IMPACT OF ICT ADOPTION ON INEQUALITY:
Evidence from Indonesian Provinces
An International Economics and Business’ Master Thesis
Student: Harry Patria
Student Number: s3732797
Supervisor: dr. A.A. (Abdul) Erumban
Co-Assessor: T.M. (Tarek) Harchaoui, PhD
University of Groningen
Faculty of Economics and Business
ii Abstract
This study investigates the relationship between ICT adoption ratio and income inequality. While the majority studies explain the impact of ICT on income inequality via labour market, this study offers a different perspective on this relationship. Fast growing ICT, in terms of ICT-based company and ICT users influence almost every human aspect nowadays. This might also influence the income structure in society, not only the employment income, but also the household income, because some studies show that there are certain types of incomes that can be acquired by means of ICT. However, these types of incomes are not covered in employment income, such as property income, consumer surplus, etc. Thus, this study seeks to show the impact of ICT on income inequality via household income channel. In addition, Indonesia has the largest internet economy in the world, valued at roughly 27 billion US dollars. Moreover, internet adoption in Indonesia increased remarkably from approximately 30% in 2012 to become 45% in 2016. These facts demonstrate the considerable impact of ICT on the lives and income of people in Indonesia. By using panel data regression, this paper shows an inverted U-shape relationship between ICT adoption and income inequality. Low ICT adoption will increase income inequality until a certain turning point, whereby higher ICT adoption will reduce income inequality in society. The first difference of the Gini coefficient with respect to the ICT adoption shows that the turning point relating to average adoption ratio of mobile phone, computer, and internet is 25%; while there is an average adoption ratio of 17% for computer and internet. Therefore, it is important for government as the policy maker to make sure that ICT adoption ratio is more than the turning point so ICT can give positive impact on income equality.
iii Table of Contents
Cover Page ... i
Abstract ... ii
Table of Contents ... iii
List of Tables ... iv
List of Figures ... v
1. Introduction ... 1
2. Literature Review ... 4
2.1. Impact of ICT on Economic Growth ... 4
2.2. Impact of ICT on Household ... 5
2.3. Impact of ICT on Income Inequality ... 6
3. Research Questions and Hypotheses ... 7
4. Some Stylised Facts: ICT and Inequality in Indonesia ... 8
4.1. ICT in Indonesia ... 8
4.2. Income Inequality in Indonesia ... 10
5. Methodology ... 11
6. Data and Variables ... 13
6.1. Dependent Variable ... 13 6.2. Independent Variables ... 13 6.3. Control Variables ... 14 7. Empirical Results ... 16 7.1. Preliminary Tests ... 16 7.2. Regression Results ... 17
7.3. Regression Results without Mobile Phone as a Proxy of ICT ... 19
7.4. Issue of Endogeneity ... 21
iv List of Tables
Table 1. Literature of the Impact of ICT adoption on households ... 5
Table 2. Number of Installs and Types of ICT Users: Some Services Applications ... 8
Table 3. Some Types of ICT Services Industries and Its Benefits in Indonesia ... 9
Table 4. Data Summary ... 15
Table 5. Skewness/Kurtosis Tests Results ... 16
Table 6. Correlation Table of Model ... 16
Table 7. Regression Result with Mobile Phone, Internet and Computer as ICT ... 17
Table 8. Regression Result with Internet and Computer as ICT ... 19
List of Figures Figure 1. Internet adoption ratio in Indonesian provinces in 2016 ... 3
Figure 2. The summary of channels of ICT influence income inequality ... 7
Figure 3. The ICT adoption ratio in Indonesia 2012-2017 ... 10
1 IMPACT OF ICT ADOPTION ON INEQUALITY: Evidence from Indonesian
Provinces
Harry Patria
1. Introduction
Most economists are paying considerable attention to issues about increasing diffusion of information and communication technology (ICT) and its impact on the economy. Initially, Nobel laureate Robert Solow (1987), shows his scepticism concerning the benefit of ICT, commenting that ‘Computers are everywhere, but not in productivity statistics’. Against Solow’s statement, further studies provide sufficient evidence that ICT positively influences production and productivity with some delays (Brynjolfsson, 1993; Triplett, 1999). Schumpeter(1942), explains productivity delay as a creative destruction process, where old technologywas replaced by a new technology that changes the business process and creates temporary productivity losses in the economy.However, after the creative destruction process ended, a lot of researches have recently demonstrated the benefit of ICT on the economy through three different channels; specifically, production effect, investment effect and
productivity effect (van Ark et al., 2016).
Regarding the first channel, Pilat et al. (2003), explain that ICT production industries, which comprise rapid technical progress and high demand, generate economic growth within a country. The second channel, higher ICT investment offers an improvement of the capital contribution on increasing output in ICT-using industries (van Ark et al, 2011). In addition, the third channel, productivity effect takes place by way of encouraging production activity by multiplying output (van Ark et al., 2011). All three channels show the positive effect of ICT, primarily on the supply side in terms of productivity and growth effects.
Other than the benefits of ICT, concerns are increasing as regards its impact on income inequality. ICT has a skill-biased characteristic, where it increases demand for high-skilled labourers, reduces the demand for medium-skilled labourers and leaves low-skilled workers unaffected (Autor et al., 2008; Autor and Acemoglu, 2010; Michaels et al., 2014; Banik et al., 2007). Regarding to the demand change, the high-skilled workers get higher income, while the wage of medium-skilled workers are decreasing. Later, the income gap between these two type of workers is expanded. Most of the study in the past looked at the relationship between ICT and income inequality from this market labour perspective.
