Research on how internet use relates to economic growth

(1)

Research on how internet use relates to

economic growth

Julius Smith 1308793 29th of August, 2007

This research tries to estimate the impact the internet has on economic growth: how does internet use relate to economic growth? No straightforward answers were expected, no straightforward answers could be given. It was found that the internet use variable has a negative relation with economic growth, except for service based economies where no relation could be found. It is expected that the internet mostly benefits service-based activities. The internet can be put to use in so many ways, that the ‘internet use’ variable is a too wide concept to link it to economic growth. A further study on (local) applications of the internet is thought to be of much value.

Supervisors:

Dr. M. Koetter

International Economics & Business, Efficiency & Growth

Dr. E.H. van Leeuwen

International Economics & Business

(2)

2

1.0 Introduction

Much has been written about the internet and the revolution it is said to bring, or have brought. However, not much statistical research exists on the economic impact of internet use so far. The adoption of internet is going tremendously fast. In the Netherlands the first public internet provider1 started 1st of May 1993. Although almost nobody knew of the internet, the first 500.000 users signed up the same day. Only twelve years later 74% of the Dutch population uses the internet on a regular basis.2

With the increase of worldwide internet users came a massive increase in its functionality. Initially focused around email and relatively static information, the internet has developed to, among other applications, a platform for intelligent machine to machine communication, an interactive exchange of all sorts of information, and the

largest trading area of the world.

Several authors describe the

internet as one of the pillars of the information revolution3, parallel to the industrial revolution. In this parallel the internet is seen as the electricity of the 21st century: a commodity that spurs economic growth. But why should internet use increase economic growth?

First this research assumes internet is expected to increase productivity. Easy access to information, fast worldwide communication possibilities to both people and machine, and one global market for goods and services are expected to have quite a positive effect on innovation and efficiency, and hence productivity. Second, the research expects internet to raise the human capital levels. Access to a vast amount of information, and facilitating communication among peers and with experts worldwide make knowledge global, and accessible to everyone.

The ITU data series on internet use by now allow for the first investigative research. No wonders are expected, solely answers to two starting questions:

Does internet use actually relate to economic growth? And if so,

How does it relate to economic growth?

1_{http://www.xs4all.nl/overxs4all/geschiedenis} 2_{http://www.itu.int/ITU-D/icteye/Indicators/}

3_{See the discussion around ‘N.G. Carr – IT Doesn’t Matter, 2003, Harvard Business Review’}

Internet is so big, so powerful and pointless that for

some people it is a complete substitute for life.

(4)

4

1.1 Research Objectives

The objective of this research is to investigate the relation between a countries’ internet use and the growth of an economy. It is expected that internet use has a positive effect on economic growth by affecting two important variables for economic growth; productivity and human capital.

Following Pilat (1996) productivity growth can be allotted to three main processes, and the internet positively influences all three. The first is innovation, the invention of new products and services which is mainly driven by knowledge (Schumpeter, 1954). The internet brings access to information, communication to clients, peers and experts, and thus is an excellent platform for knowledge accumulation (Tuomi, 2000). Second is the reduction of technical inefficiencies, which is strongly stimulated by competition. Michael Porter (2001) not only found that the internet strongly increases competition, also the internet offers large opportunities for operational effectiveness. The third process Pilat describes to increases productivity is technological diffusion; improving productivity by using processes from elsewhere. Technological diffusion is stimulated by trade and knowledge transfer. Freund & Weinhold (2003) found internet to increase international trade. And by offering all current knowledge transfer technologies (writing, listening, and viewing) combined and more, the internet probably is the best technology for knowledge transfer in human history.

The fact that the internet is such a good knowledge transfer technology (Phipps and Merisotis, 2000), and platform for knowledge accumulation (Tuomi, 2000) also explains why internet is expected to have a positive effect on human capital.

To conclude the following hypotheses are stated:

Main Hypothesis: Internet use is positively related to economic growth Hypothesis 1 Internet is a technology that relates to growth via productivity.

(5)

5

1.2 Overview of the study

The goal of this research is to find whether internet use relates to economic growth, and how internet use relates to economic growth. The neoclassical growth model (ch 2.1) is used as a framework for this empirical research.

Internet use is expected to relate to economic growth via two possibilities (ch 2.2). On the one hand internet use is expected to increase productivity; and thus provide sustaining growth (ch 2.1). On the other hand internet is expected to increase the human capital, which is a part of total capital [K] in the growth model (ch 2.1)

Prior econometric research on economic growth in relation to internet use is close to non-existent (ch 2.2). A model is built based on variables from the literature and growth theories (ch 2.3).

Being one of the first quantitative researchers linking economic growth to internet use, it serves indicative purposes mainly. Plain OLS analysis was found to be optimal for this situation (ch 3), combined with the analysis of internet use as an instrumental variable on human capital (ch 3).

Neoclassical growth theory

Prior Growth Findings

(6)

6

2. Theoretical Model

This research focuses on the effects of internet use on economic growth. Two main streams for studying economic growth can be found in the literature. On the one hand there is the endogenous growth model, in which growth is determined by the variables inside the model (Romer, 1990). On the other hand is the exogenous growth model, in which sustaining economical growth is only brought by technological growth, and technological growth is exogenous (Solow, 1956)

Both models have advantages and pitfalls. The endogenous growth models predict a spiral of growth, as the progression of knowledge on economic

growth would enable ever better policies. Endogenous growth models also explain the differences in growth between countries.

Neoclassical, exogenous growth models state that only unexplainable technological change can bring economic growth, which is rather simplistic. Also the neoclassical’s model’s inclination of output convergence between countries can hardly be observed. However, the idea of ‘invisible hands’ always balancing the economy back to its steady state combined with the idea that only productivity gains brings real economic gain does remain rather attractive.

As this research is searching for an opening solely, a simple but tested model fits this research perfectly. Furthermore, from an economic perspective internet is a rather ‘unexplainable’ technological progression. These two facts led to the decision to base this growth research on the neoclassical growth model.

