measurement Internet

Advancement of indices assessing a nation'stelecommunications development status:A PLS structuralequation analysis of over100countriesTorsten J.Gerpott n,1,Nima Ahmadi1Mercator School of Management,University of Duisburg-Essen,Lotharstr.65,47057Duisburg,Germanya r t i c l e i n f oAvailable online3February2015Keywords:AdoptionComposite indicesCountry rankingITU data basePartial Least Squares structural equationmodelingSupplyUsagea b s t r a c tThe literature has already presented quite a number of composite metrics seeking toquantify the achievement level of countries with regard to their telecommunicationinfrastructures and their use by end customers.However,prior measures leave room forextensions.Therefore,this paper introduces a new second-order overall index assessingthe availability,adoption and usage intensity of telecommunication networks and servicesat the country level.The proposed Telecommunications Development Index(TDI)integrates11indicators,which are grouped into three first-level subindices labeledSupply,Adoption and Usage.Indicator and subindex weights used in combining theminto higher-order measures are computed drawing on Partial Least Squares(PLS)structural equation modeling(SEM)techniques.The modeling rests on the assumptionthat TDI weights should be outcome-specific and thus constructed in a way thatmaximizes the indicators'and subindices'capability,respectively to predict sociallydesirable national performance criteria such as per capita gross domestic product(GDP)change or a country's Human Development Index(HDI)value.Based on data from111countries,this approach was used to calculate weights of the indicators and subindicesmerged into a total TDI predicting either GDP per capita change or a country's HDI.In bothTDI variants the Adoption subindex variable fixed broadband subscriptions per householdand the Supply subindex variable international bandwidth capacity available per Internetuser achieved the highest relative weights of all indicators in the prediction of the twostudied social outcome criteria.The PLS SEM results were used to calculate scores of thetwo TDI variations for each of the sample countries.TDI-based rankings of the countrieswere compared with a ranking derived from the Information and CommunicationTechnologies Development Index of the International Telecommunication Union.&2014Elsevier Ltd.All rights reserved.1.IntroductionIt is widely acknowledged that the availability and use of telecommunication networks and services are strongly linked to positively valued socio-economic achievements of countries.Consequently,several studies have been carried out which strive to capture the degree of sophistication of the telecommunications infrastructure and demand status at the countryContents lists available at ScienceDirectURL:/locate/telpolTelecommunications Policy/10.1016/j.telpol.2014.11.0070308-5961/&2014Elsevier Ltd.All rightsreserved.n Corresponding author.E-mail addresses:torsten.gerpott@uni-due.de(T.J.Gerpott),nima.ahmadi@uni-due.de(N.Ahmadi).1Tel.:þ492033793109;fax:þ492033792656.Telecommunications Policy39(2015)93–111level for a large number of nations in a single composite index.In the remainder of this study the term “telecommunication development index ”(TDI)is used to refer to such measures of a latent construct which encompasses several features.Among others,Barzilai-Nahon (2006,pp.274–275),Bruno,Esposito,Genovese,and Gwebu (2011,pp.18–21)and Hanafizadeh,Hanafizadeh,and Khodabakhshi (2009,p.245)provide overviews of relevant index propositions.A common key characteristic of the various TDI is that they mathematically aggregate a set of distinct indicators either in a single step or in at least two steps through the calculation of subindices.Furthermore,at the lowest level of analysis the individual indicators typically do not have a uniform measurement unit and an obvious way of weighting the indicators or higher-order subindices in the computation of the total index is missing (Saisana &Tarantola,2002,p.5;Vicente Cuervo &López Menéndez,2006,p.758).Politicians and executives of telecommunication firms use TDI to justify the inevitability or needlessness of a variety of public interventions as,for instance,state subsidies for the roll-out of fiber access networks in rural areas,financial aids for low income households to hook them up to the Internet or the imposition of ceilings for telephone line installation fees paid by residential customers (Al-mutawkkil,Heshmati,&Hwang,2009,pp.176–177).Scholars focusing on the telecommunications sector are more interested in the conceptual underpinnings and in methodological aspects (dimensionality,weighting algorithm etc.)of composite measures assessing the telecommunica-tions development (TD)level of countries.Regardless of who relies on TDI and of what is intended to be achieved through TDI analysis,capturing the status of the availability and use of telecommunication infrastructures and services at the country level in a single composite index is a highly complex endeavor.This holds even more if the index