Measuring the global information society explaining digital inequality by economic level and education standard

Transcription

1 IOP Conference Seres: Materals Scence and Engneerng PAPER OPEN ACCESS Measurng the global nformaton socety explanng dgtal nequalty by economc level and educaton standard To cte ths artcle: H Ünver 2017 IOP Conf. Ser.: Mater. Sc. Eng Related content - Cross-country Analyss of ICT and Educaton Indcators: An Exploratory Study Ahmad R Pratama - The Development of Statstcs Textbook Supported wth ICT and Portfolo-Based Assessment Putraj Hendkawat and Florentna Yun Arn - A framework to apply ICT for bequeathng the cultural hertage to next generaton R Syah, T E Nurad and M K M Nasuton Vew the artcle onlne for updates and enhancements. Ths content was downloaded from IP address on 29/01/2018 at 17:47

2 Internatonal Conference on Recent Trends n Physcs 2016 (ICRTP2016) Journal of Physcs: Conference Seres 755 (2016) do: / /755/1/ Measurng the global nformaton socety explanng dgtal nequalty by economc level and educaton standard H Ünver Ulm Unversty, Insttute Databases/Artfcal Intellgence & Research Insttute for Appled Knowledge Processng/n (FAW/n), Lse-Metner-Str. 9, Ulm/Germany E-mal: halt.uenver@un-ulm.de Abstract A man focus of ths research paper s to nvestgate on the explanaton of the dgtal nequalty or dgtal dvde by economc level and educaton standard of about 150 countres worldwde. Inequalty regardng GDP per capta, lteracy and the so-called UN Educaton Index seem to be mportant factors affectng ICT usage, n partcular Internet penetraton, moble phone usage and also moble Internet servces. Emprcal methods and (multvarate) regresson analyss wth lnear and non-lnear functons are useful methods to measure some crucal factors of a country or culture towards becomng nformaton and knowledge based socety. Overall, the study concludes that the convergence regardng ICT usage proceeds worldwde faster than the convergence n terms of economc wealth and educaton n general. The results based on a large data analyss show that the dgtal dvde s declnng over more than a decade between 2000 and 2013, snce more people worldwde use moble phones and the Internet. But a hgh dgtal nequalty explaned to a sgnfcant extent by the functonal relaton between technology penetraton rates, educaton level and average ncome stll exsts. Furthermore t supports the actons of countres at UN/G20/OECD level for provdng ICT access to all people for a more balanced world n context of sustanable development by postulatng that polcymakers need to promote comprehensve educaton worldwde by means of usng ICT. 1. Introducton Indeed, one of the most mportant terms n the feld of nformaton socety s 'dgtal dvde'. Hstorcally, t was hgh on the agenda of the European Unon [26], a topc of the World Summt on the Informaton Socety (WSIS) took place n Geneva n 2003 [33] and Tuns n 2005 [34] as well as at the World Summt on Sustanable Development (WSSD) n Johannesburg [32]. Nowadays, a lot of ssues related to dgtal dvde are dscussed at the UN Internet Governance Forum [11, 12, 13] by usng the term dgtal nequalty as a more adequate and precse descrpton of the technologcal stuaton of a country. 1.1 Terms As we know, good nfrastructure (traffc nfrastructure, technologcal nfrastructure, housng nfrastructure, energy nfrastructure etc.) s one of the most mportant components of wealth and economc power [27]. The economc strength of a country can be quantfed by the (fnancal) value of all goods and servces produced wthn a gven tme perod n a country [20]. GDP per capta (GDPpC) s one of the man ndcators of economc analyss, both n spatal and n temporal nternatonal comparsons. For Content from ths work may be used under the terms of the Creatve Commons Attrbuton 3.0 lcence. Any further dstrbuton of ths work must mantan attrbuton to the author(s) and the ttle of the work, journal ctaton and DOI. Publshed under lcence by Ltd 1

3 nternatonal analyss, the purchasng power party (PPP) s used. The socal balance or the dstrbuton of ncome s, n addton to GDP per capta, an mportant parameter for understandng the socal stuaton n a country. It s emprcally and theoretcally well known that nether too much nor too lttle nequalty s good for a country [17] [35]. A country needs to nvest n educaton and research to mantan economc welfare. In a global perspectve, t s stll mportant to observe the number of people beng able to read and wrte. The adult lteracy rate s the percentage of the populaton age 15 and above who can read and wrte a short, smple statement concernng ther everyday lfe [29]. Generally, lteracy also encompasses numeracy, the ablty to make smple arthmetc calculatons lke addng numbers. The UN Educaton Index s a more complex measure to compare between dfferent educatons standards n countres. It s a combnaton of expected and average school tme of ndvduals n a country [465]. Statstcally possble repettons of classes are ncluded. From a mathematcal vewpont, the expected years of schoolng s of a predctve character. Innovaton and technology promote economy and educaton, enablng convenence n communcaton, whch s the basc operaton of a socety wth enormous effects on the performance of a country [19]. One ndex for the technologcal development of a country, n terms of technologcal deployment and communcaton, s to measure the percentage of populaton usng any (dgtal) technology, whle gnorng the type of access (prvate/ shared), connecton (wred/wreless) or locaton (household, school, offce, cafe). For example, the Internet penetraton rate (IPR) s the estmated number of Internet users out of a total populaton. Ths ncludes those usng the Internet from any devce (ncludng moble phones) over the last 12 months, measured through household surveys. If household surveys are not avalable, an estmate can be derved based on the number of Internet subscrptons [15]. The moble phone connecton s becomng more and more mportant due to the fact of technologcal convergence whch provdes the use of the Internet from any moble devce. The moble phone penetraton rate (MPR) ncludes both the number of moble phone subscrptons and the number of actve prepad accounts that have been used wthn the last three months. It does not nclude the connectons va data cards or USB - modems, connectons of publc moble data servces, prvate trunked rado, telepont, rado pagng and telemetry servces [14]. 1.2 Data and Methods The World Bank provdes data regardng the economc development level (GDP, GDPpC, ncome dstrbuton) of countres [31]. Technology penetraton rates (MPR, IPR) are avalable (partly free of charge) on the Internatonal Telecommuncaton Unon webste [16], where also some other ICT key ndcators can be downloaded. Data for measurng the educaton level of a country and ts populaton can be downloaded from the statstcal dvson of the UNESCO [30] and UN Development Programme webstes [465]. On those webstes data s avalable for natonal adult/youth lteracy rates, enrolment ratos and educaton ndex by nternatonal standard classfcaton of educaton level (ISCED). Ths study focuses on adult lteracy rate and educaton ndex. Due to the lack of data regardng lteracy rate for many developed and developng countres, polynomal nterpolaton s used to have enough data for par-by-par comparson concernng technology penetraton rates. The data management process n ths work s essentally based on the authortatve archtecture of data warehouse systems [4]. Fgure 1 shows the ETL process (extract, transformaton, load), whch s appled for the data management. The extracton component (1) has the functon to procure data from the dfferent data sources. Among other thngs, there are questons regardng the exchange of data, lke formats to be consdered. Furthermore, certan data subsets need to be extracted from the sources, why separaton s of mportance. A decsve factor here s also the queston of the tme of extracton. Ths s done as needed, whereas a perodc, event-based or mmedate extracton s generally possble. After extracton, the needed data for analyss s stored n the work area. Step (2a) n fgure 1 s the so-called transformaton component. The data obtaned from dfferent sources must be standardzed n a certan way n order to make a vald processng possble. The unfyng and processng of the data s carred out n ths component. The loadng up component n step (3) passes on the prepared data from the workspace to the correspondng database (Access DB). Furthermore, ths component s responsble for the hstorczaton of the data n a certan regard. Ths means old data s not smply deleted or overwrtten but 2

