Diffusion of digital mobile telephony: Are developing countries different?

ARTICLE IN PRESS

0308-5961/$ - se

doi:10.1016/j.te

�Tel.:+358 9

E-mail addr

Telecommunications Policy 30 (2006) 46–63

www.elsevierbusinessandmanagement.com/locate/telpol

Diffusion of digital mobile telephony: Are developingcountries different?

Petri Rouvinen�

ETLA, The Research Institute of the Finnish Economy, ETLA, Lonnrotinkatu 4 B, FIN-00120 Helsinki, Finland

Abstract

Factors determining the diffusion of digital mobile telephony across developed and developing countries are studied

with the aid of a Gompertz model. After controlling for other factors, the speed of diffusion per se is not significantly

different between the two groups of countries. Standards competition hinders and market competition promotes diffusion

in both groups. Various factors are, however, more important in a developing country context: having a large potential

user base, accumulating network effects, being open, commanding a high (non-telecom) technological level, and

introducing innovation(s) complementing mobile telephony. Late entrants experience faster diffusion promoting cross-

country convergence.

r 2005 Elsevier Ltd. All rights reserved.

JEL Classification: L96; O30; O10

Keywords: Mobile telephony; Technology diffusion; Gompertz model; Developing countries

1. Introduction

In recent decades wireline (fixed) telephony has increasingly been complemented and also replaced bywireless (mobile), which currently dominates in terms of worldwide usage (ITU, 2002a, b). It should bepointed out that while the role of the Internet in economic development has been emphasized (see, e.g., Clarke& Wallsten, 2004; Kenny, 2003; Lucas & Sylla, 2003; Petrazzini, 1999; Qureshi, 2003), mobile telephony—withwell over twice as many users worldwide – also holds considerable potential in this respect.

This paper addresses some deficiencies in the previous literature in studying the socio-economic factorsdriving the diffusion of digital mobile telephony. The possible differences between developed and developingcountries with respect to these factors are of particular interest.

In both fixed and mobile telephony analog technologies have been replaced by digital ones. Indeed, theworldwide breakthrough of mobile telephony is associated with the commercial introduction of digitaltechnologies in the early 1990s. Among the developed countries the average penetration rate (users perpopulation) of analog mobile telephony peaked at less than five per cent in the mid-1990s, whereas the

e front matter r 2005 Elsevier Ltd. All rights reserved.

lpol.2005.06.014

609900; +358 9 60990202 (direct), +358 50 3673474 (mobile); fax:+358 9 601753.

ess: [email protected].

www.elsevierbusinessandmanagement.com/locate/telpol

ARTICLE IN PRESSP. Rouvinen / Telecommunications Policy 30 (2006) 46–63 47

penetration of digital mobile telephony is currently some fifty per cent. The corresponding penetration ratesamong the developing countries are some ten times lower. Although the switch from analog to digital mobiletelephony has also been somewhat slower among developing countries, the diffusion patterns per se seem to bemore similar in the case of digital mobile telephony.

There are many reasons for the success of digital mobile telephony. First, by economizing on the use of thelimited radio spectrum, digitalization made the current levels of mobile telephony usage technically possible.Secondly, combined with other industry developments, digital mobile telephony offered end users a moreattractive bundle in terms of price, quality, and services. In many countries competition was first introduced indigital mobile telephony with direct consequences for user cost: since the basic 2G offering was fairly standard(a voice call of reasonable quality), this naturally lead to a price competition among two or more operators.

Further, Digital mobile telephony had advanced data transmission (short messaging service etc.) andimproved voice quality. In part thanks to the lower power consumption of digital mobile telephony, smallerand lighter end-user terminals (handsets) became available. Thirdly, with the expanding user base, networkeffects and economies of scale in both production and use rapidly accumulated. In short, with digitalizationmobile telephony truly became a worldwide consumer market.

Dekimpe, Parker, and Sarvary (1998), Ahn and Lee (1999), Burki and Aslam (2000), Gruber (2001), Gruberand Verboven (2001a, b), Liikanen, Stoneman, and Toivanen (2001), Koski and Kretschmer (2002), andMadden, Coble-Neal, and Dalzell (2004) are among the studies modeling cross-country mobile telephonydiffusion. Some aspects of these studies are summarized in Table 1.

Several things are noteworthy in Table 1. First, with the exception of Dekimpe et al. (1998), and Gruber andVerboven (2001a, b), the number of countries included in the analyses are relatively low, and none of thestudies explicitly focus on comparing developed and developing countries. Secondly, with the possibleexceptions of Liikanen et al. (2001), and Koski and Kretschmer (2002), the sets of (non-telecom) socio-economic explanatory variables remain rather modest, and only GDP per capita and a population measure areshared across studies. Thirdly, again with the exceptions of Liikanen et al. (2001), and Koski and Kretschmer(2002), the dependent variable combines both analog and digital mobile telephony, although most studiesacknowledge their important differences.

2. The model

Mobile telephony diffusion is studied with the aid of a Gompertz growth model, which in the past has beenused to study, for example, the spreading of computers (Stoneman, 1983, Ch. 10) and the Internet (Kiiski &Pohjola, 2002). Although a wealth of alternatives exist (see, e.g., Stoneman, 2002), the Gompertz model isparsimonious, linear in parameters, and allows for simple inclusion of socio-economic explanatory variables.Furthermore, interestingly Madden et al. (Madden & Coble-Neal, 2001; Madden et al., 2004) end up with aspecification that is identical to the one derived below, although their starting point is a dynamic optimizationproblem of an economic agent rather than a diffusion model.

Let Ni;t be the number of mobile telephony users in country i at time t. Over time it tends towards its post-diffusion or equilibrium level N�i;t along an S-shaped path. The Gompertz growth model specifies the rate ofchange as

lnNi;t � lnNi;t�1 ¼ aðlnN�i;t � lnNi;t�1Þ (1)

where a is the speed of adjustment. The equilibrium level N�i;t is a function of past supply and demand factors(denoted by a vector X i;t) including—but not limited to—availability, disposable income, and user cost

lnN�i;t ¼ Xbi;t (2)

where b is a vector of coefficients. Inserting (2) into (1) yields

lnNi;t � lnNi;t�1 ¼ ab0 lnX i;t�1 � a lnNi;t�1 (3)

which is estimated with an appropriate econometric method as soon as X i;t�1 and the stochastic error structurehave been specified. The strict interpretation of b s in the above model is that they shift the equilibrium stockof adopters.

ARTICLE IN PRESS

Table 1

Some economic studies modeling cross-country mobile telephony diffusion

Study Dependent Independent variables Countries Period Findings

Dekimpe et al.

(1998)

Mobile penetr. GNP per cap., pop. growth, # of

major pop. centers, # of

competing systems, death rate,

communism dummy, # of ethnic

groups.

184 1979–1992 High wealth, ethnic

homogeneity and low death

rate promote diffusion.

Ahn and Lee

(1999)

Mobile penetr. GDP per cap., fixed penetr. and

digitalization rate, mobile user

cost.

64 1997 High GDP per cap. and fixed

penetr. promote diffusion.

Burki and Aslam

(2000)

Mobile users GDP, pop., fixed penetr., digital

mobile dummy, analog and

digital mobile competition

dummies.

25 (Asian) 1986–1998 Analog to digital mobile

transition changed diffusion

patterns. Competition

promotes diffusion.

Gruber (2001) Mobile penetr. GDP per cap., sh. of urban pop.,

fixed penetr. and wait time,

digital mobile competition

dummy, # of mobile operators,

market transition index.

10 (EU

accession)

Introd–1997 Late mobile adoption &

multiple operators and high

fixed penetr. and long wait

times promote diffusion.