designer strives to base the measure on raw data which are accessible not just for a few (developed)nations but for a large number of countries.In particular,the selection of variables (“indicators ”)to be entered in a TDI and the weighting of indicators or of subindices introduced at intermediate levels between the basic variables and the overall index raise a number of intriguing questions.Although these questions have been addressed in quite a number of publications there is still room for improved responses to them.The present work intends to contribute to such improved answers.It extends the literature mainly in three ways.First,we develop a proposal of a focused set of 11indicators and their summary in three first-level subindices which in turn are merged in a (second-order)overall TDI.In spite of the set's parsimony in terms of the number of considered variables,it covers the supply of telecommunication infrastructures and services in a country,their adoption by end customers as well as the intensity of usage of access lines.These three facets are less well amalgamated in earlier measures assessing a country's TD status.Second,this study considers that first-order TD subindicies constructed in earlier research (e.g.,Al-mutawkkil et al.,2009;Hanafizadeh,Saghaei,&Hanafizadeh,2009;ITU,2013;Waverman,Dasgupta,&Rajala,2011)are linked in means-end-chains.For instance,a subindex measuring the take-up rate of network access options has impacts on the values of another subindex capturing a country's use situation.Hence,in deriving appropriate weights of TDI indicators or subindices such functional relationships need to be taken into account.Third,this investigation makes the point that weights of indicators or subindices entered into an overall TDI should vary depending on the socio-economic target criterion,which is to be predicted by the index.The most prominent instance for such a criterion is a country's growth in its gross domestic product per capita.The criterion-related weighting is achieved by applying a variance-based structural equation modeling (SEM)technique –Partial Least Squares (PLS)–to empirically estimate TDI scores of 111countries for two societal outcome criteria.The remainder of this article is organized as follows:The next section reviews the literature on assessing a country's TD level in a composite index in order to explain potential areas of improvement with respect to the content areas covered in a TDI,the consideration of relations between its first-level subindices and the derivation of weights at the detailed indicator and at the subindex level.Section 3describes the structure of the subindices merged in the TDI proposed here.Furthermore,it details the data sources and indicators used in the empirical construction of three subindices for two TDI each of which aims at predicting a specific desirable societal outcome criterion.Section 4reports the results of the empirical estimation of indicator and subindex weights in the overall TDI based on a PLS SEM modeling approach.Final conclusions and suggestions for future research are presented in Section 5.2.Literature review2.1.Indicator selection issuesFrom a high level perspective,two conceptual avenues to selecting indicators for measuring the state of a country's telecommunications infrastructure availability and use are distinguishable.First,one strand of research views the TD status as part of a broader subject area dealing with the adoption and usage of information and communication technologies (ICT)in consumer households,corporations and the public sector.This type of work seeks to capture a country's “ICT-ization or Infostate ”(Orbicom,2003,p.IX ),level of “ICT development ”(ITU,2013,p.17),“digitization ”(Katz &Koutroumpis,2013,p.314),“digitalization level ”(Billon,Lera-Lopez,&Mario,2010,p.46),“level of ICT adoption ”(Billon,Marco,&Lera-Lopez,2009,p.596),“information society level ”(Çilan,Bolat,&Co şkun,2009,p.98)or “readiness …to use ICTs ”(Bilbao-Osorio,Crotti,Dutta,&Lanvin,2014,p.6)by choosing indicators that are not limited to the roll-out of telecommunications infrastructures and the uptake of access or service offers by end customers.Rather,this research stream also incorporates variables,which aim at capturing human skills or preconditions for the effective use of ICT in private and business contexts (e.g.,enrollment in a specific level of education as a percentage of the official school-age population or labor force;see ITU,T.J.Gerpott,N.Ahmadi /Telecommunications Policy 39(2015)93–11194T.J.Gerpott,N.Ahmadi/Telecommunications Policy39(2015)93–11195 2013,p.211;Katz&Koutroumpis,2013,p.316;World Wide Web Foundation,2013,p.9).Furthermore,it places heavy emphasis on the proliferation of computer hardware(e.g.,secure Internet servers),individual access to the Internet and inequalities in ICT/Internet use rates between countries or various population groups which are typically summarized with the term“digital divide”(Billon et al.,2010,p.41).Probably the most prominent example for this approach is the“ICT Development Index(IDI)”developed