4 rather provded wth a tmestamp and stored. A change at a data set therefore has the consequence that two data sets exst wth a dfferent tmestamp. In step (4a) the correspondng data are brought together for analyss purposes from dfferent sources. For ths use the structured query language (SQL) s used. Step 5 serves now to fetch the data for regresson analyss. The correspondng data are exported from the database program. Afterwards the data are loaded n Step (6a) nto the statstcs program SPSS or n step (6b) nto the numerc program ScLab. It comes n step (7a) or (7b) to the correlaton and regresson analyss by usng the numerc program ScLab or SPSS, n whch some queres (e.g. the categorzaton of the countres after economy performance) also take place n step (7c) drectly at the Access DB. Fgure 1. Adapted ETL-process. Correlaton quantfes the degree to whch two varables are related, and ths regards a lnear form of mutual dependency [8] [23]. Regresson fnds the best lne or functon (method of least squares), whch predct the dependent varable from the ndependent varable [117] [22]. The best ft functon from the class of (test) functons s determned by usng the least squares method. For ths purpose, one mnmzes the sum of the mean square of errors (MSE), wheren the coeffcent of determnaton (R2) s a qualty measure for the adapton of ndvdual functons. The functon whch fts most to the data by means of method of least squares s then fnally used for the relaton between the dependent (e.g. moble phone usage) and ndependent varable (e.g. lteracy) n a certan year. A lot of known functons n the felds of economy, natural, physcal and socal scence are used to dentfy the relaton between technology penetraton and dfferent factors n a country. Interdscplnary reasons for usng some known functons are as follows: 1. Lnear functons are easy and one of the frst approach, f humans try to understand real-lfe phenomena wth gven data. 2. Exponental functons are often used to represent growth and decay, lke populaton growth or deprecatons 3

5 3. Logarthmc functons are commonplace n scentfc formulae, and n measurements of the complexty of algorthms and n many other dfferent applcatons lke n the measurement of earthquakes and sound 4. Logstc functons are used to model real-lfe quanttes whose growth levels off because the rate of growth changes, from an ncreasng growth rate to a decreasng growth rate. 5. Planck's law descrbes orgnally the electromagnetc radaton emtted by a black body n thermal equlbrum at a defnte temperature. Anyway, n general, t descrbes an ntal ncrease of the dependent varable at hgher levels of the ndependent varable, and then a decrease of the dependent varable at hgher levels of the ndependent varable. Functon 1. Lnear Formular y a b x 2. Exponental 3. Logarthmc 4. Logstc 5. Planck s Law y y y y a b e ( c x d ) a b log( x c) a ( d x b c e ) b a x ( d x ) c e 1 Only bvarate correlatons between two varables can be examned wth a correlaton coeffcent and smple lnear regresson. If one would lke to nvestgate on the relatonshp between several varables, there s the so-called multvarate and multple regresson analyss. There s a dependent varable Z (e.g. IPR) and two ndependent varables (e.g. GDPpC and Educaton Index). Furthermore, t s assumed that a smple regresson analyss for the relatonshp between the nternet user rate and the GDPpC provdes a non-lnear functon f(x). The functon g(y) corresponds to the non-lnear relatonshp between the IPR and the Educaton Index. The correspondng multple regresson model wth the correspondng weghts A and B s: The underlyng functons f(x) and g(y) are of e.g. 3-parametrc logarthmc type: Now, based on the method of least squares t s to mnmze the functon ( A, B) : For ths we calculate the partal dervatves of ( A, B) the approprate mnmum. below. These equatons are set to zero to fnd 4