2.1 Neoclassical growth theory

The model used in this research is based on the extended Barro model (2000). And Barro also uses the Solow (1956) growth model to base his model on. Solow looks at growth from a macro-economic view and sees technological progress as exogenous. Before continuing with the Barro model an explanation on the Solow model is given. The Solow growth model states that output [Y] is a function of capital [K], labor [L], and technology [A]. Technology can be equaled to productivity. The resulting function is shown in Equation 1:

Equation 1 , 

As the growth model is assumes a constant returns to scale, the entire equation can be divided by labor in order to get output per worker, or labor productivity:

Technology growth is a measure of

ignorance

(7)

7

Equation 2 , or in different terms: y 

The three variables in Equation 1 are not static, but influenced by other variables. Capital [K] is influenced by two variables, namely the savings rate [s], and capital depreciation [δ]. Labor [L] logically is influenced by the labor force growth [n], and technology growth [A] is influenced by productivity growth [g]. Looking at the impact of these variables on the per worker function (Equation 2) we find that all these three variables impact the capital stock.

The main prediction of the Solow growth model is that a long term equilibrium exists between all these variables. This equilibrium is called the “steady state” and represented by Equation 3

Equation 3

In this steady state only technological progression [A], or actually productivity growth [g], can create sustaining economic growth. Changes in savings, depreciation, or labor force will only lead to a short term change in output growth before it reaches a new equilibrium in which the economy is balanced again: a new steady state.

This is shown in Figure 1. Although output will increase when there are more investments, it will stabilize in the new equilibrium, and also decrease when investments are decreased later. The only sustainable growth is created in the green line; by exogenous technology growth.

Although in theory the same level of technological growth is available to everyone, Pavitt (1984) found that for different economic sectors a different level of technological change exists, as specific knowledge is not generally applicable. Another theoretic assumption (Solow, 1956) with regard to technological growth is that all countries worldwide have access to the same technology. However, Comin & Hobijn (2004) empirically found that in practice consistent lags in technology adoption exist around the world. So, technological growth changes per country and per economic sector.

Another difference between countries can show from the differences in levels of investments. In Figure 1 it shows that the higher the savings rate (investment), the higher the output, and the lower the

k*gold … ..… … … … … … … … …

(8)

8 consumption is (output-savings). It was Phelps (1961)4 that saw that the consumptions vs. savings ratio is a current vs. future consumption dilemma. Referring to the biblical adagio “do unto others as you would have them do unto you” he named the optimal the golden rule steady state. In the golden rule steady state current and future consumption are maximized. The golden rule optimum is represented by k*gold in Figure 1. Although it is a steady state, it is not self-sustaining. The golden rule optimum should be created by adjusting the savings rate, and/or [δ], [n], and [g] to bring the economy at the golden rule investment level. In order to keep the golden rule optimum, the four variables should be adjusted to counterbalance changes. In practice the less difficult variable to alter is the savings rate (Phelps, 1961).

Now what does this neoclassical growth theory tells us about growth? Economic growth can be seen as the difference between the economic output at a certain point, and the economic output at an earlier point. The economic output can be written as Equation 1. Now look at the economic output at a future point in time, represented by future values of the same economic output equation:

Equation 4  

Future output differs from the current output due to changes in A and k. But, how do At+1 and kt+1

differ from A and k? According to Solow (1956) the technological progression [A] is set by the technological growth rate [g].

Equation 5 _! 1 

When an economy is not in the steady state, the differences in actual investments and needed investments will converge the economy to a new steady state automatically (Solow, 1956). In a steady state the capital stock is equal to the depreciation of capital, the change in labor force, and change in technological growth. Taking all this into account k_$! can be stated to be:

Equation 6 _! % & 

Now that both current and future output has been described, let’s look at output growth:

Equation 7 ' ()*+ ( )*+)*+ ! ,

From Equation 7 it can be seen that economic growth, aside from the steady state growth, depends on five variables, namely: technological growth [g], savings or investment rate [s], the capital stock [k], capital depreciation [δ], and labor population growth [n]. So when combining these variables to a function with GDP growth in the short run, assuming linearity, it should have the form of Equation 8. In the long run only technological growth creates sustainable growth.

(9)

9

Equation 8 ' -_! & -_. -_/% & -₀ -₁ 2

Now the basic Solow model is extended with another sort of capital, namely human capital. Instead of Equation 1, with only capital and labor, capital [H] is added. In the per worker function this brings us Equation 9.

Equation 9 y , 3

Whereas the change in physical capital is determined by the savings rate [s] the change in human capital is determined by the education rate [sh](Sachs & Warner, 1997). However, Mankiw, Romer

and Weil (1992) found that the physical savings rate also has an impact on the level of human capital. Equation 8 should thus also be extended to include the human capital savingsrate [sh]. Another

addition to the Solow model is more controversial; the addition of openness. Various authors (among others, Harrison, 1994) have found openness to be a variable of growth. However it remains elusive how openness should be fit in the Solow model. And as more exogenous variables are expected to be of significance for an empirical analysis we follow Sachs and Warner (1997) by adding another variable [Z] for the exogenous influences. All together this leads to Equation 10 which will be used for the research.

Equation 10 ' -_! & -_. -_/%₄& -₀ -₁ -₅%₆ -₇8 2

2.2 Prior empirical findings on the research variables

This section draws upon empiric evidence found regarding the variables important to this research. The main focus has been on the common ground that shows from the researches by Barro (1991, 1997), Levine & Renelt (1992), Sala-i-Martin (1997), Sachs & Warner (1997), Easterly & Levine (1999), and Doppelhofer (2000).

All the above named researches showed the importance of capital intensity [k] and population growth [n], as predicted by the Solow growth theory. The proxy used for capital intensity [k] is the initial GDP. According to the convergence theory (Solow, 1956) it was found that the lower the capital intensity [k], the higher the expected GDP growth. Population growth is important as it has an impact on the per capita GDP data, but also impacts the workforce growth [n] (See Equation 10). It is thus expected to be negatively related to per capita economic growth. Those authors that significantly included investments [sk] in their researches found a positive relation with economic growth, as could

(10)

10 Similar to physical capital the convergence theory (Solow, 1956) this also applies to the human capital stock, resulting in a negative relation to economic growth (Sala-i-Martin, 1997). On the other hand, a high capital stock enables fast adoption of technological advances and thus increases economic growth, resulting in a positive relation to economic growth (Barro & Sala-i-Martin, 1997). The saving rate of human capital [sh] is expected to have a positive effect on economic growth (Sachs

& Warner, 1997). Another discussion is on which proxy fits human capital best (Fuente and Domenech, 2001 / Barro, 2001 / Self and Grabowski, 2004). Schooling enrollment rates represent the current flow of human capital, and thus the savings rate of human capital [sk]. Life-expectancy can be

used as a proxy for the current stock of Human Capital (Barro, 1991). Both flow and stock have deficiencies; Life-expectancy might underestimate the countries’ available human capital when the investment policy has been increased in the last decennia (or two). Flow on the other hand might overestimate the capital stock in such an event. Due to the remaining discussions no expectation on the human capital proxy can be made. The schooling enrollment rates are expected to have a positive effect on economic growth.