by the International Telecommunication Union(ITU)in2008(see ITU,2013,pp.17–22,209–215).The IDI is the weighted average of three subindices labeled“ICT access”composed of five equally weighted variables,“ICT use”formed by three equally weighted variables and“ICT skills”built by three equally weighted indicators.In calculating the overall IDI the weights of the first two subindices amount to0.4each and the weight of the skills subindex is0.2.Critics of the IDI argue that the measure mixes independent inputs,such as the proportion of households with a computer,with dependent outputs as for instance,the proportion of the population using the Internet which results in misleading double-counting or-weighting, respectively of indicators.Additionally,they highlight that the IDI provides no convincing justification for assigning some subscription variables to the ICT access subindex(e.g.,mobile cellular telephone subscriptions per100inhabitants)and others as the number of wireless broadband subscriptions per100people to the ICT use subindex(James,2012).A second study group deliberately focuses its assessment efforts on the availability,uptake and use of telecommunication networks and services.For instance,Al-mutawkkil et al.(2009)proposed a composite“Telecommunication Index”which is computed as the unweighted average of a fixed telephone network subindex(based on five indicators such as main telephone lines per1000population),an Internet subindex(containing four variables like Internet subscribers per1000 inhabitants)and a mobile network subindex(combining three indicators as,e.g.,cellular mobile telephone subscribers per 1000population).A key disadvantage of the first research stream is that the inclusion of a broad variety of indicators and subindices, respectively and particularly of computer hardware as well as human(ICT)skill level variables taken as proxies for preconditions of the effective IT use blurs the meaning of a composite index(Al-mutawkkil et al.,2009,p.178;Ferreira, 2008,p.15;James,2012,p.592).This in turn makes it very hard to derive conclusions regarding economic or technology policy actions,which may be needed to improve the position of a nation's telecommunications situation relative to other countries.Therefore,the present investigation follows the second research stream by designing a composite measure,which concentrates on telecommunication networks and services use.Such an attempt is worthwhile because earlier TDI suggested in the literature often had somewhat deviating thrusts and leave room for improvement in at least three areas.First,they mostly miss important facets or rely on questionable indicators of a country's TD status.Although it is undisputed that the physical roll-out of telecommunication infrastructure and the supply of transmission capacity are necessary prerequisites for the end customer uptake of network access offerings and service usage,very few TDI contain variables assessing the deployment of networks and their data transport capabilities.Positive exceptions are Bilbao-Osorio et al.(2014),Katz and Koutroumpis(2013),and Hanafizadeh, Hanafizadeh et al.(2009)who included the percentage of a country's population covered by mobile cellular networks in a synthetic ICT status index and Waverman et al.(2011)who added the percentage of a country's inhabitants covered by third generation(3G)mobile networks to an index of“useful connectivity”.Acknowledging the pivotal role of physical deployment and transmission capacity for subsequent access take-up and service usage,the present investigation selected additional indicators of the supply side availability of networks and transport capacity regardless of whether the deployed production facilities result in end customer access and service demand.In addition,many past TD measures treat network use as a binarily scaled phenomenon,which is either present or not. For instance,the use subindex in the IDI of the ITU contains three variables which measure use by looking at subscription rates of fixed and wireless broadband access lines(ITU,2013,p.21).However,there is ample evidence suggesting that the average usage intensity of telecommunication services varies strongly between countries and their residents(Gerpott,2010, p.62;ITU,2011a,pp.29–246).Hence,past use operationalizations overwhelmingly neglect that use is better measured in the sense of usage intensity based on steadily-scaled indicators(in particular frequency,minutes or volume of usage;cf. Barzilai-Nahon,2006,p.274).Among the few positive outliers are Orbicom(2003,p.127)and Waverman et al.