r
1:...
(J:

a.

Sap..." ..

i. .:

labvvuwp. 2

. ,.
a»... SW34:
I.-“~:I ‘
I :. .

‘3‘. 8‘
3.68.3.3
Hawthfiuéfie. r...
aI-ump . ‘Ufimfifi!

a 13.05 5
31.19“.
s!

4.2.. .. .
72.91....
16340:!
is. 23......
2.9.:
14....

 

 

2“} 131). ‘ ,. .45.... .
mama: flaggaw: max...

.2. 1.1:

 

. .51 . .-

 

“F93 LIBRARY
Mei); Mic: 2-. , State
University

 

 

 

This is to certify that the
thesis entitled

THE EFFECT OF REPEAT ADOPTERS UPON SECOND
GENERATION NEW PRODUCT ADOPTION

presented by
Kwong Chan

has been accepted towards fulfillment
of the requirements for the

Ph.D. degree in Business Administration

 

 

 

“apt/Professor’s Signature

August 29, 2006

 

Date

MSU is an Affirmative Action/Equal Opportunity Institution

 

.-.—-—.—o-c—o—o-u—.—-o-o—-¢-o--o-o-o—o—-—-—p—-—o-o-o—.—o—¢—o—o—--o—o---n—n-n-n—o-o—o—.—-—.-—o-n-a-o-o—o—a-.--—-.-o—.—-

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2/05 p:/C|RC/DateDue.indd-p.1

 

THE EFFECT OF REPEAT ADOPTERS UPON SECOND
GENERATION NEW PRODUCT ADOPTION

By

Kwong Chan

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
Department of Marketing and Supply Chain Management

2006

 

 

ABSTRACT

THE EFFECT OF REPEAT ADOPTERS UPON SECOND
GENERATION NEW PRODUCT ADOPTION

By

Kwong Chan

A second product generation is offered to two adopter populations: repeat
adopters, who consist of innovative adopters that purchased the first generation
product, and a larger market of imitative new-adopters that did not. New product
adoption theory suggests that the first adopter population plays a key role in
stimulating adoption behavior in the second imitative population. This
importance stems from credibility gained from first generation product
expenence.

In this study, social connections between adopters are measured using a
demographic similarity ratio, and piecewise Weibull regression used to analyze
cross-generation product adoption. The analysis used a secondary survey
dataset of new automobile owners.

The findings indicate purchase by new adopters of a second generation
product is accelerated by repeat adopters and to a lesser extent, new adopters.
Comparative analyses conducted on buyers of existing products indicate repeat
buyers have no influence upon new buyers, but for some existing products new

buyers stimulate other new buyers to early purchase.

— .‘Plfi

 

 

 

 

Copyright by
KWONG CHAN
2006

 

ACKNOWLEDGMENTS

My thanks to Dr Roger Calantone for your unwavering support of
research. Your guidance means no problem ever seems too small to be
interesting, or too large to be impossible. It is hard to find anything uninteresting
after knowing you.

Dr Jonathan Bohlmann, I thank you for your wit as much as your
guidance. As much as I tried to make a square into a circle you thankfully didn’t
let me. Dr Ram Narasimhan, thank you for your courses that are as much about
life as about modeling. I have benefited in the most unexpected ways from the

classes I took from you. Dr Tomas Hult, thank you for your help in formulating an

 

approach to the doctoral program and supporting my work.
I cannot say enough how my work has benefited from my doctoral cohort.
The line between work and life can disappear in the company of such motivated

people. It has been my pleasure to share our ideas for all these years.

 

TABLE OF CONTENTS

 

LIST OF TABLES ............................................................................................... vii
Chapter 1 ............................................................................................................. 1
INTRODUCTION ............................................................................................... 1
Word of Mouth Effects are Difficult to Capture in New Product Diffusion ....... 1
Summary ........................................................................................................ 3
Demographic Similarity and Internal Influence ............................................... 3
Context of the Investigation ............................................................................ 5
Chapter 2 ............................................................................................................. 6
LITERATURE REVIEW ..................................................................................... 6
Second Generation New Product Adoption .................................................... 6
Influence between Adopter Categories .......................................................... 8
Adopter Segments, Influence and Adoption ................................................... 9
Social Network Models of Diffusion .............................................................. 11
Similarity as a Measure of Influence ............................................................ 15
Summary ...................................................................................................... 18
Adopter Segments for Different Product Generations .................................. 19
Development of Hypotheses ........................................................................ 21
Chapter 3 ........................................................................................................... 23
METHODOLOGY ............................................................................................ 23
Data Collection ............................................................................................. 24
Selection of Product Model Generations ...................................................... 26
Generation Years versus Sale Years ........................................................... 27
Sample Descriptive Statistics ....................................................................... 33
Development of Measures ........................................................................... 4O
Dependent Variable: Year of Purchase ........................................................ 40
Adopter Groups: Susceptibility and Infectiousness ...................................... 41
The Demographic Similarity Ratio: Characteristics ...................................... 48
Empirical Model Estimation: Piecewise Weibull versus Cox Regression ..... 51
Regressions ................................................................................................. 55
Sensitivity Tests for the Demographic Similarity Covariates ........................ 58
Chapter 4 ........................................................................................................... 60
RESULTS ........................................................................................................ 60
Piecewise Weibull Regression versus Cox Regression ............................... 60
Comparing Between Degree Similarity Analyses, and Between Adopter
Group Analyses ........................................................................................... 89
Inferring New and Repeat Adopter Effects through Model Selection ........... 89
Sensitivity Tests ........................................................................................... 93
Hypothesis Testing ...................................................................................... 96

 

 

Chapter 5 ........................................................................................................... 99
New and Repeat Adopters ........................................................................... 99
Propensity .................................................................................................. 101
Sensitivity Results ...................................................................................... 101
Influence in Diffusion .................................................................................. 103
Comparison with an All Adopter Group ...................................................... 105
Comparing All Adopters versus New and Repeat Adopter Groups ............ 106
Comparison with Existing Products ............................................................ 108
Positive Influence Coefficients ................................................................... 112
Demographic Similarity as a Classification Tool ........................................ 113
Value to Managers ..................................................................................... 114
Study Limitations ........................................................................................ 115
Future Research Extensions ...................................................................... 116
Conclusion ................................................................................................. 1 18

APPENDIX A .................................................................................................... 120

Second Generation Automobiles Model Quarterly and Annual Sample Size
Distributions ................................................................................................... 1 20
APPENDIX B .................................................................................................... 124
Complete Second Generation Automobile Results for Piecewise Weibull
Regressions ................................................................................................... 124
REFERENCES .................................................................................................. 173

vi

LIST OF TABLES

Table 1: Variables in the Dataset ........................................................................ 25
Table 2: Vehicle Models Retained and Removed According to Insufficient
Repeat Adopter Sample Size .............................................................. 32
Table 3: Age Characteristics for New and Repeat Adopters ............................... 33
Table 4: Household Income Characteristics for New Adopters ........................... 34
Table 5: Household Income Characteristics for Repeat Adopters ....................... 35
Table 6: Education Categories for New Adopters ............................................... 35
Table 7: Education Categories for Repeat Adopters ........................................... 36
Table 8: Occupation Categories for New Adopters ............................................. 36
Table 9: Occupation Categories for Repeat Adopters ......................................... 37
Table 10: New Adopter Gender Characteristics .................................................. 38
Table 11: Repeat Adopter Gender Characteristics .............................................. 38
Table 12: New Adopter Race Characteristics ..................................................... 39
Table 13: Repeat Adopter Race Characteristics ................................................. 39
Table 14: Variable Properties for Demographic Characteristics .......................... 44
Table 15: Analyses by Group and Covariate ....................................................... 58

Table 16: Dodge Neon Piecewise Weibull Regression Results, New
Adopters Degree 5 Similarity .............................................................. 65

Table 17: Dodge Neon Piecewise Weibull Regression Results, New
Adopters Degree 4 Similarity .............................................................. 66

Table 18: Dodge Neon Piecewise Weibull Regression Results, New
Adopters Degree 3 Similarity .............................................................. 67

Table 19: Dodge Neon Piecewise Weibull Regression Results, Repeat
Adopters Degree 5 Similarity .............................................................. 68

Table 20: Dodge Neon Piecewise Weibull Regression Results, Repeat
Adopters Degree 4 Similarity .............................................................. 69

vii

Table 21: Dodge Neon Piecewise Weibull Regression Results, Repeat
Adopters Degree 3 Similarity .............................................................. 70

Table 22: Jeep Grand Cherokee Piecewise Weibull Regression Results,
New Adopters Degree 5 Similarity ...................................................... 71

Table 23: Jeep Grand Cherokee Piecewise Weibull Regression Results,
New Adopters Degree 4 Similarity ...................................................... 72

Table 24: Jeep Grand Cherokee Piecewise Weibull Regression Results,
New Adopters Degree 3 Similarity ...................................................... 73

Table 25: Jeep Grand Cherokee Piecewise Weibull Regression Results,
Repeat Adopters Degree 5 Similarity .................................................. 74 r

Table 26: Jeep Grand Cherokee Piecewise Weibull Regression Results,
Repeat Adopters Degree 4 Similarity .................................................. 75

Table 27: Jeep Grand Cherokee Piecewise Weibull Regression Results,
Repeat Adopters Degree 3 Similarity .................................................. 76

 

Table 28: Mazda Miata Piecewise Weibull Regression Results, New
Adopters Degree 5 Similarity .............................................................. 77

Table 29: Mazda Miata Piecewise Weibull Regression Results, New
Adopters Degree 4 Similarity .............................................................. 78

Table 30: Mazda Miata Piecewise Weibull Regression Results, New
Adopters Degree 3 Similarity .............................................................. 79

Table 31: Mazda Miata Piecewise Weibull Regression Results, Repeat
Adopters Degree 5 Similarity .............................................................. 80

Table 32: Mazda Miata Piecewise Weibull Regression Results, Repeat
Adopters Degree 4 Similarity .............................................................. 81

Table 33: Mazda Miata Piecewise Weibull Regression Results, Repeat
Adopters Degree 3 Similarity .............................................................. 82

 

Table 34: Toyota Avalon Piecewise Weibull Regression Results, New
Adopters Degree 5 Similarity .............................................................. 83

Table 35: Toyota Avalon Piecewise Weibull Regression Results, New
Adopters Degree 4 Similarity .............................................................. 84

Table 36: Toyota Avalon Piecewise Weibull Regression Results, New
Adopters Degree 3 Similarity .............................................................. 85

viii

Table 37:

Table 38:

Table 39:

Table 40:

Table 41:

Table 42:

Table 43:

Table 44:

Table 45:

Table 46:

Table 47:
Table 48:

Table 49:

Table 50:

Table 51:

Table 52:

Table 53:

Toyota Avalon Piecewise Weibull Regression Results, Repeat

Adopters Degree 5 Similarity .............................................................. 86
Toyota Avalon Piecewise Weibull Regression Results, New

Adopters Degree 4 Similarity .............................................................. 87
Toyota Avalon Piecewise Weibull Regression Results, New

Adopters Degree 3 Similarity .............................................................. 88
Identified Prior New and Repeat Effects upon Current New and

Repeat Adopter Adoption Time ........................................................... 92
Demographic Degree Similarity Sensitivity Test Results for New
Adopters, Dodge Neon and Jeep Grand Cherokee ............................ 94
Demographic Degree Similarity Sensitivity Test Results for New
Adopters, Mazda Miata and Toyota Avalon ........................................ 94
Demographic Degree Similarity Sensitivity Test Results for

Repeat Adopters, Dodge Neon and Jeep Grand Cherokee ................ 95
Demographic Degree Similarity Sensitivity Test Results for

Repeat Adopters, Mazda Miata and Toyota Avalon ............................ 95
Naive Analyses, Identified ALL Effects upon All Adopter

Adoption Time ................................................................................... 105
Expanded Naive Analyses, Identified NEW and REPEAT Effects

upon AII Adopter Adoption Time ....................................................... 106
Existing Automobile Models Analyzed ............................................... 109

Existing Products: Identified Prior New and Repeat Effects upon
Current New and Repeat Adopter Adoption Time ............................. 110

Demographic Similarity Coefficient Effects of Prior Adopters
upon New Adopter Time to Adoption ................................................ 112

Dodge Neon New Adopter Sample Size by Survey Quarter and .
Calendar Year ................................................................................... 121

Dodge Neon Repeat Adopter Sample Size by Survey Quarter
and Calendar Year ............................................................................ 121

Jeep Grand Cherokee New Adopter Sample Size by Survey
Quarter and Calendar Year ............................................................... 121

Jeep Grand Cherokee Repeat Adopter Sample Size by Survey
Quarter and Calendar Year ............................................................... 122

Table 54:

Table 55:

Table 56:

Table 57:

Table 58:

Table 59:

Table 60:

Table 61:

Table 62:

Table 63:

Table 64:

Table 65:

Table 66:

Table 67:

Table 68:

Table 69:

Mazda Miata New Adopter Sample Size by Survey Quarter and
Calendar Year ................................................................................... 122
Mazda Miata Repeat Adopter Sample Size by Survey Quarter

and Calendar Year ............................................................................ 122

Toyota Avalon Legacy New Adopter Sample Size by Survey

Quarter and Calendar Year ............................................................... 122

Toyota Avalon Repeat Adopter Sample Size by Survey Quarter

and Calendar Year ............................................................................ 123
Dodge Neon Piecewise Weibull Regression for Degree

Similarity 5 Covariates ...................................................................... 125
Dodge Neon Piecewise Weibull Regression for Degree

Similarity 4 Covariates ...................................................................... 129
Dodge Neon Piecewise Weibull Regression for Degree

Similarity 3 Covariates ...................................................................... 133

Jeep Grand Cherokee Piecewise Weibull Regression for Degree
Similarity 5 Covariates ...................................................................... 137

Jeep Grand Cherokee Piecewise Weibull Regression for Degree
Similarity 4 Covariates ...................................................................... 141

Jeep Grand Cherokee Piecewise Weibull Regression for Degree
Similarity 3 Covariates ...................................................................... 145
Mazda Miata Piecewise Weibull Regression for Degree

Similarity 5 Covariates ...................................................................... 149
Mazda Miata Piecewise Weibull Regression for Degree

Similarity 4 Covariates ...................................................................... 153
Mazda Miata Piecewise Weibull Regression for Degree

Similarity 3 Covariates ...................................................................... 157

Toyota Avalon Piecewise Weibull Regression for Degree

Similarity 5 Covariates ...................................................................... 161

Toyota Avalon Piecewise Weibull Regression for Degree

Similarity 4 Covariates ...................................................................... 165

Toyota Avalon Piecewise Weibull Regression for Degree

Similarity 3 Covariates ...................................................................... 169

Chapter 1
INTRODUCTION

Word of Mouth Effects are Difficult to Capture in New Product Diffusion

In new product diffusion, word of mouth describes communication
between individuals in an adopting population. This communication occurs when
individuals view product performance as uncertain and consequently utilize
interpersonal contacts to gather relevant information. Widespread dissemination
of information can accelerate adoption (Chatterjee and Eliashberg 1990), while a
lack of such information flowing through interpersonal channels can lead to
delayed market acceptance (Goldenberg, Libai and Muller 2002) and even
market failure (Moore 1991 ). The importance of word of mouth is reflected in
work that examines disproportionately influential individuals who may play a key
role in catalyzing an adopting population to purchase (Rogers 1995; Feick and
Price 1991).

Most diffusion models are able to account for word of mouth effects, and
the majority of approaches are able to yield sales forecasts through the use of a
mathematical functional form. The Bass Diffusion Model (Bass 1969) for
example is defined by a market potential estimate, and coefficients of external
and internal influence. The coefficient of internal influence captures the result of
word of mouth effects, but this coefficient yields little insight into the mechanisms
by which influence travels through a population. Purchase times are used as the
basis for model estimation, however the times at which different individuals

receive and transmit influence remain unknown. Diffusion models are thus

largely restricted to the realm of forecasting, and provide only limited guidance
for marketing communications practice. An improved understanding of the
processes that underlie the coefficient of internal influence would be invaluable to
managers that wish to utilize interpersonal communication channels in marketing
programs.

In contrast to diffusion studies, network models of diffusion focus primarily
upon the mechanisms by which information flows through individuals. Forecasts
of effect (such as sales rates), if possible, are mere by-products of network
models. Relative to diffusion models, data requirements for network models are
onerous. Relational models characterize the opinion leadership of an individual
by consulting all other individuals in the network, structural models use patterns
of present and absent ties to compare the network positions of network
members (Valente 1995), while models of network spatial heterogeneity (Strang
and Tuma 1993) require assessment of infectiousness, susceptibility, proximity
and propensity at the individual level.

Clearly network models provide great insight into diffusion processes
through individual-level relationships, behavior and characteristics. Equally clear
is the comprehensive level of data needed to undertake network modeling. This
greater data need alone has arguably resulted in a predominance of aggregate-
level diffusion research that uses more accessible population-level sales data.
This leaves a large middle-ground where data may be less rich than social
network data, but more comprehensive than aggregate level sales data.

Methods of broaching this gap such as through the use of demographic similarity

(De Bruyn and Lilien 2004) have begun to appear, although no benchmark

approach currently exists.

Summary

Past work has explored adoption patterns at a broad population level more
suitable for sales forecasts than specific marketing policy, or at a network level
where data requirements exceed those typically available in a new product E‘
diffusion context. There is a need to achieve greater insight at the individual

level in a broader market level adoption context.

 

Demographic Similarity and Internal Influence

This research examines internal influence in new product diffusion by
using a social network rationale to apply demographic similarity as a proxy for
internal influence, and link this proxy to adoption time. This measure assesses
internal influence effects upon a potential adopter by determining the proportion
of past adopters who have common demographic characteristics.

The importance of the social network perspective to new product diffusion

is used to highlight the tradeoff between the need for empirical detail and

 

difficulty in obtaining data of sufficient richness. An approach is proposed and
tested that uses a demographic similarity ratio to link past and present adopters
across and within two product generations. By examining the same individuals
over two successive product generations, a method of identifying adopter

influence is developed that examines the effects of “prior adopters” that have

already purchased, upon “current adopters” that have yet to. Prior adopters
consist of both first-time adopters of a second generation product who are
termed prior New Adapters; and more experienced adopters who have
purchased bath the first and second product generations and are termed prior
Repeat Adapters. Current adopters have yet to purchase and may be potential
first-time adopters who are termed current New Adapters, or potential repeat
adopters who are termed current Repeat Adapters.

Specifically, this research uses a second successive automobile product
generation context to identify the importance of 1m New and Repeat Adapters

in acceleration of second generation product adoption among current New and

 

Repeat Adapters.

The research questions are:
1) Is purchase by current New Adapters accelerated by influence from both
prior New Adapters and prior Repeat Adapters?
2) Is purchase by current Repeat Adapters accelerated by influence from
both prior New Adapters and prior Repeat Adapters?
These questions are investigated through examination of repeat and new
adapters over time. The results indicate that New Adapters do respond to both

prior New and Repeat Adapters, whereas Repeat Adapters respond to neither.

Context of the Investigation

Secondary data were obtained from a market research firm that collected
national level survey response data for buyers of new automobiles for the period
September 1998 to December 2004. The data contain information covering two
successive generations of automobile purchase by an individual. This allows for
the examination of repeat and new adopters as possible senders and receivers

of influence in a product adoption context.

Chapter 2
LITERATURE REVIEW

This chapter contrasts aggregate and network models of diffusion by examining
how each investigates segments of adopters. The information rich, but relatively
data intensive qualities of the network approach are highlighted, and a

compromise approach suggested.

This research aims to develop a cross-generational approach to describe how
repeat and new adopter segments influence one another during adoption of a
second product generation. To develop a conceptual model, prior work that links
adapters across segments, and overtime, is reviewed over three sections. The
first section examines how the current state of cross-generational new product
research views adopters of different generations as exerting influence upon one
another; the second section examines how aggregate diffusion models and
individual-level social network perspectives describe inter-adopter influence; and
the third section proposes a means of linking adopter segments through the use

of demographic similarity.

Second Generation New Product Adoption

The competitive significance of new products to firms has led to a desire
to understand and predict new product adoption in the marketplace. However
despite a general agreement that early market entry is advantageous (Bowman

and Gatignon 1996; Symanski, Troy and Bharadwaj 1995; Kalyanaram and

Urban 1992), a successful initial first generation market entry alone does not
guarantee long-term competitiveness or profitability (Boulding and Christen 2003;
Bohlmann, Golder and Mitra 2002; Venkatesh, Carpenter and Krishnamurthi
1998). The problem of predicting the adoption of successive product generations
has received less attention, but is no less important, as competitor entry and
mass market acceptance often occur during adaption of the second, or later
product generations (Christensen 1997). Furthermore, it is common for a second
generation product launch to involve greater resource commitment because of
the need to recoup fixed development costs, expand production capacity and
broaden marketing effort.

Existing cross-generational product diffusion studies have focused upon
market level adoption rather than the individual level communication processes
that are theorized to underlie market acceptance of new products. The
forecasting benefits of market level aggregate studies is clear, however it is the
nature of the underlying word of mouth processes that is most informative for
development of interpersonal influence theory and managerial action.
Unfortunately it has proven far easier to observe the aggregate market result of
such communication than identify individual-level behaviors that drive word of
mouth influence.

The relatively few studies in multigenerational new product adoption have
focused upon identification of differences in the diffusion patterns between
product generations. Differences may be due to technological (Kim, Srivastava

and Han 2001), marketing (Danaher, Hardie and Putsis 2001), related-product

category (Kim, Chang and Shocker 2000) or user-related cross-generational
influence factors (Danaher, Hardie and Putsis 2001, Islam and Meade 1997).
Consideration of interpersonal influence has focused upon the role
adopters of a newer generation play in both attracting completely new adopters
of the product, and convincing adopters of prior generations to re-adopt the
newer generation (Danaher, Hardie and Putsis 2001; Kim, Chang and Shocker
2000; Norton and Bass 1987), or the ability of the installed base of an earlier
technology to enhance diffusion of a later technology generation (Kim, Chang
and Shocker 2000; Islam and Meade 1997). These studies treat adopters of
different generations as different adopter groups, and indicate influence can be
described between adopters of succeeding and prior generations through distinct

generational segments.

Influence between Adapter Categories

Adapters may be categorized as “innovators” and “imitators”. Innovators
make adoption decisions largely independent of others, while imitators are
heavily influenced by other peole (Bass 1969). Innovators rely proportionally
more upon external sources of information such as advertising and critical
reviews, have a lower risk threshold (Valente 1995), and are opinion leaders for
following adopters (Rogers 1995). lmitators are a larger proportion of the market
(Muller and Yogev 2005) and are thus critical to market success. In contrast to
innovators, imitators pay great attention to those around them when making

adoption decisions. While innovators are key in influencing following adopters,

 

 

their impact may vary according to different innovator types. Earlier innovators
have personal and social characteristics that differ from later innovators, having
lower risk adversity, more extreme behavior, and fewer social connections. Later
innovators tend to be more socially connected, and may act as ‘market mavens’
who influence adoption among imitative adopters (Feick and Price 1987). A lack
of influence between innovative and imitative segments may be responsible for
the “chasm” in some diffusion curves that may indicate product failure (Moore
1991) or the “saddle” that is associated with delayed market acceptance
(Goldenberg, Libai and Muller 2002). The importance of influence between
innovative and imitative segments has led to investigations of diffusion dynamics

across market segments.

Adopter Segments, Influence and Adoption

The most utilized diffusion model in new product adoption is the Bass
Diffusion Model (Bass 1969) that requires just three parameters for estimation:
market potential, a coefficient of internal influence, and a coefficient of external
influence. Internal influence is analogous to the word of mouth effect, while
external influence is often interpreted as a measure of media effects. Each
parameter may be interpreted as a summary result of underlying mechanisms.
Differences in parameter coefficients indicate faster or slower adoption rates, but
do not reveal whether differences are due to factors such as product features or

characteristics of the adopting population. The desire to identify specific reasons

for different adoption rates has led to a stream of work that examines new
product diffusion across contexts and time.

The use of diffusion models to investigate processes that generate the
adoption curve have focused upon examination of model parameters across
different market contexts, including product-markets (T alukdar, Sudhir and
Ainslie 2002; Lenk and Rao 1990) and country-markets (T alukdar et al. 2002;
Dekimpe, Parker and Sarvary 1998; Gatignon, Eliashberg and Robertson 1989).
Talukdar et al. (2002) provide evidence internal influence varies according to
ethnic diversity and experience of prior adopters that are found to be positively
related to internal influence. This class of research provides evidence adopter
characteristics are heterogeneous between adoption populations. Differences in
internal influence have also been investigated in different populations over time.

The possibility that influence flows from earlier to later segments has been
examined in the context of early and late adoption in different country markets.
This so-called ‘lead-Iag’ effect (Ganesh and Kumar1996; Kalish, Mahajan, and
Muller 1995; Takada 1991) varies according to factors such as social mixing and
geographic proximity (Ganesh, Kumar and Subramaniam 1997; Putsis,
Balasubramanian, Kaplan and Sen 1991), although it has proven difficult to
consider lead-lag and mixing effects in the same framework. An exception is
Kumar and Krishnan (2002) who applied the Generalized Bass Model (GBM) to
simultaneously consider lead-lag and mixing effects, however a key limitation
arising from the use of the GBM is the need for mixing to influence both early and

late adopters identically, negating the possibility of testing for differences in

10

characteristics across adopter types. It is also difficult to examine detailed
diffusion processes using country-based segments, as adapters in different
countries may behave along dimensions that transcend political borders
(Hofstede, Wedel and Steenkamp 2002).

These studies investigate influence between distinct populations rather
than endogenous effects of earlier adopters upon later adopters within the same
population. Such variation provides support for heterogeneous processes for
internal influence across but not within market segments, because population-
level diffusion models preclude examination of differences across individuals
within the same segment (Strang and Tuma 1993). Despite these limitations,
research across segments and overtime has demonstrated the utility of
considering adaption timing and inter-segment proximity as important
determinants of new product adoption. In an effort to explicitly examine
heterogeneous internal influence processes, social network models have been
used to describe differences in the behavior of individuals during adoption of an

innovafion.

Social Network Models of Diffusion

By allowing each individual adapter to have their own pattern of
relationships, social network models of diffusion treat each individual as a distinct
“market segment”. Aggregation of these individuals yields collective behavior

that may empirically resemble a diffusion curve (Strang and Tuma 1993). Two

11

 

perspectives to modeling social relationships are relational and structural
networks (Valente 1995).

Diffusion effects can be explained from the relational network perspective
that utilizes information concerning the nature of relations between people.
Inward and outward communication to and from individuals can be mapped
through nominations of alters (people other than the individual of focus) in the
network with whom communication occurs (e.g. Rogers and Kincaid 1981;
Coleman, Katz and Menzel 1966; Rogers and Beal 1958). Frequent nominations
of an individual by network alters is indicative of strong influence, and early
adoption of an innovation by individuals of this type can facilitate the linked alters
to subsequently adopt. Nominations made by an individual toward others reflects
connectedness to the network and suggests influence is received earlier,
although empirical evidence here is mixed (Valente 1995:37). The relational
network perspective thus emphasizes the role of an adopter’s immediate
personal network in explaining adoption behavior.

The structural network perspective is an alternative view that considers
the architecture of relations across all adopters both immediate and distant in the
network, and thus focuses upon the pattern and strength of relations rather than
their nature (Valente 1995). Networks that are highly centralized feature groups
of individuals who are densely connected within a group, but loosely connected
to other groups. These networks are not conducive to rapid diffusion in a
population because information remains confined in network cliques. In contrast,

diffusion may be enhanced in a network when weak ties act as bridges that

12

connect relatively distant network groups (Granovetter 1983). The relational and
structural views highlight in different ways, the role of micro-level adopter
connections, and macro-level network features in diffusion.

