of GDP)
0.30 0.69

117

I do not run a separate factor analysis for cultural variables, because the
cultural dimensions which I use already represent distinct factors (Hofstede 2001;
House et al 2004; Van den Bulte and Stremersch 2004).
3.5.2 Descriptive Statistics on Takeoff
I first examine our data for outliers by simultaneously examining the plots of
time-to-takeoff across products and countries. I find one observation (Dishwasher in
US) to be an extreme outlier and delete it from our analysis.
Takeoff occurs in 80% of the 429 country x category combinations. Takeoff
has occurred in all countries for very old and/or very useful categories (e.g. Washing
machine, Internet, Cellular phone). Lack of takeoff may be due to the effect of the
hypothesized explanatory variables or insufficient time to take off for younger
categories in particular countries. The latter problem is referred to in the hazard
model literature as censored data. The advantage of the hazard model is that it can
estimate the effects of the independent variables on censored data.
Table 3 shows the mean time-to-takeoff across categories for each country.
Countries vary widely in terms of the mean time-to-takeoff. What are the reasons for
these differences? The Hazard analysis seeks to answer this question.

118
Table 3: Time-to-takeoff across Countries
Mean Std Total
Japan 5.4 3.3 14
Norway 5.7 2.4 15
Sweden 6.1 2.9 15
Netherlands 6.1 3.7 16
Denmark 6.1 2.6 15
US 6.23.4 14
Switzerland 6.3 3.4 15
Austria 6.4 3.3 15
Belgium 6.5 2.5 16
Canada 6.9 5.2 12
Finland 7.0 2.6 15
Germany 7.1 4.3 15
South Korea 7.2 3.3 12
Venezuela 7.3 4.5 12
UK 8.04.5 14
France 8.2 3.5 15
Italy 8.34.0 15
Spain 8.5 4.0 14
Chile 8.5 5.7 11
Mexico 8.7 3.7 11
Portugal 8.8 4.5 15
Greece 9.0 4.4 14
Brazil 9.3 4.9 11
Thailand 10.2 6.3 12
Egypt 12.1 5.3 13
Morocco 12.3 6.3 12
India 12.4 5.0 14
Philippines 12.6 7.1 13
Indonesia 13.6 6.2 15
Vietnam 13.9 5.6 14
China 13.9 6.1 16

119

3.5.3 Tests of Hypotheses via Hazard Model
I estimate the hazard model in Equation 3, assuming a Weibull baseline
distribution (A subsequent subsection tests the robustness of this assumption). The
dependent variable is the log of the time-to-takeoff. Note, except for the cultural
variables, Product Vintage and Product Class, all independent variables are time
specific. A positive sign for the estimated coefficient indicates that an increase in the
independent variable is associated with a lengthening of the time-to-takeoff. I
estimate the hazard model for 27 out of 31 countries (373 observations). I drop
Belgium, Chile, Norway and Vietnam because they were not included in the GLOBE
study from which I obtain the measure for the cultural variables.
I first run the model for each of the independent factors or variables
separately (see Uni-variate Analysis in Table 4). As expected, Product Vintage has a
coefficient which is both negative and significantly different from zero. i.e., products
commercialized later in time takeoff faster than those earlier in time. I find
successive generation of communication products (Cellular phone, Internet and
Broadband) have shorter times-to-takeoff. Figure 5 shows that time-to-takeoff has
typically been declining over calendar time.

120
Figure 5: Time-to-Takeoff over Calendar Time
Mean Time-to-Takeoff over Calendar Time
0.0
2.0
4.0
6.0
8.0
10.0
12.0
14.0
16.0
1908 1915 1936 1939 1967 1972 1975 1976 1979 1983 1988 1994 1994 1996 1996 1996
Product Vintage
Mean Time-to-takeoff (n Years)
Mean time-to-takeoff Linear (Mean time-to-takeoff)

As expected, Prior Takeoffs also have an effect that is negative and
significantly different from zero in Table 4. This implies the existence of learning or
diffusion effects between neighboring countries that shorten time-to-takeoff.

121
Table 4: Estimates from the Hazard Model
Uni-variate Analysis Multivariate
Analysis
Construct Beta
(T-stats)
Log
likelihood
R square-like Beta
(T-stats)
Product Vintage -0.01
(-7.29)
-365.72 0.07 -0.005
(-2.14)
Prior Takeoffs -0.09
(-10.15)
-354.82 0.10 -0.02
(-2.05)
Product class
(Work =1)
0.51
(7.29)
-366.75 0.07 0.20
(2.01)
Population Density 0.00
(1)
-393.17 0.00
Wealth -0.32
(-12.79)
-327.42 0.17 -0.08
(-1.90)
Openness 0.01
(0.40)
-393.41 0.00
Economic Disparity 0.02
(3.94)
-385.02 0.02 0.00
(-0.80)
Performance
Orientation
0.17
(1.83)
-391.87 0.00
Uncertainty Avoidance -0.29
(-4.81)
-382.2 0.03 0.20
(2.95)
In-group Collectivism 0.41
(11.52)
-332.28 0.16 0.33
(4.01)
Power Distance 0.47
(6.45)
-375.78 0.04 0.01
(0.04)
Religiosity 0.01
(6.62)
-370.68 0.06 0.0
(1.20)
Log-Likelihood -286.79
R square-like 0.27

As expected, work products are associated with a longer time-to-takeoff than
fun products. Descriptive analysis suggests that the mean time-to-takeoff of fun
products is 7 years while that for work products is almost double at 12 years (see
Table 5), with much of the difference being attributed to developing countries.

122

Table 5: Time-to-Takeoff by Product Class and Economic Development
All Countries Developed Countries Developing Countries
Class Mean
(Std)
Total %
Take
off
Mean
(Std)

Total %
Take
off

Mean
(Std)

Total %
Take
off
Fun

7.3
(3.9)
305 81

6.2
(3.2)
184 94% 8.9
(4.5)
121 60%
Work

11.8
(6)
124 78 8.9
(4.4)
80 99% 17.0
(5.2)
44 41%

In the hazard models, an increase in Wealth is associated with a shorter time-
to-takeoff (Table 4) while an increase in Economic Disparity is associated with a
longer time-to-takeoff, as hypothesized. Contrary to hypotheses, the effect of
Openness and Population Density are not significantly different from zero. The
effect of Performance Orientation is not in the expected direction. An increase in
Uncertainty Avoidance is associated with a shorter time-to-takeoff. High levels of
Collectivism, Power Distance and Religiosity are each associated with a longer time-
to-takeoff as hypothesized.
I next explore the relative strength of all the significant independent variables
by a multivariate hazard model. All variables retain their significance and expected
direction, except for the Economic Disparity and the three cultural variables of
Religiosity, Power Distance, and Uncertainty Avoidance. These results indicate that
the effects of most of the cultural models are not robust to specification. On the other
hand, the effects of Product Class, Prior Takeoffs, Product Vintage, Wealth,

123
Collectivism are strong, robust, and in the expected direction. This model explains
27% of the variance. These results indicate that both Economics and Culture
determine differences in time-to-takeoff. To complement and enrich the above
analysis I next consider how takeoff varies across cultural clusters of countries.
3.5.4 Differences across Cultural Clusters
Much research suggests the existence of distinct cultural clusters of countries
(Gupta and Hanges 2004; Ronen and Shenkar 1985). I identify eight cultural clusters
(see Tables 6, 7 and 8 for a description of these clusters and the logic for their
classifications based on prior research (Ashkanasy, Trevor-Roberts and Earnshaw
2002; Gupta and Hanges 2004; Gupta et al 2002; Jesuino 2002; Kabasakal and
Bodur 2002; Szabo et al 2002; Ronen and Shenkar 1985). Countries within these
clusters exhibit similar culture because of geographic proximity, common language,
common ethnicity, or shared history. I compare the five cultural variables using the
measures from the GLOBE Study and World Values Survey. For each variable, I
present the mean and the standard deviation within a cluster. Note that except in the
case of religiosity for Confucian Asia, the means are more than twice the values of
the standard deviation within the cluster, justifying the grouping of these countries
within a cluster. Also, the means are often significantly different from the mean for
the rest of the countries in many cases, supporting inter-cluster variation.

124
Table 6: Cultural Clusters in Europe and America
Cultural
clusters
Nordic
Europe
Anglo Germanic Europe
Countries Sweden
Denmark
Finland
Canada
USA
Austria
Germany
Switzerland
Netherlands
UK
Logic for
Cluster
Geographic
proximity
Common
Nordic
history,
religion and
languages
Ethnic and
linguistic
similarities
Secular with strong
legal infrastructures
Linguistic and
religious similarities
Tradition of
orderliness,
standards, and rules
In-group
Collectivism
3.8* (0.3) 4.2* (0) 4.2* (0.4)
Power
Distance
4.5 (0.6) 4.85* (0) 4.9 (0.5)
Religiosity 8.4* (3.2) 47.8 (14) 18.1* (6.6)
Performance
Orientation
3.9( 0.3) 4.5** (0) 4.3** (0.2)
Uncertainty
Avoidance
5.2** (0.2) 4.4 (0.3) 4.9** (0.3)

**Significantly higher than mean of rest of countries
*Significantly lower than mean of rest of countries (p<0.10 or p <0.05)
Standard deviations in parentheses

125

Table 7: Cultural Clusters in Latin America and Europe
Cultural
clusters
Latin America Latin Europe
Countries Brazil
Mexico
Venezuela
France
Italy
Portugal
Spain
Greece
Logic for
Cluster
Roman law heritage,
common Spanish or
Portuguese languages
Similar Emphasis on
family living, food,
clothing, and lifestyle
Shared history of Roman
empire
Roman Catholic tradition
and languages based on
Latin
Paternalistic role of State
Similar emphasis on
family living, food,
clothing, and lifestyle
In-group
Collectivism
5.5** (0.3) 5.1(0.5)
Power Distance 5.3** (0.1) 5.4** (0.1)
Religiosity 64.7** (4.8) 29.1(13.6)
Performance
Orientation
3.8 (0.4) 3.7* (0.3)
Uncertainty
Avoidance
3.7* (0.4) 3.9* (0.4)

**Significantly higher than mean of rest of countries
*Significantly lower than mean of rest of countries (p<0.10 or p <0.05)
Standard deviations in parentheses

126
Table 8: Cultural Clusters in Asia and Africa
Cultural
clusters
Confucian
Asia
Northern Africa Southern Asia
Countries China
Japan
South Korea

Egypt
Morocco

India
Indonesia
Philippines
Thailand

Logic for
Cluster
Historical influence
of China
Confucianism
Emphasis on
hierarchy, diligence,
self-sacrifice and
delayed gratification
Influence of Arab
invasion, Islamic
legal and moral
code and the Arabic
language
Geographical
proximity to
Northern Rim
Peaceful
coexistence of
diverse religions,
languages,
customs and
cuisines
Similarity in
values, such as
morality, respect
for elders and
conservation of
resources
In-group
Collectivism
5.3 (0.6) 5.8** (0.2) 5.9**(0.3)
Power
Distance
5.2 (0.3) 5.4 (0.6) 5.4** (0.2)
Religiosity 11.3 * (12.9) 69.5**(4.1) 57.8 (29.8)
Performance
Orientation
4.4 **(0.2) 4.1 (0.2) 4.3 (0.2)
Uncertainty
Avoidance
4.2 (0.7) 3.9 ( 0.3) 4.0* (0.1)

**Significantly higher than mean of rest of countries
*Significantly lower than mean of rest of countries (p<0.10 or p <0.05)
Standard deviations in parentheses

127
Table 9 shows the differences in time-to-takeoff across the eight distinct
cultural clusters. Here again, the results show distinct differences in mean time-to-
takeoff between clusters with low standard deviations within clusters, across both
work and fun products. I consider the average of work and fun products for each
country in each cultural cluster and run both uni-variate ANOVA as well as
MANOVA. The tests indicate significant differences across the cultural clusters (For
Wilks Lambda and Pillai’s Trace, Prob > F =0.002). As further evidence of the
strength of culture, note how Latin countries across both Europe and America have
very similar mean times-to-takeoff, despite being geographically separate.
Table 9: Time-to-takeoff across Cultural Clusters
Cultural Cluster
Average
all
products
Std all
products
Average
fun
products
Std fun
products
Average
work
products
Std
work
products
Nordic Europe 6.44 -2.7 6 -2.4 7.33 -3.3
Anglo America 6.54 -4.2 5.3 -2.2 11.6 -2.4
Germanic
Europe 6.8 -3.9 5.8 -3.3 9.2 -3.8
Latin America 8.2 -3.3 7.4 -2.7 12.6 -4.4
Latin Europe 8.6 -4.1 7.6 -3.3 10.6 -5
Confucian Asia 9 -4.6 7.8 -4.2 13.2 -2.7
Confucian Asia
w/o Japan 11 -6.1 9.1 -4.1 18 -3.8
North Africa 12.2 -5.8 9.2 -3.7 17.4 -5
Southern 12.3 -6.2 9.7 -4.5 18.4 -4.9

128
Table 10 examines the impact of cultural clusters on time-to-takeoff via the
hazard model. I include the year of commercialization, prior takeoffs, and product
class. I do not include the cultural and economic variables because they are highly
collinear with cultural clusters. Countries in the Confucian Asia cluster, Latin Europe,
Latin America, North Africa, and Southern Asia see significantly slower times-to-
takeoff of products than those in the Nordic cluster, which is excluded and serves as
a comparison group. Results include Japan both in and outside the Confucian Asia.
Table 10: Hazard Model with Cultural Clusters
Hazard Model with Japan
in Confucian Asia
Hazard Model without
Japan in Confucian Asia
Log likelihood -292.32 -266.44
R square-like 0.26 0.29
Observations 373 358
Beta T-stats Beta T-stats
Product Vintage -0.01 -2.75 -0.01 -2.57
Prior Takeoffs -0.02 -1.79 -0.02 -1.92
Product Class
(Work=1)
0.19 1.92 0.20 1.97
Anglo America -0.03 -0.21 -0.03 -0.23
Germanic Europe 0.07 0.82 0.07 0.87
Latin Europe 0.27 3.12 0.27 3.21
Latin America 0.40 3.23 0.39 3.28
North Africa 0.77 5.00 0.75 5.09
Confucian Asia 0.50 4.29 0.68 5.20
Southern Asia 0.87 6.98 0.85 7.12
Nordic Europe

In Table 9, Fun products seem to takeoff faster than work products within
every cultural cluster However, the differences across cultural clusters for work
products are higher than the differences across the cultural clusters for work products.
This result suggests Work products are more culture-bound than fun products

129
because the former relate to food and clothing habits, which are immersed in cultural
traditions. Such cultural products may take off rapidly in some countries where they
match the culture (e.g. rice cooker in Japan or coffee maker in the US) and slowly in
other countries where they do not match the culture (e.g. coffee maker in China or
rice cooker in Germany). On the other hand, fun products (e.g. Cellular phones,
cameras) are used in a similar manner all over the world. Hence, their times-to-
takeoff is likely to vary less dramatically across countries. This intuition finds
support from separate analyses on work and fun products.
In Table 11, variables that are significantly different from zero from
individual hazard models are run in a joint regression by product class. For fun
products, the effects of product vintage, prior takeoffs, economics and in-group
collectivism are significantly different from zero and in the expected direction.
Results for economic openness are also significantly different from zero and in the
expected direction indicating that not only the absolute value of wealth but also
economic linkages across countries are important in influencing time-to-takeoff of
fun products. For work products however, only the effects of culture are significantly
different from zero and in the expected direction. Not only high levels of in-group
collectivism but also religiosity impact time-to-takeoff of work products.