(2011,pp.48, 53)who incorporated national fixed or mobile voice minutes per capita or Katz and Koutroumpis(2013,p.316)who included SMS sent per capita in their composite TD measures.Therefore,this study extends most relevant earlier work by including several variables capturing the intensity of usage of a general telecommunication service category per access line or person.Second,numerous extant TD measures enter price data for certain telecommunication services or revenues generated from them as proxies for“affordability”into a composite index(e.g.,Al-mutawkkil et al.,2009;Bilbao-Osorio et al.,2014; Hanafizadeh,Saghaei et al.,2009;Trkman,Blazic,&Turk,2008;Vicente Cuervo&López Menéndez,2006).This procedure is not convincing because prices are drivers of access take-up rates and intensity of service usage and revenues are mathematically dependent on prices and subscription numbers(Ferreira,2008,pp.8–10).Thus,merging price and subscription indicators into a single composite index leads to double-counting/-weighting of certain features of a country's TD situation.Furthermore,cross-national comparisons of price levels for baskets of telecommunication services raise difficult and controversial methodological issues(e.g.,basket composition,operator selection).They complicate the interpretation of a TDI and may hamper its acceptance by decision makers both in the public and the private sector (Waverman et al.,2011,pp.41–42).These considerations are confirmed by the fact that in2009the ITU dropped tariff variables from its current ICT status index.Instead,it discusses price data and affordability of fixed and mobile networkservices independent of its overall IDI (ITU,2009,pp.11–12,2013,pp.77–126).Hence,the present work avoids a problem of earlier composite TD assessments by not relying on price and revenue related variables.Third,TDI are typically construed in a way which fails to account for means-ends-relationships between non-financial indicators or subindices merged in an overall aggregated measure.This defect results in accidental over-or underweightings of indicators or subindices merged in an overall index because one variable strongly affects the lower boundary of the values of another variable (Hanafizadeh,Saghaei et al.,2009,p.387;James,2012,pp.588–589).For instance,the United Nations'Department of Economic and Social Affairs introduced a “telecommunication infrastructure index ”which selects (among others)“fixed internet subscriptions per 100inhabitants ”and “internet users per 100inhabitants ”as indicators (United Nations,2012,p.124).Regrettably,the value of the second variable is conditional on the number of Internet subscriptions.Thus,adding these two indicators with other variables,which are at best weakly associated with Internet access adoption and use (e.g.,mobile network subscription rate),leads to an overweighting of Internet-related access and use facets in the infrastructure index developed by the United Nations.Conversely,in case that one subscale of a composite TDI has strong positive effects on another subindex the neglect of this functional chain in deriving weights for the various subindices of an overall measure results in an underweighting of the subscales located at the initial parts of the causal chain.For instance,regarding the IDI of the ITU Barzilai-Nahon (2006,pp.273–275)indicates that its access subindex influences its use subscale.However,neither the ITU nor a recent analysis of Kyriakidou,Michalakelis,and Sphicopoulos (2013),who use covariance-based SEM techniques to design a modified version of the IDI,account for this linkage in suggesting weights of the access and use subindices merged in the overall IDI.In a similar vein,the Communications Committee of the EU member states argues the advanced mobile network (LTE)coverage influences the actual take-up rate of mobile broadband offers in a country (Communications Committee,2014,p.35).Hence,the present work seeks to overcome the third area of concern of some extant research by (a)applying statistical procedures which reduce double-counting distortions in the aggregation of indicators in a subindex and (b)formulating a structural model of subindices entered into an overall TDI which accounts for means-end-connections between subindices.22.2.Indicator and subindex weighting issuesThe construction of any TDI requires the weighting of the underlying indicators and of subindices in case that the com-posite overall measure is built hierarchically as a higher-order construct with at least one intermediate (subindex/-scale)aggregation level.Overviews of procedures to handling this weighting problem are provided by Saisana and Tarantola (2002,pp.11–51),Nardo et al.(2005,pp.61–75)and Al-mutawkkil et al.(2009,p.180).Some index designers such as Bilbao-Osorio et al.(2014,p.34)or Lam and Shiu (2010,p.188)aggregate indicators or subscales by taking their arithmetic mean without explicitly explaining that this procedure assumes that each variable or subindex has the same importance.But most authors expressly reflect on or at least report the weights which they assign to the indicators or subindices in computing a composite TDI.Overall,the approaches to weighting index components can be grouped into expert opinions and statistical techniques.According to Badasyan,Shideler,and Silva (2011,p.934)most of the measures of a country's IT development or TD status “are heavily based on subjective judgements ”.For instance,Orbicom (2003,p.131)assigns equal weights to the two subindices entered into its overall “Infostate ”index and justifies