The relationships that characterize each individual may be thought of as
analogous to the covariates used in aggregate level diffusion research. At the
aggregate level, covariates influence adoption, while at the individual level it is
relationships with others that influence adoption. Social network models however
do not capture the notion of path dependent influence overtime. Relationships,
and therefore influence, are assumed static in most social network diffusion
studies, while in aggregate diffusion research, influence varies according to the
number of individuals that have adopted at a given point in time. A unified
longitudinal explanation for variation in relational effects attributed to inward and
outward flows of information, and the network structure that facilitates these
flows, is provided by heterogeneous diffusion models.

In a heterogeneous diffusion model, susceptibility, infectiousness,
proximity, and propensity provide a rich description of the impact behavior in a
social system has upon current individuals (Greve, Tuma and Strang 2001 ).
Susceptibility is sensitivity to network influence, infectiousness refers to the ability
to influence other network members, proximity describes the social closeness
between a focal individual and network alters, and propensity is the intrinsic
tendency to adopt independent of network influence. Formulated by Strang and
Tuma (1993), the four dimensions of this approach capture both individual-

specific and network-derived determinants of adoption. The contagion concepts

13

 

 

of susceptibility and infectiousness are analogous to the relational network
concepts of ‘connectedness’ (nominations made), and ‘opinion leadership’
(nominations received) respectively, and proximity is analogous to the structural
network concept of tie strength. Propensity is consistent with the concept of an
individual’s risk-return profile (e.g. Chatterjee and Eliashberg 1990), and also
captures the influence of external sources such as media communication. While
social network models view network connections as static, heterogeneous
diffusion models incorporate temporal ordering and can model time varying
network conditions on the basis of adoption behavior. Each adoption event alters
network influence conditions that affect those that have yet to adopt.

The ability to monitor transmission of influence concurrent with these four
contagion dimensions facilitates accurate attribution of cause and effect, but
increases data collection requirements relative to aggregate level diffusion
studies. Determination of inward network connections, outward connectedness,
the strength of each of these, together with the risk-utility profile for every adopter
can be difficult to achieve in a new product adoption context. This is reflected in
the relatively small sample sizes used in social network research compared to
market level diffusion studies. Three of the most famous studies in network
diffusion: medical innovation (Coleman, Katz and Menzel 1966), Korean family
planning (Rogers and Kincaid 1981) and Brazilian farmers (Rogers, Ascroft and
Roling 1970), use sample sizes of 125, 1047 and 692 respectively (Valente
1995). The three population level products examined in Mahajan, Mason and

Srinivasan (1986): room air conditioners, color televisions, and clothes dryers

14

have sample sizes of approximately 14.4 million, 32.3 million, and 12.5 million
units respectively.

The heterogeneous diffusion model is a form of diffusion model that is
able to incorporate social network features that aggregate population-level
models generally do not. These enhancements allow for a more detailed
attribution of effects than traditional diffusion studies that assume homogeneity of
adopter characteristics. Nevertheless, data requirements make it difficult to
apply to aggregate level studies that consider sample sizes several orders of
magnitude greater than those normally used in social network research. An
intermediate approach to measurement of network relationships is the use of

common characteristics among adopting individuals.

Similarity as a Measure of Influence

Network theory has been particularly informative in explaining the
mechanics of diffusion, providing evidence that information travels through a
population according to both the number and type of linkages between
individuals in a social network (Valente 1995; Burt 1987; Granovetter 1973).
According to models of diffusion, an individual’s decision to adopt an innovation
is based upon the behavior of others that are socially proximal and to which they
have homophilous ties (Rogers 1995). These individuals may adopt according to
the behavior of others for reasons of uncertainty reduction (Coleman, Katz and
Menzel 1966) or status conformity (Burt 1987). While these two concepts deal

with information collection and imitation respectively, both are based upon social

15

 

 

proximity. Individuals reduce their uncertainty toward a new product by
consulting and gathering information from people that are socially close.
Similarly, individuals aim to conform to people with whom they compare
themselves - people that are also socially similar. Thus while uncertainty
reduction relies upon consulting close contacts, and social conformity relies upon
imitating one’s social competitors (Dimaggio and Powell 1983), both rely upon
influence from others that are socially proximal.

Common demographic attributes provide an avenue for measuring social
proximity between members of a network. Organizations may imitate strategies
of firms with similar resource endowments (Greve 1998), government institutions
have been observed to act following prior actions by peer institutions of similar
political orientation (Soule and Zylan 1997), and individuals may adopt a product
upon observation of purchase by others with whom they identify. Mimicry of this
kind occurs when an imitating party views the behavior of an acting party as
similarly relevant to their own circumstance. This observation by potential
adopters of prior adopters is an example of direct influence arising from
homophily in social characteristics (Rogers 1983). Such similarity in social
characteristics is also associated with indirect influence.

Influence by virtue of category membership can occur even when no direct
relationships exist. “Cultural linkage” (Strang and Meyer 1993) or “nonrelational
channels of diffusion” (McAdam and Rucht 1993) refer to avenues by which
socially unconnected actors often behave similarly (Soule and Zylan 1997).

Similar behavior across unconnected individuals can arise due to similar

16

circumstance dictating congruent needs. Organizations endowed with similar
resources that face common competitive pressures may adopt familial strategies
even in the absence of direct inter-organizational relationships. An explanation
for such “mimetic adoption” (Greve 1998) is provided by the concept of structural
equivalence across multiple individuals (Burt 1987), the presence of which
facilitates diffusion amongst these structurally equivalent actors. Actors in similar
network positions may adopt similar innovations, due to observance or path
dependence. In observance, analysis of alters in similar network positions can
substitute for an actor’s own grounded information search (Bikhchandani,

Hirshleifer and Welch 1992). The behavior of a network alter may thus be

 

imitated as a resource-saving action that reduces decision making time and
effort. However path dependence requires no such observation as distinct actors
with similar histories may independently innovate similar solutions to a problem -
a phenomenon commonly cited by academics and artists as an alternative
explanation to plagiarism when pieces of work asserted as different bear an
uncanny resemblance to each another.

In the context of new product adoption, social proximity can be used to E
assess path dependence in the diffusion process. Every adopter’s decision is

dependent upon the behavior of prior adopters, with socially close prior adopters i“

 

having greater influence than more socially distant prior adopters. An absence of
socially close prior adopters does not necessarily prevent adoption by

subsequent adopters, but instead delays the decision to adopt while imitative

17

influences are transmitted through more heterophilous indirect social network
channels.

There are therefore two reasons to consider common attributes of
individuals in the investigation of diffusion mechanics: the known propensity for
direct influence to travel directly between individuals with interpersonal
connections that are often associated with similar demographic characteristics,
and the tendency for unconnected parties to behave similarly due to common

needs and resource endowments.

Summary

Segments do not behave in isolation of one another. Diffusion studies
capture this through population-level parameter variance, and social network
studies capture this through atomistic relationships between individual adopters.
Diffusion studies provide valuable forecasts of product adoption but offer limited
insight into the mechanisms that underlie diffusion. Social network studies
provide a detailed analysis of the diffusion process, but require highly detailed
data that confines studies to small sample sizes and specific contexts. This data
barrier has lead to a presumption of a static network structure. The
heterogeneous diffusion model incorporates dynamic social network relations
into the traditional aggregate level diffusion approach, but at the cost of importing
the need for detailed individual-level network data. An intermediate avenue for
directly measuring social network relations is the use of common characteristics

among adopting individuals to proxy for hypothesized direct and indirect

18

influence. The following section places inter-segment influence into a cross-

generational context.

Adopter Segments for Different Product Generations

Prior work has considered individuals who order prior to launch as
innovative relative to post launch buyers (Mae and Fader 2002), and by
extension it is also reasonable to view adopters of a first generation product as
innovative, and adopters of a second generation product as imitative market
segments. By definition the imitative segment is influenced by the innovative
segment, and in general, innovative adapters are more influential in enhancing
diffusion (Midgley, Morrison and Roberts 1992; Czepiel 1975), however some
innovators such as “market mavens” (Feick and Price 1991) are more influential
than others.

Heterogeneity in a second product generation market may be modeled
through two populations: repeat adopters that have already purchased the
previous first generation product, and new adopters that have not. These are
referred to as generation 1 and generation 2 adopter populations respectively. It
is useful to consider these two populations separately because their adoption
behaviors are influenced by different factors, and the adoption behavior of each
population can influence adoption behavior in the other differently. An
understanding of these within- and between- population dynamics can inform us
of how successive generations of a new product diffuse when some adopters

have prior product experience and others do not.

19

 

Generation 2 adopters have no prior experience with the first generation
product. These adopters may have been exposed to the first generation product
through market or word-of-mouth information, but they did not purchase the first
generation product. This lack of purchase is the key distinction between
generation 1 and 2 adopter populations. Because they did not purchase the first
product generation, generation 2 members are considered less innovative than
generation 1 members. Their decision-making threshold is on average higher
than that of generation 1 members, and their information comes from fewer
external (media) and more internal (interpersonal) sources. Generation 2
adapters are more imitative than generation 1 adapters and their adoption
decision is therefore more reliant upon consultation with others.

During adoption of the first product generation, generation 1 members
follow a diffusion process for the same reasons generation 2 members follow a
diffusion process. However when buying the second generation, generation 1
members differ in adoption behavior because they possess product knowledge
gained from experience. For these veteran adopters of the second generation
product, the decision to purchase is based less upon marketing information and
behavioral imitation, and more upon marginal utility from comparisons of different
product generations. Consistent with diffusion, adoption of the second
generation product still follows a hazard process, however its form is dependent
upon adoption decisions arising from judgments of marginal utility provided by
the second generation product, rather than the network effects that influence the

generation 2 population.

20

 

Development of Hypotheses

Adapters of a second generation product are of two kinds: New Adapters
who buy generation 2 but not generation 1, and Repeat Adapters who buy
generation 2 and have also previously purchase generation 1. New Adapters
rely upon prior adapters for advice more than Repeat Adapters, whose greater
experiential knowledge lessens their need to seek information from the social
network. Repeat Adapters are thus less susceptible to influence than New
Adapters. The same experiential knowledge that makes Repeat Adapters less
susceptible also makes them more influential. Experienced Repeat Adapters
have greater knowledge-based credibility than New Adapters. Influence from
Repeat Adapters is therefore stronger than influence from New Adapters, and
Repeat Adapters may be said to be more infectious than New Adapters. These

assertions are stated formally in Hypotheses 1 to 5:

Hypothesis 1. For second generation products, purchase by New Adapters

is accelerated by social connection to prior New Adapters.

Hypothesis 2. For second generation products, purchase by New Adapters

is accelerated by social connection to prior Repeat Adapters.

Hypothesis 3. For second generation products, purchase by New Adapters
is accelerated more by social connection to prior Repeat

Adapters, than social connection to prior New Adapters.

21

Hypothesis 4. For second generation products, purchase by Repeat
Adapters is not accelerated by social connection to prior

New Adapters.
Hypothesis 5. For second generation products, purchase by Repeat

Adapters is not accelerated by social connection to prior

Repeat Adapters.

22

Chapter 3

METHODOLOGY

This chapter presents the hypotheses to be tested, data collection, product
model generation selection, development of an influence measure, and the

approach used for empirical estimation of the model.

A cross-generational product context was used to examine differences in word of

mouth influence and adoption time response between new and repeat adopters.

5 hypotheses were tested:

Hypothesis 1. For second generation products, purchase by New Adapters

is accelerated by social connection to prior New Adapters.

Hypothesis 2. For second generation products, purchase by New Adapters

is accelerated by social connection to prior Repeat Adapters.

Hypothesis 3. For second generation products, purchase by New Adapters
is accelerated more by social connection to prior Repeat

Adapters, than social connection to prior New Adapters.

23

 

., f" I.

 

Hypothesis 4. For second generation products, purchase by Repeat
Adapters is not accelerated by social connection to prior

New Adapters.

Hypothesis 5. For second generation products, purchase by Repeat
Adapters is not accelerated by social connection to prior

Repeat Adapters.

The methodology followed four phases: (1) data collection; (2) selection of
product model generations; (3) definition of measures for the heterogeneous
diffusion concepts of infectiousness, susceptibility, propensity, and proximity; and

(4) empirical model estimation.

Data Collection

A secondary new passenger car dataset was obtained from a market
research firm. Passenger cars are a separate vehicle category from “light
trucks”. This means vehicle types such as pickup trucks and many sports utility
vehicles are not contained within the dataset. The distinction between passenger
cars and light trucks has become less distinct in recent years. Sports utility
vehicles for example are variously classified as passenger cars or light trucks.
For future studies it is important to note this lack of discrimination is likely to

increase as manufacturers expand hybridized models that include technical

24

features of both passenger cars and sports utility vehicles (and even pickup
trucks).

This dataset consisted of survey response data gathered from individuals
who purchased new passenger vehicles within the time period spanning
September 1998 to December 2004. The market research firm compiled mailing
lists from national-level vehicle registration data across all states of the United
States. The protocol utilized stamped, self-addressed envelopes, and provided
assurance that no sales solicitation would result from survey response. The
survey instrument included questions concerning current and most recent vehicle
ownership, and demographic information. The data were deemed suitable for
this research because of respondent information concerning both current and
previous automobile ownership allowing for cross-generational investigation.
The subset of variables obtained from the data and used in the research are

shown in Table 1.

Table 1: Variables in the Dataset

 

Purchase Variables Demflraphic Variables
Calendar Date of Purchase Household Income
Vehicle Make Occupation

Vehicle Model Year Level of Education
Vehicle Size Classification Gender

Date new automobile purchased Marital Status

Brand of automobile Household Size
Automobile model year Age

Size class of automobile Race

Previously driven automobile model Geographic State

Previously driven automobile model year
Model and year of second household vehicle
Model and year of third household vehicle

 

25

The sampling strategy applied by the research firm used target sample size
requirements. In general, more popular vehicle models were sampled more
heavily than less popular models, although differences in target sample sizes are
not proportional to market level differences in sales from year to year. While
market level sales for an automobile model may vary overtime, target sample
sizes remain largely unchanged. Therefore, sample sizes from year to year do
not reflect national level vehicle sales levels.

Surveys were administered on a quarterly basis and target samples for the
first and second quarters of each year are double those of the third and forth
quarters. Samples were drawn randomly from the database. If a target sample
size was not reached in a quarter, further random samples were drawn and
additional surveys mailed until the target sample size for the quarter was
approached. A breakdown of quarterly and annual sample distributions for each
second generation automobile model used in the study are presented in

Appendix A.

Selection of Product Model Generations

Branding and manufacturing practices in the industry result in automobile
models that are often related in content and lack technological distinction from
one another. The same components may underlie mass-market and prestige
versions of the same model. Such models are almost mechanically identical but
for branding purposes they are targeted at different segments. For example the

Ford Taurus and Mercury Sable are mainstream and prestige versions of

26

automobiles that have most of their technology in common. Consistent with the
market segmentation intentions of manufacturers that design these models to
appeal to different market segments, these models are treated as distinct
automobiles for the purposes of this study.

A further complication arises from the practice of rebadging vehicles. For
example the Geo Tracker, was also sold as the Chevrolet Tracker, and as the
GMC Tracker, and in turn these models were re-branded from the original Suzuki
Escudo vehicle model. Because customers may conceivably cross shop such
automobile models within and between generations, these “twin” automobile
models were excluded from this study if they were sold under different brand
names. While such automobiles are not viewed by consumers as perfectly
identical, consumers do perceive substantial similarity between rebadged

automobiles (Sullivan 1998).

Generation Years versus Sale Years

To investigate the complete adoption process, a subset of the data were
used for motor vehicles whose most recent model generation began and finished
within the 6 year window 1999-2004. The model year rather than the sales year
of the vehicle was the basis for automobile model generation designation as
model years reflect manufacturer changes in product content. Because a
particular model year is usually available late in the prior calendar year (1999
model years for example are usually first sold in 1998), the number of sale years

is often one year greater than the number of model generation years. Thus for

27

 

 

inclusion in the sample, individuals must have purchased a vehicle model whose
model generation began and finished within the years 1999 to 2004. This
distinction is necessary because while model years are used to designate a
product as generation 1 or generation 2, it is the sales year of adoption that is the
dependent variable.

Automobile model generations used in this research are defined by the
manufacturer, and usually consist of substantial cosmetic and mechanical
changes from a prior generation. Not all succeeding generations are ‘all new’ as
most models between generations share components. Global manufacturing
and product development means this is also often true between vehicle models
and even between different manufacturers. Practices such as these led to the
choice of using manufacturer definitions for model generations, rather than
attempt to create an index of new technology content to distinguish between
generations.

Within product generations, manufacturers usually incorporate minor
technology content updates. Because continual feature incorporation is the
benchmark practice in the industry, these updates can be viewed as
maintenance of a product’s quality and feature content rather than improvements
or generational changes. This is consistent with consumer perception that
vehicles from different model years are poor substitutes for one another
(Copeland, Dunn and Hall 2005).

This research is concerned with second generation products. For these

products, the first generation did not replace a prior vehicle generation of the

28

same or different brand name, and were built on vehicle platforms different from
any related current or preceding model at the time of first generation product
launch. Automobiles preceded by more than one generation are classified as
existing products, and are not the focus of this study.

“G2” is the designation for models of the most recent generation. “G1” is
the designation for models of the generation immediately prior. Based upon the
requirement that the G2 sales period fall within 1999-2004 inclusively, 6
automobile models were isolated for use in this study.

This study calculates an influence ratio and uses the resultant measure
together with additional propensity covariates in regression analyses. Because
of missing values for some of these propensity covariates, some individuals
could not be used in regression analyses, but were still retained for calculation of
the influence ratio. Regression analyses thus utilize a sub-sample of individuals
who have a complete set of propensity covariates.

The proposed measure of influence in this study (defined later in this
chapter) is calculated from the number of similar and dissimilar prior adopters at
yearly intervals. In the current sample repeat adopter sample size is always
substantially lower than that of new adapters, and any micronumerosity sample
size issues arise from repeat adopter restrictions. Because the proposed
influence measure is a ratio of individuals, and adapters are classified
dichotomously as either similar or dissimilar, the binomial procedure for
determining the required sample size for evaluation of proportions in a population

is used, given by:

29

n = pq*(Z../E)2

where:

n: required sample size

p: proportion in the population
q=(1-p)

=critical value from the normal distribution

a: desired confidence interval

E = margin for error
Note that the proportion of prior similar adopters is unknown. In the binomial
distribution greatest variance and therefore the most conservative sample size
estimate is provided if it is assumed the proportion of similar adopters is half
(p=0.5). Choosing a 90% confidence interval (2:165) and a 10% margin of error
(E=0.1) for estimation of the proportion, the required sample size is calculated
thus:

(0.5)(0.5)*(1.65/0.1)2
68.1

3
II II

68 repeat adopters per year is therefore an upper bound on the desired sample
size. It is unlikely half of all prior adopters will be similar to current adopters. A
less conservative estimate is provided if it is assumed that 10% of all prior
adapters are similar:

(0.1)(0.9)*(1.65/0.1)2
24.5

n

The value of 24.5 adopters provides an approximate lower bound for the required
sample size. Based upon these sample size guidelines and the repeated yearly
calculations of the influence ratio used in this study, an automobile model was

only included in the study if the number of repeat adopters per sale year on

30

average was greater than or equal to 20. This excluded 2 of the 6 models.
These two models were the Honda Odyssey and Land Rover Discovery that had
markedly lower numbers of repeat adopters per year than even the Mazda Miata
which was the last automobile included. Automobile models retained and

removed according to repeat adopter sample size are presented in Table 2.

31

 

 

8.2 8 ea 8 89 m8. k 38 82 E. 8385 86m 98..
as 8. RE 8 88. m an .2. 88. 82 88.60 88:
U0>OEO¢
09. mm: 89 com 32 88 m 38 88 co_m>< 588
m. a we 88 mm. 8: .88 k 38 82 as: 88.2
v. 6 m 5 3.: one at 52 k 38 82 8.686 290 82.
men 8. 8: 8m 99 23 o 88 88 :82 880
8582..

23> 23 6353 2928 9.0523 «3323 9883‘ 23> one to?
«@588... .35 88...: .26 so: .88: 252 __< 28 «a «a «a .28:

mum aEEmm Lax—on,“ Banan— Eflafiams a. 9.5623 ua>oEam new oasfiam 2305. a_aEa> "N min...

32

The 4 retained automobile models belong to four different vehicle classes:
compact sedan (Dodge Neon), sports utility vehicle (Jeep Grand Cherokee),
convertible (Mazda Miata), and large sedan (Toyota Avalon). These four

vehicles are therefore diverse in product characteristics.

Sample Descriptive Statistics

Descriptive statistics presented for adapters in Tables 3 to 13 reveal
several differences between the 4 automobile models and between New and

Repeat Adopter groups.

Table 3: Age Characteristics for New and Repeat Adapters

Jeep

 

 

Dodge Mazda Toyota

Neon ngnkie Miata Avalon

New Adapters N 1,704 1,616 1,117 1,690
Mean 40.69 45.94 49.22 60.57

Std. Dev. 16.808 13.357 11.777 12.652

Repeat Adapters N 200 415 146 241
Mean 45.97 52.32 50.89 61 .84

Std. Dev. 17.231 1 1.390 1 1.344 1 1.663

 

 

 

 

 

 

For each automobile, Repeat Adapters always have a higher mean age than
New Adapters. Distribution of household income by automobile and adapter
group are displayed for New Adapters and Repeat Adapters in Table 4 and Table

5 respectively.

33

Table 4: Household Income Characteristics for New Adapters

 

 

 

 

 

 

 

 

 

 

 

 

Dodge Neon Jgfigreoiaegd Mazda Miata Toyota Avalon

Freq % Freq % Freq % Freq %
$5533)?“ 85 4.7 4 0.2 3 0.3 10 0.5
$2333“ 212 11.7 28 1.6 12 1.0 32 1.7
3:333 289 15.9 70 4.1 38 3.2 84 4.5
32:33? 258 14.2 96 5.6 77 6.5 103 5.6
$543333“ 231 12.7 179 10.4 106 8.9 141 7.6
gigg- 214 11.8 217 12.6 115 9.7 179 9.7
233:333' 142 7.8 295 17.1 204 17.2 239 12.9
$3333 85 4.7 202 1 1.7 164 13.8 202 10.9
$3333 25 1.4 125 7.3 95 8.0 101 5.5
$133333 8 0.4 102 5.9 90 7.6 94 5.1
323333 6 0.3 43 2.5 37 3.1 38 2.1
3233200 OR 5 0.3 64 3.7 44 3.7 38 2.1
Total 1,560 86.0 1,425 82.8 985 83.1 1,261 68.2
Missing 255 14.0 296 17.2 200 16.9 587 31.8
Total 1,815 100.0 1,721 100.0 1,185 100.0 1,848 100.0

 

The $75,000-$99,999 is the most prevalent category for all but the Dodge Neon

whose lower l ncome profile, likely reflects its economy vehicle status.

34

 

Table 5: Household Income Characteristics for Repeat Adapters

 

 

 

 

 

 

 

 

 

 

 

 

Dodge Neon Jgiggiaegd Mazda Miata Toyota Avalon

Freq % Freq % Freq % Freq %
LESS THAN
$15000 0 0 2 0.5 1 0.7 1 0.4
$15,000- $24,999 27 12.9 5 1.2 3 2.0 2 0.8
$25,000- $34,999 31 14.8 10 2.3 1 0.7 4 1.5
$35,000- $44,999 30 14.4 14 3.3 7 4.6 11 4.2
$45,000- $59,999 30 14.4 35 8.1 16 10.5 25 9.6
$60,000- $74,999 27 12.9 37 8.6 17 11.1 24 9.2
$75,000- $99,999 19 9.1 74 17.2 28 18.3 36 13.8
$100000- $124999 10 4.8 61 14.2 25 16.3 24 9.2
$125000- $149999 3 1.4 32 7.4 12 7.8 23 8.8
$150000- $199999 3 1.4 46 10.7 8 5.2 14 5.4
$200000- $249999 0 0.0 11 2.6 5 3.3 8 3.1
$250,000 OR
MORE 0 0.0 27 6.3 7 4.6 6 2.3
Total 180 86.1 354 82.3 130 85.0 178 68.5
Missing 29 13.9 76 17.7 23 15.0 82 31.5
Total 209 100.0 430 100.0 153 100.0 260 100.0

 

 

For Repeat Adapters the $75,000-$99,999 category is again the most prevalent

except for the Dodge Neon. Categories of education are presented next in Table

6 and 7.

Table 6: Education Categories for New Adapters

Jeep Grand

 

 

 

 

 

 

 

 

 

 

 

 

Dodge Neon Cherokee Mazda Miata Toyota Avalon
Freq % Freq °/a Freq % ‘ Freq %
GRADE SCHOOL 25 1.4 14 0.8 8 0.7 16 0.9
HIGH SCHOOL 540 29.8 241 14.0 109 9.2 273 14.8
TRADENOCATIONAL 158 8.7 86 5.0 39 3.3 66 3.6
SOME COLLEGE 577 31.8 390 22.7 268 22.6 452 24.5
COLLEGE 335 18 5 592 34 4 378 31 9 499 27 0
GRADUATE ' ' ' '
gngGRADUATE 135 7.4 350 20.3 360 30.4 478 25.9
Total 1,770 97.5 ~ 1,673 97.2 1,162 98.1 1,784 96.5
Missi_ng 45 2.5 48 2.8 23 1.9 64 3.5
Total 1,815 100.0 1,721 100.0 1,185 100.0 1,848 100.0

 

35

 

Except for the Dodge Neon “College Graduate” is the most prevalent educational

category.

Table 7: Education Categories for Repeat Adapters

Jeep Grand

 

 

 

Dodge Neon Cherokee Mazda Miata Toyota Avalon

Freg % Freq % Freq % Freq %
GRADE SCHOOL 2 1.0 0 0.0 0 0.0 1 0.4
HIGH SCHOOL 61 29.2 61 14.2 11 7.2 29 11.2
TRADE/VOCATIONAL 22 10.5 20 4.7 4 2.6 5 1.9
SOME COLLEGE 54 25.8 89 20.7 33 21.6 60 23.1
COLLEGE
GRADUATE 46 22.0 146 34.0 47 30.7 82 31.5
ggETG RADUATE 19 9.1 1 10 25.6 54 35.3 79 30.4
Total 204 97.6 426 99.1 149 97.4 256 98.5
Missing 5 2.4 4 0.9 4 2.6 4 1 .5
Total 209 1 00.0 430 100.0 1 53 1 00.0 260 1 00.0

 

 

 

 

 

 

 

 

 

 

 

A Similar pattern is apparent for Repeat Adapters, except for the Dodge Neon
whose more prevalent category is now “High School” instead of “Some College”.
However there is evidence for all models that those at or around the college level
have increased their level of education relative to New Adapters. Occupation

categories are examined next in Tables 8 and 9.

Table 8: Occupation Categories for New Adapters

 

 

 

 

Dodge Neon Jgiggggd Mazda Miata Toyota Avalon

Freq % Freq % Freq % Freq %

TECH/SALES/ADMIN 265 14.6 231 13.4 181 15.3 152 8.2
MGR/PROFESSIONAL 428 23.6 731 42.5 576 48.6 501 27.1
CRAFT/REPAIR 61 3.4 40 2.3 18 1.5 19 1.0
OPERATOR/LABORER 177 9.8 52 3.0 18 1.5 15 0.8
SERVICE 159 8.8 64 3.7 49 4.1 68 3.7
FARM/FOREST/FISH 5 0.3 7 0.4 3 0.3 4 0.2
ARMED SERVICES 29 1.6 23 1.3 8 0.7 3 0.2
OTHER 332 18.3 188 10.9 83 7.0 122 6.6
Total 1 .456 80.2 1 .336 77.6 936 79.0 884 47.8
Missing 359 19.8 385 22.4 249 21 .0 964 52.2
Total 1,815 100.0 1,721 100.0 1 ,185 100.0 1,848 100.0

 

 

 

 

 

 

 

 

 

36

 

Disregarding the occupation category “Other” the most common occupation

category is always “Mgr/Profession” followed by the “T ech/Sales/Admin".