130
Table 11: Hazard Model for Fun versus Work Products
Fun products

Work products

Variables Beta T-stats Beta T-stats
Product Vintage -0.02 -4.7
Prior Takeoffs -0.02 -2.7 -0.03 -0.86
Wealth -0.09 -1.7 -0.02 -0.16
Openness -0.05 -1.6
Economic disparity 0.00 -0.5 -0.01 -0.59
In-group
Collectivism
0.27 2.9 0.49 3.02
Power distance 0.00 0.0 -0.03 -0.16
Religiosity 0.00 -0.1 0.01 1.94
Uncertainty
avoidance
0.19 2.8 0.27 1.73
Observation 267 105
Log likelihood -177.41 -84.3
R Square-like 0.29 0.28

Has there been a convergence in takeoff over time for both long time-to-
takeoff countries and shorter time-to-takeoff countries? In the next section, I
examine convergence patterns over time.
3.6 CONVERGENCE IN TIME-TO-TAKEOFF
Though our results indicate substantial differences in time-to-takeoff across
countries, a key issue is whether takeoff patterns across countries are converging or
diverging? I use the word Convergence to refer to the decrease over time in the range
of dates of takeoff across the same set of countries. Convergence in product takeoff
may occur due to several reasons. First, economists have documented convergence in
wealth across countries over time, especially in countries that were previously poor
(Sala-i-Martin 1996). Second, most countries are enjoying better access to the media,

131
which facilitates the diffusion of new products, but here again , the improvement is
greatest in countries that were furthest behind. Third, most countries are also
experiencing a greater similarity in culture due to increasing inter-country
communication and travel, common education curriculum, use of English, exposure
to western practices, adoption of common cultural activities such as movies and
music, and diffusion of Eastern religions and philosophies (such as yoga and
Buddhism). Thus, cultural differences that caused divergence could be dissolving,
albeit slowly (Dorfman and House 2004). Indeed, a fear of such a trend and the need
to maintain cultural uniqueness may be seen in Europe (Dorfman and House 2004).
For instance, the French government has taken several measures to prevent the
contamination of French culture by American culture.
To measure convergence, I take the time spread between the two countries,
where takeoff occurs first and last, in any single product category. I then plot the
time spread of the product category across any one of two scales: the year of first
takeoff or the year of first commercialization for the respective product category. If
convergence is occurring, then the curve should slope downwards over time. If
divergence is occurring, then the curve should slope upwards over time. If neither is
occurring, then the curve should be flat.
Since this measure requires takeoff to have occurred, I do not include
countries where takeoff has not occurred. In the interest of consistency, I also need to
include the same set of countries in each category. As a result, for this analysis, I can
consider only 14 product categories in 18 countries. The countries are Japan, US,

132
Canada and 15 countries of Western Europe. I include all the products in our sample
except MP3 players and hand-held computer (I have data only till 2003 for the
former and not for all the countries for the latter). This sample covers 246
observations.
The results are in Figures 6 and 7 with year of first takeoff and first
commercialization respectively as the X-axis. Figure 6 shows a dramatic, downward,
almost linear plunge over time indicating a strong convergence in time-to-takeoff.
The time spread between the first and last takeoff drops from over 50 years in 1950
to 5 years in 2000. A simple regression of time spread on year of first takeoff yields
a coefficient that is negative and significantly different from zero (T stats of -6.6, R
2

= 0.78). Figure 7 shows a similar pattern. A regression of time spread on year of first
commercialization also yields a coefficient that is negative and significantly different
from zero (T stats of -5.1, R
2
= 0.68). The sharp increase in the first few periods of
both graphs could be due to the negative effects of the World Wars and the
Depression, though I cannot draw any firm conclusions because of the small sample
size.

133
Figure 6: Time Spread by Year of First Takeoff
0
5
10
15
20
25
30
35
40
45
50
55
1930 1935 1940 1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005
Year of first takeoff
Tim e spread in takeoff in years
Time spread
Linear (Time spread)

Figure 7: Time Spread by Year of First Commercialization
0
5
10
15
20
25
30
35
40
45
50
55
1905 1910 1915 1920 1925 1930 1935 1940 1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000
Year of first commercialization
Time spread in takeoff in years
Time spread
Linear (Time spread)

134
3.7 TESTS OF ROBUSTNESS
Apart from examining different distributional assumptions, two tests of
robustness on the baseline hazard and alternate measure of takeoff are carried out.
3.7.1 Baseline Distribution
I considered several alternate baseline distributions such as the Log-Normal,
Log-logistic, Exponential, Weibull, and Gamma, each yielding a different parameter
estimates of the Hazard model. In order to determine the best distribution function, I
compare non-nested models using the SBC (Schwarz's Bayesian Criterion), as
suggested by Allison (1995), Srinivasan, Lilien and Rangaswamy (2004), and Pliner
(2005). SBC is calculated using the formula: -2*log-likelihood + k*(# of parameters),
where k = log (n), with n representing the number of observations. Lower values of
SBC indicate better fit. I find that the Weibull model generally outperforms the other
models using the SBC criteria, with a SBC value of 637.9 for Model 13 and 625.37
for Model 14.
3.7.2 Measure of Takeoff
I use an operational measure (achieving 2% penetration) to determine the
year of takeoff. I evaluate the robustness of our results using a sub-set of data where
I can demonstrate that takeoff actually occurred.
For 187 product-country combinations in our original data set, I was able to
collect both sales and penetration data. These include established categories such as
work products, CD player and PC for developed countries (89 observations) and new

135
categories such as DVD player, Digital camera, MP3 player and Hand-held computer
where I have data for both developed and developing countries (98 observations).
For all of these product-country combinations, I compare the year of takeoff as
measured by our 2% penetration rule to the year of takeoff as measured by the rule
proposed by Tellis, Stremersch and Yin (2003).I find that overall, in 87% of the
cases, the absolute differences in the year of takeoff between the two rules are less
than or equal to two years, while they match exactly in 37% of the cases (Table 12).
Table 12: Comparison of 2% Rule and Threshold Rule of Takeoff
Variable Estimates Prob> Chi
square
Variable Estimates
Total 65 67 34 24 190
% 34 35 18 13
Cumulative % 34 69 87 100

For 158 product-country combinations in the OECD countries alone, I use the
Tellis, Stremersch and Yin (2003) rule to examine the mean penetration at takeoff. I
find that the mean penetration at takeoff is 2%, which adds further validity to using
the 2% rule.
Thus, our rule has the advantages of being simple, consistently applied across
all categories and countries, and relatively similar to that proposed by Tellis,
Stremersch and Yin (2003). In the absence of adequate data, following this rule
seems a good alternative to the option of dropping those categories for which I do
not have adequate data.

136
3.8 DISCUSSION
This section summarizes the key findings, discusses questions and
implications of findings, and lists limitations of the study.
3.8.1 Summary
This study leads to several new findings:
• Time-to-takeoff is getting shorter over calendar time.
• In addition, there is strong convergence in time-to-takeoff over calendar
time among developed countries.
• Despite these two effects, differences across countries are quite strong.
o Products takeoff fastest in Japan and Norway, followed by other
Nordic countries, US and some countries of Mid-western Europe.
o Newly developed countries of Asia (e.g., South Korea) see faster
times-to-takeoff of products than established, major European
countries (e.g., France, Italy) with centuries of industrialization.
o Latin countries across Europe and South America have similar
times to takeoff.
o Despite the recent and rapid increases in the GDP of emerging
markets such as China, India, Philippines, these countries still
substantially lag other countries in time-to-takeoff of new
products.

137
• Takeoff is not a purely cultural phenomenon. Differences in both
Economics (Wealth) and Culture (In-group collectivism) account for
differences in time-to-takeoff across countries and regions.
• The mean time-to-takeoff varies considerably between developing
countries (11 years) and developed countries (7 years). The mean time-to-
takeoff varies between 6 and 12 years across cultural clusters.
• Time-to-takeoff varies considerably between fun products (7 years) and
work products (12 years).
o Fun products takeoff substantially faster than work products
within each cultural cluster.
o Takeoff of fun products also shows smaller differences across
cultural clusters than work products do.
o Takeoff of fun products is driven more by dynamic economic
variables and is be converging faster over time than work products.
3.8.2 Questions
These findings raise three important questions.
First, can time-to-takeoff serve as an indicator of the innovativeness of a
country? Researchers across disciplines and global policy makers have long debated
which countries rank high on innovativeness (The Task Force on the Future of
American Innovation Report 2006). Prior research has measured this innovativeness
either by input measures such as R&D and scientific talent (e.g. Furman, Porter and

138
Stern 2002) or by surveys of consumers (e.g. Steenkamp, Hofstede and Wedel 1999).
However, an alternate viewpoint holds that innovativeness is better defined by the
willingness and ability of consumers to acquire and use new products and
technologies (Bhide 2006; Tellis, Stremersch and Yin 2003). When based on hard
data, such a measure of innovativeness is also less prone to self-report and cultural
biases as is survey data. I find significant and robust differences across countries in
the times-to-takeoff. These differences persist within classes of products and across
time. Thus, they could serve as a metric of the innovativeness of the nation itself.
However, when doing so, we need to keep in mind that the differences in time-to-
takeoff, and hence innovativeness are due to both wealth and culture of the country.
Second, why does Japan not fit in with the cultural cluster of Confucian
Asia? Cultural clusters explain differences in times-to-takeoff across countries, with
one notable exception, Japan. It has the shortest time-to-takeoff even though it is
sometimes grouped in the Confucian cluster, which shows slow times to takeoff. I
propose one explanation for this anomaly, consumerism. Consumerism has been
defined by Stearns (2001) as a societal trait in which many people formulate their
goals in life partly through acquiring goods that they clearly do not need for
subsistence or for routine appearance. Authors claim that consumerism has
flourished in Japan due to a combination of factors: a major thrust by the government
to promote product development and consumption, a strong native desire of the
Japanese to produce and own the best products, investment in new products rather
than land (which is scarce) as symbols of economic progress, and a broader

139

admiration of Western (materialistic) values (Stearns 2001). In Japan, modern
consumerism may have overwhelmed older Confucian values, leading to one of the
most aggressive and dynamic markets for consumer goods
6
. Unfortunately, scales for
this construct are unavailable across all countries, so I could not test this explanation.
Third, manufacturers have not introduced major new work products (except
microwave oven) recently, whereas they have introduced a large number of fun
products. So, are the distinctions between work and fun products indistinguishable
from that between older and newer vintage products? An examination of Table 4 can
address this issue. Note that more fun products that work products have taken off in
developing countries, even though they have been introduced more recently. Thus,
the distinction between fun and work products seems intrinsic to these products
rather than a function of product vintage.
3.8.3 Implications
The study’s findings have the following strategic and research implications.
First, researchers have debated the merits of a waterfall strategy (staggering
the commercialization of new products across countries) versus a sprinkler strategy
(simultaneously introducing the new product across countries). For instance,
Chryssochoidis and Wong (1998) and Gielens and Dekimpe (2001) argue for a

6
In the 1950s and 1960s, Japanese consumers referred to the three Ss as major life
goals: senpuki, sentakuki and suihanki (fan, washing machine, and electric rice
cooker). This was followed by the three Cs, in the late 1960: ska, kura, kara terebi
(car, air conditioner, and color TV), and by the Js in the late 1970s - jueru, jetto,
jutaku– jewels, jetting and a house”.

140
simultaneous launch to minimize product failure risk due to delayed roll outs and
competitive environments. However, Kalish, Mahajan and Muller (1995) argue that
conditions such as long product life cycles, small size or slow growth of a foreign
market make a waterfall strategy more preferable. Mitra and Golder (2002) suggest
that firms enter countries where they have greater economic and cultural knowledge
based on operating in similar other countries. Tellis, Stremersch and Yin (2003)
argue that a waterfall strategy greatly reduces the scale of operation and exposure to
risk of product failure, and increases senior management support when takeoff
occurs quickly in the most innovative countries.
We believe market strategy should depend considerably on the type of
products. Because times-to-takeoff of fun products are more similar across countries
and converging faster over time than that for work products, they probably have a
universal appeal across cultures. Hence, a sprinkler strategy might be feasible for fun
products. However, work products are culturally bound and adopted in some cultures
more readily than in others. In such categories, a waterfall strategy might be more
profitable. By introducing first in countries or cultural clusters where the products
are more conducive to the culture, product managers can lower risk, increase odds of
success, win support of senior management, and use the confidence, revenues, and
lessons gained from those countries and regions to market the product in less
accepting countries. In this respect, even small differences in times-to-takeoff of 1 to
3 years may represent enough real time differences to execute a waterfall strategy.

141
Second, should one choose a waterfall strategy, authors have debated about
which countries are the best to introduce a new product first. I recommend one of
two sets of strategies. If a manager wishes to launch a new product in an innovative
and large market, then the best countries to launch in would be the Japan or the US.
However, if a manager wishes to test-market in a small but highly innovative country,
then the best countries to launch in would be in the Nordic cluster, Switzerland or
Netherlands. In addition to these countries, South Korea also shows promise as a
relatively small country with a relatively short time-to-takeoff of new products. For
example, it leads the world in penetration of Broadband and 3G technologies.
Third, in addition to country innovativeness, managers need to consider the
economics of scale, especially between marketing to giants such as China and India
and to small countries such as Norway. For example, Cellular phone subscribers are
growing by 6 million a month in India in 2006 (The Economist). The annualized
sales of Cellular firms in countries like India and China would dwarf the entire
population of most European countries. The issue of scale becomes especially
critical in conjunction with population concentration. If one country’s adopters are
concentrated in a small easily accessed portion of the country and yet another
country’s adopters are dispersed more widely, then the former may be a superior
option to the latter.

