 

,. 24.53 4.

... . “hem, :
. .. .
waging.»

 

 

 

 

 

 

 

 

 

.2‘
5.. y A: 5
‘ 3....3‘951

.9 A I :3
it 2.3...»
u. i 1 ix

 

an.
i
. .
.‘a 3-
IA

 

 

 

 

 

9 .9) (JV-5.3..
, , («n ’-
Juiuai

..

 

l

 

v 5.39..

S
1 5. P... . .1:
75.: ‘ 4.3.:

4.1. . . .. u
.55. :1 Flair
3:.

 

 

 

 

1 e LiRP .ARY

M 'r-:---.a iate
Oq .5 3L; u?“ \_.j L‘
2’0 Unwerssry 2

 

This is to certify that the
dissertation entitled

Income Inequality, Market Potential, and Diffusion of Mobile
Telephony

presented by

Sungjoong Kim

has been accepted towards fulfillment
of the requirements for the

Ph.D. degree in Media and Information Studies

 

 

/me

Major Professor’s Signature
I
// f/I/ a?

Date

 

MSU is an affirmative-action, equal-opportunity employer

 

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

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1 5/08 K:IProj/Acc&Pres/CIRCIDateDueindd

 

INCOME INEQUALITY, MARKET POTENTIAL, AND DIFFUSION OF MOBILE
TELEPHONY

By

Sungioong Kim

A DISSERTATION
Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of
DOCTOR OF PHILOSPHY
Media and Information Studies

2009

ABSTRACT

INCOME INEQUALITY, MARKET POTENTIAL AND DIFFUSION OF MOBILE
TELEPHONY

BY

Sungjoong Kim

The diffusion of many previous innovations eventually slowed down and reached

an equilibrium level. Despite continued rapid growth, it is possible that the diffusion of

mobile telephony will also begin to decelerate and reach a saturation level. Whether

universal service can be achieved with the help of mobile telephony will therefore depend

considerably on whether the diffusion of mobile telephony will stagnate before such

universality is reached. One key question in developing countries is whether inequality

will limit or delay the adoption of mobile telephony.

The goal of this dissertation is to contribute to a better understanding of these

issues. It investigates mobile telephony diffusion focusing on the effects of income and

other forms of inequality on two core aspects of diffusion: the saturation level (market

potential) and the speed of diffusion. The dissertation theorizes that market potential and

the rate of acceptance are functions of demand—side factors, supply-side factors. and

social conditions and tests corresponding hypotheses empirically.

A two-step approach was used to accomplish these goals. In a first step, three

statistical models of the diffusion of innovations (Bass, Gompertz, and logistic) were

employed to estimate market potential and parameters reflecting diffusion speed for 160

countries. The factors determining the variation of these parameter estimates across

countries were examined in a second step. For this purpose, regression analysis was used

to investigate the effects of supply-side, demand-side, and socio-cultural factors on the

diffusion parameters. Thus, diffusion models were used predominantly to obtain

estimates for the dependent variables used in the second part of the analysis. This second

phase is the main innovation and contribution of the dissertation.

The study revealed that income inequality had a statistically significant negative

effect on the speed of diffusion but not on the market potential. The dissertation also

found a statistically significant association between price and market potential. The effect

of income inequality on the speed of diffusion implies that it will most likely take longer

to achieve universal service in a society with highly unequal income distribution. The

association between price and market potential suggests that regulators could accelerate

the diffusion process by coaxing suppliers to provide inexpensive calling plans. Overall,

the dissertation contributes additional theoretical insights and empirical evidence to the

mobile telephony and the universal service literature.

Copyright By
Sungjoong Kim
2009

TABLE OF CONTENTS

List of Tables ...................................................................................................................... v
List of Figures .................................................................................................................... vi
1. Introduction ..................................................................................................................... 1

2. Literature Review

2.1. History of mobile telephony ................................................................................. 7
2.2. Diffusion of innovations ..................................................................................... 10
2.3. Research on mobile telephony diffusion ............................................................ 19
2.4. Income inequality and diffusion ......................................................................... 24

3. Theoretical Model and Hypotheses

3.]. Theoretical model ............................................................................................... 32
3.2. Hypotheses related to market potential ............................................................... 36
3.3. Hypotheses related to the rate of acceptance ...................................................... 41

4. Approaches and Methods

4.1. Statistical models of diffusion ............................................................................ 50
4.1.1. Bass diffusion model ................................................................................ 50
4.1.2. Logistic model .......................................................................................... 54
4.1.3. Gompertz model ....................................................................................... 55

4.2. Regression models for market potential and rate of acceptance ......................... 57

4.3. Data ..................................................................................................................... 59
4.3.1. Data collection and operational definitions. ............................................. 59
4.3.2. Data sets for regression analysis ............................................................... 62
4.3.3. Hypotheses test ......................................................................................... 67

5. Empirical Findings

5.1. Phase I: Estimation of diffusion parameters ....................................................... 73
5.2. Phase II: Determinants of diffusion parameters ................................................. 78
5.2.1. Factors explaining market potential ......................................................... 78

5.2.2. Test of hypotheses related to market potential ......................................... 87

5.2.3. Factors influencing the rate of acceptance ............................................... 88
5.2.4. Test of hypotheses related to the rate of acceptance ................................ 93

6. Dicussion

6.1. Discussion of findings ........................................................................................ 95

7. Concluding Remarks

7.]. Implications ...................................................................................................... 109
7.2. Limitations and questions for future study ....................................................... 110
Appendix I ...................................................................................................................... 117
Appendix II. .................................................................................................................... 119
Appendix HI. ................................................................................................................... 121
References ....................................................................................................................... 126

v i

List of Tables

Table 1. Estimated parameter values ................................................................................ 66
able 2. Operational definitions of variables ...................................................................... 69
Table 3. Descriptive statistics of the independent variables (BDM) ................................ 70
Table 4. Descriptive statistics of the independent variables (Gompertz) ......................... 71
Table 5. Descriptive statistics of the independent variables (Logistic) ............................ 72
Table 6. Correlation of the estimated ceiling values (12:20) ............................................. 77
Table 7. Correlation of the estimated ceiling values (11:46) ............................................. 78
Table 8. Regression on estimated ceiling values (BDM) ............................................... 83
Table 9. Regression on estimated ceiling values (Gompertz model) .............................. 85
Table 10. Regression on estimated ceiling values (Logistic model) .............................. 86
Table I 1. Results of market potential related hypotheses testing ..................................... 87
Table 12. Regression on estimated rate of acceptance (Gompertz model) ..................... 90
Table I3. Regression on estimated rate of acceptance (Logistic model) ........................ 91
Table 14. Results of the rate of acceptance related hypotheses testing ............................ 94

List of Figures

Figure 1. Theoretical model .............................................................................................. 47
Figure 2. Analysis process flow chart ............................................................................... 49
Figure 3. Number of countries in data sets ....................................................................... 65
Figure 4. Distribution of estimated ceiling ....................................................................... 76

viii

1. Introduction

Mobile telephony has experienced remarkable growth in the past two decades,
not only in the wealthiest parts of the globe but also in poorer regions. According to
Madden et al. (2004), the number of mobile subscribers doubled annually during the
19903, from 11 million in 1990 to halfa billion in 1999. The International
Telecommunication Union (ITU) reported that the total number of mobile phone
subscribers surpassed the number of fixed line subscribers in 2003, reaching 3 billion in
August 2007.l Global mobile penetration surpassed 50 % of the population in early 2008
and was expected to reach 61% by the end of 2008.2 An industry white paper predicts
that the number of mobile subscribers will grow to as many as 5 billion by 2012, driven
largely by fast market expansion in developing countries (Ericsson, 2007). In some
countries, the number of mobile subscriptions is larger than the total population3, for
example in Lithuania (138.1%) and Luxembourg (151.6%). With the introduction of low
cost handsets and inexpensive calling plans, the mobile industry in developing countries

is rapidly growing even in countries where fixed voice service had not been very

 

See http://wviw.itu.int/ITU-I)/icl/statistics/ict/index.html

See http://www.newkeralaxom/ttipstory-fullnews-26209Jitml

This is possible because a single user can have multiple subscriptions. The ITU reports mobile
penetration by the number of subscriptions.

’d I») -—'

successful.4

Such rapid development of mobile markets led to optimistic expectations as to
the possibility of achieving universal service or at least universal access5 in countries
with very low fixed line penetration and to hopes of overcoming the Digital Divide with
the help of wireless broadband (James. 2003; Ericsson, 2007). At first glance it seems
that there is finally a technology that could achieve the dream of universal service at a
global level if mobile telephony were to continue to grow at such fast speeds. But will the
present trends continue unabated until universal service is reached? Previous diffusion
processes of innovations eventually slowed down and reached an equilibrium
(“saturation”) level. A relevant question for policy-makers and other stakeholder is.
therefore, whether the diffusion of mobile telephony might slow down or even stop at
some point. If this is the case, corollary questions are at which level such a saturation
might occur and what factors might influence the process and outcomes.

The existing literature on mobile diffusion provides only partial answers.

Previous studies on mobile diffusion focus on the variables affecting the speed of

 

4 According to Ericsson (2007). the number of mobile subscribers outnumbers the number of the mainline
telephone in most countries. Among 160 countries of which mobile diffusion patterns were investigated in
this study, there were only ten countries that the number of fixed line telephone exceeded the number of
mobile subscriptions in 2005. In 37 countries. mobile subscriptions outnumbered ﬁxed line by more than
ﬁve times and as many as more than ten times in 13 countries.

5 Unlike the concept of universal service, which implies having subscription to a type of voice service. the
term universal access refers to a state in which every resident in a given area has access to phone service
with or without actually subscribing to a voice service. For example, if there is a public telephone in a town
that every resident may use. all townspeople have access.

diffusion rather than those influencing the size of potential demand. This tendency is

partly due to a widespread methodological practice in diffusion of innovation studies,

which is to transform the non-linear statistical models into linear forms.6 Because of this

practice, the question of what affects the ceiling level in mobile telephony diffusion has

been left largely unanswered. However, unless one assumes that mobile telephony will

reach the same saturation level in all countries, finding the determinants of the market

potential must be considered as important as discovering the factors that are influencing

the speed of diffusion.

Countries at a similar level of economic development may nonetheless have

socio-economic differences that may cause the size of potential demand for a new

product to differ considerably. One important candidate for such structural differences are

forms of inequality, from economic to other types. Income inequality may inﬂuence both

aspects of diffusion: market potential and the speed of diffusion. From an economic point

of view, different degrees of income inequality may result in a different proportion of

people who can afford mobile telephone service. Furthermore, high socio-cultural

 

6 Because of the usual S-shape of diffusion of innovations, the statistical models are in non-linear forms —
for example, a logistic equation. To solve the equations, in other words to acquire parameter values, it is
necessary to run non-linear regressions. However. currently available techniques for non-linear regression
do not allow a very sophisticated analysis compared to available techniques for linear regression. To avoid
such a limitation, researchers often transform the formula into a linear form. The transformed linear
equation is still not solvable unless the value for one of the parameters is known or fixed. A common
solution is to assume the ceiling level as a known value — for instance to apply a common ceiling value of l
(100% of potential adopters) to all of the countries (,Stoneman. 2001 ).

inequality, which often causes the relationships between different classes or socio-cultural

groups to be less harmonious, may slow down the diffusion process by impeding

information flows about an innovation between different social groups. Especially in

developing countries. where high levels of inequality have persisted and have been

identified as obstacles to economic development, mobile diffusion may show quite

different patterns depending on the level of inequality.

The goal of this dissertation is to investigate whether socio-economic differences

among and within nations, especially inequality. have a discernible effect on the diffusion

of mobile telephony. To this end, the dissertation focuses on the relationship between

income distribution on the one hand as well as market potential and speed of diffusion on

the other.

A complication in investigating the factors influencing the ceiling is that the

diffusion of mobile telephony in many countries is still in progress. However, the ceiling

can only be observed empirically after the market has reached its saturation point. Thus,

we need an alternative means to acquire the eventual ceiling values. One way is to predict

it from the available preliminary observations. Statistical models of diffusion of

innovation can provide a methodological solution to this problem as they allow an

estimation of diffusion parameters from longitudinal adoption data.

Diffusion processes may follow different patterns, and it is a matter of actual

observation which specification, if any, produces the best fit. This dissertation will use

three widely used statistical diffusion models to acquire estimates of the market potential

and the rate of acceptance (or equivalent characteristic parameters) for every country,

with longitudinal data of mobile penetration as input. This first step was carried out for

160 countries for which data was available. The market potential determines the level at

which the diffusion of a new product will end while the rate of acceptance decides how

fast the market will reach this national saturation level. Only countries in which mobile

diffusion could be described by one of the three models were included in the next step of

analysis. For these 160 countries. the diffusion parameters derived from step one were

used as dependent variables in phase two of the study. Regression analysis was used to

identify the factors that influence them. Thus, the first part of the investigation can be

seen as a preliminary stage to acquire data for the second phase of the analysis. The

primary goal of the dissertation is to investigate the factors that influence market

potential and the speed of diffusion (rather than a comparison of performance of the three

diffusion models).

The dissertation is organized as follows: the next chapter reviews the literature

on the diffusion of innovations, mobile diffusion, and on the effects of income inequality

on diffusion of innovations. The following chapter presents the theoretical model,

research questions, and hypotheses. Chapter four presents empirical methods and the data,

including information on data collection and a detailed description of the variables.

Chapter five reports findings and the following chapter six is dedicated to a discussion of

findings. Concluding remarks offer an outlook on future research questions.

2. Literature Review

This chapter reviews the literature pertinent to the topic. The first section is a

brief discussion of the history of mobile telephony. The next section reviews the research

on diffusion of innovations focusing on contributions relevant to mobile telephony

diffusion. A review of studies on income inequality and the diffusion of innovations is

provided in the third section.

2.1. History of mobile telephony

Observing the mobile industry at the turn of the Millennium, Gans et al. (2001)

divided the history of the sector into four periods: pre-cellular. first generation (10),

second-generation (2G), and third-generation (36). In the meantime, the next chapter in

mobile communication is being written with the gradual emergence of the next

generation of technologies, often summarily referred to as “4G”. Mobile technology

was invented at the beginning of the twentieth century. Before the introduction of cellular

technology, due to technological limitations, scarce spectrum was used inefficiently,

limiting its commercial potential (Gruber & Verboven. 2001b). The invention of cellular

K

technology in the I960s was a major technological breakthrough that opened the
possibility of mobile voice service as used today. In most countries, cellular mobile
telephone service was not launched until the 1980s often because of regulatory delays.
Japan is one of the few exceptions and started mobile cellular service in 1979 (Gans et al.,
2001 ).

First generation mobile telephones used analog technology. Several incompatible
technology standards were developed and adopted independently in different countries.
The US. opted to have a single standard, the Advanced Mobile Phone System (AMPS).
In contrast, multiple standards were adopted in European countries, including TACS
(Total Access Communications System), NMT (Nordic Mobile Telephone), and C-450.
The use of spectrum had become far more efficient with cellular technology and
subsequent technological improvements. Mobile telephony, however, was not widely
adopted because of its high price. During the early stages of market development carriers
mainly served businesses demand (Gans et al., 2001). On average, penetration of analog
mobile telephony in developed countries remained at less than five percent by the mid-
19903 (Rouvinen, 2006).

It was the introduction of 2G mobile telephones based on digital technology that

facilitated wide adoption. The first 26 service was launched in 1992 in Finland (Koski &

8

Kretschmer, 2006). Digital had several advantages over analog technology: first,

compared to analog service, due to a more efficient use of spectrum. the number of users

that could be served using a given bandwidth increased by 3-4 times even for the first

versions of digital technology. It increased further with technological improvements in

subsequent years (Gruber & Verboven, 2001b). Second. digital mobile telephony offered

better sound quality. fewer dropped calls, lower prices, and a greater variety of ancillary

services. In addition, data transmission was made possible including short message

service (SMS) and email (Gruber & Verboven, 2001b; Madden et al., 2004). Contrary to

the experience with analog service, the US. did not mandate a single national standard

for digital mobile. It was left to the industrial consortia and market coordination to

develop standards with the only condition that they were backwards compatible. In the

end, three (mutually incompatible) 2G standards emerged in the US. market (Gruber,

2005). On the other hand, European countries adopted a single 2G standard proposed by

the Groupe Spéciale Mobile (GSM) - which was later renamed to Global System for

Mobile Communications (Gans et al., 2001). The introduction of digital technology

accelerated diffusion considerably (Gruber & Verboven, 2001b).

Transition from 2G to 3G has been gradual. 2.5G and 2.75 G services offer

limited versions of the features that full-ﬂedged 3G services provide. 3G services were

9

first introduced in Japan and South Korea in 200]. Even though 2G was also based on
digital technology. it was still voice-centric (Tilson, 2006). Data services delivered via 2G
networks were mostly limited to simple text message services. 3G mobile telephony was
developed to meet anticipated demand for mobile data service. The success of the fixed
Internet sparked the interest of the suppliers of mobile data services (Tilson, 2006). In
addition, mobile voice services had reached or neared saturation levels in many richer
countries by the end of the twentieth century. Wireless companies shifted their focus to
the realization of “Wireless World Wide Web” services (Tanguturi, 2006). On top of
further improved voice quality, 3G allowed offering multi-media services and broadband
Internet access through mobile handsets. Six radio interfaces7 are approved under the

IT U-created global IMT-2000 (the International Mobile Telecommunications-2000)
standard. Except in the East Asian markets, 3G so far has achieved slower than

anticipated market acceptance.

2.2. Diffusion of innovations

According to Rogers (2003). the history of diffusion studies traces back to “early

 

7 Originally, the ITU approved ﬁve radio interfaces — IMT DS (Direct Sequence). IMT-MC (Multi-Carrier),
IMT-TD (Time-Division). IMT-SC (Single Carrier). and IMT-FT (Frequency Time). On Oct. l8th of 2008,
The ITU Radiocommunication Assembly formally recognized technology derived from IEEE 802.16 as the
sixth interface of IMT—2000.

IO

sociology” at the beginning of 20th century. Rogers (1976) credited Gabriel Tarde (1903)

with proposing the typical S-shaped diffusion curve and saw him as an important opinion

leader from the very beginning of the diffusion of innovation research tradition. Diffusion

of innovation studies had been conducted in several disciplines largely independent of

each other until 1960s, including anthropology, rural and medical sociology, education,

communication, marketing, and others.

It was the research of Ryan and Gross (1943) on hybrid com diffusion in Iowa

from which the revolutionary paradigm in diffusion research arose (Rogers, 1976). The

authors reported that the rate of adoption followed an S-shaped pattern. Furthermore,

innovators were more cosmopolitan and had higher socio-economic status. In addition,

even though the main source of product information was the salesperson, it was

interpersonal communication with fellow farmers that was most influential in the

adoption decision. However, modeling and forecasting diffusion of innovation came

much later with several pioneering works in the l950s and 1960s (Meade & Islam, 2006).

Rogers (2003) defined diffusion as “the process by which an innovation, that is a

new idea, is communicated through certain channels over time among the members of a

social system.“ There are four elements that affect the process: characteristics of

innovation, social system (individual, oroup, organization, or country), time, and

II

communication channel. Rogers distinguished five (perceived) characteristics of

innovation that may affect the diffusion process: trialability, relative advantage,

compatibility, observability — further divided into result demonstrability and visibility

(Van Slyke et al., 2007) - , and complexity.

Diffusion of innovation research has been carried out in various disciplines;

anthropology, communication, economics, education, geography, marketing, political

science, sociology. and others. Everett Rogers (McGrath & Zell. 2001) credited Jack L.

Walker (1966) with starting the tradition of diffusion of innovations study in political

science and Torsten Hagerstrand (1967) in geography. The types of innovations

investigated are also very diverse including technology or products that embody it,

ideology, information/knowledge/news, practice, policy, and so on. In a study that set up

a model for following news event studies, Deutschmann and Danielson (1957)

investigated diffusion of three news events — President Eisenhower’s stroke, the Explorer

I satellite, and Alaskan statehood, and found S-shaped patterns with flattening of the

curves due to “shut down for the night.” Rogers (2000) summarized that news event

studies had found that the news event diffusion process varies by the perceived salience

of the news events and timing of the news. Tremayne (2007) investigated a relative new

phenomenon of diffusion of dynamic contents (i.e., video clip) in online newspapers.