Respondents for the Toyota Avalon appear particularly reluctant to provide

information concerning their occupation with over half of responses coded as

‘Missing”. This reluctance is possibly due to the mean age of Avalon

respondents exceeding 60 years, meaning a substantial number of respondents

are retired relative to buyers of the other automobiles.

Table 9: Occupation Categories for Repeat Adapters

 

 

 

 

 

 

 

 

 

 

 

 

Dodge Neon ngggiaegd Mazda Miata Toyota Avalon

Freq % Freq % Freq % Freq °/o
TECH/SALES/ADMIN 26 12.4 55 12.8 19 12.4 28 10.8
MGR/PROFESSIONAL 57 27.3 197 45.8 83 54.2 81 31.2
CRAFT/REPAIR 3 1.4 7 1.6 3 2.0 2 0.8
OPERATOR/LABORER 18 8.6 16 3.7 1 0.7 1 0.4
SERVICE 18 8.6 11 2.6 4 2.6 5 1.9
FARM/FOREST/FISH 0 0.0 0 0.0 0 0.0 2 0.8
ARMED SERVICES 3 1.4 5 1.2 1 0.7 0 0.0
OTHER 34 16.3 29 6.7 9 5.9 10 3.8
Total 159 76.1 320 74.4 120 78.4 129 49.6
Missing 50 23.9 110 25.6 33 21.6 131 50.4
Total 209 100.0 430 100.0 153 100.0 260 100.0

 

 

Results for Repeat Adapters show the same pattern of occupational category

responses observed for New Adapters. Again more than half of respondents for

the Toyota Avalon did not provide their occupational category. Gender

characteristics are presented next in Tables 10 and 11.

37

Table 10: New Adopter Gender Characteristics

 

 

 

 

Dodge Neon Jgigreoaaegd Mazda Miata Toyota Avalon

Freq % Freq % Freq % Freq %

MALE 718 39.6 860 50.0 679 57.3 982 53.1
FEMALE 1,017 56.0 790 45.9 486 41.0 778 42.1
Total 1 .735 95.6 1 .650 95.9 1 .165 98.3 1 ,760 95.2
Missigq 80 4.4 71 4.1 20 1.7 88 4.8
Total 1,815 100.0 1,721 100.0 1,185 100.0 1,848 100.0

 

 

 

 

 

 

 

 

 

 

Males are a higher proportion of adapters for all automobiles except the Dodge

Neon.

Table 11: Repeat Adopter Gender Characteristics

Jeep Grand

 

 

 

 

Dodge Neon Cherokee Mazda Miata Toyota Avalon

Freq % Freq % Freq % Freq °/o
MALE 86 41 .1 256 59.5 85 55.6 158 60.8
FEMALE 121 57.9 168 39.1 67 43.8 92 35.4
Total 207 99.0 424 98.6 1 52 99.3 250 96.2
Missing 2 1.0 6 1.4 1 0.7 10 3.8
Total 209 100.0 430 100.0 153 100.0 260 100.0

 

 

 

 

 

 

 

 

 

 

AS for New Adapters, males are a higher proportion except for the Dodge Neon.

Race distribution characteristics are examined next in Tables 12 and 13.

38

Table 12: New Adopter Race Characteristics

Jeep Grand

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Dodge Neon Cherokee Mazda Miata Toyota Avalon
Freq % Freq °/o Freq % Freq %
WHITE ONLY 1,435 79.1 1,451 84.3 1,020 86.1 1,539 83.3
HISPANIC+HISP-WH 101 5.6 60 3.5 36 3.0 29 1.6
ASIAN+ASIAN-WH 19 1.0 32 1.9 36 3.0 77 4.2
BLACK+BLACK-WH 113 6.2 49 2.8 11 0.9 61 3.3
NAT AM+NAT AM-WH 40 2.2 29 1.7 16 1.4 34 1.8
OTHER+OTHER-WH 30 1.7 23 1.3 10 0.8 14 0.8
ADDTL DUAL RACES 11 0.6 3 0.2 2 0.2 1 0.1
> 2 RACES 6 0.3 0 0 4 0.3 4 0.2
Total 1,755 96.7 1,647 95.7 1,135 95.8 1,759 95.2
Missing 60 3.3 74 4.3 50 4.2 89 4.8
Total 1,815 100.0 1,721 100.0 1,185 100.0 1,848 100.0
Table 13: Repeat Adopter Race Characteristics
Dodge Neon Jgiggiaegd Mazda Miata Toyota Avalon
Freq % Freq % Freq % Freq %
WHITE ONLY 176 84.2 395 91.9 136 88.9 233 89.6
HISPANIC+HISP-WH 7 3.3 5 1.2 1 0.7 2 0.8
ASIAN+ASIAN-WH 2 1.0 2 0.5 6 3.9 6 2.3
BLACK+BLACK-WH 6 2.9 9 2.1 1 0.7 3 1.2
NAT AM+NAT AM-WH 6 2.9 9 2.1 3 2.0 4 1.5
OTHER+OTHER-WH 3 1 .4 2 0.5 1 0.7 0 0.0
ADDTL DUAL RACES 0 0.0 1 0.2 0 0.0 0 0.0
> 2 RACES 0 0.0 1 0.2 0 0.0 1 0.4
Total 200 95.7 424 98.6 148 96.7 249 95.8
Missing 9 4.3 6 1 .4 5 3.3 1 1 4.2
Total 209 1 00.0 430 1 00.0 153 1 00.0 260 1 00.0

 

 

 

 

 

 

 

 

 

 

 

Race category distributions are quite consistent across New and Repeat Adopter

groups. “White Only” is always the most common mode of race category. The

dominant proportion of “White Only" respondents is lowest for the Dodge Neon in

both New and Repeat Adopter groups.

39

 

Development of Measures

Both individual characteristics and social effects can influence purchase
overtime. Individual characteristics are treated as unchanging overtime, while
social influence is a time varying construct. Following Strang and Tuma’s (1993)
interpretation of factors that affect the adoption decision, the concepts of
infectiousness, susceptibility, propensity, and proximity are used to capture
individual and network effects.

To capture these dynamics, a panel-type dataset was constructed as
deemed suitable for infrequently purchased durableS with new technology
content (Kim, Srivastava and Han 2001; Hsiao 1986). Propensity factors remain
constant overtime for each individual. The status of an individual as susceptible
and infectious are fixed according to New or Repeat Adopter status. The
measure of proximity between susceptible and infectious adopters is calculated
for each time period and is time varying. The dependent variable, measures of
infectiousness, susceptibility and propensity, are defined before turning to

development of the proximity measure.

Dependent Variable: Year of Purchase

An individual’s time to adoption is the dependent variable. Purchase
timing in the dataset is measuring with calendar date precision, however sample
Size guidelines for estimation of proportions require independent time varying

covariates to be assessed over 12 month intervals. To maintain parity between

40

dependent and independent variable time intervals, time to adoption was

measured in years.

Adopter Groups: Susceptibility and Infectiousness

Respondents were divided into one of two categories:
. New Adapters: An individual was classified as a new adopter if they
purchased a vehicle model in G2 and did not own Gt of that model.
0 Repeat Adapters: An individual was classified as a repeat adopter if they
purchased both G1 and G2.
New Adapters represent an adopting population that is more susceptible to
influence than Repeat Adapters. Prior Repeat Adapters represent a population

that is more Infectious than prior New Adapters.

Propensity

Propensity is the likelihood of adoption by an individual, independent of
network factors. Propensity is assumed to be constant and is measured by the
age of the vehicle replaced, number of vehicles owned, and household income.
As a durable good, the age of the vehicle currently owned, and the number of
vehicles owned, clearly impact upon the need to purchase a replacementvehicle.
Mechanical reliability of durable goods declines overtime and the greater the
number of alternative substitutes owned (number of vehicles), the lower the need

for a replacement. Household income was measured because higher income

41

individuals are known to have higher tolerance for risk in risk-return utility

functions and adopt more quickly.

Proximity: the Demographic Similarity Ratio

Determining whether two individuals are of identical character and social
status is difficult. Studies of networks and diffusion have consequently utilized
more visible qualities as proxy measures of similarity. Similarities consistent with
Structural network equivalence at the organizational level have been measured
along geographic lines (Greve 1998; Davis and Greve 1997; Soule and Zylan
1997), organizational size (Davis and Greve 1997) and market Size (Greve
1998). Indirect relationships in these studies were found to predict adoption of
innovations where more Similar organizations adopted similar innovations at
similar times. De Bruyn and Lilien (2004) tracked direct relations at the individual
level and found age, education, sex and occupational similarities provided
avenues for word of mouth influence in a consumer decision making context.

To measure the influence of prior adopters upon current adopters an index
of demographic similarity was constructed. This index assessed the proportion
of prior adopters that are demographically similar to a current adopter. This time
varying measure of proximity between individuals provides a means of linking
past and present adopters.

The demographic characteristics used in this ratio calculation are:
0 Household income (12 categories)

. Age (numeric)

42

Race (8 categories)
Level of education (6 categories)
Occupation (8 categories)

Gender (dichotomous)

Demographic factors were selected for likely time invariance. With the exception

of “Age” all demographic characteristics were assumed to be constant for each

individual. A tolerance range of plus or minus 5 years was used to account for

change in age. Individuals were deemed Similar along demographic lines it:

they were of the same income class;

they were within plus or minus 5 years in age;
of the same race;

of the same level of education;

of the same occupation;

of the same gender.

Variable properties for the 6 demographic characteristics are presented in Table

14.

43

Table 14: Variable Properties for Demographic Characteristics

 

Income Occupation Race Education Gender Age
LESS THAN TECH/SALES/ GRADE

$15000 ADMIN WHITE ONLY SCHOOL MALE NUMERIC
$15,000- MGR/ HISPANIC+HI HIGH FEMALE
$24,999 PROFESSIONAL SP-WH SCHOOL

$25,000- CRAFT ASIAN+ASIAN TRADE/

$34,999 IREPAIR -WH VOCATIONAL

$35,000- OPERATOR! BLACK+BLAC SOME

$44,999 LABORER K-WH COLLEGE

$45,000- SERVICE NAT AM+NAT COLLEGE

$59,999 AM-WH GRADUATE

$60,000- FARM/ OTHER+OTH POSTGRADU

$74,999 FOREST/FISH ER-WH ATE COL

$75,000- ARMED ADDTL DUAL

$99,999 SERVICES RACES

$100000-

$124999 OTHER > 2 RACES

$125000-

$149999

$150000-

$199999

$200000-

$249999

$250,000

OR MORE

 

While degrees of similarity from 1 to 6 could potentially be examined, the

nature of demographic measures such as race and gender mean almost no

individuals have zero, or all 6 characteristics in common, and almost all

individuals have one or two demographic features in common. This is because

gender has only two categories, and for all car models the predominant race

representation was ‘white’. Therefore to establish the degree of similarity needed

to informatively observe influence effects, three ratios with increasingly stringent

similarity criteria were calculated:

0 3 or more demographic features in common;

0 4 or more demographic features in common;

0 5 or more demographic features in common;

44

These are referred to in the study as degree 3, 4, and 5 similarity respectively.
Degree 3 similarity requires a past adapter to have 3 or more characteristics in
common and thus includes individuals of degree 4 and 5 similarities. Degree 4
similarity requires 4 or more common characteristics and thus includes
individuals of degree 5 similarity. The different similarity degrees are thus nested
sets of individuals with more strict higher degrees consisting of subsets of
individuals who also appear in less strict lower degrees. The algebraic

expression used to calculate the demographic similarity ratio is developed next.

Let adapter 1 be fixed according to their year Y, of adoption given by:

Y(i);

All prior adapters are compared to j, given by:

Y(i) where Y(i) < Y(j) ;

These prior adapters are denoted collectively as set members:

l(j) where l(j) {i=Y(i) < Y(j)} ;

Individual Y(j) is compared to all prior adopters l(j) along 6 demographic

dimensions according to the following decision rules:

Sij(age) = 1 if lagej — agei| s 5 , 0 otherwise;
Sij(inC) = 1 if incj — inc; = 0 , 0 otherwise;
Sij(edU) = 1 if eduj — edui = , 0 otherwise;
Sij(gen) = 1 if gen; — geni = , 0 otherwise;
Sij(rac) = 1 if racj — raci = , 0 otherwise;
Sij(OCC) = 1 if OCCj - occi = 0 , 0 otherwise;

45

where:
age = numeric age of adopter;
inc = income level of adopter;
edu = education level of adopter;
gen = gender of adopter;
rac = race category of adopter;
occ = occupational category of adopter;

To sum the 6 dummy results:

let h = [age,inc,edu,gen,rac,occ]
= [1 ,...,6]

Adopter j is deemed Similar to prior adopter i when ZhSijh is sufficiently high. To
be sufficiently high, common characteristics must satisfy the 3, 4, or 5 degree

cutoff. Similarity is thus dichotomous and is given by:

Dcij = 1 If ZhSiJ-h 2 C, lCIG),

where:
c = 3, 4, 5 cutoff points reflecting degree 3, 4, and 5 Similarity

The proportion of Similar prior adopters is the sum of Similar prior adopters

divided by the sum of all prior adopters. This ratio can be expressed thus:

Zi€:l(j) IDcij

2161(1) 1

For each ratio the numerator is the cumulative number of prior adopters with c or
more demographic characteristics in common with the current adapter, and the
denominator is the total number of prior adopters.

These ratios were calculated from all individuals that adopt in a prior

calendar year to a current adopter. Consequently individuals who adopt in year

46

one have no demographic similarity ratio as no individuals adopt prior to their
adoption year. An individual adopting in year 2 has a single demographic
similarity ratio calculated from all individuals who adopted in year 1; an individual
adopting in year 3 has two demographic similarity ratios based upon all prior
adapters in year 1 and another based upon all prior adapters in both year 1 and
year 2, and so on. These demographic similarity ratios are thus time-varying
covariates with the ratio calculated repeatedly for each year prior to an adopter’s
purchase year.
Three forms of this ratio are calculated:

0 a ratio for all prior adopters (new adapters and repeat adopters);

o a ratio for prior new adopters;

. a ratio for prior repeat adopters.
Respectively, these 3 forms assess overall prior effects, effects due to prior new
adapters, and effects due to prior repeat adopters.

The ratio is calculated for each individual for every time period of exposure
prior to purchase, from every prior adapter in the dataset. At each comparison,
this involves comparison of the 6 demographic dimensions between individuals.
Direct comparisons of demographic characteristics results were time consuming.
Using the spreadsheet program Microsoft Excel, calculation times for a Single
year of exposure ranged between 2-8 hours depending upon sample size for the
automobile model. To improve the efficiency of the comparison process a matrix
for each individual was created to store the accumulation of similar prior

individuals as each year was calculated. The PROCIML statement in SAS v9.1

47

was used to conduct the matrix algebra manipulations. The improvement in
computation time was dramatic. Calculation time for all time periods for most
automobile models took less than an hour. The cost of utilizing this cumulative
approach is an inability to determine the source of similarity between individuals.
Between current and prior adopters, it is not possible to determine which
demographic characteristics are similar, only that a certain number (3, 4 or 5) are
in common. Sensitivity Simulation results partially compensate for this
abstractness through stepwise removal of each of the 6 demographic
characteristics. This involved repeating the dataset construction and regression

analyses 6 times, once for each characteristic omitted.

The Demographic Similarity Ratio: Characteristics

The demographic similarity ratio determines heterogeneous internal
influence by taking a sample of adopters. The ratio differs from the internal
influence parameter used in diffusion models, and iS designed to assess network
influence from a sub-sample of adopters. These two key points are discussed
next.

The demographic similarity measure is a sampling ratio. The greater the
proportion of prior adopters with demographic characteristics in common with a
current adapter, the higher the ratio for that adopter. Lower ratios indicate prior
adopters have fewer demographic characteristics in common. As a ratio of two
cumulative functions, the demographic similarity ratio may increase or decrease

at different points in time according to the number of connected versus

48

unconnected adopters. Each individual contributes equally to the ratio which is
determined uniquely for each potential adopter until adoption occurs. This is
similar in principle to diffusion models that interpret word of mouth through an
influence parameter. For example the Bass Diffusion Model (Bass 1969) is of

the form:

n(t) = Dim-N(t)] + (q/m)*N(t)lm-N(t)l

where:

n(t) = number of adopters at time t;

N(t) = cumulative number of adopters;

m = market potential;

p = coefficient of innovation;

q = coefficient of imitation.

Notably, the contribution of an adapter to N(t) declines as the cumulative number
of adopters increase. Because N(t) is cumulative, it describes a monotonically
increasing function, and as a consequence, each individual’s proportional
contribution to N(t) declines overtime. At all times, each adopter makes an
equal contribution to N(t) and therefore an equal contribution to the measure of
internal influence, q. This means that all potential adopters receive influence
equally.

It is stressed that the demographic Similarity ratio measures the proportion
of connected adopters, whereas internal influence in diffusion models generally
assess all prior adapters and do not allow for heterogeneous influence between
past and potential adopters. Like the Bass Model, the demographic similarity

ratio weights all adopters equally, a key difference however is that a prior adopter

only exerts influence when similar to a potential adopter, whereas a dissimilar

49

prior adopter lowers the ratio. However it would be incorrect to conclude from
the ratio that adoption by an unconnected individual reduces influence in the
network. Adapters may exert positive or no influence, but not negative influence.
Although some adoption models have proposed modeling negative influence
(Sharif and Ramanathan 1984; Midgely 1976), the similarity ratio does not.
Because adoption data are not at the population level, it is not possible to simply
accumulate all prior connected adopters. It is instead necessary to observe the
available sample of adapters to measure if the proportion of connected to
unconnected adopters changes overtime, and apply this as an indicator of
influence exerted by connected adopters at the population level. In the sample
ratio, dissimilar individuals model lack of influence and not negative influence. It
is thus important the sample provides a valid representation of influence.

The sampling assumption holds that the ratio measure is a reliable and
valid indicator of influence from prior to current adopters. The data collection
process randomly sampled a population level list of all new vehicle registrants.
The classification of individuals into new and repeat categories, and choice of
demographic characteristics as a measure of influence in the research are
therefore unrelated to the process used to obtain data. It is therefore reasonable
to assume the proportion of new and repeat adapters in the sample is similar to
that in the population, provided the sample size is large enough to offset
statistical fluctuations.

The data in this research are themselves a sample of the overall market

for each automobile and to the extent this sample may not be representative of

50

the market, any derived ratios may also be non representative. This cannot be
controlled for in the present study, however the problem of small samples was
partially accounted for by ensuring that ratios for the smaller repeat adopter
group (repeat adopters groups are always smaller than new adopter samples)
are calculated from 20 or more adopters per calendar year. While this decision
removed two second generation models from the study, partial vindication of this
decision is provided by the observation that there is no consistent pattern in the
results related to sample size of new or repeat adopters. 20 repeat adapters in a
year is too small a number to be deemed representative of the market, but large

enough to claim sufficient representation of the data in the sample.

Empirical Model Estimation: Piecewise Weibull versus Cox Regression

Diffusion models assess the rate at which members of a susceptible
population transition from a potential adapter to a past adopter state. The size of
the susceptible population must be specified and the rate at which individuals
transition from a potential adapter to adopter state defines the distribution of the
adoption function. If the size of the potential adopting population is unknown, or
the number of total adopters at all points in time is unknown, diffusion models
cannot be used to describe new product adoption. The data in the research
represent a subset of the market selected by event occurrence up to a target
sample level, and therefore cannot be used to provide estimates of the potential

market or the total number of adopters. Hazard rate models are more flexible

51

than diffusion models because it is not necessary to observe the entire

population.

Hazard Rate Models

Hazard rate models examine the probability of event occurence overtime.
Parametric hazard rate models specify the distribution of the hazard (such as an
exponential distribution) while semi-parametric models leave the distribution of
the hazard unspecified (as in Cox Regression). Both parametric and semi-
parametric models may be modeled as proportional hazards where differences in
hazards overtime between individuals remain constant. Consideration of factors
where effects upon individual hazards vary overtime, results in non-
proportionality between the hazards of individuals. Incorporation of time-varying
covariates in either parametric or semi-parametric hazard rate models is thus a
means of modeling non-proportional hazard rates.

The choice of a parametric distribution is largely an empirical question
determined by model fit. Even a strong a priori rationale for specification of a
particular parametric distribution must be weighed against the possibility the
underlying process that generates the hazard is less than fully understood.
lmportantly, incorrect specification can result in inaccurate estimation of covariate
coefficients. This provides a strong argument for leaving the baseline hazard
unspecified through the use of semi-parametric Cox regression.

The data in this research are known to have several sampling characteristics:

1) All individuals experience an event;

52

2) There are target sample sizes for each year that are not based upon
market level sales;
3) Sample responses obtained in the first and second quarters are

approximately twice the number of those in the third and fourth.

Respectively this means: (1) there are no right-censored data; (2) sample sizes
are not representative of market sales levels; and (3) the distribution of events
within model years is higher during the first half of the year. These sample
characteristics preclude determination of the effect of time as a cause of
purchase. In the context of consumer durables (in this case automobiles),
product factors such as durability mean time undoubtedly has an influence upon
time to the next vehicle adoption, however sample characteristics do not allow
time to be tested in a causational manner. A hazard rate therefore cannot be
specified with the aim of determining the effect of time upon event occurrence.
The influence of covariates however may still be investigated if the effect of time
(or more accurately, the distribution of events in the sample) is suitably
accounted for.

The sampling strategy varies across quarters, but iS consistent within a 12
month period. A piecewise approach to regression may therefore account for
between year variations, and the choice of a suitable parametric distribution can
account for within year variations. Because the distribution of events within a
year is not constant (as first and second quarter samples sizes are twice that of

third and fourth quarters) a piecewise exponential distribution is not appropriate.

53

By virtue of a more flexible specification, the piecewise Weibull is able to account
for the non-constant event occurrence within each year. In addition, a piecewise
approach is necessary to incorporate the time-varying covariates that are used to
assess demographic similarity in this study. The piecewise Weibull model in

hazard form is Specified as:

h,(t) = cjht.exp(cj-1)

for a“ s t < a]

where:

i = individual i

t=fime

c = shape parameter

A = rate parameter

j = number of intervals

a = cut point for the interval, a0 = 0, aJ=°°

Thus, the piecewise Weibull allows the hazard Shape and scale parameters to
vary for each interval j given by [aj-1, aj], thereby accounting for within year
sampling variation uniquely across different years. The non-piecewise form of
the model does not partition by intervals, and does not incorporate the interval
indicator j.

Alternatively, Cox regression allows for an unspecified baseline hazard,
removing the need for selection of a parametric distribution. Events must still
conform to a failure time process but the distribution of events overtime need not
be parametric inform. Purchase events for automobiles in the current sample

are consistent with a failure time process that by reason of the data collection

strategy, appear non-parametric in nature. The attraction of Cox regression can

54

be Shown by contrasting the regression equations for Weibull regression versus
Cox regression. Estimation of hazard rate models is typically undertaken in log
form to ensure time remains positive as a dependent variable, and allow for an
additive linear relationship among the variables. The Weibull model in log form
is:

'09 hi“) = G.lOQ t + B1Xi1+ BgXiz + + kaik

where o.log t represents the underlying time-based (Weibull) hazard distribution
that must be estimated together with all covariates. The Cox model in log form

can be represented as:

'09 “10) = (1(1) + BIXH + Bzxiz + + kaik

where C(t) represents any underlying hazard distribution that need not be
Specified due to the method of partial likelihood estimation used in Cox
regression. Leaving the underlying hazard unspecified removes the possibility of
modeling event distribution overtime incorrectly, but reduces estimation
efficiency for the covariates. Both piecewise Weibull and Cox regression were

used in the research, and the results compared.

Regressions

The time-varying demographic similarity ratios, and static propensity factors,
were entered as explanatory variables in regressions upon the log of adoption

time for separate New and Repeat Adopter groups. For each adopter group, 5

55

regressions were estimated to assess the effect of the demographic similarity
ratios upon time to adoption:

1. A baseline regression with static covariates only;
log(timej) = o(t) + [31(incj) + [32(vehj) + 83(vehagej) + ujt
2. ALL prior adopters;
log(timej) = C(t) + [3,0an + [32(vehj) + 83(vehagej) + [34(ALth)
+ ujt
3. Prior NEW adopters;
log(timej) = o(t) + 81(incj) + 82(vehj) + 83(vehagej) + 84(NEth)
+ ujj
4. Prior REPEAT adopters;
log(timej) = o(t) + 8,0an + 82(vehj) + 83(vehagej) +
B4(REPEATjt) + “it
5. Prior NEW adapters and prior REPEAT adopters.

log(timej) = o(t) + 8,0an + 82(vehj) + [33(vehagej) + B4(NEW,-,)

+ 85(REPEATjt) + ujt
where:
time = adoption time in years since year of product launch
j = adopterj
t = year of adoption, 1 < t s T
inc = income of adopter
veh = number of vehicles owned by adopter
vehage = calendar model year of vehicle previously driven
ALL = demographic similarity ratio calculated from prior All Adapters
NEW = demographic similarity ratio calculated from prior New Adapters
REPEAT = demographic similarity ratio calculated from prior Repeat
adopters

56

The time varying covariate ALL in regression form 2 is used to assist in
hypothesis testing of simultaneous prior New and Repeat Adapters. Should
NEW and REPEAT have similar effects in regression models 3 and 4, it is likely
simultaneous consideration in model 5 would result in misleading conclusions.
Inflated variance due to collinearity could result in one or both of NEW and
REPEAT coefficients estimated as insignificant when in fact both groups are
exerting significant influence. ALL is a demographic similarity ratio calculated
from both NEW and REPEAT adopters collectively and is entered as a single
regressor in regression 2, thus avoiding possible collinearity that could occur if
NEW and REPEAT (with similar effects in regressions 3 and 4) were entered as
distinct covariates in regression form 5, while still allowing for simultaneous
consideration of New and Repeat Adopter influence effects.

Regression 2 may also be interpreted as a relaxation of the assumption
prior adopters exert influence differentially according to New and Repeat status.
This means if regressions 3 and 4 Show similar effects for NEW and REPEAT,
and the coefficient for ALL in regression 2 is consistent with NEW and REPEAT,
effects cannot be separated by New and Repeat Adopter segments, as influence
stems from a single prior adopter group rather than distinct New and Repeat ,
groups. Substantively this would mean New and Repeat Adapters have equal

infectiousness. Regression analyses 1 to 5 are represented in Table 15.

57

Table 15: Analyses by Group and Covariate

 

 

Static ALL NEW REPEAT
Covariates

New Adapters X

x x

x X

x x

X x x
Repeat Adapters X

x x

x x

x x

x X x

 

To examine the prevalence of any influence effects, regressions were conducted
for each of the 3, 4, and 5 degree demographic similarity ratios.

Relative to the baseline model with only static covariates, a log likelihood
statistic was used to assess improvement in fit for models with time varying
covariates. Statistical significance was assessed by comparing twice the
difference in log likelihood statistics with 5% critical values of the Chi-square
distribution. Log likelihood comparisons could be made within demographic
similarity degrees but not between, as different degree similarity ratios are
calculated from different sets of adopters. Regressions between degrees are

thus non-nested.

Sensitivity Tests for the Demographic Similarity Covariates

Robustness tests were conducted for each degree similarity measure

through sequential removal of each demographic characteristic and recalculation

58

of the similarity covariate. Regressions were then repeated with the reduced set

covariates.

59

Chapter 4

RESULTS

Piecewise Weibull Regression versus Cox Regression

The piecewise Weibull distribution was chosen for its ability to model the
distribution of the known sample fluctuations in purchase levels overtime that
were an inherent feature of the dataset. Cox regression utilizes partial likelihood
estimation which leaves the baseline distribution for events overtime
unspecified. Since the piecewise Weibull was chosen to account for sample
variation overtime rather than estimate the effect of time upon purchase, Cox
regression is a natural alternative to determine the suitability of the piecewise
Weibull distribution. Support for the piecewise Weibull is provided if covariate
effects upon purchase time for both methods are similar. The statistical software
package SAS v9.1 was used to conduct all regression analyses.