142
3.8.4 Limitations and Further Research
Five obvious limitations suggest areas for future research. First, due to data
limitations, I use a heuristic of 2% to measure the point of takeoff. However, I use
this rule consistently across a wider variety of countries and categories, and similar
heuristics have been used in prior literature. Second, I do not account for the role of
variables such as price declines, quality improvements or competition within product
markets (Agarwal and Bayus 2002, Golder and Tellis 1997, Jain, Mahajan and
Muller 1991, Mahajan, Muller and Bass 1995, Van den Bulte 2000). Third, I do not
consider differences in time-to-takeoff within a country. Fourth, there is a high
amount of correlation among groups of variables. However, I mitigate the effects
somewhat by considering Wealth as a factor of related dimensions. Fifth, an
extension of this study to products other than consumer durables and high-tech
services will lead us to a better intuition about the phenomenon of takeoff.

143

CHAPTER 4: GETTING A GRIP ON THE SADDLE: CYCLES,
CHASMS, OR CASCADES?
7
4.1 INTRODUCTION
A pattern not commonly considered in diffusion literature is the phenomenon
of a ‘Saddle’, a pattern of a sudden decline in sales of a new product following a
rapid growth phase. A product here refers to all categories that are close substitutes,
fulfilling a distinct need, for example DVD player or Personal Computer. The term
Sales refers to unit sales for the product category.
Golder and Tellis (2004) find a period of level, slowly increasing, or
temporarily decreasing product category sales in 22 out of 23 product categories in
the U.S. market. Goldenberg, Libai and Muller (2002), who term this phenomenon
the ‘Saddle’, find it in 50% of the 32 product categories that they examine.
Researchers had as early as in the 1970s proposed the possibility of such a
sharp slowdown in sales (Cox 1967; Wasson 1978). However, for many decades,
there was no substantial empirical evidence to support this pattern of slowdown (Day
1981). Nearly twenty years later, two papers have found some empirical evidence of
this phenomenon of a sudden decline in sales following a rapid growth phase
(Goldenberg, Libai and Muller 2002; Golder and Tellis 2004).

7
This essay forms the basis of a working paper in the same name co-authored with
Gerard J. Tellis

144
The compelling new evidence indicates a clear need to revisit and rethink the
phenomenon of the product life cycle. The pattern of a sharp decline averaging 15-
20% in sales and lasting on an average more than 2 years has grave consequences for
new product marketers. Managers, who believe the phenomenon to be isolated to
their product alone, may, on seeing such a sharp deviation in trend from the expected
bell-shaped curve, pull the plug on new products they think are performing poorly.
Or their processes may be inadequately geared to meeting a sudden slowdown in
sales, and to the subsequent growth in sales (Chandrasekaran and Tellis 2006).
Goldenberg, Libai and Muller (2002) coined the term “Saddle”, and define it
as a pattern in which an initial peak predates a trough of sufficient depth and duration
to exclude random fluctuations, which is followed by sales eventually exceeding the
initial peak. A limitation of their study is that they cannot give an indication of the
timing of the Saddle. We define the Saddle more parsimoniously as the ‘first trough
in sales after the post takeoff peak’ (See Figure 8). Linking the phenomenon of the
Saddle to the concept of the Takeoff has two advantages. One, it indicates that the
phenomenon occurs after the product sees a rapid increase in sales, and is not an
indication of a product that failed in the market place. Two, empirically, it can help
identify consistent patterns of the Saddle across different products and countries.
Figures 9 and 10 illustrate the presence of a Saddle across products in different
categories.

145
Figure 8: Saddle in New Product Sales

Figure 9: Saddle in Microwave Oven Sales in UK
Microwave Sales in UK
0
500
1000
1500
2000
2500
1960 1970 1980 1990 2000 2010
Year
Sales in 000 Units

Slowdown
Time
Sales
Takeoff
The Saddle
Introduction

146
Figure 10: Saddle in PC sales in USA
Personal Computer Sales in US
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
1977 1979 1981 1983 1985 1987 1989 1991 1993 1995
Year
Sales in '000 units
PC Sales

The aim of this research paper is to help develop a better understanding of the
phenomenon of the Saddle, given the paucity of literature in this area. The following
questions are examined. One, how pervasive is the Saddle across products and
countries? Two, why does a Saddle occur?
Academically, this paper hopes to lend fresh insights into the concept of the
product life cycle, by examining a new and interesting phenomenon. Managerially,
this paper hopes to help product planners prepare for a sudden downturn in sales,
immediately after they have attained certain initial expectations of rapid growth.

147
4.2 THEORY AND HYPOTHESES
This section proposes to examine potential drivers of the Saddle: Chasms in
Adopter Segments, Information Cascades, Business Cycles, Technological Changes
and Repurchase Lags.
4.2.1 Chasms in adopter segments
The product life cycle has been viewed as a social phenomenon by various
studies on the adoption cycle or the diffusion process. Rogers (1995) classified the
adopters in the diffusion process into five distinct groups. The innovators comprise
the first 2 to 3 % of the market, and are people who are inclined to seek out and
pioneer the use of new products. The early adopters comprise the next 12 to 15% of
the adopters. They are the opinion leaders, and their expertise is respected by the
others, and hence they become the major source of influence for the buying
population. The early majority are those who increasingly depend on the early
adopters for trial and legitimization of the offering and comprise the next 34% of
adopters of the new product. The late majority are the third of the market that lag
behind in adoption of the product. The skeptics form the final 12 to 15% of the
market who adopt the product as it reaches saturation. Traditionally, it has been
thought that the diffusion process can be fuelled by simply targeting the innovators
and the early majority, triggering word of mouth communication across the other
adopter segments (Rogers 1995, Mahajan and Muller 1998). However, some
researchers debate this notion of continuity. Mahajan and Muller (1998) posit

148
conditions wherein the majority should be treated differently from the early market.
Moore (1991) suggested that there is a ‘chasm’ between the early market dominated
by a few visionary customers and the main stream market dominated by a large
block of pragmatic customers (Figure 11)
Figure 11: Chasms in Adopter Segments

Late market (64%
of Total Market)
Early market
(16% of Total
Market)
Chasm
Goldenberg, Libai and Muller (2002) argue that this weak communication
between the early market and a late market may lead these markets to adopt products
at differing rates, creating a Saddle (See Figure 12).

149
Figure 12: Saddle due to Chasms
Total market
Main market
Early market

H1: Saddles are more likely to occur during the presence of a chasm in
adopter segments than in the absence of a chasm in adopter segments
4.2.2 Informational Cascades
Individuals make behavioral choices by either direct examination of the
alternatives, which may be costly and time-consuming; or rely on the information of
others. Informational Cascades occur when individuals, who have similar
information, face similar decision problems, alternatives and payoffs, make similar
choices (Bikhchandani, Hirshleifer and Welch 1998, 1992).
Information cascades refer to the tendency of a market to snowball onto a
popular behavior (See Figure 13)

150
Figure 13: Information Cascades in Consumer Markets

Initial consumers adopt
based on utility

In consumer markets, initially consumers may carefully consider alternatives
while making product choices. Their adoption gives a signal to the rest of the market.
New consumers may make evaluations based both on their information, as well as on
signals provided by the market. Their actions provide a further signal to the market.
Soon, consumers may increasingly rely on signals provided by the actions of earlier
adopters rather than making a careful evaluation of alternatives. This triggers a
positive cascade in sales. Such cascades can lead to an acceleration of sales.
In Figure 14, the smooth curve shows the sales as would have been
determined on the basis of utility alone. The dotted line shows acceleration in sales
triggered by a positive cascade. However, cascades are extremely fragile as they are
based on very little information. A positive cascade can be reversed by new
Signal
New consumers adopt primarily
due to influence of signal
More consumers follow suit
strengthening signal
Stronger
Signal
Positive
Cascade

151
information that affects the personal preferences and decisions of a small number of
people (Golder and Tellis 2004). Any new public information that suggests a
different optimal course, or creating a suspicion among people that circumstances
have changed, may demonstrate the inherent fragility of the cascade. Any shock may
thus be accompanied by a deceleration in sales triggered by a negative cascade. The
effect of the negative cascades may wear off in time and sales based on utility picks
up, leading to a recovery.
Figure 14: Saddle Due to Negative Cascades
Shock
Deceleration in sales due to
Hence, I hypothesize,
H2: Saddles are more likely to occur during the presence of negative
informational cascades than during an absence of information cascades.
4.2.3 Business Cycles
Much of marketing research has concentrated on the positive aspect of
income changes on diffusion and market potential (Golder and Tellis 1998, Talukdar,
Acceleration in
sales due to Positive
cascade
Negative cascade
Utility-based
sales curve
Saddle

152
Sudhir and Ainslie 2002). However, even if product sales are growing rapidly,
reversals in the economy would lead to a Saddle in sales (Golder and Tellis 2004).
Figure 15 shows a slowdown in sales in VCR sales in USA occurring during
a time of economic recession.
Figure 15: Saddle and Recessions
.
Sales slowdown
during a Recession
0
500
1000
1500
2000
2500
3000
3500
4000
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
Recession
VCRs

Researchers in marketing have recently paid more attention on the impact of
business cycles on product sales (Deleersnyder et al 2004, Lamey et al 2007).
Recessions may lead to a Saddle in product sales for three reasons: One, recessions
shrink income and depress buying power. Hence, consumers are more likely to cut
back on discretionary expenditure such as spending on consumer durables during
times of recessions, and postpone these purchases to times of expansion; two,
consumers lose trust during recessions more quickly than re-acquiring confidence
during the following expansion, leading to cyclical asymmetries in the purchase of

153
durables and three, firms may engage in counter cyclical pricing or cut expenditures
on advertising (Bachus and Kehlow 1992; Deleersnyder et al 2004, Lamey et al 2007,
Woodford and Rottemberg 1998), which may aggravate the sensitivity of consumers
during recessions depressing sales even further.
Hence, I hypothesize,
H3: Saddles are more likely to occur during times of recessions than during
times of expansions
4.2.4 Control Variables
There are three important control variables: Time since takeoff, technological
cycles and replacement cycles.
Time since Takeoff
The greater the time since the product took off, the more likely it is to expect
a Saddle in product sales.
Technological Changes
During times of technological change, consumers may engage in
leapfrogging behavior by waiting for the next-generation product, or engage in
purchase postponement in expectation of product improvements (Weiss and John
1989, Weiss 1994). Hence, the greater the level of technological changes the more
likely it is to expect a Saddle in product sales.

154
Repurchase Lags
Product sales volume are often thought to be composed of two elements-
initial purchases and replacement purchases (Harrell and Taylor 1981, Midgley
1981). Typically, in the early stages of the product life cycle, the initial purchases
constitute the majority of sale volume, however as saturation approaches, the
replacement component becomes dominant. Hence, we may expect a Saddle in total
sales to occur if first purchases are on a decline and repurchases have not taken off
4.3 METHOD
This section describes the model, measures, data and sources.
4.3.1 Model
I use a discrete time hazard model to determine the drivers of the occurrence
of the Saddle. In the discrete time approach, the event history is modelled as a series
of independent success/failure trials. The unit of analysis is a category country year.
The first year for each category country combination in the discrete time analysis is
the year after takeoff, and the category country combination is observed every year
thereafter till the event either occurs or the observation is censored.
Each category i in country c has J
ic
observations, one per year of risk. An
Event indicator Event
icj
indicates whether a Saddle occurred in year j for a category i
in country c (where 0=no event, 1=event). The probability that category i in country
c experiences the event in year j given no prior event occurrence is h (t
icj
) such that

155
[ ] ) .. ( exp 1 / 1 ) (
2 2 1 1 picj p icj icj j icj
x x x t h β β β α + + + + − + = (1)

The hazard of the event occurring is hence a function of the baseline hazard
term α
j
as well as the covariates x
icj
. A polynomial function of α
j
is considered such
that
2
3 2 1
time time
j
α α α α + + = (2)
where time refers to the time since Takeoff to the observation year. The
other covariates that we consider are Recession in year j, Chasm in Adopter Segment,
Negative Information Cascade. We also examine the impact of two control variables-
Repurchase Lags and Technological Changes.
In the year when the event is experienced (when event =1), each category-
country combination contributes h(t
icj
) to likelihood function. In all time periods
when event is not experienced (when Event=0), it contributes 1-h(t
icj
) to likelihood
function. The likelihood function for the model is therefore:
() ()
) 1 (
11 1
1 ) (
icj
ic
icj
Event
icj
n
i
C
c
J
j
Event
icj
t h t h L
−
== =
− =
∏∏∏
(3)

Standard logistic regression applied to this category-period data set provides
estimates of the parameters that maximize the log-likelihood (Singer and Willett
1993, Allison 1982)

156
4.3.2 Measures
In the following section, we describe the measures for the year of takeoff, the
year of Saddle, business cycles, chasms in adopter segments, negative cascades and
the control variables.
Year of Takeoff
The year of Takeoff is the first year in which an individual category’s growth
rate relative to the base sales crosses a threshold. We use the threshold rule followed
by Stremersch and Tellis (2004) and Tellis, Stremersch, and Yin (2003).
The threshold is a standard plot of growth in sales across products and
countries for various levels of market penetration to provide for a standard
comparison across several countries. Figure 16 shows a graph of the threshold in
sales as adapted from Tellis, Stremersch and Yin (2003).
Figure 16: Threshold Rule for Takeoff

Threshold Rule
0
100
200
300
400
500
600
700
0. 1
0. 3
0. 5
0. 7
0. 9
1.1
1.3
1 .5
1 .7
1. 9
2. 1
2. 3
2. 5
2. 7
2. 9
3.1
3 .3
3. 5
Penetration
S a l es G ro w th R a t e