12

Using Roger’s theory as well as agenda—setting, general systems, and social change

theories, Strodthoff and his colleagues (1985) analyzed the diffusion of environmentalism

in the US. and the roles of mass media and social movement in the process. There is also

substantial literature on information technology (IT) diffusion. Prescott and conger

(1995) provided a review of the research tradition on IT diffusion. Rogers and Peterson

(2007) investigated factors that inﬂuence diffusion process of clean air ordinances in US.

communities.

Depending on the type of social system a study focuses on, theory and method

may differ greatly. Previous studies (Downs & Mohr, 1976; Tomatzky &Klein, 1982)

pointed out that it is not reasonable to generalize ﬁndings from the research on individual

adoption to the organizational innovation process. In a meta-analysis of studies on

organizational innovation, Damanpour (1991) listed 13 determinants of the diffusion

process (i.e., specialization, functional differentiation, formalization, centralization,

administrative intensity, vertical differentiation). Different factors are at work at national

level as well. In a cross-national diffusion research, Gatignon et. al. (1989) reported that

country-specific factors such as cosmopolitanism, mobility and women in labor force had

effects on the adoption process.

It is not possible. nor is it necessary, to provide a comprehensive review of all

13

strands of diffusion research in the context of dissertation. Rather, the discussion will
focus on research that is directly relevant to the topic of mobile telephony. most
importantly the research on diffusion of product and service innovation.

Even though the idea that the pattern of diffusion of innovations often follows an
S-shaped curve had been introduced from the very beginning of the research tradition, it
was not until the 1950s that researchers began to propose statistical models that fit the
pattern. Griliches (1.957) was among the first researchers to do so. In a study on hybrid
com diffusion among US farmers, he reported that the pattern of hybrid corn diffusion
followed an S-shaped curve.8 The observed pattern of diffusion was represented by a
logistic curve.9 However, a statistical model per se does not explain why the pattern is S-
shaped. Thus, an open question was why diffusion followed this particular pattern.

Meade and Islam (2006. p.522) pointed out that. “the two extreme hypotheses that
explain this shape are those based on the dynamics of a (broadly homogeneous)
population and those based on the heterogeneity of the population.” Bass (1969)
proposed that there are two characteristics of individuals that affect the diffusion of

innovations: their tendency to innovate: and their tendency to imitate others. The Bass

 

8 Ryan and Gross (1943) did not propose a statistical model.

9 Later, Dixon (1980) re—examined the same data and argued that the Gompertz curve is a better
representation of the diffusion process. The Gompertz curve is asymmetric and has an inﬂection point at
one—third of the number of potential adopters while that of a logistic curve, which is symmetric, is at the
half point.

14

diffusion model assumes that the chance of purchasing a new product is determined (1)
by a person’s own inherent attitude towards an innovation. and (2) by influences (or
pressures) from those who have already adopted the product. In his diffusion model, if
the tendency to innovate (represented by the coefficient of innovation p) is smaller than
that to imitate (represented by the coefficient of imitation q) the resulting curve is S-
shaped.IO In other words. Bass theorized that the diffusion of innovations follows an S-
shaped path because the influence from the adopters has greater impact on the decision to
purchase a new product than the desire to innovate of potential adopters.ll The Bass
diffusion model was developed to fit an S—shaped curve. For diffusion patterns that
deviate considerably from an S-shaped path, the Bass model may not be the best
approach.

In contrast, Rogers (2003) suggested a heterogeneous distribution of the
propensity to innovate in the population as the main cause of the particular shape of the
diffusion path. Adopters are grouped into five categories depending on the time of the
adoption; the first to adopt are “innovators” (approximately 2.5% of the eventual adopters

according to his studies). followed by “early adopters” (13.5%), then “early majority”

 

‘0 Talukdar et al. (2002) used the term “external influence” for innovation and “internal inﬂuence” for
imitation and this dissertation will use the same terminology.

” In BDM, The relationship between two parameters determines the shape of the graph. In the case when p
>0. q = 0, a pure innovation scenario. the diffusion path will resemble a modiﬁed exponential function.
When p = 0, q > 0. a pure imitation case. it becomes identical to the logistic model (Meade & Islam. 2006)

1:3

(34%), the “late majority” (34%), and lastly “Iaggards” (16%). Individuals have different

threshold levels for adoption with innovators having the lowest. If thresholds are

distributed normally, the diffusion path takes an S—curve form. Of course, this condition

may not be met in all societies. The exact form of the diffusion path will depend on the

skewness of the threshold distribution. Higher levels of education, literacy, social status

and wealth are related to favorable attitudes towards innovation (or a lower adoption

threshold).

Heterogeneity of income distribution, the fact that income is not equally

distributed among the population, has often been considered as one of the main causes of

an S-curved diffusion path (Meade & Islam. 2006). Usually, as diffusion of an innovation

proceeds, the price falls with technological improvements reducing production and

service costs. Price can decrease further if the market expands to reach a critical size that

enables economies of scale. When the price of a new product approaches or falls below

the threshold price of a potential customer — which will be determined by his income and

preferences — the likelihood of a purchase increases. If income is normally distributed and

the price falls monotonically, an S-curve diffusion path will result (Van den Bulte &

Stremersch, 2004; Meade & Islam, 2006). Liebemrann and Paroush (1982) reported that

income inequality, price and advertising are main drivers of diffusion. From a meta-

16

analysis of research that used the Bass Diffusion Model (henceforth: BDM), Van den

Bulte and Stremersch (2004) found that income distribution has a significant effect on the

shape of the diffusion curve.

The three diffusion processes used in this dissertation are ‘epidemic’ models that

view the process of diffusion as a rather automatic process. It mainly occurs through

contacts between adopters and potential adopters similar to the process of an epidemic. A

potential adopter’s probability of purchase is predicted solely by the proportion of

adopters in the total number of potential adopters. In other words, the only independent

variable is the number of adopters. Acknowledging the fact that diffusion of innovations

does not occur in a static environment, scholars have attempted to make diffusion models

‘dynamic’ by incorporating variables that influence parameter values. This was carried

out by the inclusion of explanatory variables in the estimation of (l) the market potential,

(2) the probability of adoption, or (3) both (Meade & Islam, 2006).

Mahajan and Peterson (1978) explained the market potential as a function of

explanatory variables, price and advertising expenditure. Proposing a new dynamic

model, they used housing starts in the US. to predict the saturation level for washing

machines. Horsky (1990) argued that the market potential is related to income (more

precisely wage) distribution and price. In his model. the saturation level increases with a

17

lower price, income growth, and with more equal distribution of wage. While examining

business telephones market in the UK. Islam and Meade (1996) compared the

performance of fixed saturation models and those of the models where saturation levels

were determined by indices that reflected changes in economic environment. The authors

found that the latter did not produce significantly better forecasts.

Robinson and Lakhani (1975) incorporated price in the probability function of

the Bass model. Horsky and Simon (1983) included advertising expenditures and found

that their model provided plausible estimates for telephone banking service. Thompson

and Teng (1984) proposed a synthesis of the above two models. Bass, Krishnan and Jain

(1994) developed a generalized Bass model as a response to these attempts. They found

the new modified model produced better forecasts than the original model. Talukdar et al.

(2002) provided a theoretical synthesis of what had been accomplished in attempts to

improve BDM by incorporating explanatory variables in parameter estimations.

Diffusion of innovations research has experienced continuous theoretical and

methodological refinement and sophistication. There has been growing acknowledgement

of the dynamic nature of diffusion process and subsequent attempts to make models more

dynamic by inclusion of explanatory variables in the parameter estimations. For the

diffusion of mobile telephony research, applying a common fixed saturation level of

100% does not seem to be appropriate, given various levels of penetration in the

countries where mobile market approaches saturation. Previous diffusion of innovation

studies provide methodological tools to investigate mobile telephony diffusion while

letting the saturation levels vary by countries.

2.3. Research on mobile telephony diffusion

Research on mobile telephony diffusion may focus on its unique characteristics

and investigate the influence of those characteristics on the diffusion pattern. On the other

hand, a study may analyze the diffusion of mobile telephony focusing on its similarities

with other new products and test whether general tendencies that have been found

inﬂuential in other studies are applicable to mobile telephony diffusion.

The diffusion of mobile telephony possesses several characteristics that make it

an interesting topic. One of them is the fact that there have been several coexisting

incompatible standards, sometimes in a single domestic market — for instance, 2G in the

United States. This variation provided a good Opportunity to observe the effects of

technology standardization. Gruber and Verboven (2001a, 2001b) found that

standardization accelerated diffusion, while Liikanen et al. (2004) did not find any

evidence of such an effect. Rouvinen (2006) reported a positive effect of standardization.

Koski and Krestschrner (2005) also found a positive effect although they pointed out that

price was higher in a single standard market.

Another interesting aspect of mobile telephony is that its history is marked by

several distinctive generations of a technology. Usually an innovation does not remain

unaltered; the technology continues to develop. Sometimes, as it is the case with mobile

telephony, the changes are sufficiently important to merit the label of a new generation.

An interesting question is what the effects are of preceding and following generations of

technology on the adoption of each other. Gruber and Verboven (2001a, 2001b)

investigated the effects of the introduction of digital technology (by including a dummy

variable) and found a significant effect on the rate of diffusion of mobile telephony. The

increased capacity of digital technology accentuated the effect of competition, which in

turn facilitated demand. Liikanen et al. (2004) found a positive effect of 1G penetration

on 2G diffusion and a negative effect of 2G on 1G diffusion.

As mobile telephony has to use scarce spectrum allocated and assigned by

govemment, the effects of regulation are also of interest. especially that of the licensing

policy — which determines the number of service providers and therefore has a strong

inﬂuence on the level of competition. Central issues are the optimum number of licenses

as well as the entry timing and entry mode of competitors. Gruber and Verboven (2001b)

found that having more than one operator has a significant effect on the speed of
diffusion in the European Union (EU) countries. A similarly designed study with an
expanded data set examined the differences in the effect of entry mode; simultaneous or
sequential entry of the competitor (Gruber and Verboven 2001a). The authors found that
entry of a competitor has a positive effect on the diffusion of mobile telephone service.
Sequential entry had a stronger effect - possibly as a result from the efforts of incumbents
to acquire as many as subscribers through more aggressive pricing before a competitor
enters the market. Rouvinen (2006) also found a positive effect of competition. Liikanen
et al. (2004), however, did not find any support for the effect of the number of licensees.
Koski and Kretschmer (2005) investigated two regulatory issues, licensing policy and the
degree of independence of the regulatory body. Their study of 2G diffusion revealed that
having competition from the beginning has a positive effect on the diffusion of digital
mobile telephony, even though pricing had been less aggressive”. The regulatory body’s
independence did not to have a significant effect.

Another interesting aspect of mobile telephony is that it typically is introduced in
markets with already established fixed line telephone service which can either be a

complement or a substitute. Understanding the relationship between fixed line and mobile

 

12 _ . . . . . . . .
They argued that less aggressive prrcmg might be due to increased non-price cornpctrtron.

21

telephony as complements, Rogers (2003) suggested that mobile telephony will benefit
from an existing fixed telephony customer base that serves as critical mass”. On the
other hand, Mariscal and Rivera (2006) pointed out that the mobile telephone has been
accepted mainly as a substitute in Latin America. Overall, the findings have been
inconclusive so far. Some studies found that higher penetration of fixed line facilitated
the diffusion of mobile telephones (Ahn & Lee, 1999; Gruber 2001; Gruber & Verboven
2001b). On the other hand. Gruber and Verboven (2001a) found a negative effect of
higher fixed line penetration in the EU, while Koski and Kretschmer (2005) found no
statistically significant effect. The wait list for fixed lines was found to have a positive
effect on mobile diffusion (Gruber 2001, Gruber & Verboven, 2001b). Koski and
Kretschmer (2005) found that liberalization of the fixed line market accelerated the
diffusion of mobile telephony. It is probable that mobile telephony will function as a
complement of fixed in the early stage of mobile diffusion and that it will become more
of a substitute in the later stage. Young (1993) provided a theoretical model in which the

relationship between old technology and new technology changes according to the

 

'3 Borrowing the idea from physics, Markus (1987) introduced the term ‘critical mass’ to the studies of
diffusion of interactive media. Critical mass is “the minimal number of adopters of an interactive
innovation for the future rate of adoption to be self-sustaining" (p. 721 , Mahler & Rogers, 1999). Mobile
telephony has an advantage that other new products normally do not have; if it is introduced to a market
where ﬁxed telephony has a sufﬁcient number of customers that can serve as an already reached critical
mass for mobile telephony. However, while such may be the case of developed countries, it may not be
entirely true for certain developing countries where mainline customer base is too small to serve as critical
mass for mobile telephony.

[\3
[\J

development of the market for the new product. During the early stage of diffusion, a new
product (or technology) is more of a complement for an old technology and even creates
new rents for the old technology. However, as the market matures, the new product’s
characteristic as a substitute to the old technology becomes more prominent.

Researchers also have paid attention to non-economic variables. Dekimpe et al.
(1998) included ethnic homogeneity as an independent variable arguing that a more
heterogeneous social system would have a negative inﬂuence on the diffusion rate and
the maximum penetration. The authors found that the number of ethnic groups had
negative effects on the number of adopters in the first year (the intercept), and penetration
growth, but did not find any significant relationship with the penetration ceiling.l4 In
addition, they found that the crude death rate, which was used as a measure of poverty,
has a negative relationship with the dependent variables. Rouvinen (2006) compared the
diffusion of digital mobile telephone in developed countries and that in developing
countries. Some differences were found between the two: the effect of market size
(population) was stronger in developing countries: the size of largest city had a

significant effect only in developed countries; and openness of the economy (the ratio of

 

'4 Even though the authors let market potential vary by countries, their model was not dynamic in that they
simply deﬁned that market potential as “the percentage of the population who is literate. lives in urban
areas and has a sufﬁcient income to afford basic telephone service” (pl 13).

[\D
C»:

imports and exports to GDP) was only significant in developing countries.

Economic factors were among the first to be considered and tested in studies of
mobile diffusion. Economic prosperity or a country’s wealth — typically represented by
GDP per capita — turned out to be an important factor in the studies by Dekimpe et al.
(1998), Ahn and Lee (1999), Gruber (2001), Gruber & Verboven (2001b), Madden et
al.(2004), and Koski and Kretschmer (2005). It had no significant effect in Gruber and

Verboven’s analysis of European countries (2001a) and in that of Rouvinen (2006).'5

2.4. Income inequality and diffusion

Inequality has not received much attention in diffusion of innovation studies.
This may be due to the fact that most types of inequality are not easy to measure, thus
making them less readily available variables in quantitative analysis. Income inequality,
especially its relationship with economic growth, has interested economists for a long
time. However, its economic effect. let alone social, cultural and political inﬂuence, is not
theoretically or empirically clear.

There have been two streams of research on income inequality; one is to

investigate the relationship between economic growth and income inequality; another is

 

'5 In both articles the authors pointed out that the result might have been caused by the high multi-
collinearity between GDP per capita and other independent variables such as main line penetration.

24

to investigate inequality and growth independently (Lundberg & Squire, 2003; Garcia-

Penalosa & Turnovsky, 2006).

In his inﬂuential work in 1955, Simon Kuznets hypothesized an inverted U-

shaped relationship between economic growth and income inequality income inequality

initially increases with economic growth but it decreases again once the economy has

reached a high level of development (Kuznets, 1955). Studies in the first group

empirically test the Kuznets’ hypothesis by running regressions of economic growth on

measures of income inequality. Even though there have been numerous empirical studies,

the findings so far have been inconclusive. Several studies (Alesina & Rodrik, 1994;

Persson & Tabellini, 1994; Alesina & Perotti, 1996) found a negative relationship. On the

other hand, other studies reported a positive relationship (Li and Zou, 1998; Forbes,

2000). Some researchers found different directions of the effects depending on the level

of development (Barro, 1997; Smith, 2001).

Theoretical explanations that have been suggested in support of a negative

relationship between inequality and growth are: (l) endogenous fiscal policy or political

economy (Persson and Tabellini, I994; Alesina and Rodrik 1994); (2) credit market

imperfections that prevent the poor from investing in human capital development (Galor

& Zeira. 1993); (3) reductions in investment resulting from socio-political instability

[\3
.1; l

caused by high inequality (Alesina and Perotti, 1996); and (4) increased fertility of the
poor (Kentor, 2001; Odedokun & Round, 2005).

A number of researchers investigated the effects of income distribution on
economic development by focusing on its relationship with demand. Murphy et al. (1989)
showed that domestic demand can increase with broadly distributed income. Lambert and
Pfahler (1997) developed a model that separates changes of income distribution into a
size effect (mean income) and a distribution effect. They examined the effects of two
types of income distribution change when the shape of the Engel curve'6 is concave or
convex to the origin. Mani (2001) investigated the effect of interactions between
inequality and demand patterns for goods on economic development. He argued that high
(initial) inequality results in shortage of initial demand for medium-skilled goods and can
lead to persistent underdevelopment and poverty. Matsuyarna (2002) argued that in order
to start the process of development towards a mass consumption society, inequality
should not be too high or too low. Too much equality would lead to a poverty trap, while
too little of it would result in premature end of development. In an attempt to explain the

modern history of economic development, Galor and Moav ( 2004) presented a unified

 

'6 An Engel curve shows how demand of a good or service changes with the change in consumers' income.
For normal goods. the Engel curve has a positive slope, meaning that the size of demand increases with the
increase in the income. On the other hand. the Engel curve for inferior goods has a negative slope. which
means that an increase in income is accompanied by the reduction in demand.

26

growth model. In the model, when accumulation of physical capital is the primary engine
of growth. inequality facilitates economic development by concentrating resources on the
individuals with higher propensities to save. However, when human capital accumulation
emerges as the main engine of development. inequality has a negative effect. It is because
the poor with credit-constraints cannot invest in activities that are necessary to
accumulate human capital (i.e., education).

Zweimuller (2000) investigated a relatively unexplored area in the income
inequality and economic growth literature: the effects of income inequality on market
demand for a new product. The author developed a theoretical model of economic
development in which the main driver of the growth is innovation. Income inequality
affects market demand for a new product, hence incentives to innovate, which in turn
affects economic development. If the homothetic preference assumption used in many
Schumpeterian models is discarded, the distribution of income will have effects on
demand because it determines the number of (potential) buyers. While changes in income
distribution do not affect the initial size of the market if the rich are sufficiently rich, it
affects how fast a market for a new product grows. Zweimuller and Brunner (2005)
suggested that in general more equality is favorable for innovation and hence for

innovation driven economic growth. Foellmi and Zweimuller (2006) distinguished two

27

effects generated from more equalin on demand for a new product: a market—size effect
and a price—effect. The market size effect means that if less concentration of wealth is due
to an increased number of the rich without increase of gross income of the rich, it will
result in relatively lower individual income of the rich. In such a case. the demand for an
innovation would increase because of the increased number of customers who can afford.
On the other hand, the price effect means that the rich are less willing to buy a new
product because of the decreased individual income and the producers are forced to lower
prices. The authors argued that the price effect always dominates the market-size effect.
Through a simulation of product diffusion, Reinstaller and Sanditov (2005) found that as
differences in socio-economic characteristics between two social groups increase, the
speed of diffusion accelerates while the saturation level. or potential demand. is lowered.
There also have been studies that investigated differences in the effects of income
distribution between developed and developing countries. The empirical findings
provided inconclusive answers. lyigun and Owen (2004) found that greater income
inequality led to reduced consumption in poor countries, but had the opposite effect in
richer countries. In studies on the relationship between income and food consumption,
Senauer (1990) found that lower-income households in developing countries were more

responsive to the price in rice consumption. while Park et al. (1996) reported that the

28

price elasticity for food was similar across different income strata in the US.