However, the recent ICT diffusion has changed almost all the aspects of human life. The benefit of ICT influence people’s income more than just in labour market. Several studies present the benefits of ICT by means of several mechanisms at a micro level. This micro level mechanism is relevant from the perspective of households. Lindbeck and Wikstrom (2000), argue that ICT can reduce the asymmetric information between economic agents1. Households may learn better methods as regards producing, pricing and marketing their products via ICT and connect
2
them to the government to match their needs and the services provided, increasing market access, reducing production costs and improving efficiency on information gathering (McNamara, 2003; Mbuyisa and Leonard, 2017). Finally, ICT can reduce transportation costs, provide up-to-date information, increase networking, make it easier to run personal businesses and increase security (Aguero et al., 2014). Thus, regarding micro level analysis, ICT may have a positive effect also.
As the rising concern of negative impact of ICT in labour market, there are also considerable attention paid to the impact of ICT diffusion on income inequality in household level as the digital divide persists (Norris, 2001). Wealthy people, who have more capital to acquire ICT, have greater opportunities to enjoy the benefits of technology quickly, while the poor need more time to benefit from ICT, to earn more capital to acquire ICT or to wait until the ICT’s price is lower and affordable for them (Ziemba, 2016). As the wealthy are able to retrieve more sophisticated ICT tools and benefit more, their welfare will increase over time and the poor will be left behind (Fontenay and Beltran, 2008). Until this point, we can see that ICT diffusion might increase inequality via two different channels: labour market and the lag of ICT adoption on households.
While the existing literature relating to the impact of ICT on income inequality via the labour market is widely available2, there are limited studies pertaining to the impact of ICT on income inequality as the effect of uneven ICT distribution among households. The impact via the household channel is important to investigate further because there are various benefits from ICT that cannot be captured by labour market analysis. While labour market only covers the income inequality from employment income, the adoption of ICT can also affect household income by reducing transaction costs, match public needs and government services, increasing efficiency relating to information gathering, as well as increase networking and security. This study aims to fill the gap in the existing literature. Firstly, we provide evidence of the impact of ICT adoption on income inequality among households in a country with a digital divide. Secondly, we apply the inverted U-shape of the relationship between economic growth and inequality by Kuznets (1955), given that initially, ICT benefits the rich and widens income inequality (Fontenay and Beltran, 2008; Ziemba, 2016). Nonetheless, in the later stage, more people have access to and benefit from ICT. In this phase, it is possible that more households have greater incomes and reduce income inequality in society. By using the panel data regression method, we study the relationship between the ICT adoption rate and inequality test if the inverted U-shaped relationship persists.
To gain a new perspective of the effect of ICT on inequality, we use data obtained from 33 provinces in Indonesia because of its fast growing ICT diffusion, in terms of ICT-based companies and ICT users. In the last decade, many start-up businesses have appeared in Indonesia, such as Tokopedia (a virtual marketplace), Gojek (online transportation service), Traveloka (online travel and hospitality agent), etc. These new types of business provide greater opportunity for households to gain additional income because Tokopedia provides a variety of goods and prices, Gojek offers easy access to transportation, while Traveloka enables house-owners to rent out their properties. UNCTAD (2015), shows that Indonesia, with the fourth biggest population in the world, was expected to have 139 million people accessing the
3
internet and to reach 5.7 million online buyers in 2014. In accordance with this phenomena, to benefit from these online services adopting ICT is essential. As there is uneven distribution concerning the adoption of ICT among households in Indonesia (Hermana and Silfianti, 2011; Sujarwoto and Tampubolon, 2016), only a limited number of people can benefit from this advantage and influence the income inequality situation in Indonesia, besides the other factors that influence income inequality in the country such as poverty and economic structure (Dartanto, 2017), education (Akita et al., 1999), and trade and foreign direct investment (Lee and Wie., 2015). In 2016, the Indonesian province with the lowest Internet adoption ratio was Papua with 19.26%, the highest was Jakarta with 76.96%. The variation in the internet adoption ratio among Indonesian provinces in 2016 is shown in Figure 1.
In the following section, we will review various existing studies that relate to the impact of ICT on growth and inequality and the other factors that influence inequality. In section three, we provide our research questions and hypothesis and then present a brief review of ICT and income inequality conditions in Indonesia in section four. The study methodology will be provided in section five. In section six, we explain the data that are used.
Figure 1. Internet adoption ratio in Indonesian provinces in 2016 (source: bps.go.id)
In section seven, we provide the empirical result of this study. By using random effect estimation, we find strong evidence that ICT adoption by households has an inverted U-shape relationship with income inequality in Indonesia. Moreover, by excluding mobile phone adoption rates from regression, for the reason that we cannot distinguish between the conventional mobile phone and smartphone, ICT adoption exhibits a stronger inverted U-shape relationship with income inequality. By applying the first difference of income inequality
formula with respect to ICT adoption, we note that the turning point for average adoption ratio
as regards mobile phones, personal computers and the internet is at 24.70%, whilst average adoption is observed to be at 17.18% with respect to personal computers and the internet. Finally, the conclusion and limitations of the study will be available in section eight.
4 2. Literature Review
In this section, we will review a number of studies related to the impact of ICT on income inequality. However, it is important to take a look at the role of ICT on economic growth previously because one of the channels of ICT influence economic growth is by means of the productivity effect which plays a vital role in determining income inequality. By recognising that ICT can influence labour productivity, we show how ICT can influence income inequality in the labour market. This channel has been extensively discussed both in developing and developed countries, as theoretical or empirical studies. Moreover, we provide literature reviews regarding the impact of ICT on households, which will reveal the role of ICT on increasing household income. Further, we look at the impact of ICT on income inequality through the household channel, as the uneven distribution of ICT persists in society.