157
Year of Saddle
Following Goldenberg, Libai and Muller (2002), the year of Saddle is the
first year after takeoff when the sales declines by 10 % or more from the initial peak
and takes more than 2 years to cross the initial peak. This is depicted in Figure 17.
Figure 17: Identification of Year of Saddle

finland_dishwasher
-
20
40
60
80
100
120
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005
Year
Unit sales
Year of
Saddle
Business Cycles
For studying the impact of business cycles on product sales, we have to first
measure them. The question then is how to isolate the cyclical component of the
GDP series i.e., separate cyclical elements from slowly evolving trends or rapidly
varying seasonal or irregular components (Baxter and King 1999). Baxter and King
(1999) treat U.S. business cycles to be cyclical components of duration no less than 6
quarters (2 years) and no more than 32 quarters (eight years), and construct an
approximate band pass filter, that passes through components of a time series
between 6 and 32 quarters, while eliminating components at lower and higher

158
)
frequencies. For annual data, Baxter and King (1999) recommend the following
filter:
( 0510 . 0 ) ( 1351 . 0 ) ( 2010 . 0 7741 . 0
3 3 2 2 1 1 + − + − + −
+ − + − + − =
t t t t t t t t
y y y y y y y c (4)
where y
t
is the original GDP series (in logs) in year t, and c
t
is the cyclical
component used in the analysis.
In the first step in this analysis therefore, I first use the Band pass filter (using
EViews software) to isolate the cyclical component of the GDP series. This is similar
to the procedure followed by Deleersnyder et al (2004).
Figure 18 illustrates an example of a cyclical component of a GDP series.
Figure 18: Cyclical Component of GDP Series for Austria

-0.05
-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
1950
1952
1954
1956
1958
1960
1962
1964
1966
1968
1970
1972
1974
1976
1978
1980
1982
1984
1986
19

159
Decreases (increases) in the cyclical component of GDP correspond to
contractions (expansions). In the second step therefore, I create an indicator variable
for the state of the economy. I label the period from a peak in the cyclical component
of GDP to the following trough as a business cycle recession, and the period from
the trough to the next peak as a business cycle expansion.
Negative Cascades
There is no known empirical measure of negative information cascades. I
have data at an aggregate cross-country level. I measure the impact of negative
cascade by the number of Saddles that occur in the same year or the year before in
other countries.
Is this valid? Consider the example of microwave oven sales in Europe. In
1989, newspaper reports from Europe reported on the results of a British government
study on 70 brands of microwave ovens that were tested for safety. It was reported
that a third of the oven brands it had tested left dangerous "cool spots" in food but
the government refused to tell consumers which were dangerous. The British
government also revealed that commercial caterers were risking their customer's
health by using domestic ovens instead of industrial ovens. At the same time,
Austrian research had discovered potentially dangerous molecular changes in baby
milk which had been heated in a microwave. If we look at the aggregate sales data at
the same time, we see a slowdown in sales 1989 in Finland, France, Germany,
Sweden, and UK and in 1990 in Austria, Ireland, Norway and Switzerland.

160
Does this pattern hold systematically across our sample? After controlling for
the economy, a coefficient that is positive and significantly different from zero in the
hazard model would imply that negative signals from other markets lead to a
slowdown in sales in the focal market.
Chasms in Adopter Segments
There are no existing measures in the literature for chasms in adopter
segments. However, there are heuristics that point at measures for the chasm.
From the earlier discussion on adopter categories above, the likely point of
transition from the early adopters to early majority occurs at a penetration level of
15-20% (Rogers 1995, Mahajan, Muller and Srivastava 1990, Mahajan and Muller
1998). If chasm explanation is correct, Saddle typically occurs at tight range of
penetration. This could be around 16% but may vary depending on the category. As
seen from the descriptive analysis later, there are 109 country-category combinations
for which the Saddle occurs. The mean penetration at Saddle for these 109
observations is 27.3. If the chasm explanation for the Saddle is correct, then we
should expect to see a Saddle occurring in a very tight range around 27.3%. To
determine if this is true or not, we develop a measure of dispersion from the mean
penetration at Saddle in the following manner: for each category country year we
determine the deviation of the actual penetration from the mean penetration at Saddle.
) 3 . 27 ( − =
icj icj
n Penetratio Abs Deviation (5)

161
If the driver of the Saddle is indeed the occurrence of a chasm in adopter
segments, then we may expect that the Saddle occur at a tight range of market
penetration. Hence, we would expect a positive coefficient in the hazard model for
the measure Deviation
icj.
In developing this measure, I am making the assumption that

the chasm or a
lack of communication among adopter segments occurs at a specific range of market
penetration, based on the definitions of adopter segments (Rogers 1995, Mahajan,
Muller and Srivastava 1990, Mahajan and Muller 1998). I test the alternate
assumption that the division of adopter segments may vary by product class. Hence, I
test the effect of an alternate measure wherein
) 7 . 18 ( 2 − =
icj icj
n Penetratio Abs Deviation if product class = 0 (fun products
alone) (6)
) 4 . 30 ( 2 − =
icj icj
n Penetratio Abs Deviation if product class = 1 (work products
alone) (7)

An argument may be made that the chasm could however occur at
penetration levels much higher or much lower than those considered here. However,
if this were indeed true, then the theory is either no longer valid or is a theory that
can never be tested.

162
Control Variables
Patent information gives a rich indication of technological activity in an
industry- specifically on trends of R&D, competitive and complementor activity,
important innovations in the field. Prior research has hence used patent counts as a
measure of innovation and technological improvements in an industry (Ahuja and
Katila 2001; Council of Competitiveness 2001; Narin, Noma, and Perry 1987;
Prabhu, Chandy and Ellis 2005; Wu 2005).
We use two measures of technological changes in an industry. Patent Count
is an indicator of the absolute level of technological activities in an industry. I
measure the item Patent Count

as the sum of all US granted patents for each product
category i in each year j. We assume this to be a valid indicator of technological
activity for the product category across countries since the most comprehensive
patent data is available for US patents, and we assume that all important innovations
will have secured US patents.
Patent Importance is an indicator of the importance of technological activities
in an industry. A patent that is frequently cited, i.e. that has a number of forward
references is another indicator of the level of technological development in an
industry (Wu, 2005). Hence, I also consider the ratio of total number of citations
(forward references) to total number of patents granted in a year j for each product
category i to create a second measure termed Patent Importance
.
Since sales data consist of both first purchase and repeat purchase sales, a lag
between first purchase and repeat purchase can lead to the formation of a slowdown

163
in sales. I create a measure called Repurchase Lag to account for the impact of the
timing of repurchase cycles on the hazard of occurrence of a Saddle. Repurchase Lag
is the difference between the time to repurchase and the time to decline in first
purchase sales. I obtain the measure for time to repurchase from various issues of the
Appliance magazine for work products (9 years for dishwasher, 10 years for
microwave oven, 12 years for freezer, 13 years for dryer and washing machine, and
15 years for refrigerator), and published papers (Stremersch and Tellis 2004) for
personal computer (4 years) and VCR (6 years). I measure the time to decline of first
purchases as the time to reach a 50% market penetration in a category. The measure
repurchase lag is the same for all years in a product-country combination, but varies
across products and countries.
A positive coefficient of Repurchase Lag in the hazard model would imply
that a delay in repurchase demand (Time to repurchase > Time to first purchase
peak) is leading to a temporary slowdown in sales.
For instance, Figure 19 compares the time to reach 50% penetration (bars)
with the time to repurchase (area curve) in ach product-country combination. In 61
observations (product-country combinations) alone, the product penetration has
crossed 50%. If the Time to reach 50% penetration is less than the repurchase time,
as in the first 11 cases, then this implies that there may be a delay in repurchase
demand which may lead to a Saddle in total sales. However, in the majority of the
cases, the time to reach 50% penetration is far greater than the repurchase time.

164
Figure 19: Repurchase Lags in Product Sales
Comparison of Repurchase Time with Time to 50% penetration
0
10
20
30
40
50
60
70
80
90
100
1 6 11 16 21 26 31 36 41 46 51 56 61
Observation
Years
repurchase time
timeto50%

4.3.3 Data and Sources
I assemble a database on historical sales and market penetration of 6 work
products and 6 fun products across 16 European countries, US and Japan. I use the
term Work Products for products that help consumers work efficiently, such as
microwave ovens, washing machines, and dryers. I use the term Fun Products for
those products that provide consumers with information, pleasure, or enjoyment,
such as personal computers and DVD players. The former groups have also been
called time-saving and the latter time-consuming products respectively (Horsky
1990).

165
I put together this time-series data from a variety of sources: subscription-
based sources (Euromonitor Global Marketing Information Database, World
Development Indicators Online, Fast Facts Database), archival search through freely
available secondary sources (Historical Statistics of Japan, Historical Statistics of
Canada, Electrical Merchandising, Merchandising, Merchandising Week and
Dealerscope journals for US, OECD statistics), and proprietary data (Database of
Tellis, Stremersch and Yin 2003), and data provided by Arvind Rangaswamy over
several hundreds of research hours.
GDP has historically been a good proxy for overall economic activity
(DeLong and Summer, 1986b)
8
. For calculating the time of business cycle recessions
across countries, I use a measure for Real GDP from the Groningen Growth and
Development Centre and the Conference Board, Total Economy Database, January
2005, (http://www.ggdc.netis). I use Total GDP, in millions of 1990 US$ (converted
at Geary Khamis PPPs).
The Delphion database (www.delphion.com) is a subscription-based database
that contains detailed historical records on patents granted and applied for both in the
US and other countries. For each product category, I create queries to extract all US
granted patents containing the name of each product category- for instance, personal
computer or microwave oven under ‘Title, Claim and Abstract’. This gives a list of
all the patents applicable to each product category over time. I use only US granted

8
See The Economist, “How to Measure Economies”, Feb 9th 2006 for a critique of
using GDP and GDP per capita as measures to reflect the economy

166
patents as this gives the most comprehensive time series patent data (the earliest date
is 1968 and the latest is 2003).
I obtain measures of replacement cycle from various issues of the Appliance
magazine for work products (9 years for dishwasher, 10 years for microwave oven,
12 years for freezer, 13 years for dryer and washing machine, and 15 years for
refrigerator), and published papers (Stremersch and Tellis 2004) for personal
computer (4 years) and VCR (6 years), as mentioned before.
4.4 RESULTS
4.4.1 Descriptive Statistics
There are 160 product country combinations where a product takeoff has
occurred as assessed by the measure proposed by Tellis, Stremersch and Yin (2003).
The mean time-to-takeoff across these cases is 7 years, with a mean penetration of
1.8%. The mean sales growth at takeoff is 200%.
110 products that took off, experience a Saddle (65%). Table 13 depicts the
number of occurrences of a Saddle in the different products in our data-set. There are
no Saddles in 3 categories- DVD Player, Digital Camera and MP3 Player. Other fun
products- VCR, PCs and MD Player, as well as all the work products see high
occurrences of the Saddle. I will explore this discrepancy further in the next section.

167
Table 13: Saddle Occurrence by Products
Product No Saddle Saddle Total
DVD Player 14 0 14
Digital Camera 16 0 16
Dishwasher 1 14 15
Dryer 3 13 16
Fridges 0 9 9
MD Players 1 11 12
MP3 Player 9 0 9
Microwave 4 13 17
PCS 0 7 7
VCR 0 13 13
Washer 1 15 16
Freezer 1 15 16
Total 50 110 160

Figure 20 depicts the average time to a Saddle from the point of takeoff. This
varies from 6 years in the case of fun products to 10 years in the case of work
products. Figure 21 depicts the depth of the Saddle which is the drop in sales from
the initial peak to the bottom of the trough. This is around 25- 27% for work and fun
products. Figure 22 depicts the duration of the Saddle, or the time from Slowdown to
Recovery. This is between 7 and10 years for fun and work products.
Figure 20: Time from Takeoff to Saddle by Product Class

0
2
4
6
8
10
12
Fun products Work products
Product Class
Years
Mean
Std

168

Figure 21: Depth of Saddle by Product Class
Depth of Trough
0
5
10
15
20
25
30
Fun products Work products
Product Class
% Overall Decline from
Initial Peak
Mean
Std

Figure 22: Duration of Saddle by Product Class
Duration of Trough
0
2
4
6
8
10
12
Fun products Work products
Product Class
Number of Years
Mean
Std

Table 14 shows descriptive statistics on occurrence of the Saddle by countries.
The proportion of occurrence of the Saddle in products across countries varies from
30% in Greece to 80% in Italy, UK, Germany, Belgium and Ireland. The average
decline in sales in the year of Saddle is 18% but varies from 5 % (Spain) to 25%
(Denmark) across the countries.

169
Table 14: Comparisons across Countries

Proportion of
Saddle
Time from
Takeoff to
saddle
Decline rate
at saddle
Duration
of saddle
Depth of
saddle
Austria 0.7 10.3 12.2 7.7 -17.7
Belgium 0.8 11.0 9.7 9.8 -25.3
Denmark 0.7 5.6 24.8 7.9 -42.5
Finland 0.7 10.6 15.012.1 -27.9
France 0.7 7.2 9.110.0 -18.6
Germany 0.8 10.5 6.8 8.5 -15.0
Greece 0.3 11.0 23.917.5 -29.4
Ireland 0.8 13.5 10.2 7.3 -29.5
Italy 0.8 8.1 17.210.4 -28.7
Japan 0.7 12.0 9.67.0 -12.0
Netherlands 0.6 7.0 10.7 11.2 -32.2
Norway 0.7 7.2 13.4 6.8 -19.2
Portugal 0.6 10.4 18.4 6.8 -29.8
Spain 0.6 14.3 4.66.7 -20.9
Sweden 0.7 11.0 11.2 7.2 -14.5
Swiss 0.5 9.5 15.412.5 -10.0
UK 0.8 8.3 14.711.3 -31.9
US 0.610.2 18.48.3 -33.3

The duration of the growth stage (or the time to Saddle) varies from 6
(Denmark) to 14 years (Spain). The average depth (this is the fall in sales from the
initial peak to the lowest point of the trough) is -26%. This varies from -15% in
Sweden and Germany to -42% in Denmark, with the US showing an average depth
of -33%. Golder and Tellis (2004) find a comparable number of -25% in their
analysis of 30+ categories in the US. The duration of the Saddle or the average time
to recovery is 7 years. Note however that there are no distinctive or consistent
country or cultural cluster specific patterns as observed in Chapter 3 comparing time-
to-takeoff of new products.