6 The result s a system of equatons, whch delvers the regresson weghts A and B after ts resoluton accordng to the followng procedure. In dong so, we can fnd out the dfferent mportance (weght) of economc level and educaton standard for the deployment of the Internet n a country. Certanly there s also a relaton between the economc power and the educaton level of a country. Based on the fndngs of ths work and the dfferent mportance of economy and educaton for the deployment of ICT, n partcular moble phone usage and Internet penetraton, t was able to suggest or advce mportant decson makers and nternatonal organsatons, such as at the Internet Governance Forum n 2014 (Turkey) and 2015 (Brazl) or at UN/G20/OECD level. 3. Lterature Overvew and Dstncton In the 1990s, the term dgtal dvde emerged to descrbe technology haves and have-nots. Current research regardng dgtal dvde has a descrptve character. Those studes emphasze the dgtal dvde by usng demographcal, economcal or educatonal data, n general at an ndvdual level of technology usage or dgtal sklls of people n a country [2]. Ths study nvestgates the dgtal dvde more from the perspectve of dgtal nequalty. There are manly four successve knds of access n the appropraton of dgtal technology [7]. These knds of access are a) motvaton, b) physcal and materal access, c) dgtal sklls and d) dgtal usage. Here, the focus s on access types b) and c) by searchng for functonal relatons between the economc power, educaton level and technology penetraton rates of countres. It s assumed that the dgtal sklls of a socety are hgher, f the educaton level of ths socety s hgher. Physcal and materal access corresponds to moble phone usage rate (MPR) and Internet penetraton rate (IPR). A number of emprcal studes have been done on the topc of dgtal dvde. A few of them hghlghted that ncome level [1] [2], ncome dstrbuton [9] [36], educaton level [5] [10] [21], sze of populaton [28] and urbanzaton [3] have essental correlaton wth Internet penetraton levels of countres. Andres et al. [2] stated that low-ncome countres have a steeper Internet dffuson curve than that of hgh-ncome countres. Although ths result s ratonal, because low-ncome countres can leapfrog technologcal 5

7 developments, t has to be mentoned, that the splttng to only two categores of ncome levels s questonable. Zhang [36] found out a postve contrbuton of GDP per capta (PPP, current nt.$) to IPR and a negatve nfluence of ncome dstrbuton measured by the Gn-Index. Here t should be noted; a hgher average ncome corresponds to a more equal ncome dstrbuton n general [18] and that's not dscussed anyway n the results of Zhang. Furthermore he dd not explan the relaton between GDP per capta and Internet penetraton n form of a detaled functon. Ksk et. al. [21] showed that the average years of schoolng s consderably a postve factor for the Internet hosts per capta n a country. Even so another study fnd out that the degree to whch the dfference n Internet rates depends upon educaton level s surprsngly small [6]. Ths s n contrast to our fndngs. In the past, studes about usng technology acceptance explan, how atttudes determne Internet penetraton [3]. Such studes show, that some of the faster rates of growth n Internet use have been among ndvduals who are older, less educated, of mnorty status or wth lower ncomes. A smlar research work to ths study examnes the relaton of the Internet penetraton rate wth the human development level over the decade from 2000 through 2010 [25]. These results support the argument that a dgtal dvde exsts between developed and developng countres. The man outcomes of ths study are that there s a postve correlaton between human development level and Internet penetraton rate and that the correlaton has become slghtly stronger from 2000 through However, Internet Usage rates assocated wth these demographc groups are lower than that of the general populaton [24]. Another study ponts out, that more educated people use the Internet more actvely and ther use s more nformaton orented, whereas the less educated seem to be nterested partcularly n the entertanment functons of the Internet. If we look to the dgtal dvde wth respect to Internet penetraton, there s stll an nequalty between developed and developng countres [25]. However, there seems to be no related scentfc work analyzng the emprcal relaton between educaton ndex and Internet penetraton, especally over the decade from 2000 through Therefore, ths ssue s tackled n ths study. Furthermore, most of the related work descrbes the relaton between economcal development, educaton level and Internet penetraton wth lnear regresson functons and wth logarthmc or exponental functons. In ths study, more non-lnear regresson models lke the logstc functon or functons accordng to Planck's law are used to descrbe the relaton between the dfferent parameters n detal. Specfcally the multvarate regresson analyss for the relaton between the ICT penetraton rates as the dependent and educaton level and ncome stuaton as the dfferent ndependent varables (descrbed by non-lnear regresson functons), s a essental contrbuton to the scentfc feld related to dgtal dvde or dgtal nequalty. 4. Emprcal Results and Functonal Analyss 4.1 Relaton between Economc Power and ICT usage GDPpC and Moble telephony (MPR) Table 1 shows the number N of analyzed countres, the mnmum (mn), maxmum (max), span (sp), mean (av), medan (md) and standard devaton (sta) for the GDP per capta (GDPpC) and moble penetraton rate (MPR) n a worldwde perspectve n the years 2000 and Table 1. Descrptve statstcs of moble penetraton rate (MPR) and GDP per capta (GDPpC)n year 2000 and If one descrbes the worldwde dgtal nequalty n the context of MPR and the average GDPpC wth the help of sp and sta as easy dsperson metrcs, t turns out that the nequalty n the use of moble phones has ncreased between 2000 and 2013, whereas also the nequalty regardng the GDPpC 6