2.1.Impact of ICT on Economic Growth
The onset of new technology can have disruptive effects initially, as envisaged by Schumpeter’s famous creative destruction process (1942). New technologies can make existing technologies obsolete faster and firms are forced to discard their capital that uses the existing technology (Erumban and Timmer, 2012). It was during this initial phase of transition that several advanced economies failed to see improvement in productivity3. However, once the deployment stage was over, and technology was in place, studies have observed the positive impact of ICT on economic growth in three different channels; specifically, production effect, investment effect, and productivity effect.
ICT-production sectors have been characterised by rapid technological progress. The
Organisation for Economic Co-operation and Development (OECD) (2001), provides evidence of a conspicuous increase in Multi-Factor Productivity (MFP)4 in ICT-producing sectors in the US and Finland from 1990-1999. This increasing MFP was then followed by very large economies of scale (OECD, 2001), whereby producing a large number of ICT products, such as semiconductors, made the price of the products lower. Moreover, the capability of advanced technology makes the development of better quality products faster, which in turn influences the price of existing lower quality products to become lower (van Ark et al, 2011). This price drop further creates considerable increase in demand and contributes to the economy.
As the price of ICT is decreasing globally, ICT has become more affordable for more firms and industries. Consequently, they are increasing their investment in ICT to gain an increase in MFP (van Ark et al., 2011). By performing growth accounting studies, Erumban and Das (2016), Timmer and van Ark (2005), Colecchia and Schreyer (2002), and Cette et al. (2002),
3 Solow (1987) was questioning the impact of ICT to growth through his famous quote “computers are everywhere but not in productivity statistics”. This scepticism was supported by Brynjolfsson (1993), Triplett (1999), and Pilat (2005)
4Multifactor productivity (MFP) can be defined as the efficiency of combination between labour and capital inputs in the production process. Changes in MFP reflect the other factors that influence the production process outside the labour and capital inputs. Therefore, MFP growth is calculated as a residual, i.e. the factors that affect the changes of GDP growth that cannot be explained by changes in labour and capital inputs. In simple words, changes in MFP is any changes in output while keeping labour and capital inputs constant. MFP is typically measured using the growth accounting model, 𝑌 = 𝐴𝐾𝛼𝐿𝛽, where Y=output, A=MFP, K=Capital, L=Labour, α and β are
5
show a positive relationship between investment in ICT and economic growth in a country. Moreover, empirical studies by Stanley et al. (2018), Acharya (2015), Spiezia (2012) and Aghaei et al. (2017), also describe the significant positive impact of ICT investment on economic growth.
By using ICT in the production process, labour can perform more efficient work to produce output, therefore, they can increase the amount of output produced. In terms of MFP, ICT provides a better method of combining capital and labour to produce output (van Ark, 2011). This study was supported by Oliner and Sichel (2002) and Jorgenson et al. (2002) in the US, Van Ark (2003) in the US and the EU, Faha and Vaumi (2015) in Cameroon and Wamboye et al (2016) in sub-Saharan countries. So, based on the previous studies above, it is clear that ICT play important role on economic growth and productivity.
2.2.Impact of ICT on households.
Instead of only influencing activity in firms and industries, ICT nowadays influences almost all aspects of human life. Countless studies provide evidence of the benefit experienced by households due to the adoption of ICT. For instance, Lindbeck and Wikstrom (2000), assert that ICT can benefit households by reducing asymmetric information between economic agents. This argument was supported by Bhavnani et al. (2008), who presented a literature study on the benefits of the mobile phone by reducing negative factors (e.g. corruption, crimes, expensiveness) and increasing the positive (e.g. education, efficiency, health) through easier access to information and transparency. Further studies even show the possibility that
household income can be increased by adopting ICT, as shown in Table 1. Table 1. Literature of the Impact of ICT Adoption on Households
Authors Impact of ICT adoption on households
Irvine and Anderson (2006)
Increasing efficiency of the business process in hospitality services in Scotland
Cramer and Krueger (2016)
Greater utilisation of ICT-equipped taxi driver (Uber) than the conventional taxi driver in the US during 2014-2015
McNamara (2003) Learn better methods for producing, pricing and marketing
their products and connects them to the government to match their needs and the services provided
Mbuyisa and Leonard (2017)
Increasing market access, reducing production costs, and rising efficiency on information gathering
Aguero et al (2014) Reduce transportation cost, provide updated information,
increase networking, make it easier to run personal businesses and increase security
Hubler and Hartje (2016) Information exchange, provide a job, weather information, mobile financial transactions, social networks, etc.
Cecchini (2003) Have better access to markets, health, education and
financial services
De Silva et al (2008) Possibility of reducing transaction costs significantly by providing access to ICTs
Aker and Blummenstock (2014)
Information sharing, money transfers, saving money and as devices to use in learning.
Brynjofsson et al (2003) Improving efficiency on book markets
Kim (2018) Positive net consumer welfare of using mobile instant
6
Authors Impact of ICT adoption on households
Bryne and Corado (2017) Consumer surplus in digital content delivery services Source: author
From these studies, we can see that there are numerous advantages to be gained by households adopting ICT and not just an increase in employment income. However, household incomes cover all of the ICT benefits in property income as other income from other non-financial
assets5. By looking at ICT adoption at the household level, we can measure the households that
experience income increases as a result of ICT. 2.3.Impact of ICT on Income Inequality
As previously mentioned, the diffusion of ICT is advantageous for economic growth via increasing productivity, both in industries and in households. However, there is growing concern about the impact of this increasing productivity on income inequality, seeing as ICT does not benefit all types of workers in industries and the presence of the ‘digital divide’ at the household level.