Gruber and

Verboven (2001a)

Mobile users GDP per cap., fixed penetr.,

digital mobile technology

dummy, analog/digital mobile

competition dummies.

15 (EU) 1992–1997 Analog to digital mobile

transition & competition

promote diffusion. Late

entrants adopt mobile faster.

Gruber and

Verboven (2001b)

Mobile users GDP per cap., fixed penetr. and

wait list, digital mobile dummy,

digital introduced w/o pre-

existing analog dummy, multiple

analog/digital standards

dummies, analog/digital general/

simult./sequential market

competition dummies.

140 1981–1997 Slow diffusion convergence

across countries. Introducing

competition has strong

immediate impact.

Incumbents pre-empt

sequential entry. Single

standard promotes diffusion.

Liikanen et al.

(2001)

Ch. in analog

and/or digital

mobile users

GDP per cap., pop., sh. of urban

pop. and pop. over 65, fixed

users/penetr., analog/digital

users/penetr., # of analog/digital

standards and years since

introd., NMT & GSM dummies,

5 measures of mobile telephony

operation, age-dependency ratio,

surface area.

80 1992–1998 Digital mobile introduction

hinders analog mobile

diffusion. Generation-specific

(analog vs. digital) results

differ from generic

(analog+digital) results:

technology shifts should be

accounted for.

Koski and

Kretschmer (2002)

Mobile penetr.,

user cost and

entry

GDP per cap., sh. of urban pop.,

telecom regulator dummy and

competition measure, analog

mobile penetr., digital mobile

subscriber and prepaid users,

digital mobile standard dummy,

market sh. of dominant digital

mobile standard, more than 2

mobile operators dummy.

32 1991–1999 Incorporating the time of

entry to digital mobile

telephony study is important.

Both between and within

standards competition

promote diffusion and lower

user cost particularly when

more than 2 operators are

present.

Madden et al.

(2004)

Mobile penetr. GDP per cap., pop., mobile user

cost.

56 1995–2000 High wealth, low users cost

and large user base promote

diffusion.

Note: Dependent refers to the dependent variable(s) in the study in question.

P. Rouvinen / Telecommunications Policy 30 (2006) 46–6348


3. Data

For the present purposes a developing country is defined as a low or lower-middle income country asdefined in the 2002 edition of the World Bank’s World Development Indicators database.1 The possibleconsequences of this choice are discussed below.

EMC’s World Cellular Database (www.emc-database.com), the ITU’s World TelecommunicationsIndicators, and the World Bank’s World Development Indicators include telephony diffusion informationfor over 200 countries and regions. Table 2 shows the year of commercial introduction of digital mobiletelephony by country. As can be seen, developing countries typically adopt later and one in ten had notadopted at all by the end of 2000.

Besides the time of adoption, the penetration rates also differ considerably between developed anddeveloping countries. As shown in Fig. 1, the average penetration rate across the developed countries was onethird, whereas the corresponding figure for the developing countries was five per cent at the end of 2000.

Table 3 defines the socio-economic variables used in the analysis. The country’s total population andpopulation in the largest city are proxies for the overall size of the market.2 Real GDP per capita controls forthe wealth and income effects. Agricultural value added to GDP and the illiteracy rate are used to account forthe country’s overall state of development. The ratio of private credit to GDP is taken as a proxy of thecountry’s level of financial development. The ratio of trade to GDP is a control for openness. Two well-knowndatasets—Freedomhouse (www.freedomhouse.org/ratings) and Polity IV (www.cidcm.umd.edu/inscr/polity)—are used to construct an index of political freedom accounting for the degree of democracy in thecountry’s political system. The PCs per capita ratio is taken as a proxy for the country’s overall (non-telecom)technological level.

The remaining variables in Table 3 relate to fixed, as well as analog and/or digital, mobile telephony. Fixedand analog mobile telephony penetration rates capture their substitutability or complementarity, as well aspossible network and/or economies of scale effects, with respect to digital mobile telephony.3 It is assumedthat fixed telephony user cost captures other relevant aspects of fixed telephony. The dependent variable isconstructed by taking a log difference of the number of digital mobile telephony users. Three indicatorscapture aspects of the digital mobile telephony market: the ‘prepaid’ dummy takes the value of one (and iszero otherwise) if prepaid mobile telephony calling cards are available in the country in question, the ‘many’dummy takes the value of one (and is zero otherwise) if the country has competing standards in digital mobiletelephony, and the ‘competition’ dummy takes the value of one (and is zero otherwise) if the country hascompeting operators in digital mobile telephony. As mobile user costs cannot be constructed separately foranalog and digital mobile telephony with the available data, a combined mobile cost is used instead. Sincemobile handset prices are not available, the investment price level is used as a proxy, that is, handsets arelikened to capital goods.4

Table 4 presents the descriptive statistics of the untransformed variables that are usable in the regressionanalysis. Developed and developing countries are reported separately in order to aid comparison of the twogroups. The differences in means would seem to suggest that developing countries are on average larger, theirGDPs per capita and (non-telecom) technological levels are eight times lower, they are more agrarian, havehigher illiteracy rates, private sector credit is less readily available, and they are less open and democratic thanthe developed countries.

As far as telecommunications related variables are concerned, the developing countries’ fixed telephonypenetration rates are four times lower but user costs are twice as high. Analog and digital (not shown) mobilepenetration rates are some ten times lower. Although in absolute terms mobile user cost is about one fifthlower in the developing countries, in relative terms (with respect to average income) it is indeed considerable.

1If the country in question is not included in this data source, it is assumed to be a developing country.2Also the inclusion(s) of the share of urban population, population density, and/or the country’s surface area were studied jointly and

separately, but test statistics did not indicate that they should be included.3Also the inclusion(s) of the fixed telephony digitalization rate and/or wait time, ownership and/or competitive status of the incumbent

operator, and fixed and/or analog mobile competition dummies were studied jointly and separately, but test statistics did not indicate that

they should be included.4Note that the time dummies will capture global price changes.

http://www.emc-database.com

http://www.freedomhouse.org/ratings

http://www.cidcm.umd.edu/inscr/polity


ARTICLE IN PRESS

Table 2

The year of commercial introduction of digital mobile telephony by country

Developed

(78 countries/regions)

Argentina

Brazil

Bahrain Brunei

Australia Belgium Canada Czech Rep.

Austria Channel Isl. Costa Rica Dominica

Andorra Greece Hungary Croatia Guam

Denmark Ireland Iceland Cyprus South Korea

Finland Italy Israel Estonia Libya Bahamas

France Japan Kuwait French Polyn. Mauritius Barbados Botswana

Gabon Luxemb. Malaysia Lebanon Oman Bermuda Cayman Isl.

Germany N. Zealand Netherlands Macao Panama Chile Dominica

Hong Kong Norway Qatar New Caled. Poland Malta Faeroe Islands

Portugal Singapore South Africa Puerto Rico Saudi Arabia Slovak Rep. Greenland Dominica

Sweden Switzerland Turkey Seychelles Slovenia Uruguay Mexico Grenada

UK USA UAE Spain Venezuela Virgin Isl. Trinidad and T. St. Lucia

1992 1993 1994 1995 1996 1997 1998 1999 2000

Nicaragua Cameroon Bulgaria Albania Bangladesh Dominican R. Algeria Anguilla

China Colombia Armenia Bolivia El Salvador Angola Benin

Fiji Congo, Rep. Azerbaijan Cape Verde French Guiana Belarus Burundi

Indonesia Georgia Bosnia-Herz. Guinea Guyana Centr. Afr. R. Chad

Iran Gibraltar Burkina Faso Martinique Moldova Congo D. R. Eq. Guinea

Madagascar India Cambodia Mozambiq. Paraguay Cuba Honduras

Morocco Jordan Cote d’Ivoire Romania Peru Ethiopia Mali

Pakistan Kyrgyz Rep. Ecuador Togo Rwanda Guatemala Marshall Isl.