Talukdar et al. (2002) is a rare empirical investigation of the relationship
between income inequality and the diffusion of innovations. The authors theorized that
unequal income distribution would negatively affect a customer‘s ability to pay, hence
would lower the ceiling. However. the analysis, using BDM, found an insignificant
relationship between the Gini index and market potential. However, it was signiﬁcantly
related to the coefficient of internal inﬂuence (q) in the analysis with consumer
products.'7 The findings provide a limited number of clues to the relationship between
income inequality and diffusion of mobile telephony. Because the authors used pooled
data of six products, it is not clear whether the insignificance of the relationship between
income inequality and market potential applies to all six products.

Considering these findings of previous studies, what conclusions can be drawn
about the relationship between income inequality and diffusion of mobile telephony? At a
theoretical level, earlier work supported the idea that income inequality has an effect on
diffusion. Despite its theoretical usefulness. the assumption of homothetic preference is

empirically questionable. The literature on consumer behavior literature provides plenty

 

7 . . ,

I The authors used pooled data of SIX products - VCR players. microwaves. camcorders. CD players.
cellular phones, and fax machines. The ﬁrst four products were considered as consumer products, and the
last two were considered to be both consumer and business products.

29

of contradicting findings (Zweimuller, 2000). If we accept heterogeneous preferences of
customers, it seems that the distribution of income will have effects on demand.

This dissertation. however, deals with a specific type of service. Although theory
would expect similar effects to be at work, it is largely an empirical question whether this
particular service follows the expected pattern or not. In other words, mobile telephony
may or may not have characteristics that cause its diffusion path to differ from the typical
diffusion pattern of innovations.

However, some preliminary inferences are possible. It is necessary to check ﬁrst
whether the diffusion of mobile telephony has shown unique tendencies that would make
its diffusion atypical. What is frequently pointed out as unique about diffusion of mobile
telephony is that it is being adopted very fast in developing countries. However, the speed
of diffusion does not affect its ceiling. Thus, fast market expansion does not suggest that
the relationship between income inequality and market potential for mobile telephony
will be different from the relationships in other products.

Secondly. the price of mobile telephony may continue to decrease to a point
where the majority of the population can afford the service. In such a case, income
inequality may not have any noticeable effect. However, predicting future price trends is

not the goal of this dissertation. The dissertation does not attempt to make any prediction

on future trend. On the contrary, the current research attempts to find out factors that

inﬂuence market potential and the rate of acceptance using past pattern of mobile

telephony diffusion. Barrantes and Galperin (2008) indicated that, for the time being, the

price of mobile is not low enough for the poor in Latin American countries to afford

service.

Lastly. nowadays it is not uncommon to witness a person with more than one

hand set. It may be possible that multiple subscriptions of the rich may compensate for

lower demand by the poor. This would not be a problem if the mobile penetration data is

available in a format that can identify the actual number of users, which is currently not

the case. However, subscription to a mobile service is subject to the law of diminishing

marginal utility. In other words, marginal utility gained from having another hand set will

diminish as the number of handsets a customer possesses increases. Hence, there will be a

certain limit to the maximum number of subscriptions for a person, however rich he may

be. In addition, higher income inequality offsets some of the additional demand created

by the rich’s tendency to buy multiple subscriptions.

To summarize, there is no conclusive a priori evidence suggesting that mobile

telephony is such a unique service that income inequality will not have an influence on its

potential demand. This remainder of this dissertation will explore this question in detail.

3. Theoretical Model and Hypotheses

In this section, I present a theoretical model for the diffusion of mobile telephony
expanding Talukdar et al.’s approach (2002). First, rationales for each variable will be
presented, followed by related hypotheses. Discussion of the hypotheses referring to
independent variables already tested in previous mobile diffusion studies will be kept at a

minimum.

3.1. Theoretical model

The dissertation investigates diffusion of mobile telephony by focusing on two
main aspects of diffusion of innovations— market potential and the speed of diffusion. In
the statistical models of diffusion of innovations used in this dissertation, the latter aspect
of diffusion of innovations is captured by parameters that represent inﬂuences from the
factors that affect the chances of adoption of potential adopters. In the BDM, they are
coefficients of internal and external inﬂuences. In the logistic model, it is often called the
rate of acceptance, and the Gompertz model also has a parameter that is equivalent to the
rate of acceptance.

In Talukdar et al.’s (2002) synthesis of various modifications of the Bass

diffusion model, the factors inﬂuencing market potential18 are (a potential adopter's)
ability to pay, the willingness to pay, and ease of access to the product. External inﬂuence
was explained as a function of consumers’ access to product related information and
consumers” inclination and ability to process non-word-of—mouth information. Finally,
internal inﬂuence was explained as dependent on population hornophily and
persuasiveness of existing adopters. While the theoretical model is in a complete form, it
suffers from the difficulty that some of the factors, for example willingness to pay and
persuasiveness of existing adopters, are subjective in nature, thus making them hard to
quantify. More importantly, as the model is based on an individual consumer’s choice and
ability, it has a limited capability to adequately reflect supply conditions. Another
potential shortcoming of the individual-based model is that there can be structural
characteristics of a society that may slow down or even prevent information trickle down
between social groups.

Taludar et al.’s work is based on the BDM. In contrast. research in the
dissertation will use three models (Bass. logistic, and Gompertz), of which two have

different structures. It was decided to use three models because there is no guarantee that

 

l . . . . . L . . . . .

8 Talukdar et al. (2002) used the term synonymous With the ceiling or saturation point. Here, It refers to
the saturation level as estimated by statistical models of diffusion of innovation using annual mobile
penetration data as input.

the BDM is the best representation of mobile telephony diffusion, and which model will
fit the data best cannot be known in advance. Presentation of the hypotheses will be
limited to those that are common to all three models.

In the model used in this dissertation, market potential is a function of demand—
related factors (income and income inequality), supply-related factors (competition,
investment. price, standardization, and market conditions of substitutes/complements),
and social conditions that may have direct or indirect effects (social/cultural/gender
inequality, urbanization, and population density). Demand side factors inﬂuence potential
adopters’ ability to pay and their preference for the service. Supply factors such as
investment and price inﬂuence or reﬂect service providers’ ability to provide the service
and to make the service attractive to a larger number of people. The model shifts the
focus from an individual customer to macro and socio-economic factors. The
modiﬁcation allows the inclusion of supply factors in the model that may be as inﬂuential
as demand factors in the determination of market potential. Supply conditions are
important because customers have different ability to pay. In other words, the size of
market potential will depend on how far the price will decrease and how much utility of a
product can be increased — for example, by improving service quality and providing

ancillary services. Carriers may be forced to exert such efforts by competition. A carrier

in a fiercely competitive market may have to exert a greater effort to squeeze any
untapped demand, while a monopolist would be less enthusiastic about serving additional
demand.

All three statistical models of diffusion of innovations used in this paper are
epidemic models in that they theorize that an innovation spreads throughout the society
mainly through a potential adopters’ contact with actual adopters through which
necessary information is transferred and social pressure or inﬂuence is exercised.
Parameters that represent such an inﬂuence are the rate of acceptance in the logistic
model ([9), its equivalent in the Gompertz model, and the coefficient of internal inﬂuence
(q) in the Bass diffusion model.'9 For an innovation to be adopted, customers need to be
aware of it first. However, mere possession of the information is often not sufficient to
trigger an actual purchase of a new product. It often involves persuasion. Thus, the
interaction between adopters and potential adopters with regard to an innovation will be
affected by the chance of encounters that will convey product information and the
effectiveness of such an interaction. In other words, the magnitude of the parameters is

likely to be inﬂuenced by how dense and effective interpersonal communication channels

 

'9 The Bass diffusion model has another parameter that also determines speed of diffusion. the coefﬁcient
of external inﬂuence ([2). However. unlike the rate of acceptance in the logistic model and its equivalent in
the Gompertz model. it represents the inﬂuences from sources other than existing adopters in the society.
Lim et al. (2003) understood the coefficient of internal inﬂuence to represent ‘word-of—mouth effect,’ and
the coefﬁcient of external influence to represent ‘mass media effect.’

in a society are. Of course, a decision to adopt an innovation can also happen through
simple observation. While such an observation could provide information, it would not as
effective as inﬂuence from persuasion by an adopter. In this dissertation, I theorize that
the rate of acceptance will be influenced by the chance and the effectiveness of

interpersonal interaction that can convey information related to a new product.

3.2. Hypotheses related to market potential
The first set of hypotheses addresses the demand factors discussed above:
income size and income inequality. Higher income is likely to increase the number of

potential adopters due to the income effect known from consumer theory.

[Hypothesis l-l] Market potential will be positively related to income.

Income distribution will be another factor in determining market potential as it
affects the ability to pay, especially of the poor. High concentration of income in a small
segment of the population will likely decrease market potential as there will be fewer
people who can afford mobile telephone service. On the other hand, Barrantes & Galperin

(2008) indicated that the poor in developing countries are willing to spend a larger share

36

of their income on mobile service than their counterparts in developed countries. Thus, it

is possible that customers have a higher preference for cellular service that will cause the

effect of income inequality on potential demand to disappear. This dissertation will

empirically explore the relationship between income inequality and market potential of

mobile telephony.

[Hypothesis l-2] Market potential will be negatively related to income inequality.

The next three hypotheses address supply side factors. Price reduction brings the

product closer to an individual’s threshold price (Meade & Islam, 2006), which results in

an increased number of customers who can afford the product. Thus, price decreases

affect the market potential because each individual has different ability to pay.

[Hypothesis 1-3] Market potential will be negatively related to price.

Another important factor is investment volume, which is likely to increase

market potential. Greater investment will likely. though not necessarily, result in a larger,

more advanced. and better maintained network, more stable connections and better sound

quality, improved customer service and so on. All of those improvements would make

mobile telephone a more attractive service, thus likely to increase market potential.

[Hypothesis 1-4] Market potential will be positively related to investment volume in the

mobile industry.

Conditions in the markets for complements and substitutes will also be important

in determining market potential.20

It is fixed line telephony in this paper. The market
conditions of fixed telephone service will have an impact on market potential for mobile
telephony because there is network effect. Rogers (2003) argued that higher penetration
of fixed telephone would accelerate the diffusion of mobile telephony, especially in the
earlier stages of mobile diffusion. if critical mass has already reached in fixed telephony
market. He essentially understood fixed line service as a complement. Higher penetration
of wired telephony also increases the benefits of joining a mobile network because a new

mobile subscriber will be connected not only to existing mobile subscribers but also to

fixed telephony subscribers as well. On the other hand, customers in developing countries

 

20 One aspect of the relationships between mobile and ﬁxed telephone that cannot be explored in this paper
is that the direction of the influence can be either positive (complement) or negative (substitute). Though it
was found that mobile telephone may become a substitute to the wired line telephone - thus. reduces
demand of wired - once its penetration has reached a certain level. An opposite effect — that high wired
penetration reduces demand of mobile - has not been found yet.

38

have adopted mobile telephone service mainly as a substitute to main line telephone
service (Mariscal & Rivera, 2006). Thus. it remains as an open question subject to
empirical investigation whether the relationship between mobile and main line telephony

is substitutive or complementary.

[Hypothesis 1-5] Market potential will be positively related to fixed line penetration.

The remaining hypotheses are related to socio-cultural factors. The level of
urbanization and the population density are likely to affect market potential because of
economies of density. In addition, there is a possibility that the urban population will find

having mobile phone convenient due to their mobile life style than the rural population.

[Hypothesis l-6] Market potential will be positively related to the level of urbanization.

[Hypothesis l-7] Market potential will be positively related to population density.

Lastly, social, cultural. and gender inequality may also affect overall demand,
albeit indirectly. Previous research on inequality and economic growth indicates that

inequality is harmful for economic growth in that it may lead to disruptive social and

political conﬂict (Gottschalk & Iustino, 2006).

[Hypothesis 18] Market potential will be negatively related to gender inequality.

It is, however, not easy to quantify socio-cultural inequality. Thus, in the paper, a
concept and measure of socio-cultural fractionalizationzl developed in a past study
(Alesina et al., 2003) will be used instead of a direct measure of socio-cultural inequality.
Fractionalization of a society is a closely related concept to inequality. While it does not
necessarily result in a less egalitarian society, it is one of the main sources of inequality.
Such inequality in turn may negatively affect economic growth through increased social

tensions and conﬂicts.

[Hypothesis 1-9] Market potential will be negatively related to ethnic, linguistic and

religious fractionalization.

 

7 . . . . . .
“I A detailed description of the measure rs provrded in the next chapter.

40

3.3. Hypotheses related to the rate of acceptance22

This dissertation theorizes that the rate of acceptance is affected by demand-side
factors. supply-side factors. and social conditions. In addition. since potential adopters
need to learn about the new product before they make purchase, factors that affect
customers’ access to product related information will affect the rate of acceptance.

The first set of hypotheses address economic factors. Income size, fixed line
penetration, price and competition have been tested in previous studies and found to be
significant in many cases. Rogers (2003) argued that a higher fixed line penetration will
positively affect speed of mobile telephony adoption. It is because more number of
mainline subscribers increases benefit from subscribing to a cellular service. In other
words, network effect is created not only by new mobile users but also emanates from
main line subscribers. On the other hand, a higher level of competition usually translates
into more aggressive business strategies of the carriers, which often involves greater

efforts to ‘persuade’ potential adopters.

[Hypothesis 2-1]. The rate of acceptance will be positively related to the income.

 

a . 1 . ,, . _ . . .

‘2 Even though the term ‘rate of acceptance applies only to the parameter b in the logistic model. here for
the sake of brevity the term also refers to its equivalents in other two models. including the coefﬁcient of
external inﬂuence in BDM.

41

[Hypothesis 2-2]. The rate of acceptance will be positively related to fixed line

penetration.

[Hypothesis 2-3]. The rate of acceptance will be negatively related to price.

[Hypothesis 2-4]. The rate of acceptance will be positively related to the level of

competition.

The following set of hypotheses is about the factors that may inﬂuence the ﬂow

of product information among the members of a society. As argued earlier, adoption of a

new product is necessarily accompanied by the diffusion of product related information —

in the form of simple knowledge transmission and that of persuasion. Thus, it is likely

that the effectiveness and density of the interpersonal communication network of a

society will inﬂuence the likelihood of a potential adopter’s purchase.

The level of inequality is one indicator of social cohesion. High inequality in a

society causes polarization, which will make communication between different social

groups less effective. Earlier research found a positive relationship between inequality

and various forms of socio-political conﬂict (Lichbach, 1989; Gupta, 1990: cited from

Gottschalk & Justino, 2006). Thus, high inequality in a society will have a negative effect

on the coefficient by reducing the chances and effectiveness of interpersonal

42

communications that transmit innovation-related information from one group to others.

[Hypothesis 2-5] The rate of acceptance will be negatively related to income inequality.

Heterogeneity of the population can be rooted not only in the socio-economic

structure of the population but also in ethnic and cultural diversity. Even though co-

existence of diverse ethnic, linguistic, and religious groups in a society does not

necessarily mean that communication among different sub-groups is infrequent and

ineffective, it is more likely to be so than in a more homogenous population. Thus, one

would expect that socio-cultural fractionalization will be negatively related to the

coefficient.

[Hypothesis 2-6] The rate of acceptance will be negatively related to ethnic, linguistic,

and religious fractionalization.

Gender inequality also will likely affect the rate of acceptance negatively. In a

society with high gender inequality, women’s need to subscribe to cellular service is

likely given lower priority compared to a more egalitarian society. In addition, women in

4 3

such settings will likely have a smaller inﬂuence on the economic decisions in her

household.

[Hypothesis 2-7] The rate of acceptance will be negatively related to gender inequality.

Population density and the level of urbanization are factors that were frequently
tested and found to be significantly related to the rate of acceptance in mobile diffusion

studies.

[Hypothesis 2-8] The rate of acceptance will be positively related to population density.
[Hypothesis 2-9] The rate of acceptance will be positively related to the level of

urbanization.

There are inﬂuences and information that is acquired through sources other than
interpersonal communication. In the logistic and Gompertz model, inﬂuences that affect
adoption and information about a new product come from a single source — existing
adopters. On the other hand, the Bass model has two separate parameters that represent

effects from two different sources. The coefficient of internal inﬂuence (q) refers to the

44

inﬂuence from adopters within a country or a society while another parameter the

coefficient of external inﬂuence (p) captures effects from the variables that affect

potential adopters’ decision independent of existing adopters in the country (Talukdar et

al., 2002). It refers to inﬂuences exercised through non-interpersonal communication

- 23 . . .
channels and those from non-domestic sources . Such inﬂuences increase wrth

“consumers’ access to product-related information”, and “consumers’ inclination and

ability to process infomiation from” non-interpersonal channels (p. 103, Talukdar et al.,

2002). In a modern society, the most likely candidate for the non-interpersonal source of

product-related information is advertisement, most of which is delivered though mass

media. Since it is difficult to acquire information about expenses on advertisements, the

influence of information through mass media. as measured by the ratio of household with

TV sets to all households. will be used in this paper to represent non-interpersonal

sources of product information.

[Hypothesis 2—10] The rate of acceptance will be positively related to access to related

information through mass-media.

 

33 The coefﬁcient of external inﬂuence (p) is similar to the parameter a in the logistic model in that it
represents the intercept or the number of adopters during the ﬁrst time interval. However. the two do not
represent the same aspect ofdiffusion as p in the Bass diffusion model affects the shape of the curve while
a in logistic model does not.

Information can also be acquired from foreign contacts or trans-border media.

The chance of acquiring information from foreign sources will depend on how frequent

interactions with foreign contacts are. A more open economy, in terms of international

trade. will likely increase such chances.

[Hypothesis 2-1 I] The rate of acceptance will be positively related to the openness of the

economy.

Learning from foreign experience is not exclusive to consumers; suppliers can

also learn from experiences abroad and better prepare themselves for the eventual

launching of domestic service. Thus, as Gruber and Verboven (2001a, 2001b) found in

their research, introductory delay will likely have a positive effect because late adopter

countries have more time to learn from experience abroad.

[Hypothesis 2-12]. The rate of acceptance will be positively related to introductory delay.

   
 

35383.. no 83—

LHUl

Saab: 22%.“; 25¢:
Ur

335:5 3:32

 

 

 

x225 b23855 Maﬁa—uses—
:25825 5353-35;

562: $22 ”cemaaﬁueuﬁ 335.:— 3 883‘

 

 

 

 

bison cessaom

/ 5239525 5:335 ”285555 32%
BacmszmbceEoEEcU 52:59:30
Ill

.:c:§§.§§m 62.5 45,532: ”E398 .2996

3:55on .2555 6.535 “Ea—:00

 

 

 

EnoE 32852:. ._ oSwE

4. Approaches and Methods

The analysis was carried out in two phases. The first phase was concerned with

the estimation of parameters of the diffusion processes. Since it is impossible to

empirically observe the market potential until the market has actually reached the

saturation point. it was necessary to acquire market potential values through other means.

The diffusion of innovation literature provides a plethora of statistical models that allow

estimations of ceiling and parameters that characterize the shape of the diffusion curve.

The underlying assumption in parameter estimation is that. diffusion pattern data closely

resembles the pattern portrayed by the statistical model employed. This dissertation used

three widely used diffusion models — Bass diffusion model, the Gompertz model, and the

logistic model - in estimating parameters of the diffusion processes. Using annual mobile

penetration growth data for 27 years since 1980 as the input. the ceiling and the rate of

acceptance and its equivalences were estimated for each of n=l60 countries. The main

goal of the study was to explain the factors inﬂuencing the relative magnitude of these

parameters. Thus. the statistical models of diffusion of innovations were employed as an

intermediate step to acquire estimates for dependent variables that would be inaccessible

otherwise.

Figure 2. Analysis process ﬂow chart

 

l,)i f fusion Data

 

 

 

JZL

 
    
   
 

Parameter Estimation based on

longitudinal data

 
 
 

   

(BDM/Logistic/Gompertz. n: 160)

Q i)

 

 

 

 

 

 

 

 

Estimated values of Estimated values of
market potential Rate of acceptance

 
   
   
   

 
     
   

Regression analysis of factors Regression analysis of factors

  

affecting market potential affecting the rate of acceptance

(cross-sectional, n=30,34.49) (cross-sectional. n=30.34,49)

In the second phase, using market potential and the rate of acceptance (and its

equivalences) estimated in the first phase as dependent variables, a set of regression

analysis was carried out to examine the factors inﬂuencing the values of the estimated

ceiling and the rate of acceptance values. Not all estimations were successful. In some

cases, parameters were statistically insignificant at 95% confidence level, or had

estimated values that did not seem plausible. In addition, widths of estimated parameter

values were very large for some countries. In the estimation using BDM, many countries

49

had coefficients of internal and external influence values that were out of acceptable

range (0 S p, q S 1). Consequently, the number of countries included in the second phase

of analysis was much smaller than that in the first phase. Each data set had different

number of countries. Data set for logistic model had 49 countries, for Gompertz 34, and

for BDM 30. Data for this second phase were cross-sectional. In other words, the

differences in the diffusion parameters derived from the longitudinal analysis were

explained by differences between countries in the data set. The ﬂowchart above describes

the analysis process.