Analyses were run for 3, 4, 5 and degree similarity. A 5% significance
level was used for selection of coefficient significance. The results were highly
congruent across piecewise Weibull and Cox regressions. As expected
parametric estimation was more sensitive than semi-parametric estimation as
some information is lost during partial likelihood estimation in Cox regression.
The piecewise Weibull analyses yielded 169 significant static covariate and 108
significant time varying covariate coefficients across degree 3, 4, and 5
demographic similarity models. Cox regression results yielded 154 significant

static and 80 significant time varying covariate coefficients. Of the 234 significant

60

coefficients identified during Cox regression, 225 were also identified by the
piecewise Weibull technique for an overall overlap identification percentage of
96.2%, Because more significant coefficients were identified by the piecewise
Weibull than the Cox regression method, Cox regression results overlapped less
with piecewise Weibull results and commonly identified 81.6% of significant
coefficients. The similarity of these results is consistent with the observation that
Cox regression provides a reasonable approximation to parametric regression
when covariates are uncorrelated (Kalbfleisch 1974). All commonly identified
coefficients inferred the same direction of effect. The choice of the piecewise
Weibull distribution to account for sampling fluctuations is therefore supported by
the comparison results.

The choice was made to utilize piecewise Weibull estimation in preference
to Cox regression because when the form of the underlying distribution is known,
parametric estimation provides more efficient estimates than semi-parametric
estimation in finite samples (Efron 1977). This greater efficiency is reflected in
the comparison results where a greater number of significant effects were
identified by piecewise Weibull regression. An additional advantage of using the
piecewise Weibull is faster computation time. The Cox regression procedure in
SAS v9.1 procedure requires recalculation of the dataset for every regression
whereas the piecewise regression procedure does not.

In these analyses, piecewise Weibull estimation yields 3 kinds of
covariates: static, time varying, and time-dependent. Normally time-dependent

covariates assess how time affects purchase (i.e. the hazard of adoption over

61

time). However in this model time-dependent covariates have no substantive
interpretation as the distribution of events overtime is a result of target sample
variations rather than market-level adoption dynamics. This means higher
purchase levels indicate larger sample sizes instead of higher rates of market
adopfion.

In this study the purpose of the time-dependent covariates in the
piecewise modeling approach is to partial out the effect of the longitudinal
sampling frames used in data collection. Similarly, the scale and shape
coefficients do not reflect the shape of the hazard distribution, but instead reflect
changes in sampling levels in a given year. The Weibull scale and Shape
parameters, in conjunction with the Cox regression results, indicate the
systematic non-uniform sample distributions within years are well accounted for
by a Weibull function. Comparative analyses conducted against a piecewise
exponential distribution (which assumes constant changes in sample distribution
within years) using log likelihood statistic comparisons indicated the piecewise
Weibull distribution provided markedly superior fit.

Abbreviated piecewise Weibull regression results for the 4 automotive
models used in the study are Shown in Tables 16 to 39. For clarity, only static
and time varying coefficient results are presented. Coefficients for sales years,
and Weibull scale and shape parameters are omitted. Chi-square results relative
to the baseline model are Shown for all models.

The dependent variable is log of adoption time, while all independent

variables are regressed in their log forms. Positive coefficients indicate purchase

62

delay and negative coefficients indicate acceleration. Coefficient interpretation is

achieved using the transformation:

100(eB -1)

which is a percentage point interpretation for static covariates and a percentage
point change interpretation for time-varying influence covariates, because the
time-varying demographic similarity ratio is a percentage. The result for the
Dodge Neon “vehicle year” coefficient (0.020) for New Adapters is transformed

thus:

100(e°'°2° -1)

= 2.02

Indicating a currently owned vehicle that is one year newer increases expected
time to adoption by 2.02%. The result for the Dodge Neon degree 3 similarity for

New Adapters and the REPEAT coefficient (-0.118) is transformed thus:

100(6'0-118 -1)

=-11.1

Indicating a 1% change in the proportion of prior similar repeat adopters,

decreases expected adoption time by 11.1%, measure in years. Caution should

63

be exercised in comparing coefficient magnitudes as the data are not capable of
determining the precise effect of time upon purchase. Coefficient estimates
assume the piecewise Weibull distribution sufficiently accounts for the quarterly
sampling variation in the data. This assumption is supported by the comparisons
with semi-parametric Cox regression, but it is possible the strength of the
assumption varies across vehicles and adapter groups. It is stressed that
emphasis should be placed upon polarity of effects rather than the absolute value
of coefficients. This reduces the chance that incorrect specification of the
piecewise Weibull distribution alters substantive conclusions regarding relative
adoption times.

In log-likelihood regressions, nested models can be compared using log-
likelihood ratio statistics. Twice the difference in log-likelihood ratio statistics
between nested models are distribution Chi-square. Models, and accompanying
model coefficients were only considered if the model Chi-square statistic
indicated significantly better overall model fit than the baseline model at a 5%
level of significance. For nested analyses, this same Chi-square statistic is used
to select the model of best fit. The 5% critical Chi-square value is 3.84 for 1
degree of freedom, and 5.99 for 2 degrees of freedom. When analyses were
repeated using a less conservative 10% level of significance for model selection,
all newly significant models were consistent in coefficient effects with those
identified using the 5% criteria, and the substantive conclusions remained the
same. Complete static covariate and log likelihood statistic results are presented

in Appendix B.

64

_aooE 35.3me 35 9 326.9 ozmzfim 8:3on 2833-30 n Eu ox

mod w a no €85.56 .

 

 

 

 

mmoo or to Fooo.v Boo 9.3 o
momo wood mooo _ 5o oooo 88 v
.83. $3. .83 888 wood. a .89: 3387
wood .ooov Fooo oomo o
33o Nooo P So wood 5... m
.83. .83 588 Boo- a 885 8387
23 Fooo.v Boo mowo o
o So «So too oooo o... N
hoof ouoo .888 wood. a Loud ooooeow
Boo Boo.v Boo 33 a
mood «So too wood .5... F
83...- .83 .83 oooo- o .838 53.82-
Pooo.v Boo Bad o 8% .80
wood :3 wood 68 8:88 o 28>
636 .086 wood. a vw P .3 P- 2283‘ 332
Emomm am: ._._< odocos :2 2.. 5.5.x 8%..9... .ooos.
.. OJ

3.35:5 m 3.500 23335 302 .3585 5.3353: 55.25 02333:. :32 canon 6.. 33a...

65

mod w a E €85ch .
.39: 8:33 m5 9 9:59 3.....sz 855:6 93345 n Eu «x

 

 

 

 

mooo onto Fooo.v Food 83 o
83 mo 8 wood F Fo.o wood 6.... F.
homo. 23. .33 .83 oooo- o .23 meoooF-
mood Fooo.v Foo.o 83 a
mood mooo F Fo.o wood 6.... o
.23. .83 .83 Boo- o .83 8982-
ouoo Fooo.v Foo.o 88 a
mo 3 «So F Fo.o oooo o... N
.23. .83 .83 oooo- o .omo.m ovvovoF-
mooo Fooo.v Foo.o vao o
83 Nooo F Fo.o ooo.o o... F
.83. .83 .83 oooo- o boos «8.82-
Fooo.v 53 god a oomo .80
wood :96 mood ow mémmmo m 98>
.83 .83 oooo- o «3.92. 9982 262
Emamm 262 ._._< amaze, :2 2.. Eu Nx gauge... .28:
I On—

b...m=E_m e 02ch 923.3 302 .233: cammemmm .3925 02303..“ :82 maven "2 29¢...

66

moo w a 2m ~52:ch .
.89: $5.33 9: 2 9:39 ozmzflm 885:6 93320 u :6 Nx

 

 

 

 

ovmo oNvo Fooo.v Foo.o mmFo o
Eoo moFo Noo.o F Fo.o mooo o... v
Eoo- Foo.o- .33 .83 mooo- o mood meovoF-
mooo Fooo.v Foo.o 83 o
nmoo Noo.o F Fo.o mooo om m
.23. .83 .83 Noo.o- o .eNoF. Noo.ovoF-
mvoo Fooo.v Foo.o 53 o
muoo Noo.o F Fo.o mooo om N
.83. .33 .83 mooo- o .83. 393.2.
NFoo Fooo.v Foo.o 83 o
Roo Noo.o F Fo.o mooo om F
.33. .33 .83 mooo- m .33 omoomoF-
Fooo.v Foo.o nmmo a oomm .80
Noo.o FFo.o mooo .3 8:32 m 93>
.ouod .onod wood. a vm F .mvo F- 9283. 262
imam: 252 ._._< $5.... :9, 8. Efx 30%.»: .28:
.- On—

>=E=§m m 358 «330.3 262 6:32“. 5.3962“. .3903 3.332.". :32 woven For 2%...

67

moo w o a 685.55 .
.mooE 2......me m5 2 9.3.9 0.6.53 858:6 233.50 u Eu Nx

 

 

 

 

No3 vooo mmNo mNo.o mmoo o
ommo mNmF FNo.o Noo.o mFoo om m
$3 .33 F.No.o .NNoo .33 o .mm...oF ooF.F..N-
FmNo FmFo ovoo Foo.o o
93 NNo.o mooo mFoo om N
Eo oooo .mooo LNoo o mNNF Emma-
Noo.o mmNo FNo.o «moo o
No: FNo.o Noo.o mFoo om m
NE... N3 .FBo .33 o .mmooF moN.FF.N-
mooo NmNo mNo.o mmoo o
NNmF FNo.o Noo.o mFoo om m
.33. mNo.o .Nnoo .83 m .3...» me.N..N-
mm S 23 2 S o Bo moo
FNo.o Noo.o 83 o... 2:38 m 28>
Noo.o mmoo mNo.o m NNF..mvN- 2283. 68mm
Emamm 252 ..._< moms? =2, 2.. Euwx 30%”... .28:

3.5.85 m 358 m..2nou< 62.2. .332. :23»..an :36; 3.2635 :32 canoe "or 0.3...

68

moo w a 3 3.8.9.63 .
.mooE o£393 9.. o. 3.3.2 0.6.33 8.5.9.3 93:09.20 n =3 Nx

 

 

 

 

NmNo oFF..o mFNo omo.o $.o.o o
vomo Foo.o NNo.o Noo.o mFoo om m
mNo.o mo...o NNo.o .83 .83 m F8... NomiN-
NNo.o moFo ovoo omo.o o
mmNo NNo.o Noo.o mFoo om N
movo mNo.o Lmoo Foo.o m omo.o 853$
vooo NFN.o omo.o omo.o o
to NNo.o Noo.o mFoo o... m
omo.o NNo.o .63 .83 m RmN ooo.m..N-
omo.o .VFNo omo.o too o
vao NNo.o Noo.o 23 o... m
9.3 FNo.o .83 .33 m omo.o 83.3-
83 omo.o 23 Q Bo Bo
FNo.o Noo.o 83 .3 8:33 m 23>
Noo.o omo.o mNo.o o Roma- 9283. .83..
Emamm .52 4.... $2.9 =2, 2.. Euux 30%."..9... .32...

9.9.55 v 0202. 9.2623 32.3. .333... c9323.. .3925 3.2506... :32 3600 now «Ba...

69

mo... w a .m 28:29... .
6.8... 2.6me m... o. 9.66. 0.6.66 60:96:... $6356-30 u t... Nx

 

 

 

 

8o... oi... vuu... m8... «8... ..
63... an... .8... N8... 65... .3. m
8.... 8.... 8.... .68... L8... m. .83 $6.73..
.8... cam... 68... 8.... a
E... .8... m8... 6...... a... 5
i=6... 8.... L8... .68... a .63.... 86.3w-
2.... 8.... mac... owe... a
mum... .8... N8... So... 6...... m.
.66... 8...... :3... .68... m. .68... www.mvw-
So... mm... m8... 8...... a
mam... .8... m8... 6...... .n... n
.«mm... 68... 6...... LS... m. .62.... 60.63.-
mm... o8... a. to .. Sm .80
.8... N8... 6...... .3. 8:38 m. 28>
mmod mmod omo.o a nwvdvm- 6.98.3 6.8mm
Guam... 2...: .:< wage. :2 2.. Eu... 3%”... .282

39.5.85 9 3.500 9.0323 630m .9362". c0635»... 55.25 3.332.. :32 $2.00 #N 29m...

70

mod w a 6 660.....90 .
.000... 05.0060 0... o. 026.0. 0.6.66 00:06:... 06:00-20 u :6 ~x

 

 

 

 

oto omo.o .ooo.v ooo.o .ooo.v o
86o oomo Noo.o o.o.o mooo o... v
Noo.o- 08.? .83 .83 .63. o o6... 80.88-
83 .ooo.v ooo.o .ooo.v o
80o Noo.o o.o.o ooo.o o... m
.o8.o. .83 .83 68.? o .68... ooo.owom-
8o.o .ooo.v ooo.o .ooo.v o
86o Noo.o o.o.o mooo o... m
53. £80 Looo .89? o 85 86.88-
8o.o .ooov ooo.o .ooov ..
ommo Noo.o o.o.o ooo.o o... .
$8.? .83 89o .89? o Rt... «888.
.ooov ooo.o .ooo.v .. 89. .80
«So o.o.o wood 0.... 2.08.. s 28>
Luoo .8o.o 88.9 o o8. .8... 298...... 302
20000 .502 ..._< 084...... :2. 2.. .20... 085.9... .28....

3.6.6.5 m 00.600 0.0.00.2 .502 .6563. 5600.50... .5903 00.3000... 000.205.. 9.0.0 000—. "Nu 030...

.504

71

8o w o .8 88:29.0. .

.0006 05.0660 0... 0. 026.0. 0.6.66 0000.050 06:08.20 n =6 Nx

 

 

 

 

:oo 8...o .ooo.v .ooo.v .ooo.v o
Looo 8..o Noo.o o.o.o mooo o... v
.08.? Looo .8o.o 88o .63. o .8...L L288-
Loo.o .ooov .ooo.v .ooo.v o
oLoo Noo.o o.o.o mooo o... m
.83. .8o.o .83 .63. o L8... L388-
80o .ooo.v ooo.o .ooo.v ..
voto Noo.o o.o.o ooo.o .3. N
S ..o- Lmoo .8o.o .83. o .8 .. . .8. 88-
omo.o .ooov ooo.o .ooo.v ..
oo..o Noo.o o.o.o mooo .o.m .
.83- .8o.o .8o.o .080. o 8%.... 88.88-
.ooo.v ooo.o .ooo.v o 88 .80
«So o.o.o mooo o... 8.88 L 28>
kg... .30... 63.0. a ova. 50w- 6.0.0004. 2.02
20mm... 302 ....< 0080.. 8.. o... 0...... 30%”... .23....

3.6.2.5 0 00.500 80.003. 302 .8300”. 2060050... .3905 00.2.0003 09.0.0.5 96.0 000.. "mm 0.6...

72

00.0 m a 6 560.556 .

.000... 05.003 0... 0. 026.0. 0.6.66 0000.0...0 06:00-30 u 0.... Nx

 

 

 

 

83 8.8 .ooo.v ooo.o .ooo.v o
omo.o .o..o Noo.o o.o.o 080 .o... 8
83. 8o.o- .8o.o .8o.o .8o.o- o 88... 88.88-
83 .ooo.v ooo.o .ooo.v o
83 Noo.o o.o.o wood .8 8
83. .8o.o .8o.o 880. 8 8L...” omo.o..8-
8o.o .ooo.V ooo.o .ooo.v o
oLo.o Noo.o o.o.o 080 .0...- N
83. .8o.o 88o .89? o 8.8 88.88-
too .ooo.v ooo.o .ooo.v ..
:3 wood o.o.o mooo .08 .
.oL..o. .8o.o .83 .83. o 88.0 8988-
.ooo.v ooo.o .ooo.v o 88 .80
«So o.o.o 8o.o 8.... 8.88 L 08>
.500 .0000 5.60.0- 0 000.58- 0.0.000... 302
5000... .502 0.... 88> 822 2.. 0.0.x confine... .28....

3.6.5.5 0 00.08 0.0.00.3 .502 00.000: 00.00050: ..:0.0.5 00.2.0003 000.2020 .865 000.. "ea 0.00...

73

00.0 m 0 6 6000.006 -
.0000. 00.6000 00. o. 026.0. 0.6.66 0000.0...0 06:00-...0 u 0.0 Nx

 

 

 

 

80o 08o ooo.o 8..o .ooo.v 0
8.8 8.3 too 080 0.o.o .00 0
06o- 8..o 8o.o 8o.o .88.? o o8.o 808...
L80 ooo.o o0..o .ooo.v a
.8.o L.o.o 8o.o 23 .00 L
8.0. .080 8o.o .080- 0 08o 8.80-
08d ooo.o 8..o .ooo.v ..
80o L.o.o 8o.o 0.o.o 0.0 0
83. .080 8o.o .080- o 8o.o 80.80-
080 ooo.o Lio .ooo.v ..
80o L.o.o 8o.o .80 .00 0
08.? .080 8o.o .080- m. 83 08.00-
ooo.o 80o 58v 0 .8 .80
:3 8o.o 23 0.0 00.88 L 28>
500.0 30.0 0000.0. 0 05.80- 0.0.0000. 6000:
80000 2.02 ._._< 0080., 20.. 0... 0.0... 085.9... .000...

:30:—

3..6..0-.5 0 00.500 0.0.023 6000: 00:00: 00.00050: .3065 00.2.0020 09.2000 000.00 000.. "mm 0.00...

74

000 0 0 .0 .08..._.0.0 .
.0000. 05.0000 0... 0. 026.0. 0.6.6.0 0000.0...0 06:00.00 u 0.0 Nx

 

75

 

 

 

0000 .R0 0000 .30 .000.v 0
0000 0000 0.00 0000 0.00 .00 0
0000 0..0- .0000 0.00 .0000- 0 .000 000.000-
0000 0000 00.0 .000.v 0
00.0 0.00 0000 0.00 .00 0
.50- .0000 .30 .0000. 0 0000 000.00.
0.00 0000 3.0 .000.v 0
0000 0.00 0000 0.00 .00 0
0000- .0000 :00 .0000. 0 0000 0005.0-
00.0 0000 .30 .000.v 0
.000 0.00 0000 0.00 .00 0
E00- .0000 :00 .0000- 0 .000 000.00-
0000 030 .000.v 0 .00 000
0.00 0000 0.00 .00 05.0000 R 0.00..
.0000 .000 .0000. 0 000.20- 0.0.003. .0080
000000 .502 ._-.< 000...; 00> 2.. 0.0.x 08%....0... .002...

3.6—.50 0 00.000 0.0.000... 6000: 0.500: 00.00050: .3062. 00.3000... 000.2000 000.0 000.. .00 0.00...

mod w a 8 :52:ch .

.268 8:38 9: 2 9:99 ozmzfim 88.3% $33.20 u Eu Nx

 

 

 

 

$3 «8.0 08.0 3.3 58v a
83 «mad :3 mNo.o Sod .3 m
mad mNo.o 68... «3.0 L3... a god ENS?
and 08.0 9:5 58v 0.
«26 Sad mNo.o «5.0 .5 n
mNo.o .83. mvod .53. u mmod mafia-
mvwd 8o.o 38 58¢ a
88 :3 omo.o mad .5 o
8o.o .83 56.0 £3? a 8o.o mafia-
Sod ooo.o 38 58v a
33 206 omo.o mad .3 m
N36 .82. Sod .53.. a $5.0 «ENE-
ooo.o mid 58v a :5 .80.
Sad omo.o mad .3 2623 h 28>
.93.: 30.0 .39? a mmmsvo- 9283. “magma
Emamm am: ._._< onus? :2 2.. Euflx Sawfly... .282

3.55.83 m 00..me «3503‘ .532". £233. 5.9.352... :33; 33600.; 09.995 3:20 32. Km min...

76

mod w a a Emozegm .

.28.: 9533 m5 2 9:39 ozmzflm 858:6 233-20 u :6 ~x

 

 

 

8o.o 33 58v 8o.o £90 a
$3 memo wood «56 8o.o .3 ¢
38.? #3. as... 635... «8.? a .«mndu 8E3-
58v 58v ooo.o 890 a
83 «8.0 8o.o 8o.o .3 m
.89? as... 585 8o.o- a .33. £39.:
8o.o 58v Bod mmmd a
$3 «8.0 Sod 80.0 .3 m
.59? as... 38... 8o.o- a 3.3.» 2983-
wood 58v Bod 50o a
mad wood 92. ooo.o .3 F
a3? .23. 38.: Sod- a .33 9.0283-
88v «5.0 «5.0 a 8mm .80
Noo.o 8o.o 80.0 .3. 8:82 n 98>
.m E... . .85 8o.o- a 898v F- 2283 252
Guam: Bu: #2 ans? :9, cc. 53x uoomfié .28:

322.....5 m 358 9.0322 202 £33.". 5.396»: .5925 3.25an 33.2 «unas— umw min...

77

mod w a E E8556 .

.268 2:33 05 2 9:59 25:me mocmootfi 95:52.5 u En Nx

 

 

 

 

«woo «woo Pooo.v ooo.o vomo o
ooo.o 33 «So moo ooo.o .3 v
.53. Soo 535 .33 ooo.o- o .33 333..
«Foo Fooov ooo.o onmo o
onoo Noo.o Soo ooo.o .3 o
.23. .2o.o .«ooo ooo.o- o .35 «8.33-
tzo Fooov F Poo Exo o
23 «So ooo.o ooo.o do w
ooo.o- .585 .33 ooo.o- o Noo.o 5083-
nowo Soov F Poo .ov.o o
98 Noo.o ooo.o ooo.o ow. F
ooo.o- ooo.o hood Boo- o ooo.o $983-
.ooov moo Noo.o o moow .80
«So Eoo ooo.o .oo 8:33 k 23>
3 Foo . Foo.o ooo.o- o ooosoo P- 2283 25.2
Emamm Bmz .12 among g3 2.. Enux ncome”... .28:

3:235 v 3..ch 93523 252 .333: 5.329: :33; 3.330.; 85.2 «nuns. "mm 035...

78

mod w a E ESP—Ema .

.89: 05.33 9: 2 9:59 ozmzfim 855:6 293-30 u Eu ox

 

 

 

 

moto two .ooov Eoo ooo.o o
Koo noto «So «So ooo.o ow. o
ooo.o mowo- ooo.o .5o.o ooo.o- o ooo.o E983-
ooo.o .ooov Noo.o oomo o
omo.o Noo.o Noo.o ooo.o .o.m o
3o.o .23 .53 Noo.o- o «So 8033-
ooo.o Fooov Noo.o omo.o o
ooo.o Noo.o Boo ooo.o .3 m
8o.o- .opoo .5o.o ooo.o- o omo.o @933.
«who Fooo.v «Foo ooo.o o
ooo.o Noo.o Boo ooo.o .o....,. F
omo.o- .o.o.o .5o.o ooo.o- o oo..o 8033-
Fooo.v «Poo mso o ooom .moo
Noo.o «Foo ooo.o .oo 8:32 n 28>
.o Poo . Foo.o ooo.o- o ooosov F- 2282 32
Guam: Bmz 4.2 mass? :9, 2.. Ecux 30%....e... .28:
. 4

39.5.85 0 09600 90525 302 62.52“. 5.3960: .5925 3.253.“. 92.2 «unus— 59 min...

79

mod w a on .505ch .
.89: 8:33 on. 9 9:59 26sz 8:295“. 9.33-20 u Eu ox

 

 

 

 

80

ooo.o omo.o omo.o Fooov ooo.o o
Noo.o omo.o Boo vooo Eoo 6...... o
8...? moto- .89? .33 omo.o- o 5.? Soon..-
ooho omo.o Fooo.v Koo o
Noo.o opoo vooo zoo o... n
ooo.o- 58.? .33 omo.o- o 85 ooo.oa-
Koo omo.o .ooo.v ooo.o o
ooo.o ooo.o vooo Soo .o.m o
omo.o- .39? .83 omo.o- o ES Coda-
Noo.o omo.o Fooov ooo.o o
ooo.o moo ooo.o zoo om o
ooo.o- .ouod. .«2o omo.o- o onto omo.oa-
omo.o .ooov ooo.o o ooo .ooo
ooo.o ooo.o moo o... 8:32 n 28>
.39? .33 omo.o- o ooo.o &. 9983 ooooom
Emamm 3m: 42 adage, :9 2.. Enux uoomfiu... .28:

3.3.2.5 m 00.68 «.5333 63¢: .35me 5.3962". .3925 3.2602“. 52.2 «Una—2 "5 22¢...

.89: 8:32. on: 9 9:99 2632.0. 8555:. 233-20 u :6 «X

moo w o a 22:29... .

 

 

 

 

ooo.o omo.o ooo.o ooo.o omo.o o
omo.o homo o:o.o ooo.o Eoo o... o
wood- vuo.o .ouoo. .oo:.o omo.o- o 53 ooo.o:m-
oBo Roo :ooo.v omo.o o
voto o:o.o ooo.o Eoo .o.m :
Noo.o- .oaoo. .oo:.o omo.o- o :Noo 3.2m-
omo.o omo.o ooo.o omo.o o
ooo.o o:o.o ooo.o Soo .o.m o
too- .23. .oo:.o omo.o- o ooo.o ooo.o:m-
mood Roo ooo.o ooo.o o
ooo.o o:o.o ooo.o oFoo o... m
omo.o .oaoo. .o~:.o omo.o- o Noo.o ooo.o:m-
:moo :ooo.v ooo.o o 3o .20
o:o.o ooo.o o:o.o o... 3:32 : 93>
.ouoo. .oo:.o omo.o- o ooo.o a- 9283 32%
imam: :52 ._._< 522 =2, 2.. Euux vacuums... .28:

332.8% e 3500 20303: «252: £23.32". 223962: =38; 03330.n— 5m=2 mugs. "mm «BE.

81

mod w a um 8:85ch .

Enos. 9532. m5 2 9:36.. ozmzflm 85.8% 9333-20 .u Eu Nx

 

 

 

 

ooo.o ooo.o ooo.o :ooov «.83 o
53 33 o:o.o ooo.o 33 3.3 o
83 33- .23. .53 omo.o- o mono ooo.o:m-
mooo omo.o ooo.o goo o
5.3 23 omo.o 23 3.3 :
omo.o .23. .33 .33. o 33 338.
.33 omo.o :ooo.v ooo.o o
two 93 wood 33 o... o
omo.o- .23. .83 omo.o- o 33 3.2m-
ooo.o omo.o ooo.o :83 o
$3 23 ooo.o 33 3.3 o
omo.o- .23. .83 omo.o- o «So .335.
R3 :ooo.v omo.o o 23 .20
93 ooo.o 23 3.3 2:32 :. 22>
.33.. .8 :d 890. n mead :N- 9283. 632”.
Guam: Bmz ._._< no.2? :9 2.. Eamon 33%... .28:

33:85 m 3.560 95303: “mama .333: 5.3352“. 55.25 32532; 32.2 «anus. "an 293...

82

mod w a F. 532ch .

.39: 3:33 9: 9 9:39 ozmzflm 8:936 938-20 u :6 Nx

 

 

 

 

83 33 58v 83 83 3
23 883 «83 23 83 .3 v
.23 .33. .23 S3- .33- 3 88.3 83::-
23 58v 83 83 a
83 83 23 83 3 m
.83 .83 83- .83- a .83 8.32-
o83 58v 83 93 3
:3 33 93 83 .3 N
33- .83 23- .83- a 8o.o 83::-
83 58v 83 F 53 a
8.3 83 23 983 .3 F
83 .33 23- .23. a 83 83.52.
Soov $3 93 a 88 .80
83 23 83 .3 8:88 o 28>
.056 3.0.0- 35.? a 80.99- 9239‘ 262
53mm 5.2 ._._< 89.: :2, 2.. Euax nook”: .322

33:—Em m 8500 «.8325 262 8.33... 5.30.62“ .3925 3.263;. co_w>< 823... 3a 038.

83

609: 3:33 9: 2 93.9 28sz 8:836 933-20 u :6 Nx

mod w a 8 €35ch .