Philippines Lao PDR Egypt Zambia Swaziland Haiti Mauritania

Russia Latvia Ghana Tunisia Jamaica Sierra Leone

Taiwan Lithuania Guadeloupe Kazakhstan Tajikistan

Thailand Malawi Guernsey Maldives Turkmenistan

Vietnam Myanmar Kenya Nepal

Namibia Lesotho Syrian Arab R.

Reunion Macedonia W. Bank, Gaza

Sri Lanka Mongolia

Suriname Senegal

Tanzania Sudan

Tonga Ukraine

Developing Uganda Yugoslavia

(102 countries/

regions)

Uzbekistan Zimbabwe

Note: Not introduced commercially in the following 21 developing countries/regions by the end of 2000: Afghanistan, Belize, Bhutan,

Comoros, Cook Island, Djibouti, Eritrea, Gambia, Guinea-Bissau, Kiribati, North Korea, Micronesia, Niger, Nigeria, Papua New

Guinea, Samoa, Sao Tome and Principe, Solomon Islands, St. Helena, Vanuatu, and Yemen Republic.


Table 5 presents the correlation matrix of the variables. With the exception of the analog mobile telephonypenetration, the number of digital mobile telephony users is statistically significantly correlated with all theexplanatory variables considered: the correlations are negative with respect to agriculture, illiteracy, and trade.The latter correlation is driven by the fact that smaller countries tend to be more open.

Fig. 2 illustrates perhaps the most interesting unconditional pairwise correlation in the data. The scatterdiagram shows the correlation between mobile telephony diffusion and average income. As suggested in someprevious studies, the correlation is rather strong.

The time-series cross-section patterns of the data in Table 6 reflect the years of mobile telephonyintroductions in Table 2, although the first annual observation of each time series is lost due to log

ARTICLE IN PRESS

12

Developing: 5

12

4

7

13

22

Developed: 34

1991 1992 1993 1994 1995 1996 1997 1998 1999 20000

10

20

30

40

Fig. 1. The average country penetration rates (user per population) of digital mobile telephony (%). Note: Including 78 developed and 123

developing countries as in Table 2.

P. Rouvinen / Telecommunications Policy 30 (2006) 46–63 51

differencing. The observations for three developed and thirty-three developing countries cannot be used in theanalysis.5

4. Estimation

Although the choice of variables in this study has been dictated by data availability, on several occasionsthere are missing values. For one variable (Illiteracy) a third and for many a fifth (Agriculture, Credit, Trade,Freedom, PCs, Fixed user cost, and Handset) of the observations are missing. Three basic alternatives havebeen employed to solve the problem of missing values: the explanatory variables with missing observations areexcluded, the cross-sectional units and/or points in time with missing observations are excluded (case-wisedelete), or some method is used to impute the missing values. The first alternative leads to a limited set ofexplanatory variables, and the second to a loss of is and/or ts. Strictly speaking the third alternative is notfeasible without additional information in cases of non-random missing observations, which is the case here.

In what follows, a fourth alternative is employed. It economizes on the use of available information withoutarbitrarily imputing missing values. Every time a missing value is encountered, it is replaced by zero. Thenovelty is, however, that a set of dummy variables is generated indicating when such replacements have beenmade. While these dummies do not have economic interpretations, they nevertheless control for the biases thatcould be introduced by an alternative imputation method.

All the regressions below are ‘fully robust’, that is, the standard deviations are heteroscedasticity consistentWhite (1980) and arbitrary dependence of observations across t (autocorrelation) is allowed. Thus, thestandard deviations are robust as long as the is are independently distributed (for discussion see Stata 2001,section 23.11). Thus, the estimations are consistent in large samples with a relatively weak set of assumptions(see, e.g., Wooldridge, 2002, Sections 7.8.1–3).

Table 7 reports the estimation results. Column 1 represents a reduced model with the completely non-missing explanatory variables. Columns 2 the full model with ‘case-wise’6 deleting the observations withmissing values, and Column 3 with the missing dummies. In some ways results in Column 1 are not entirelyunsatisfactory: all of the explanatory variables are statistically significant at conventional levels and thefindings from previous literature are largely confirmed. The developing dummy is not statistically significant(wrongly) suggesting that developing countries might not be too different. If the case-wise delete rule is applied(Col. 2), over sixty per cent of the observations are lost.

5As can be seen in Eq. (1)-(3), the model is ‘in differences’ – thus at least two observations are needed. Thus, observations for the 3

developed and 12 developing countries introducing mobile digital telephony in, and the 21 developing countries that had not done so by

2000, are lost.6‘‘Case-wise delete’’ refers to a situation where only observations with completely non-missing data are used. Thus, if the regression

includes 30 variables and the observation in question has non-missing data for 29, and missing data for 1, it is not to be included in the

regression in any way upon employing the ‘‘case-wise delete’’ rule.

ARTICLE IN PRESS

Table

3

Definitionsofthevariables

Category

Variable

Description

Type

Unit

Source

Transf.

Market

size

Population,total

Thetotalnumber

ofthecountry’sresidents

Count

1000

WDI(ITU)

Nat.log

Market

size

Population,city

Thetotalnumber

ofresidents

inthelargestcity

Count

1000

ITU

Nat.log

Wealth

Income

GDPper

capitain

USdollars

and1995prices

Ratio

US$

WDI(ITU,WB,IM

F)

Nat.log

Development

Agriculture

Valueadded

inagriculture

per

GDP

Share

%WDI

Nat.log

Development

Illiteracy

Illiterate

tothetotalpop.(ages

15andabove)

Ratio

%WDI

Nat.log

Finance

Credit

Financialresources

provided

totheprivate

sector(regardless

ofthesource)

per

GDP

Ratio

%WDI

Nat.log

Openness

Trade

Thesum

ofexportsandim

portsofgoodsandservices

per

GDP

Ratio

%WDI

Nat.log

Dem

ocracy

Freedom

Index

ofpoliticalfreedom

from

autocracy

(0)to

dem

ocracy

(100)

Index

0–100

POLIT

YIV

(FH)

Nat.loga

Technology

PCs

Self-contained

computers

designed

tobeusedbyanindividualper

capita

Ratio

%WDI(ITU)

Nat.log

Fixed

Fixed,penetr.

Fixed

telephonemainlines

connectingacustomer

tothepublicsw

itched

network

per

cap.

Ratio

%IT

U(W

DI)

Nat.log

Fixed

Fixed,usercost

Monthly

chargefor120minutesoffixed

callingwithin

thesameexchangein

USdollars

and95p.

Price

US$

ITU

(WDI)

Nat.log

Analogmob.

Analog,penetr.

Analogmobiletelephonyusers

per

cap.

Ratio

%EMC,IT

U,WDI

Nat.loga

Digitalmob.

Digital,users

Digitalmobiletelephonyusers

Count

1000

EMC,IT

U,WDI

Nat.log

Digitalmob.

Digital,prepaid

Digitalmobiletelephonycanbeaccessvia

prepaid

callingcardsin

thecountryin

question

Dummy

0,1

EMC

–

Digitalmob.

Digital,many

Digitalmobiletelephonyhasmore

thanonenetwork

standard

inthecountryin

question

Dummy

0,1

EMC

–

Digitalmob.

Digital,comp.