170
4.4.2 Logistic Regression
In over 30% of the observations, a Saddle has not yet occurred. Does the
occurrence of a Saddle depend on product class, region or product age? I run simple
logistic regressions on the probability of the occurrence of a Saddle (a 0/1 indicator)
on product class, region, and age of product
I find that consumer electronics are less likely to see a Saddle than home
appliances (p<0.0001), with an R Square value of 0.21. There are no significant
differences across regions in the occurrence of a Saddle. Products launched before
1990 are more likely to see a Saddle in sales.
Now, products launched after 1990 include consumer electronics such as
DVD player, Digital Camera, MP3 player, wherein no Saddle occurs. But these
products have also been observed for less than 10 years and have just experienced
takeoff across countries. The mean time to Saddle from Takeoff is 10 years. This
indicates that we may not be seeing a Saddle in these three categories because there
are insufficient data points after takeoff, and these products have not reached the
initial peak before the Saddle. These results are supported by descriptive statistics in
Table 15 below.
Table 15: Saddle by New versus Old Categories
State All New Old
Saddle 68% 20% 90%
No Saddle 32% 80% 10%
Total 100% 100% 100%

171
The results from the descriptive statistics and the logistic regression indicate
that given sufficient time, a Saddle almost always occurs. Given that a Saddle occurs,
why does it occur? I examine this question in a sub-sample of those categories where
the Saddle has occurred.
4.4.3 Discrete Time Hazard Model
Next, I examine the results of the discrete time hazard model for 109
category * country combinations observed over time leading to 990 observations.
In the individual runs of the hazard model, Saddles are also more likely to
occur the longer the time since takeoff (model 1 in Table 16), the stronger the effect
of negative cascades across countries (model 2), and with the presence of a recession
in the year of the Saddle (model 3), but is not associated with a Chasm in Adopter
Segments (model 4), measured by Deviation
icj
with a single mean across product
classes. This result does not change if the measure Deviation2
icj
(with differing
means is used instead in the individual model.
The full model (model 5) includes the effect of time since takeoff, presence
of negative cascades across countries, presence of recession in the year of the Saddle,
and effect of chasms.
There is a strong positive effect of Time on the hazard of a Saddle. The effect
seems to increase at a declining rate over time as indicated by the variable measuring
the square of the time since takeoff. Negative Cascades also has a strong positive
effect i.e., the greater the number of prior Saddles, the higher the hazard of a Saddle,

172
as seen from the results. There is a strong positive effect of Recession in the year of
the Saddle on the hazard of the Saddle. However, there is no significant effect of
Chasms in Adopter Segments, measured by Deviation
icj
on the hazard of the Saddle.
The adjusted R square of the final model is 14%. The hit rates for the final model is
68%, with a sensitivity score of 65% and a specificity score of 69%.
Table 16: Hazard Model Results
Models 1 2 3 4 5 Odds
Ratio
Adjusted R
square
0.07 0.05 0.02 0 0.14
Observations 990 990 953 977 940
Time Since
Takeoff
0.13
(<0.0001)
0.24
(0.008)
1.26
Square of Time
Since Takeoff
-0.005
(0.23)
0.99
Negative
Cascades
0.32
(<0.0001)
0.30
(<0.0001)
1.34
Recession
same year
0.60
(0.005)
0.67
(0.004)
1.95
Chasm in
Adopter
Segments
0.002
(0.88)
0.007
(0.48)
1.008

In order to support and enrich the hazard analysis, I next examine descriptive
statistics relating to business cycles and chasms.

173
Figure 23 examines the occurrence of Takeoff and Saddle by Recession and
Expansion.
Figure 23: Takeoff and Saddle by Expansion and Recession
0%
10%
20%
30%
40%
50%
60%
Expansion Recession
State
Percentage of all Categories
Takeoff
Saddle

In Figure 23, While 52% of the applicable products took off during times of
economic expansion, 36% took off during times of economic contraction (missing
values for the rest). However, 38% of applicable products experienced a Saddle
during times of expansion while 53% experienced a Saddle during times of
recessions. Hence, Takeoffs show greater association with expansion and Saddles
with recessions
Figure 24 examines the number of takeoffs and Saddles at different levels of
market penetration. All products took off before reaching a market penetration level
of 7%. The average penetration across all products is 2%. The mean penetration at
takeoff varies between 1% and 3% with an average of 2% across countries. The
dispersion of penetration at takeoff is very tight around the mean.

174
Figure 24: Takeoff and Saddle by Market Penetration

0
20
40
60
80
100
120
140
160
No. of
takeoffs/saddles
<5 5 to 10 10 to 20 20 to 40 40 to 60 60 to 80 >= 80
Market penetration
No. of Takeoffs
No. of saddles
This stands in marked contrast to the number of Saddles at different level of
market penetration. The mean penetration at Saddle is 27%, with a standard
deviation of 21. Hence, the dispersion of penetration at Saddle is very loose around
the mean.
In the next section, I consider the impact of the two control variables:
technological changes and repurchase lags in the hazard model.
4.4.4 Impact of Technological Changes
I next run the hazard model with the two item measure of technological
change. The measures are Patent Count and Patent Importance. The results are
shown in Table 17.

175
Table 17: Hazard Model with Technological Change

Parameter Estimate Pr> Chi square
Intercept -4.9<.0001
Time since Takeoff 0.22 0.02
Square of Time Since
Takeoff
-0.004 0.36
Negative Cascades 0.25 0.001
Recession same year 0.84 0.0007
Chasm in Adopter
Segments
0.001 0.88
Patent Count 0.002 0.25
Patent Importance 0.05 0.003
R Square 0.16
Observations 714

Time since Takeoff, Negative Cascades, and Recessions have coefficients
that are positive and different from zero while the coefficient for Chasms is not
significantly different from zero. The coefficient for Patent Count is positive but not
significantly different from zero while the coefficient for Patent Importance is
positive and significantly different from zero. This indicates that the hazard of a
Saddle increases significantly during times of important technological change.
4.4.5 Impact of Repurchase Lags
Only in 64 cases has a category reached a market penetration level of 50%.
This may imply that in many cases, the first cycle peak is not obtained before the
time of the Saddle, weakening the replacement cycles argument. The mean time to
reach the 50% penetration is 21 years with a standard deviation of 11 years.

176
The results with the inclusion of the variable Repurchase Lag is as follows.
Table 18: Hazard Model with Repurchase Lag
Variable Estimates Prob> Chi
square
Intercept -5.24 <.0001
Time since Takeoff 0.32 0.01
Square of Time Since
Takeoff
-0.01 0.12
Negative Cascades 0.37 0.0003
Recession same year 0.76 0.01
Chasm in Adopter Segments 0.02 0.21
Repurchase Lag -0.004 0.59
Adjusted R Square 0.18
Observations 600

As before, Time since Takeoff, Negative Cascades, and Recessions have
coefficients that are positive and different from zero while the coefficient for Chasms
is not significantly different from zero. The coefficient for Repurchase Lag is not
significantly different from zero.
Hence, I find that among the control variables, technological change
measured by Patent Importance increases the likelihood of observing the Saddle
while Repurchase Lag does not seem to be an important driver. In the next section, I
examine the robustness of some of the other independent variables.

177
4.5 ROBUSTNESS ANALYSIS
4.5.1 Magnitude of Business Cycles

In the previous sections, a 0/1 indicator variable was used as a measure for
recessions- i.e., either a recession occurs or not. Alternate measures could be the
Magnitude of expansion during times of expansion, and the magnitude of recession
during times of recession (Lamey et al 2007).
To see whether magnitude of recessions and expansions provide better
explanatory powers, similar to Lamey et al (2007) I create two variables exp
cj
and
rec
cj
.
The variable exp
cj
measures the magnitude of expansion in a country c when the
economy is doing well ( ∆gdp
j
c
>0 where ∆gdp
j
c
refers to the growth in the cyclical
component of the GDP relative to the previous year) by comparing the rise in
cyclical component at a time j with respect to the prior trough. Similarly, the variable
rec
cj
measures the magnitude of contraction in a country c by comparing the fall in
cyclical component at a time j with respect to the prior peak.

178
These measures are as follows:
c
j
c
j cj
hgdp priortroug gdp − = exp if (8) 0 > ∆
c
j
gdp
Otherwise
0 exp =
cj
(9)
0 =
cj
rec (10)
Otherwise
c
j
c
j cj
gdp dp priorpeakg rec − = if (11) 0 < ∆
c
j
gdp

Table 19 shows the result of the hazard model with Magnitude of Expansion
included along with Time since takeoff, Square of Time since takeoff, Negative
cascades, and Chasms in adopter segments. The coefficient of magnitude of
economic expansion is negative and significantly different from zero. Other results
do not vary with inclusion of new measures in the model.

179
Table 19: Hazard Model with Magnitude of Economic Expansion
Parameter Estimates Prob>
Chi square
Intercept -3.87 <.0001
Time since
takeoff
0.24 0.008
Square of Time
since takeoff
-0.005 0.24
Negative
cascades
0.29 <.0001
Magnitude of
Expansion
-18.1 0.03
Chasms in
adopter segments
0.007 0.51
Adjusted R
Square
0.13
Observations 940

I also examine the model with Magnitude of Recession included along with
the other variables. However, its coefficient is not significantly different from zero.
This result points to asymmetries between magnitude of expansion and contraction.
4.5.2 Use of the Hodrick and Prescott filter

There is much literature on the pros and cons of several business cycle filters.
Every filter has its advantages and disadvantages. The Baxter and King filter only
retains fluctuations at business-cycle periodicities (in other words a band-pass filter).
However, with the Baxter and King filter, we lose some data-points.
Hence, I validate the results by adopting the Hodrick and Prescott (HP) filter
to extract the business-cycle fluctuations from each individual series. The HP filter
decomposes a time series into a trend component, which varies smoothly over time,

180
and a cyclical component, by fitting a smooth curve through a set of data points
(Stock and Watson 1999). The HP filter however does not disentangle short-term
from business cycle fluctuations and is what is termed a high pass filter. The results
do not vary with the use of the HP filter- recession remains a strong driver of the
occurrence of a Saddle.
I can also use the growth rate in GDP as an indicator of economic
fluctuations. However, the growth rate strongly overweighs short-term fluctuations
as it is a first-difference filter. Since the focus of this paper is on business cycle
fluctuations, this is not an appropriate filter.
4.5.3 Use of Leading Economic Indicators
Can leading indicators explain the occurrence of the Saddle? As before, I
consider Decreases (increases) in the cyclical component of GDP to correspond to
contractions (expansions), and label the period from a peak in the cyclical
component of GDP to the following trough as a business cycle recession, and the
period from the trough to the next peak as a business cycle expansion. However, I
assume that consumers also make one year-ahead forecasts of the economic status. I
create a new leading economy indicator Leading Economy
cj
which equals 0 if there
is no recession in the current year j in a country c, or the next year j+1; which equals
1 if there is a recession either in the current year or the next year; and which equals 2
if there is a recession both in the current year and the next year.

181
Table 20: Hazard Model with Leading Economy Indicator
Variable Estimates Prob> Chi
square
Intercept -4.84 <.0001
Time since Takeoff 0.27 0.002
Square of Time Since
Takeoff
-0.006 0.12
Negative Cascades 0.29 <.0001
Leading Economy 0.59 0.0006
Chasm in Adopter Segments 0.008 0.46
Adjusted R Square 0.15
Observations 931

The coefficient for Leading Economy
cj
in Table 20 is positive and strongly
different from zero. This indicates that a recession in the same year or an expectation
of a recession in the next year increases the probability of occurrence of a Saddle.
4.5.4 Robustness of Results on Repurchase Cycles
The results from the hazard model indicated that a lag in repurchase demand
is not a driver of the slowdown in sales that we see if so pervasive across categories
and countries. In this section, I test the robustness of this finding in another way.
I check if the pattern of the Saddle is replicated at the first purchases level at
the same or similar time as at the aggregate sales level. I first derive the first
purchases curve by subtracting from sales in the current year the sales n years back,
where n is the replacement time for the category (assumed the same across countries).
I then plot the aggregate sales curve and the first purchases curve.

182
Figure 25: Washing Machine Sales
spain_washer
0
200
400
600
800
1000
1200
1400
1600
1964
1966
1968
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
sales
first purchases

In Figure 25 both overall sales and First purchases see a slowdown occur in
the same year. The timing of the drop in both first purchases curve and total sales
occurs at similar times in 81 out 99 cases where a saddle occurs in all the work
products, PCs and VCRs.
4.6 SUMMARY OF FINDINGS
To sum up, I find that the Saddle is pervasive across the countries and
categories considered in this study. It occurs on average 8-10 years from the time of
takeoff. It leads to a sales decline of -18% on average in the year of Saddle and an
average depth of 27% from the year of the Saddle to the year of Recovery. The
average time to recovery is 7 years. Recessions, Negative Cascades, Time since
Takeoffs are dominant drivers of the Saddle.

183
4.7 CONTRIBUTIONS
This paper makes the following academic and managerial contributions: One
Many factors have been hypothesized to influence the sequence and duration of the
four stages of the traditional product life cycle. Day (1981) criticizes the product life
cycle literature, as not having done enough systematic research into the reasons for
the differences in the shapes. The current study attempts to fill this gap. .A sustained
decline in sales with an average depth of 27% lasting for more than 5 years should
cause grave concerns to managers and lead to premature withdrawal of products. I
develop the first integrated test of multiple theories in this literature stream and the
first study of sales Saddle in an international context. I highlight the importance of
business cycles in driving the Saddle. The Chasm argument has been heavily
popularized in the literature as a dominant driver of a temporary sales slowdown. I
refute it.
4.8 MANAGERIAL IMPLICATIONS
I point out certain key implications of this study.
One, marketers are constantly engaged in devising ways of creating and
managing consumer expectations. This study however adds a word of caution by
pointing out to the effects of ‘over-hype’. While creating expectations and buzz may
cause positive cascades in sales, i also point out at the fragility of cascades and the
response to adverse market news, a subject that warrants further attention in
marketing.