 

 

 

 

2 3 83 83 B3 883 a
F F 3 83 83 m :3 83 .3 v
E3 83- .83 23- .83. a 83 88.99-
83 58v 83 93 a
83 N83 23 83 .3 m
83 .83 8:3- .83. a 83 83.28.
83 58v .23 83 a
83 8o.o 23 83 .3 N
83- .83 83- .83- a 53 83::-
83 83 83 23 a
83 83 8:3 83 .3 F
8o.o- .23 83- .83. a 83 8.22-
58.v 83 93 a 88 .80
83 93 83 .3 8:88 o 98>
.m Ed v 5.0- .N 5.0. a 89m 5 F- 9232 262
imam: 252 ._._< 888 =2, 8. tawx 30%.”... .802

33:—ca v 0285 883.3 252 .8385 c2383.“ :33? 3.330.."— :o_a>< 826... "mm 23...

84

.38: 35.3me 35 2 328.9 2838 838:6 98.8-50 u :6 «x

mod w a 8. 8.8on8 .

 

 

 

 

383 83 58v 38.3 83 3
83 8 3 83 m 53 83 .3 v
83- 83 .83 83- .83- 3 38.8 83.88-
83 83 83 83 3
83 ~83 83 83 .3 3
383- .83 83- .83. 3 33 38.88-
.83 83v 383 83 3
33 83 83 83 .3 3
~83 .83 $3- .83. 3 83 83.88-
83 68v 83 83 3
83 83 83 83 .3 F
83- .83 83- .83. 3 83 83.88-
83.v 83 83 3 88 .80
83 83 83 .3 8:88 3 28>
.3 5.3 v 8o.o- .u 56. a 08.3 E F. 883034. 262
.283 Bmz ._._< ommcg =2 2.. 83.x 38333.... .822

323.35 m 3.500 80322 302 8.33m 3.39.9.8". .3925 3.3033 :o_a>< Ego... "on 333...

85

mod w 3 .3 Enocigm .
.335 35.338. 8.. o. 32.3.9 3.8.33.3 33:98.6 93:33.30 u Eu Nx

 

 

 

 

83 .83 839 83 83 3
33 :3 N83 33 83 .3 3
.83.. 83 .83 83- 383 3 .82. 33.38-
83 83 83 883 3
33.3 «83 33 83 .3 3
.38.. L83 83- 83 3 .83. 8338-
.83 58v 833 .83 3
83 83 33 33 .3 3
RE .383 83- 83 3 33 83.38-
333 83.v 833 883 3
833 «83 33 33 .3 3
.8; .383 83- 83 3 L83 3.38-
83 3.3 83 3 38 .80
83 333 83 .3 8:383 3 98>
.336 v 5.0- n 5.3 3 mom-vmw- 93.303. .3338.
3.333: .532 ..._< 882, 33> 2.. Eumx 38.3.9383 .82.

3.32.53 3 3..me 283213333. .8332“. 3.83.50: .3995 3.2.30.3. co.u>< 82.0... Km 23...

86

.3308 35.3333 35 2 3233.3. 3333333 333333333 333333-20 u Eu 3x

mod w 3 33 3:323:93 .

 

 

 

 

333.3 33.3 333.3 333.3 333 3
333.3 33.3 333.3 :33 33.3 3.3 3
333.3 333 .3333 333.3- 333.3 3 33.3 33.333-
3333 333.3 333.3 33.3 3
33.3 333.3 :33 33.3 3.3 n
33.3 .3333. 333.3- 35.3 3 333.3 333.333-
333 333.3 333.3 333.3 3
333.3 3333 3.3.3 333 3.3 3
333.3 .333 83.3. 333.3 3 3:; 33.333-
3333 333.3 333.3 333.3 3
333.3 333.3 :33 33.3 3.3 3
333.3 .3333 333.3- 333.3 3 333.3 £3333-
333.3 3.33 32.3 3 333 .330
333.3 333.3 33.3 3.3 35.3333 3 3.33
.356 v 5.0- n 5.3 a momémw- 33.333. 333333.
.2333: 332 ._._< 33.23., 33> 3:. 3.3.x 385.33....33 .3332

3.33.85 3 33530 238033 332 .9333: 32333533. .3933 33.2.3333 :o_3>< 805... "33 333...

87

.339: 353333 35 2 3233. 3333333 33:33.53 333333330 u Ev 3x

mod w 3 .3 3:335:93 .

 

 

 

 

 

 

 

 

333.3 3:3 333.3 333.3 333.3 3
333.3 333.3 3333 333.3 33.3 3.3 3
333- 333.3 .3333 333.3- 333.3 3 33.3 333.333-
333.3 333.3 333.3 333.3 3
33.3 333.3 33.3 33.3 3.3 3
E3 .3333 33.3- 333.3 3 3333 333.333-
333.3 333.3 333.3 333.3 3
333.3 333.3 333.3 33.3 3.3 3
333.3 .3333 333.3- 333.3 3 33.3 333.333-
333.3 333.3 333.3 3333 “,1
333.3 333.3 333 $33 3.3 3
:33 .3333 333.3- 333.3 3 3333 333.333-
3333 3.33 333.3 3 333 .330
.333 333.3 333.3 3.3 35.3333 3 3.333,
.353 v 5.5- 3 5.0 3 333.3333. 33333393. 333333.
35333 23.2 ._._< 33%.; 33> 33. 3.3 .x 38339....33 .3332

33.2.85 3 33.330 3.3323 332 .3333”... 5.3333333. .3933 33.3333... :o_3>< 823... "33 333...

88

Comparing Between Degree Similarity Analyses, and Between Adopter

Group Analyses

The magnitude and significance of static covariates are virtually identical
across 3, 4, and 5 degree similarity models. There is a greater prevalence of
significant regression coefficients for new adopters than repeat adopters. For
each automobile model, static covariate results are consistent across degree
similarities. The time varying influence covariates show a wider variety of results
as some models have significant coefficients across all degree similarities in new
and repeat adopter groups (as with the Dodge Neon) while others vary across
both degree similarity, and new and repeat adopter group analyses (such as for
the Mazda Miata).

Significant results at lower degrees of similarity are generally
accompanied by significant results at higher degrees of similarity although the
reverse is not the case. When results are apparent across different degrees,
coefficients increase in magnitude the higher the degree of similarity. In general,
this indicates the more similar current adopters are to prior adopters, the stronger

the effect.

Inferring New and Repeat Adopter Effects through Model Selection

For regression results conducted on current New Adopter groups, models
2 and 3 are nested within model 4, and for regressions conducted on current

Repeat Adopter groups models 6 and 7 are nested within model 8. These

89

models utilize influence covariates calculated from prior new and repeat adopter
groups. Models 1 and 5 utilize influence calculated from all prior adopters and
are non-nested with models 2, 3, and 4; and 6, 7 and 8 respectively.

For nested analyses, the model of best fit is determined by Chi-square
significance tests. An unrestricted model (such as 4 or 8) has better fit than
restricted models (against 2 or 4; or 6 or 7) when the improvement in Chi-square
is significant. If the unrestricted model does not have better fit, a restricted model
is chosen. For Mazda Miata New Adopter group degree 5 similarity results,
model 4 fits significantly better than model 2 (20.352-8.049=12.303, >3.84, 1
d.f.), however model 4 does not fit significantly better than model 3 (20.352-
19.529=0.823, <3.84, 1 d.f.). Thus model 3 is the best model and indicates
purchase by New Adopters is accelerated by prior Repeat Adopters. The same
comparisons and conclusions occur for Dodge Neon New Adopter degree 4 and
5 similarity results. For Jeep Grand Cherokee New Adopter degree 4 similarity
results, model 3 has better fit than model 4, because model 4 does not fit
significantly better, (7.486-6.987=0.499, <3.84, 1 d.f.). This same conclusion is
reached for the Mazda Miata New Adopter degree 4 similarity results.

When both restricted models are significant, and the unrestricted model
does not have better fit than either, there is no statistical basis for choosing
between restricted models. If the coefficients for the restricted models are of the
same sign, the most likely reason for lack of improved fit in the unrestricted
model is collinearity between NEW and REPEAT. In this instance it is

inappropriate to use the unrestricted analyses to test simultaneous effects of

90

NEW and REPEAT. An equivalent means of testing NEW and REPEAT when
effects seem collinear is to combine them as a single group to assess
demographic similarity. This is achieved in non-nested analyses 1 and 5 that are
the correct model choice when restricted models indicate similar effects of NEW
and REPEAT, and unrestricted analyses do not fit significantly better than
restricted models.

This occurs for the Dodge Neon degree 3 similarity New Adopter group,
where models 2 (x2 = 4.060, >3.84 1 d.f.) and 3 (x2 = 4.324, >3.84 1 d.f.) fit
better than the baseline model, and model 4 does not fit significantly better than
models 2 (4.945-4.060=0.885, < 3.84 1 d.f.) or 3 (4.945-4.324=O.621, < 3.84 1
d.f.). Non-nested model 1 is a better model than model 4 because its single
demographic similarity ratio covariate ALL consists of aggregated new and
repeat adopters instead of the two separate NEW and REPEAT ratios used in
model 4. This eliminates the problem of collinearity that occurs in model 4. This
inference is supported by the ALL coefficient for model 1 ([3=-0.194, p=0.012)
that is consistent in sign with the model 2 NEW coefficient (B=-0.153, p=0.045),
and model 3 REPEAT coefficient (B=-O.118, p=0.038). When model 1 instead of
model 4 is chosen to examine simultaneous new and repeat effects, the
simultaneous NEW and REPEAT adopter effects in models 2 and 3 are of the
same polarity as ALL in model 1. The same comparison procedure for model

selection is followed for Dodge Neon repeat adopter degree 3 analyses.

91

Table 40 presents the identified prior NEW and REPEAT effects upon
current New and Repeat Adopters. Coefficients effects are summarized as faster

(-) or slower (+) effects upon time to adoption.

Table 40: Identified Prior New and Repeat Effects upon Current New and
Repeat Adopter Adoption Time

 

5 degree 4 degree 3 degree
similarity similarity similarity
Adopter Model NEW REPEAT NEW REPEAT NEW REPEAT
Group
New Dodge Neon
Jeep Grand Cherokee
Mazda Miata
Toyota Avalon
Repeat Dodge Neon + + 4-
Jeep Grand Cherokee
Mazda Miata
Toyota Avalon +
( - ) indicates faster time to adoption
( + ) indicates slower time to adoption

+I|o

Static Covariates

With one exception, significant coefficients for income always accelerate
time to adoption (for Dodge Neon Repeat Adopters higher income delays
adoption). As expected, the greater the number of household vehicles the
longer the time to adoption. In line with expectations the more recent the model
year of the vehicle driven previously, the longer the time to adoption. The Mazda

Miata Repeat Adopter group is the only exception.

92

Time Varying Covariates
Coefficients for the effects of prior adopters upon New Adopters were
predominantly negative. The few coefficients for the effect of prior Adopters

upon Repeat Adopters were consistently positive.

Effects upon Repeat Adapters

For the Dodge Neon, later repeat adopters have demographic
characteristics in common with prior new adopters. This implies that late repeat
adopters, but not early repeat adopters, are similar to prior new adopters. No

other consistent effects are apparent.

Sensitivity Tests

To determine relative impact of the 6 demographic dimensions upon the
influence covariate results, analyses for new and repeat adopters were repeated
6 times, each with one of the demographic dimensions absent. For calculation of
the demographic similarity ratios, this means only 5 dimensions were available.
Thus degree 5 similarity requires all of 5 dimensions to be similar, degree 4
similarity requires 4 of 5 dimensions to be similar, and degree 3 similarity
requires 3 of 5 dimensions to be similar. The impact upon the prevalence of
significant results when removing each demographic dimension from the
influence covariate is presented in Tables 41 and 42 for New Adopters and

Tables 43 and 44 for Repeat Adopters.

93

Table 41: Demographic Degree Similarity Sensitivity Test Results for New
Adopters, Dodge Neon and Jeep Grand Cherokee

 

5 degree 4 degree 3 degree
similarity similarity similarity
Model Dimension NEW REPEAT NEW REPEAT NEW REPEAT
Removed

Dodge Neon none - - - -
age - - - -
education - - - - -
gender - -
income - - - -
occupation - -
race -

Jeep Grand none - -

Cherokee age . .
educafion
gender
income - - -
occupafion
race

( - ) indicates faster time to adoption
( + ) indicates slower time to adoption

 

 

 

Table 42: Demographic Degree Similarity Sensitivity Test Results for New
Adopters, Mazda Miata and Toyota Avalon

 

 

 

5 degree 4 degree 3 degree
similarity similarity similarity
Model Dimension NEW REPEAT NEW REPEAT NEW REPEAT
Removed
Mazda Miata none - -
age - - - -
education - -
gender - -
income - - -
occupation .
race - -
Toyota Avalon none - 4-
age - +
education
gender + 4-
income - + - +
occupation - + - + -
race + + +

 

( - ) indicates faster time to adoption
( + ) indicates slower time to adoption

94

Table 43: Demographic Degree Similarity Sensitivity Test Results for
Repeat Adopters, Dodge Neon and Jeep Grand Cherokee

 

5 degree 4 degree 3 degree
similarity similarity similarity

 

Model Dimension NEW REPEAT NEW REPEAT NEW REPEAT
Removed
Dodge Neon none 4—
age +
educafion +
gender +
income + +
occupation +
race +
Jeep Grand none
Cherokee age
education
gender
income
occupation
race

( - ) indicates faster time to adoption
( + ) indicates slower time to adoption

4.
++++

+
++

Table 44: Demographic Degree Similarity Sensitivity Test Results for
Repeat Adopters, Mazda Miata and Toyota Avalon

 

5 degree 4 degree 3 degree
similarity similarity similarity

 

Model Dimension NEW REPEAT NEW REPEAT NEW REPEAT
Removed
Mazda Miata none
age
educafion -
gender -
income -
occupafion -
race -
Toyota Avalon none
age 4-
education
gender
income 4-
occupation
race 4-

( - ) indicates faster time to adoption
( + ) indicates slower time to adoption

++++

95

Hypothesis Testing

Hypothesis 1. For second generation products, purchase by New Adopters
is accelerated by social connection to prior New Adopters.
The results presented in Table 40 provide partial support for Hypothesis 1.
Significant negative effects upon time to adoption are observed for just one
degree similarity for the Dodge Neon (degree 3) and Toyota Avalon (degree 5).
The presence of only 2 accelerative effects of prior New Adopters upon current

New Adopters provides weak support for Hypothesis 1.

Hypothesis 2. For second generation products, purchase by New Adopters
is accelerated by social connection to prior Repeat Adopters.
Significant negative coefficients for REPEAT are present for the Dodge
Neon (degrees 3, 4, and 5), Jeep Grand Cherokee (degrees 4 and 5), and
Mazda Miata (degrees 4 and 5), but not the Toyota Avalon that shows a single
positive coefficient (degree 5). This indicates purchase by current New Adopters
is predominantly accelerated by prior Repeat Adopters. Three of the four second
generation automobiles show this effect, providing strong support for

Hypothesis 2.

Hypothesis 3. For second generation products, purchase by New Adopters
is accelerated more by social connection to prior Repeat

Adopters, than social connection to prior New Adopters.

96

For the Dodge Neon, Repeat Adopters accelerate time to adoption across
degree 3, 4 and 5 similarities. NEW accelerates time to adoption for only degree
3. For both the Jeep Grand Cherokee and Mazda Miata, REPEAT accelerates
purchase in New Adopters at degree 4 and 5 similarities, and no NEW effects are
observed. For each of the three second generation products that show
consistent effects of prior adopters, REPEAT effects are consistently more
prevalent than NEW across different degrees of. demographic similarity. This

provides strong support for Hypothesis 3.

Hypothesis 4. For second generation products, purchase by Repeat
Adopters is not accelerated by social connection to prior
New Adopters.

The results displayed in Table 40 for current Repeat Adopters show
positive coefficients for the Dodge Neon for NEW at degree similarities 3 and 5,
and no negative coefficients for NEW. This indicates that purchase by Repeat
Adopters is not accelerated by prior New Adopters, and the positive coefficients
even suggest Repeat Adopters of the Dodge Neon lack social connections to

prior New Adopters. This provides strong support for Hypothesis 4.
Hypothesis 5. For second generation products, purchase by Repeat

Adopters is not accelerated by social connection to prior

Repeat Adopters.

97

There are no negative coefficients to indicate purchase by Repeat
Adopters is accelerated by prior Repeat Adopters. However there are positive
REPEAT coefficients for the Dodge Neon at degree similarity 3, and for the
Toyota Avalon at degree similarity 5, which provides preliminary evidence
Repeat Adopters lack social connections to New Adopters. The complete lack of
negative coefficients for prior Repeat Adopter effects upon current Repeat

Adopters provides strong support for Hypothesis 5.

98

Chapter 5

DISCUSSION

This chapter discusses the research results and conducts several post-hoc
analyses to aid in explanation of the findings by relaxing the assumption that new
and repeat adapters are different, and applying the method to existing products.

Evidence is found the method may be of value for existing products.

The results of the research can be summarized thus:
a For second generation products, new adopters are susceptible to
influence from prior adopters, and repeat adopters are not.
0 For second generation products, repeat adopters are more infectious than

new adopters.

New and Repeat Adopters

New Adopters are susceptible to influence, and are clearly an imitative
segment. Early purchase in this group is associated with prevalent connections
to earlier adopters. This supports the assertion that later adopter categories
such as the late majority and laggards have fewer social connections than earlier
categories such as early adopters and the early majority (Rogers 1995:283).

A possible explanation for the lack of significant results for the Toyota
Avalon is its relative lack of innovativeness as a new product. The first and

second generation Toyota Avalon models were built on a modified stretched

99

platform version of the simultaneously available smaller Toyota Camry that was
already in its third generation when the Avalon was launched. In addition, the
Toyota Avalon engine was almost identical to that used in the Toyota Camry.
Adopters of the Toyota Avalon may not have viewed product performance as
uncertain, due to a strong positive brand image of the Toyota Camry which was
(and remains) one of the highest selling passenger cars in the United States. A
sufficient level of media information would make consultation of the social
network unnecessary.

There is a clear lack of susceptibility for Repeat Adopters although the
smaller sample size and corresponding lower statistical power for these tests
prevents direct comparisons with New Adopter susceptibility results. This study
thus provides prima facie evidence that adopters of a prior generation are
effective in stimulating adoption for the next product generation, a finding
consistent with Kim, Chang and Shocker (2000) and Islam and Meade (1997),
although the results provide no evidence second generation adopters encourage
first generation adopters to switch.

Repeat Adopters are more infectious than New Adopters, indicating
innovative adopters have greater credibility gained from product experience with
the first generation product. It is also possible that as an innovative segment,
Repeat Adopters are more outwardly connected than New Adopters. The
difference between outward and inward connections cannot be determined in this
study because the similarity measure is symmetrical. A relational network

approach to inward and outward network nominations could distinguish between

100

connection flows, but the data requirement is outside the scope of the data used
in this investigation. In either case, the existence of strong influence between
innovative and imitative groups means firms should target innovators in order to

improve the chances of new product launch success (Mahajan and Muller 1998).

Propensity

The variables used to capture an individual’s propensity to adopt
independent of social influence are predominantly in the expected direction with
no discernible difference in effect by New versus Repeat Adopter groups. While
prior work has found that innovators have inherently higher propensity, in this
study there appears to be no difference in the role propensity plays in making the

adoption decision by innovators versus imitators.

Sensitivity Results

In sensitivity tests each of the six demographic variables were removed in
sequence from the calculation of the demographic similarity ratio, and all
regression and model selection procedures repeated. This placed a more strict
constraint upon similarity requirements as individuals have one fewer dimension
for comparison. Most results retain their significance in the face of demographic
dimension removal. Some changes occur for persistence across similarity
degrees. The single notable change in result for Repeat Adopters is the

omission of gender for the Toyota Avalon that results in persistent positive

101

coefficients when all other analyses suggest a lack of consistent effects. Several
differences are observed for New Adopters.

Omission of race causes the single biggest change for the Dodge Neon
where accelerative effects of prior Repeat Adopters almost disappear. This
same effect occurs for the Jeep Grand Cherokee when gender or race is omitted
and for the Mazda Miata when occupation is omitted. These dimensions seem to
be particularly important to the flow of influence between prior Repeat and
current New Adopters. The interpretation for the Dodge Neon is that similarities
of age, education, gender, income, and occupation are not effective in
establishing influence without simultaneously considering race. Similarly, for the
Jeep Grand Cherokee, past Repeat Adopters are connected to current New
Adopters largely through gender and race, and the connection is largely through
occupational similarity for the Mazda Miata. Consideration of these omitted
factors alone would likely be insufficient to link demographic similarity and
adoption time, but must be used in conjunction with additional demographic
factors.

The results for the Toyota Avalon are in contrast with the others in that the
incidence of significant coefficients increases rather than decreases with removal
of gender, income, occupation, and race. Positive coefficients appear for prior
Repeat Adopter effects indicating that early New Adopters are not similar to prior
Repeat Adopters, while later New Adopters are. A possible explanation is the
relative lack of innovativeness for the Toyota Avalon identified previously. This

would result in low usage of the social network for product information collection,

102

meaning information may have to travel through indirect network ties. The
negative coefficients when income and occupation are omitted also indicate that
New Adopters may be linked to prior New Adopters. These negative coefficients
indicate New Adopter similarity with earlier prior New Adopters, and the positive
coefficients indicate relative New Adopter dissimilarity to the earlier Repeat
Adopters. This suggests early New Adopters are a distinct market segment from
early Repeat Adopters.

Sensitivity analyses provide evidence of hierarchical importance in the use
of demographic characteristics to proxy for internal influence. Future use of
demographic characteristics to measure influence should therefore asses relative

importance of the demographic dimensions used.

Influence in Diffusion

Aggregate level diffusion studies investigate internal influence
mechanisms by examining the variance of diffusion parameters across adopting
populations (e.g. Kumar and Krishnan 2002). This necessitates consideration of
multiple adoption contexts at the population level. Network studies obtain
detailed insights in just a single adoption context, but each individual’s position in
the network must be described by examining their specific social relations with
others. The demographic similarity approach used in this research provides
greater insight into the diffusion process compared to aggregate level diffusion
studies, and uses individual level information more parsimonious than typically

used in network studies of diffusion. This practice of inferring word of mouth

103

influence from demographic comparisons thus provides an intermediate means
of modeling internal influence in an adopting population that is suitable for
samples sizes between those normally used in aggregate and individual level
studies. Indeed demographic level customer data (such as scanner or survey
data) are quite common in marketing contexts, holding promise for more
widespread application of the technique developed here.

This study modeled influence in a manner consistent with the influence
factors used in the heterogeneous diffusion model proposed by Strang and Tuma
(1993), whose most important feature is arguably the incorporation of time
varying internal influence. However the approaches used here and by Strang
and Tuma differ in their measurement of propensity, proximity, infectiousness
and susceptibility, and in addition the heterogeneous model assumes population
level data rather than samples of the kind used in this study. Nevertheless, the
current demographic similarity results indicate the factors proposed as key to
capturing adoption behavior by Strang and Tuma have explanatory power
beyond population level samples and individual level relationship data.

While other studies have also utilized an imitation rationale to infer
influence by virtue of similarity (e.g. Greve 1998; Soule and Zylan 1997), further
validation for the use of demographic similarity as a proxy measure for internal
influence can be provided through use of simultaneous relation and demographic
assessment of the kind applied by De Bruyn and Lilien (2004), who utilized online
referrals to determine the role of network actors in influencing others of varying

demographic similarity and social proximity; thus allowing for simultaneous

104

assessment of network relations and individual demographic characteristics. A
useful starting point would be analysis of existing social network diffusion data
that customarily include extensive information concerning relationships,
demographics and adoption timing. Common incidence of relationship and
demographics may then be first ascertained before incorporation of a

demographic similarity index in a longitudinal context

Comparison with an All Adopter Group

A naive set of regression analyses were conducted for a single adopter
group to determine the value of treating individuals as distinct new and repeat
adopters. In these models current adopters are not treated as two distinct new
and repeat groups, but as a single “all adopter” group. Table 45 presents the
identified effects for prior ALL adopters upon current All Adopters. To aid in
understanding of these naive regression results, additional expanded analyses
were also conducted that incorporated distinct prior NEW and REPEAT

covariates. These results are presented in Table 46.

Table 45: Naive Analyses, Identified ALL Effects upon All Adopter Adoption
Time

 

 

 

 

5 degree 4 degree 3 degree
similarity similarity similarity
Model All All All
Dodge Neon
Jeep Grand Cherokee -
Mazda Miata -

 

Toyota Avalon

( - ) indicates faster time to adoption
( + ) indicates slower time to adoption

 

 

 

 

105

Table 46: Expanded Naive Analyses, Identified NEW and REPEAT Effects
upon All Adopter Adoption Time

 

5 degree similarity 4 degree similarity 3 degree similarity

Model NEW REPEAT NEW REPEAT NEW REPEAT
Dodge Neon
Jeep Grand Cherokee - - - .
Mazda Miata - . . -

Toyota Avalon

( - ) indicates faster time to adoption
( + ) indicates slower time to adoption

 

 

 

 

 

 

 

 

Comparing All Adopters versus New and Repeat Adopter Groups

Naive analyses for All Adopters treat New and Repeat Adopters as a
single group, thereby assuming that all potential adopters are equally
susceptible. Table 45 presents the simplest class of naive results. This analysis
yields only two results, negative coefficients for the Jeep Grand Cherokee for
degree 5 and the Mazda Miata for degree 4. These are suggestive of word of
mouth effects but are not prevalent across similarity degrees. Expanding the
naive analyses to include distinct prior New and Repeat adopter effects upon All
Adopters is more informative.

Expanded naive analyses reveal more prevalent effects across similarity
degrees. Diffusion effects, or the absence of, are identified correctly for the Jeep
Grand Cherokee, Mazda Miata and Toyota Avalon; but are incorrectly absent for
the Dodge Neon. This incorrect result is likely due to the opposite effects found
in the separate New and Repeat Adopter analyses for the Dodge Neon that have
offset each other in the All Adopter analyses. The diffusion effects from prior

New and Repeat groups for the Jeep Grand Cherokee are consistent with the

106

New Adopter group findings, but results in misattribution of effect for the Mazda
Miata where influence now appears to stem largely from New rather than Repeat
Adopters.

Conclusions based upon All Adopter analyses are substantially different
compared to 2 of 4 New Adopter analyses, and 3 of 4 Repeat Adopter analyses;
however when viewed as a complementary analysis, the results for All Adopters
complement the separate New and Repeat analyses. Absent All Adopter results
can be due to contrasting New and Repeat effects, or absent New and Repeat
effects. These are apparent for the Dodge Neon and Toyota Avalon respectively.
The Jeep Grand Cherokee expanded all adopter analyses (Table 46), are largely
consistent with New Adopter results, but not with Repeat Adopter results (Table
40). This indicates Repeat Adopters are quite infectious, with effects pervasive
over New and All Adopter analyses. For the Mazda Miata, All Adopter analyses
would attribute influence largely to prior New Adopters, but separate analyses
identify influence as stemming from Repeat Adopters. The All Adopter analyses
therefore mask the infectious effect of Repeat Adopters.

While All Adopter analyses can potentially reveal the presence of diffusion
effects, they do not reveal the relative susceptibility of New Adopters, and relative
infectiousness of Repeat Adopters. Thus All Adopter analyses provide additional
insight and support the findings of the New and Repeat analyses, but in isolation

do not accurately reveal attribution of word of mouth effects.

107

Comparison with Existing Products

What happens when the demographic similarity ratio is applied to buyers
of existing products instead of adopters of new products? This is a natural
question to ask as the marketplace for existing products is information rich
compared to that for new products, and interpersonal word of mouth effects are
hypothetically absent when there is little need to seek information from the social
network. Compared to new products, buyers of existing products can assess
likely product performance with greater confidence from available media and do
not need to rely upon inter-personal information. Social connections between
buyers are not used to exchange information regarding product performance.
This makes the purchase decision insensitive to the actions of prior buyers. This
is true for both first-time and loyal buyers of existing products who represent
inexperienced and experienced market segments. Thus the presence of inter-
personal influence for adopters of new products, and its absence for buyers of
existing products would provide a means of distinguishing between new product
“adopter segments” and existing product “market segments”.