The majority of studies use the labour market channel to measure the impact of ICT on income inequality. The skill-biased characteristics of technology is suggested as the main reason for changes in the wage structure (Bound and Johnson, 1989). Autor et al. (1998), suggest that computer technology plays a crucial role in increasing demand for high-skilled labourers relative to the medium-skilled and low-skilled labourers. Further explanation by Acemoglu and Autor (2010), shows that the introduction of new technology might increase the wages of high skill labourers and reduce the wages of labourers whose jobs can be replaced by new technology. This study is supported by Michaels et al. (2014), who maintain that there was a shift in demand from medium-skilled labour to high-skilled labour in fast-growing ICT industries in the US, Japan and nine European countries from 1980 until 2004.
Some other studies attempt to offer a different perspective on the relationship between ICT and income inequality. Uneven ICT distribution among households means that only a limited number of people can enjoy the benefits of ICT and increase their income. In this situation, there will be an increasing income gap between households with ICT and ones without ICT (Fontenay and Beltran, 2008; Ziemba, 2016). However, as the adoption of ICT has increased, more people benefit and earn higher incomes, which in turn further reduces the income inequality level. Asongu (2015), highlights the positive income equalising effect of mobile phones in 52 African countries from 2002-2009. A further study by James (2016), used the mobile phone as the proxy of ICT to study its impact on income distribution between eleven countries in Africa in 2011. By involving not only the adoption rate but also the intensive use of mobile phones, James (2016), found that mobile phone usage has an income equalising effect between countries. The summary of the channels of ICT influence income inequality is shown in Figure 2.
7
Figure 2. The summary of the channels of ICT influence income inequality
In the light of the interesting previous literature, we are unable to find any study about the impact of ICT adoption on income inequality in Indonesia. The most prominent channel of the relationship between ICT and income inequality, as identified by the previous literature is by way of the supply side/labour market, but the benefit of ICT for households and the presence of the digital divide might also play a vital role in determining income inequality. While the income inequality study via labour market is widely available, the household channel, which can also change the income structure by reducing transaction cost, provide consumer surplus, etc., is less looked at. It is important to do this study because ICT diffusion influence almost all the aspects of human life more than just in labour market. In 2019, Indonesia has the largest internet economy in the world with US$27 billion economic value and 49% compound annual growth rate during 2015-20186. Also, four technology companies in Indonesia turned out to be
billion dollars “unicorn”7 and become part of 300 biggest technology companies in the world.
This diffusion changes not only the wage of labour, but also change the income structure in household level. Given the data limitations about the wage structure in Indonesia, we are not able to investigate the impact of ICT adoption on income inequality in the labour market. So this thesis will examine the impact of ICT adoption in households on income inequality in Indonesia. In addition, the relationship between ICT adoption and income inequality is remains unclear because some studies show the possibility of ICT increasing income inequality (Fontenay and Beltran, 2008; Ziemba, 2016), though other researches demonstrate the equalising effect of ICT on income distribution (Asongu, 2015; James, 2016). Therefore, in this case, there might be an opportunity to implement the inverted U-shaped relationship of economic growth on income inequality introduced by Kuznets (1955).
3. Research Questions and Hypotheses
By observing the household adoption rate of ICT, we can obtain information about how far people in society benefit from it; where a high adoption ratio shows that more people gain an
6 From an article provided in www.forbes.com (published 14 May 2019).
8
advantage from it. By controlling other factors that affect inequality, we can see how the adoption of ICT can increase household income and influence income inequality. This study will contribute in two ways by providing evidence of the impact of the adoption of ICT on income inequality among households in a country with a digital divide and show the possibility of implementing the inverted U-shape relationship between ICT adoption ratio and income inequality.
Thus, in this study, we seek to answer the impact of ICT adoption ratio on income inequality in Indonesia’s provinces. For this, we use the data on Indonesian provinces because of the fast growing technology companies and ICT adoption. As previously mentioned, in the last decade, four ICT-based companies in Indonesia have become companies with valuation more than one billion US dollars and Indonesia has the largest internet economy in the world. After reviewing the existing studies related to the impact of ICT, the following hypotheses is postulated:
The ICT adoption rate has a significant impact on the income inequality in Indonesia and the relationship is an inverted U-shape.
In this research, we use the mobile phone, computer and internet access as the proxies of ICT because these tools are the most common ICT products that are used by households compared to other types of ICT’s proxies (BPS, 2018). Most of the benefits of ICT can be obtained by using a mobile phone as studied by Asongu (2015) and James (2016). Nonetheless, the mobile phone has limited capacity to increase people’s productivity. A computer has better capacity to increase productivity, nevertheless, to access ICT and benefit from it, a computer requires internet access.
4. Some Stylised Facts: ICT and Inequality in Indonesia
4.1.ICT in Indonesia
In the last decade, ICT service industries in Indonesia have been growing rapidly. MIKTI (Indonesia Digital Creative Industry Society)8 released a database of ICT-based companies in Indonesia and shows that there were more than nine hundred ICT companies located in Indonesia at the end of 2018. Moreover, CB Insights describes that there are four ICT-based companies in Indonesia which are valued at more than one billion US dollars; specifically, Gojek, Tokopedia, Traveloka and Bukalapak. Google Play shows the numbers of users that use the application to access the ICT services provided. More than fifty million people have installed Gojek on their mobile phones and more than a hundred million people have installed Grab. The number of installs and types of users for some other applications are shown in the Table 2. These large numbers show that these ICT services industries play a significant role in increasing household income in Indonesia.