Digitalmobiletelephonyhastw

oormore

competingoperators

inthecountryin

question

Dummy

0,1

ITU-P

–

Both

mob.

Mobile,

usercost

Monthly

chargefor120minutesoflocalmobilepeak-tim

ecallingin

USdollars

and95p.

Price

US$

ITU

(EMC)

Nat.log

Both

mob.

Mobile,

handset

Price

level

ofinvestm

ent(a

proxyforhandsetprices)

Ratio

%PWT

Nat.log

aIn

order

toaddress

theproblem

ofzeros,thenaturallogarithm

wastaken

ofthevariable

valueplusone.

Sources:EMC¼

EMC’sworldcellulardatabase,FH¼

freedom

house

(www.freedomhouse.org)worldcountryratings,IM

F¼

IMF’sInternationalFinancialStatistics,IT

U¼

ITU’sworldtelecommunicationsindicators,IT

U-P:IT

U(2002a,b),POLIT

Y

IV:politicalregim

echaracteristics

andtransitions(M

arshallandJaggers,www.cidcm

.umd.edu/inscr/polity)data

set,PWT¼

PennWorldTable

(ver.6.1

byHeston,Summersand

Bettina,pwt.econ.upenn.edu),WB¼

WorldBank’sglobaldevelopmentnetwork

growth

database

(Easterly

andSew

adeh)macroeconomic

timeseries,WDI¼

WorldBank’sworld

developmentindicators.Ifdata

foragiven

countrywasunavailableattheprimary

source(s),itwastaken

from

thenextsourcelisted

intheparenthesis.Ifmultiplesources

are

listed

but

noneare

intheparenthesis,thedata

usedforthecountry

iattime

tisthelargestnon-m

issingvalueamongthesources

listed.


http://www.freedomhouse.org


ARTICLE IN PRESS

Table

4

Descriptivestatistics

(a)75developed

(1993–2000unbalancedpanel)

(b)90developing(1994–2000unb.panel)

Obs.

Mean

St.dev.

Min.

Max.

Obs.

Mean

St.dev.

Min.

Max.

Population,total

392

22,138

44,715

35

281,550

Pop.

335

61,677

200,447

27

1,262,460

Population,city

389

2668

3542

118,131

City

331

2993

3678

14

17,419

Income

392

16,181

10,134

2693

44,603

Inc.

335

2043

4097

84

27,681

Agriculture

265

4.89

3.58

0.10

18.52

Agric.

297

24.50

13.27

2.25

59.91

Illiteracy

190

10.05

7.69

0.27

33.16

Illit.

265

23.84

19.67

0.20

78.57

Credit

320

76.10

45.75

6.54

203.17

Credit

288

26.83

27.33

1.72

165.45

Trade

320

93.22

63.76

15.99

341.35

Trade

289

74.07

35.06

1.13

207.79

Freedom

342

82.48

31.45

0100

Free

298

57.53

30.68

5100

PCs

335

17.68

13.06

0.10

58.52

PCs

255

2.52

5.08

0.07

54.74

Fixed,penetr.

392

40.71

18.42

2.92

100

Fixed

335

11.22

15.91

0.04

85.92

Fixed,usercost

341

13.65

6.89

3.03

32.13

F.cost

268

6.01

3.89

0.23

18.22

Analog,penetr.

392

3.65

4.56

018.33

Analog

335

0.40

1.06

010.70

Digital,users

392

3220

8707

070,530

Digital

335

786

5305

084,533

Digital,prepaid

392

0.49

0.50

01

Prepaid

335

0.47

0.50

01

Digital,many

392

0.22

0.41

01

Many

335

0.29

0.45

01

Digital,comp.

392

0.72

0.45

01

Comp.

335

0.71

0.45

01

Mobile,

usercost

367

61.55

33.13

9.70

183.14

M.cost

296

51.03

34.20

0.78

276.61

Mobile,

handset

305

94.04

31.97

34.34

264.55

H-set

258

66.74

42.06

15.17

318.79

No

te:Theaboveare

thedescriptivestatisticsoftheuntransform

edvariables,thatis,naturallogarithmsorlagshavenotbeentaken

inorder

toaid

interpretation.Since

thedependent

variableisalogdifference,theinform

ationonagiven

countryisonly

usablefrom

thesecondpositiveobservationonwards.Thus,theobservationsforthe21countriesthathadnot

introduceddigitalmobiletelephonyby,andforthe15countriesintroducingitin,2000are

notusable.Theabovetable

refers

totheusable

observations.


ARTICLE IN PRESS

Table

5

Pairwisecorrelations(correlationcoefficients

multiplied

byonehundred)

Pop.

City

Inc.

Agric.

Illit.

Credit

Trade

Free

PCs

Fixed

F.cost

Analog

Digital

Prepaid

Many

Comp.

M.cost

H-set

Population,total

100

Population,city

61

100

Income

�9

�4

100

Agriculture

6�9

�59

100

Illiteracy

10

7�34

53

100

Credit

11

18

64

�53

�29

100

Trade

�21�30

16

�27

�29

23

100

Freedom

�7

741

�49

�35

33

�2

100

PCs

�10�9

85

�59

�37

57

24

45

100

Fixed,penetr.

�12�9

88

�68

�57

62

20

57

87

100

Fixed,usercost

�1

167

�48

�7

41

�12

39

58

51

100

Analog,penetr.

�3

047

�37

�30

33

�9

33

51

48

37

100

Digital,users

33

39

23

�20

�12

38

�14

15

28

22

18

6100

Digital,prepaid

28

�7

�11

�9

11

19

14

4�5

318

100

Digital,many

22

41

�8

�1

�18

17

1�3

1�12

�4

26

14

9100

Digital,comp.

13

23

�2

�17

4�2

�11

35

9�5

14

315

22

13

100

Mobile,

usercost�13�6

29

�17

�17

17

�7

26

13

25

35

6�2

�20

�10

15

100

Mobile,

handset

�20�17

39

�26

10

17

26

24

28

53

13

2�14

�18

122

100

No

te:Underliningindicatesthatthecorrelationcoefficientisstatisticallysignificantatthe1%

level.


ARTICLE IN PRESS

Table 6

Data patterns and their frequencies (i.e., countries with the pattern in question)

Developed countries/regions

Freq. Yrs. Obs. Sh. (%) 1993 1994 1995 1996 1997 1998 1999 2000

14 4 56 19 1 1 1 1

13 5 65 17 1 1 1 1 1

12 6 72 16 1 1 1 1 1 1

12 7 84 16 1 1 1 1 1 1 1

10 8 80 13 1 1 1 1 1 1 1 1

7 2 14 9 1 1

7 3 21 9 1 1 1

75 2–8 392 100

21 4 84 23 1 1 1 1

20 5 100 22 1 1 1 1 1

15 1 15 17 1

13 6 78 14 1 1 1 1 1 1

10 2 20 11 1 1

9 3 27 10 1 1 1

1 4 4 1 1 1 1 1

1 7 7 1 1 1 1 1 1 1 1

90 1–7 335 100

Note: Patterns after log-differencing the dependent variable, that is, the observations usable in the analysis.

50 50000

GDP per capita in US dollars and 1995 prices (log scale)

0.10

0.1

1

10

100

1000

Dig

ital m

obile

tele

phon

y us

ers

per

1,00

0 re

side

nts

(log

sca

le)

Fig. 2. Mobile telephony diffusion and average income in 2000. Note: Excluding the 21 countries that had not introduced digital mobile

telephony by 2000.