The same procedure was followed to test for evidence of internal influence
between early and later buyers of existing products. 18 automobile models were
identified with complete generation sales years within 1999-2004, of these 8
satisfied the sample size requirements. The piecewise Weibull procedure again

overlapped substantially with Cox regression results indicating sampling variation

108

was suitably accounted for. Vehicle details and abbreviated analyses are shown

in Tables 47 and 48.

Table 47: Existing Automobile Models Analyzed

 

Model G2 62 62 All New Repeat New Repeat
start and sale Sub- Sub-

years Sample Sample
Buick LeSabre 2000 2004 6 7003 1964 983 804 522
Chevrolet Monte Carlo 2000 2004 6 4295 1737 336 1 136 241
Ford Mustang 1999 2004 7 5521 1788 211 1202 152
Ford Taurus 2000 2004 6 5050 1032 272 572 191
Mercury Sable 2000 2004 6 3114 1287 264 793 165
Pontiac Bonneville 2000 2004 6 4414 1286 627 766 429
Pontiac Grand Am 1999 2004 7 7381 2005 372 1287 266
Subaru Legacy 2000 2004 6 6340 3337 641 2179 439

 

There are three types of results of the prior New Buyer covariate upon current

adopters of non-new automobiles. The first effect is observed among a group

that includes the Mercury Sable, Pontiac Grand Am and Subaru Legacy where

there are negative coefficients for the ‘new’ component of NEW and REPEAT.

The effect is consistent across all degree similarities, and suggests current New

Adopters are similar to prior New Adopters. The second effect is observed in a

group that includes the Chevrolet Monte Carlo, Ford Taurus, and Pontiac

Bonneville that show positive coefficients for NEW for degree similarity 5, and as

part of ALL in degree similarity 4. A significant positive NEW coefficient is also

present for the Pontiac Bonneville at degree similarity 3. These positive

coefficients indicate current New Buyers are dissimilar to prior New Adopters for

these models. The third type of result is a lack of significant effect upon the

Buick LeSabre and Ford Mustang where there are no consistently significant

109

coefficients across degree similarities, and no demographic similarities are

apparent.

Table 48: Existing Products: Identified Prior New and Repeat Effects upon
Current New and Repeat Adopter Adoption Time

 

5 degree 4 degree 3 degree
similarity similarity similarity
Model NEW REPEAT NEW REPEAT NEW REPEAT

 

New Buick LeSabre - - 4-
Ford Mustang
Mercury Sable - + -
Pontiac Grand Am
Subaru Legacy
Chevrolet Monte Carlo
Ford Taurus
Pontiac Bonneville

Repeat Buick LeSabre

Ford Mustang
Mercury Sable
Pontiac Grand Am -
Subaru Legacy - + - + - +
Chevrolet Monte Carlo -
Ford Taurus 4-
Pontiac Bonneville +

( - ) indicates faster time to adoption

( + ) indicates slower time to adoption

+++nu
+
+++-
++++++
I
+++++

 

 

Significant coefficients for REPEAT are always positive. Present for the
Mercury Sable, Pontiac Grand Am, Subaru Legacy, Chevrolet Monte Carlo, Ford
Taurus and Pontiac Bonneville, this dissimilarity of prior Repeat Adopters to
current New Adopters is observed across 6 of the 8 non-new automobiles.

A persistent positive coefficient for the effect of prior Repeat Buyers upon
current New Buyer time to adoption indicates New Buyers are not influenced to
purchase early by Repeat Buyers. This implies that early new and early repeat
buyers are distinct market segments that have few demographic connections.

However later New Buyers have demographic qualities in common with Repeat

110

Buyers, but these common characteristics do not result in accelerated adoption.
This can indicate that later New Buyers are influenced by prior Repeat Buyers
whereas earlier New Buyers are not, in which case any influence that may flow to
later New Buyers is delayed while influence is transmitted through indirect
network channels; or that later New Buyers make similar decisions to prior repeat
buyers (but later in time) due to similar resource sets and circumstances. If
similar demographics can be used to proxy for similar resources, the implication
is that given similar resource sets, new buyers take longer to reach a decision
and thus have a higher threshold than repeat buyers due to greater risk adversity
and a need to accumulate more information.

Early New Buyers are thus always different from Repeat Buyers. This is
consistent with the market segmentation view that buyers of varying loyalty are
driven by different factors and exhibit different purchase behavior (Yim and
Kannan 1999; Blattberg, Buesing and Sen 1980; Starr and Rubinson 1978). The
results here indicate this is true even if there are common demographic qualities.

These results for existing products provide insight into the findings for the
Toyota Avalon, where the lack of significant results in initial analyses, together
with the significant results found during sensitivity analyses, reveal the Toyota
Avalon bears strong resemblance to existing products, thereby supporting the
previously offered explanation that its use of an existing modified platform and
engine led to its perception as a less innovative product. The results for second

generation and existing products can be summarized as contrasting influence

111

effects from prior New and Repeat Adopters, upon current New Adopters. These

contrasting effects are presented in Table 49.

Table 49: Demographic Similarity Coefficient Effects of Prior Adopters
upon New Adopter Time to Adoption

 

Product Prior New Adopter Prior Repeat Adopter

 

Second Generation
Dodge Neon

Jeep Grand Cherokee
Mazda Miata

Existing 1
Mercury Sable
Pontiac Grand Am
Subaru Legacy

Existing 2

Chevrolet Monte Carlo

Ford Taurus + +
Pontiac Bonneville

 

Positive Influence Coefficients

Negative coefficients for demographic similarity influence variables
indicate accelerated purchase, and a lack of significance indicates no
accelerative effects. Interpretation of positive coefficients is somewhat more
complex. The presence of positive coefficients indicates social connection to
prior adopters is associated with later adoption. This means either information
travels slowly through the social network, or that early and later buyers have
similar resources (indicated through similar demographic characteristics) but
different decision making thresholds. Positive coefficients occur most often for

Repeat Adopters and Repeat Buyers and both of these groups have little reason

112

to utilize social contacts for information collection. Positive coefficients therefore
indicate differences in decision making thresholds among adopters with similar

resource endowments.

Demographic Similarity as a Classification Tool

In a longitudinal purchase context, there appears to be value in
considering demographics as a proxy for channels of influence. If word of mouth
effects are expected during the process of product adoption, there is value in
placing emphasis on demographic characteristics when forming market
segments. The demographic similarity ratio used in this study was found to
differentiate between second generation versus existing products, and between 2
different kinds of existing product buyer groups.

In forming adopter categories, Rogers (1995:280) only requires groups to
be exhaustive, mutually exclusive, and classified upon a clear principle. In
determining the utility of market segments: identifiability, substantiality,
accessibility, stability, responsiveness, and actionability are often used (Wedel
and Kamakura 2000:4). Rogers” three criteria, and the first two market segment
criteria aid in conceptual distinction of groups, but do not assist in efforts to
communicate with individuals within these groups. Accessibility, stability, ‘
responsiveness and actionability reflect the managerial desire to inform, and
measure the response to potential buyers in a population. By identifying distinct
effects exerted by infectious Repeat Adopters of a second generation product,

future investigations now have an additional basis for forming market segments.

113

Together with the ability to specify the more important demographic dimensions
of the similarity ratio, consideration of New and Repeat groups can improve the

ability of researchers and managers to identify key customers.

Value to Managers

Marketing data are typically rich in demographic information that has
primarily been used to form market segments with the goal of predicting
behavioral response. The time varying effects of the demographic similarity ratio
upon adoption time in this study suggest a customer’s demographic information
can be used to predict not only their own response, but the responses of other
individuals.

For second generation products, Repeat Adopters are an identifiable
segment that can be marketed to. Their relatively small segment size enables
more efficient marketing communication compared to targeting the entire
adopting population. Demographic information can be used to both target
Repeat Adopters for repurchase, and to stimulate influence in a potential New
Adopting population. Care should be exercised in determining that a product is
perceived by the market as new. The results for the Toyota Avalon in this study
indicate some new products are not viewed as innovative by consumers and
therefore do not stimulate word of mouth effects involving Repeat Adopters.

Some existing products are amenable to word of mouth marketing, but this
is only the case for the effect of prior new buyers upon current new buyers.

Consequently it is important that marketers track in real time the demographic

114

profiles of first time buyers of existing products. If demographic similarity reveals
accelerated purchase among first time buyers, mechanisms may be implemented
to encourage these recently identified buyers to share their purchase experience.
Repeat Adopters and Buyers are loyal consumers whose main barrier to
early purchase is their decision making threshold. These individuals would be
difficult to reach through word of mouth communication but should be receptive

to resource-related firm actions such as financial incentives.

Study Limitations

The data in the investigation are only a sample of the adopting population.
This means network information is incomplete and influence can only be inferred
rather than modeled explicitly. There is potential for misattribution of effect when
data are incomplete (Greve, Strang and Tuma 2001), where infectiousness,
susceptibility, proximity and propensity are mistaken for one another.

Due to smaller sample size and thus lower statistical power, the results for
Repeat Adopters are a weaker test than for New Adopters. Conclusions drawn
regarding the difference in the susceptibility of New versus Repeat groups are
therefore less certain. Bootstrap sampling or investigation of product contexts
where new and repeat adopters are more equally distributed would improve
confidence in susceptibility findings.

The study considered data for automobiles which is a highly specific
product context. The diffusion of automobiles as a technological innovation

started more than a century ago in the United States around the year 1900 when

115

ownership penetration was at a low 1 per 1000 Americans, and reached
saturation point sometime in the 19603 (Fischer and Carroll 1988). It is during
this diffusion phase that automobiles may be reasonably described as a “new to
the world” innovative product (Garcia and Calantone 2002). Automobiles are
currently best described as a durable consumer good in a mature stage of the
product life cycle where purchase is driven primarily by replacement. Despite the
maturity of automobiles as a technological innovation, their high expense and
relationship to social status (Fredrick 1971) suggest that purchase is highly
dependent upon information concerning product performance and comparisons
with choices made by others. The assertion that new automobile introductions
generate word of mouth is supported by the incorporation of word of mouth into
forecast models of new automobile acceptance (Urban, Hauser and Roberts
1990)

The moderate innovativeness of new automobiles mean this investigation
has presented only a conservative test of the theory. Replications for products of
higher innovativeness may reasonably be expected to yield stronger results, as
word of mouth influence should play an even greater role in the adoption

decisions of individuals.

Future Research Extensions

The modeling approach used in this research identified the presence of
demographic links between past and present individuals, but did not provide

substantial detail on the nature of these links. The programming approach used

116

to calculate demographic similarity was able to determine similarity degree, but
not the specific demographic dimensions of similarity or their relative strengths.

The means of establishing similarity treated demographic dimensions as
qualitative, prevented consideration of varied levels of similarity for each
demographic dimension. For example individuals are judged dissimilar if their
incomes are not of the same category but there is no allowance for the degree of
difference between adopters along the 12 income categories. Weighting of
similarity by the level of difference for ordered demographic dimensions such as
income and education, would provide a more detailed assessment of similarity,
potentially improving the power of the demographic similarity ratio. The lack of
weighting in this study reduced the capacity to detect more subtle influence
effects. This approach was adopted to conserve computation time. To improve
understanding of influence and managerial usefulness, specific demographic
dimensions should be ranked in importance of influence using distance weights
within dimensions, and relative weights across dimensions.

While results for existing products were consistent with market
segmentation theory that distinguishes between new and repeat adopters; for 3
automobile models new buyers appear to respond to the actions of earlier new
buyers, implying that actions by members of one segment alter membership in
another. Should this latent change (Wedel and Kamakura 2000) finding be more
than a methodological artifact, this has implications for market segmentation
research that may need to entertain the possibility that latent change is an

assumption for some products rather than the exception (Bdckenholt and

117

Langeheine 1996; Farley, Lehmann and Winer 1987; Calantone and Sawyer

1978)

Conclusion

Demographic similarity as a measure of internal influence is able to
distinguish between New and Repeat Adopters of second generation products.
Repeat Adopters accelerate purchase by New Adopters and represent an
actionable market segment that is of use to managers due to their relatively small
size. Together with a method of decomposing demographic similarity influence
effects, Repeat Adopters may provide an efficient means of stimulating adoption
in a New Adopter segment. There is a promising avenue for the use of
demographic similarity to capture word of mouth effects generated by New

Adopters of existing products.

118

APPENDICES

119

APPENDIX A

Second Generation Automobiles Model Quarterly and Annual Sample Size
Distributions

120

Calendar Year

Table 50: Dodge Neon New Adopter Sample Size by Survey Quarter and

 

 

 

 

 

 

 

 

 

Quarter Total
1 2 3 4

Survey Year 1,999 0 0 55 50 105
2,000 97 115 45 62 319

2,001 121 120 51 76 368

2,002 121 125 50 41 337

2,003 137 146 63 51 397

2,004 87 97 51 54 289

Total 563 603 315 334 1 ,81 5

 

 

Table 51: Dodge Neon Repeat Adopter Sample Size by Survey Quarter and

 

 

 

 

 

 

Calendar Year
Quarter Total
1 2 3 4

Survey Year 1,999 0 0 9 2 11
2,000 14 14 4 8 40

2,001 16 12 6 12 46

2,002 26 21 10 5 62

2,003 6 13 4 6 29

2,004 9 7 2 3 21

Total 71 67 35 36 209

 

 

 

 

 

Quarter and Calendar Year

Table 52: Jeep Grand Cherokee New Adopter Sample Size by Survey

 

 

 

 

 

 

 

 

Quarter Total
1 2 3 4

Survey Year 1,999 131 113 39 108 391
2,000 35 76 32 27 1 70

2,001 90 74 27 40 231

2,002 66 82 40 70 258

2,003 88 1 1 8 1 07 140 453

2,004 87 70 29 32 218

Total 497 533 274 417 1 ,721

 

 

 

121

 

 

 

Table 53: Jeep Grand Cherokee Repeat Adopter Sample Size by Survey
Quarter and Calendar Year

 

 

 

 

 

Quarter Total
1 2 3 4

Survey Year 1,999 28 44 11 32 115
2,000 14 26 9 9 58

2,001 37 26 4 10 77

2,002 21 14 8 10 53

2,003 16 20 28 16 80

2,004 18 14 7 8 47

Total 134 144 67 85 430

 

 

 

 

 

 

Table 54: Mazda Miata New Adopter Sample Size by Survey Quarter and
Calendar Year

,

 

 

 

 

 

 

 

 

 

 

Quarter Total
1 2 3 4

Survey Year 1,999 42 56 22 31 151
2,000 4 60 28 20 1 12

2,001 99 89 49 51 288

2,002 83 1 09 49 43 284

2,003 0 1 14 66 52 232

2,004 6 28 44 40 1 18

Total 234 456 258 237 1 .185

 

Table 55: Mazda Miata Repeat Adopter Sample Size by Survey Quarter and

 

 

 

 

 

Calendar Year
Quarter Total
1 2 3 4

Survey Year 1,999 0 6 4 2 12
2,000 2 17 1 6 26

2,001 14 14 7 7 42

2,002 19 1 1 2 9 41

2,003 0 8 3 6 17

2,004 0 1 7 7 15

Total 35 57 24 37 1 53

 

 

 

 

 

 

Table 56: Toyota Avalon Legacy New Adopter Sample Size by Survey
Quarter and Calendar Year

 

 

 

 

 

Quarter Total
1 2 3 4

Survey Year 2,000 90 132 49 52 323
2,001 1 09 1 08 51 54 322

2,002 116 145 59 184 504

2,003 1 10 1 39 79 59 387

2,004 116 102 46 48 312

Total 541 626 284 397 1 ,848

 

 

 

 

 

 

122

 

 

 

 

Table 57: Toyota Avalon Repeat Adopter Sample Size by Survey Quarter
and Calendar Year

 

 

 

 

 

 

 

 

 

Quarter Total
1 2 3 4

Survey Year 2,000 20 24 6 6 56
2,001 19 12 4 7 42

2,002 20 16 3 19 58

2,003 12 20 8 8 48

2,004 23 17 6 10 56

Total 94 89 27 50 260

 

 

 

123

APPENDIX B

Complete Second Generation Automobile Results for Piecewise Weibull
Regressions

124

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

333.3 33.3 33339 333.3 33.3 383v 3
333.3 333.3 333.3 :33 333.3 333.3 3 3.3
333.3- 333- 333.3 333.3 333.3- 333.33- 333.: 333.3333- 3
333.3 3333.v 333.3 333.3 3333.v 3
333.3 333.3 :33 333.3 333.3 3 3.3
333.3. 333.3 333.3 333.3- 333.33- 3333 333.333.- 3
333.3 3333.v 333.3 333.3 3333.v 3
333.3 333.3 :33 333.3 333.3 3 3.3
333. 3. 333.3 333.3 333.3- 333.33- 333.3 333.3333- 3.
333.3 3333.V 333.3 333.3 3333.v 3
333.3 333.3 :33 333.3 333.3 3 3.3
333..- 3333 333.3 333.3- 333.33- 333.3 333.3333- 3

3333.v 333.3 333.3 3333.v 3 3333 .330

333.3 :33 333.3 333.3 3 3.3 3 98>

333.3 333.3 333.3- 33.33. 333.3333- 33 3.333393 332

53333 332 d< 3322, 33> 3:. E. Efx 385.3%... .28:

 

 

 

 

 

 

 

 

 

 

 

 

 

3232330 3 3:35.35 3530 .83 5.33353: .3925 33253035 cooz canon "33 333...

 

125

.. 3 :4... LI. I'l

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

333.3 3333.v 383v .333.v 3
333.3 333.3 333.3 333.3 333.3 333.3 3 3.3
333.3 333.3 333.3 333.3 333.3 333.3 333.3 d
. 333.3 333.3 3333.v 3333.v 3
333.3 333.3 333.3 333.3 333.3 333.3 3 3.3
333.3 333.3 333.3 333.3 333.3 333.3 333.3 3
. 333.3 3333.v 3333.v 3333.v 3
333.3 333.3 333.3 333.3 333.3 333.3 3 3.3
333.3 333.3 333.3 333.3 333.3 333.3 333.3 3
. 333.3 3333.v 3333.v .333.v 3
333.3 333.3 333.3 333.3 333.3 333.3 3 3.3
333.3 333.3 333.3 333.3 333.3 333.3 333.3 3
. 333.3 3333.v 3333.v 3333.v 3 3333 .330
333.3 333.3 333.3 333.3 333.3 333.3 3 .33 3 3.39,
23.3 333.3 333.3 333.3 333.3 333.3 333.3 3 3.3.333. 332
332.3 3.83 3 3 3 3 3 .333:

 

 

 

 

 

 

 

 

 

 

 

 

3.3383 "33 3.333

126

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

333.3 333.3 333.3 333.3 333.3 333.3 3
333.3 333.. 333.3 333.3 333.3 333.33 33 3.3
333.3 333.3 333.3 333.3 333.3 333.33- 333.33 33533. 3
333.3 333.3 333.3 333.3 333.3 3
3.33 333.3 333.3 38.3 333.33 3 3.3
333.3 333.3 333.3 333.3 333.33- 33.3 33.333. 3
333.3 333.3 333.3 333.3 333.3 3
333.3 333.3 333.3 333.3 333.33 3 3.3
33.3 333.3 333.3 333.3 333.33- 333.33 333.333- 3
333.3 333.3 333.3 333.3 333.3 4
333.3 333.3 333.3 35.3 333.33 3 3.3
333.3 333.3 333.3 333.3 333.33- 33.3 33.333- 3

333.3 333.3 32.3 333.3 3 333 .330

333.3 333.3 333.3 333.33 3 .33 3 28>

333.3 333.3 333.3 333.33- 333.333- 3 33333533333

5333.3 .532 ....< 332.? 33> 3:. E. £3333 383.3533. .333:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

3.33333 "33 3.333

127

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

333.3 333.3 333.3 333.3 3
333.3 333.3 333.3 333.3 333.3 333.3 33 3.3
333.3 333.3 333.3 333.3- 333.3 333.3- 333.3 3
. 333.3 333.3 333.3 3 33.3 3
333.3 333.3 333.3 333.3 333.3 333.3 3 3.3
333.3 333.3 333.3 333.3 333.3 333.3 333.3 3
. 333.3 333.3 333.3 333.3 3
333.3 333.3 333.3 333.3 333.3 333.3 3 3.3
333.3 333.3 333.3 333.3- 333.3- 333.3- 333.3 3
. 333.3 333.3 333.3 333.3 3
333.3 333.3 333.3 333.3 333.3 333.3 3 3.3
333.3 333.3 333.3 333.3- 333.3 333.3- 333.3 3
. 333.3 333.3 333.3 333.3 3 333 .330
333.3 333.3 333.3 333.3 333.3 333.3 3 3.3 3 3.33.,
333.3 333.3 333.3 333.3 333.3 333.3 333.3 3 233333533333
33333 3.333 3 3 3 3 3 .333:

 

 

 

 

 

 

 

 

 

 

 

 

3.33333 "33 3.333

128

 

I‘- t nil-uh ‘tii

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

333.3 333.3 68v 83.3 333 .333.v 3
32.3 3333 333.3 :33 333.3 333.3 3 3.3
333.3. 32.3. 333.3 333.3 333.3- 333.33- 333.3 33.3337 3
333.3 533v 53.3 33.3 533.v a
333.3 333.3 :33 333.3 33.3 3 .33
333.3. 333.3 333.3 333.3- 333.33- 333.3 333.3337 3
333.3 533 53.3 333.3 533 3
333.3 333.3 :33 333.3 333.3 3 3.3
3.3.3. 333.3 333.3 333.3- 333.33- 333.3 33333? 3
333.3 .333.v 53.3 33.3 633 3
333.3 333.3 :33 333.3 333.3 3 .33
333.3. 333.3 333.3 333.3- 333.33- 333.3 333.3337 3
3 3333 .330
w .Dd O mhwm>
Q 332333. 332
5333: 332 ..._< 333...; 33> 2.. E. Euax 3335.33.33. .28:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

3333.330 3 3.33:85 33.53 .8 20.33352... :33? 33.33023. c032 3300 "mm 333...

129

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.. I... .iall
. 333.3 533.v F333.v .333.v 3
333.3 333.3 333.3 333.3 333.3 333.3 3 .33
35.3 333.3 333.3 333.3 33.3 333.3 333.3 3
. 333.3 .333.v .333.v 533 3
333.3 333.3 . 333.3 333.3 333.3 333.3 3 .33
333.3 333.3 333.3 333.3 333.3 333.3 333.3 3
. 333.3 533 .333.v 533 3
333.3 333.3 333.3 333.3 333.3 333.3 3 .33
3.3.3 333.3 333.3 333.3 33.3 32.3 333.3 3
. 333.3 533 .333.v F3333 3
333.3 333.3 333.3 333.3 333.3 333.3 3 .33
33.3 333.3 333.3 333.3 33.3 .333 333.3 3
37 3333 330
F .33 3 3.33>
n 3333233. 262
333...... 3.333 3 3 3 3 3 .333...

 

 

 

 

 

 

 

 

 

 

 

 

.6338. "33 3.333

130

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

333.3 3.33 3.3.3 333.3 333.3 333.3 3
333.3 .333 333.3 333.3 3.3.3 333.33 3. .33
333.3 333.3 333.3 333.3 333.3 333.3- .333 333.333- 3
3333 33.3 333.3 333.3 333.3 3
333.3 333.3 333.3 3.3.3 333.33 3 .33
333.3 333.3 333.3 .333 3.3.3. 333.3 333.33. 3
333.3 3.3.3 333.3 333.3 333.3 a
3.3.3 333.3 333.3 3.3.3 333.33 3 .33
333.3 333.3 333.3 333.3 33.33- 333.3 333.333- 3
333.3 3.3.3 333.3 333.3 333.3 .1
333.3 333.3 333.3 3.3.3 333.33 3 .33
333.3 333.3 333.3 333.3 333.3- 333.3 333.33. 3

3 333 330

o .Ud o mumm>

a 3.9352 $333.3

5333: .532 j... 333|..3> 33> 3... .3. 3.3333 333....3..._33. .3335.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.3338. "33 3.33..

131

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

3.3.3 3333 333.3 333.3 3
333.3 333.3 .3.3 33.3 33.3 33.3 3. .33
333.3 3333 333.3 333.3 ...3 .333 33.3 3
. 3333 333.3 333.3 33.3 3
333.3 333.3 . 33.3 33.3 33.3 33.3 3 .33
333.3 333.3 333.3 333.3 33.3 333.3 3.3.3 3
. 333.3 333.3 333.3 333.3 3
333.3 333.3 . 33.3 33.3 .3.3 33.3 3 .33
333.3 3333 333.3 :33- 333.3 333.3 33.3 3 [-1
. 333.3 333.3 .333 333.3 3
333.3 333.3 33.3 33.3 .3.3 33.3 3 .33
333.3 333.3 333.3 3.3.3- 333.3 333.3 33.3 3
3 333 330
3 .33 3 3.33..
3 3.3.3033. .33.qu
33333 3.333 3 3 3 3 3 .333...

 

 

 

 

 

 

 

 

 

 

 

 

333333. "33 3.33.

132

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

030.0 033.0 .000.v .00.0 30.0 .000.v a
030.0 00.0 300.0 . .0.0 000.0 300.0 m .03
330.0- 300.0- 0300 000.0 000.0- 0 .000- 0.30.3 50.030 .- 0
000.0 .000.v .00.0 00 ..0 .000.v 3
300.0 300.0 . .0.0 000.0 .30..” 3 .33
0.. ..0. 030.0 000.0 300.0- 03 .00- 330.3 300.030 .- a
0.0.0 .000.v .00.0 5.0 .000.v 3
030.0 300.0 . .0.0 000.0 030.0 0 .03
mm .0. 030.0 000.0 000.0- 0 5.00- 000.3 300. .30 .- a
m .0.0 .000.v .00.0 00 ..0 .000.v 0
330.0 300.0 . .0.0 000.0 030.0 3 0.3
v0 .0- 030.0 000.0 000.0- 033.00- 036 000.000 .- a
a 0030 .300
. .03 0 3.38.
3 3.9003. 2.32
h<mnmm gmz j< mums? 30> 0:. E. .5. 3x too£_oxfimo_ .2322

 

 

 

 

 

 

 

 

 

 

 

 

 

 

383.3330 0 3.32.8.0 333030 30. 5.330303: :33? 33.2.3an .632 00.300 "00 033...