Openness determines the speed to which an economy can return to its steady state after a change in the levels (Baldwin, 2003). And openness is expected to have an effect on the distribution of technology, and to which extent it has access to technical change, and (human) capital (Irwin, 2000). In is thus expected that the more open the economy is, the more it is fully competitive, as assumed in the Solow model. Various proxies have been used to measure openness, such as real exchange rate distortions (Kormendi & Meguire, 185), number of years of open economy (Sachs & Warner, 1997), and all sorts of combinations of imports and exports. For this research we will use the index as created by Summer & Heston (1991), which is the sum of imports and exports, as these represent one of the main pillars below openness: trade.

(11)

11

2.3 Internet and economic growth

The pursuit of this research is to find the effect and place ‘internet use’ has in the neo-classical growth model. Plainly stated: If internet use has an effect on the economic growth of countries, how does this affect economic growth?

Since the internet is adopted during the last 15 years, data is starting to become available on internet use just now. Cross-Country or even Panel on

evidence the relation between internet and economic growth therefore is sparse.

From the variables that influence economical growth, two variables can be expected to be positively influenced by internet use. First of all technological progression; empirical research already found that ICT investment causes economic growth (van Ark et al., 2002). So it can be expected that internet use has an effect on productivity as well. On the other hand there is an influence of internet use on human capital (Agarwal & Day, 1998), or human capital to be more specific. Internet use is found to both benefit quality and availability of education, and thus can be expected to positively influence the instrumental variable ‘human capital stock’.

Although there undoubtedly are voices that state internet use also influences the savings rate, depreciation or even population growth of a country, this research will not deal with the possible influences of internet use on these variables

2.3.1 Internet as Technological growth

Although Solow once wittingly stated “You can see the computer age everywhere but in the productivity statistics5,” by now many authors (e.g. Van Ark et al, 2002) have found firm relations between ICT and productivity. Kenny (2002) found that there are so many influencing technologies, that the influence of an individual technology is likely to be small. The question is thus whether the internet can be seen as an individual technology, or the enabler for many new technologies: similar to the mass-introduction of electricity.

Gordon (2000) compares computers and the internet with the second Industrial Revolution, but finds only productivity growth due to ICT in the manufacturing of durables. Van Ark, Inklaar and McGuire (2002) performed a cross-industry empirical research and found a dissimilar outcome. In their research they found that the productivity in ICT-using services industries to be significantly higher than the non-ICT using services industry. A further remarkable outcome was the difference in

5_{Also known as the Solow computer paradox, originally stated in New York Review of Books, July 12, 1987}

You can see the computer age everywhere

but in the statistics

(12)

12 productivity growth between the US and the EU. A more recent study by Van Ark, Inklaar, and McGuire (2003) finds that this difference, apart from structural impediments, can be attributed to higher ICT investments in the US. In his latest research Gordon (2003) also found a similar outcome.

This relation between ICT investments and productivity growth has been underwritten by many authors in single country analysis. Oulton (2002) for example found this causation for the UK economy, and Yoo and Kwak (2004) for the Korean economy. The Korean research did not only find that ICT investments directly affect economic development, but also that economic development influences IT investment; creating a spiral of productivity: IT use increases productivity and thus GDP, and a higher GDP increases IT use. On firm level Brynjolfsson and Hitt (2003) show that ICT investments do not increase productivity in the short term. However, in the long run they find highly increased productivity due to these investments. This finding might stroke with a theory that people are gradually learning how to optimally use ICT investments, just how it took a while to find the optimal applications for electricity.

In an effort to find a relation between productivity growth and internet use Blinder (2000) found that productivity growth happened in the U.S. showed rather in line with internet adoption. Time will tell whether the internet really brings us the information revolution, but current empirical evidence clearly shows that investments in ICT increase productivity. As internet use can be seen as the actual usage of ICT investments, this leads to the expectation that (private) internet use show increasing productivity. Internet use as technological growth is thus expected to be positively related to economic growth.

2.3.2 Internet use influencing Human Capital

Apart from internet use as a technological advance that influences the productivity of the production factors, it is also expected to change the level of the capital factor. The focus here is not on physical, but the level of human capital. Human capital is a broad concept of investments in knowledge, skills, and health. The main method to increase, or invest in, the amount of human capital is via education and training (Becker, 1994). In order to find out what the effect of internet use on human capital is, it is thus essential to focus on the effects internet use has on education and training and learning.

Agarwal and Day (1998) found that internet use in education increases the quality of the education in two manners: quality

and availability. The students’ quality of learning and their retention increases, and the perception and communication with the teacher increases. Another example of availability is distance-based learning,

When I took office, only high energy physicists had ever

heard of what is called the Worldwide Web....

Now even my cat has its own page

(13)

13 in which internet strongly reduces the cost of education, especially in remote areas (Phipps and Merisotis, 2000). But apart from traditional educational activities Levin and Arafeh (2002) found that current internet savvy youth uses the internet for a dozen other education related purposes. By some, this evolving of the way internet is used is expected to just have started. (Pence, 2007).

In short it can be stated that the internet is an enabler for low cost communication and conferencing, and information access and retrieval. The internet can thus be expected to radically increase the educational opportunities for its users. Internet use is thus expected to be positively influencing the instrumental variable ‘Human Capital Stock’

3. Methodology

For testing the hypotheses, empirical analysis is used based on an averaged cross country data set. On this cross country data set an ordinary least squares regression analysis is performed to find the relations between the various variables.

The research focuses of this research is on providing direction. The dataset did not prove to be suited for panel data research, mostly due to its abundance of missing data points and autocorrelation problems. Therefore the choice has been made to focus on the core of this research: the impact of ‘internet use,’ and settle with the conservative results of the averaged cross-country OLS regression.