For example Baltagi (1995, 4) and Hsiao (2003, 311–2) have suggested that multicollinearity is less likely tobe a problem in panels, primarily because of more informative and variable data.7 As multicollinearity isnevertheless commonplace in growth regressions, it is nevertheless a potential cause of concern in the currentcontext. The variance inflation factors (VIFs) are commonly used to detect multicollinearity (Kennedy, 1998,p. 190). VIFs do not have clear critical values but a common rule-of-thumb is that individual VIFs exceeding

7Also note, however, with certain panel estimators multicollinearity there may be a particularly series problem due to the large number

of dummy variables (at least implicitly) involved.

ARTICLE IN PRESS

Table

7

FullyrobustOLSestimationsofthereducedmodel(C

ol.1),aswellasthefullmodelwiththecase-w

isedeleted

(Col.2)andfull(C

ol.3)samples(dependentvariable:logdifference

of

mobiletelephonyusers)

Variable

Column1

Column2

Column3

Coeff.

Std

err.

Sg.

Coeff.

Std

err.

Sg.

Coeff.

Std

err.

Sg.

Digital,users

a0.385

(0.042)

***

0.380

(0.040)

***

0.407

(0.044)

***

Constant

ab0

�3.119

(0.864)

***

�5.550

(1.310)

***

�2.862

(0.951)

***

Developing

country

ab1

�0.032

(0.108)

�0.260

(0.085)

***

�0.071

(0.089)

Population,total

ab2

0.305

(0.058)

***

0.332

(0.051)

***

0.339

(0.055)

***

Population,city

ab3

0.091

(0.044)

**

0.140

(0.045)

***

0.086

(0.046)

*

Income

ab4

0.165

(0.067)

**

0.132

(0.098)

0.040

(0.061)

Agrarian

ab5

0.111

(0.047)

**

�0.102

(0.051)

**

Illiteracy

ab6

�0.041

(0.031)

�0.022

(0.036)

Credit

ab7

0.009

(0.045)

0.055

(0.051)

Trade

ab8

0.294

(0.097)

***

0.083

(0.061)

Freedom

ab9

�0.020

(0.043)

�0.026

(0.036)

PCs

ab10

0.111

(0.068)

+0.022

(0.037)

Fixed,penetr.

ab11

0.203

(0.040)

***

0.111

(0.068)

+0.198

(0.054)

***

Fixed,usercost

ab12

0.098

(0.059)

+0.041

(0.038)

Analog,penetr.

ab13

0.089

(0.034)

***

0.100

(0.059)

*0.107

(0.037)

***

Digital,prepaid

ab14

0.229

(0.068)

***

0.084

(0.095)

0.241

(0.069)

***

Digital,many

ab15

�0.195

(0.083)

**

�0.128

(0.101)

�0.249

(0.074)

***

Digital,comp.

ab16

0.155

(0.075)

**

0.371

(0.092)

***

0.193

(0.077)

**

Mobile,

usercost

ab17

0.006

(0.054)

�0.009

(0.039)

Mobile,

handset

ab18

0.127

(0.099)

0.083

(0.081)

Year¼

1993

ab19

�0.020

(0.242)

�0.077

(0.170)

�0.068

(0.221)

Year¼

1994

ab20

�0.181

(0.190)

�0.472

(0.425)

�0.226

(0.184)

Year¼

1995

ab21

0.045

(0.102)

�0.099

(0.144)

0.013

(0.098)

Year¼

1997

ab22

0.123

(0.060)

**

0.144

(0.102)

0.119

(0.061)

*

Year¼

1998

ab23

0.171

(0.089)

*0.178

(0.138)

0.170

(0.085)

**

Year¼

1999

ab24

0.366

(0.101)

***

0.352

(0.138)

**

0.357

(0.097)

***

Year¼

2000

ab25

0.561

(0.121)

***

0.678

(0.184)

***

0.552

(0.114)

***

Missingdummies:Agric.;Illit.;Credit;Trade;

Free;

PCs,F.&

M.cost;H-set

Observations

727

274

727

R-squared

0.55

0.54

0.58

No.ofcountries

165

79

165

VIF

,mean

3.48

4.57

8.01

VIF

sabove10

Inc.;Fixed,PCs.

H-set

&missing;Inc.;

Trade&

missing;Fixed,Pop.

No

te:***,**,*,and+

respectivelyindicate

significance

atthe1%

,5%

,10%

,and15%

level.Standard

errors

inparentheses.



ten (Belsley, Kuh, & Welsch, 1980) and a model’s mean VIF exceeding six warrant further investigation. Ascan be seen in the bottom of Table 7, VIFs for Column 1 are below these values. VIFs for Column 2 suggestthat multicollinearity might be causing problems. Indeed, income is correlated with many of the otherexplanatory variables, particularly with fixed line penetration and PCs. VIFs for Column 3 suggest thatadditionally the missing dummies might be causing problems. Recall that multicollinearity causes thecoefficient estimates to become unstable with greatly inflated standard errors. Note, however, that variableswith large VIFs are (at least weakly) statistically significant, suggesting that multicollinearity—even thoughpresent—is not too severe. If the handset missing dummy is dropped and re-estimation is conducted of thespecification in Column 3, the mean VIF comes down to 5.80, although a few individual VIFs remain above10. Further analysis will be carried out without this dummy. Dropping income would further reduce the meanand especially the individual VIFs, but this is hardly an option; in fact all the three main variables causingproblems are essential to the model, and thus the commonly employed solution of dropping the problematicvariable(s) is not an option here.

Simply introducing the developing country dummy, as shown in Table 7, is a rather crude way of capturingthe difference between developed and developing countries. The two leftmost columns of Table 8 allow formaximum generality in this respect by performing separate regressions for the developed (Col. 1) anddeveloping (Col. 2) countries. Column 3 formally tests whether the developed and developing countries effectsare indeed different by including (a) the unaltered explanatory variables and (b) their interactions with thedeveloping country dummy. A Chow test on the joint significance of the interacted explanatory variables ishighly statistically significant confirming that the developing countries are indeed different.8

The leftmost column in Table 8 suggests, that the effects of five socio-economic and two telecommunica-tions variables, as well as two missing dummies (not shown), are statistically significantly different between thedeveloped and developing countries.9 Table 9 presents the final model (Col. 1), where this information isexploited by estimating separate coefficients for these variables and joint ones for the others. The table alsoprovides—primarily for reference—a within estimation (Col. 2) of the same parameters (when estimable).

The final results (Col. 1) suggest that the speed of adjustment (Digital, users) in digital mobile telephonydoes not differ between developed and developing countries. The absolute size of the coefficient is slightlylower than the corresponding (cross-sectional) estimates in the Internet study of Kiiski and Pohjola (2002).The market size effect (Population, total) is stronger in developing countries. The population in the largest city(Population, city), measuring the size of the largest pool of demand that can be captured with relative ease, isonly statistically significant in developed countries.10 Interestingly the wealth effect (Income) is not statisticallysignificant after controlling for the other factors, which goes against the findings in the earlier literature.11

More agrarian (Agriculture) countries have lower diffusion. Openness (Trade) boosts diffusion only in thedeveloping countries. Somewhat surprisingly a more democratic regime (Freedom) implies a lower level ofdiffusion in the developing countries. One possible explanation is that in the developing countries democracyis sometimes associated with political instability. It may also be that more authoritarian regimes take action topromote mobile diffusion because of possible military applications of the civilian technology. The overall(non-telecom) technological level seems to boost diffusion only in developing countries. The positive signs forfixed and analog mobile penetration rates indicate that network effects are present at the country level. Analogmobile penetration is a much more important factor in the developing country case. This may also be anindication of two additional factors, that is, the general economic potential for, as well as the cultural andsocial acceptability of, mobile telephony. The availability of prepaid mobile calling options is understandablymore important in developing countries, as the payment systems needed for subscription services are likely tobe less sophisticated. Standards competition (Digital, many) hinders and market competition (Digital, comp.)promotes diffusion. The rising and (mostly) statistically significant coefficients of the time dummies areinterpreted as an indication of worldwide network effects and economies of scale. With a globally expanding

8To be precise, a more robust Wald test using the generalized variance-covariance matrix is performed: F(32, 164) ¼ 3.26, i.e.,

Probability 4 F ¼ 0.0000.9As this test is slightly biased against finding differences, statistical significance at the 15% level is the criterion employed here.10This may be an indication of the fact that in the developing countries the size of the largest city sometimes measures the size of the

slums rather than the potential user base.11One should nevertheless keep in mind the possible multicolinearity problem discussed above.