133

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

[I F’s-9...... E
3000 .000.v 500v .000.v 0
030.0 000.0 . 000.0 000.0 000.0 030.0 m 0.0
35.3 003.0 000.0 030.0 50.0 02.0 033.0 0
. 300.0 500.v 600.v 500v 0
030.0 000.0 000.0 030.0 030.0 030.0 3 0.0
35.3 000.0 000.0 :00 03.0 530 033.0 0 h;
. 300.0 500.v 500.v 600.v 0
030.0 000.0 000.0 000.0 000.0 030.0 0 0.0
3.0.3 0030 000.0 000.0 300.0 02.0 033.0 0
. 030.0 .000.v 500v 500v 0
030.0 000.0 000.0 030.0 000.0 030.0 3 0.0
38.3 000.0 000.0 300.0 39.0 02.0 333.0 33
9 003m .80
w 0.3 m mhmm>
3 923003. 262
2.5.» 208 0 m 3 3 3 .28:

 

 

 

 

 

 

 

 

 

 

 

 

.0..an "00 2033

134

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

000.0 033.0 333.0 000.0 330.0 003.0 0
003.0 330.0 330.0 300.0 05.0 333.33 03 0......
003.0 003.0 030.0 000.0 300.0 003.00- 003.0 030.33. 0
300.0 033.0 000.0 030.0 303.0 0
033.0 330.0 300.0 030.0 003.33 0 0.0
330.0 030.0 300.0 000.0 003.00- 030.0 000. 33. 0.
030.0 003.0 030.0 030.0 003.0 0
033.0 330.0 300.0 030.0 033.33 0 .00 1
030.0 030.0 330.0 000.0 003.00- 003.0 003.033-
05.0 003.0 300.0 030.0 303.0 0
033.0 330.0 300.0 030.0 000.33 3 0.0
300.0 030.0 000.0 300.0 303.00- 033.0 300.03. 0
0 300 .000
o .93 m mumm>
3 00000003. 000003...
5000.... >002 ._._< 0022, 00> 0:. E. 503x 0005.00.00. .0002

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.6053 "00 0.003

135

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

030.0 033.0 000.0 003.0 0
003.0 030.0 . 000.0 303.0 003.0 003.0 03 .00
300.3 030.0 000.0 300.0 300.0 300.0 333.0 0
. 300.0 003.0 030.0 303.0 0
003.0 030.0 . 003.0 303.0 033.0 003.0 0 0.0
000.3 030.0 000.0 303.0 300.0 033.0 0
000.0 000.0 300.0 033.0 0
003.0 030.0 3030 303.0 003.0 003.0 0 .00
000.3 030.0 300.0 000.0 300.0 333.0
000.0 300.0 300.0 300.0 0
003.0 030.0 303.0 303.0 003.0 003.0 3 0.0
000.3 030.0 :00- 000.0 000.0 303.0 03
3 300 .000
m 6.3 o wbmo>
00000003. 03033.
00000 0.000 0 3 0 3 .0022

 

 

 

 

 

 

 

 

 

 

 

136

 

40.0080 ”00 0.003

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

03 3.0 033.0 3000.v 000.0 _.000.v 3000.v a
033.0 003.0 «00.0 0 _.0.0 000.0 003.3 0 0.3
N000- 303.0- 0N0.0 030.0 3300- 33 F .00- 0 30.3 000.0303- 0
030.0 3000.v 000.0 _.000.v 3000.v a
30 w .0 300.0 0 5.0 000.0 003.3 3 0.3
000.0- 0300 000.0 «30.0- 03000- 030.0 000.0003- 0
000.0 _.000.v 000.0 F000.v _.000.v 0
033.0 300.0 0 5.0 000.0 333.3 0 .0.3
3 33.0- 330.0 300.0 3N0.0- 030.30- 003.3 000.0003- 0
000.0 3000.v 000.0 3000.v 3000.v 0
003.0 300.0 0 _.0.0 000.0 003.3 3 0.0
300.0- 0300 300.0 330.0- 300.00- 33 3.3 300.0303- 0
_.000.v 000.0 3000.v 3000.v 0 3003 .000
300.0 0 _.0.0 000.0 0 3313 P 0.3 3 200>
330.0 000.0 0N0.0- 030. 30- 030. 003- d 3.000003. 302
P<mamm gmz -_-_< amazo> c0> 0:. 0:. EU 3x UGO—595mg .0602

 

 

 

 

 

 

 

 

 

 

 

 

 

 

00000030 0 3:30:50 00.000 .8 02000.00: =:£0>> 00.30003 000.2000 000.5 000.. #0 030.3

137

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

03.0 500v 500v 500.v 58v 9
000.0 000.0 0.0.0 3.0.0 03.0 03.0 20.0 0 0.0
000.0 30.0 000.0 000.0- 0&0 000.0 000.0 000.0 0
. «5.0 58v 500.v F000.v 58v 0
000.0 000.0 000.0 :00 03.0 $0.0 50.0 v 0.0
000.0 300 000.0 000.0- 000.0 000.0 000.0 :00 0
. 000.0 500v 500v 500v F000.v 0
000.0 000.0 000.0 30.0 0.0.0 2.0.0 :00 0 0.0
000.0 30.0 000.0 000.0- 05.0 000.0 000.0 3000 0
. 000.0 500v 500v 500.V 500v 0
08.0 000.0 03.0 30.0 03.0 $0.0 20.0 m 0.0
0.0.0 3.00 000.0 000.0- «:00 «00.0 000 000.0 0
. 000.0 500.v 500v 500.v 58v 0- 0000 .000
000.0 000.0 03.0 30.0 03.0 $0.0 30.0 F 0.0 0 93>
000.0 3.00 000.0 0000. 0.00 000.0 000.0 000.0 0 0.0503. 202
0020 0.80 n 0 0 e 0 N .0022

 

 

 

 

 

 

 

 

 

 

 

 

 

.5280 :0 0.00»

138

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

03.0 000.0 000.0 003.0 .000.v 000.0 0
000.0 0000 0.0.0 000.0 05.0 03.00 0. 0.0
00.0. 000.0 000.0 0.00.0 0.00.0. 0000:- 000.0 000.000- 0
03.0 000.0 003.0 3000.v 300.0 0
000.0 000.0 000.0 06.0 000.00 0 0.0
003.0- 000.0 0.00 000.0. 30.03- 0000 000.000- 0
000.0 000.0 03.0 .000.v 50.0 0
000.0 05.0 000.0 05.0 000.00 0 .00
003.0- 000.0 :00 000.0- 0000:- 000.0 000.000- 0
000.0 000.0 03.0 500.v 000.0 0
03.0 05.0 000.0 030.0 000.00 0 0.0
.000- 000.0 30.0 000.0- 030:- 000.0 000.03- 0

000.0 03.0 500v 50.0 q .00 .000

000.0 000.0 000.0 000.00 0 .00 0 28>

000.0 50.0 000.0- 0000:- 000.000- 0 92000500000

5000.. .502 4.2 0.32.? 00> 2.. E. 0.0% 0005.05.00. .28:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.8053 :0 0.000

139

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.000 000.0 000.0 000.0 0...0 0

0...0 000.0 00.0 0.0 0...0 0...0 00.0 0. 0.0
.000 000.0 000.0 000.0 03.0 000.0 000.0 00.0 0

. 000.0 000.0 000.0 000.0 3.0 m
0...0 000.0 . 00.0 .0..0 0...0 0...0 00.0 0 0.0
.000 000.0 000.0 000.0 03.0 00.00 000.0 00.0 0

. .000 000.0 .000 000.0 0...0 0
0...0 000.0 00.0 .0..0 0...0 2.0 00.0 0 0.0
.000 000.0 000.0 000.0 03.0 000.0 000.0 00.0 -

. .000 000.0 .000 000.0 3.0 0
0...0 000.0 00.0 .0..0 0...0 0...0 00.0 0 0.0
.000 000.0 000.0 000.0 03.0 000.0 000.0 00.0 0

. .030 000.0 .000 000.0 3.0 0 .00 .000
0...0 000.0 00.0 .0..0 0...0 2.0 00.0 0 0.0 0 0.00..
.000 000.0 000.0 000.0 03.0 000.0 .000 00.0 0 92000500000
000.... 0.80 0 0 0 0 0 0 .000...

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.6008. ".0 0.000

140

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

..0.0 000.0 .000.v .000.v .000.v .000.v 0
000.0 00.0 000.0 0.0.0 000.0 000.0 0 0.0
000.0- 0000 000.0 000.0 .000- 000.00- 000.0 00.0000- 0
000.0 .000.v .000.v .000.v .000.v 0
000.0 000.0 0.0.0 0000 000.0 0 0.0
000.0. 000.0 000.0 .000- 00000- 000.0 000.0000- 0
000.0 .000.v 000.0 .000.v .000.v 0
00.0 0000 0.0.0 000.0 000.0 0 0.0
3.0- 000.0 000.0 000.0- 00000- 00... .00..000- 0
000.0 .000.v 000.0 .000.v .000.v 0
00.0 000.0 0.0.0 000.0 000.0 0 0.0
000.0. 000.0 000.0 000.0. 000.00- 0000 000.0000- 0
0 0000 .000
. 0.0 3 0.00>
3 0.0.0000. 2.02
5000: .502 4.2 00000.. 00> 0... E. Euux 0000.00.30. .000...

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0000.00.60 0 3.8.5.10- 000000 .2 20.000500 .3903 00.3000... 09.9.0.5 0:000 000... .30 0.00...

141

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.000 .000.v .000.v .000.v .000.v 0
000.0 000.0 .30 3.0 03.0 03.0 03.0 0 0.0
0.0.0 300 000.0 000.0- .000 000.0 000.0 000.0 0
. 0000 .000.v .000.v .000.v .000.v a
000.0 0000 03.0 03.0 03.0 03.0 03.0 v 0.0
0.0.0 300 000.0 .000- .000 000.0 000.0 000.0 0
. 0000 .000.v .000.v .000.v .000.v 0
000.0 000.0 03.0 03.0 03.0 03.0 03.0 0 0.0
000.0 300 000.0 0 .00- 0 .00 000.0 000.0 000.0 0
. 000.0 .000.v .000.v .000.v .000.v 9
000.0 000.0 03.0 330 03.0 03.0 03.0 0 0.0
0.0.0 300 000.0 0.00- 0.00 000.0 .000 000.0 0
0 0000 .000
. 0.0 0 0.00>
0 0.0.000... 2.02
000:0 0.000 0 0 0 v 0 0 .0005.

 

 

 

 

 

 

 

 

 

 

 

 

 

00.000. .00 0.000

142

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0000 .000 000.0 .0..0 .000.v .000 0
000.0 0000 0.0.0 0000 0.00 000.00 0. 0.0
000.0 0..0- 000.0 000.0 0000- 0000..- 0000 000.000- 0
000.0 000.0 00.0 .000.v .000 0
00.0 0.0.0 0000 0.00 000.00 0 0.0
..0.0- 0000 .000 0000- .000..- 0000 000.000- 0
0.00 000.0 00.0 .000.v .000 0
0000 0.0.0 0000 0.0.0 0.0.00 0 0.0
000.0- 0000 .000 000.0. 000.0..- 0000 000.000- 0
000.0 000.0 00.0 .000.v .000 0
.000 0.0.0 0000 0.00 000.00 0 0.0
000.0- 0000 .000 0000. 000.0..- .000 000.000- 0

0 .00 .000

m .U.m 0 m.mm>

0 0.0.0000. .0000...

5.000.. .502 0-_< 00000.. 00> 00. .0. .00... 0000000000. .000...

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.0008. .00 0.000

143

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0000 0000 .000 000.0 0. .0 0
0. .0 0000 . 00.0 00.0 0. .0 . . .0 3.0 0. 0.0
.000 000.0 000.0 000.0 000.0 000.0 000.0 00 .0 a
. 000.0 000.0 .000 000.0 0...0 0
0...0 0000 00.0 .0.0 0...0 0...0 3.0 0 0.0
.000 000.0 000.0 000.0 000.0 0000 0000 00 .0 0
. 0000 000.0 .000 000.0 0. .0 0
0...0 000.0 00.0 .03 0...0 0...0 3.0 0 0.0
.000 000.0 000.0 0000 000.0 000.0 000.0 00 .0 a
. 000.0 000.0 .000 0000 0. .0 0
0...0 0000 00.0 .0..0 0...0 0...0 3.0 0 0.0
.000 0000 000.0 000.0 000.0 0000 000.0 00.0 a
0 .00 .000
0 0.0 0 0.00..
0 00.002 .0000...
00$? 0.000 0 0 m 0 0 0 .0005.

 

 

 

 

 

 

 

 

 

 

 

 

 

.0008. .00 0.00.

144

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0000 000.0 .000.v 000.0 .000.v .000.v 0
000.0 .0.0 0000 0.0.0 0000 000.0 0 0.0
000.0- 000.0- 000.0 000.0 000.0- 00000- 0 .0... 000.0000- 0
000.0 .000.v 000.0 .000.v .000.v a
000.0 0000 0.0.0 0000 000.0 0 0.0
00.0- 0000 000.0 0000- 00000- 000.0 000.0000- 0
0000 .000.v 000.0 .000.v .000.v a
000.0 0000 0 .00 0000 000.0 0 0.0
00.0- 000.0 000.0 000.0- 000.00- 000.0 000.0000- 0
0 .00 .000.v 000.0 .000.v .000.v 0
.000 0000 0.0.0 0000 000.0 0 0.0
00.0- 0000 000.0 000.0. 000.00- 000.0 000.0000- 0

0 0000 .000

. 0.0 0 0.00>

0 0.0.0000. 2.02

.5000: .502 .13 00000> 00> 0... E. .0... 0x 0005.00.30. .0005.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0000.550 0 00.0.2.5 00.000 .0. 5.000.000 .3902. 00.2.0030 0000.000 000.6 000.. .00 0.00...

145

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0.0.0 .000.v .000.v .000.v .000.v 0
000.0 000.0 :00 30.0 03.0 03.0 03.0 0 .00
0.0.0 30.0 0000 0.0.0- .000 00.0 000.0 030.0 0
. 000.0 .000.v .000.v .000.v .000.v 0
000.0 000.0 03.0 30.0 03.0 2.0.0 030 v .00
000.0 30.0 000.0 000.0- .000 000.0 000.0 03.0 0
. 000.0 .000.v .000.v .000.v .000.v 0
000.0 000.0 000.0 30.0 03.0 03.0 03.0 0 .0.»
0.0.0 :00 0000 0.0.0- 0.0.0 000.0 000.0 30.0 0
. 000.0 .000.v .000.v .000.v .000.V a
000.0 000.0 030 3.0.0 03.0 03.0 03.0 0 .00
20.0 30.0 000.0 0.00. 0.0.0 0000 000.0 03.0 0
0 0000 .20
F 6.0 0 900>
0 0.03004. 302
2:2.» 0.8» .- 0 0 e 0 a .28:

 

 

 

 

 

 

 

 

 

 

 

 

 

.8008. ”00 0.000

146

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

000.0 000.0 000.0 .000 .000.v 000.0 0
000.0 0000 0.0.0 0000 0.0.0 000.00 0. 0.0
0.00 000.0 000.0 000.0 000.0- 000.0:- 0000 000.000- 0
000.0 000.0 000.0 .000.v 000.0 0
0000 0.0.0 000.0 0.00 000.00 0 0.0
000.0 000.0 000.0 000.0- 000.0:- 0000 000.000- 0
000.0 000.0 .000 .000.v 000.0 0
020 0.0.0 000.0 0.0.0 0.0.00 0 0.0
000.0 000.0 .000 000.0. 0000:- 000.0 000.000- 0
000.0 000.0 000.0 .000.v 000.0 0
020 0.0.0 0000 0.00 000.00 0 0.0
000.0 000.0 .000 000.0. 000.0:- 0000 000.000- 0

0 .00 .000

m .00 0. m.mm>

0 0.0.003. 000000

5000.. .502 ._._< 0022. 00> 2.. E. 0.3x 32:00.02 .03:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.6008. "00 2000

147

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.. a... hilfl-.-...E-:-.i
000.0 000.0 50.0 0000 0:0 0
«:0 000.0 000.0 50 0:0 03.0 000.0 0. 0.0
000.0 000.0 000.0 000.0 000.0 000.0 000.0 000.0 0
. 000.0 000.0 000.0 0000 0:0 0
0:0 000.0 000.0 .000 0:0 :00 000.0 0 0.0
000.0 000.0 000.0 000.0 000.0 000.0 000.0 000.0 0
. 000.0 000.0 000.0 0000 0:0 0
«:0 000.0 000.0 000.0 0:0 02.0 08.0 0 0.0
000.0 000.0 000.0 000.0 000.0 000.0 000.0 000.0 0
. .000 000.0 000.0 000.0 0:0 0
0:0 000.0 000.0 000.0 0:0 2.0 000.0 0 .00
000.0 000.0 000.0 000.0 000.0 000.0 000.0 000.0 0
0 000 .000
m 6.0 0 0.mm>
0.0.0004. 000000..
32.0 0.80 0 0 0 0 0 0 .0022

 

 

 

 

 

 

 

 

 

 

 

 

 

.0..an "00 0.000

148

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

000.0 000.0 .000.v 000.0 000.0 .000.v 0
00.0 000.0 0000 0.0.0 0000 000.0 0 0.0
000.0- 0000- 0.0.0 000.0 000.0- 000.00- «00.00 000.000.- 0
.000.v .000.v 000.0 000.0 .000.v 0
00.0 0000 0.0.0 000.0 000.0 0 0.0
000.0- 0.00 000.0 000.0- 000.00- 0000. 000.000.- 0
000.0 .000.v 000.0 000.0 .000.v 0
000.0 0000 0.0.0 0000 000.0 0 0.0
.000. 0.0.0 000.0 000.0- 000.00- 000.0 0.0000.- 0
000.0 .000.v 000.0 .000 .000.v 0
0.0.0 000.0 0.0.0 000.0 000.0 0 0.0
«00.0- 0.0.0 000.0 000.0- 0.0.00- 000.0 0.0000.- 0

.000.v 0.0.0 0.0.0 .000.v 0 0000 .000

0000 0.00 000.0 000.0 . .00 0 0.00..

0.0.0 .000 000.0- .0000- 000.000.- 0 0.0.009. 202

0.0000... .502 ._..< 03.2. 00> 2.. E. 0.3x 0005.00.02 .28....

 

 

 

 

 

 

 

 

 

 

 

 

 

 

00.00030 0 03.0.3.0 00.000 .0. 02000.00: .3503 003.000.... 0.0:... 0000.2 .00 0.00...

149

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0000 .000.v 0000 .000.v .000.v 0
.000 ..00 0000 0000 0000 0000 0000 0 0.0
000.0 .000 0000 0000 0.00 00.0 0000 0000 0
. 0000 .000.v 0000 .000.v .000.v 0
.000 ..00 0000 0000 0000 0000 0000 0 0.0
000.0 .000 0000 0000 0.00 00.0 0000 0000 0
. 0000 .000.v 0000 .000.v .000.v 0
.000 :00 0000 0000 0000 0000 0000 0 0.0
.000 .000 0000 .000 0.00 00.0 0000 0000 0
. 0000 .000.v 0000 .000.v .000.v 0
.000 ..00 0000 0000 0000 0000 0000 0 0.0
.000 .000 0000 0000 0.0.0 00.0 0000 0000 0
. 0000 .000.v 0000 .000.v .000.v q 0000 .000
.000 :00 0000 0000 0000 0000 0000 . 0.0 0 0.00..
000.0 0000 0000 0000 0.00 00.0 0000 0000 0 0.0.0090 202
00000 0.000 0 0 0 0 0 0 .000...

 

 

 

 

 

 

 

 

 

 

 

 

 

.8008. "00 0.000

150

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0000 0000 0000 .000.v 0000 0.00 0
0000 0000 0.00 0000 0.00 .0000 0. 0.0
0000- 00.0- 0000- 00.... 0000- 00...0 000.. .00.0.0- 0
00.0 0000 .000.v .000 0.00 0
0000 0.00 0000 0.00 000.00 0 0.0
0000- 0000. 00.0 0000- 00000 000.. 0000.0- .fi
0000 0000 .000.v 0000 .000 0
0000 0.00 0000 0.00 000.00 0 0.0
0000- 0000. 00.0 0000- 00000 0000 0.0.0.0- 0
0000 0000 .000.v 0000 0000 0
0000 0.00 0000 0.00 000.00 0 0.0
0000- 0000. 00.0 0000- 00000 00.0 0000.0- 0

0000 .000.v 0000 0000 0 000 .000

0.00 0000 0.00 000.00 0 0.0 0 0.00..

0000- 00.0 0000- 000.00 0000.0- 0 0.0.003 .0800

5000.... .502 ._-.< 0020.. 00> 0:. .0. 0.00.. .80....0..._0o_ .000...

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.0008. .00 0.000

151

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0000 00.0 0000 0000 0000 0
0.00 0000 00.0 00.0 00.0 00.0 00.0 0. 0.0
.000 0000 0000 00.0 0.00 0000 .0.0 .000 0
. .000 00.0 0000 0000 0000 0
0.00 0000 00.0 00.0 00.0 00.0 00.0 0 0.0
000.0 0000 0000 00.0 0000 0000 00.0 0000 0
. 00.0 0000 0.00 0000 0000 0
0.00 0000 00.0 00.0 00.0 00.0 00.0 0 0.0
000.0 0000 0000 0000 0000 0000 00.0 0000 0
. 00.0 00.0 0000 0000 0000 0
0.00 0000 00.0 00.0 00.0 .0.0 00.0 0 0.0
000.0 0000 0000 00.0 0000 .000 00.0 0000 0
. 00.0 00.0 0000 0000 0000 9 000 .000
0.00 0000 00.0 00.0 00.0 00.0 00.0 0 0.0 0 0.00..
000.0 0000 0000 00.0 0000 0000 00.0 0000 0 90.08.0080,.
00000 0.000 0 0 0 0 0 0 .000...

 

 

 

 

 

 

 

 

 

 

 

 

 

.8008. "00 0.000

152

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0. . -2..-

0000 0000 .000.v 0.00 0000 .000.v 0
0000 00.0 080 0.00 0000 0.0.0 0 .00
.0.0- 0.00 0.00 0000 0000- 000.00- 00.0 000.000.- 0
0.00 .000.v 0.00 0000 .000.v 0
0000 0000 0.00 0000 000.0 0 .00
00.0. 0.00 0000 0000- 000.00- 00.0 000.000.- 0
00.0 .000.v :00 0000 .000.v 0
0...0 0000 0.00 0000 000.0 0 .00
00.0- 0.00 0000 0000- 20.00. 000.0 .00.000.- 0
00.0 .000.v :00 .000 .000.v 0
0...0 0000 0.00 0000 000.0 0 0.0
00.0- 0.00 0000 0000- 0.0.00- 000.0 000.000.- 0

0 0000 .000

. .00 0 0.00>

0 0.0303 2.02

50000 .502 ._..< 00E? 00> 00. .0. 0.00.. 080000.00. .000...

 

 

 

 

 

 

 

 

 

 

 

 

 

 

030.030 0 3.5.8.0 00.000. .0. 02000.00... ..:0.0>> 00.30005 0.0.5. 030.2 "mm 0.00...

153

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

000.0 .000.v 000.0 .000.v .000.v 0
.000 :00 000.0 000.0 000.0 000.0 000.0 0 0.0
000.0 000.0 000.0 000.0 0.0.0 000.0 000.0 000.0 0
. 000.0 500v 000.0 .000.v .000.v 0
.000 :00 000.0 000.0 000.0 000.0 000.0 0 0.0
000.0 000.0 000.0 000.0 00.0 00.0 000.0 000.0 0
. 000.0 .000.v 000.0 800v .000.v 0
000.0 :00 000.0 000.0 000.0 000.0 000.0 0 0.0
000.0 000.0 000.0 000.0 000.0 000.0 000.0 000.0 0
. 000.0 .000.v 000.0 .000.v .000.v 0
.000 :00 . 000.0 000.0 000.0 000.0 000.0 0 0.0
000.0 000.0 000.0 000.0 .000 000.0 000.0 000.0 0
0 0000 .000
_. .Ud N w..mm>
a 0.000004. 2.02
002.0 0.000 0 0 0 0 0 0 .000:

 

 

 

 

 

 

 

 

 

 

 

 

 

.6008. .00 0300

154

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

000.0 000.0 000.0 000.0 000.0 000.0 0
000.0 0000 0.0.0 0000 0.0.0 000.00 0. 0.0
000.0- 000.0 000.0- 00.0 000.0- 000.00 .000 0000.0- 0
000.0 000.0 .000.v 000.0 000.0 0
00.0 0.0.0 0000 0.0.0 000.00 0 0.0
000.0- 000.0- 00.0 000.0- 000.00 0000 .00.0.0- 0
000.0 000.0 000.0 000.0 000.0 0
0000 0.0.0 0000 0.0.0 000.00 0 0.0
0.00. 000.0- 00.0 000.0- 000.00 000.0 0000.0- 0
000.0 000.0 000.0 000.0 000.0 0
000.0 0.0.0 0000 0.0.0 ..0.00 0 0.0
000.0 000.0. 00.0 000.0- 000.00 000.0 0000.0- 0

0 000 .000

0 .U.m N 9mm>

.0 0.0.0004. 00000.0

2000.. 2.02 0...0 00000.. 00> 00. .0. .00... 000009.000. .000:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

000000. "00 0.00..

155

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.0..0 0...0 000.0 000.0 000.0 0
0.0.0 0000 00.0 00.0 00.0 00.0 00.0 0. 0.0
.000 000.0 000.0 00.0 000.0 000.0 00.0 000.0 0
. 00.0 ....0 000.0 000.0 000.0 0
0.0.0 0000 00.0 00.0 00.0 00.0 .0..0 0 0.0
000.0 000.0 000.0 00.0 000.0 000.0 00.0 000.0 0
. 00.0 .0..0 000.0 000.0 000.0 0
0.0.0 0000 00.0 00.0 00.0 .0..0 00.0 0 0.0
.000 000.0 000.0 00.0 000.0 000.0 00.0 000.0 a
. 00.0 00.0 .000 000.0 000.0 0
0.0.0 0000 00.0 00.0 00.0 00.0 00.0 0 0.0
000.0 000.0 000.0 00.0 000.0 000.0 00.0 000.0 0
0 000 .000
0 .U.m h m.mm>
0 0.0.0000. .0000”.
00000 0.000 0 0 0 0 0 0 .000:

 

 

 

 

 

 

 

 

 

 

 

 

 

0.00000. .00 0.000

156

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

[Ill h _ . .—
00 . .o 0 .0..0 .000.v 0 5.0 000.0 .000.v 0
.50 55 000.0 05.0 000.0 000.0 m 0.0
000.0 mm .d- 0 5.0 50.0 0000- 0.00.00- on .0 000.000 .-
0000 .000.v 05.0 000.0 .000.v 0
000.0 000.0 N 5.0 000.0 000.0 0 0.0
5.0 0 5.0 50.0 50.0- 0.00.00- 0 5.0 000.50 .- a
000.0 .000.v 05.0 000.0 .000.v 0
000.0 000.0 m 5.0 moo-o 000.0 m 0.0
05.0- 05.0 50.0 000.0. 000.00- .00-o 000.50.- 0
«00.0 .000.v m 5.0 000.0 .000.v 0
000.0 000.0 m 5.0 000.0 00...0 N 0.0
000.0. 0 5.0 50.0 ooo.o- 000.00. 00 .0 000.50. .- 0
0 0000 .000
. 0.0 n 0.00>
0 0.0.0000. 2.02
h<m0mm Bmz .13 0mg? 00> 0:. 2.. Eu 0x 00053530. .0022

 

 

 

 

 

 

 

 

 

 

 

 

 

 

00.0..050 0 320.55 00.00.. .0. 00.00050... .3903 00.2.0020 0.0.5. 00005. .00 0.0.0.0

157

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

000.0 .000.v 0.0.0 .000.v .000.v 0
50.0 :00 000.0 000.0 000.0 000.0 000.0 0 .00
000.0 «00.0 000.0 000.0 0.0.0 .90 50.0 000.0 0
. 20.0 .000.v 000.0 .000.v .000.v 0
50.0 :00 :00 000.0 000.0 000.0 000.0 0 .00
000.0 000.0 000.0 .000 0.0.0 00...0 000.0 000.0 0
. 000.0 .000.v 000.0 .000.v .000.v 0
50.0 :00 000.0 000.0 000.0 000.0 000.0 m .0..0.
000.0 «00.0 000.0 000.0 0.00 00...0 000.0 000.0 0
. 000.0 .000.v 000.0 .000.v .000.v 0
50.0 :00 50.0 000.0 000.0 000.0 000.0 m .0.»
000.0 000.0 000.0 000.0 20.0 09.0 000.0 000.0 0
0 0000 .20
_. .00 N. mhmm>
a 0.0.002 302
8.2.0 0.8... N. 0 0 0 0 u .000:

 

 

 

 

 

 

 

 

 

 

 

 

 

.0008. ”8 20¢

158

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

000.0 000.0 000.0 .000.v 000.0 000.0 0
.0..0 000.0 20.0 000.0 30.0 000.00 0. .00
00.0 00.0- 0000- .0..0 000.0- 000.00 000.0 0000.0- 0
000.0 000.0 000.0 000.0 000.0 0
030 0.0.0 000.0 20.0 000.00 0 .00
000.0 000.0- 00.... 000.0- 000.00 00.0 0000.0- 0
000.0 000.0 .000.v 000.0 000.0 0
20.0 20.0 000.0 30.0 000.00 0 .00
000.0- 000.0- 00.0 000.0- 000.00 000.0 {0.0.0- 0
000.0 000.0 000.0 .000 000.0 0
0.0.0 20.0 000.0 30.0 000.00 0 .00
000.0- 000.0- 09.0 000.0- 000.00 30.0 0000.0- 0

0 000 .80

m .90 - w.mm>

0 0.0.000< .0000...