In the first hypothesis internet use is expected to be directly related to economic growth. Based on the theory (See Equation 10), we find the almost same equation as used by Sachs and Warner (1997). Rewriting this equation to match with the variable names used in this research Equation 11 is found.

Equation 11 9_: -_! -_. -_/; -₀% -₁%₆ -₁9<= 2

Equation 11 is the regression in which internet use is entered as technological growth. [lnygr] is the

logarithm of average GDP growth. [k] is the initial GDP value. [n] is the average population growth. [sk] is the average investment share of GDP, [O] is the average openness as share of GDP. [sh] is the

human capital growth variable for which the gross enrollment of tertiary education per hundred habitants is used. [lniu] is the internet use variable, and lastly, ε is the error term.

(14)

14 instrumented by internet use. Equation 13 shows how the human capital variable [sh] is instrumented

by internet use.

Equation 12 9_: -_! -_. -_/; -₀% -₁%> 2 ₆ Equation 13 %> -₆ _?@_A -_!9<= 2

3.1 Variables

In this section the included variables and their chosen proxies are explained. The section will further go into detail into the sources of the data series and the processing methods to optimize their explanatory power.

The economic growth [lnygr] source is the PPP adjusted per capita GDP series of the International

Monetary Fund World Economic Outlook Database 2007. These per capita GDP series are adjusted to current international dollars using WEO purchasing power parity data6. The growth rates have been calculated7, and the geometric mean (CAGR) of the available points is taken. A log distribution of this variable has been chosen to have a more normal distribution of the values. For initial GDP [k] the per capita GDP in 1960 often is used in empiric research to proxy initial capital intensity (e.g. Barro, 1991). However, this study researches economic growth starting from 1999. Therefore the initial GDP [k] value is the countries’ GDP for the year 1998. The source is the Penn World Table RGDPL series. Although economic growth can be calculated from the same source as initial GDP, there has been chosen to use a different source8. Test regressions have confirmed that the model holds when the same source is used9. Based on previous growth research a negative relation between initial GDP and economic growth is expected, see section 2.2. For population growth [n] the actual population growth has been selected. In the short run a large birthrate, with mortality fixed, will decrease the per capita income: same workforce, larger population. Vice versa, a sudden increase in the relative workforce would bring per capita GDP growth. In order to match per capita GDP growth, the variable should thus be (change in workforce)/(change in population). However, for the majority of the countries included in this research no reliable source could be found to provide the workforce growth. The population series are sourced from the Penn World Table 6.2 database, and population growth has been calculated10. The available points for the time period 1999-2005 are averaged using the

6_{For more information see the IMF World Economic Outlook Database, April 2007 edition} 7_{GDP growth =}()

()B+& 1

8_{Using the same source data for two variables will lead to autocorrelation. Furthermore, it was found that the PWT growth variable was not}

fully compatible with GDP related variables from other sources than the Penn World Table. Also see footnote 9.

9

See the appendix for further explanation.

10_{[pgr] =}CDEFGH ?D)

(15)

15 geometric mean. Based on previous growth researches a negative relation between population growth and economic growth is expected11

For the investment share of GDP [sk] the Investment Share of RGDPL series from Penn World Table

6.2 data is used12. The available points for the time period 1999-2005 are averaged using the regular mean13. Based on previous growth researches a positive relation between the savings rate [sk] and

economic growth is expected, see section 2.2. For the openness index the Penn World Table 6.2 openness index has been used. The available points for the time period 1999-2005 are averaged using regular mean. Openness in the Penn World Table 6.2 is defined as the sum of both imports and exports as share of GDP13. Based on previous growth researches a positive relation between openness and economic growth is expected, see section 2.2.Although the current literature acknowledges the effect of human capital on economic growth, discussion about the optimal proxy remains (See section 2.2 Prior empirical findings). For this research it was found that the tertiary net enrollment rates [sh]

gave the best result of the found proxies in section 2.2. This data has been selected from the World Bank education statistical (EDSTATS) database. The available points for the time period 1999-2005 are averaged using the regular mean. Based on previous growth researches a positive relation between [sh] and economic growth is expected, see section 2.2.

For the internet use variable [lniu] data has been used from the ITU, the UN telecommunications division. It represents the internet users per 100 inhabitants of a country14.

In order to discriminate countries along economic activity, also variables for service [Serv], manufacturing [Manu] and agricultural activities [Agri] have been used. These three variables represent the share of GDP earned by the respective activity, and is sourced from the Statistique Canada world indicator database.

11_{See section “2.2 Prior empirical findings”}

12_{For more information see: http://pwt.econ.upenn.edu/php_site/pwt62}

(16)

16

3.2 Descriptive Statistics

In this section the data will be described. Apart from the whole dataset of 148 countries the set has also been stratified according to economic activity. As found in chapter 2 technological growth rates differ per economic activity (Pavitt, 1984). But it is also found that the effects of ICT investments differ per economic activity (Jorgenson & Stiroh, 2000). These subsets allow for the investigation of the effects of internet use on the different sorts of economic activity. The subsets originally have been created by dividing the total dataset in quartiles. However, the quartiles subsets did not render significant regression models. Therefore the whole dataset is split into halves, and thirds. Quartiles have only been regressed, but so sample size was not large enough for significance.

3.2.1 All country dataset

148 countries have been included in this dataset, for which the list of countries can be found in the appendix “Datasets” The

descriptives for this dataset are shown in Table 1.

The S variable represents the percentage of GDP derived from service activities, M represents the percentage of GDP derived from manufacturing activities, and A represents the percentage

of GDP derived from agricultural activities.

All variables but k represent percentage values, the values for k represent year 2000 international dollars.

In Table 2 the results of the pair-wise correlation between the variables used in this research are shown. The first row of each variable shows the coefficients, while the second row shows the significance. In this correlation overview, and in the following regressions the [k] variable has been scaled down in order to provide for better overview. [k] represents, unless otherwise states, thousands of year 2000 international dollars.