8 ncreased. Lookng closer at the data for the year 2000 n fgure 1 one can recognze that states wth a low GDPpC have low MPR. The greater the GDPpC values become, the greater also the MPR values become. Movng nto the drecton of hgher GDPpC the MPR frst reaches a maxmum and afterwards drops agan. In ths ntal stuaton the Planck s functon (black) returns a very good adjustment (wthn the consdered class of functons and accordng to the method of the squares) of the relatonshp between the GDPpC and the MPR n Fgure 1. moble phone penetraton rate (MPR) as a functon of GDP per Capta (GDPpC), year 2000 (left) und 2013 (rght). More than a decade later, a frst observaton s that the data concernng 2013 reach sgnfcantly hgher values for MPR than the data n Furthermore the MPR s over 100% n many countres n Ths means that there s averaged more than one moble devce per person n a country. Overall, all states have ncreased ther MPR, so that the catch-up n the moble phone usage s very clear worldwde. In fgure 1 s also a lst of selected functons such as lnear (red), logrtm (blue), monod1 (pnk), monod2 (gold), logst (green), maxlog (brown) und Planck (black). The 4-parameter functon n equaton 5 (see secton 1.2) wth the parameters a, b, c, and d (expresson of the Planck functon) delvers the best ft for After verfcaton of the partcular functons for each year between 2000 and 2013, t turns out that only n 2013 the maxlog functon delvers a better adaptaton than the Planck functon. Table 2 shows the dfferent fttng functons and the assocated values for the sum of squared errors (MSE or MSE/N) and the coeffcent of determnaton (R2). Table 2. Sum of mean square of errors (MSE) and coeffcent of determnaton (R2) for regresson MPR = f(bippk) In fgure 2 one can study the development of the best ft functons for the relatonshp between GDPpC and MPR from 2000 (blue) tll 2013 (black). The result shows that the MPR has ncreased n all areas of GDPpC from year to year and has taken a smlarly good development over all GDPpC ranges n more than a decade. Ths result s not only an artefact of the adaptaton n the optmsaton process; rather the nvestgaton of the database lad for ths optmsaton delvers the same result. Based on fgure 2 one can recognse, that the MPR has rased over all GDPpC areas.e. accordngly n all countres. Ths can be seen on the one hand n the hgher course of the curve for 2013 (black) as the curve course for 2000 (blue). On 7

9 the other hand, the data ponts for the year 2000 (blue) are not or barely overlad wth those data ponts for In relatve terms, the shft of centre of data s stronger upwards (n the drecton of MPR) than to the rght (n the drecton of GDP). Fgure 2. development of moble phone penetraton rate (MPR) as a functon of GDP per Capta (GDPpC) between 2000 and 2013 (left); data ponts, best ft functon, centre (rght). If one performs a growth analyss over the whole tme from 2000 to 2013 for the GDPpC and the MPR n the sutable GDPpC categores A, B, C, D and E, where category A stands for the rchest and category E for the poorest countres, one s able to examne, to what extent the GDP growth and the MPR growth are connected wthn and between these categores. The bar chart n fgure 3 shows on the x-axs, the respectve GDPpC category and on the y-axs the growth of GDPpC (blue) and MPR (burgundy). Hence the hghest growth of GDPpC and MPR took place n category C (medum GDPpC). Both growth rates le n category C wth about 112 %. In the categorya (hghest GDPpC) and B (hgh GDPpC) s a smlarly hgh MPR growth of approx. 94% and 94. 5%, whereby the GDPpC growth les n category A wth about 62 % and n category B wth approx. 100 %. The GDPpC growth s n category D (low BIPpE) 109 % wth a MPR-growth of 90 %. In category E (lowest GDP) was held the slghtest MPR growth, namely by about 65 % wth a GDPpC growth of nearly 77 %. The result ndcates that the GDP and the MPR growth have a strong coherence (hgh correlaton), especally f one does not consder the rchest countres (category A). Ths does not surprse f one mnds that the rchest states can not show such a hgh performance as economcally weaker states, because they start from a clearly hgher level of GDPpC. Fgure 3. Growth rate (total) for GDPpC (blue) and MPR (burgundy) between 2000 and Ths study confrms to some extent the assumpton or hypothess, that the catchng up generally n the use of ICT and n partcular n the moble phone usage runs faster worldwde than catchng up wth respect to prosperty n general. The sze and/or the growth/shrnkng of the populaton have no sgnfcant nfluence on the MPR. A hypothess represented n ths work s that we wll experence a convergence n the ICT use. Ths thess s confrmed so far for the moble phone usage, as that n 2013 the worldwde 8

10 average MPR has reached about 100 % and that the MPR growth between 2000 and 2013 over all GDPpC areas took place at a smlarly hgh level, also due to the fact that n many countres there s more than one moble phone connecton and that the avalablty of prepad systems has favoured the moble phone usage n the poorer regons of the world ( moble mracle ). In ths context today one can descrbe the stuaton n the followng way: The world populaton calls up moble and has carred out the jump of the localengaged communcaton (fxed network) to the person engaged communcaton (moble phone) GDPpC and Internet Usage (IPR) Aberrantly to the consderatons n the prevous secton, the queston gets answered, how bg the amount of a country s n the world gross domestc product (world GDP) and whch amount a country has n the worldwde Internet users (world IPR). The populaton sze s factored out n the consderatons of the economc achevement and the Internet usage. Fgure 4 shows on x-axs the economc achevement of a country as a porton n the world GDP and on the y- axs the Internet usage extent of a country as a porton n the world IPR n the year 2000 (left) and 2013 (rght). Fgure 4. Proporton of world IPR as a functon of the proporton of world GDP, 2000 (left) and 2013 (rght). The Unted States has accounted for about 23 % of world economc output and about 30 % of Internet users (worldwde). If one dvdes the states nto one group above and one below the regresson lne, the USA, Japan, Brtan and Korea are located above and states lke Chna, Inda, Brazl, Russa and the socalled BRIC states, but also France and Italy below the lne. As fgure 4 shows, there s a hgh correlaton n the year Furthermore, t s n such a way, that the relatons after 13-year development n 2013 stll can be descrbed well by a lnear model,.e. that stll a hgh correlaton s gven. Ths easy analyss ponts to a narrow respect between the economc power of a country and the Internet use n ths country. Ths s also vald, f one gnores the populaton number wth regard to the economc power. In a per capta consderaton the dfferences between the rcher and poorer countres are stll much greater. Ths s also due to the large populaton of countres such as Chna and Inda. It s nterestng n ths context that n 2013 all the BRIC countres are above the lnear regresson model and the USA, Japan and Germany le under t. Ths development s a consequence of that the BRIC states, whch have a relatvely large populaton, ncreased ther IPR between 2000 and 2013 sgnfcantly. Snce the algnment of communcaton s stll faster than the economc performance, the sze of populaton of BRIC states leads to a correspondngly large Internet populaton. Table 3 shows for the GDPpC and the IPR the values of the descrptve statstcs. The IPR has n 2000 a maxmum of 52% (Norway) and ths value ncreases tll 2013 to approxmately 96.5% (Iceland). The IPR mnmum value of 0.01% (Dem. Rep. of Congo) n 2000 ncreases slghtly and s almost 1% (Ertrea) n There are agan the Afrcan countres, as n the MPR, whch have the lowest values for the IPR. The average ncreases over the same perod from nearly 8% to about 41.6%. It s nterestng, that the medan of 39.2% n 2013 s nearly at the same level as the average value, although the medan of nearly 9