5.0000 .502 -_._< 002.2, 00> 2.. .c. 50.x 000....9.._0o_ .28....

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.0008. "00 0...0.

159

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

. 02.0 000.0 000.0 R00 000.0 0
20.0 000.0 00.0 00.0 00.0 .20 02.0 2 .0...
000.0 000.0 000.0 .20 000.0 .000 00.0 000.0 0
. 02.0 .0..0 000.0 000.0 000.0 0
20.0 000.0 . 00.0 00.0 02.0 .20 02.0 0 0.0
000.0 000.0 000.0 02.0 000.0 000.0 00.0 000.0 0
. 02.0 02.0 .0..0 .000 000.0 0
20.0 000.0 . 00.0 03.0 00.0 .20 02.0 0 .00
02.0 000.0 000.0 02.0 000.0 000.0 .0..0 000.0 0
. .20 02.0 000.0 000.0 000.0 a
20.0 000.0 00.0 02.0 02.0 02.0 02.0 0 .00
000.0 000.0 000.0 02.0 000.0 000.0 00.0 000.0 0
0 000 .80
o 6.0 N m..mm>
0 0.0.0004. .0000...
002.0 0.80 .- 0 0 0 0 0 .000...

 

 

 

 

 

 

 

 

 

 

 

 

 

.6008. "8 0.00»

160

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0000 000.0 2000.v 000.0 000.0 500v 0
200 000.0 000.0 20.0 000.0 02.0 0 0.0
:00 0000. 20.0 0.0.0- 050- 20.00- 000.02 000.22- 0
20.0 2000.v 000.0 000.0 2000.v 0
02.0 000.0 20.0 000.0 02.0 0 0.0
000.0 20.0 000.0- 0.00. 2:00- 000.0 200.02. a
000.0 2000.v 000.0 20.0 2000.v m
0200 000.0 200 0000 E0 0 0.0
.000- 200 20.0- 20.0- 00000- 0000 000.22- 0
000.0 2000.v 000.0 :00 2000.v 4
000.0 0000 20.0 0000 02.0 0 0.0
02.0 20.0 20.0- 0.00. 020.00- 0000 000.022- 4

2000.v 0000 20.0 2000.v 0 0000 .000

000.0 20.0 000.0 02.0 F 0.0 0 0.08,

20.0 020.0- 020.0. 000.00- 000.22- 0 20.003 202

50000 .502 0.2 000..? 00> 2.. E. Evux 085.9... 0. .28....

 

 

 

 

 

 

 

 

 

 

 

 

 

 

020.330 0 322.85 00500 .2 5.00050: :36? 00.30030. co.0>< 08.3... .00 0.0.0...

161

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0000 000.0 20009 2000.0. 0
000.0 000.0 000.0 0000 0000 200.0 0 0.0
000.0 000.0 0000 0000 0000 02.0 002.0 0
. 0000 000.0 .000.v 2000.v 0
000.0 0000 000.0 0000 200.0 200.0 0 0.0
000.0 000.0 000.0 0000 000.0 0.00 02.0 01
. 000.0 002.0 2000.v 2000.v 0
0000 20.0 0000 0000 200.0 2000 0 0.0
000.0 0000 000.0 000.0 2000 020.0 02.0 9
. 0000 02.0 2000.v 2000.v 0
000.0 20.0 000.0 000.0 200.0 200.0 0 0.0
000.0 0000 0000 0000 0000 020.0 02.0 9
. 000.0 0020 .000.v 600v 0 0000 .000
000.0 20.0 0000 000.0 2000 2000 2 0.0 0 0.00>
000.0 000.0 0000 000.0 0000 020.0 02.0 0 0.0.003. 302
002.0 0.000 0 0 0 0 0 .000...

 

 

 

 

 

 

 

 

 

 

 

 

.928. "00 0.000

162

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

000.0 000.0 .000.v 0000 0000 .000.v a
0000 ..0.0 000.0 .000 0.0.0 0.0.00 2 0.0
20.. .000 0000 000.0- 0000 000.00.- 000.0 20.000- 0
0000 .000.v .000 .000 .000.v 0
0000 000.0 .000 0.0.0 02.00 0 0.0
00... 000.0 000.0- 0000 .00.00.- 000.0 000.000- 0
000.0 .000.v 0000 .000 000.0 0
000.0 0000 .000 0.0.0 000.00 0 0.0
.0... 000.0 000.0- 0000 000.00.- 000.0 000.000- 0
000.0 .000.v 0000 .000 .000.v 0
000.0 000.0 .000 0.0.0 000.00 0 0.0
.00.. 000.0 000.0- 0000 00000.- 000.0 00000. 0

000.0 0000 02.0 0000 0 000 .000

.000 0000 0.00 000.00 0 0.0 0 0.00..

0000 0.0.0- 0.0.0 00002. 000.000- 0 000.002.0000...

5000.. .502 ....< 00000.. 00> 0:. .0. £000. .82...0.....§ .0..0:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.0008. .00 0.000

163

 

udJu-1ll‘ifi..b . .J

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

000.0 02.0 0000 .000 0
20.0 000.0 02.0 02.0 0...0 0...0 2 0.0
000.0 .000 000.0 0000 00.0 000.0 0000 0
. 0000 0...0 0000 000.0 0
0.0.0 0000 00.0 02.0 2.0 02.0 0 0.0
000.0 000.0 000.0 0000 00.0 .000 20.0 0
. .000 02.0 20.0 0000 0
20.0 000.0 .20 00.0 2.0 02.0 0 0.0
.000 000.0 000.0 20.0 00.0 0000 000.0 a
. 000.0 020 0.0.0 0000 0
0.0.0 0000 .20 02.0 2.0 02.0 0 0.0
.000 0000 0000 20.0 00.0 0000 000.0 0
. 0000 00.0 20.0 000.0 0 000 .000
0.00 000.0 02.0 02.0 2.0 02.0 0 0.0 0 0.00.0
000.0 000.0 000.0 .000 02.0 0000 02.0 0 0.008500%:
0030 0.000 0 0 0 0 0 .0005.

 

 

 

 

 

 

 

 

 

 

 

 

.0008. .00 0.000

164

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2 .0 00 .0 .000.v .000 000.0 .000.v 0
. . .0 00.0 0000 0 .00 000.0 .0..0 0 0.0
00.0 000.0- 0.0.0 0.0.0- 0.0.0. 000.00- 000.0 0000.2- 0
000.0 .000.v 0000 0.0.0 .000.v 0
000.0 0000 0.0.0 0000 00.0 0 0.0
000.0 200 0.0.0- 0.00. .00.00- 0000 000.0.2- 0
000.0 .000.v 0.0.0 0000 .000.v 0
0000 0000 20.0 000.0 .20 0 0.0
0000- 20.0 0.00. 0.0.0. 000.00- .000 0000.2. 0
0000 .000.v 000.0 20.0 .000.v 2
000.0 0000 0.0.0 0000 00.0 0 0.0
000.0- 20.0 20.0- 0.00- 00000- 000.0 .0022- 0

0 0000 .000

P 6.0 @ 0..wo>

a 0.0.003. >>0z

0.00000. 2.02 ..._< 0.9002 00> 00. .0. £00.. 080000.000. .000...

 

 

 

 

 

 

 

 

 

 

 

 

 

 

00.0.0030 0 3:00.500 00.000 .0. 5.000.000. .3903 00.3000... co.0>< 0.0.3... .00 0.00...

165

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.000 0000 .000.v .000.v 0
000.0 0000 0000 0000 0000 000.0 0 0.0
000.0 000.0 000.0 000.0 0000 0000 02.0 0
. .000 02.0 .000.v .000.v 0
000.0 20.0 0000 000.0 .000 .000 0 0.0
000.0 000.0 0000 000.0 0000 0.0.0 02.0 0
. 000.0 02.0 .000.v .000.v 0
0000 0.0.0 0000 0000 0000 .000 0 0.0
000.0 000.0 000.0 000.0 000.0 0.00 02.0 0
. 000.0 00.0 .000.v .000.v 0
0000 0.0.0 0000 0000 .000 .000 0 0.0
000.0 000.0 000.0 000.0 .000 0.0.0 02.0 0
0 0000 .000
. 0.0 0 0.00>
a 0.0.82 302
00000 0.000 0 0 0 0 0 .0002

 

 

 

 

 

 

 

 

 

 

 

 

00.0.8. 00 0.000

166

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0000 .000 0000 000.0 00.0 0000 0
000.0 000.0 0000 .000 0.00 000.00 2 0.0
0000 02.0 0000 0000- .000 000.00.- 0000 20.000. 0
0000 0000 000.0 .0..0 0000 0
0.0.0 0000 .000 0.00 000.00 0 0.0
.000 0000 000.0- 200 000.00.- 000.0 000.000- 0
02.0 0000 0000 0000 000.0 0
0000 0000 .000 0.0.0 000.00 0 0.0
000.0 .000 000.0- 000.0 000.00.- 20.. 000.000- 0
0000 0000 000.0 000.0 000.0 0
20.0 0000 .000 0.0.0 .300 0 0.0
0000 000.0 000.? 000.0 000.00.- 000.0 000.000- .0

0 000 .000

0 6.0 O 0.mm>

0 0.0.08... .0000...

0.000.. .502 .3... 00002 00> 00. .0. .00.... 080._0..__a. .0022

 

 

 

 

 

 

 

 

 

 

 

 

 

 

00008. .00 0.000

167

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Ill-lat.

000.0 00.0 20.0 000.0 0
0.0.0 0000 02.0 02.0 2.0 02.0 2 0.0
0.0.0 0000 000.0 0000 00.0 000.0 02.0 0
. 000.0 02.0 000.0 000.0 0
0.0.0 0000 02.0 02.0 2.0 02.0 0 0.0
000.0 000.0 000.0 000.0 .0..0 0000 02.0 0
. 0000 00.0 0.0.0 0000 0
0.0.0 000. 02.0 02.0 2.0 02.0 0 0.0
0.0.0 0000 0000 0000 02.0 000.0 02.0 0
. 000.0 00.0 0.0.0 0000 0
0.0.0 0000 .0..0 02.0 0...0 02.0 0 0.0
0.0.0 0000 0000 0000 02.0 0000 02.0 0

0 000 .000

0 0.0 0 0.00..

a 0.0.003. .0000...

00000 0.000 0 0 0 0 0 .03....

 

 

 

 

 

 

 

 

 

 

 

 

.0008. .00 0.000

168

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

000.0 000.0 .000.v 000.0 0.00 .000.v 0
000.0 0. .0 000.0 0 .00 0000 00 .0 0 0.0
2.0- 0...0 20.0 200- ..00- .0000- 000.. 0000.2- 0
000.0 .000.v 0000 0.0.0 .000.v 0
000.0 0000 0.0.0 0000 .0..0 0 0.0
000.0- 200 0.0.0- 0.0.0. 00000- .000 0000.2. 0
000.0 .000.v 0000 0.0.0 .000.v 0
.000 000.0 0.00 000.0 02.0 0 0.0
000.0 0.0.0 0.0.0- 0.0.0. 00000. 0000 000.22- 0
0000 .000.v 000.0 20.0 .000.v 0
000.0 0000 0.0.0 0000 02.0 0 0.0
.000- 0.0.0 0.0.0- 0.0.0. 0.0.00- 000.0 000.22- 0

0 0000 .000

. 0.0 0 0.00..

a 0.0.003. 2.02

0.00000 .502 0.00 0 000> 00> 00. .0. .00... 080.00%... .002...

 

 

 

 

 

 

 

 

 

 

 

 

 

 

020.060 0 3_.0..E.0 00.00.. .0. 02000.00... ..:0.0>> 00.3000... :o_0>< 0.2.0... .00 0.00...

169

 

1 .. 1! 1...... r0 .

...

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

000.0 0000 .000.v .000.v 0
000.0 0000 000.0 0000 000.0 .000 0 0.0
000.0 000.0 0000 000.0 0000 000.0 000.0 0
. 000.0 00.0 .000.v .000.v 0
000.0 20.0 . 0000 000.0 000.0 .000 0 0.0
000.0 0000 000.0 000.0 0000 0.0.0 02.0 0
. 000.0 .0..0 .000.v .000.v 0
0000 0.0.0 0000 000.0 .000 .000 0 0.0
000.0 000.0 000.0 0000 0000 0.0.0 02.0 0
. 000.0 .0..0 .000.v .000.v 0
0000 0.0.0 0000 000.0 .000 .000 0 0.0
000.0 0000 000.0 000.0 0000 0.0.0 02.0 0
0 0000 .000
w 6.0 0.mo>
0 0.0.003. 2.02
00000 0.000 0 0 0 0 0 .000...

 

 

 

 

 

 

 

 

 

 

 

 

.0008. "00 0.000

170

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0.0.0.0 0 F 5 000.0 000.0 000.0 000.0 0
000.0 20.0 000.0 05.0 v5.0 50.9 or 6.0
005- 03.0 000.0 000.0- 000.0 000.00? 05.0 50.000- 0
000.0 000.0 0 5.0 mm 5 000.0 0
«.0 F .0 000.0 05.0 05.0 5 v.00 - 0 6.0
Ed 50.0 05.0- 05.0 {0.007 000.0 000.000- 0
05.0 000.0 000.0 Rod 000.0 0
000.0 000.0 05.0 05.0 000.00 0 6.0
000.0 50.0 000.0- v8.0 000.00? v5.0 000.000- 0
000.0 000.0 50.0 005 000.0 0
00.0.0 000.0 05.0 v 5.0 00 0.0..‘ n .0..0
— 5.0 50.0 000.0- 000.0 F 5.00 T 005.0 000.000- 0

0 00v .000

0 6.0 0 0.08»

0 0.2002 «080$

._.<mn_mm 252 ._._< onus? 30> 2: E. Eu «x 0005.9... 0. .000—z

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.5280 50 2.5

171

 

 

W1.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

000.0 03.0 000.0 000.0 0
30.0 000.0 000.0 0000 0:0 E0 0_ 0.0
000.0 0.00.0 0000 060. $00 0000 000.0 0
. 000.0 000.0 30.0 000.0 0
08.0 000.0 000.0 000.0 0:0 0:0 0 0.0
«00.0 03.0 000.0 000.0 09.0 000.0 000.0 0
. 000.0 000.0 200 £00 0
30.0 000.0 000.0 000.0 0:0 000.0 0 0.0
000.0 30.0 000.0 000.0 000.0 0&0 000.0 0
. #00 000.0 05.0 000.0 0
05.0 000.0 000.0 0000 0:0 0:0 0 0.0
000.0 30.0 000.0 000.0 «.000 0.0.0.0 0000 0
0 00¢ .000
0 0.0 0 98>
0 0.20000. .0002".
8.20 0.8» 0 m e 0 u .28:

 

 

 

 

 

 

 

 

 

 

 

 

.3080 "00 0.0....»

172

REFERENCES

Bass, Frank M. (1969), “A New Product Growth Model for Consumer Durables,”
Management Science, 15(5):215-228.

Bikhchandani, Sushil, David Hirshleifer and lvo Welch (1992), “A Theory of Fads,
Fashion, Custom and Cultural Change as Information Cascades,” The Journal of
Political Economy, 100(5):992- 1026.

Blattberg, Robert C., Thomas Buesing and Subrata K. Sen (1980),
“Segmentation Strategies for New National Brands,” Journal of Marketing,
44(4):59-67.

Bdckenholt, U. and Ft. Langeheine (1996), “Latent Change in Recurrent Choice
Data,” Psychometrika, 61:285-302.

Bohlmann, Jonathan D., Peter N. Golder and Debanjan Mitra (2002),
“Deconstructing the Pioneer’s Advantage: Examining Vintage Effects and
Consumer Valuations of Quality and Variety,” Management Science, 48(9):1175-
1195.

Boulding, William and Christen, Markus (2003), ”Sustainable Pioneering
Advantage? Profit Implications of Market Entry Order,” Marketing Science, 22(3):
371 -392.

Bowman, Douglas, and Hubert Gatignon (1996), “Order of Entry as a Moderator
of the Effect of the Marketing Mix on Market Share,” Marketing Science, 15(3):
222-243.

Burt, RS. (1987) “Social contagion and innovation: Cohesion versus structural
equivalence,” American Journal of Sociology. 92: 1287-1335.

Calantone, Roger J. and Alan G. Sawyer (1978), “T he Stability of Benefit
Segments,” Joumal of Marketing Research, 15(3): 395-404.

Chatterjee, Rabikar, and Eliashberg, Jehoshua (1990), "T he Innovation Diffusion
Process in a Heterogeneous Population: A Micromodeling Approach,”
Management Science, 36(9): 1057(23).

Christensen, Clayton, M. (1997) The lnnovator’s Dilemma, Harvard Business
School Press.

Coleman, J.S., E. Katz, and H. Menzel (1966), Medical Innovation: A diffusion
study, New York: Bobs Merrill.

173

' 2...;
.. OL'FI

 

 

Copeland, Adam, Wendy Dunn, and George Hall (2005), “Prices, Production and
Inventories over the Automotive Model Year”, Finance and Economics
Discussion Series Divisions of Research and Statistics and Monetary Affairs
Federal Reserve Board, Washington, 0.0. Working Paper 2005-25

Czepiel, J. A. (1975), “Patterns of interorganizational Communication and
Diffusion of a Major Technological Innovation in a Competitive Industrial
Community,” Academy of Management Journal, 18(1): 6-24.

Danaher, Peter J., Bruce G.S. Hardie, and William P. Putsis (2001), “Marketing-
Mix Variables and the Diffusion of Successive Generations of a Technological
Innovation,” Journal of Marketing Research, 38(4):501-514.

De Bruyn, Arnaud and Gary L. Lilien (2004), “A Multi-stage model of Word of
Mouth Through Electronic Referrals,” Pennsylvania State University Working
Paper.

Dekimpe, Marnik G., Philip M. Parker and Miklos Sarvary (1998), "Forecasting
the Diffusion Pattern of Cellular Telecommunications on the Country-level Using
a Staged Estimation Approach," Technological Forecasting and Social Change,
57, 105-132.

DiMaggio, P.J. and WW. Powell (1983), “The Iron Cage Revisited: Institutional
Isomorphism and Collective Rationality in Organizational Fields,” American
Sociological Review, 48:147-160.

Efron, Bradley. (1977) “The Efficiency of Cox’s Likelihood Function for Censored
Data,” Journal of the American Statistical Association, 72(359): 557-565.

Farley, John U., Donald R. Lehmann, and Russel S. Winer (1987), “Stability of
Membership in Market Segments Identified with a Disaggregate Consumption
Model,” Journal of Business Research, 15(4):313-328.

Feick, Lawrence F. and Linda L. Price (1987), “The Market Maven: A Diffuser of
Marketplace Information,” Journal of Marketing, 51 (January). 83-97.

Fischer, Claude S. and Glenn R. Carroll (1988), “Telephone and Automobile
Diffusion in the United States, 1902-1937,” American Journal of Sociology,
93(5):1 153-1 178.

Fredrick, May E. (1971), “T he Effect of Social-Class on Brand-Loyalty", California
Management Review, 14(1 ):81 -87.

Ganesh, Jaishankar and V. Kumar (1996), “Capturing the cross-national Ieaming

effect: An analysis of an industrial technology diffusion,” Journal Academy of
Marketing Science, 24(4): 328(10).

174

 

 

Ganesh, Jaishankar, V. Kumar, and Velavan Subramaniam (1997), “Learning
Effect in Multinational Diffusion of Consumer Durables: An Exploratory
Investigation,” Journal of the Academy of Marketing Science, 25(3):214-219.

Garcia, Rosanna and Roger Calantone (2002), “A critical look at technological
innovation typology and innovativeness terminology: A literature review,”
Journal of Product Innovation Management, 19(2): 110-132.

Gatignon, Hubert, Jehoshua Eliashberg, and Thomas S. Robertson (1989),
“Modeling Multinational Diffusion Patterns: An Efficient Methodology,”
Marketing Science, 8(3):231-247.

Goldenberg, Jacob, Barak Libai and Eitan Muller (2002), “Riding the Saddle:
How Cross Market Communications can Create a Major Slump in Sales,” Journal
of Marketing, 66(2):1-17.

Granovetter, M. (1973) “The Strength of Weak Ties,” American Journal of
Sociology, 78:1 360-1380.

 

Granovetter, M. (1983), “The Strength of Weak Ties: A Network Theory
Revisited," Sociological Theory, 1:201 -233.

Greve, Henrich R. David Strang, and Nancy B. Tuma, (1995), "Specification and
Estimation of Heterogeneous Diffusion Processes," in RV. Marsden, ed.,
Sociological Methodology. Oxford: Blackwell: 377-420.

Greve, Henrich R. (1998), “Managerial Cognition and the Mimetic Adoption of
Market Positions: What You See is What You Do”, Strategic Management
Journal, 19:967-988

Greve, Henrich R., Nancy Brandon Tuma, and David Strang (2001), “Estimation
of diffusion processes from incomplete data: A simulation study,” Sociological
Methodology and Research, 29(4): 435-467. * “1

Hofstede, Ter, Frenkel, Michel Wedel, and Jan-Benedict E.M. Steenkamp (2002),
“Identifying Spatial Segments in International Markets,” Marketing Science, . 1
21 (Spring):160-177.

 

Hsiao, C. (1986) Analysis of Panel Data, New York: Cambridge University Press.
Islam, T. and N. Meade (1997), “The Diffusion of Successive Generations of a

Technology: A More General Model,” Technological Forecasting and Social
Change, 56(1): 49-60.

175

 

Kalbfleisch, John D. (1974), “Some Efficiency Calculations for Survival
Distributions,” Biometrika, 61 (1 ):31 -38.

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(2): 105-119.

Kalyanaram, Gurumurthy and Glen L. Urban (1992), “Dynamic Effects of the
Order of Entry on Market Share, Trial Penetration, and Repeat Purchases for
Frequently Purchased Consumer Goods,” Marketing Science, 11(3): 235-240.

Kim, Namwoon, Dae Ryun Chang and Allan D. Shocker (2000), “Modeling
lntercategory and Generational Dynamics for A Growing Information Technology
Industry,” Management Science, 46(4):496-512.

Kim, Namwoon, Rajendra K. Srivistava and Jin K. Han (2001), “Consumer
Decision-making in a Multi-Generational Choice Set Context,” Journal of
Business Research, 53(3):123-136.

Kumar, V. and Trichy V. Krishnan (2002), “Multinational Diffusion Models: An
Alternative Framework,” Marketing Science, 21(3): 318-331.

Lenk, Peter J. and Amber G. Rao (1990), “New Models from Old: Forecasting
product Adoption by Hierarchical Bayes Procedures,” Marketing Science 9(1):
42-57.

Mahajan, Vijay and Eitan Muller (1998), ‘When is It Worthwhile Targeting the
Majority Instead of the Innovators in a New Product Launch?,” Journal of
Marketing Research, 35(November):488-495.

Mahajan, Vijay, Charlotte H. Mason and V. Srinivasan (1986), “An Evaluation of
Estimation Procedures for New Product Diffusion Models” in Innovation Diffusion
Models of New Product Acceptance, V. Mahajan and Y. Wing (Eds.), Cambridge,
MA: Ballinger: 203-234.

McAdam, Doug and Dieter Rucht (1993), "Cross-National Diffusion of Movement
Ideas: The American 'New Left' and the European New Social Movements”,
The Annals of the American Academy of Political and Social Sciences.

Midgley, David F. (1976), “A Simple Mathematical Theory of Innovative
Behavior,” Journal of Consumer Research, 3(1):31-42.

Midgley, David F., Pamela D. Morrison, and John H. Roberts (1992), “The Effect

of Network Structure in Industrial Diffusion Processes,” Research Policy, 21 (6):
533(20)

176

 

Moe, Wendy W. and Fader Peter S. (2002), “Using Advance Purchase Orders to
Forecast New Product Sales,” Marketing Science, 21(3): 347(20).

Moore, G. A. (1991). Crossing the Chasm. New York: Harper Business.

Muller, Eitan and Guy Yogev (2005), “When Does the Majority Become a
Majority? Empirical Analysis of the Time at which Main Market Adopters
Purchase the Bulk of Our Sales,” Working Paper, Tel Aviv University, Israel.

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, 22 (Sept): 1069-1086.

Putsis, William P Jr, Sridhar Balasubramanian, Edward H. Kaplan, and Subrata
K. Sen (1991), “Mixing Behavior in Cross-country Diffusion,” Marketing Science,
16(4): 354-369.

Rogers, EM. and DJ. Kincaid (1981 ), Communication Networks: A New
Paradigm for Research. New York: Free Press.

Rogers, EM. and GM. Beal (1958), “The Importance of Personal Influence in
Adoption of Technological Changes”, Social Forces, 36(4):329-334.

Rogers, E.M. J.R. Ascroft and N. Roling (1970), “Diffusion of innovations in
Brazil, Nigeria, and India,” Unpublished Report, Michigan State University, East
Lansing, MI.

Rogers, Everett M. (1995), Diffusion of Innovations. 5th ed. The Free Press. New
York.

Sharif, M. Nawaz and K. Ramanathan (1984), “Temporal Models of Innovation
Diffusion,” IEEE Transactions on Engineering Management, 31 (2)76-77.

Soule, Sarah A. and Yvonne Zylan (1997), “Runaway train? The diffusion of
state-level reform in ADC/AFDC eligibility requirements, 1950-1967,” American
Journal of Sociology, 103(3): 733-763.

Starr, Martin K. and Joel R. Rubinson (1978), “A Loyalty Group Segmentation
Model for Brand Purchasing Simulation,” Journal of Marketing Research, 15(3):
378-383.

Strang, D. and NB. Tuma (1993), “Spatial and temporal heterogeneity in
diffusion,” American Journal of Sociology. 99: 614-639.

177

 

 

Strong, D. and J. W. Meyer, (1994), “Institutional Conditions for Diffusion”. In
W.R. Scott, J.W. Meyer, and Associates (eds), Institutional Environments and
Organizations: Structural complexity and Individualism. Sage, Thousand Oaks,
CA:100-112.

Sullivan, Mary W. (1998), “How Brand Names Affect the Demand for Twin
Automobiles,” Journal of Marketing Research, 32(2):154-166.

Szymanski, David M, Lisa C. Troy, and Sundar G. Bharadwaj (1995), “Order of
Entry and Business Performance: An Empirical Synthesis and Reexamination,”
Journal of Marketing, 59(4):17-33.

Takada, Jain (1991), “Cross-national Analysis of Diffusion of Consumer Durable
Goods in Pacific-Rim Countries, Joumal of Marketing, 54:48-54.

Talukdar, Debabrata, K. Sudhir, and Andrew Ainslie (2002), “Investigating New
Product Diffusion Across Products and Countries,” Marketing Science, 21(1): 97-
116.

Urban, Glen L., John R. Hauser and John H. Roberts (1990), “Prelaunch
Forecasting of Automobiles,” Management Science, 36(4):401-421.

Valente, Thomas, W. (1995), Network Models of the Diffusion of Innovations,
Hampton Press Inc. Broadway Cresskill, NJ

Venkatesh Shankar, Gregory S Carpenter, Lakshman Krishnamurthi (1998),
“Late Mover Advantage: How Innovative Late Entrants Outsell Pioneers,” Journal
of Marketing Research, 35(1): 54-70.

Wedel, Michel and Wagner Kamakura (2000), Market Segmentation: Conceptual
and Methodological Foundations, 2"d ed, Dordrecht: Kluwer.

Yim, Chi Kin, and PK. Kannan (1999), “Consumer Behavioral Loyalty: A
Segmentation Model and Analysis,” Journal of Business Research, 44(2):75-92.

178

 

 

   

IIjllljljjjllljjljljjlljljjljl