According to Carter, Hill, Griffiths, and Judge (2001) a general rule of thumb for signifying multincollinearity is that a correlation between variables should not exceed 0.8. Using this rule of thumb strictly, no multicollinearity problems are expected. Relations between internet use and initial

Variable Obs Mean Std. Dev. Min Max

ygr 148 4.94 % 3.10 -3.39 % 22.21 % k 148 $ 7859 8308 $ 459 $ 41388 n 148 1.33 % 1.20 -5.27 % 3.38 % O 148 87.91 % 49.46 2.14 % 301.0 % sk 148 13.80 % 7.59 2.21 % 44.99 % sh 148 24.67 % 22.74 0.36 % 85.12 % iu 148 11.91 % 15.30 0.06 % 67.22 % Serv 148 54.27 % 14.59 4.30 % 90.00 % Manu 148 14.29 % 7.31 1.60 % 37.20 % Agri 148 15.31 % 13.57 0.10 % 61.80 %

(17)

17 GDP, and relations between internet use and [sh] should however be looked upon closely. The

correlation between [lniu] and [sh] does enable the use of instrumental variable techniques, as it is the

third requirement (Greene, 2000).

Table 2; Pair wise Correlations, all countries

lnygr k n O sk sh lniu Serv Manu Agri

k -0.032 1 0.696 n -0.349 -0.345 1 0.000 0.000 O 0.194 0.284 -0.177 1 0.018 0.001 0.031 sk 0.208 0.583 -0.339 0.233 1 0.011 0.000 0.000 0.004 sh 0.188 0.654 -0.573 0.046 0.436 1 0.023 0.000 0.000 0.580 0.000 lniu 0.045 0.765 -0.550 0.285 0.546 0.758 1 0.589 0.000 0.000 0.001 0.000 0.000 Serv -0.121 0.589 -0.366 0.193 0.325 0.528 0.650 1 0.142 0.000 0.000 0.019 0.000 0.000 0.000 Manu 0.181 0.163 -0.367 0.110 0.159 0.423 0.355 0.141 1 0.028 0.048 0.000 0.184 0.054 0.000 0.000 0.087 Agri -0.196 -0.637 0.455 -0.292 -0.476 -0.617 -0.774 -0.624 -0.325 1 0.017 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

Interpreting the correlations between the variables, no unexpected results are found. Interesting is the confirmation of expectations with regard to economic activities. A strong positive relation between service activities and internet use, and a strong negative relation between agricultural activities and internet use is found.

Agricultural economies are further found to have a significantly lower openness, investment, and human capital share of GDP than manufacturing, and service economies in particular. Also agricultural activities display a negative relation with economic growth.

(18)

18

3.2.2 Service based economies

The countries in this subset represent a part of the total set that have the highest share of their GDP earned by service activities.

In Figure 2 the total (all country) dataset is shown distributed with regard to the percentage of GDP created by service activities. It is this dataset that has been spliced into two in order to capture the top half, and third of this distribution. Also regressions with quartile sets are run, but show much less significant results, most likely due to the limited number of observations.

Table 3 shows the descriptive for this subset. In this

subset an average 65.76% of GDP is derived from services; 15.74% from manufacturing and 7.72% from agricultural activities.

The descriptives for this subset follow expectations. Compared to the ‘all countries’ dataset the service-based economies show a higher mean initial GDP, Openness, Investment share, Tertiary enrollment, and on average twice as much internet use. Population growth, also according to expectation, is lower than the overall set. On the other hand, the average GDP growth hardly differs from the ‘all country’ set, but for a lower standard deviation.

Table 3; Descriptives, Service-Based Economies 50 percentile

ygr 74 4.96 % 2.06 -0.14 % 10.63 % k 74 $ 11957 9414 $ 709 $ 41388 n 74 0.91 % 0.98 -0.93 % 3.25 % O 74 96.92 % 51.96 2.14 % 301.0 % sk 74 16.47 % 7.58 2.21 % 44.99 % sh 74 36.29 % 23.75 0.87 % 85.12 % iu 74 20.01 % 17.50 0.33 % 67.22 % Serv 74 65.76 % 8.07 54.90 % 90.00 % Manu 74 15.74 % 5.69 2.70 % 28.40 % Agri 74 7.72 % 6.91 0.10 % 28.90 % 0 .0 1 .0 2 .0 3 D e n s it y 0 20 40 60 80 100 %service

(19)

19

3.2.3 Economies with a high level of manufacturing activities

The countries in this set represent the half of the total set that have the highest share of their GDP earned by manufacturing activities.

The total data set as shown in Figure 3 has been split into two sets. The top half subset, with the most manufacturing activities, is discussed here. Also for this subset regressions with quartile sets are run, and also show much less significance.

The descriptive to this manufacturing set are shown in Table 4. It shows that the

average percentage of manufacturing (20.11%) to contribute to the GDP is still lower than percentage earned by service activities (56.68%). To the least, not all countries in this subset have manufacturing

activities as primary economic source.

The descriptive of this subset follow expectations. The average initial GDP is situated between the total set average and the average for service-based economies. The same is true for openness, investments, tertiary education, and internet use.

Table 4; Descriptives, Economies with a high level of manufacturing activities

ygr 74 5.51 % 2.80 -3.39 % 14.61 % k 74 $ 8543 7393 $ 504 $ 28274 n 74 0.88 % 1.21 -5.27 % 3.25 % O 74 89.34 % 46.25 2.14 % 211.9 % sk 74 14.48 % 6.75 3.38 % 29.49 % sh 74 33.48 % 21.43 0.86 % 85.12 % Iu 74 14.21 % 15.03 0.06 % 58.65 % Serv 74 56.68 % 10.41 26.10 % 75.90 % Manu 74 20.11 % 5.18 14.10 % 37.20 % Agri 74 11.14 % 8.95 0.90 % 46.00 % 0 .0 1 .0 2 .0 3 .0 4 .0 5 D e n s it y 0 10 20 30 40 %manuf

(20)

20

4. Empirical Analysis

In this section the hypotheses are tested, using the methodology from chapter 3. First these hypotheses will be tested for the whole dataset, they will then be tested for countries which derive a large part of their GDP from services activities, and finally the hypothesis will be tested for countries which derive a large part of their GDP from manufacturing activities. This stratification has been applied because it is found that technological growth rates differ per economic activity (Pavitt, 1984), and because the effects of ICT investments differ per economic activity (Jorgenson & Stiroh, 2000).

The two hypotheses to be tested are:

Hypothesis 1 Internet is a technology that relates to growth via productivity.

Hypothesis 2 Internet is related to human capital and thereby indirectly is related to income growth.