11 2% n 2000 s on a four tmes lower level as the average. Ths s a consequence of the sgnfcantly hgher balance of the proportons tll 2013, as further noted n another context above. Table 3. Descrptve statstcs of Internet penetraton rate (IPR) and GDP per capta (GDPpC)n year 2000 and We consder further n detal the emprcal relatonshp between GDPpC and IPR per country. Here are sgnfcantly greater dfferences to be expected, as above n consderng the economc output (GDP) of each state. The relatonshp s no longer lnear. In fgure 4.12 (left), t s evdent that states wth a lower GDPpC have a lower IPR. The greater the value of the GDPpC becomes, the greater also becomes the IPR, ndeed, for qute bg values of the GDPpC t falls agan. Ths s not surprsng n ths respect, as nearly every Internet access n year 2000 occurred va a fxed network connecton by dal-up (modem). The moble Internet access hardly exsted worldwde. So the fxed network was a necessary condton for the Internet connecton, whch was (only) developed n rcher countres. The Scandnavan countres Denmark, Sweden, Norway and Fnland have worldwde one of the hghest IPR n But also Australa, Germany, Great Brtan, Japan, Canada, Korea, Swtzerland, Sngapore and the USA have hgh values for the IPR. There are countres such as Bahran, Brune, Kuwat, Oman, Saud Araba, and the Unted Arab Emrates, but also surprsngly Luxembourg, whch have a hgh GDPpC and n relaton to ths huge GDPpC value a relatvely low IPR value compared to the remanng rch states. These are prmarly the ol states that enforce a form of Planck's functon for the relatonshp between GDPpC and IPR. Reason s found n the n parts of nature borrowed prosperty, manly resultng from the sale of ol and gas, not from an nternatonally compettve, dversfed productve economy, whch tself would requre massve dssemnaton of technology, Internet usage and nnovaton n ths area. Fgure 5. Internet penetraton rate (IPR) as a functon of GDP per Capta (GDPpC), year 2000 (left) und 2013 (rght). The values for the IPR have n 2013 ncreased sgnfcantly compared to The statement from 2000 that states wth low GDPpC have a rather low IPR s also vald for 2013, n spte of occurred catchng-up processes. Furthermore the obvous correlaton s vald that hgher GDPpC values cause hgher IPR values. Table 4.12 lsts the results of the regresson analyss wth the values for the sum of squared errors (MSE or MSE/N) and the coeffcent of determnaton (R2) for all functons tested for the years 2000 and As already mentoned, the Planck functon fts best, followed by the functons maxlog and logst wth a smlarly good adjustment n In 2013, the Planck functon agan has the best ft, but also the rest of the tested functons, except for the lnear, descrbe the relatonshp smlarly 10

12 well. Ths ponts to the fact that the functons wth an nhbton lose relatvely n strength and functons wth saturaton wn relatvely n strength for ths relatonshp. Table 4. Sum of mean square of errors (MSE) and coeffcent of determnaton (R2) for regresson MPR = f(bippk). Fgure 6 shows the development of the best ft functon for 2000 (blue) to 2013 (black). For all years between 2000 and 2013 the Planck functon acheves the best ft. The results for the ft functons show that the IPR has expanded smlar for all GDPpC areas between 2000 and The (Planck) fttng functons for the years 2008 and 2009 run n ths respect slghtly dfferently, because states wth the bggest GDPpC have lower IPR values, than durng the years before. However, ths s not approprate as the data show. So ths s an artefact of the regresson optmzaton. Ths results from the fact that the countres n the medum GDPpC area between 2008 and 2009 tend to have a larger (relatve) ncrease n ther IPR than the countres n the top GDPpC area. The reasons for ths are to be found n the begnnng of the moble Internet usage n countres wth a medum GDPpC. One can observe the same phenomenon n Fgure 6. Development of Internet penetraton rate (IPR) as a functon of GDP per Capta (GDPpC) between 2000 and 2013 (left); data ponts, best ft functon, centre (rght) In fgure 6 s to be recognsed once more that the IPR ncreased over all GDPpC areas n all countres. The curve for 2000 (blue) runs clearly below the curve for 2013 (black) and there s no ntersecton of the curves. Furthermore, there s hardly a superposton of the two pont clouds. The shft of focus s more upwards (towards IPR) than to the rght (towards GDPpC) relatvely regarded. If one carres out a growth analyss from 2000 to 2013 for the GDP and the IPR n the correspondng GDPpC categores A, B, C, D and E (dvson, see appendx A.1) the relatonshp between GDP growth and IPR growth can be specfed. Fgure 4.15 shows on the x-axs the respectve GDPpC category and on the y-axs the growth rates for the GDPpC (blue) and the IPR (red). The respectve number refers to the overall growth between 2000 and Here category A stands for the rchest and category E for the poorest countres. 11