Multiple statistical models of diffusion were used because it was impossible to

know in advance which model presented a better fit with the data; variations of all three

models have been used in previous mobile diffusion studies. The following section

introduces the three diffusion models. The regression models for market potential and the

rate of acceptance will be introduced in the following sections.

4.1. Statistical models of diffusion

4.1.1. Bass diffusion model

The Bass diffusion model can be categorized as an epidemic model in that it

theorizes an individual’s inherent desire to imitate others as the main driver of the

diffusion of innovations, which is represented by the coefficient of internal inﬂuence. An
additional coefficient of external influence distinguishes BDM from other epidemic
models. The parameters in other epidemic models represent only the inﬂuence from
adopters, while the coefficient of external inﬂuence represents inﬂuences from sources
other than adopters.

In the Bass diffusion model, the number of adopters between time interval t-l and

r, X(t) is;

X(t) =pm + (q-p)N(,_i,- q/mN2(,_|, (05p, q 5 1,0 S m, t=l,2, 3...., T) (I)

Where Nu-” is the cumulative number of adopters at time t-I, and m is the
eventual number of adopters or market potential. p is the coefficient of innovation, and q
is the coefficient of imitation. Since N“), = 0, pm represents the number of adopters that
purchased the product during the first time interval of adoption. Originally, Bass used the
ordinary least-square estimation (OLS) method to obtain the values of m, p, and q. It has
been, however, repeatedly reported that OLS is prone to wrong parameter signs. In

addition, it was also found that the BDM does not usually perform well when the number

of observations is less than 10 (Meade & lslarn, 2006).24

Two major approaches were proposed and have been used frequently instead of
the original OLS approach. One is maximum likelihood estimation (MLE), introduced by
Schmittlein and Mahajan (1982), and the other is nonlinear least squares estimation
(NLS), proposed by Srinivasan and Mason (1986). Both alternatives use the probability
density function of adoption in BDM — which is not exactly identical to the function for
the number of adopters (Schmittlein & Mahajan, I982) - to derive their statistical models.
Schmittlein and Mahajan (I982), Srinivasan and Mason (1986) showed that the
performance of the two approaches is equivalent and better than that of OLS. Despite
their equivalence, NLS turned out to be the more popular of the two (Meade & Islam,
2006). For the analysis with the Bass model, NLS will be used to determine its
parameters in this paper.

The adopters’ probability density function ﬁt) for adoption at time t is:

fit) = (p + qF(I))(1 - Fit» (2)

It has the following distribution function:

 

24 . - - . -
The latter weakness strll applies to alternative methodological approaches.

(.31
[\3

Fm = {1 - expr - (p + (Nil/l 1+<q/p>expt- (worn (3)

Sales (the number of adopters) at ith time interval (r,--,, r,-) X(z’) is given by:

X(i) = "IlFtti ) - F(t.--1)l + “i. (4)

By substituting F(r) in (4) using equation (3), we have;

X(i) = m(l 1 — €XPI - (p + q) t.- M 1+(q/p)eXpl- (p+q) ti] }

- {1 - eXpl - (p + q) 11-1 }/I 1+(q/p)eXpl- (p+q) till I) + at (5)

Where it,- is and an additive error term with variance 5:, and i = l. 2, 3, 4....., T.

Equation (5) was used in this paper to derive estimates for parameter values.

BDM has restrictions on parameter values. Both coefficient of external inﬂuence
([2) and coefficient of internal inﬂuence (q) should have values between 0 and 1 (0 S p, q
< I). In addition, in order to be S-shaped, the coefficient of external inﬂuence (p) should

have a value smaller than that of the coefficient of internal inﬂuence ((1)

4.1.2. Logistic model

The logistic model was first proposed by Griliches (1957) and has been used in
numerous diffusion studies. One of the reasons for the popularity of the logistic model as
a representation of growth is because “it may fit empirical data better than other functions
with similar shapes” (p. 58. Oliver, 1964). It has a simpler structure from BDM and is
free from the problem of unacceptable parameter values. BDM is identical to the logistic
model in the pure imitation scenario (p = 0, q >0). In some countries. diffusion of mobile
telephony exhibits deviations from an ideal S-shaped curve, possibly due to the
introduction of a new generation of technology and occasional economic turbulence. For
the countries with such mobile diffusion history, the logistic model may fit better than the
more complex BDM.

The basic logistic equation is written as:

P” : Kl/[l+e-((1i+l)it)] (6)

Where Pi, represents the proportion of total adopters in country i, at time t, and K
is the ceiling value. The parameter b, which is often referred to as the “rate of

acceptance”, determines the rate of increase in P over time. The rate of acceptance is,

even though it does not exactly represent the speed of adoption”, related to how fast a
market will be saturated. The above equation is often transformed into a linear formzé.
The transformed linear equation best describes the role of the rate of acceptance (Dixon,

1980). The equation can be written as:

Iog(Pir/Ki-Pir) = a, + by (7)

This dissertation uses the original form — equation (4) - and employs the iterative
non-linear square method to estimate parameters, as had been suggested by Oliver

(1964).27

4.1.3. Gompertz model
In a review of Griliches’ work, Dixon (1980) proposed the Gompertz model as a
better representation of hybrid corn diffusion. He found that the diffusion pattern of

hybrid corn often showed deviations from the logistic curve. Diffusion was skewed with

 

25 As it can be seen from the statistical models. the parameter represents the chance of adoption of a
potential adopter, while ‘speed’ means how many years it will take before the market is saturated. In the
Bass model, the chance of adoption is represented by two parameters. coefﬁcient of external inﬂuence and
coefﬁcient of internal inﬂuence.

26 Stoneman (2002) indicates that even though it is a very common practice in diffusion literature.
transformation itself does not solve the problem of non-linearity. The left side of equation still contains a
parameter of which value is not known. To circumvent this problem. a usual approach is to treat the ceiling
as a known value.

37 He argued that “there is no substitute for full least squares in estimating logistic growth function” (p.65).

OI
CI

long tails and tended to be asymmetric, while a ‘true’ logistic curve should be symmetric.
He found that the Gompertz function provided a better fit with hybrid corn diffusion.

The equation for the Gompertz function may be written as:

Pi! = Kiai/m (8)

Where Pi, represents the proportion of total adopters of country i, at time t, and K
is the ceiling value, a and ,6 are parameters. The rate of growth equation for Gompertz

model can be written as:

dep/(If = In ’81:” Pg,(]nP-IIIK,') (9)

Equation (9) was used in the parameter estimation in this paper.28 In ,6 has a
similar role to the rate of acceptance in the logistic model (Dixon, 1980).
Estimation was carried out using a statistical software package, NLREG, which

allows running a non-linear regression without transforming original equation into a

 

28 Dixon (1980) used equation (7) in parameter estimation instead of (8). He, however, ﬁxed the value of
ceiling at l. A trial with mobile diffusion data using equation (7), albeit without ﬁxed ceiling value. was not
successful. Thus, I employed equation (8) that has one less parameter to estimate.

56

linear form. First, mobile penetration data from 1980 to 2005 or 2006 for 160 countries
was converted into required formats and fed to the software.29 Next, one country at a
time, the software repeatedly applies different set of values of parameters — each trial is

called iteration - until best estimates for parameters were found.

4.2. Regression models for market potential and rate of acceptance

This section introduces the regression models for market potential and that for
the rate of acceptance, which are based on the theoretical models discussed in the
previous chapter. The sources and definitions of variables are described in the following

section.

The model for market potential is specified as:

Market Potential = a + ,8, [GDP per capita] + ,8; [Income inequality] + ,65[Price] +

,64[Investment] + [3’5[Fixed line penetration] + Bo [Socio-cultural fractionalization] + ,6;

[Gender inequality] + [38 [Urbanization] + ,89 [Population density] + e,- (10)

 

29 As it can be seen from the equations, the Gompertz and Logistic models required cumulative sales as
input while BDM required incremental sales. Thus, the number of countries analyzed was slightly smaller
for BDM (12:152 compared to 160 for others).

OI

Where 8,- is an error term.

The model for the rate of acceptance is written as below; a, is again an error term;

Rate of acceptance = a + ,8, [GDP/capita] + [33[Fixed line penetration] + [33[Price]

+ﬂ4[Competition] +/i’,-[Income inequality] +ﬂ6 [Socio-cultural fractionalization]+

,87[Gender inequality] + ﬂdpopulation density] + [39[Urbanization] + ,[1’10[Mass media] +

ﬂu[Openness of economy] + ﬂlgﬂntroductory delay]+ a, (l I)

Since the coefficient of internal inﬂuence in BDM represents only the inﬂuence

from existing adopters, the regression model is divided into two equations: that for the

coefficient of internal inﬂuence and that for the coefficient of external inﬂuence. The

model for the coefficient of internal inﬂuence is specified as;

Coefficient of internal inﬂuence = a + /3,[Income inequality] +/J’3[Socio-cultural

fractionalization]+,B_.[Gender inequality] +ﬂ4[Population density] + /3_5[Urbanization] +

8i (17-)

vi]
CL;

The coefficient of external inﬂuence represents inﬂuences from sources other

. . . . . 30
than adopters wrthm the society. The regressron model rs ;

Coefficient of external inﬂuence = a + [MMass media] + ,l)’3[Openness of economy] +

/)’.;[Introductory delay]+ e,- (13)

4.3. Data
4.3.1. Data collection and operational definitions

Table 2 on page 69 provides a summary of the variables. Information related to
mobile and fixed telephony including penetration, price, and investment was collected
from the ITU World Telecommunication/1C T Indicators database. Information about
income inequality, measured by the Gini coefficient, was collected from the online
version of the World Bank World Development Indicators database and was partially
augmented by information from the World Factbook, Central Intelligence Agency (CIA).
The rest of the data was collected from the World Development Indicators database

except the number of carriers per year and the number of standards, for which

 

30 The variables such as GDP per capita. ﬁxed line penetration, price. and competition could not be
included in the either of regression models for parameters in BDM because theoretically they could not
ﬁt into any of them. Talukdar et al. (2002) did not include such variable either.

59

information was collected from the Telegeography’s GlobalComm Database.

Mobile penetration was measured by the number of mobile cellular subscribers
per 100 inhabitants. The estimated market potential (or ceiling) was also expressed in the
number of mobile cellular subscribers per 100 inhabitants. Income was measured by GDP
per capita. Fixed line penetration was measured by the number of main lines per 100
habitants as reported by the ITU. The monthly subscription charge (in current US S) was
used as a measure for the absolute price, while absolute price divided by GDP/capita was
employed to represent the relative price.3 I Due to missing data on investment in mobile
industry, investment was approximated by the share of investment in the
telecommunication sector in GDP.

The ratio of females in the labor force was used to as an indicator of gender
inequality. Fractionalization measures developed and employed by a previous study
(Alesina et al., 2003) were used to represent social and cultural heterogeneity. There are
ethnic, linguistic, and religious fractionalizations. The authors collected the information
about ethnic, linguistic, and religious groups in each country from the Encyclopedia

Britannica. The Encyclopedia “reports the shares of languages spoken as mother tongues,

 

3‘ Though price was measured in two different ways, absolute price was only included in order to gain

additional information. Results from the regression analyses with absolute price speciﬁcation were not

considered in the hypothesis tests because it would be price compared to income. not price itself , which
may affect market potential.

60

generally based on census data” (p. 159). Though the authors indicated that the
information about religion is less controversial and subjective than that of language or
ethnicity. it does not distinguish sects within a same religion. The formula for the index
N .,
is: FRACTj = l - 2 5“,]. where s,-,- is the share of group i (i = l N) in countryj.
r=l
Similar to previous research (Rouvinen, 2006), openness of the economy was
measured by the ratio of exports and imports to GDP. The average number of major
. . . . . . 32
servrce provrders was used to represent the level of competrtron 1n mobile markets. The
level of urbanization was represented by the percentage of population living in urban
areas. Finally, introductory delay was calculated by counting the number of years had
. . .. . . . . . 33
passed srnce 1980 until mobile vorce servrce was introduced in country 1.
The regressions explaining the factors inﬂuencing the estimated market potential
and the rate of acceptance are cross-sectional. However, the estimated values for the two
parameters are the outcomes of cumulative sales. Thus, it raised a question whether
independent variables should be measured as averages or by using the most recent data.

Initially, it was planned to use both averages and latest values, for instance the average

fixed line penetration from 1980 to 2006 and the main line penetration in 2006. However.

 

32 The information in GlobalComm database was extensive but sometimes lacked necessary information in
early days of mobile telephony. It has to be augmented by information from various sources such as
academic literature. web resources and etc. Thus, there can be some margin of error especially in the
number of major carriers in the l980s.

33 1980 was the ﬁrst year the ITU World Telecomnmnir'ution/IC T database reports a positive number of
mobile subscribers in any country — Finland.

61

the latter approach encountered a difficulty in that for some variables, for instance Gini

coefficient, latest data point was as old as 10 years. Though regressions that used latest

data were also carried out, they did not produce any significantly different results from

those that used average data. Thus. only the results of regression analyses that used

average values are presented.

4.3.2. Data sets for regression analysis

Only countries that had at least 10 data entries — in other words those with

mobile penetration data for more than 10 consecutive years (ending at either 2005 or

2006) — were included in the parameter estimation. The annual penetration data of mobile

telephony in 160 countries from 1980 to 2006 was analyzed with the three diffusion

models. Some of the countries were missing mobile penetration information for 2006.

Rather than using penetration data only until 2005 for all the countries, all the available

information was used. In other words, some countries had penetration data including that

of 2006, and the other countries had penetration data only until 2005. It was because an

additional observation likely resulted in a more reliable estimate. The parameter values

were estimated for each country based on longitudinal data. For the estimation of

diffusion related parameters, NLREG. a statistical software package for non-linear

62

regression was used.

The number of countries yielding estimated values of parameters within
acceptable range differed greatly by diffusion model. Both the coefficient of external
inﬂuence (p) and the coefficient of internal inﬂuence (q) in BDM should satisfy the
condition of 0 S p, q < I. For the logistic and the Gompertz model, there were cases for
which the estimated ceiling values did not appear to be plausible.34 In addition, such
cases were accompanied by very wide confidence interval, which means that the likely
range of the market potential is very large. Those cases were not included in the second
phase analysis. In addition, p-values of estimated parameter values varied substantially.
Countries with parameters that had p-values greater than 0.05 were excluded. Appendix I
in page 123 provides detailed information about the estimated parameter values.

Even though the p-values of parameters were usually very small, the widths of
confidence intervals (95% level) of the estimated ceiling values were quite large,
sometimes greater than 100% of the estimated value. Thus it was deemed appropriate to
further limit the number of countries in order to minimize the margin of error. For the
Gompertz and logistic models, the threshold was set at 50% of the predicted value. In

other words. if the predicted ceiling value was 100, the width of confidence interval

 

34 For some countries, the estimated ceiling was a 7 digit number. The cut—off was made at 400 (per l00
inhabitants).

63

should be less than 50 to be included in the data set. However, due to a lower number of

countries with acceptable parameter values, the threshold for BDM was set at 70%. The

parameter estimation using the BDM produced insignificant coefficient of external

inﬂuence at 95% confidence level for the most of countries. The coefficient of external

inﬂuence was significant in only nine countries. Rather than discarding results from the

BDM parameter estimation altogether, it was decided to proceed with the analyses only

on market potential and the coefficient of internal inﬂuence. Thus countries for regression

data set were selected by examining estimated values of market potential and the

coefficient of internal inﬂuence only. However, the number of countries included in the

regression analyses was further reduced due to missing data. In the end, the number of

countries included in the data sets was 49 countries for the logistic model, 34 for

Gompertz, and 30 for BDM. For the regression analyses of the three parameters, SPSS

was used.

64

Figure 3. Number of countries in data sets (total: 160)

 

Countries with unacceptable parameter values or missing data

108 Logistic

Gompertz BDM

14

 

 

 

The data sets for the logistic model and the BDM shared 27 countries. This

means that almost all the countries in the BDM data set were also included in the logistic

model data set. The data set for the Gompertz model and that of BDM shared 20

countries, meaning that one-third of countries in BDM data set were not in the data set of

the Gompertz model. On the other hand, all of the countries in the Gompertz model were

also included in the data set of the logistic model. However, this does not mean that the

two models produced similar estimated values. In terms of the correlation between

estimated ceiling values. the logistic model and the Gompertz model showed the greatest

difference.

Table 1. Estimated parameter values

 

 

 

 

 

Minimum Maximum Mean Std. Deviation
Bass diffusion model (21230)
Market potential 3.49 165.52 98.51 31.55
Coefﬁcient of internal Inﬂuence .23 .99 .59 .19
Gompertz model (n=34)
Market potential 6.00 167.13 107.71 28.77
Rate of acceptance .19 1.03 .42 .28
Logistic model (n=49)
Market potential 5.74 162.39 94.82 28.77
Rate of acceptance .29 1.56 .68 .28

 

 

The 160 countries for which mobile penetration data were analyzed in the
diffusion model phase had an average GDP per capita of $ l3,167.4, average ﬁxed line
penetration of 21.9 (lines per 100 habitants), and average mobile penetration of 50.7
(subscribers per 100 habitants) in 2005. The data set for the logistic model had an average
‘ GDP per capita of 55 18.0076, Gini coefficient of 34 and fixed line penetration of 32.6.35
The average price was $ 32.1. and it was 0.44% of the average annual income.36

Compared to the average of the full set of 160 countries, the countries included

in the phase-two data sets were substantially richer, and had a higher penetration of main

 

35 As indicated above, independent variables were measured by the average value of the past 27 years
(from 1980 to 2006). The statistics of the countries reported here are averages. not the latest data.

3" The mean absolute price of the countries in the logistic model data set is higher than those of other two
models. It was because of Bosnia and Herzegovina of which absolute price was $ 284.55. Without Bosnia
and Herzegovina. the average absolute price is $ 26.3 that accounts for 0.31% of average income.

6 (3

line telephone. This tendency was also found in the countries in the Gompertz model data
set. The data set contained countries that had even greater GDP per capita of $ 20,058. 1 ,
and fixed line penetration of 35.9 on average. On the other hand, the average price was 35
28.9 and accounted for 0.26% of the income. The data set for BDM was not an exception.
The average income was 33 18,3635. and the average penetration of main line was 33.2,
compared to 21.9 of the entire countries. The average price for the countries in the BDM
data set was 33 27.6. Detailed descriptive statistics of the variables in the three data sets

are presented in tables 3, 4, and 5 on pages 70-72.

4.3.3. Hypotheses tests

The fact that three models were used in parameter estimation and the fact that
regression analyses on estimated parameter values from the three models produced
different results poses a question of how hypotheses should be tested. A very strict way of
hypothesis testing would accept a hypothesis as supported only when all the regression
results produce supporting evidence. Naturally. this approach would lead to more
rejections. In addition, there is a problem that the patterns of diffusion of innovations
captured by the three models are not identical. It is possible that mobile diffusion data of

a country fits very well with one of the models but not with the other two. Even in

countries where all three models produced acceptable results, the estimated parameter

values from one of the models may be not as good an estimation as those from the other

two. In other words. it does not necessarily make the evidence of an independent

variable’s effect stronger when more number of regressions find significant effects. A

more appropriate approach is therefore to carry out hypotheses tests for each diffusion

model separately. This is the approach adopted in this dissertation.

 

Bases:

:Beﬁumm

Eon 5:833 E 3535:: cc— 5: $029523 252: .6 39:2: 2:.