Hypothesis 1 will be tested using a regression in which internet use is entered as an exogenous technological growth variable. When the internet use variable is significantly related to economic growth, the hypothesis can be accepted.

(21)

21

4.1 All Countries

In this section the whole dataset is used to test the hypotheses. Fourteen regressions have been run and shown in Table 5. All large sized figures represent the coefficients, and all small sized figures represent the significance. The main variable used in the research is the per capita log amount of internet users [lniu]. All the variables, but internet use, show the expected sign. The other proxies related to internet use have been added in order to support the findings with the [lniu] variable.

In order to explain why internet use has a negative impact on economic growth, several regressions are run. First the variable is tested. Instead of internet use regressions III and IV tested the growth of internet use; with no result. So the possible reasons for internet use to be negatively related to economic growth will have to be tested.

One cause of the negative productivity might be that internet access alone does not increase productivity. It are the applications available through internet that raise productivity, not the access to the internet. In regressions V and VI this idea is tested. The amount of hosts per capita roughly proxies the number of websites per capita, and is thus a quantitative measure of the available internet applications15. Unfortunately this brings no results, although testing this for countries with a large tourism or SMB16 sector might change the results, this is not in line with the research.

Another possibility might be distorted or noisy data. Regressions VII to XI are similar to III to VI but with another set of proxies. Internet use is proxied by connectivity, which is a mix of access to various network technologies17. The availability of internet applications is proxied by the “e-services” ranking, which is a qualitative measure18. Regressions VII and VIII show the results for the connectivity ranking, but lack significance. Regression IX shows a significant and negative coefficient for the availability of services. In Regression XI it is found that both connectivity and e-services are significant on a 90% level, and connectivity even shows a positive coefficient. This finding strengthens the expectation that internet use only measures a part of what is relevant towards economic growth. The total effect of both variables however remains negative.

Another possibility for the decreasing productivity of internet use might be found in the economic cycle: especially the economic downturn of 2000~2001. The ‘internet bubble burst,’ might have had an amplified impact on countries that had a high level of internet adoption. In order to rule out this possibility regression XII shows the standard regression for the countries with data available between

15_{On the one hand the amount of websites per country might not be a good measure for a truly global place. On the other hand not all}

applications have global use (i.e. online tax administration), and to a certain extend language barriers remain.

16

SMB: small and medium sized businesses

17_{See the Economist Intelligence Unit website for more information on the “e-readiness” ranking and its subcategories “connectivity” and}

“e-services” http://www.eiu.com/site_info.asp?info_name=eiu_2006_e_readiness_rankings&rf=0

18_{A ranking which measures the availability of consulting and technical support services, the availability of back office support and the}

(22)

22 2001 and 2006. Unfortunately this rendered a different set of countries in which a bias is expected19. Between these years most countries showed positive economic growth rates. As the internet use variable no longer is significant between 2001 and 2006, the ‘internet bubble burst’ might very well have been of impact in the total set. However, when the investment variable is omitted [lniu] is significant and shows a negative relation. But because collinearity between [sk] and [lniu] is not

expected (correlation = 0.5663), and a different set of countries is used, the only conclusion that can be drawn is that [lniu] is not significant for the years after 2001 for a different set of countries.

Looking at hypothesis 1, there is a relation between internet use and economic growth, but, the main hypothesis assumes the relation between internet use and economic growth to be positive. Therefore the main hypothesis must be rejected, and thus so must hypothesis 1. Internet use does not positively relate to economic growth as an exogenous technological growth variable.

In regression XIV the second hypothesis is tested; does internet use relate to economic growth via human capital? Although the third rule20 for applying instrumental variable techniques has been broken, the human capital growth variable is taken as instrumented, endogenous variable: [sh] is

instrumented by [lniu]. The regression shows that the instrumented variable [sh] is not significant.

Also initial GDP [k] and openness [O] show insignificance. Finally the R-square is rather low. The second hypothesis must clearly be rejected; internet use is not indirectly related to GDP growth via human capital.

19_{See the appendix for descriptives, but one reason for the expected bias is that the post 2001 set shows a much higher mean initial GDP.} 20_{The instrument may not directly act on the dependant [lny}

(23)

Table 5; All country dataset

21_{When the investment variable [s}

k] is omitted, [lniu] is significant and shows a negative coefficient.

22_{iugr is the mean growth of internet users per capita, iugr2 is the geometric (CAGR) average of internet users per capita. Internet hosts are the numbers of hosts per capita, also from the ITU dataset}

23_{Connectivity and e-services both are rankings from the combined Economist Intelligence Unit and IBM research on countries’ “e-readiness,” see appendix for more information}

(24)

4.2 Service based economies

In this section the same hypotheses are tested, but now using the dataset for service based economies. The total dataset has been split in halves and thirds. In Table 6 the stratified sets are shown, regressions I, II and IV represent the top 50% of the total dataset with regard to the amount of service based activities. Regressions III and V represent the top 33% of the whole dataset.

Table 6; Service-based economies

lnygr I 50 percentile II 50 percentile III 33 percentile IV 50 percentile instrumented V 33 percentile instrumented k -0.00121 -0.00123 0.00000 -0.00125 -0.00109 0.000 0.001 0.021 0.003 0.021 n -0.68974 -0.68635 -0.28676 -0.65292 -0.10418 0.004 0.005 0.369 0.059 0.824 O 0.01327 0.01320 0.00013 0.01372 0.01536 0.001 0.001 0.007 0.007 0.013 sk 0.03375 0.03393 0.00023 0.03436 0.00966 0.259 0.261 0.661 0.255 0.867 sh 0.03734 0.03645 0.00029 0.04061 0.05603 0.002 0.007 0.076 0.107 0.134

lniu 0.00041 0.00328 instrument instrument

0.883 0.502 ε 0.72669 0.72486 0.31443 0.68833 0.12834 0.003 0.004 0.337 0.055 0.790 Wald 0.000 0.000 0.000 0.000 0.000 R-sq 0.438 0.438 0.299 0.437 0.253 # of obs. 74 74 49 74 49

In regressions II to III the outcomes of the regression with [lniu] as exogenous technological growth variable are shown. All variables show the expected sign. Internet use shows to be insignificantly related to economic growth. The coefficient of [lniu] is positive, and thus differs from the regression on the whole dataset. However due to its insignificance no conclusions can be drawn on the coefficients sign. The positive sign of the internet use variable is also found in the 33-percentile (and quartile) dataset, but there also lacks significance. Concluding, the first hypothesis must be rejected for this subset; internet use is not directly related to economic growth for service based economies.