184
Two, this study also points at a need to plan manufacturing and inventory to
account for a slowdown in sales, as well as plan market entry strategies during the
different phases of the product life cycle. The analysis is carried out at a broad
category level. While the category level sales sees a slowdown in sales, brand level
sales may either be constant, increasing or decreasing. For instance, five out of seven
countries see a Saddle in personal computer sales between 1985 and 1990. The
largest market- United States sees a sales slump in the PC market which begins in
1985 and lasts for 8 years before the initial peak is recovered. In 1985, the global
sales of IBM PC and its clones however increase while that of other brands such as
Apple II, Macintosh, Atari 400, Commodore 64, trs-80 see either a temporary or
permanent decline in sales (see http://arstechnica.com/articles/culture/total-
share.ars/1). During this time, there is one new brand entrant – Amiga into the
market, which also sees a permanent decline in sales by 1994.
4.9 FUTURE DIRECTIONS
The dominant role played by recessions in influencing the occurrence of the
Saddle points to a need to research on how to better manage the allocation of
marketing expenditures such as Advertising, promotion and sale staff over time.
Prior literature (Deleersnyder et al 2004) has pointed out that some firms in fact
engage in counter cyclical pricing strategies as a quick fix solution to maintaining
margins. I suggest that these may in fact aggravate the cyclical sensitivities leading
to a worse fall in sales, as the slowdown may in fact be sustained for many years

185
4.10 LIMITATIONS
Certain other drivers have been proposed in the literature which I have not
specifically examined in this study. Wasson (1978) was among the first researchers
to specifically incorporate a specific period of slowdown in following a period of
rapid growth, in his theoretical exposition of the product life cycle sales. He termed
this period competitive turbulence, and argued that during this time the supply of
potential buyers reduced and the market reaches a degree of over capacity, leading to
a competitive battle for permanent market position. I am unable to test this
explanation for the slowdown in sales as I do not have data on market competition
for each category and country. However, it appears unlikely that competitive
turbulence would be a systematic cause across these diverse categories and across
several countries.
Van den Bulte and Joshi argue a drop in sales may occur when the product
moves from Influentials to Imitators. I am unable to tease out the proportion of the
sales to Influentials and proportion of sales to Imitators and hence am not able to
examine this version of the dual-segments argument.
Comparable data on price, promotions and market competition for these
different industries in the different countries is not available. Price declines and
market entry have been shown in prior literature to be dominant drivers of takeoff,
and may also impact slowdown.

186
REFERENCES
Agarwal, Rajshree, Bayus, B Bayus (2004), “Creating and Surviving in New
Industries”, In J.A.C. Baum & A.M. Mcgahan (Eds.), Business Strategy Over
the Industry Life Cycle: Advances in Strategic Management, V. 21. Oxford
UK: JAI/Elsevier, Forthcoming.
Agarwal, Rajshree, Barry L. Bayus (2002), “Market Evolution and Sales Takeoff of
Product Innovations”, Management Science, 48 (8), 1024 -1041.
Allison, Paul D. (1995), Survival Analysis Using SAS: A Practical Guide, Cary, NC:
SAS Institute Inc.
Allison, Paul D. (1982), “Discrete-Time Methods for the Analysis of Event History
Data” In Sociological Methodology (Ed. S. Leinhardt), San Francisco:
Jossey-Bass
Ashkanasy, Neal. M., Trevor-Roberts, Edwin, Earnshaw, Louise (2002), “The Anglo
Cluster: Legacy of the British Empire”, Journal of World Business, 37, 28-39
Backus, D.K., and P.J. Kehoe (1992), “International Evidence on the Historical
Properties of Business Cycles”, American Economic Review 82 (4), 864-888.
Bailey, Norman T.J. (1957), The Mathematical Theory of Epidemics, 1st Ed.
London: Griffin.
Bailey, Norman T.J. (1975), The Mathematical Theory of Infectious Diseases and Its
Applications. London: Charles Griffin and Company.
Bass, Frank M. (1969), “A New Product Growth Model for Consumer Durables”
Management Science, 15 (5), 215–227.
Bass, Frank M. (1980), “The Relationship between Diffusion Rates, Experience
Curves and Demand Elasticities for Consumer Durable Technological
Innovations”, Journal of Business, 53 (2), 51–68.
Bass, Frank M., Trichy V. Krishnan, and Dipak C. Jain (1994), “Why the Bass
Model Fits Without Decision Variables”, Marketing Science, 13 (3), 203–
223.
Bayus, Barry. (1987), “Forecasting Sales of New Contingent Products: An
Application to the Compact Disc Market.” Journal of Product Innovation
Management, 4, 243–255.

187
Bayus, Barry (1993), “High-Definition Television: Assessing Demand Forecasts For
A Next Generation Consumer Durable” Management Science, 39 (11), 1319–
1333.
Bayus, Barry and Sunil Gupta (1992), “An Empirical Analysis of Consumer Durable
Replacement Intentions”, International Journal of Research in Marketing, 9,
257–267.
Baxter, Marianne, and Robert G. King (1992), “Measuring Business Cycles:
Approximate Band-Pass Filters For Economic Time Series”, The Review of
Economics and Statistics, 81 (4): 575-593.
Bayus, Barry, S. Hong and R.P. Labe Jr. (1989), “Developing and Using Forecasting
Models of Consumer Durables”, Journal of Product Innovation Management,
6, 5–19.
Bemmaor, A. (1994), “Modeling the Diffusion of New Durable Goods: Word-of-
Mouth Effect versus Consumer Heterogeneity.” Research Traditions in
Marketing, Ed. Gilles Laurent, Gary L. Lilien and Bernard Pras, 201–229,
Boston: Kluwer.
Bemmaor, A. and Yanghyuk Lee (2002), “The Impact of Heterogeneity and Ill-
Conditioning on Diffusion Model Parameter Estimates”, Marketing Science,
21, 209–220.
Bernhardt, Irwinand Kenneth M. Mackenzie (1972), “Some Problems in Using
Diffusion Models For New Products.” Management Science, 19, 187–200.
Bhargava, Subhash, Raj K Bhargavaand Ashok Jain (1991), “Requirement of
Dimensional Consistency in Model Equations: Diffusion Models
Incorporating Price and their Applications”, Technological Forecasting and
Social Change, 41, 177–188.
Bikhchandani, Sushil, David Hirshleifer and Ivo Welch (1992), “A Theory of Fads,
Fashion, Custom and Cultural Change as Information Cascades”, Journal of
Political Economy, 100 (5), 992–1026.
Blattberg, Robert and John Golanty (1978), “Tracker: An Early Test Market
Forecasting and Diagnostic Model For New Product Planning”, Journal of
Marketing Research, 15 (2), 192–202.
Bretschneider, Stuart I. and Barry Bozeman (1986), “Adaptive Diffusion Models For
The Growth of Robotics in New York State Industry” , Technological
Forecasting and Social Change, 30, 111–121.

188
Bretschneider, Stuart I. and Vijay Mahajan (1980), “Adaptive Technological
Substitution Models”, Technological Forecasting and Social Change, 18,
129–139.
Bronnenberg, Bart J. and Carl F Mela (2004), “Market Adoption and Retailer
Rollout of New Brands”, Marketing Science, 23 (4), 500.
Brown, Lawrence (1981), Innovation Diffusion: A New Perspective, London:
Methuen.
Bowden, Sue and Avner offer (1994), “Household Appliances and the Use of Time:
The United States and Britain since the 1920s,” Economic History Review,
47 (4), 725-48.
Bucklin, Louis P., and Sanjit Sengupta (1993), “The Co-Diffusion of
Complementary Innovations: Supermarket Scanners and UPC Symbols.”
Journal of Product Innovation Management, 10, 148–160.
Burns, A. M., and W. C. Mitchell (1946), Measuring Business Cycles, New York:
National Bureau of Economic Research
Carl, Dale, Vipin Gupta and Mansour Javidan (2004), “Power Distance”, In House
Et Al., Ed. “Culture, Leadership, and Organizations: The GLOBE Study of
62 Societies”, Sage Publications Inc.
Chandrasekaran, Deepa and Gerard J. Tellis (2006), "A Critical Review of
Marketing Research on Diffusion of New Products", Review of Marketing
Research, Volume 3, Ed. Naresh K. Malhotra, ME Sharpe, Inc
Chandrashekaran, Murali and Rajiv K. Sinha (1995) , “Isolating the Determinants of
Innovativeness: A Split-Population Tobit (SPOT) Duration Model of Timing
and Volume of First and Repeat Purchase”, Journal of Marketing Research,
32 (4), 444
Chandy, Rajesh K., and Gerard J. Tellis (2000), “The Incumbent’s Curse?
Incumbency, Size and Radical Product Innovation”, Journal of Marketing,
64, 1-17
Chatterjee, Rabikar, and Jehoshua Eliashberg (1990), “The Innovation Diffusion
Process in a Heterogeneous Population: A Micro Modeling Approach”
,Management Science, 36, 1057–1079.
Chow, Gregory C. (1967), “Technological Change and the Demand for Computers”,
American Economic Review, 57 (5), 1117–1130.

189
Chryssochoidis G.M. and V. Wong (1998), “Rolling Out New Product across
Country Markets: An Empirical Study of Causes of Delays,” Journal of
Product Innovation Management, 15 (1), 16-41.
Cohen, Morris A., Teck H. Ho and Hirofumi Matsuo (2000), “Operations Planning
in the Presence of Innovation Diffusion Dynamics”, In New Product
Diffusion Models, Ed. V. Mahajan, E. Muller, and Y. Wind, Pp. 237-259,
Boston: Kluwer Academic.
Coleman, James, Elihu Katz and Herbert Menzel (1966), Medical Innovation, New
York: Bobbs-Merrill.
Cox, W.E. Jr. (1967), “Product Life Cycles as Marketing Models”, Journal of
Business, 40, 375–384.
Danaher, Peter J., Bruce G.S. Hardie, and William P. Putsis Jr. (2001), “Marketing-
Mix Variables and the Diffusion of Successive Generations of a
Technological Innovation”, Journal of Marketing Research, 38, 501–514.
Day, George. (1981), “The Product Life Cycle: Analysis and Application Issues”,
Journal of Marketing, 41, 60–67.
Dekimpe, Marnik, Philip Parker and Miklos Sarvary (1998), “Staged Estimation of
International Diffusion Models: An Application to Global Cellular Telephone
Adoption.” Technological Forecasting and Social Change, 57, 105–132.
Dekimpe, Marnik, Philip Parker and Miklos Sarvary (2000a), “Global Diffusion of
Technological Innovations: A Coupled-Hazard Approach”, Journal of
Marketing Research, 37 (1), 47–59.
Dekimpe, Marnik, Philip Parker and Miklos Sarvary (2000b), “Multimarket and
Global Diffusion”, In New Product Diffusion Models, Ed. Vijay Mahajan,
Eitan Muller, and Yoram Wind, Boston: Kluwer Academic.
Deleersnyder, Barbara, Marnik Dekimpe, Miklos Sarvary, and Philip Parker (2004),
“Weathering Tight Economic Cycles: The Sales Evolution of Consumer
Durables over the Business Cycle”, Quantitative Marketing and Economics,
4, 347–383.
Deninger, Klaus and Lyn Squire (1996), “Measuring Income Inequality: A New
Database,” url http://Www.Worldbank.Org/Research/Growth/Dddeisqu.Htm.

190
Dickson, Marcus W., Renee S. Beshears and Vipin Gupta (2004), “The Impact of
Societal Culture and Industry on Organizational Cultures: Theoretical
Explanations”, In House Et Al., Ed. “Culture, Leadership, and Organizations:
The GLOBE Study of 62 Societies”, Sage Publications Inc.
Dorfman, Peter W. and Robert J. House (2004), “Cultural Influences on
Organizational Leadership: Literature Review, Theoretical Rationale, and
GLOBE Project Goals” In House Et Al, Ed. “Culture, Leadership, and
Organizations: The GLOBE Study of 62 Societies”, Sage Publications Inc.
Easingwood, Christopher (1987), “Early Product Lifecycle Forms For Infrequently
Purchased Major Products”, International Journal of Research in Marketing,
4 (1), 3–9.
Easingwood, Christopher (1989), “An Analogical Approach to Long Term
Forecasting of Consumer Durable Sales”, International Journal of
Forecasting, 5 (1), 69–82.
Easingwood, Christopher, Vijay Mahajan and Eitan Muller (1983), “A Non-Uniform
Influence Innovation Diffusion Model of New Product Acceptance”,
Marketing Science, 2 (3), 273–295
Elberse, Anita and Jehoshua 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, Jehoshua and Mohanbir S. Sawhney (1994), “Modeling Goes to
Hollywood: Predicting Individual Differences in Movie Enjoyment”,
Management Science, 40 (9), 1151–1173
Eliashberg, Jehoshua and Steven M. Shugan (1997), “Film Critics: Influencers or
Predictors?”, Journal of Marketing, 61, 68–78
Eliashberg, Jehoshua, Jedid-Jah Jonker, Mohanbir S. Sawhney, and Berend
Wierenga (2000), “MOVIEMOD: An Implementable Decision Support
System for Pre-Release Market Evaluation of Motion Pictures”, Marketing
Science, 19 (3), 226–243
Feder, Gershon and Gerald O’Mara (1982), “On Information and Innovation
Diffusion: A Bayesian Approach”, American Journal of Agricultural
Economics, 64, 145–147
Fisher, J.C., and R.H. Pry (1971), “A Simple Substitution Model of Technological
Change”, Technological Forecasting and Social Change, 3, 75–88

191
Foster, Joseph A., Peter N. Golder, and Gerard J. Tellis (2004), “Predicting Sales
Takeoff For Whirlpool’s New Personal Valet”, Marketing Science, 23 (2),
180–191
Fourt, L., and Joseph Woodlock (1960), “Early Prediction of Market Success of New
Grocery Products,” Journal of Marketing, 25(2), 31–38
Frances, Philip Hans (1994), “Modeling New Product Sales: An Application of
Cointegration Analysis”, International Journal of Research in Marketing, 11,
491–502
Fudenberg, Drew, and Jean Tirole (1985), “Preemption and Rent Equalization in the
Adoption of New Technology”, Review of Economic Studies, 52, 383–402
Furman, J. L., Porter, M.E. and S. Stern (2002), “The Determinants of National
Innovative Capacity,” Research Policy, 31, 899–933
Ganesh, Jaishankar, and V. Kumar (1996), “Capturing the Cross-National Learning
Effect: An Analysis of an Industrial Technology Diffusion”, Journal of the
Academy of Marketing Science, 24 (4), 328–337
Ganesh, Jaishankar, V. Kumar, and V. Subramaniam (1997), “Learning Effect in
Multinational Diffusion of Consumer Durables: An Exploratory
Investigation”, Journal of the Academy of Marketing Science, 25 (3), 214–
228
Garber, Tal, Jacob Goldenberg, Barak Libai, and Eitan Muller (2004), “From
Density to Destiny: Using Spatial Dimension of Sales Data for Early
Prediction of New Product Success”, Marketing Science, 23 (3), 419–428
Gatignon, Hubert, Jehoshua Eliashberg, and Thomas S. Robertson (1989),
“Modeling Multinational Diffusion Patterns: An Efficient Methodology”,
Marketing Science, 8 (3), 231–247
Gelfand, Michele J., Dharm P.S. Bhawuk, Lisa Hisae Nishii and David J. Bechtold
(2004), “Individualism and Collectivism”, In House, Robert J., Paul J
Hanges, Mansour Javidan, Peter W. Dorfman, Vipin Gupta , (2004), Ed.
“Culture, Leadership, and Organizations: The GLOBE Study of 62
Societies”, Sage Publications Inc.
Gielens, Katrijn, Marnik G. Dekimpe (2001), “Do International Entry Decisions of
Retail Chains Matter in the Long Run?”, International Journal of Research in
Marketing, 18, 235–259