13 Fgure 7. Growth rate (total) for GDPpC (blue) and IPR (burgundy) between 2000 and The hghest average IPR growth of 54.4% took place n the hghest GDPpC category A, but n ths category also the slghtest GDPpC growth of 62.5% s gven what, nevertheless, absolutely means a hgh growth, because the rchest countres have a hgher startng level than poorer countres. Ths fndng s nterestng that the average (absolute) IPR growth s lower, the poorer the group of states s. Ths means that n category B, the second strongest IPR growth takes place wth 50%, followed by the category C wth 39.4%, category D wth 30.5% and category E wth 9.6%. On the one hand the thess that the catchng-up n the usage of Internet gets confrmed generally. But on the other hand ths fndng shows that regardng to the global spread of Internet access, especally n the developng countres stll much s to be done. In summary, the emprcal relatonshp between the GDP and the IPR both n 2000 and n 2013 s postve. The Planck functon modelled ths context (wthn the consdered class of functons) best for all years from 2000 to The course of the Planck functon n 2000 s smlar to the the one of the relatonshp between the GDPpC and FTR, suggestng also the then exstng physcal connecton of Internet and Fxed Telephony (see secton 4.4.2). Up to 2013 almost all states have ncreased ther IPR and on average the rchest states show the hghest rse and the poorest states the lowest. Wth the relatve ncrease t s exactly the reverse. Ths observaton underlnes the thess that the catchng-up n Internet usage around the world runs (clearly) faster than catchng up wth regard to prosperty n general. 4.2 Relaton between Educaton Level and ICT usage In a socety and culture that s dependent on the wrtten language n addton to oral communcaton, the lteracy of people s a dstnct advantage compared to llteracy. The use of the Internet for example s restrcted sgnfcantly for llterate people, because t works on a text-based nature untl today manly. As a comparson, the use of fxed or moble phones requres (almost) no lteracy sklls Lteracy and Moble Telephony To use the moble phone, you have to be able to dentfy at least the dgts 0 to 9. Ths requres no sklls for readng and wrtng of letters or words. The dentfcaton of numbers and correspondng basc operatons such as addng numbers s part of sklls are ndcated n the adult lteracy rate (ALR). It s expected that the emprcal relatonshp between the ALR and the moble phone rate (MPR) s low, but should be lower n 2012 or functonally run sgnfcantly dfferent than n The reason for ths s the rapd spread of moble telephony n ths perod across all countres n the world, also wthn llterate people. Table 5 shows the descrptve statstcs for the ntersecton of the data for ALR and MPR for the years 2000 and Table 5. Descrptve statstcs of adult lteracy rate (ALR) and moble phone penetraton rate, year 2000 and

14 In 2000, the global average ALR s about 81.2%, whle ths rate ncreased to about 86.2% n 2012, although n the same perod, the world populaton has ncreased by almost 16% from 6 to 7 bllon people. Worldwde development programs relatng to the MDGs and smlar campagns contrbute to ths progress. The sp of the ALR has decreased worldwde from about 90% to almost 70%, whch s also a postve sgnal n the drecton of mprovng global levels of educaton. The data confrms that most countres wth a low ALR are located on the Afrcan contnent. In Afrca, the hgh populaton growth s a major challenge when t comes to ncrease the ALR n total. The Baltc countres belong to the countres wth the hghest ALR. These hgh values for the ALR are also consequences of the culturally hgh mportance of educaton n the old Sovet Unon. In both years the medan value s wth 89.1% n 2000 and 93.8% n 2012 slghtly hgher than the average. Most tmes t s the other way round, for example, at ncome dstrbutons,.e. the average s (often sgnfcantly) hgher than the medan. What s t the reasonng? In many dstrbuton stuatons there are great outlers upwards as very hgh ncomes that amount a multple of the average ncome. In the present educaton data t s dfferent. Almost all states are wth regard to the ALR anyway at a level greater than 50%, many greater than 80%. An ALR = 100% s the upper lmt. There are no outlers upwards, only those down. The medan s about 90% n The mean value s therefore determned by smaller ALR close to 50% and lower. Ths drves the mean value downwards n the drecton of about 80%, what s less than the medan. In 2000 the data ponts are strewn along the ALR and MPR axs, whch s why the exponental model and the power functon descrbe the relaton well, whch means that only for hgh values of the ALR also accordngly hgh values of the MPR are reached. It s probably the case that n 2000 the moble telephony was closely lnked for fnancal reasons wth the fxed telephony, although the respectve physcal nfrastructure was decoupled. Fgure 8. Moble phone penetraton rate (MPR) as a functon of GDP per capta (GDPpC), year 2000 (left) and 2013 (rght). The relatonshp between the ALR and MPR changes up to 2012 so far that also for relatvely low values of the ALR hgh values of the MPR exsts. Ths change s clearly vsble n the functonal modellng of the relatonshp between ALR and MPR, because n 2012 the lnear, exponental and power functon have almost dentcal values for the sum of squared errors (MSE or MSE / n) and the coeffcent of determnaton (R2). Table 6. Sum of mean square of errors (MSE) and coeffcent of determnaton (R2) for regresson MPR = f(alr). 13