In regressions V and VI the human capital variable is taken as instrumented, endogenous variable. [sh]

(25)

25

4.3 Economies with a high level of manufacturing activities

Similar to the last section the total dataset has been split in halves and thirds, this time with regard to manufacturing activities. In the Table 7 the stratified sets are shown, regressions I, II and IV represent the top 50% of the total dataset with regard to the amount of service based activities. Regressions III and V represent the top 33% of the whole dataset.

Table 7; Economies with a high level of manufacturing activities

I 50 percentile II 50 percentile III 33 percentile IV 50 percentile instrumented V 33 percentile instrumented k -0.00206 -0.00152 -0.00161 -0.00067 0.00000 0.001 0.014 0.018 0.437 0.576 n -0.69981 -0.76182 -0.67570 -1.12662 -0.97494 0.006 0.002 0.008 0.476 0.219 O 0.01453 0.01469 0.01418 0.00979 0.01040 0.016 0.012 0.027 0.002 0.007 sk 0.02872 0.04910 0.03422 0.00268 0.00299 0.582 0.337 0.530 0.169 0.178 sh 0.04310 0.06589 0.05357 -0.03186 -0.02521 0.016 0.001 0.012 0.964 0.963

lniu -0.00739 -0.00610 instrument instrument

0.017 0.080 ε 0.74533 0.80657 0.72770 1.19705 1.04853 0.004 0.001 0.005 0.001 0.006 Wald 0.000 0.000 0.000 0.000 0.000 R-sq 0.332 0.387 0.441 0.150 0.243 # of obs. 74 74 49 74 49

Again, all basic variables show the expected sign. And similar to the total dataset the internet use variable shows a negative coefficient.

Regressions II and III are used to test hypothesis 1. Based on the outcome, the main hypothesis is rejected, and thus is the first hypothesis; internet does not positively relate to economic growth.

For the second hypothesis the human capital variable is taken as instrumented, endogenous variable. [sh] is instrumented by [lniu]. Similar to the ‘all country’ data set the second requirement for using

instrumental techniques is not met24. Looking at the significance of the instrumental regressions IV and V it is found that only population growth and the error term shows significance. The second hypothesis must thus be rejected. Internet use is not related to economic growth via human capital.

(26)

26

4.4 Summary of the empirical findings

Countries with a high level of manufacturing activities have the highest mean growth of GDP. Service-based economies show the lowest standard deviation in GDP growth. Both service and agricultural economic activities are negatively related to economic growth.

There exists a positive correlation between service-based activities and the amount of internet users, the more an economy is service based, the more internet users, and/or vice versa. The same is true for economies with a high level of manufacturing based activities, but with roughly half the effect. The relation between agricultural activities and internet use is negative; the more agricultural activities in an economy, the less internet users it has.

The same relations between economic activities and tertiary education are found, but for the difference between service and manufacturing economic activities. Both show an almost similar positive relation to tertiary education enrollment.

The relation between internet use and economic growth was found to be negative on average. Countries with a high level of manufacturing activities only show a marginally less negative relation between internet use and economic growth. Service-based economies show no relation between internet use and economic growth.

The internet use variable is found to proxy the sought concept, but might include some noise. The explained variation in the regressions is lower than the theory predicts, probably due to the (too) wide set of countries25.

Internet use only partly describes technological change, the available applications also are important.

For the period after the internet bubble burst in 2000-2001, which might have hurt economies with a high number of internet users most, the relation between internet use and economic growth only is significant when the investment variable is omitted. The internet bubble burst has thus been of influence, and distorted the relation.

Internet use was not found to indirectly influence economic growth through tertiary education enrollment but was found to be positively related to tertiary education enrollment26

(27)

27

5. Conclusions and Discussion

The outcome of this research is contradictory to the expectations, and common belief. It was expected that either a positive relation between internet use and economic growth was found, parallel to the findings of ICT investments (van Ark et al, 2002). Or that indirectly internet use was related to economic growth via human capital (Agarwal & Day, 1998). Instead internet use is negatively related to economic growth. But then again, in 1987, 6 years after the introduction of the IBM pc, long after the introduction of the mainframe, Solow stated that the computer age did not effect GDP growth.

Although the data is not perfect, it is believed that the found relations do match reality. The fact that

internet use is negatively related to growth in the average economy, but insignificant to service-based economies, partly is according to expectations: the relation is more positive for service-based

economies.

Previous research proved ICT investments to benefactor GDP growth: but ICT investments mostly are done by businesses. The internet use variable focuses on the population, which might very well explain the difference. Analyzing the countries with the highest GDP growth, it could be argued that internet use has only a marginal impact on daily life for the “high growth countries”. Most economic activities are rather straightforward, as is the needed communication and education; internet might not offer extra efficiency but only unneeded functionality (for now). As investments are one of the most important pillars for growth in these “high growth countries”, it might even be argued that internet connectivity is a sub-optimal investment at this stage. The functionality brought by the internet can be expected to mostly benefit service activities, for which only a small (but growing) market exists.

Due to the wide application possibilities of the internet’s functionalities, service economies are not expected to benefit from internet use evenly. Jamaica, Estonia, and the U.K. for example show rather similar adoption levels. But it can be expected that each have different applications bringing most economic benefit. Where the Jamaican population can be expected to benefit through tourism from its internet use, the U.K. population might be benefited most by the efficiency of online government services. Also here the distinction between business ICT investments and population’s internet is important; the advantages internet brings to Swiss banks mostly are reaped through ICT investments, not due to the local population using internet (but the Swiss banks are prone to foreign populations’ internet use).

So in order to locally benefit from a population’s internet use, not only adoption is necessary but also

(28)

28 It is argued that the usefulness of the offered applications for a large part is determined locally: from online banking, last-minute holiday bookings, to tax administration; the availability is geographically dependant. The part of a country’s population that has access to the internet is half the necessity for the virtual world to bring the real world economic growth. The country’s availability and economic usefulness of local internet functionality is the other.