192
Goldenberg, Jacob, Barak Libai, Eitan Muller (2002), “Riding the Saddle: How
Cross-Market Communications Can Create A Major Slump in Sales,” Journal
of Marketing, 66, 1-16
Goldenberg, Jacob, Barak Libai, Eitan Muller (2001), “Using Complex Systems
Analysis to Advance Marketing Theory Development: Modeling
Heterogeneity Effects On New Product Growth Through Stochastic Cellular
Automata,” Academy of Marketing Science Review,
Http://Www.Findarticles.Com/P/Articles/Mi_Qa3896/Is_200101/Ai_N89447
61, Last Accessed April 4 2006)
Goldenberg, Jacob, Barak Libai, and Eitan Muller (2001b), “Talk of the Network: A
Complex Systems Look at the Underlying Process of Word-of-Mouth”,
Marketing Letters, 12 (3), 209–221
Golder, Peter N., Gerard J. Tellis (2004), “Going, Going, Gone: Cascades, Diffusion,
and Turning Points of the Product Life Cycle”, Marketing Science, 23 (2),
207-218
Golder, Peter N., Gerard J. Tellis (1998), “Beyond Diffusion: An Affordability
Model of the Growth of New Consumer Durables”, Journal of Forecasting,
17 (3/4), 259-280
Golder, Peter N. and Gerard J. Tellis (1997), “Will It Ever Fly? Modeling the
Takeoff of Really New Consumer Durables,” Marketing Science, 16(3), 256-
70
Gort, Michael, Steven Klepper (1982), “Time Paths in the Diffusion of Product
Innovations,” The Economic Journal, 92(367), 630-53
Gupta, Sachin, Dipak C. Jain, and Mohanbir S. Sawhney (1999), “Modeling the
Evolution of Markets with Indirect Network Externalities: An Application to
Digital Television”, Marketing Science, 18, 396–416
Gruber, Harald (2005), The Economics of Mobile Telecommunications, Cambridge
University Press
Gupta, Vipin and Paul J. Hanges (2004), “Regional and Climate Clustering of
Societal Clusters”, In House Et Al .Ed. “Culture, Leadership, and
Organizations: The GLOBE Study of 62 Societies”, Sage Publications Inc.
Gupta, Vipin, Surie, Gita, Javidan, Mansour and Chhokar, Jagdeep (2002) “Southern
Asia Cluster: Where the Old Meets the New?”, Journal of World Business,
37, 16-27

193
Hagerstrand, Torsten (1953), Innovation Diffusion as A Spatial Process, Chicago:
University of Chicago Press
Hahn, Minhi, Sehoon Park, Lakshman Krishnamurthi, and Andris Zoltners (1994),
“Analysis of New Product Diffusion Using A Four Segment Trial-Repeat
Model”, Marketing Science, 13 (3), 224–247
Hartfield, Edward F. (1989), “The Divergent Economic Development of China and
Japan” In Confucianism and Economic Development: An Oriental
Alternative? Ed. Hung-Chao Tai, The Washington Institute Press, 92-115
Heeler, R., and Thomas Hustad (1980), “Problems in Predicting New Product
Growth for Consumer Durables”, Management Science, 26 (10), 1007–1020
Helsen, Kristaan, Kamel Jedidi, and Wayne Desarbo (1993), “A New Approach to
Country Segmentation Utilizing Multinational Diffusion Patterns”, Journal of
Marketing, 57 (4), 60–71
Herfindahl, O. C. (1950), Concentration in the Steel Industry Ph. D. Thesis,
Columbia University
Heston, Alan Robert Summers and Bettina Aten (2002), Penn World Table Version
6.1, Center for International Comparisons at the University of Pennsylvania
(CICUP)
Hiebert, L Dean (1974), “Risk, Learning and the Adoption of Fertilizer Responsive
Seed Varieties”, American Journal of Agricultural Economics, 56, 764–768
Hjorth, Urban (1980), “A Reliability Distribution with Increasing, Decreasing,
Constant and Bathtub-Shaped Failure Rates”, Technometrics, 21 (2), 99–107
Ho, Teck-Hua, Sergei Savin, and Christian Terwiesch (2002), “Managing Demand
and Sales Dynamics in New Product Diffusion under Supply Constraint”,
Management Science, 48 (2), 187–206
Hofstede, Geert (2001), Culture’s Consequences: Comparing Values, Behaviors,
Institutions and Organizations across Nations, 2nd Ed., Sage Publications,
Thousand Oaks
Horsky, Dan (1990), “A Diffusion Model Incorporating Product Benefits, Price,
Income and Information”, Marketing Science, 9, 342–365
Horsky, Dan, and Leonard Simon (1983), “Advertising and the Diffusion of New
Products”, Marketing Science, 2 (1), 1–17

194
Hossain, Ferdaus and Benjamin Onyango (2004), “Product Attributes and Consumer
Acceptance of Nutritionally Beneficial Genetically Modified Foods”,
International Journal of Consumer Studies, 28(3), 255-267
House, Robert J., Paul J Hanges, Mansour Javidan, Peter W. Dorfman and Vipin
Gupta , (2004), Ed. “Culture, Leadership, and Organizations: The GLOBE
Study of 62 Societies”, Sage Publications Inc
Infosino, William J. (1986), “Forecasting New Product Sales from Likelihood of
Purchase Ratings”, Marketing Science, 5 (4), 372–390
Jain Dipak, and Ram C. Rao (1990), “Effect of Price on the Demand for Durables:
Modeling, Estimation and Findings”, Journal of Business and Economic
Statistics, 8 (2), 163–170
Jain, Dipak, Vijay Mahajan, and Eitan Muller (1991), “Innovation Diffusion in the
Presence of Supply Restrictions”, Marketing Science, 10 (1), 83–90
Javidan, Mansour (2004), “Performance Orientation”, In House Et Al., Ed. “Culture,
Leadership, and Organizations: The GLOBE Study of 62 Societies”, Sage
Publications Inc.
Jesuino, Jorge. C (2002), “Latin Europe Cluster: from South to North”, Journal of
World Business, 81-89
Jones, Morgan, and Christopher J. Ritz (1991), “Incorporating Distribution into New
Products Diffusion Models”, International Journal of Research in Marketing,
8, 91–112
Kabasakal, Hayat and Bodur, Muzaffer (2002), “Arabic Cluster: A Bridge between
East and West”, Journal of World Business, 37, 40-54
Kalish, Shlomo, Vijay Mahajan and Eitan Muller (1995), “Waterfall and Sprinkler
New-Product Strategies in Competitive Global Markets,” International
Journal of Research in Marketing, 12, 105-19
Kalish, Shlomo (1985), “A New Product Adoption Model with Pricing, Advertising
and Uncertainty”, Management Science, 31, 1569–1585
Kamakura, Wagner, and Siva K. Balasubramanian (1987), “Long-Term Forecasting
With Innovation Diffusion Models: The Impact of Replacement Purchases”,
Journal of Forecasting, 6, 1–19

195
Kamakura, Wagner, and Siva K. Balasubramanian (1988) , “Long-Term View of the
Diffusion of Durables: A Study of the Role of Price and Adoption Influence
Processes Via Tests of Nested Models”, International Journal of Research In
Marketing, 5, 1–13
Karshenas, M., and P. Stoneman (1993), “Rank, Stock, Order, and Epidemic Effects
in the Diffusion of New Process Technologies: An Empirical Model”, Rand
Journal of Economics, 24, 503–528
Kim, Namwoon, Dae Ryun Chang, and Allan D. Shocker (2000), “Modeling Inter-
Category and Generational Dynamics for a Growing Information Technology
Industry”, Management Science, 46 (4), 496–512
Kohli, Rajeev, Donald Lehmann, and Jae Pae (1999), “Extent and Impact of
Incubation Time in New Product Diffusion”, Journal of Product Innovation
Management, 16, 134–144
Krishnan, Trichy V., Frank Bass, and Dipak Jain (1999), “Optimal Pricing Strategy
for New Products”, Management Science, 45 (12), 1650–1663
Krishnan, Trichy V., Frank Bass, and V. Kumar (2000), “Impact of a Late Entrant on
the Diffusion of a New Product/Service”, Journal of Marketing Research, 37
(2), 269–278
Kumar V., and Trichy V. Krishnan. (2002) “Multinational Diffusion Models: An
Alternative Framework”, Marketing Science, 21 (3), 318–330
Kumar, V., Jaishankar Ganesh, and Raj Echambadi (1998), “Cross National
Diffusion Research: What Do We Know and How Certain Are We?”, Journal
of Product Innovation Management, 15, 255–268
Lamey, Lien, Barbara Deleersnyder, Marnik G. Dekimpe, and Jan-Benedict E.M.
Steenkamp (2007), “How Business Cycles Contribute to Private-Label
Success: Evidence from the U.S. and Europe”, Journal of Marketing, 71(1),
1-15
Lavaraj, U.A., and A.P. Gore (1990), “On Interpreting Probability Distributions
Fitted to Times of First Adoption”, Technological Forecasting and Social
Change, 37, 355–370
Lee, Jonathan A., Peter Boatwright, and Wagner A. Kamakura (2003), “Bayesian
Model for Pre-Launch Sales Forecasting of Recorded Music”, Management
Science, 49 (2), 179–196

196
Lehmann, Donald R., and Charles B. Weinberg (2000), “Sales through Sequential
Distribution Channels: An Application to Movies and Videos”, Journal of
Marketing, 64 (3), 18–33
Lenk, Peter J., and Ambar G. Rao (1990), “New Models from Old: Forecasting
Product Adoption by Hierarchical Bayes Procedures”, Marketing Science, 9
(1), 42–53
Lilien, Gary I., Ambar G. Rao, and Shlomo Kalish (1981), “Bayesian Estimation and
Control of Detailing Effort in A Repeat-Purchase Diffusion Environment”,
Management Science, 27 (5), 493–506
Lindridge, Andrew (2005), “Religiosity and the Construction of A Cultural-
Consumption Identity”, Journal of Consumer Marketing, 22(3), 142-151
Lynn, Michael and Gelb, Betsy D. (1996), “Identifying Innovative National Markets
For Technical Consumer Goods”, International Marketing Review, 13(6), 43-
52
Mahajan, Vijay, and Eitan Muller (1996), “Timing, Diffusion and Substitution of
Successive Generations of Technological Innovations: The IBM Mainframe
Case”, Technological Forecasting and Social Change, 51, 109–132
Mahajan, Vijay, Eitan Muller, and Frank M. Bass (1990), “New Product Diffusion
Models In Marketing: A Review and Directions for Research”, Journal of
Marketing, 54, 1–26
Mahajan, Vijay, Eitan Muller and Frank M. Bass (1995), “Diffusion of New
Products: Empirical Generalizations and Managerial Uses”, Marketing
Science, 14(3), Part 2 of 2: Special Issue on Empirical Generalizations in
Marketing, G79-G88
Mahajan, Vijay, Eitan Muller, and Rajendra K. Srivastava (1990), “Determination of
Adopter Categories Using Innovation Diffusion Models”, 27 (1), 37–50
Mahajan, Vijay, Eitan Muller, and Yoram Wind (2000a), “New Product Diffusion
Models: from Theory to Practice”, New Product Diffusion Models, Ed.
Mahajan, V., Eitan Muller, Yoram Wind, Pp. 3-24, Boston: Kluwer
Academic
Mahajan, Vijay, Eitan Muller, and Yoram Wind (2000b), New Product Diffusion
Models, Boston: Kluwer Academic
Mahajan, Vijay, and Robert Peterson (1978), “Innovation Diffusion in a Dynamic
Potential Adopter Population”, Management Science, 24 (15), 1589–1597

197
Mahajan, Vijay, and Robert Peterson (1979) “Integrating Time and Space In
Technological Substitution Models”, Technological Forecasting and Social
Change, 14, 231–241
Mahajan, Vijay, Subhash Sharma, and Robert D. Buzzell (1993), “Assessing the
Impact of Competitive Entry on Market Expansion and Incumbent Sales”,
Journal of Marketing, 57, 39–52
Mansfield, Edwin (1961), “Technical Change and the Rate of Imitation”,
Econometrica, 29, 741–766
Mcclelland, David C (1976), The Achieving Society, Irvington Publishers Inc. New
York
Miller, A. S., and Hoffmann, J. P. (1995), “Risk and Religion: An Explanation of
Gender Differences in Religiosity”, Journal for the Scientific Study of
Religion, 34, 63–75
Miller, Alan S. and Rodney Stark (2002), “Gender and Religiousness: Can
Socialization Explanation Be Saved?”, American Journal of Sociology,
107(6), 1399-1423
Mitra, Debanjan and Peter N. Golder (2002), “Whose Culture Matters? Near-Market
Knowledge and Its Impact on Foreign Market Entry Timing”, Journal of
Marketing Research, 39 (August), 350-65
Moe, Wendy, and Peter Fader (2002), “Using Advanced Purchase Orders to Forecast
New Product Sales”, Marketing Science, 21, 347–364
Moore, Geoffrey A. (1991), Crossing the Chasm: Marketing and Selling Technology
Products to Mainstream Customers, New York: Harpercollins
Morrill, Richard, Gary L. Gaile, and Grant Ian Thrall (1988), Spatial Diffusion,
Newbury Park, CA: Sage
Nakata, Cheryl and K. Sivakumar (1996), “National Culture and New Product
Development: An Integrative Review”, Journal of Marketing, 60, 61-72
Narin, Francis, Elliot Noma, and Ross Perry (1987), “Patents as Indicators of
Corporate Technological Strength”, Research Policy, 16 (2), 143–55
Neelamegham, Ramya, and Pradeep Chintagunta (1999) “A Bayesian Model to
Forecast New Product Performance in Domestic and International Markets”,
Marketing Science, 18 (2), 115–36