this with the remark that the authors “have no knowledge basis to do otherwise ”.Conversely,in construing the IDI the ITU (2013,p.21)applies differentiated weightings of 0.4for its access and use subindices,respectively and of 0.2for its skills subindex,but at the same time allocates the same weight to all indicators entered into each subindex.This is worrying because due to differences in the number of indicators in the access and use subindices (five versus three)the procedure implies that weights for individual indicators in the access subindex deviate from the base indicator weights in the use subindex notwithstanding that the ITU provides no reason to justify this unbalanced weighting.3Regardless of whether equal or unequal weights are assigned at the indicator or subindex level,quite many authors do not even make the effort to develop substantial arguments to support their position regarding the “adequate ”weighting scheme in construing an ICT or TD index (e.g.,Çilan et al.,2009,p.100;United Nations,2012,p.124).The unease with the almost unlimited “discretion of the researcher which weights to assign ”(Hilbert,2011,p.720)in general and in particular the argument that “it is unlikely that all variables in an index truly are equally important ”(Wallsten,2009,p.17)has motivated a considerable number of investigations which resort to multivariate statistical techniques in order to handle the problem of how to weight indicators and subindices,respectively.Probably the most frequently applied quantitative method to derive such weights is exploratory factor analysis (e.g.,Al-mutawkkil et al.,2009;Billon et al.,2009,2010;Bruno et al.,2011;Corrocher &Ordanini,2002;Hanafizadeh,Saghaei et al.,2009;Pradhan,Arvin,Norman,&Bele,2014;Trkman et al.,2008;Vicente Cuervo &López Menéndez,2006).This technique transforms a given set of indicators or subindices into a smaller number of synthetic components,factors or dimensions,respectively through2Our means-end argument should not be taken to imply that the relationship between one subindex and another is purely unidirectional.For instance,access adoption of consumers is not only driven by infrastructure supply but also likely to have an impact on future network roll-out.Instead,our approach assumes that the means-end-link of one subindex to another is relatively stronger than the effect in the opposite direction.3The ITU (2013,p.21)claims that the weighting of the IDI indicators and subindices were chosen based on factor analysis results.This is surprising since the relevant statistics reported in an earlier ITU publication (ITU,2009,pp.77–82)are not in line with this proposition.For instance,according to the factor analysis the weight of the indicator “mobile-cellular telephone subscriptions per 100inhabitants ”in the access subindex weight ought to be almost three times as high as the weight of the fixed telephone subscription rate variable.However,the IDI assigns equal weights to both indicators.T.J.Gerpott,N.Ahmadi /Telecommunications Policy 39(2015)93–11196T.J.Gerpott,N.Ahmadi/Telecommunications Policy39(2015)93–11197 linear combinations of the original variables.There are different ways to get weights from the loadings of the variables on the latent dimensions.The most widespread approach is to estimate factor scores for each case(¼country)in a data set.In the course of this estimation a matrix is derived which contains so-called factor score coefficients which are multiplied with the original(standardized)indicator values in order to compute a country's score for each latent component retained in the analysis(see,for example,Vicente Cuervo&López Menéndez,2006).These coefficients equate to the weights assigned to the indicators or subindices in aggregating them to an overall index value.Unfortunately,a considerable share of the relevant publications does not disclose this coefficient matrix(e.g.,Billon et al.,2010;Bruno et al.,2011;Katz&Koutroumpis, 2013;Pradhan et al.,2014).In this case,research recipients are left in the dark with regard to indicator weightings in the composite measure.The method described has two significant disadvantages.First,it results in indexes which are difficult to interpret because each indicator is entered into each index.Second,it does not allow for a hierarchical ordering of indices since it provides little support in obtaining weights for subindices combined in the overall measure.An approach which is based on the results of exploratory factor analysis but does not share these two shortcomings was developed by Nicoletti,Scarpetta,and Boylaud(2000).The authors suggest to assign each indicator to the factor on which it has its highest loading.Indicator weights are then computed by dividing the squared factor loading of a variable by the sum of the squared loadings of all indicators.Thus,indicator weights correspond to the share of variance of an indicator in the overall variance of a composite index.Subindex weights are calculated by dividing the total variance of all factors kept in an analysis (¼sum of their eigenvalues)into the variance of