An initial attempt to test this local functionality of the internet has been performed with the EIU27 ranking on e-services, and proofed successful. The amount of hosts (roughly equal to websites) per capita did not show any significant results, but is expected to be a too quantitative proxy. For future research it would thus be interesting to categorize the available local applications of the internet and use this category set to regress against economic growth to find their economic impact.

(29)

29

References

Abramovitz, M. 1956, Resource and Output Trends in the United States since 1870, The American

Economic Review

Agarwal, R. & Day, A.E. 1998, The impact of Internet on Economic Education, Journal of Economic

Education

Ark, B. van; Inklaar, R. & McGuckin, R. 2002,"Changing Gear" Productivity, ICT and Service Industries: Europe and the United States, GGDC Research Memorandum

Banerjee, A. 1999, Panel data unit roots and cointegration: an overview, Oxford bulletin of economics

and statistics

Barro, R. & Sala-i-Martin, X. 1997, Technology Diffusion, convergence, and Growth, Journal of

Economic Growth

Barro, R. 1991, Economic growth in a cross section of countries, The Quarterly Journal of

Economics

Barro, R. 1996, Determinants of economic growth: a cross-country empirical study, National Bureau

of Economic Research.

Barro, R. 2000, Inequality and growth in a panel of countries, Journal of Economic Growth Barro, R. and Sala-i-Martin, X. 1995, Economic growth, New York, McGraw-Hill

Blinder, S.A. 2000, The Internet and the New Economy, Brookings Policy Brief

Brynjolfsson, E. & Hitt, L.M. 2003, Computing Productivity: Firm level evidence, Review of

Economics & Statistics

Carter Hill, R., Griffiths, W. and Judge, G. 2001, Undergraduate econometrics, New York, Wiley and

Sons Inc.

Castells, M. 2002, The Internet Galaxy, Oxford University Press

Comin, D. & Hobijn, B. 2004, Neoclassical growth and the adoption of Technologies, National

Bureau of Economic Research

Doppelhofer, G., Miller, R. and Sala-i-Martin, X. 2000, Determinants of long-term growth: a Bayesian Averaging of Classical Estimates (BACE) approach, Organisation for Economic Co-operation and Development.

Easterly, W. & Levine, R. 1997, Africa's Growth Tragedy: Policies and Ethnic divisions, The

Quarterly Journal of Economics

Ferdinand, P.C. 2000, The internet, democracy, and democratization, Frank Cass Publishers Freund, C.L. & Weinhold, D. 2003, The effect of Internet on International Trade, Journal of

International Economics

Fuente, A. de la & Doménech, R. 2001, Schooling Data, Technological Diffusion, and the Neoclassical Model, The American Economic Review

Gordon, R.J. 2000, Does the "new" economy measure up to the great inventions of the past?, National Bureau of Economic Research

Greene, W.H. 2000, Econometric Analysis, Fourth edition, Prentice Hall

Grootaert, C. & Bastelaert, van T. 2002, Understanding and measuring Social Capital. A Multidisciplinary Tool for Practitioners, The World Bank

Gujarati, D. 2003, Basic Econometrics, 4th edition New York: McGraw Hill

Harrison, A. 1994, Openness and Growth; a time series, cross-country analysis for developing countries, journal of Development Economics

Husted, S. and Melvin, M. 2001, International Economics, Boston: Addison Wesley Longman. Jorgenson, D.W. & Stiroh, K.J. 1999, Information Technology and Growth, The American Economic

(30)

30 Kenny, C. 2002, The Internet and Economic Growth in Least Developed Countries; a case of

managing expectations?, World Institute for Development Economic Research

Kormendi, R.C. & Meguire, P.G. 1985, Macroeconomic Determinants of Economic Growth; Cross Country Evidence, Journal of Monetary Economics

Levin, D. & Arafeh, S. 2002, The Digital Disconnect: The Widening Gap between internet Savvy students and their Schools, Pew Internet and American Life Project

Levine, R. and Revelt, D. 1992, A sensitivity analysis of cross-country growth regressions, The American Economic Review

Litan, R.E. & Rivlin, A.M. 2001, Beyond the dot.coms : the economic promise of the Internet, Washington, D.C : Brookings Institution Press

Lucas, R.E. 1988, On the Mechanics of Economic Development, Journal of Monetary Economics Mankiw, N., Romer D. and Weil, D. 1992, A contribution to the empirics of economic growth,

Quarterly Journal of Economics

Mankiw, N.G. 2000, Macroeconomics, Fourth edition, Worth Publishers

Oulton, N. 2002, ICT and Productivity Growth in the United Kingdom, Oxford Review of Economic Policy

Pavitt, K. 1984, Sectoral Patterns of Technical Change: Towards a Taxonomy and a theory, Research Policy

Pence, H.E. 2007, Preparing for the Real Web Generation, Journal of Educational Technology Systems

Phelps, E.S. 1961, The Golden Rule of Accumulation: a fable for growthmen, The American Economic Review

Philbeam, K. 1998, International Finance, Second Edition, Palgrave

Phipps, R. & Merisotis, J. 2000, Quality on the Line: Benchmark for Success in Internet-Based Distance Learning, National Education Assocation

Pilat, D. 1996, Competition, Productivity, and Efficiency, OECD Economic studies no.27 Porter, M.E. 2001, Strategy and the Internet, Harvard Business Review

Romer, P.M. 1990, Endogenous Technological Change, Journal of Political Economy

Sachs, J.D. & Warner, A.M. 1997, Fundamental sources of Long-Run Growth, The American Economic Review

Sala-i-Martin, X. 1997, I just ran four million regressions, National Bureau of Economic Research Schumpeter, J. 1954, The History of Economic Analysis, Taylor & Francis

Self, S. & Grabowski, R. 2004, Does education at all levels cause growth? India, a case study, Economics of Education Review

Solow, R. 1956, A contribution to the theory of economic growth, The Quarterly Journal of Economics

Solow, R. 1994, Perspectives on Growth Theory, Journal of Economic Perspectives

Summer, R. & Heston, A. 1991, The Penn World Table (Mark 5): An Expanded Set of International Comparisons, 1950-1988, Quarterly Journal of Economics

Tuomi, I 2000, Internet, Innovation, and Open Source; Actors in the Network, First Monday / Finnish national fund for R&D

Research on how internet use relates to economic growth