198
Norton, John A. and Frank M. Bass (1987), “A Diffusion Theory Model of Adoption
and Substitution for Successive Generations of High-Technology Products”,
Management Science, 33 (9), 1069–1086
Norton, John A. and Frank M. Bass (1992), “Evolution of Technological
Generations: the Law of Capture”, Sloan Management Review, 33 (2), 66–77
Olson, Jerome and Seungmook Choi (1985), “A Product Diffusion Model
Incorporating Repeat Purchases”, Technological Forecasting and Social
Change, 27, 385–397
Oren, Shmuel S. and Rick Schwartz (1988), “Diffusion of New Products in Risk-
Sensitive Markets”, Journal of Forecasting, 7, 231–287
Parker, Philip (1994), “Aggregate Diffusion Forecasting Models in Marketing: A
Critical Review”, International Journal of Forecasting, 10, 353–80
Parker, Philip and Hubert Gatignon (1994), “Specifying Competitive Effects in
Diffusion Models: An Empirical Analysis”, International Journal of Research
in Marketing, 11, 17–39
Perkins, Richard and Eric Neumayer (2004), “The International Diffusion of New
Technologies: A Multi-Technology Analysis of Latecomer Advantage and
Global Economic Integration”, Development and Comp Systems 0407001,
The Former Econwpa
Pliner, Vadim (2005), “Predictive Modeling Using Survival Analysis”, NESUG
2005 Proceedings, URL:
Http://Www.Nesug.Org/Html/Proceedings/Nesug05/Pos/Pos6.Pdf, Last
Accessed May 5 2006
Prabhu, Jaideep C., Rajesh K. Chandy, and Mark E. Ellis (2005), “The Impact of
Acquisitions on Innovation: Poison Pill, Placebo, or Tonic?”, Journal of
Marketing, 69, 114–130
Putsis, William P. Jr., Sridhar Balasubramanian, Edward Kaplan, and Subrata Sen
(1997), “Mixing Behavior in Cross-Country Diffusion”, Marketing Science,
16 (4), 354–369
Putsis, William P. Jr. and V. Srinivasan (2000), “Estimation Techniques for Macro
Diffusion Models”, in New Product Diffusion Models, Ed. Mahajan, V.,
Eitan Muller, Yoram Wind, 263-291, Boston: Kluwer Academic.

199
Rangaswamy, Arvind and Sunil Gupta (2000), “Innovation Adoption and Diffusion
in the Digital Environment: Some Research Opportunities” in New Product
Diffusion Models, Ed. Mahajan, V., Eitan Muller, Yoram Wind, 75-96,
Boston: Kluwer Academic.
Redmond, William. (1994), “Diffusion at Sub-National Levels: A Regional Analysis
of New Product Growth”, Journal of Product Innovation Management, 11,
201–212
Reinganum, Jennifer F. (1981), “Market Structure and the Diffusion of New
Technology”, Bell Journal of Economics, RAND, 12 (2), 618–624
Roberts, John. H. and James L. Lattin (2000), “Disaggregate-Level Diffusion
Models,” In New Product Diffusion Models, Ed. V. Mahajan, Eitan Muller,
and Yoram Wind, 207- 236 Boston: Kluwer Academic
Roberts, John H. and Glen Urban (1988), ‘‘Modeling Multivariate Utility, Risk, and
Belief Dynamics for New Consumer Durable Brand Choice”, Management
Science, 34 (2), 167–185
Robinson, Bruce and Chet Lakhani (1975), “Dynamic Price Models for New Product
Planning”, Management Science, 21, 1113–1122
Ronen, S. and Shenkar, O. (1985), “Clustering Countries on Attitudinal Dimensions:
A Review and Synthesis”, Academy of Management Review, 10(3), 435-454
Rogers, Everett (1995), Diffusion of Innovations, New York: Free Press
Sawhney, Mohanbir S., and Jehoshua Eliashberg (1996), “A Parsimonious Model for
Forecasting Gross Box-office Revenues of Motion Pictures”, Marketing
Science, 15 (2), 113–131
Schmittlein, D. and Vijay Mahajan (1982), “Maximum Likelihood Estimation for an
Innovation Diffusion Model of New Product Acceptance”, Marketing
Science, 1 (1), 57–78
Sharma, Praveen, and S.C. Bhargava (1994), “A Non-Homogeneous Non-Uniform
Influence Model of Innovation Diffusion”, Technological Forecasting and
Social Change, 46, 279–288
Shocker, Allan D., Barry L. Bayus and Namwoon Kim (2004), “Product
Complements and Substitutes in the Real World: The Relevance of Other
Products”, Journal of Marketing, 68 (1), 28–40

200
Shugan, Steven (2000), “Recent Research in the Motion Picture Industry” Available
At Http://Bear.Cba.Ufl.Edu/Shugan/Profile/RRMPI.Pdf, Last Accessed April
4 2006
Shugan, Steven and Joffre Swait (N.D.), “Enabling Movie Design and Cumulative
Box office Predictions Using Historical Data and Consumer Intent-to-View”,
University of Florida, Working Paper
Simon, H. and K. Sebastian (1987), “Diffusion and Advertising: The German
Telephone Company”, Management Science, 33, 451–466
Singer, J.D. and Willet, J.B. (1993), “It’s About Time: Using Discrete-Time Survival
Analysis to Study Duration and the Timing of Events”, Journal of
Educational Statistics, 18: 155-195
Sinha, Rajiv K.and Murali Chandrashekaran (1992), “A Split Hazard Model for
Analyzing the Diffusion of Innovations”, Journal of Marketing Research, 29
(1), 116
Song, Inseong and Pradeep Chintagunta (2003), “A Micromodel of New Product
Adoption with Heterogeneous and Forward Looking Consumers: Application
to the Digital Camera Category”, Quantitative Marketing and Economics, 1,
371–407
Srinivasan V., and Charlotte Mason (1986), “Nonlinear Least Squares Estimation of
New Product Diffusion Models”, Marketing Science, 5 (2), 169–178
Srinivasan, Raji, Gary L. Lilien and Arvind Rangaswamy (2004), “First In, First
Out? The Effects of Network Externalities on Pioneer Survival,” Journal of
Marketing, 68, 41–58
Srinivasan, Raji, Gary L. Lilien, and Arvind Rangaswamy (2006) “The Emergence
of Dominant Designs”, Journal of Marketing, 70(2), 1-17
Steenkamp, Jan-Benedict, Frenkel Ter Hofstede and Michel Wedel (1999), “A
Cross-National Investigation into the Individual and National Cultural
Antecedents of Consumer Innovativeness,” Journal of Marketing, 63(2), 55-
69
Steffens, Paul R. (2002), “A Model of Multiple Ownership as a Diffusion Process”,
Technological Forecasting and Social Change, 70, 901–917
Stoneman, Paul (1981), “Intra-Firm Diffusion, Bayesian Learning and Profitability”,
Economic Journal, 91, 375–388

201
Stoneman, Paul (2002), The Economics of Technological Diffusion Cambridge, MA:
Blackwell
Stremersch, Stefan and Gerard J. Tellis (2004), “Understanding and Managing
International Growth of New Products”, International Journal of Research in
Marketing, 21, 421-438
Stulz, Rene M, and Rohan Williamson (2003), “Culture, Openness and Finance”,
Journal of Financial Economics, 70 (3), 313-349
Sully De Luque, Mary and Mansour Javidan (2004), “Uncertainty Avoidance”, In
House Et al., Ed. “Culture, Leadership, and Organizations: The GLOBE
Study of 62 Societies”, Sage Publications Inc.
Sultan, Fareena, John U. Farley, and Donald R. Lehmann (1990), “A Meta-Analysis
of Diffusion Models”, Journal of Marketing Research, 27, 70–77
Szabo, Erna, Brodbeck, Felix. C., Den Hartog, Deanne. N., Reber, Gerhard, Weibler,
Jurgen and Wunderer, Rolf (2002), “The Germanic Europe Cluster: Where
Employees Have A Voice”, Journal of World Business, 37 (1), 55-68
Takada, Hirozu, and Dipak Jain (1991), “Cross-National Analysis of Diffusion of
Consumer Durable Goods in Pacific Rim Countries”, Journal of Marketing,
55, 48–54
Talukdar, Debabrata, K. Sudhir, and Andrew Ainslie (2002), “Investigating New
Product Diffusion across Products and Countries”, Marketing Science, 21 (1),
97–114
Tellis, Gerard J., Stefan Stremersch, and Eden Yin (2003), “The International
Takeoff of New Products: The Role of Economics, Culture and Country
Innovativeness”, Marketing Science, 22 (2), 188–208
Thorelli H.B., and Burnett S.C. (1981), “The Nature of Plcs for Industrial Goods
Businesses”, Journal of Marketing, 45, 97-108
Triandis, Harry C. (1995), Individualism versus Collectivism, Westview Press, Inc.,
Boulder
Urban, Glen L., Bruce D. Weinberg, and John R. Hauser (1996), “Premarket
Forecasting of Really-New Products”, Journal of Marketing, 60, 47–60

202
Urban, Glen L., John R. Hauser, William J. Qualls, Bruce D. Weinberg, Jonathan D.
Bohlmann, and Roberta A. Chicos (1997), “Information Acceleration:
Validation and Lessons from the Field”, Journal of Marketing Research, 34,
143–153
Van Den Bulte, Christophe (2000) “New Product Diffusion Acceleration:
Measurement and Analysis”, Marketing Science, 19 (4), 366–380
Van Den Bulte, Christophe, and Gary Lilien (1997), “Bias and Systematic Change in
the Parameter Estimates of Macro-Level Diffusion Models”, Marketing
Science, 16 (4), 338–353
Van Den Bulte, Christophe, and Gary Lilien (2001), “Medical Innovation Revisited:
Social Contagion versus Marketing Effort”, American Journal of Sociology,
106 (5), 1409–1435
Van Den Bulte, Christophe and Stefan Stremersch (2004), “Social Contagion and
Income Heterogeneity in New Product Diffusion: A Meta-Analytic Test”,
Marketing Science, 23 (4), 530–544
Van Den Bulte, Christophe and Yogesh V. Joshi (2007), “New Product Diffusion
with Influentials and Imitators”, Marketing Science, Forthcoming
Venkatesan, Rajkumar, Trichy V. Krishnan, and V. Kumar (2004), “Evolutionary
Estimation of Macro-Level Diffusion Models Using Genetic Algorithms: An
Alternative to Nonlinear Least Squares”, Marketing Science, 23 (3), 451–464
Wacziarg, Romain and Karen Horn Welch (2003), “Trade Liberalization and
Growth: New Evidence”, NBER Working Paper #10152, Last Downloaded
Jan 24 2007, URL: Http://Papers.Nber.Org/Papers/W10152
Wasson, Chester (1978), Dynamic Competitive Strategy and Product Life Cycles, 3d
Ed. Austin, TX: Austin Press
Wells, Louis T. (1968), “A Product Life Cycle for International Trade?”, Journal of
Marketing, 32, 1-6
Wilkes, Robert E., Burnett, John J., Howell, Roy D. (1986), “On the Meaning and
Measurement of Religiosity in Consumer Research”, Academy of Marketing
Science Journal, 14(1), 47-57
Webster’s New World College Dictionary (2004), 4th Edition, Cleveland: Wiley

203
Xie, Jinhong, Michael Song, Marvin Sirbu, and Qiong Wang (1997), “Kalman Filter
Estimation of New Product Diffusion Models”, Journal of Marketing
Research, 34, 378–393
Yeniyurt, Sengun and Townsend, Janell D. (2003), “Does Culture Explain
Acceptance of New Products in a Country? An Empirical Investigation”,
International Marketing Review, 20(4), 377-96
Zufryden, Fred S. (1996), “Linking Advertising to Box office Performance of New
Film Releases: A Marketing Planning Model”, Journal of Advertising
Research, 36 (4), 29–41
Zufryden, Fred S. (2000), “New Film Web Site Promotion and Box office
Performance”, Journal of Advertising Research, 40 (1/2), 55-64

Abstract (if available)

Abstract This dissertation examines how and why diffusion of new products varies across products, countries and time. The word product refers to a product category and not the brand. -- Chapter 1 gives an overview of the three essays that form part of the dissertation.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays on the role of entry strategy and quality strategy in market and consumer response

PDF

Essays on consumer conversations in social media

PDF

Essays on the competition between new and used durable products

PDF

Essays in retail management

PDF

Essays on product trade-ins: Implications for consumer demand and retailer behavior

PDF

Essays on the luxury fashion market

PDF

Understanding virality of YouTube video ads: dynamics, drivers, and effects

PDF

Scale construction and effects of brand authenticity

PDF

Quality investment and advertising: an empirical analysis of the auto industry

PDF

Empirical essays on alliances and innovation in the biopharmaceutical industry

PDF

Three essays on distance: examing the role of institutional distance on foreign firm entry, local isomorphism strategy and subsidiary performance

PDF

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

PDF

Essays on delegated asset management in illiquid markets

PDF

Essays on commercial media and advertising

PDF

Essays on supply chain management

PDF

Essays on interest rate determination in open economies

PDF

How product designs and presentations influence consumers’ post-acquisition decisions

PDF

The effects of a customer's comparative processing with positive and negative information on product choice

PDF

Essays on information and financial economics

PDF

Essays on the unintended consequences of digital platform designs

Asset Metadata

Creator Chandrasekaran, Deepa (author)

Core Title Essays on the turning points of the product life cycle

School Marshall School of Business

Degree Doctor of Philosophy

Degree Program Management

Publication Date 04/26/2009

Defense Date 03/09/2007

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

Tag diffusion,hazard models,innovation management,international marketing strategy,new product,OAI-PMH Harvest

Language English

Advisor Tellis, Gerard J. (committee chair), James, Gareth (committee member), Luo, Lan (committee member), Siddarth, S. (committee member)

Creator Email dchandra@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m455

Unique identifier UC1161690

Identifier etd-Chandrasekaran-20070426 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-493444 (legacy record id),usctheses-m455 (legacy record id)

Legacy Identifier etd-Chandrasekaran-20070426.pdf

Dmrecord 493444

Document Type Dissertation

Rights Chandrasekaran, Deepa

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