8:538: :o Bax
332:: .382

vista.» .5653:

 

moom ...—m .o :53?~
moon ...: :0 «Ewe—xx
moon J: Lo 553?.
boom 5.:

moon 15% 25>?

NOON .3.:

meow. 155 2.53

meow 155 2.53

mean .3.:

boom .3.:

mean ..,—cam 35>»

boom .3.:

xOCN .xcwm 25>»

wooN .<_U Mxcom 155 2:03

50m .3.:
xCCm .Eqﬁwoowﬂo...

330% 2.23.2 some .6 :Etogoa 38:3 he :5: on: I _
Essa ocmswE— soap .8 :Etoqoa 83:3 :0 E3 on: I _
339:0. 2:50 some E :Eteaoa 3:53 ..o :5: on: I _

5m >._.. E3 @2333: he es

n50 E tomonogE he as

corona: E

BEBE?» 250E :0 55:5: 03:3: : :E: owi 8E: woman: ﬁne» :0 5::5: of.
.83: 53:: E :ozﬂzaoq :o as

8:8 8::— E 2:88 :o a:

3.8 55953:: 55:32

298 .5: :00 2 339:8 :8 53:53:: 355E Co 8:
EﬁtawE wEo—S .3 ﬁes—53:2. .955: Edam—smea—

mQO E 553 EogoEsEEooBB :o EoEamo>E E a:
Gauze: _::o::::2E 38.28 Ev «:93 8: ABC $23.8: and.
E96598 EU

35:52:. 9: .5: oc: E9: .5 5:83: 2;

cch I cwi WES: 897 Beta: :0 New. 52.: 83E :23 £2.93 .5 55:5: omEo><

553222.25 szwzom
553220:an osmeEA
:EEEEEEUE: NEE—am
EBE :5.me EC: .8:—

bce:0oo :o 32:30
:33: b35395:—

:EEEENED

EEEVQE 5:50

8:: 3303.:

out: 03.23.,—

555: :ozﬂzaom

~5an:6—

EoEQEoD—a EEo:ooo ..c .33
bzwscoE 0E3:—

:E:Eo:o: 0:: onE
5:59:00

«3:29 «51:23::—

 

 

3.55

 

5:35 35:88::

 

“...—ES.»

 

 

SEES.» mo m:o:m:¢o_o 3:09:25 .m 03:...

69

 

 

 

«in. 83.. SS. :8. 8:35:25“: 255:»:
2mm. 2%. 82. $5. cosﬁacozuﬁ 2:35..
£8 88. 83 e S. 83:225.“: 0:55
3m 3:. 8.2. on: o. 5.. :52 225...
R. m: 3.: E. 22532:
3.. am. _- mm.m mum- 82:55:55 me: out: 535.3—
%: 5m. $3 5 S. 8:: 2:29:
2 .N 35 5S 8.: 8E 2283
2.. _ Z. 8.: mx. Gueéée. ms: :55: do:
$.35. 8.2a 33$. :3 :25: 5:230:
R: :3 _ woman: 8.823 5.82 3&8 5: :20
xx: x _ .2 9% Q. 5:355: BE
38 5.8 8.3 35 5:52.35
3.x 3?. an? 5. _ N 520580 :6
53:50 .2m :32 E=Ec§2 E:E_E_2

 

 

 

 

 

 

Gm": .355 33:93, E5333E 55 Mo 8533 ozatomoo .m 23:.

 

 

 

38. 8%. SS. 38. 8.35.2253. 323.3.
o _ cm. 3%. 3%. :8. 83:20:02,. 32sz;
out. ems. 5%. 88. gosauscoﬁé 25cm
30 9: x _ .m on... 53358
mm. _ n 8.8 8. EN S. _ N 5282 $2525
86 as 8.: 8. 3% 9.8855
saw Few mi; 8.3. 28:. an:
m: cm. 3 8.2V 83 Eggs 35w
2: ©wa 3% G. 8F. 228,?
S. 8. m3 ~25. 8E 262$
OS ES. 8?. E. _ N 56508 26
N ﬂ .wa Sam 38% SN 22% Susana
3.2 2.2 3.3 3.3 5:35:25
8.2 3.2 3% 2; 252.28 BE
8.98.2 8&8? 8.833 3.33 2&8 an E0
5:359 .9m :32 838332 SEES—2

 

 

 

 

 

 

cum": 5.6anva 33.3? “concomouﬁ 05 HQ 859% “Sunﬂomoa .v 05¢

71

 

 

 

8NN. 83.. 3N8. 88. 8828888 258:3.
88. .88. 88. _N8. 885.8888 82:8:
8 _ N. 28. 88. 88. 88328828 255
w _ ._ v: 88 NN._ 882.58
8.8 8.8 9; ~ 8 8. _ N 5.88m $2580
8+ NM: 8.: 8. 38 888:5
8.2 8.8 8.8 8.8 38.: £2
88 5.8 8.8 8.8 8:885 5800
_ 2. ENN 8.38 S. SE 2282
E ._ 3.. :8 88. 8E 283
:8 8.8 8.8 E. _ N 83808 ED
888; 8.8m 8.88 8N 88% 8:28am
8.: 8.8 8.8. 8.8 88885
8.2 8.8 8.8 3.. 8:888 85
8.8:: 8.88.»: 8.888 5.83 5&8 .8 86
5:353. .9m :82 E:E_§..2 EEEEE

 

 

 

 

 

 

3v": dammed moist? 8385865 05 mo woumcﬁm oﬁatomoa .m 28d.

5. Empirical Findings

5.1. Phase 1: Estimation of diffusion parameters

Two parameters, market potential and the rate of acceptance, were estimated by
three diffusion models.37 The estimation used annual mobile penetration data from 1980
to 2006 for 160 countries as reported by the ITU. Each model showed a different level of
fit to mobile diffusion data.

The logistic model fit reasonably well with the data in terms of parameter values
and the significance of parameters. For 113 of the 160 countries, applying a 95%
statistical significance level. the analysis produced parameter estimates within
permissible value ranges. The average estimated ceiling value was 88.2. Even though p-
values for the ceiling estimates were very small (0.0001 or less for 90 countries), the
widths of the confidence intervals for the parameter estimates were substantial. This
width was evaluated by dividing the range of the confidence interval (maximum —
minimum) by the estimated ceiling value. It was less than 20% of the predicted ceiling
value for 49 countries: for 79 countries the width was 50%; and the average width was

86.6%. Countries for which the width of the confidence interval was below 50% were

 

37 ~ . . . . . .
for the BDM, three parameters were estimated - market potential. the coetfncnent of Internal Influence.
and the coefficient of external inﬂuence.

selected for the regression analysis whereas the others were excluded. This was a

pragmatic choice intended to limit further analysis to case where the endogenous values

were established with a level of accuracy deemed acceptable by the analyst.

In the estimation using the Gompertz model, 80 of 160 countries yielded

parameter estimates with acceptable values at the 95% significance level. Market

potential for these countries averaged at 101.8. In general, estimated ceilings were greater

than those derived using the logistic model. The widths of confidence intervals for the

market potential at 95% levels were even greater than those derived from the logistic

model. The widths were less than 50% of the predicted ceiling for 40 countries; 100% for

65 countries: and the average width for all 80 countries was 226.4%. A wider average

means that the Gompertz model performs worse than the logistic model in temis of the

number of countries of which pattern of mobile diffusion a model fits well with. However,

since diffusion patterns are not identical across countries, whether a certain model is a

good fit or not should be judged for each country.

The selection of countries for the regression analysis was made with the same

criteria to that of logistic model - the width of the confidence should be below 50% of the

predicted ceiling value.

Estimation of parameters with the (more complicated) BDM yielded fewer

acceptable parameter values. In more than half of the countries the data did not produce

parameter values that were within the range that is compatible with the theoretical

foundations of the BDM. The widths of the confidence intervals for national ceiling

values were even larger than in the other models. Furthermore, the coefficient of external

influence was below the 95% significance level for most of the countries. Only for nine

countries were the parameter values significant at the .05 level. For another five countries

they were significant at the 0.1 level. Parameters were also estimated with the OLS

technique but the results were worse than those from the analysis with NLS.

There were some countries of which current mobile penetrations are very close

or even above the predicted ceiling values. Still, the number of countries in which the

market appears to be saturated varied by the models due to the differences in predicted

ceiling values. For instance, Portugal’s mobile penetration in 2006 stood at 115.95, while

the logistic model predicted Portuguese mobile market will be saturated at 111.54, the

Gompertz 117.72, and the BDM 117.49. The logistic model data set had 21 countries of

which the estimated ceiling values were lower than the current penetration. The

Gompertz model data set had 9 countries, and the BDM data set 6.

t:-
VI

Figure 4. Distribution of the estimated ceiling values

[1 Gompertz

I Logistic

I Bass

 

0 2 4 6 8 10 12 14 16

Number of countries

Estimates generated by the logistic model predicted that more than half of the
countries (28 countries out of 49) in the data set would not reach 100% penetration if past
pattern of mobile diffusion continue to hold. The number was 15 (of 30) countries for the
BDM, and 12 (of 34) for the Gompertz model. According to the logistic model estimation,
saturation levels in 4 countries were predicted to be below 50% of the population.

In the subsets of countries for which acceptable parameter values were derived,
strong correlations among estimated ceiling values from the three models existed. This is
not unexpected and solidifies the ﬁndings somewhat. After all the diffusion patterns

depicted by the three models are not radically different from one another and one would

76

expect some level of consistency in the findings. Correlation analyses were carried out to

compare the estimated ceiling values. The estimated ceiling values for the 20 countries

that were included in all three data sets were quite similar. The Pearson correlation

coefficient between the ceiling values from BDM and the Gompertz model was .944,

between BDM and the logistic model .955, and the correlation between the logistic model

and the Gompertz model .939.

Table 6. Correlation coefficients between estimated ceiling values (n=20)

 

Models Bass Gompertz Logistic

Bass 1
Gompertz 944* l
Logistic 955* 939* 1

 

* Signiﬁcant at .01 level.

However, once we expanded the comparison to include all the countries (12:46)

for which at least one model. but not all models, produced ceiling values within

acceptable range, there were considerable differences. In other words, the additional 26

countries are those of which estimated parameters from one or two models were

‘1

insignificant. or had too wide confidence intervals. The difference was especially large

between the ceiling values from the logistic model estimation and those from the

Gompertz model (r = .719).

Table 7. Correlation coefficients between estimated ceiling values (12:46)

 

Models Bass Gompertz Logistic

Bass 1
Gompertz 926* 1
Logistic 846* .719* 1

 

* Signiﬁcant at .01 level.

5.2. Phase II: Determinants of diffusion parameters

5.2.1. Factors explaining market potential

The factors influencing the estimated market potential (as derived from the

diffusion models) were examined using regression analyses. The specification of the

empirical model was identical for all three models.

The regression model for market potential (in the chapter 4) has GDP per capita.

income inequality, price, investment, fixed line penetration, socio-cultural

fractionalization. gender inequality. urbanization. and population density as independent
variables. In addition to variables in the empirical model for market potential,
independent variable in the model for the rate of acceptance also includes mass media,
openness of economy and introductory delay as well.

Not all the factors in the theoretical model were included in the empirical model.
The theoretical model of the dissertation includes standardization as one of supply factors
that may influence market potential and the rate of acceptance. Various variables intended
to capture the degree of standardization were used, including the number of 2G standards,
the number of 2G and 3G standards, and a standardization dummy variable, but they
turned out not to be signiﬁcant. Moreover, the inclusion of standardization caused a
substantial decrease in the F -value and adjusted R2. In the ﬁnal empirical models, the
variable of standardization was therefore not included.

There was a noticeable resurgence of penetration growth in some developed
countries in recent years, which seemed to be coincided with the introduction of 3G. Even
though it was not included in the theoretical model, to check whether it was due to the
introduction of 3G service, a dummy variable was created and tested. The test, however,
did not provide any supporting evidence and the inclusion of the variable did not improve

the explanatory power of the model. One of the likely reasons for this outcome is that the

79

majority of countries in the data set have already started 3G service, making the
comparison less meaningful. The variable was therefore dropped from the analysis.

In some cases, different variables could be used to represent a factor. For
example, two model specifications were estimated throughout with regard to price —- one
with the relative price and another with the absolute price. The analysis using the absolute
price measure was carried out to acquire additional information rather than as a part of
hypothesis test.

Cross-national analyses often suffer from the problem of high correlations among
independent variables.38 Such high correlations were also present in the regression
analyses on factors that affect the rate of acceptance. As a consequence, some of the
variables had relatively high VIF39 values. For instance, GDP/capita and fixed line
penetration were strongly correlated throughout the data sets, a tendency also found in
previous research (Gruber & Verboven 2001a, Rouvinen, 2006). It was especially high in
the sets of countries derived from the logistic model (r = .899) and the Gompertz model
(r = .852) data sets. GDP per capita and fixed line penetration were also the variables

with the highest VIF values. In some cases, fixed line penetration generated a VIP value

 

38 See the appendix III on pages 127-131 for detailed information
39 Variance Inflation Factor (VIF) is an index of multi-collinearity. The index measures the increase of the
variance in a coefﬁcient because of the collinearity.

81.)

above 10. However. exclusion of one of the two did not improve the ,B-value of the other

substantially and resulted in reduced overall model fit. In addition, though the

GDP/capita and fixed line penetration is strongly correlated, we cannot conclude that

they are representing a same phenomenon. As it is reported below, both of the variables

were found to significantly affect market potential predicted by the logistic model, even

with the presence of multi-collinearity. Thus, it was decided to keep both of the variables

absent a better alternative specification. Other variables that had relatively high VIF

values - though not so high as to raise a serious concern - were mass media (measured by

percentage of households with TV set), ethnic and linguistic fractionalization.

Summary results of the regression analyses are presented in table 8 on page 83,

table 9, and 10 on pages 85-86. Each regression analysis produced slightly different

outcomes. In the analysis using the estimated ceiling from the logistic model, GDP per

capita ([3 = .544, p = .026) as well as fixed line penetration (,8 = - .518, p = .057) showed

significant effects despite the problem of multi-collinearity (VIF = 5.938 for GDP/capita,

and 7.559 for fixed line). The effect of the latter variable was negative contrary to the

hypothesis. The other two independent variables that were found to have significant

effects were urbanization (,6 = .414, p: .01 l) and relative price ([3 = -.430, p: .004), with

each variable showing the expected direction of the effect. For purposes of comparison, a

81

model specification using the absolute rather than the relative price was also examined.
The outcome was almost identical to that of the relative price specification. An exception
was that gender inequality was significant, albeit barely (,8 = .237, p = .094) in the
analysis using the absolute price specification. The other variables, including inequality
measures, did not show any significant effect on estimated market potential.

Using the parameter estimates from the Gompertz model as dependent variable
revealed a somewhat different pattern. GDP per capita (ﬂ = .527, p = .06) and
urbanization (ﬂ = .402. p = .029) were found to have significant positive effects on the
ceiling, while gender inequality (,6 = .308, p: .1)40 and price (p = - .48, p: .012)
negatively affected market potential. Fixed line penetration was not significant (,8 = .477,
p = .13), possibly due to a high level of multi-collinearity with GDP per capita. On the
other hand, in the absolute price specification, GDP per capita ([3 = .732, p = .021) and
urbanization (ﬂ = .522, p = .008) were found to have significant positive effects with the
ceiling. Gender inequality negatively affected the estimated ceiling value. However,

unlike the relative price. the absolute price was not significant.

 

40 . j . . . . . . .
Because gender inequality was measured by the ratio of female in labor force. the positive value ofthe
standardized [i should be interpreted as a negative effect of gender inequality.

0C
(\3

Table 8. Factors inﬂuencing estimated ceiling values (BDM) (n=30)

 

Relative price

Absolute price

 

speciﬁcation speciﬁcation
Model (D.V. = market
potential)
F-value 5.698*** 4.889***
Adj. R3 .641 .596
Variables Standardized ,6 Standardized ,6
GDP/capita .756** .725**
Gini coeﬂieient .155 .150
F ixed Penetratirm. - .399 - .120
Population Density .179 .198
Urbanization .238 .349*
Price -.379 -.070
Investment .321* .366**
Female labor force .291* .226
F meti onal i:ati on
- Ethnicity - .329 - .258
- Language .464* .351
- Religion - .042 - .096

 

* significant at 0.1 level, ** significant at 0.05 level, *** significant at 0.01 level

83

Results from the analysis using the parameter estimates from the BDM as

dependent variable differ from the logistic and Gompertz findings. GDP per capita ([3

= .756, p: .011), investment ([3 = .321, p: .056), and linguistic fractionalization ([3 = .464,

p: .099) were found to be positively related to estimated market potential. Gender

inequality (1): .099) was found to be negatively related to ceiling values. If the absolute

price specification was used, GDP per capita (ﬂ = .725, p: .032) and investment (,8 = .366,

‘3
II

.044) had significant effects. In addition, urbanization was significant (,8 = .349,

p: .095). However. gender inequality (p: .207) and linguistic fractionalization (p: .213)

were no longer significant.

To summarize. GDP per capita had a significant effect on the estimated market

potential in all three models. Urbanization and (relative) price had significant effects on

estimated ceiling values in the logistic and the Gompertz models. Gender inequality had a

negative effect on estimated market potential in the BDM and the Gompertz model. Fixed

line penetration was significant only in the diffusion pattern predicted by the logistic

model. Investment was significantly associated with market potential predicted by the

BDM.

84

Table 9. Factors influencing estimated ceiling values (Gompertz model)

Number of countries = 34

 

Relative price Absolute price

specification specification
5

Model (D.V. = market

potential)

F -value 3.825*** 2.828**
Adj. R3 .485 .379
Variables Standardized [i Standardized ,8
GDP/capita .527* .732**
Gini coefﬁcient .022 .027
F ixed Penetration. - .477 - .562
Population Density .085 .074
Urbanization .402** .522**
Price - .480** - .271
Investment .133 .012
Female labor force .308* .417**

F racti onal i :ati on

— Ethnicity -.082 -.038
- language .144 .036
- Religion -. 120 -. 107

 

* significant at 0.1 level, ** signiﬁcant at 0.05 level, *** significant at 0.01 level

00
1,11

Table 10. Factors inﬂuencing estimated ceiling values (Logistic model)

Number of countries = 49

 

 

Relative price Absolute price
specification specification
Model (D.V. = market
potential)
F—value 6.641*** 5.569***
Adj. R3 .569 .517
Variables
GDP/capita .544** .703***
Gini coeﬂicient - .041 - .041
F ixed Penetration. - .518* - .480*
Population Density .044 .033
Urbanization .414** .501***
Price - .430*** - .267**
Investment .060 .073
Female labor force .21 l .237*
F racti onalizati on
- Ethnicity - .032 -0.41
- language .026 - .027
- Religion -.099 -.1 l7

 

* significant at 0.1 level, ** significant at 0.05 level, *** significant at 0.01 level

88

5.2.2. Test of hypotheses related to market potential

Table 1 1. Tests of hypotheses relating to market potential

 

Hypothesis Variable BDM Gompertz Logistic
m
H. 1-1 Income X X X

1-2 Income inequality
1-3 Price X X
1-4 Investment X
Fixed penetration
. 1-6 Urbanization X X
. 1-7 Population Density
. 1-8 Gender Inequality X X

 

 

:mzmzzza
'31.

. 1-9 Fractionalization

 

 

 

 

 

X = supported

Table l 1 above summarizes the results of the hypotheses tests. In the regression

on the ceiling values estimated by the logistic model, GDP per capita, fixed line

penetration. urbanization, and relative price significantly affected market potential.

However, contrary to hypothesis 1-5, fixed line penetration had a negative effect. In the

regression using the absolute price specification, absolute price turned out to be

significant. Thus, hypotheses l-l (income), 1-3 (price). and 1-6 (urbanization) were

supported in the regression analysis on market potential estimated by the logistic model.

‘1

(I)

Hypotheses 1-2 (income inequality), 1-4 (investment), 1-5 (fixed line penetration), 1-7
(population density), 18 (gender inequality), and 1-9 (fractionalization) were not
supponed.

The regression on the market potential values estimated by the Gompertz model
found that GDP per capita. urbanization, gender inequality, and relative price had
significant inﬂuences on the saturation level. Thus, hypotheses l-l (income), 1-3 (price),
and 1-6 (urbanization), 1—8 (gender inequality) were supported in the regression analysis
on market potential estimated by the Gompertz model.