each factor or subindex,respectively(cf.Nicoletti et al.,2000,p.22).With exploratory factor analysis all indicators are typically related to every composite index by a factor loading. Therefore,it is exploratory in the sense that the factor calculation is purely data driven and does not set any a priori constraints on the number of meaningful latent dimensions to be extracted and on which indicator ought to be strongly related to which underlying factor.This approach is appropriate if the number of factors and the assignment of indicators to latent constructs cannot be derived from conceptual considerations.However,in the field of TD status assessment theoretical ideas concerning the indicators and subindices of aggregate TDI are available for quite some time.Thus,a few researchers have started to apply confirmatory(instead of exploratory)factor analysis in conjunction with covariance-based SEM to build composite subindices and to aggregate them to a higher-order overall assessment(Grigorovici,Schement,& Taylor,2004;Kyriakidou et al.,2013).Studies using exploratory or confirmatory factor analysis techniques have yielded interesting insights with respect to the weights of individual indicators,the existence of latent assessment dimensions and the resulting possibility to introduce subindices.Nevertheless,they share a drawback:They calculate composite indices without incorporating the conceptual principle that a TD measure is not an end in itself in their statistical algorithms.Rather its“ultimate”purpose is to predict how changes in TD indicators(influenced by specific political/regulatory interventions)are linked to the attainment of socially valuable outcomes such as economic growth or improvement of the overall living conditions among a country's residents(Barzilai-Nahon,2006,p.270;Grigorovici et al.,2004,p.193;Hilbert,2011,pp.728,733;Saisana&Tarantola,2002, p.11;Waverman et al.,2011,pp.7–8).In principle,two avenues exist to link TDI construction to societal outcome criteria.First,it can be hypothesized that each country has unique political priorities in terms of telecommunications infrastructure and services.Therefore,TD-related indicator and subindex weights should be country-specific.Based on this view,Badasyan et al.(2011)and Waverman et al.(2011)have applied statistical techniques such as the“benefits of doubt”approach to derive TDI variable weights which vary across countries. Second,it can be argued that most of the relevant literature takes a country's TD status as an“enabler”to promote the achievement of a limited set of overarching societal targets as,for instance,promotion of economic development or human well-being(ITU,2013,pp.17–18;Pradhan et al.,2014).From this it follows that indicator and subindex weights of composite TD measures ought to be chosen as a function of clearly named societal objectives.This,in turn,has two implications.First,indicator and subindex weights may vary across outcome criteria studied.Second,various TDI need to be built in a way that the ability of each measure to predict its corresponding outcome criterion is maximized.Given that the second approach,which considers specific outcomes,has been largely overlooked in the literature,it is taken up in the present study.One statistical technique,which fits with the ends-oriented derivation of weights in TDI,is PLS SEM(cf.Chin,1998;Hair, Sarstedt,Ringle,&Mena,2012).It generates weights for individual indicators and subindices aggregated to a superordinated overall TDI in a way that the variance of a societally desirable correlate of the availability and use of networks and services explained by the intermediate level constructs and its indicators included in a conceptual model is maximized.In this way, PLS achieves the best possible reproduction of the raw data set.The focus on data reproduction is a crucial advantage of PLS over covariance-based SEM methods as exemplified in the popular LISREL and AMOS software packages(for an application of these methods to derive revised weights for the three IDI subindices see Kyriakidou et al.,2013).Covariance-based SEM estimate variable weights in a manner that the covariances of the observed indicators are reproduced as good as possible.In contrast,“the variance-based approach of PLS shifts the orientation from causal model/theory testing to component-based predictive modeling”(Chin&Newsted,1999,p.312).In addition,PLS has further strengths relative to covariance-based SEM. Among other things,these include less severely distorted parameter estimates in cases where variable distributions in a data set are extremely non-normal,less demanding minimum sample size requirements and greater ease in simultaneously incorporating formative and reflective measurement models of latent study constructs(Hair,Ringle,&Sarstedt,2011, pp.143–144;Weiber&Mühlhaus,2014,pp.73–78).Particularly,the question of choosing an appropriate measurement model has yet received no attention in studies which proposed composite ICT or telecommunications development indices.A measurement model posits relationships between。