Table 2. Number of Installs and Types of ICT Users: Some Services Applications
Applications Type of users Number of installs
Gojek Customer 50,000,000++
Grab Customer 100,000,000++
9
Applications Type of users Number of installs
Uber Customer 100,000,000++
Traveloka Customer 10,000,000++
Tokopedia Customer & Partner 50,000,000++
Bukalapak Customer & Partner 10,000,000++
Grab Driver Partner 10,000,000++
Airbnb Customer 50,000,000++
Uber Driver Partner 50,000,000++
Source: Google Play Store (accessed 01 July 2019)
These ICT services offer various new mechanisms to conventional business processes, such as online transportation and delivery services, besides online retail markets and entertainment. The changes to these business processes were subsequently followed by changes to earning structures, such as in transportation and delivery services; taxi drivers and couriers typically receive sporadic wages, but online transportation and delivery partners earn their income based on the work or services they completed. Conversely, these ICT-based services can also benefit the customer by reducing transaction costs, provide greater variation regarding products and prices, as well as easier access to education, health and security, etc. While the change to the structure of individuals’ earnings can be captured in employment income in the labour market, the other benefit of ICTs can only be measured as household income. The new ICT services industries and their benefits are shown in Table 3.
Table 3. Some Types of ICT Services Industries and Its Benefits in Indonesia
Sectors Type of Services Company Benefits
Local transportation services Online Ride-sharing Gojek, Grab, Uber, etc
Consumers: reducing transaction cost, Partners: increasing income, easier access to customers
Retail market Virtual
marketplace,
e-commerce
Amazon, Alibaba, Tokopedia, Bukalapak, etc
Consumers: variation in prices and products, reducing transaction cost
Partners: wider market access, consumers’ data Travelling services Online travel agent Traveloka, Tiket.com, etc
Consumers: easier and faster access to information, reducing transaction cost Partners: wider market access
Hospitality House-sharing Airbnb, Airy,
Zenroom, etc
Consumers: variation in prices and products, reducing transaction cost
Partners: Property income, wider market access, low marketing cost
Entertainment Video sharing,
Online Gaming
Youtube, Gemscool, Lyto
Consumers: inexpensive services, reduction transaction cost, access to learning
Partners: advertising income Social Networking Social media, video conference, etc. Whatsapp, Facebook,
Customers: cheap networking cost, update information
Providers: advertising income Financial Technology Lending money, banking, payment, mutual funds OVO, Doku, Bareksa, T-Cash
Customers: easier financial access Partners: gain interest income
10
However, individuals can only access all of the ICT services above via the internet. In Indonesia, mobile phones, tablets and personal computers are the most common devices used by people to access the internet (BPS, 2018). Therefore, by looking at mobile phones, tablets, personal computers and internet adoption in households, we can measure the number of households that enjoy all of those benefits and increase their household income. The share of people with access to the internet has increased significantly from about 30% in 2012 to more than 55% in 2017 (www.bps.go.id). While data relating to tablet adoption in Indonesian households is not available, the adoption rate for mobile phones, computers and the internet in Indonesia from 2012-2017 was described in Figure 3. As the digital divide persists in Indonesia (Hermana and Silfianti, 2011; Sujarwoto and Tampubolon, 2016), not all the households can gain an income from the available ICT services. The households that adopt ICT have high opportunity to increase their income from ICT and leave the households without ICT behind. Hence, this situation might influence inequality conditions in Indonesia, beside the other factors that have been investigated previously.
Figure 3. The ICT adoption ratio in Indonesia 2012-2017 (source: bps.go.id)
4.2.Income Inequality in Indonesia
Over the last two decades, income inequality in Indonesia has demonstrated an increasing trend. Gini coefficients grew from approximately 0.32 in 1998 to roughly 0.38 in 2017. However, the Gini coefficient shows a decreasing trend during 2011-2017, which is in line with the growing of ICT adoption in that period. These trend changes might imply a strong relationship between ICT adoption and income inequality in Indonesia. Statistics for Gini coefficients in Indonesia from 1998-2017 is shown in Figure 4.
Even though there are numerous studies relating to income inequality in Indonesia9, studies on income inequality determinants in Indonesian provinces are limited. Certain studies show that economic structure plays a crucial role in income inequality in Indonesia’s provinces (Akita et al., 2011; Dartanto et al., 2017; Afandi et al., 2017). Dartanto et al. (2017), concluded that the
11
agricultural, industrial and service sectors have a significant effect on income inequality in Indonesia. While the agricultural and industrial sectors have a negative impact on income inequality, an increase in service sectors is estimated to have a positive impact on income inequality. The findings are supported by Afandi et al. (2017), that show the negative impact of agriculture on income inequality. However, in their study Afandi et al. (2017), divide the service sectors into several types, namely trade, hotel and restaurant; transportations and communications; financial, real estate and business services and services, while each of the types provides different impacts. Moreover, Dartanto et al. (2017) and Afandi et al. (2017), show the important effect of poverty ratio on income inequality, where the increase in poverty share will reduce income inequality. Other factors considered to influence income inequality in Indonesia are foreign direct investment (Dartanto et al., 2017; Lee and Wie, 2015), education (Dartanto et al., 2017; Afandi et al., 2017) and government expenditure (Dartanto, 2017). An increase in FDI has a positive impact on income inequality due to the skill-bias characteristic of technological changes. While Afandi et al. (2017), demonstrate the positive impact of education on income inequality, Dartanto et al. (2017), fail to provide evidence of the education effect on income inequality. It should be noted that the government through infrastructure expenditure has proven to have a negative impact on income inequality in Indonesia (Dartanto et al., 2017). Certain other limited studies in Indonesia show the importance of GDP per capita in determining the income inequality level with negative effects.