The analysis on ceiling values from BDM found that GDP per capita, investment,
and gender inequality were significantly associated with the estimated market potential.
Thus. hypotheses 1-1 (income). 1-4 (investment), and 1-8 (gender inequality) were

supponed.

5.2.3. Factors influencing the rate of acceptance

Tables 12 and 13 on pages 90 and 91 summarize the results of the regression
analyses using the estimated rates of acceptance as dependent variable. The overall
regression analysis using the estimated rates of acceptance by the logistic model, was

significant (F = 4.732. adj. R2 =.521, p = 0.001 ). Income inequality (3 = -.275, p=.061)

and religious fractionalization ([3 = -.335. p=.021) negatively affected the rate of
acceptance. Introductory delay ([3 = .493, p=.004) had a positive effect. Gender inequality
was also found to be significant (ﬂ = -.267, p=.08) but had positive effects contrary to
expectation. Additional analysis with absolute price specification showed a similar
outcome. The model using absolute prices was also significant (F = 4.57, adj. R2 =.51, p
= 0.001). Income inequality was found to slow down the adoption ([3 = - .281,p=.06l).
The results seems to indicate that introductory delay sped up adoption (5 = .575, p=.003).
Gender inequality showed a significant positive effect ([3 = -.285. p: .65). Lastly,
fractionalization measure of religion was found to be negatively associated with the

estimated rate of adjustment ([3 = -.322, p=.021).

89

Table 12. Factors influencing the estimated rate of acceptance (Gompertz model)

Number of countries = 34

 

 

Relative price Absolute price
specification specification
Model (D.V. = rate of
acceptance)
F -value 3.518*** 3.518***
Adj. R3 .517 .517
Variables
GDP/capita - .189 - .189
Gini coeﬁ‘icient - .410** - .409**
Fixed Penetration. - .072 — .059
Population Density - .004 - .011
-Urbanization - .209 - .21 1
Price .026 .023
Competition .236 .234
Introductory Delay .452*** .452***
Openness of Economy - .072 - .061
Female labor force - .517** - .523**
Mass Media .041 .028
Fractional izati on
- Ethnicity - .137 - .132
- Language .312 .305
- Religion - .191 - .185

 

* significant at 0.1 level, ** significant at 0.05 level, *** significant at 0.01 level

90

Table 13. Factors influencing the estimated rate of acceptance (Logistic model)

Number of countries = 49

 

 

Relative price Absolute price
specification specification
Model (D.V. = rate of
acceptance)
F -value 4.732*** 4.570***
Adj. R3 .521 .510
Variables
GDP/capita - .031 -.085
Gini coeﬁ‘icient - .275* - .281*
F ixed Penetration. - .202 - .217
Population Density .101 .093
Urbanization .001 -.020
Price .187 .112
Competition .074 .080
Introductory Delay .569*** .575***
Openness of Economy - .1 l8 - .108
Female laborforce - .267* - 285*
Mass Media .137 .125
F racti onal i za ti on
- Ethnicity .090 .092
- language. - .161 - .154
- Religion - .335** - .322**

 

* significant at 0.1 level, ** significant at 0.05 level, *** significant at 0.01 level

91

Regression analyses of the factors influencing the rate of acceptance from the
Gompertz model showed similar results. The analysis found the model significant (F-
3.518. adj. R2 =.517. p = 0.006). Introductory delay (ﬂ = .452. p=.049) positively affected
rate of acceptance. Gini coefficient (,6 = - .410, p=.043) and ratio of female in labor force
(ﬂ = - .517, p=.022) were negatively associated with the estimated parameter value.
Additional analysis with absolute price measure also found a very similar result. The
model using the absolute price had the same F—value and Adjusted R2 with the model
using the relative price. The same variables were found to have significant effects as the
previous analysis, though standardized ,8 values and p-values were slightly different.

The Bass diffusion model has two parameters that together determine the rate of
acceptance: the coefficient of external influence (q) and the coefficient of internal
inﬂuence (p). Most of the estimated values for the coefficient of external inﬂuence were
insignificant at 95% level. The number of countries to which the model fit was too low to
ensure sufficient degrees of freedom to run a regression. Therefore, regression analysis
was run only on the coefficient of internal inﬂuence. The regression model for the
coefficient of internal influence was not significant overall. Again, the model proved not

to be significant.

QC)
[\3

5.2.4. Test of hypotheses related to the rate of acceptance

Here, the hypotheses were reviewed in the same way that the hypotheses related

to market potential were tested. Table 14 on page 94 provides a summary of the tests.

The regression analysis on the rate of acceptance estimated by the logistic model

found that the Gini coefficient, introductory delay, and religious fractionalization had

significant effects. Thus, hypothesis 2-5 (income inequality) and hypothesis 2-12

(introductory delay) were supported by the diffusion pattern predicted by the logistic

model. Hypotheses 2-1 (income), 2-2 (fixed line penetration), 2-3 (price), 2-4

(competition), 2—6 (fractionalization), 2-7 (gender inequality), 2—8 (population density), 2-

9 (urbanization). 2—10 (mass media), and 2-1 1 (openness of economy) were not supported.

The analysis with the estimated rate of acceptance derived from the Gompertz

model found that the Gini coefficient. introductory delay, and gender inequality had

significant effects. No other variables were found to be significantly associated with the

rate of growth. Thus, hypothesis 2—5 (income inequality) and hypothesis 2—12

(introductory delay) were supported by the diffusion pattern predicted by the Gompertz

model.

93

Table 14. Results of the rate of acceptance related hypothesis testing

 

 

 

 

 

 

 

Hypothesis Variable Gompertz Logistic

H. 2-1 Income

H. 2—2 Fixed penetration

H. 2-3 Price

H. 2-4 Competition

H. 2-5 Income inequality X X

H. 2-6 Fractionalization

H. 2-7 Gender Inequality

H. 2-8 Population density

H. 29 Urbanization

H. 2-10 Mass media

H. 2-11 Openness Economy

H. 2-12 Introductory Delay X X
X = supported

04

6. Discussion

6.1. Discussion of findings

This dissertation started with a question about the effects of inequality on the
diffusion of mobile telephony. To study this question, the role of inequality among
other factors that influence the shape of diffusion processes had to be examined. The
dissertation set out to develop a general model of the factors that shape the diffusion
of mobile telephony. exploring the effects of inequality in the context of other
variables that can reasonably be expected to influence diffusion. As several of these
factors are specific to nations, 3 cross-national study design seemed appropriate. In a
first step. estimates for the market potential and the rate of acceptance were generated
using three widely used diffusion models. In a second step, regression analyses were
performed to examine the factors that inﬂuence the characteristics of the diffusion
process.

As ceiling values for mobile service adoption can only be observed in
countries that have actually reached saturation, statistical models of the diffusion of
innovations were used to generate estimates for all other countries. Whereas a
comparative evaluation of the alternative diffusion models was not the primary goal

of this paper, the outcomes, nevertheless, deserve a brief discussion. In terms of the

VA
\f‘
V

number of countries for which the analyses produced acceptable parameter values and

acceptable confidence intervals, the logistic model ﬁt the data best. In other words,

there are more countries in which mobile diffusion showed patterns that are best

approximated by a logistic function rather than one of the others. There are two

distinct differences between a logistic function and a Gompertz function: the graph of

the latter is more skewed to the right; and its inﬂection point is at one-third of the

ceiling compared to the half point for the logistic function (Dixon, 1980). The data

seems to suggest that mobile diffusion has a more symmetric pattern. This is further

supported by the fact that the majority of countries that were included in the analysis

were developed countries. In these countries. it took many years before the market

started to expand more rapidly with the introduction of 2G services. Diffusion curves

for these countries indeed have long right tails. However, the lengthy left-tails

resulted from the many years of slow growth offset them, resulting in an overall more

symmetric diffusion pattern.

Numerous studies employing the Bass diffusion model are testimony to its

usefulness. Nonetheless, it has inherent methodological weaknesses. As Meade and

Islam (2006) pointed out, the number of data points upon which the estimation is

based seems to affect the performance of the model. in particular whether estimated

parameter values are within the acceptable range. On average. in the countries for

96

which the Bass model yielded acceptable (but not necessarily statistically signiﬁcant)
parameter values. mobile telephony was introduced 18 years ago. In countries where
the Bass model did not yield acceptable parameters, mobile services had been
available for only 15 years. In the fourteen countries for which the values for the
coefficient of external inﬂuence are within the acceptable range (p S .1). diffusion
patterns appear to be smooth without any sudden slow down or take off. In other
words, mobile diffusion in those countries resembles an ideal S-shaped path compared
to other countries where sudden breaks and accelerations are visible.

All data sets contained more developed countries than developing countries.
In fact. the number of low income countries was too small to run an independent
regression.“ There are a couple of possible explanations as to why only a small
number of low income countries were included in the data sets. First, it may be simply
because the diffusion patterns described by the three models do not ﬁt the diffusion
pattern of mobile telephony in the low income countries. A future study with an
alternative statistical model of diffusion of innovations would be able to answer
whether such is the case. Another possibility is that as low income countries
introduced later than developed countries. smaller number of data points may have

caused estimated parameter values to have large variance. In addition, in many of

 

4' It was true even to the data set for the logistic model that had the largest number of countries. In the
logistic model data set. there were 32 high-income countries. 6 upper middle income, 9 lower-middle
income. and only 2 low-income countries according to the income classiﬁcation used by the ITU.

£17

countries with a relatively short history of mobile telephony. diffusion may not have
reached the inﬂection point yet. If this is the case, it could be an additional cause of
greater variance in parameter values. Those tendencies may have resulted in the
inclusion of fewer low-income countries that met the dual threshold of estimated
ceiling values that satisfied the conceptual model requirements and the additional
selection criterion of confidence intervals whose width was below 50% of the
predicted ceiling value. Subsequent studies which can take advantage of more
matured mobile mtu'kets may be able to include a higher number of low-income
countries.

Even though the ceiling values derived from the three diffusion models were
highly correlated. there were still differences. These differences indicate that the
patterns of diffusion of innovations represented by the models are different. Testing
hypotheses for each model separately was, therefore, a plausible approach.

The estimated ceiling values were below 100 (We) in some countries. The
BDM and the logistic model predicted that about half of the countries in the data sets
would not be able to achieve universal service42 if past pattem of mobile telephony

diffusion remains unchanged. The Gompertz model also predicted that about one third

 

43 How we define universal service for mobile telephony is an open question. For main line telephony.
which is designed to be used by family, universal service is reached when every household subscribes
to the service. ()n the other hand. mobile telephony is a technology developed to be used by an
individual. This dissertation assumes that the universal service for mobile telephony is achieved when
every individual has a subscription.

9 8

of the countries in the data set would not reach a saturation level of 100%“. The

ﬁndings suggest that regulatory efforts will be necessary to make universal service a

realistic goal in some countries.

The analyses did not reveal any signiﬁcant relationship between market

potential and income inequality. This is compatible with the findings of a previous

study (Talukdar et al., 2002) that, in an empirical study of six products, also failed to

ﬁnd evidence of such a relationship. However, a signiﬁcant effect of income

inequality on the rate of acceptance (and its equivalent parameters) was detected in

our study. Several factors may explain that the diffusion ceiling is apparently not

inﬂuenced by inequality. Barrantes and Galperin (2008) reported that the poor in

developing countries are willing to spend considerably more on mobile than their

counterparts in developed countries. The higher propensity to spend for mobile

service may compensate the (ceteris paribus) reduction of demand caused by lower

income. In addition. innovations such as low-cost hand sets and prepaid plans have

lowered the cost of mobile service signiﬁcantly. Prepaid plans are not cheaper than

postpaid plans but have lower or no ﬁxed costs and provide users with more control

over variable expenses. Therefore, they allow ﬁnancially constrained customers to

purchase mobile service (Samarajiva, 2006; Barrantes & Galperin, 2008). Even

 

43 Certain customers may elect not to purchase mobile service even though they can afford to do so
simply because they do not want to. Thus, universal service may be achieved at below 100%
penetration. On the other hand. since the ITU reports the number of subscribers than users, 100%
penetration may not mean every individual subscribes to a mobile service.

91')

though the expense for a mobile service may take a large share of subscribers” income
in developing countries. our findings seem to suggest that the cost may have been
reduced sufficiently to neutralize any effect of income inequality on expected
saturation levels.

Another possibility is that higher consumption by the rich may compensate
for lower consumption by poor people. Traditional diffusion models study the initial,
ﬁrst adoption decision in favor of a new product. This is a unavoidable limitation
since if repetitive purchase were to be included the model, it would cease to be a
model of diffusion of innovations but become a demand model for a product, which
could follow any shape. However. it is not uncommon for some users to have more
than one subscription. especially in developed countries. It may be possible that in
some of the countries rich individual’s tendency to own multiple subscriptions makes
up for the lower demand due to the inability to pay of the poor segments of the
population. There is no available information. however. to verify this hypothesis. Due
to the fact that mobile penetration data report the number of subscribers rather than
the number of owners. multiple subscriptions of a single person cannot be
distinguished from single subscriptions of as many persons.

One could interpret the current data format of mobile penetration to contain

two different innovations — mobile telephone service for business and that for private

1 ()0

use. In fact. diffusion of innovations research has investigated business ﬁxed

telephone service and residential ﬁxed telephone service separately (i.e., Islam &

Meade, 1996). Cellular service for business use and that for private use can be

regarded as two different innovations in that they are purchased and used for different

purposes. In addition, different factors would inﬂuence adoption: for business phones,

it would depend on the type of business, available budget. company policy towards

telecommunications services and so on. Some of factors in this dissertation many not

inﬂuence the adoption of business mobile telephone service. However. there is no

available cross-national data that distinguishes two types of usage.

A third possibility is that the relationship between market potential and

income inequality is more complicated than captured in the analysis. First, income

inequality may work differently depending on the level of development. Some studies

(Senauer, 1990; Park et al., 1996; lyigun & Owen; 2004) found differences in the

relationships between income inequality and consumption in developing and

developed countries. If the effect of income inequality is less pronounced in

developed countries, the fact that the data set is skewed towards developed countries

may have weakened the effect. Unfortunately, the number of low income countries in

the data set was too small to run independent regressions on subsets of countries. A

future study with a sufﬁcient number of developing countries in the data set may be

101

able to provide an answer.

In addition. there is a possibility that the relationship between the two may

not be linear. Wei (2001) reported from surveys on mobile non-adopters conducted

two years apart that income had become less inﬂuential as the market approached

saturation. Thus, it is possible that the relationship may differ depending on the

maturity of the mobile market. Secondly, the relationship may be indirect. Income

inequality may have an influence on market potential not identiﬁed by the one-step

regression model in this dissertation. A path model might be an alternative

speciﬁcation to explore such indirect influences. This dissertation did not investigate

the question further because it would have required a different theoretical and

methodological approach, which would have made another dissertation.

Price for mobile service had a statistically signiﬁcant inﬂuence on the ceiling

values. Some customers choose not to subscribe to the service because the price was

too high compared to their income and purchasing power. Even though mobile service

price is likely to continue to drop, so far it apparently has been too expensive for some

potential adopters. The signiﬁcance of the effects of price as well as that of income

reafﬁrms that the decision to purchase mobile service depends on the income level

and the cost of the service compared to income.

Another ﬁnding is a negative relationship between mainline penetration and

market potential. The outcome suggests that in the set of countries that were included
in the analysis mobile telephony is a substitute to fixed line service. A previous study
by Mariscal and Rivera (2006) found that in developing countries mobile telephony is
mainly adopted as a substitute for mainline telephony. The finding in the dissertation
is compatible with Young‘s (1993) econometric model as well as Sawhney’s
infrastructure development model (2003) in which a new product functions as a
complement in the early stage of adoption and gradually becomes a substitute as the
market matures.44 The cellular markets in developing countries may not have
matured yet. but the relationship between the mobile and the ﬁxed line telephony was
mainly substitutes even at the early stages of diffusion in developing countries. Before
the introduction of mobile telephony, many developing countries, especially poorer
ones, had a large pool of unmet demand for telephone service that either mobile or
ﬁxed voice service could satisfy. In this situation. ﬁxed and mobile telephony
compete for a new subscriber from the very introduction of cellular service. making
cellular service a substitute to ﬁxed line.
The level of urbanization had a signiﬁcant positive effect on market potential

in both the Gompertz and logistic models. while population density did not. In

addition. the analysis with the absolute price using the BDM produced ceiling values

 

4.; . . . . . .
However. Sawhney saw the relationship between mainline and mobile telephony as an exception
and predicted that they would coexist in complementary relationship.

103

also found a significant effect of urbanization on market potential. Economies of
density were one of the reasons why the two variables were hypothesized to affect the
ceiling values. The insigniﬁcant association between population density and the
ceiling value may imply that the reason why higher levels of urbanization increase the
ceiling values may not be related to cost effectiveness.45 In addition, neither variable
had a statistically significant effect on the rate of acceptance. The rate of acceptance
in an epidemic model is related to the chance of social interaction between adopters
and potential adopters. through which information is transferred and social pressure is
exercised. A higher rate of acceptance can be interpreted as either greater chance of
the interaction or greater effectiveness of the interaction, or both. Greater social
pressure would positively affect market potential because it would increase the chance
of adoption of a potential adopter who is ﬁnancially capable of the purchase but
would chose not to otherwise.

The ﬁnding that neither population density nor the level of urbanization is
significantly related to the rate of acceptance seems to suggest that the increased
market potential is not due to greater social pressure either. A possible answer is that

the urban population may have a higher preference for mobile communication due to

 

45 It is still possible that level of urbanization is signiﬁcantly related to market potential due to
economics ofdensity. A country may have a high level of urbanization but relatively low population
density. In such a case, mobile service providers can still beneﬁt from economies of density at least in
the urban areas where the majority of the population is. The ﬁndings only suggest that the signiﬁcance
of level of urbanization is less likely because of economies of density but do not eliminate the
possibility.

101

their mobile life style. Also it may be because there are more people in urban areas
who are likely to have favorable attitudes towards an innovation because of higher
education, wealth. and social status.

One interesting observation was that estimates for Eastern European countries,
including countries that once belonged to the former Soviet Union, revealed relatively
high ceiling values compared to the countries with similar levels of economic
development in other regions. The average ceiling value estimated from the logistic
model for those (seven) countries was 108.8 ‘70 (of the population) while the average
for the entire data set was only 93.3 %. This observation may be another indication
that low availability of ﬁxed line telephone service increases the market potential for
cellular service. Gruber (2001) found that long waitlists for mainline telephone
service - in other words: the inability of the incumbent telecommunications carrier to
meet effective demand - was one of the reasons for the fast growth of mobile
subscriptions in Eastern Europe. In addition, competition contributed to faster market
expansion.

The negative association between income inequality and the rate of
acceptance suggested by theoretical considerations was found in both the logistic and
Gompertz models. The effect may be due to economic and socio-cultural reasons. In a

society with very unequal income distribution. there will be more people whose

105

threshold prices are relatively low. At any given price level there will be fewer

subscribers compared to a more egalitarian society. If the price decreases at similar

speeds across countries, it would take more time for a society with high income

inequality to reach a certain level of subscribers than for a country with low income

inequality. On the other hand, high income inequality. usually coupled with other

types of inequality, may be one of the underlying causes of tensions and conﬂicts

between social groups that disrupt economic development and communication among

the groups. This in turn negatively affects growth of the mobile market and the flow

of information about the new product.

Introductory delay was positively related to the rate of growth. This

corroborates ﬁndings of previous studies (Gruber & Verboven, 2001a; 2001b) that

late-starter countries catch up with early adopters — not in terms of the saturation level

but in terms of the time it takes for the market to become saturated. The authors

suggested that learning from the experience of other countries reduces uncertainty for

both investors and customers and consequently increases the chance of adoption.

Another beneﬁt of later introduction of the service is that the risk of technological

ﬂaws will likely be reduced. Takada and Jain (1991) reported a similar tendency on a

study of diffusion of general consumer durables such as car, TV, and washing machine.