ARTICLE IN PRESS

Table

8

Fullyrobust

OLSestimationsofthefullmodel

withthedeveloped

(Col.1)anddeveloping(C

ol.2)countrysubsamples,aswellaswiththefullsample

andthedevelopingdummy

interactions(C

ol.3)

Variable

Column1

Column2

Column3

Developed

Developing

(a)Unaltered

variabl.

(b)Developinginteract.

Coeff.

Std

err.

Sg.

Coeff.

Std

err.

Sg.

Coeff.

Std

err.

Sg.

Coeff.

Std

err.

Sg.

Digital,users

a0.412

(0.032)

***

0.449

(0.088)

***

0.411

(0.032)

***

0.038

(0.088)

Constant

ab0

�3.175

(0.898)

***

�2.976

(1.131)

**

�3.061

(0.750)

***

Population,total

ab2

0.284

(0.051)

***

0.482

(0.097)

***

0.283

(0.050)

***

0.201

(0.101)

**

Population,city

ab3

0.143

(0.041)

***

�0.062

(0.058)

0.143

(0.041)

***

�0.204

(0.072)

***

Income

ab4

0.129

(0.086)

+0.046

(0.090)

0.121

(0.079)

+�0.069

(0.088)

Agrarian

ab5

�0.024

(0.048)

�0.130

(0.101)

�0.028

(0.045)

�0.098

(0.104)

Illiteracy

ab6

�0.007

(0.041)

�0.013

(0.050)

�0.008

(0.041)

�0.006

(0.065)

Credit

ab7

0.099

(0.050)

*0.039

(0.072)

0.099

(0.050)

**

�0.059

(0.086)

Trade

ab8

0.021

(0.088)

0.189

(0.073)

**

0.015

(0.083)

0.177

(0.104)

*

Freedom

ab9

�0.007

(0.043)

�0.121

(0.063)

*�0.009

(0.042)

�0.111

(0.075)

+

PCs

ab10

�0.107

(0.075)

0.092

(0.045)

**

�0.106

(0.073)

+0.197

(0.086)

**

Fixed,penetr.

ab11

0.252

(0.091)

***

0.129

(0.069)

*0.255

(0.088)

***

�0.127

(0.112)

Fixed,usercost

ab12

�0.039

(0.080)

0.040

(0.046)

�0.037

(0.080)

0.077

(0.093)

Analog,penetr.

ab13

0.084

(0.035)

**

0.540

(0.128)

***

0.085

(0.035)

**

0.453

(0.132)

***

Digital,prepaid

ab14

0.120

(0.067)

*0.331

(0.115)

***

0.120

(0.067)

*0.211

(0.133)

+

Digital,many

ab15

�0.194

(0.099)

*�0.355

(0.140)

**

�0.194

(0.099)

*�0.161

(0.171)

Digital,comp.

ab16

0.236

(0.081)

***

0.222

(0.105)

**

0.236

(0.081)

***

�0.013

(0.131)

Mobile,

usercost

ab17

0.015

(0.068)

�0.015

(0.057)

0.014

(0.068)

�0.028

(0.091)

Mobile,

handset

ab18

0.021

(0.023)

0.023

(0.038)

0.021

(0.023)

0.001

(0.045)

Year¼

1993

ab19

�0.151

(0.231)

�0.149

(0.231)

Year¼

1994

ab20

�0.335

(0.174)

*�0.160

(0.182)

�0.334

(0.173)

*0.174

(0.252)

Year¼

1995

ab21

�0.072

(0.091)

0.006

(0.236)

�0.072

(0.091)

0.077

(0.254)

Year¼

1997

ab22

0.111

(0.068)

+0.170

(0.105)

+0.109

(0.069)

+0.063

(0.117)

Year¼

1998

ab23

0.207

(0.096)

**

0.301

(0.177)

*0.204

(0.098)

**

0.100

(0.182)

Year¼

1999

ab24

0.441

(0.103)

***

0.469

(0.202)

**

0.437

(0.103)

***

0.037

(0.195)

Year¼

2000

ab25

0.585

(0.108)

***

0.727

(0.236)

***

0.579

(0.108)

***

0.153

(0.221)

Missingdummies:Agric.;Illit.;Credit;Trade;

Free;

PCs;F.cost;M.cost;H-set.

(Alsotheinteracted

versionsofthemissingdummiesare

included

inColumn3).

Observations

392

335

727

R-squared

0.72

0.56

0.64

No.ofcountries

75

90

165

No

te:***,**,*,and+respectivelyindicate

significance

atthe1%

,5%

,10%

,and15%

level.Standard

errors

inparentheses.


ARTICLE IN PRESS

Table

9

FullyrobustOLS(C

ol.1)andwithin

(Col.2)estimationofthefinalmodelwithseparate

coefficientsforthedeveloped

anddevelopingcountriesifapplicable(dependentvariable:log

difference

ofmobiletelephonyusers)

Variable

Column1(O

LS)

Column2(w

ithin)

Joint/developed

Developing

Joint/developed

Developing

Coeff.

Std

err.

Sg.

Coeff.

Std

dev.

Sg.

Coeff.

Std

err.

Sg.

Coeff.

Std

err.

Sg.

Digital,users

a0.422

(0.042)

***

0.636

(0.022)

***

Constant

ab0

�2.588

(0.696)

***

�18.570

(8.935)

**

Population,total

ab2

0.303

(0.049)

***

0.424

(0.068)

***

0.439

(1.050)

3.464

(1.553)

**

Population,city

ab3

0.156

(0.040)

***

�0.044

(0.058)

�0.180

(0.521)

0.656

(0.410)

+

Income

ab4

0.044

(0.056)

0.224

(0.116)

*

Agrarian

ab5

�0.069

(0.041)

*0.017

(0.108)

Illiteracy

ab6

�0.008

(0.032)

�0.200

(0.501)

Credit

ab7

0.058

(0.040)

0.008

(0.085)

Trade

ab8

0.049

(0.068)

0.154

(0.060)

**

0.461

(0.194)

**

0.419

(0.188)

**

Freedom

ab9

0.019

(0.041)

�0.138

(0.059)

**

0.481

(0.568)

�0.069

(0.117)

PCs

ab10

0.002

(0.040)

0.068

(0.035)

*�0.018

(0.068)

0.097

(0.049)

*

Fixed,penetr.

ab11

0.160

(0.055)

***

�0.490

(0.201)

**

Fixed,usercost

ab12

0.041

(0.038)

�0.029

(0.053)

Analog,penetr.

ab13

0.103

(0.037)

***

0.506

(0.120)

***

0.406

(0.069)

***

0.653

(0.258)

**

Digital,prepaid

ab14

0.108

(0.071)

+0.325

(0.079)

***

0.139

(0.074)

*0.146

(0.082)

*

Digital,many

ab15

�0.292

(0.076)

***

0.003

(0.140)

Digital,comp.

ab16

0.219

(0.069)

***

Mobile,

usercost

ab17

�0.015

(0.039)

�0.025

(0.050)

Mobile,

handset

ab18

0.017

(0.022)

0.018

(0.027)

Year¼

1993

ab19

�0.231

(0.209)

�1.176

(0.186)

***

Year¼

1994

ab20

�0.322

(0.173)

*�0.849

(0.125)

***

Year¼

1995

ab21

�0.043

(0.094)

�0.283

(0.087)

***

Year¼

1997

ab22

0.124

(0.058)

**

0.313

(0.066)

***

Year¼

1998

ab23

0.217

(0.087)

**

0.633

(0.086)

***

Year¼

1999

ab24

0.403

(0.101)

***

1.022

(0.110)

***

Year¼

2000

ab25

0.599

(0.118)

***

1.457

(0.138)

***

Jointmissingdummies:Illit.;Credit;

Jointmissingdummies:Credit;Trade;

Trade;



Separate

missingdummies:Agric.,Free.