Figure 4. Gini coefficient in Indonesia 1998-2017 (source: World Bank)
5. Methodology
In this section, we will discuss the regression methods that are used to estimate the relationship between the ICT adoption rate and income inequality. The model we use in this thesis postulates a relationship between income inequality and ICT adoption. That is whether ICT adoption ratios have any impact on income inequality. In addition, we also hypothesise that the impact of the ICT adoption ratio on income inequality can be an inverted U-shape. That is, we expect ICT to have a positive significant impact on income inequality in low ICT adoption ratios, but ICT to have a negative significant impact on income inequality in high ICT adoption ratios. To test the significant impact of ICT adoption on income inequality in Indonesian provinces, we follow the model that was used by Dartanto (2017), to measure the impact of
12
structural change on income inequality and add ICT variables into the model. Moreover, to test if the relationship is an inverted U-shape, we implement the approach employed by Haans et al. (2016), by adding the ICT square variable into the model. The positive impact of ICT and the negative impact of ICT square variables on income inequality imply an inverted U-shape relationship between ICT and income inequality.
𝐺𝐼𝑁𝐼𝑖𝑡 = 𝛼𝑖+ 𝛽1𝐼𝐶𝑇𝑖𝑡+ 𝛽2𝐼𝐶𝑇𝑖𝑡2+ 𝛽3𝑅𝐺𝐷𝑃𝑖𝑡+ 𝛽4𝐴𝐺𝑅𝐼𝑖𝑡+ 𝛽5𝑀𝐴𝑁𝑈𝑖𝑡+ 𝛽6𝐹𝐼𝑁𝑖𝑡+ 𝛽7𝐺𝑂𝑉𝑖𝑡+
𝛽8𝐸𝐷𝑈𝑖𝑡+ 𝛽9𝐹𝐷𝐼𝑖𝑡+ 𝛽10𝑃𝑂𝑉𝑖𝑡+ 𝜀𝑖𝑡
where,
GINI = the Gini coefficient;
ICT = the average of adoption rate of mobile phone, computer, and internet
(percentage);
RGDP = real GDP per capita (in million rupiahs);
AGRI = share of agriculture on GDP (percentage);
MANU = share of manufacture on GDP (percentage);
FIN = share of finance on GDP (percentage);
GOV = share of government expenditure on GDP (percentage);
EDU = average years of schooling (years);
FDI = foreign direct investment ratio (in million US dollars);
POV = poverty rate (percentage);
i = province, i=1, …, 33;
t = year t=2012, 2013, …, 2016.
Since our data consist of 33 provinces and five-year periods, we use panel data analysis. To verify that the data is distributed normally, we perform a normality test, which is the skewness-kurtosis test. For the data that are not normally distributed, we transform the data with natural log transformation because some limited studies show a better distribution of data by using natural log transformation. Panel data can be estimated using pooled OLS (POLS), fixed effects (FE) or random effects (RE) estimation methods. The main difference between these methods lies in the assumption of the coefficients of the variable. While POLS assumes that coefficients
β1 β2 β3 do not differ between individuals (𝑦𝑖𝑡 = 𝛽1+ 𝛽2𝑥2𝑖𝑡+ 𝛽3𝑥3𝑖𝑡+ 𝜀𝑖𝑡), FE assumes
different intercepts β1i between individuals but keep constant slope coefficients β2 β3 (𝑦𝑖𝑡 = 𝛽1𝑖+ 𝛽2𝑥2𝑖𝑡+ 𝛽3𝑥3𝑖𝑡+ 𝜀𝑖𝑡). In RE, all individual differences are captured by the intercept coefficient (𝛽1𝑖 = 𝛽̅1+ 𝑢𝑖) but presented as an average and random value so Model l becomes
𝑦𝑖𝑡 = 𝛽̅1+ 𝛽2𝑥2𝑖𝑡+ 𝛽3𝑥3𝑖𝑡+ 𝑣𝑖𝑡, where 𝑣𝑖𝑡 = 𝜀𝑖𝑡+ 𝑢𝑖𝑡. FE is chosen if we are interested in the individual effects of the sample, while RE is preferred if the individual effect is not our interest. Moreover, POLS is preferred when there are no random or fixed individual differences among sample members. However, in our study, we perform a number of statistical tests to make the appropriate choice; specifically:
1. Chow test to choose between POLS and FE, 2. Hausman test to choose between FE and RE,
3. Lagrange Multiplier (LM) test to choose between POLS and RE.
13
as in the previous study. Therefore, the significance of control variables is also considered to choose the model in this study.
6. Data and Variables
The entire data used in this study were collected from the Indonesian Bureau of Statistics (bps.go.id) Dynamic Table10 session. Since the objective of the thesis is to understand the impact of ICT penetration on income inequality, we need to measure income inequality, ICT adoption and several possible control variables. We use the Gini coefficient as a measurement of income inequality, besides multiple indicators, which are, mobile phone, personal computer and internet adoption ratio to capture ICT adoption as the main explanatory variable. The study is conducted across 33 provinces throughout Indonesia11 for the period 2012-2016. In addition, real GDP per capita, share of agriculture, manufacturing, finance, trade and government on GDP, poverty rate, average years of schooling, and foreign direct investment in each province are used as control variables that may have a significant impact on inequality.