15 Catchng up n the moble telephony use n the developng and emergng countres n whch worldwde most llterates lve, decsvely contrbutes to ths postve development. The 100% mark of the MPR s acheved from an ALR of about 60% Educaton Index and Internet Usage The use of the Internet stll requres at least the ablty to read and to wrte. But what s the functonal relatonshp between the Internet penetraton rate (IPR) and the level of educaton as measured by the Educaton Index (EI)? A basc thess s as follows: The hgher the EI of countres s, the hgher the IPR should be. Another theory, whch s descrbed on the basc thess wth the mportance of the educatonal level of countres for the IPR decreases contnuously between 2000 and 2012 and wll most lkely decrease n the future. Table 4.31 shows the descrptve statstcs for the ntersecton of the data for EI and IPR for the years 2000 and Table 7. Descrptve statstcs of Internet penetraton rate (IPR) and Educaton Index (EI), year 2000 and A frst sght at the pont cloud n fgure xx shows that n both years a postve correlaton exsts between the EI and the IPR. If one looks at the prevous secton about the relatonshp between the educatonal standard, measured by the Alphabetzaton Rate (ALR), and the IPR, the _rst d_erence s to be observed here. Whle ndeed a hgh ALR s a necessary but not a su_cent condton for a hgh IPR, you have at the top of EI also correspondngly hgh, but no low values for the IPR n One can see ths also n the fact that many data ponts are spread along the EI axs up to a value of about 0.6 and up from ths EI value the IPR ncreases. A compresson of the data along the IPR axs s here, as opposed to the correlaton between ALR and IPR, n 2012 not avalable. One also notes ths d_erence wth the comparson of the sngle regresson models. The hyperbolc functon (yellow green) delvers the worst adaptaton for the connecton between EI and IPR.22 The relatonshp between EI and the IPR s modelled wth the help of a power functon (green) best of all, followed by the exponental functon (blue) whch s almost as good as the power functon. The lnear regresson model delvers (red) only partly a good adaptaton (see table 8). Fgure 9. Internet penetraton rate (IPR) as a Educaton Index (EI), year 2000 (left) and 2012 (rght). 14

16 Table 8. Sum of mean square of errors (MSE) and coeffcent of determnaton (R2) for regresson IPR = f(ei) Up to 2012 the dsperson of the pont cloud changes what s to be seen good optcally. The power functon gves n 2012, as well as n 2000, the best ft wthn the consdered class of functons. The exponental functon delvers the second-best adaptng. The lnear model approxmates the data n 2012 sgnfcantly better than n Bascally all three functons delver a smlarly good modellng of the connecton between the EI and the IPR. A hgh EI delvers a hgh value of the IPR and, however, already a mddle EI level s suffcent to reach rather hgh values of the IPR. Countres wth a low EI also tend to have a low IPR, although there are also countres that, despte a relatvely low EI have rather hgh values for ther IPR. Catchng up n Internet use has progressed worldwde for all countres faster than the development n the area of educaton. Ths fact suggests that global convergence processes n Internet use run sgnfcantly faster than the convergence n the context of educaton. Interestng s the negatve value of the coeffcent of determnaton (R2), whch s obtaned for the hyperbolc functon n 2012 (see table 8) and occurs n ths work for the frst tme n ths form. Ths s frst contra ntutve, because one would thnk the R2 to be a squared sze, because t s symbolc descrpton. Ths would then always have values greater than or equal to zero. However, ths s not the case here. It does not concern a sze n the square, but from the defnton follows that n the case n whch the adaptaton of a functon to the data s worse than a horzontal lne at the level of the average value, the sze R2 s negatve. In ths example one can nterpret ths as follows: The negatve value of the R2 says that the carred out modellng of the data wth certan (here as hyperbolc assumed) functon s worse than the modellng of the data wth a steady functon whch runs at the level of the average of the IPR. 4.3 Multvarate Regresson By unvarate regresson (only) relatonshps between two varables can be examned. If one uses nstead multple varables for predctng a sze, one goes to the feld of multvarate and multple regresson analyss (see secton xx). When consderng several ndependent varables n general, the queston arses, whch of them affects the dependent varable at most or at least. The effect of each explanatory varable on the dependent varable s measured (ndrectly va the assocated estmators) by the respectve regresson coeffcent. If the ndependent varables own dfferent unts, the dmensons of the regresson coeffcents cannot be compared (drectly). Standardsed regresson coeffcents compensate the effect of dfferent scales. Thus all varables of the regresson model receve a statstcally unform unt. The contact wth the negatve weghts whch can appear n the optmsaton process just lke that s nterestng also. Comparng the dmenson of the nfluence of the respectve ndependent varables one has to use the absolute value and compare them wth each other. In ths work the regresson coeffcents (A, B, C and D) are used as weght varables that specfy whch mportance the respectve predctor enters the forecast Combnaton of the factors MPR, GDPpC, EI, ALR Lookng at the moble phone rate (MPR) as the dependent varable and the GDP per capta (GDPpC), the Educaton Index (EI) and the adult lteracy rate (ALR) as the (three) ndependent varables, the queston s to whch extent the ndvdual ndependent varables affect the MPR. Besdes, the nteracton between three explanatory varables s to be consdered once more mplctly wth. The approach s the followng: Based on the results n prevous sectons the presumpton s that the MPR s related to the GDPpC most, the EI second most and the ALR lowest. The relatonshp between the GDPpC and the MPR was 15