Developed

missingdummy:Agric.

Observations

727

727

R-squared

0.63

0.76

No.ofcountries

165

165

No


significance

atthe1%

,5%

,10%,and15%

level.Standard

errors

inparentheses.


ARTICLE IN PRESSP. Rouvinen / Telecommunications Policy 30 (2006) 46–6360

user base, the quality-adjusted12 real prices of both digital mobile telephony network equipment and handsetshave dropped constantly, and simultaneously uncertainties with respect to standards and dominant designshave diminished. Indeed, late entrants will experience more rapid diffusion once other things have beenaccounted for.

The results in Column 1 of Table 9 are largely driven by cross-country variation in the data. In Column 2 ofthe table, this variation has been removed by employing the ‘within’ estimator. A few interesting differencesemerge. First, income effects are now observed. Both population parameters are positive and significant(although the largest city only at the 15% level), but only for developing countries. One possible interpretationis that it pays to have a dynamic (growing) economy, especially for a developing country. Secondly, increasingopenness of the economy seems to have a positive and equally important effect for both types of countries.Thirdly, the coefficient for fixed telecommunications usage is now significant and negative indicatingsubstitution from fixed to mobile.

The term ‘developing country’ typically refers to a nation that has not achieved a high degree ofindustrialization and/or does not enjoy a high standard of living, which often refers to the number of goodsand services provided but may also also refer to the life expectancy or general health of the inhabitants. So farthis paper has employed an exogenous and generally accepted definition, that is, a developing country hasbeen defined as a low or lower-middle income country in the 2002 edition of the World Bank’s WorldDevelopment Indicators. The possible consequences of this choice are studied in Table 10 by re-estimating thefinal model with alternative definitions of a developing country.13 Developing countries are defined as follows:Column 2—countries having an above average ratio of agricultural value added to GDP, Column 3—countries having commercial energy use per capita in the lower quartile, Column 4—countries having a belowaverage share of newborns immunized against measles, Columns 5 and 6—countries that are above themedian GDP per capita (Income) in the beginning or/and at the end of the observation period,14 and Column7—countries having a below mean factor score derived via a principal factor analysis of the four variablesunderlying definitions in Columns 2–6. In Columns 2–4, countries having missing values are defined asdeveloping.15 Column 1 replicates some of the results in Table 9 and is provided for reference.

The results in Table 10 indicate that qualitatively the results remain largely the same regardless of thedeveloping country definition employed. Note, however, that the largest city is not consistently significant forthe developed countries and the analog penetration is not consistently significant for the developing countries.There are also some indications of positive income effects and benefits of financial development, as well assigns of economic substitution from fixed to mobile.

5. Conclusion

Developing countries are indeed different when it comes to the diffusion of digital mobile telephony, but notin the most obvious ways. The speed of adjustment is not too different from their developed worldcounterparts, and the income effect is not the primary driver of differences in diffusion patterns but ratherrelated factors.

As the continuously rising and significant time dummies reveal, a country benefits from being a late entrantin digital mobile telephony, arguably because the earlier adopters have carried the burdens of accumulatingglobal critical mass and resolving uncertainties related to standards and dominant designs. As developingcountries are typically late adopters, they are the main beneficiaries of this phenomenon.

It seems quite natural that in a developing country case having a large potential user base is more important,as the average revenue per user is likely to be lower. Also network effects play a more important role indeveloping countries, possibly because infrastructure and logistics are, at least in relative terms, moreexpensive to build and maintain.

12Availability may in this context considered to be one aspect of quality.13The need for these re-estimations was kindly pointed out by an anonymous referee.14There may well be countries that are at quite similar stages of development but nevertheless fall under different categories upon

employing a simple criteria with a given threshold value. There may also be countries that cross that threshold during the observation

period. Columns 5 and 6 attempt to study the effects of this by classifying the rising and falling countries differently.15The shares of the missing observations are: Column 2, 14%; Column 3, 36%; and Column 4, 22%.

ARTICLE IN PRESS

Table

10

Fullyrobust

OLSestimationofthefinalmodel

withalternativedevelopingcountrydefinitions(dependentvariable:logdifference

ofmobiletelephonyusers)

Official

Agric.

Energy

Immuniz.

Start/end

Start+

end

Factor

Joint:digital,users

a0.422

***

0.414

***

0.411

***

0.405

***

0.418

***

0.417

***

0.410

***

Joint:constant

ab0

�2.588

***

�3.076

***

�2.958

***

�3.094

***

�2.302

***

�2.597

***

�2.667

***

D-ped:pop.,total

ab2

0.303

***

0.348

***

0.370

***

0.363

***

0.332

***

0.297

***

0.311

***

D-ping:pop.,total

0.424

***

0.369

***

0.346

***

0.328

***

0.412

***

0.428

***

0.400

***

D-ped:pop.,city

ab3

0.156

***

0.080

*0.060

0.074

0.121

**

0.141

***

0.126

**

D-ping:pop.,city

�0.044

0.043

0.085

+0.089

+�0.037

�0.038

�0.003

Joint:income

ab4

0.044

0.065

0.105

*0.086

0.012

0.038

0.058

Joint:agrarian

ab5

�0.069

*�0.079

*�0.108

**

�0.100

**

�0.091

**

�0.095

**

�0.093

*

Joint:Illiteracy

ab6

�0.008

�0.018

�0.018

�0.010

0.009

�0.008

0.006

Joint:credit

ab7

0.058

0.063

0.054

0.040

0.053

0.067

*0.047

D-ped:trade

ab8

0.049

0.074

0.062

0.076

0.028

0.040

0.027

D-ping:trade

0.154

**

0.125

**

0.132

**

0.131

**

0.174

***

0.158

***

0.167

***

D-ped:freedom

ab9

0.019

0.017

�0.042

�0.024

0.035

0.039

0.041

D-ping:freedom

�0.138

**

�0.059

�0.101

+�0.041

�0.147

**

�0.139

**

�0.140

**

D-ped:PCs

ab10

0.002

0.021

�0.027

0.001

0.023

0.024

0.031

D-ping:PCs

0.068

*0.032

0.080

*0.035

0.091

**

0.118

**

0.109

**

Joint:fixed,

penetr.

ab11

0.160

***

0.175

***

0.172

***

0.178

***

0.154

***

0.139

***

0.152

***

Joint:fixed,userc.

ab12

0.041

0.050

0.035

0.054

0.019

0.062

+0.033

D-ped:analog,

pen.

ab13

0.103

***

0.141

***

0.191

***

0.150

***

0.083

**

0.106

***

0.096

**

D-ping:analog,

pen.