6.1.Dependent Variable
Gini Coefficient (GINI) is the most common indicator that is used to measure income
inequality. This coefficient is obtained from the Lorenz curve, which is the sort of people from the lowest income to the highest income in a country or region and shows the population accumulation on the x axis and income accumulation on the y axis. If the value of GINI is one, the population is considered to have perfect inequality, while zero indicates a perfectly equal population. There are various other measurements that are used to estimate inequality, such as the top 1% and income of the top 10%, in addition to interdecile p50/p10, Palma ratio, s80/s20 quintile ratio, etc. We use GINI in this study because of the availability of the data. The Indonesian Bureau of Statistics provides data on the Gini coefficient for urban and rural areas and the combination of both areas. The website (bps.go.id) provides an option for a Gini coefficient for the first and second half of the year and also for the whole year, from the year 2002 until 2018. However, the data is only available for the first half year, while the second half and the whole year’s option result in ‘no data’. For this study, we use the combined data of both rural and urban areas that represent the entire situation of a province in Indonesia. For the period, we use the year 2012-2016 because of the lack of data on some other indicators.
6.2. Independent Variables
Average ICT adoption ratio (ICT): The average value of mobile phone, computer and
internet adoption ratio in a province in Indonesia, where mobile phone adoption ratio shows the percentage of households that own or have the right to use a mobile phone in a province; computer adoption ratio indicates the percentage of households that own or have the right to use a personal computer in a province, whereas internet access ratio offers the ratio between the number of households that accessed the internet at least once in the last three months before the survey was completed and the total number of households in a province. The use of these
10 A session that is available on the Indonesian Bureau of Statistics website (bps.go.id) which allows users to choose which type of data they want to use and in what kind of table the data will be presented.
14
three tools as proxies of ICT is important because they are the most general and affordable ICT devices that can be used by households to benefit from ICT. However, mobile phone adoption ratio does not provide information regarding the type of mobile phone that is owned by households, where smartphone provides a greater benefit than a conventional mobile phone. An increase in the average ICT adoption ratio shows that there are new people who use ICT in their daily lives who can be exposed to the benefit of ICT.
In this study, we expect increasing income inequality on low average ICT adoption ratio because, in the early phase, only wealthy people have access to ICT, while the poor are left behind, creating a large gap between them (Fontenay and Beltran, 2008; Ziemba, 2016). However, as the adoption ratio is increased, more people have access to ICT and enjoy the benefit of ICT, which in turn increases their incomes and finally reduce the income gap in society (Asongu, 2015; James, 2016). Thus, in this study, we expect a positive significant relationship between ICT and GINI, and negative significant relationship between ICTsquare and GINI.
6.3.Control Variables
Economic Structure (AGRI, MANU, FIN, GOV, TRADE): the GDP share of agricultural
(AGRI), manufacturing (MANU), financial service (FIN), government expenditure (GOV) and trade sector (TRADE) on total GDP. It is important to control these indicators for the reason that certain studies show the significant impact of structural economic changes to inequality. An increase in the share of agricultural and manufacturing sectors will reduce income inequality, while an increase in service sectors will increase income inequality (Dartanto et al., 2017), because in Indonesia, agriculture and manufacturing are labour-intensive sectors with low-income workers, while the service sector is a capital-intensive sector with high-income workers. An increasing share in labour-intensive sectors will be followed by an increase in demand in this sector, increase their wages and reduce the income gap to the high-income workers. This situation is also relevant for the increase in share on capital intensive sectors. An increasing share of service sector is followed by an increase in the employee’s income. However, an increase in the income in this high-income sector, will widen the gap to the low-income workers and increase overall low-income inequality. Moreover, an increase in government share implies an increase in public investment. Dartanto et al. (2017), provide evidence that an increase in public investment in infrastructure will lower income inequality in society. Trade plays an essential part in the Indonesian economy via informal trade and food stall activities (Afandi et al., 2017). These non-formal businesses typically involve low-income workers. Therefore an increasing share in this sector will enhance their income and reduce income inequality.
Regional Gross Domestic Product (RGDP): Regional gross domestic products provide the
per capita expenditure in each province during 2012-2016. This variable applies constant price in 2010 (in million rupiahs). In Indonesia, the shift from agriculture to manufacturing or service sectors is the main cause of the increase in real GDP per capita (Dartanto et al., 2017). While a shift from agriculture to manufacturing reduces income inequality, the movement from agriculture to service sector increases income inequality. Consequently, in this situation, the direction is still ambiguous.
Poverty Rate (POV): number of people who live under the poverty line in a province in
15
food poverty line is described as the minimum expenditure that must be spent by an individual to gain 2,100 kilocalorie food per capita per day. Non-food poverty line is the minimum expenditure to gain commodities such as housing, clothing, education and health, based on the Basic Needs Commodity Package Survey 2014 in Indonesia. There are two periods of time available for this data yearly, first and second half-year. For this study, we use the average of the first and second half year. Afandi et al. (2017), show evidence of the positive relationship between poverty rate and income inequality in Indonesia. An increase in poverty rate will increase income inequality. The study explains that the increase in per capita income reduces the number of poor people, but at the same time, the top income earners enjoy a greater increase in their income.
Average Years of Schooling (EDU): the average years of schooling for people aged more than
25 years in a province. Finishing basic school measured as six-years of schooling, finishing primary high school measured as nine-years of schooling, whilst finishing secondary high school measured as twelve-years of schooling, without considering any longer periods of schooling in those schools. Higher education is measured as years attempted in certificates. Using high school participation rate, Dartanto et al. (2017), did not find evidence concerning the role of education to determine income inequality. However, Afandi et al. (2017), show a positive relationship between education and income inequality by applying college participation rate in their study. The high cost of university education limits the participation of poor people in this type of education, which in turn further increases income inequality. By using average years of schooling to measure education, this study follows Dartanto et al. (2017), who assert that education does not play a crucial role in determining income inequality because of the low average years of schooling in Indonesia.