Gender inequality exhibited a significant effect on market potential in the

106

BDM and Gompertz models. Though it was not statistically signiﬁcant. gender
inequality also had the expected effect in the logistic model. More active participation
in economic activity means that more women have independent income sources,
hence more freedom to purchase products for their needs. In addition, working
women may have a higher demand for cellular service than housewives. However,
gender inequality was positively associated with the rate of acceptance in the
Gompertz and the logistic models. Thus. the ﬁndings suggest that a higher ratio of
females in labor force increases the saturation level but slows down the adoption. It
may be the case that a higher ratio of female in labor force increases purchasing
power of households so that potential demand increases. However, the increased
income may not result in faster adoption, which agrees to the insigniﬁcance effect of
GDP per capita on the rate of acceptance in this dissertation. Access at the workplace
to phone service may explain why increased female participation in the labor force
does not result in faster adoption of cellular service. On the other hand, it is possible
that the findings are simply statistical artifacts or due to an unidentiﬁed relationship
between the ratio of female in labor force and other measures of independent variable.
Subsequent research may be able to shed more light on that issue.

Socio-cultural fractionalization produced few statistically signiﬁcant results.

Exceptions were religious fractionalization in the regression explaining the rate of

107

acceptance derived from the logistic model, and linguistic fractionalization in the

regression explaining the ceiling value gained from the BDM model. There are two

possible interpretations of the overall insigniﬁcance of the fractionalization measure.

First, it may simply be because the degree of fractionalization does not have a direct

effect on the diffusion of mobile telephony. Second. increased chances of social

tension and conﬂict may not necessarily result in a signiﬁcant negative effect on the

economy. The fractionalization measures are not identical with inequality measures,

as was elaborated earlier. Fractionalization only increases the chance of socio-cultural

inequality. This dissertation theorized that high inequality may be harmful for the

growth of mobile market because (1) it may cause social tension and conﬂicts; which

(2) in turn may be detrimental to the economy: and (3) interrupt the ﬂow of

information between social groups. Our observations suggest that even though high

inequality in a society may increase social tensions, it may not reach a level that will

significantly affect economic processes in mobile markets.

108

7. Concluding Remarks

7.1. Implications

This dissertation developed a theoretical model building on and expanding an
approach used by Talukdar et. al. (2002). A major modification was to shift the focus
from an individual’s characteristics and capabilities to socio-economic elements. The
shift enabled the inclusion of economic factors that inﬂuence supply conditions as
well as socio-cultural factors, both of which are not easily ﬁt into an individual-based
theoretical model. The findings suggest that supply conditions are as important as
demand factors in understanding the diffusion of mobile telephony.

With regard to the diffusion literature the ﬁndings of this paper suggest that a
deeper understanding of the diffusion of innovations in mobile service requires
considering both service providers’ abilities to create and meet the demand as well as
consumers' ability to pay. This dissertation provided some evidence that income
inequality affects mobile telephony diffusion. It also provided additional empirical
evidence of the effects of variables. such as ﬁxed line penetration, that have been
tested in previous studies.

The dependent variables examined in the dissertation reﬂect two important

aspects of universal service. The market potential is directly related to the question of

109

whether universal service is a feasible goal if the market is left to itself. The rate of

acceptance is related to how fast universal service will be achieved (if at all). The

variables that had a signiﬁcant influence on the ceiling are income (all three models),

price. level of urbanization (logistic model and the Gompertz model), gender

inequality (Gompertz and BDM), fixed line penetration (logistic only), and

investment (BDM only). It was income inequality, introductory delay (logistic model

and the Gompertz model), and gender inequality (Gompertz model only) for the rate

of acceptance. For policy makers, the signiﬁcance of the price variable on market

potential suggests that they should target their regulatory efforts towards provision of

inexpensive calling plans in order to achieve universal access. For businesses the

ﬁndings in the dissertation provide insight into additional factors to be considered in a

market forecasts. such as how active the female population is in the economy.

7.2. Limitations and Questions for Future Study

One potential limitation of the methodological approach employed in this

dissertation is that the values of the dependent variables had to be estimated, which

means the reliability of ﬁndings in the dissertation depends on the accuracy of

estimated parameter values. In addition, the estimated values of the market potential

had relatively wide confidence intervals in many cases. An alternative is to assume

110

that a group ofcountries has identical market potentials. which contradicts the

empirical evidence found in this dissertation. Another alternative is to determine the

ceiling values outside of the diffusion model although no straightforward methods

exist to do so.

It may be interesting to explore the applicability of a ‘critical mass’

phenomenon in the context of mobile telephony in more detail. Mahler and Rogers

(1999) and Rogers (2003) indicated that the diffusion of an interactive medium

generally requires more time before reaching critical mass compared to other new

products. Once the critical mass is reached, it tends to a show faster rate of diffusion,

resulting in a steeper S-shaped pattern. Lim et al. (2003) reported that the late take-off

of interactive media diffusion is because potential adopters have a relatively

homogenous threshold value. However, whether such a tendency can be applied to

mobile telephony is unclear. Mahler and Rogers (1999) and Rogers (2003), for

example. argued that mobile telephony has an advantage of already reached critical

mass because there is already established ﬁxed line customer base. If their argument

were true, mobile telephony should require less time reaching critical mass. showing

relatively faster rate of diffusion at early stage of diffusion. compared to other

interactive media. A future study that compares diffusion patterns of mobile telephony

to those of other interactive media may be able to provide an answer to this open

111

question.

Because countries were included in the data sets only if they showed mobile
diffusion patterns that could be explained with one of the three statistical diffusion
models. the data sets are not random drawings from the population. The data sets
contained more developed countries than developing countries. resulting in higher
average GDP per capita, fixed line penetration, and years of adoption. There were not
enough developing countries to run independent regressions on a subset representing
only low income nations. This prevented testing whether income inequality had
different effects dependent on the level of economic development. A future study may
be able to work with a broader data set once the diffusion of mobile telephony in
developing countries reaches a more mature stage and better data is available.

The dissertation has limitations in allowing the generalization of the ﬁndings
to all countries. The inclusion of countries in the data sets largely relied on parameter
estimation. A country was not included in a data set when estimated parameters had a
very wide confidence interval. was insigniﬁcant. or was out of acceptable value range.
Such results mean that the particular model may not be a good representation of
mobile diffusion pattern in that country.46 It further means that ﬁndings from the

regression analyses with the data set are not applicable to a country of which data the

 

4b .. . . . . . . .. .

A more delinite judgment Will have to wait until there is sullicient data because model may have not
appeared to fit the data well due to small number of data points that may have caused greater variances
parameter values.

particular model did not fit. The result from each regression analysis should be

considered as applicable only to the countries in corresponding data set and cannot be

generalized to the countries that. were left out from the data set.

In terms of the relationship between income inequality and demand, this

dissertation provides additional insights to the literature. We hypothesized a direct

relationship between income inequality and the diffusion of mobile telephony

diffusion. At this point in time, there is no theoretical and empirical information

available that would suggest a different relation. Thus, the dissertation can be seen

as a ﬁrst step in exploring the phenomenon and future studies could expand the

analysis to investigate additional channels of inﬂuence.

Income inequality was measured by the Gini coefficient. The Gini coefficient

is not the only metric for income inequality nor is it free of limitations. As any

constructed measure, it necessarily relies on assumptions and has limitations. This

dissertation used the Gini coefﬁcient because it is most widely available. Future

studies may try to employ altemative speciﬁcations of income inequality,

This study also included socio-cultural inequality under the assumptions that:

(l) diffusion of mobile telephony is a social as well as economic phenomenon

influenced not only by economic factors but also by socio-cultural factor would have

influences; (2) high income inequality usually accompanies other types of inequality,

113

compounding the effect of each other. This paper is one of the few studies that have

attempted to include socio-cultural inequality. Future studies of the relationship

between socio-cultural inequality and diffusion of mobile telephony could reﬁne the

current approach. The related hypotheses were built on assumptions that were not

empirically tested. In addition. fractionalization measure may not have been the best

measure to represent socio-cultural inequality. A “right” measure for socio-cultural

inequality is difficult to find. A previous study (Dekimpe et al., 1998) used the crude

death rate as the indicator, which may capture some of the phenomenon but is closer

to a national health care measure. As an alternative, this dissertation used a

fractionalization measure. Fractionalization is not identical with socio-cultural

inequality but is one of the strongest causes of the phenomenon. In addition, it was a

measure that led to the largest number of countries in the data set. There are other

alternatives such as HDI (Human Development Index). GDI (Gender Development

index) that have been reported by the United Nations Development Program (UNDP)

in the annual Human Development Report. They are also not free from problems in

that GDP per capita is one of the components of the indices and data is missing for

many countries. Future studies may therefore attempt to use alternative socio-cultural

inequality measures.

This dissertation focused on investigation of the effect of socio-economic

factors on the diffusion of mobile telephony. Another avenue for future studies may be

to include an explicit feedback effect from mobile telephone diffusion to socio-

economic factors. Future research on the effect of the mobile telephony industry not

only on economic development btit also on other socio-cultural factors would also

contribute to a better understanding of mobile telephony diffusion. In addition. such

research will be able to provide a better understanding of the benefits that mobile

telephony can bring to people in need.

The approach this dissertation employed is only one of several possible ways to

investigate mobile telephony diffusion. Statistical research necessarily abstracts from

the ﬁne-grained historical contexts or unique socio-cultural environments in which an

innovation is embedded and the implications of these factors on the diffusion process.

It only gives a higher-level understanding of how mobile telephony is actually being

adopted and used in poorer countries, though there are scattered evidence that suggest

the pattern may be different from that in developed countries. Arnould (1989)

suggested that a novel consumer good of Western origin goes through a process of

reordering. reconstruction and resignifying in the process of adoption in materially

deprived society. Research using different approaches. for example iii-depth case

studies or ethnographic studies, may be able to provide complementary understanding.

The dissertation investigated diffusion of mobile telephony. which has shown

115

a remarkably fast market expansion, especially in developing countries. Contrary to

optimistic expectations, the findings suggest that universal service through mobile

telephony may not be a realistic goal in some countries if mobile telephony continues

to follow the diffusion patterns exhibited so far. It means that government intervention

may be necessary in those countries. The ﬁndings also suggest that in order to achieve

universal service sooner, regulatory efforts to provide inexpensive calling plans to the

poor might be useful.

110

Appendix I. Estimated ceilings and rates of acceptance

 

Country

Algeria
Australia
Austria
Belarus
Belgium
Bosnia
Canada
Chile
China
Cyprus
Czech
Republic
Denmark
Estonia
Finland
France
Germany
Greece
Hong Kong
Hungary
Iceland
Indonesia
Ireland
Israel

Italy
Jamaica
Japan
Korea. Rep.
Latvia
Luxembourg
Macedonia
Maldives
Malta

Mauritius

 

 

 

 

 

 

 

 

Logistic BDM Gompertz
Market Rate of Market Coef. of Market Rate of
potential acceptance potential internal Inf. potential acceptance
73.38 1.56 69.46 1.03
116.51 0.34 121.68 0.31
103.68 0.71 109.75 0.42
88.03 0.93 88.60 0.93 1 1 1.43 0.49
89.95 0.90 91.63 0.54
49.46 0.81
71.33 0.30 83.02 0.25
82.78 0.53 9-22 0.45 94.72 0.29
39.65 0.56 46.11 0.46 49.13 0.29
123.86 0.42 1 15.54 0.47
115.91 0.81 114.94 0.83 117.87 0.52
1 12.67 0.41 113.93 0.40 130.54 0.21
154.97 0.43
107.10 0.41 110.19 0.39 124.07 0.21
79.68 0.72 75.63 0.81 83.34 0.44
98.84 0.63 105.46 0.38
91.07 0.87 93.76 0.53
133.72 0.45 144.50 0.39 148.08 0.25
99.75 0.71 95.66 0.78 106.82 0.40
106.13 0.60 89.58 0.90 115.79 0.30
54.09 0.58
106.83 0.59 105.06 0.63 1 14.52 0.36
120.20 0.52 127.83 0.46 132.14 0.30
129.35 0.53 147.61 0.42 146.74 0.30
93.86 0.71
76.89 0.50 83.52 0.42 82.00 0.30
77.58 0.74 79.71 0.48
119.85 0.51 134.81 0.44 165.53 0.24
162.39 0.54 165.52 0.52 167.13 0.34
76.20 0.82
88.42 0.94
79.91 1.31 82.45 0.65
74.20 0.48 89.53 0.38

 

117

 

 

Country

Netherlands
New Zealand
Nigeria
Norway
Philippines
Portugal
Russia
Singapore
Slovak
Republic
Slovenia
Spain
Sweden
Switzerland
Tajikistan
Tunisia

UK

United States
Uzbekistan

Viet Nam

 

 

 

 

 

 

 

 

Logistic BDM Gompertz
Market Rate of Market Coe f. of Market Rate of
potential acceptance potential internal Inf. ‘ potential acceptance
97.35 0.71 102.63 0.44
98.24 0.41
41.07 0.97 45.49 0.90
112.78 (1.38 117.12 0.35 132.06 0.19
58.63 0.58
111.54 0.70 117.49 0.55 117.72 0.38
127.18 0.99 138.39 0.56
108.85 0.50
92.89 0.71 94.81 0.68 98.32 0.41
90.30 1.21 90.26 0.76
100.27 0.73 88.64 0.96 105.27 0.44
111.34 0.40 107.20 0.43 123.72 0.21
93.29 0.68 100.78 0.37
5.74 1.41 6.00 0.88
74.78 1.10
114.21 0.55 125.09 0.31
97.88 0.29 118.96 0.23
3.49 0.87
28.20 0.99

 

118

 

Appendix II.Mobile penetr

  

Algeria

Australia

: Austria

Belarus

' Belgium

Bosnia/Herzegovina

Canada

Chile

. China

 

Cyprus
Czech Republic
Denmark
Estonia
Finland
France
Germany
Greece
Hongkon g
Hungary
Iceland
Indonesia
Ireland
Israel

Italy

Jamaica

:Japan

. Korea (Rep. of)

Latvia
Luxembourg
Maldives
Malta

: Mauritius

. Netherlands

. New Zealand

E Nigeria

El” L- .2

 

0

0

0.4913

000000

0
0
0
0

        

 

ation (s

1.5188

0
0
0.1290
0
0
0
0.0463
0
0
0
0
0.9014
0
1.3798
0
0.0018
0
0.0806
0
0
0.0012
0.0085
0
0.01 12
0
0.051 1
0
0
0.0109
0
0
0
0.0331
0
0

110

 

ubscriptions/100 habitants)
1 1995

 

1990

0.0019
1.0838
0.9599

0
0.4302

0
2.1561
0.1063
0.0016
0.5376

0
2.8862

0
5.1595
0.4994
0.3769

0
2.4358
0.0255
3.9146
0.0101
0.7137
0.3216
0.4613

0
0.7023
0.1844

0
0.2158

0

0
0.2078
0.5284
1.5969

0
4.6323

0.0167
12.4121
4.8254
0.0575
2.3246
0
8.8098
1.3822
0.2976
7.0672
0.4731
15.7285
2.0521
20.0720
2.2517
4.5528
2.6124
12.9688
2.5864

1 1.5319
0.1082
4.3819
7.9277
6.8425
1.8070
9.3272
3.6839
0.5931
6.5506
0
2.8998
1.0458
3.4788
9.9363
0.0126
22.4555

 

 

0.2827
44.7022
76.3521

0.4921
54.8453

2.4699
28.3459
22.3623

6.6924
32.1538
42.3052
63.1058
38.7021
72.0368
49.3308
58.5975
56.1515
81.7306
30.1685
76.4754

1.7789
64.9872
70.1799
73.7303
14.1298
52.6193
58.3152
16.5541
69.1617

2.8278
29.3446
15.0809
67.2731
39.9710

0.0263

71.5898

 

41.5163
91.3917
105.8143
42.0241
89.9977
40.8080
52.5098
67.7884
29.8976
93.7129
1 15.2239
100.3428
108.7509
100.4001
79.4896
95.8664
92.2697
123.4678
92.2955
103.4017
21.0564
102.9412
1 12.4203
123.0785
74.7440
75.3287
79.3933
81.1271
154.8280
66.7508
80.7930
52.7573
97.1471
87.6148
14.1315

 

102.9041

 

 

Year

 

Philippines
Portugal

Russia
Singapore
Slovak Republic
Slovenia

Spain

Sweden
Switzerland
T.F.Y.R. Macedonia
Tunisia

United Kingdom
United States
Uzbekistan

Viet Nam

 

 

0
0
0
0
0
0
0

 

1985

0.8743
0
0
0
0.0883
0.1427
0
0

1990

 

0
0.0663
0
1.6976
0
0
0.1371
5.3686
1.8202
0
0.01 17
1.9354
2.1242
0
0

 

1995

0.7197
3.4357
0.0597
8.6794
0.2295
1.3736
2.4100
22.7214
6.3541
0
0.0356
9.7860
12.6881
0.0164
0.0321

 

2000

8.4372
66.4952
2.2265
68.3824
23.0236
61.0855
59.9140
71.7558
64.3434
5.7600
1.2460
72.7035
38.9019
0.2155
1.0023

 

2005

41.2980
109.0844
83.6237
100.7630
84.0654
89.4373
100.0054
100.6968
91.6228
62.0104
56.3229
109.7266
71.4966
2.7075

1 1.3882

 

 

:32 ad a 285%: ...... .35. Be a Became ...... .55. 3 a 38%? ..

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

08.. *cm_. 1.5m. 3:. owo. ting. smom. c _ or Eo. 08.. N8. mo. .- =o_w=om

c9: *Icoh. .23..de 2o. moor ...MZN. *Nmmr sionmr 433.. 9:. 5%.. owmzwcj

coo.— ..aaw 5v. Nwor mac. cc _. .5..er ......mmmr ......wvmr ... _ an. $55.. 22:52

95.— vmﬁ .1 _N.- .4103: .22.. _ mm: 3.34%.. sin—W- on _. 4.352.- out;

89.. .I. :oNr *v _ N. 3b. mvor moo. 3 _. coo. doc dam

coo; who. .3. .3..me 3... 3.403..- *vmm. .5250

coo; ......aomvr $394.- ......izmr *mi. 3.2%.. 56>:—

ooo._ 33$. *3.qu w. _.- 4.5%. Sat:

25.. 133. $4034.- ***xmw. teacu—

ooo._ 4.55.- 43%. saw

95.. .."mcmr E6

ooo._ wizou

=o_w:om own—sweet— bBEﬁm REA doa dam 5280 3.62: Sat: 63:”— QDO 25 9.5pr

man: 23:82. Hod—HE 16on oummwod $56568 macaw—8.50 .2: 638.

.838 coca—BSD .E iguana.

.26. S... a .588; ...... ....so. 8... .... .588... ...... .35. ..o a .5828 ..

 

 

_oloc._ N2. .3...me 3...—3. Fe. vﬂ. m9.- assnhv. ......svwnr ZN: ea". 1.23.. .....immmr 8.42m.- ..aamwv. coca
coo; 03. ......SN. 3;.- omo. m2. of. mmor mam. c2. o5. o8. m8: .....immmr comwnom
coo.— .....:..o$.. .353. 05. OD. ......i-wm. ....Inwvr woor ow_. ...Nmmr iammr ......Nnmr m8. owaawSJ
8o.— 92. simmer m9. .....iev. .....iomvr ammo. an. ......wnur ......ammr ......wvmr Kc. axoaﬁm
coo; at... 1&2“.- ..smN-N. owe. 03.- moor ...SS. Sou :0.- NS. doom—demo
coo; mmor to. S _.- SN.- 3 ﬂ. o5. mvor moo. Kc. don dam
8c; ...vmmr omor m2. 5.. EC. 2:. m8. lama.- 5.5885
So.— sssmbmr mace acmm. .....imoer 145%.- ...Imocr ......mom. 8:5
coo; N2. ......smhmr Items. .....ioK. 3.53. .53.- «:82
coo; .....icvmr “No. .....me 3:. ...,...owmr 5280
RX: 37 1.403.- 1.42m.- ooor 85
cos; .....imnc. 2.390. .....iommr Sat:
OR: 3.5%. .2339..- taxi
o8.— .........Nov.- n50
.aouo<
coo.—
uo 8am
8.5 : em » doom den :2 .330 «622 5950 E5 535 3me mac .383.
awzom mamas-— 5:55 .555 don E3800 we 8am

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

at:

323388 mo 88 I Eco—b ocmmwod $56568 maouﬂobou .N-E £an

122

.25. S... a 28%»... ...... ..o>o. no... a Baa-Ea... .... ..o>o. ..o a Became ..