0.506

***

0.026

�0.003

0.083

0.403

***

0.462

***

0.481

***

D-ped:dig.,

prepaid

ab14

0.108

+0.162

**

0.215

***

0.232

***

0.112

*0.152

**

0.141

**

D-ping:dig.,

prepaid

0.325

***

0.362

***

0.279

***

0.234

**

0.383

***

0.347

***

0.337

***

Joint:Digital,

many

ab15

�0.292

***

�0.272

***

�0.279

***

�0.297

***

�0.275

***

�0.312

***

�0.311

***

Joint:digital,

comp.

ab16

0.219

***

0.206

***

0.254

***

0.195

**

0.196

***

0.211

***

0.191

**

Joint:mob.,user

c.

ab17

�0.015

�0.006

�0.029

�0.015

�0.006

0.004

�0.005

Joint:mob.,h-set

ab18

0.017

0.029

0.023

0.036

0.024

0.015

0.022

Thetimedummies(Y

ear¼

y)are

included

butnotreported

intheinterest

ofspace.

Jointmissingdummies:Illit.;Credit;Trade;


Separate

missingdummies:Agric.,Free.

Observations

727

727

727

727

727

727

727

R-squared

0.63

0.59

0.60

0.59

0.62

0.61

0.60

No.ofcountries

165

165

165

165

165

165

165

No


significance

atthe1%

,5%

,10%,and15%

level.Standard

errors

inparentheses.


ARTICLE IN PRESSP. Rouvinen / Telecommunications Policy 30 (2006) 46–6362

More open developing countries tend to have higher diffusion rates. It seems plausible to argue thatopenness drives a need for advanced real-time communication. In some cases openness may also be related toa large volume of foreign roaming customers that are very lucrative to network operators and thus provide anadditional incentive for network expansion.

The finding that developing countries with more democratic regimes seem to have lower diffusion rates isperhaps somewhat puzzling. As speculated above, it might be that the variable is also standing in for otherthings, such as political instability or the communication needs of the military supporting an authoritarianregime. Further investigation of the issue is called for.

In developing countries high diffusion of non-telecom technology is associated with high mobilepenetration. Besides controlling for the general level of technological infrastructure, the measure may alsocapture social and cultural acceptability of technology, which especially in developing countries cannot betaken for granted. It is also consistent with the existence of technological complementarities.

For both developed and developing countries, standards competition hinders and market competitionpromotes diffusion. The coefficient estimates would seem to suggest that standards competition is moredetrimental in a developing country, although the difference between the developed and developing countriesis not statistically significant. It is also interesting to note that even if one were to estimate separate marketcompetition coefficients, they would be nearly identical. Thus, competing for customers is indeed equallybeneficial in both developed and developing countries, even though the competition contexts are often wildlydifferent.

Acknowledgements

Part of the Wireless Communication Research Program (brie-etla.org) of BRIE, the Berkeley Roundtableon the International Economy at the University of California at Berkeley, and ETLA, the Research Instituteof the Finnish Economy. The author would like to thank Derek Jones, Heli Koski, Tobias Kretschmer, MattiPohjola, Pekka Yla-Anttila, and the participants of the UNU/WIDER Conference on the New Economy inDevelopment for comments and suggestions. This work was completed while the author was a Jean MonnetFellow at the European University Institute in Florence, Italy. The hospitality of the Robert Schuman Centrefor Advanced Studies, as well as the financial support of the Academy of Finland and the Yrjo JahnssonFoundation, are gratefully acknowledged.

References

Ahn, H., & Lee, M.-H. (1999). An Econometric Analysis of the Demand for Access to Mobile Telephone Networks. Information

Economics and Policy, 11(3), 297–305.

Baltagi, B. H. (1995). Econometric Analysis of Panel Data. Chichester, UK: Wiley.

Belsley, D. A., Kuh, E., & Welsch, R. E. (1980). Regression Diagnostics: Identifying Influential Data and Sources of Collinearity.

New York: Wiley.

Burki, A. A., & Aslam, S. (2000). The Role of Digital Technology and Regulations in the Diffusion of Mobile Phones in Asia. Pakistan

Development Review, 39(4), 741–748.

Clarke, G., & Wallsten, S. (2004). Has the Internet Increased Trade? Evidence from Industrial and Developing Countries. The World Bank

Policy Research Working Paper, 3215.

Dekimpe, M. G., Parker, P. M., & Sarvary, M. (1998). Staged Estimation of International Diffusion Models: An Application to Global

Cellular Telephone Adoption. Technological Forecasting and Social Change, 57(1-2), 105–132.

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., & Verboven, F. (2001a). The Diffusion of Mobile Telecommunications Services in the European Union Countries. European

Economic Review, 45(3), 577–588.

Gruber, H., & Verboven, F. (2001b). The Evolution of Markets under Entry and Standards Regulation—The Case of Global Mobile

Telecommunications. International Journal of Industrial Organization, 19(7), 1189–1212.

Hsiao, C. (2003). Analysis of Panel Data (2nd Ed.). Cambridge, UK: Cambridge University Press.

ITU. (2002a). Trends in Telecom Reform.Geneva: International Telecommunication Union.

ITU. (2002b). World Development Report 2002 – Reinventing Telecoms.Geneva: International Telecommunication Union.

Kennedy, P. (1998). A Guide to Econometrics (4th ed). Cambridge, MA: MIT Press.


Kenny, C. (2003). The Internet and Economic Growth in Less-developed Countries: A Case of Managing Expectations? Oxford

Development Studies, 31(1), 99–113.

Kiiski, S., & Pohjola, M. (2002). Cross-Country Diffusion of the Internet. Information Economics and Policy, 14(2), 297–310.

Koski, H., Kretschmer, T. (2002). Entry, Standards, and Competition: Firm Strategies and the Diffusion of Mobile Telephony. ETLA

Discussion Papers, 824.

Liikanen, J., Stoneman, P., & Toivanen, O. (2001). Intergenerational Effects in the Diffusion of New Technology: The Case of Mobile

Phones. Swedish School of Economics and Business Administration Working Papers, 3.

Lucas, H. C., Jr., & Sylla, R. (2003). The Global Impact of the Internet: Widening the Economic Gap Between Wealthy and Poor Nations?

Prometheus, 21(1), 20–22.

Madden, G., & Coble-Neal, G. (2001). Internet Forecasting and the Economics of Networks. In D. G. Loomis, & L. D. Taylor (Eds.),

Forecasting the Internet: Understanding the Data Communications Revolution. Boston: Kluwer Academic Publishers.

Madden, G., Coble-Neal, G., & Dalzell, B. (2004). A Dynamic Model of Mobile Telephony Subscription Incorporating a Network Effect.

Telecommunications Policy, 28(2), 133–144.

Petrazzini, B. (1999). The Internet in Developing Countries. Communications of the ACM, 42(6), 31–36.

Qureshi, S. (2003). Movement in the information technology for development debate: How can it meet the challenges of global

competition? Information Technology for Development, 10(3), 147–149.

Stata. (2001). Stata Statistical Software: Release 7.0 User’s Guide (software manual). College Station, TX: Stata Corporation.

Stoneman, P. (1983). The Economic Analysis of Technological Change. Oxford, UK: Oxford University Press.

Stoneman, P. (2002). The Economics of Technological Diffusion. Oxford, UK: Blackwell Publishers.

White, H. (1980). A Heteroscedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroscedasticity. Econometrica,

48, 817–838.

Wooldridge, J. M. (2002). Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT Press.

Diffusion of digital mobile telephony: Are developing countries different?

Documents

Transcript of Diffusion of digital mobile telephony: Are developing countries different?