17 modelled n 2000 usng a Planck functon, whch wll be used now as class of underlyng transformaton functons f1 for the GDPpC data. In 2012, the transformaton functon f1 s a MaxLog functon. The relatonshp between MPR and EI has been descrbed both n 2000 and 2012 usng a power functon, whch s now used as a transformaton functon f2 for the EI data. Furthermore, the relatonshp between MPR and ALR n 2000 and 2012, was modelled by an exponental functon f3. For 2000, the followng regresson weghts A, B, C and D result: The result shows that the MPR s related most strongly n 2000 to the GDPpC. It further shows that the EI affects the MPR the second strongest. Besdes, the MPR s determned dfferently strongly by GDPpC and EI. In comparson to these both explanatory varables the ALR has a lower nfluence on the MPR. Consderng the proportonal nfluence of all ndependent varables, the followng relatonshp s found: The GDPpC affects the MPR to 48.7%, the EI to 32.1% and the ALR to 19.2%. How can ths result be nterpreted? In 2000, the moble phone has been used almost exclusvely n the more rcher states. Thus, the GDPpC was the determnng factor for the MPR worldwde. In these countres, furthermore, the ALR s consstently very hgh,.e. n ths group of states the ALR has no great mportance for the heght of the MPR for the purposes of a d_erentaton characterstc. The lower meanng of the EI concernng the MPR s determned above all by states lke Greece, Hong Kong, Italy, Korea, Portugal, Sngapore, Span etc., that have a relatvely hgh MPR, but a low EI have n comparson to states lke Belgum, Denmark, Germany, Fnland, the Netherlands, Norway or Sweden. It can be assumed that the ALR n 2012 wll have an even lower mpact on the MPR, because frstly, the ALR has ncreased worldwde and on the other, the moble phone s now used almost all over the world by many people. Moreover, n ths context s to be expected that the GDPpC and the EI wll determne the MPR agan smlarly strongly. For 2012, the followng regresson weghts A, B, C, D result: The MPR s stll nfluenced strongest by the GDPpC, followed by the EI, ndeed, wth the dfference that now both explanatory varables exert an accordng to porton smlarly strong nfluence on the MPR. Ths result corresponds to the expectatons. The GDPpC has the strongest nfluence on the MPR n 2000 and 2012, but decreases proportonally n ths perod. The EI ncreases t s proportonally nfluence durng the consdered perod, whle the connecton wth the ALR snks a lttle bt. But what could be the reasons for ths? Worldwde, the MPR has massvely ncreased between 2000 and 2012, so that almost the entre world populaton nowadays uses the moble phone. One could logcally conclude from ths that there mght be no more recognzable connecton between the MPR and the ndependent varables, because the moble phone s used by everybody no matter whether somebody s rch or poor, educated or uneducated, able to and wrte or not. In spte of ths fact the analyses delver a dfferentated relatonshp between MPR, GDPpC, EI and ALR. Ths s due to the fact that the MPR s not lmted upwards. There are many countres wth a MPR sgnfcantly over 100%, because one person can have and has more than one moble phone access. A MPR value over 100% s recognzable over all GDPpC areas. So ths s n fact not only n rch states the case. Furthermore one can not observe any compensatory effects by the growth of the populaton n a country or n the whole world wth regard to the ncrease of the MPR. Another reason could le n the fact that the MPR shows a saturaton effect n 2012 worldwde ndeed, but, however, there are stll some countres n Afrca wth a low prosperty level and educatonal standard (Ethopa, Burund, Ertrea, Krbat, Congo, Mozambque, Nger, Rwanda, Serra Leone, Togo, Chad, Uganda, Central Afrca etc. ) n whch stll less than half of the populaton has a moble phone. 16

18 4.3.2 Combnaton of the factors IPR, GDPpC, EI, ALR If one looks at the Internet penetraton rate (IPR) as the dependent varable and the GDP per capta (GDPpC), the Educaton ndex (EI) and the adult lteracy rate (ALR) as the ndependent varables, s the queston agan wth whch weght the sngle ndependent varables n_uence the IPR. Here are mplctly the nteractons between the three explanatory varables to consder wth. It s placed the followng approach: For 2000 the followng regresson weghts A, B, C and D result: The result shows that the IPR s most related to the GDPpC, second most to the EI and lowest to the ALR. Here the GDPpC and the EI have a smlar nfluence on the IPR. Ths ndcates once more to a (possble) hgh correlaton of economc power and educaton level of a country. Compared to GDP and EI the ALR has (sgnfcantly) less mpact on the IPR. A negatve regresson weght, as t s the case for C, s an artefact of the optmzaton. Concernng the contents the negatve regresson coeffcent C concernng the ALR means that wth rsng ALR values there s a smaller probablty for the MPR to accept hgh values. In comparson to that postve regresson coeffcents, as t s the case for the GDPpC and the EI, wth rsng values for the GDPpC and the EI lead to rsng probabltes for hgh MPR values. Consderng the proportonal nfluence of all ndependent varables, one fnds the followng relatonshp: The GDPpC affects the IPR to 44.9%, the EI to 39.1% and the ALR to 16.0%. Though the readng and wrtng ablty s one of the basc competence of people to be able to use the Internet, ndeed, the access to the Internet s frst a queston of cost. A wde lteracy of the populaton s further a necessary condton for a correspondngly hgh level of prosperty n a country, but not a suffcent condton. A hgh level of prosperty n turn allows more lkely to be able to afford Internet access. In 2000 an Internet connecton s to be found almost only n rch countres and n these countres nearly the whole populaton s alphabetsed. From ths results the low correlaton between IPR and ALR. Wthn ths group of rch countres there are dfferences regardng the educaton level n form of the EI, what explans the strong meanng of the EI. What s the stuaton after 12 years of development n 2012? The followng regresson proportons A, B, C and D result: It becomes clear frst of all that all the regresson coeffcents have reduced (accordng to amount). Ths s related to the used transformaton functons. It also turns out that the IPR s stll most related to the GDPpC followed by the EI. The IPR s nfluenced by the GDPpC to 57.3%, by the EI to 40.5% and the ALR to 2.2%. The IPR has rsen between 2000 and 2012 worldwde on just 40%. In ths tme frame the bggest (absolute) growth concernng the IPR has taken place n the rch states, what explans the ncrease of the meanng of the GDPpC as explanatory varable. In comparson wth that the slghtest IPR growth took place n the poorest states. Wthn the group of rch states an accordngly hgh EI s gven. Furthermore, these countres already have an ALR very close to the 100% lmt, that consequently has almost no sgnfcance for the explanaton of IPR level. Now the IPR has ncreased worldwde, also as a result of access to the moble Internet. How s t to be explaned then that the weght of the ALR s so low n 2012? The reason could le n the fact that wthn the countres n whch the Internet s wdespread,.e. a hgh IPR, there are (clearly) hgher dfferences wth regard to the EI than wth regard to the ALR, because most countres have a very smlar ALR. The result s clear, because the ALR wthn the states, whch use the Internet, s ether already very hgh before 2000 or have rsen untl 2012 on a very hgh level. By technologcal means, as already descrbed on top, the narrow couplng of the Internet access wth the fxed 17