 

 

—ooo._ m .o. ......amD... ......N _ v. mmv. mom: ......somv. ......amwor ......aoohr ......aoonr Em. moor outn—
ooo. m. ..- ow..- ......momr ......cw..- of. mm..- 2...- Nmor ...Eo. 1.3%.. .55 dos.
ooo. ......Vwm. mm... mm .. ......nwm. ......mnmr ...... .mm.. .mm: ...mvm. cmmr 5.33.
8o. ......Pw. I... E _ .- 3m. mmm: ...mmm: :- _ .- teem. moor uwsswcﬁ
ooo._ ......nm _ .- ...vom. ... Em.- ..aﬂumr Loom..- ...owm. oo . .- 56:55
So. wo..- .....va timov. .21-mm. ago—m: ......nmm. $9.60
So. .....cwvcmr .4480.- ..atvr omm. :5. 75>:—
ooo._ 4&3. ......amoo. tel-mm.- aaaomo. moo
ooo._ ...,...mnn. ......ime..- ......tnwm. Bx.”—
ooo._ Em.- ..**_oo. :35
ooo._ .2.- E5
8.... 8:30
out; don—don cog—om owmzwcj 5.25.5 8650 .82.. ABC Bin. 555 .50 wﬁzou

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

om":

9:282. Hod—BE 1.25... .5256 8mm: 3.5.0568 macaw—2.50 .m-E 635.

123

.25. ...... a 88...? ...... ..o>o. mod 3 .588... .... ..o>o. .... a .5828 ..

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ooo._ 32. ot. 2m. 33.. mo..- ......mom. 8.35;.- .........mov.- ***va.- boo.- .....cwvmor outn—
ooo._ com. boo. N _. _.- mvmr mm _. *wvm. mom. .2. wmo. N _o. .50 don
ooo. _. ooo. mmo. ......mom. ......Nhﬁ Nhor C ..- .mmr ooo.- 55.- 5.33%
So. time... .5... an..- 9.... S. .- ooo.- ooo. mm... 8.333
ooo._ omo. 5.. owor ...mwmr moor who. ooo. 33:55
ooo._ mi. om..- om_. moo. ......wmmr .33. .9260
ooo._ *smvmr ......Lummr ......i-nmr omor o _ Nr Engage—
ooo._ ......owm. 3.....va omor .....cwovm. 533.53.:
ooo._ ...,...wa *aoamr ......womv. 05:58:...
8.... .32.- ......oom. 333.50
ooo._ 35.. 08.0 .50
8.... 8:60
02.5 .50 dog :o_w=o~_ omaswcﬁ 562.50 65.50 258535 ..235535 95:55”. n_Q0 .300 .50 @560
VMHQ Comaﬂgom gov—.88 l 3605 NUOQEOCV mucomomCDOo COUQ—OCOU Jim: D—tht

:32 So a “505%; t; .92 so a 28%? i .65. 3 a 28%? *

 

—oco._ o5. 2N. *wmm. _w_. *wmmr 02. “nib—v. ***~_v.- war 58.. 133.- $39!- "wimmvr suing. ootm
ooo._ ooo. NNO. than. “Invom. 09 .- moo. nmor Imam. G8,- who.. 2 _.- .mmr om. .. comwzom

ooo._ ***mnw. 58. So.. .3va *ivm. wmor woo. ooo. 03. $7 ooo.- *mcm. owmzwcj

A50.— N. ..- 3.... CNN. *Imov. C ..- one. who. 95.. *mri 93.. inoam. ESE—tm

ooo._ 5:. twig“. 95.. 00—. *mvmr wno. *mvm. mom. _C. :9. doc do;

ooo._ cwor mm _ .- NC. 1mm. *th. owo. ooo. _ 5. mm _ .- cosaanoU

ooo._ NE. ivom. w_ _.- ooo.- Him—m. mmo. 00—. me. doom .EEO

ooo._ **mmm.- ﬁmor mt. ivcmr 1.556.- $339.- ***ovo. AEQQAEE

ooo._ CE. 12%.. $0359. ***©O©. ***ooo. "Irma”- «=52
ooo._ **wmm.- cm... 02. N8. :50. 53:00
ooo._ Omor *inoomr **omm.- :0. E6

ooo._ ***©mm. $2"va $39M- =OENNEN€D

ooo._ ***mww. *ibmwr taxi
ooo._ “3.313.- ADO

. .Eooo<

cos _ ho 2mm

 

our—m cog—om owgmcﬁ $255 due :0: doom >203 «622 $2.00 £5 :23 Bin mDO .Eooo<
dam 53:50 .550 .82: NEmﬁD .6 2mm

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

an: Aoocwaoooa mo 28 I 589:0va $56508 scam—850 6-5 28g

125

 

References

Ahn, H., & Lee, M. (1999). “An Econometric Analysis of the Demand for Access to
Mobile Telephone Networks.” Inﬁmnation Economics and Policy, 1 1: 297—305.

Alesina, A., & Rodrik, D. (1994). “Distributive Politics and Economic Growth.” The
Quarterly Journal of Economics, 109(2): 465-490.

Alesina, A., & Perroti, R. (1996). “Income Distribution, Political Instability, and

Investment.” European Economic Review, 40: 1203—1228.

Alesina, A., Devleeschuwer, A., Easterly, W., Kurlat, S., & Wacziarg, R. (2003).

“Fractionalization.” Journal of Economic Growth, 8: 155-194.

Arnould, E. J. (1989). “Towards a Broadened Theory of Preference and the Diffusion of
Innovations: Case of Zinder province, Niger Republic.” The Journal of Consumer
Research, 16: 239-267.

Barrantes. R., & Galperin, H. (2008). “Can the Poor Afford Mobile Telephone? Evidence

from Latin America.” Telecommunications Policy, 32: 521-530.

Barro, R. J. (2000). “Inequality and Growth in a Panel of Countries”. Journal of
Economic Growth, 5: 5—32.

Bass, F. M. (1969). “A New Product Growth for Model Consumer Durables.”
Management Science, 15(5): 215-227.

Bass, F. M., Krishnan, T. V., & Jain, D. C. (1994). “Why the Bass Model Fits without
Decision Variables.” Marketing Science, 13(3). 203— 223.

Damanpour. F. (l99l). “Organizational Innovation: a meta-analysis of effects of

determinants and moderators.” Academy of Management Journal, 34(3): 555-590.

Dekimpe, M. G, Parker, P. M., & Sarvary, M. (1998). “Staged Estimation of

International Diffusion Models: an application to global cellular telephone

126

adoption.” Dchnological Forecasting and Social Change. 57(1): 105-132.

Deutscgmann P. J., & Danielson, W. A. (1960). “Diffusion of Knowledge of the Major
News Story.” Journalism Quarterly, 37: 345-355.

Dixon, R. (1980). “Hybrid Corn Revisited.” Econometrica. 48(6): 1451-1461.

Downs, G. W., & Mohr, L. B. (1976). “Conceptual issues in the study of innovation.”
Administrative Science Quarterly, 21: 700-714.

Foellmi, R. & Zweimuller, J. (2006). “Income Distribution and Demand-Induced
Innovations.” Review of Economic Studies, 73: 941-960.

Forbes, K. J. (2000). “A Reassessment of the Relationship between Inequality and
Growth”, American Economic Review, 90(4): 869—87.

Gans, J. S., King, S. P., & Wright, J. (2001). “Wireless Communications” in Handbook of
Telecommunications Economics, Volume 1., M. E. Cave et al. eds., pp. 241-285.
Amsterdam: North-Holland/Elsevier.

Galor, 0., & Zeira, J. (1993). “Income Distribution and Macroeconomics” Review of
Economic Studies, 60: 35—52.

Galor, 0., & Moav, O. (2004). “From Physical to Human Capital Accumulation:
inequality and the process of development.” Review of Economic Studies, 71:
1001-1026.

Garcia-Penalosa, C., & Turnovsky, S. J. (2006). “Growth and Income Inequality: a

canonical model.” Economic Theory, 28: 25-49.

Gatignon, H., Eliashberg. J ., & Robertson, T. S. (1989). “Modeling Multinational
Diffusion Patterns: an efficient methodology.” Marketing Science, 8(3): 231-247.

Gottschalk, R. & Justino, P. (2006). “Introduction” in Overcoming Inequality in [min
America: issues and challenges for the twenty-ﬁrst century, Gottschalk, R. & P.
Justino eds, pp. 1-14. London: Routledge.

197

H

Griliches, Z. (1957). “Hybrid Corn: an exploration in the economics of technological
change.” Econometrica, 25(4): 501-522.

Gruber, H. (2001). “Competition and Innovation: the diffusion of mobile
telecommunications in Central and Eastern Europe.” Information Economics and
Policy. 13(1): 19-34.

Gruber, H. (2005). “The Economics of Mobile Telecommunications.” Cambridge:

Cambridge University Press.

Gruber, H. & Verbovan, F. (2001a). “The Diffusion of Mobile Telecommunication

Services in the European Union.” European Economic Review, 45: 577— 588.

Gruber, H. & Verbovan, F. (2001b). “The Evolution of Markets Under Entry and
Standards Regulation: the case of global mobile telecommunications.”

International Journal of Industrial Organization, 19: 1189-1212.

Horsky, D. (1990). “A Diffusion Model Incorporating Product Beneﬁts, Price, Income
and Information.” Marketing Science, 9(4): 342-365.

Horsky, D., & Simon, L. S. (1983). “Advertising and the Diffusion of New Products.”
Marketing Science, 2(1): 1— 17.

Islam, T. & Meade, N. (1996). “Forecasting the Development of the Market for Business
Telephones in the UK.” Journal of the Operational Research Society, 47: 906-
918.

Iyi gun, M. F. & Owen, A. L. (2004). “Income Inequality, Financial Development. and

Macroeconomic Fluctuations”, The Economic Journal, 114: 352-376.

James, J. (2003). “Bridging the Global Digital Divide.” Northampton, MA: Edward
Elgar.

Karshenas, M., & Stoneman, P. (1992). “A Flexible Model of Technological Diffusion

Incorporating Economic Factors with an Application to the Spread of Colour
Television Ownership in the UK.” Journal of Forecasting, l 1, 577— 601.

128

 

Kentor, J. (2001). “The Long Term Effects of Globalization on Income Inequality,
Population Growth, and Economic Development.” Social Problems, 48(4): 435-
455.

Koski, H., & Kretschmer, T. (2006). “Entry, Standards and Competition: firm strategies
and the diffusion of mobile telephony.” Review of Industrial Organization. 26(1):
89-113.

Kuznets, S. (1955). “Economic Growth and Income Inequality.” The American Economic
Review, 45(1): 1-28.

Lambert, P. J. & Pfahler, W. (1997). “Market Demand and Income Distribution: a
theoretical exploration.” Bulletin of Economic Research, 49(2): 137-151.

Lee, M., & Cho, Y. (2007). “The Diffusion of Mobile Telecommunication Serivces in
Korea.” Applied Economic Letters. 14: 477-481.

Liebermann, E. & Paroush, J. (1982). “Economic Aspects of Diffusion Models.” Journal
of Economics and Business, 34, 95—100.

Li, H., & Zou, H. (1998). “Income Inequality is not Harmful for Growth: theory and
evidence.” Review of Development Economics, 2(3): 318-334.

Liikanen, J ., Stoneman, P., & Toivanen, O. (2004). “Intergenerational Effects in the

Diffusion of New Technology: the case of mobile phones.” International Journal
of Industrial Organization, 22(8-9): 1137-1154.

Lim, B., Choi, M., & Park, M. (2003). “The Late Take-Off Phenomenon in the Diffusion
of Telecommunication Services: network effect and critical mass.” Information

Economics and Policy, 15: 537-557.

Lundberg, M., & Squire, L. (2003). “The Simultaneous Evolution of Growth and
Inequality.” The Economic Journal, 113: 326-344.

Madden, G, Coble-Neal, G, & Dalzell, B. (2004). “A Dynamic Model of Mobile

Telephony Subscription Incorporating a Network Effect.” Telecommunications

129

 

Policy. 28(1): 133-144.

Mahajan, V., & Peterson, R. A. (1978). “Innovation Diffusion in a Dynamic Potential
Adopter Population.” Management Science, 24(15): 1589-1597.

Mahler, A., & Rogers, E. M. (1999). ”The Diffusion of Interactive Communication
Innovations and the Critical Mass: the adoption of telecommunications services

by German banks.” Telecommunications Policy. 23: 719-740.

Mariscal, J ., & Rivera, E. (2005). “New trends in the Latin American
Telecommunications Market: Telefonica & Telmex.” Telecommunications Policy,
29: 755-777.

Matusyama, K. (2002). “The Rise of Mass Consumption Societies.” The Journal of
Political Economy, 110(5), 1035-1070.

Mani, A. (2001). ”Income Distribution and Demand Constraint.” Journal of Economic
Growth, 6: 107-133.

Markus, M. L. (1987). “Toward a ‘Critical Mass’ Theory of Mass Media: universal

access, interdependence and diffusion.” Communication Research, 14: 491-511.

McGrath, C., & ZeIl, D. (2001). “The Future of Innovation Diffusion Research and Its
Implications for Management: a conversation with Everett Rogers.” Journal of
Management Inquiry, 10(4): 386-391..

Meade, N., & Islam, T. (2006). “Modeling and Forecasting the Diffusion Innovations —A
25-year Review.” International Journal of Forecasting. 22(?): 519-545.

Murphy, K., Shleifer, A., & Vishny, R. (1989). “Income Distribution, Market Size, and
Industrialization.” The Quarterly Journal of Economics, 104(3): 537-564.

Odedokun, M. O., & Round, J. I. (2005). “Determinants of Income Inequality and Its
Effects on Economic Growth: evidence from African countries.” African
Development Review, 287-327.

Oliver, F. R. (1964). “Methods of Estimating the Logistic Growth Function.” Applied
Statistics. 13(2): 57-66.

Park. J. L., Holcomb, R. B., Raper, K. C., & Capps, Jr. 0. (1996). “A Demand Systems
Analysis of Food Commodities by US. Households Segmented by Income.”

American Journal of A gricultural
Economics, 78(2): 290-300.

Persson, T., & Tabellini, G (1994). “Is Inequality Hamiful for Growth?” The American
Economic Review, 84(3): 600-621.

Prescott, M. B., & Conger, S. A. (1995). “Information Technology Innovations: a
classification by IT locus of impact and research approach.” Data base for

advances in information systems, 26(2-3): 20-41.

Reinstaller, A., & Sanditov, B. (2005). “Social Structure and Consumption: on the

diffusion of consumer good innovation.” Journal of Evolutionary Economics, 15:
505-531.

Robinson, B., & Lakhani, C. (1975). “Dynamic Pricing Models for New Product
Planning.” Management Science, 10: 1113— 1 122.

Rogers, E. M. (1976). “New Product Adoption and Diffusion.” The Journal of Consumer
Research. 2(4): 290-301.

Rogers, E. M. (2000). “Reflections on News Event Diffusion Research.” Journalism and
Mass Communication Quarterly, 77(3): 561-576.

Rogers, E. M. (2003). “Diffusion of Innovations (Fifth Edition ).” New York: The Free

Press.

Rogers, E. M. & Peterson, J. C. (2007). “Diffusion of Clean Air Ordinances in the
Southwestern United States.” Health, Education & Behavior, 35(5): 683-697.

Rouvinen, P. (2006). “Diffusion of Digital Mobile Telephony: are developing countries
different?” Telecommunications Policy. 30(1): 133-144.

131

 

Ryan, 8., & Gross, N. C. (1943). “The Diffusion of Hybrid Seed Corn in Two Iowa

Communities.” Rural Sociology, 8: 15-24.

Samarajiva, R. (2006). “Preconditions for Effective Deployment of Wireless
Technologies for Development in the Asia-Pacific.” Information Technologies
and International Development, 3(2): 57-71.

Sawhney, H. (2003). “Wi-Fi Networks and Re-run of the Cycle.” Info: the journal of

policy, regulation and strategyfor telecommunications. 5-6: 25-33.

Schmittlein, D. C. & Mahajan, V. (1982). “Maximum Likelihood Estimation for an
Innovation Diffusion Model of New Product Acceptance.” Marketing Science,
1(1): 57-78.

Senauer, B. (1990). “Household Behavior and Nutrition in Developing Countries.” Food
Policy, 15:408-417.

Smith, D. (2001). “International Evidence on How Income Inequality and Credit Market
Imperfections affect Private Saving Rates.” Journal of Development Economics,
64: 103-127.

Srinivasan, V. & Mason, C. H. (1986). “Nonlinear Least Square Estimation of New
Product Diffusion Models.” Marketing Science, 5(2): 169-178.

Stoneman, P. (2001). “The Economics of Technological Diﬂusion”, Oxford, MA:
Blackwell.

Strodthoff, G G, Hawkins, R. P., & Schoenfeld, A. C. (1985). “Media Roles in a Social
Movement: a model of ideology diffusion.” Journal of Conunimication, 35(2):

134-153.

Takada. H., & Jain, D. (1991). “Cross-National Analysis of Diffusion of Consumer
Durable Goods in Pacific Rim Countries.” Journal of Marketing. 55(2): 48-54.

Tan guturi, V. P. & Harmantzis, F. C. (2006). “Migration to 3G Wireless Broadband

Internet and Real Options: the case of an operator in India.” Telecommunications

13?.

Policy, 30(7): 400-419.

Talukdar, D., Sudhir, K., & Ainslie, A. (2002). “Investigating New Product Diffusion

across Products and Countries.” Marketing Science, 21(1): 97— 1 l4.

Telefonaktiebolaget LM Ericsson. (2007). “Can Mobile Communications Close the
Digital Divide?” Ericsson White Papers.

Ten g, J. T. C., Grover, V., &s Giittler, W. (2002). “Information Technology Innovations:
general diffusion patterns and its relationships to innovation characteristics.”

IEEE Transactions on Engineering Management, 49(1): 13-27.

Tilson, D., & Lyytinen, K. (2006). “The 3G Transition: changes in the US wireless
industry.” Telecommunications Policy, 30(10): 569-586.

Thomson, G. L., & Teng, J. T. (1984). “Optimal pricing and advertising policies for new
product oligopoly models.” Marketing Science, 3: I48—168.

Tornatzky, L. G, & sKlein, K. J. (1982). “Innovation characteristics and innovation
adoption- implementation: a meta-analysis of findings.” IEEE Transactions on

Engineering Management, 29: 28-45.

Tramayne, M., Weiss. A. S. & Alves, R. C. (2007). “From Product to Service: the
diffusion of dynamic content in online newspapers.” Journalism and Mass
Communication Quarterly, 84(4): 825-839.

Valletti, T. M. (1999). “A Model of Competition in Mobile Communications.”

Information Economics and Policy, 1 l(l):61-72.

Van den Bulte, C. & Stremersch, S. (2004). “Social Contagion and Income Heterogeneity
in New Product Diffusion: a meta-analytic test.” Marketing Science, 23(4): 530-
544.

Van Slyke, C., Ilie, V., Lou, H., & Stafford, T. (2007). “Perceived Critical Mass and the
Adoption of Communication Technology.” European Journal of Information
Systems. 16: 270-283.

Wei, R. (2001). “From Luxury to Utility: a longitudinal analysis of cell phone laggards.”
Jourt-talism and Mass C ommtmication Quarterly, 78(4): 702-719.

Young, A. (1993). “Substitution and Complementarity in Endogenous Innovation.” The
Quarterly Journal of Economics, 108(3): 755-807.

Zweimuller, J. (2000). “Schumpeterian Entrepreneurs Meet Engel’s Law: the impact of
inequality on innovation-driven growth.” Journal of Economic Growth, 5: 185-

206.

Zweimuller, J., & Brunner, J. K. (2005). “Innovation and Growth with Rich and Poor
Consumers.” Metroeocnomica. 56(2): 23—262.

1. 3 4

111111111111111111111111111'“