Foreign Direct Investment (FDI): total investment from foreign countries that are coming to
a province in one year. This investment was measured in million US dollars. The previous study offers the positive effect of foreign direct investment on income inequality. Research by Lee and Wie (2015) and Dartanto et al. (2017), reveal the positive impact of FDI on income inequality in Indonesia. However Lee and Wie (2015), treat FDI as a channel for incoming technical changes, which are likely to benefit skilled-workers rather than unskilled-workers, Dartanto et al. (2017), explain that FDI increases the income from capital rather than wages. The summary of the data used in this study is shown in Table 4.
Table 4 Data Summary
Variable Observation Mean Std. Dev. Min Max
16 7. Empirical Result
7.1.Preliminary Tests
Prior to regressing the model, we complete various preliminary tests concerning the data; specifically, normality and correlation tests. By using the skewness kurtosis test, the normality test result shows that the data is not normally distributed (see Table 5). This might happen because of the limited size of the data. To gain better distributed data, we perform natural logarithm transformations for all the data because certain limited studies show the benefit of natural logarithm transformation in removing the outlier data. By implementing the natural logarithm transformation, the model becomes:
𝑙𝑛𝐺𝐼𝑁𝐼𝑖𝑡 = 𝛼𝑖+ 𝛽1𝑙𝑛𝐼𝐶𝑇𝑖𝑡+ 𝛽2𝑙𝑛𝐼𝐶𝑇𝑖𝑡2+ 𝛽3𝑙𝑛𝑅𝐺𝐷𝑃𝑖𝑡+ 𝛽4𝑙𝑛𝐴𝐺𝑅𝐼𝑖𝑡+ 𝛽5𝑙𝑛𝑀𝐴𝑁𝑈𝑖𝑡+ 𝛽6𝑙𝑛𝐹𝐼𝑁𝑖𝑡
+ 𝛽7𝑙𝑛𝐺𝑂𝑉𝑖𝑡+ 𝛽8𝑙𝑛𝐸𝐷𝑈𝑖𝑡+ 𝛽9𝑙𝑛𝐹𝐷𝐼𝑖𝑡+ 𝛽10𝑙𝑛𝑃𝑂𝑉𝑖𝑡+ 𝜀𝑖𝑡
Table 5. Skewness/Kurtosis Tests Results
Variable Obs Pr(Skewness) Pr (Kurtosis) Joint test
Adj chi2 (2) Prob>chi2
GINI 165 0.571 0.028 5.12 0.077 ICT 165 0.822 0.05 3.92 0.14 RGDP 165 0.000 0.000 73.47 0.000 AGRI 165 0.705 0.088 3.10 0.212 MANU 165 0.000 0.744 14.95 0.000 FIN 165 0.000 0.000 . 0.000 GOV 165 0.000 0.000 45.99 0.000 TRADE 165 0.617 0.000 13.02 0.001 EDU 165 0.031 0.258 5.77 0.055 FDI 165 0.000 0.000 . 0.000 POV 165 0.000 0.16 17.58 0.000
Based on the new form of data, we confirm the correlation between all the variables in the model. However, we excluded the square of average log ICT because it must be highly correlated to ICT. We have no problem with that. The correlations between variables are shown in Table 6.
Table 6. Correlation Table of Model
17
As we can see, most of the variables have small correlations. A strong correlation can be seen in the correlation between education and ICT. It tells us that education influences the decision of someone in relation to acquiring ICT products. Moreover, the correlation is positive, where educated people are more likely to require ICT tools. Otherwise, ICT devices have a negative correlation with poverty. It might imply that poor people are less likely to own ICT. Moreover, there is also a high correlation between the share of manufacturing industry on GDP and the share of government expenditure on GDP. With this high correlation, we might fall into a multi-collinearity problem. A few limited studies show that relaxing the variables, which are suspected to cause multi-collinearity, can solve this problem. Therefore, we will test the relationship with and without education and manufacturing share on GDP in the regression. Furthermore, we undertake some regressions following the formulated model. In undertaking the regression, we perform it using a full model and partial model where we relax one ICT proxy to ascertain the impact of one ICT without the existence of other ICT tools. Moreover, we also relax the education and manufacturing share of GDP variables to avoid the presence of the multi-collinearity problem.
7.2.Regression Results
The regression results are shown in Table 7.
Table 7. Regression Result with Mobile Phone, Internet and Computer as ICT
Variable
Full Model Model Without Education Model Without Manufacture share Fixed Effect Random Effect Fixed Effect Random Effect Fixed Effect Random Effect lnICT 1.003* 1.052** .996* 1.056** 1.093** 1.056** lnICT2 -.151* -.164** -.153* -.165** -.166** -.164** lnRGDP .065 -.049 .057 -.053 .069 -.049 lnAGRI .296* -.060** .302* -.061** .335** -.061** lnMANU -.055 .001 -.053 -.000 lnFIN -.008 .114** -.014 .111** .026 .113** lnGOV -.229* -.076** -.234* -.081** -.235* -.077*** lnTRADE -.374** -.179*** -.371** -.183*** -.337* -.178*** lnEDU -.184 -.031 -.117 -.031 lnFDI -.003 -.002 -.003 -.003 -.003 -.002 lnPOV .020 .088*** .026 .088*** .025 .088*** _cons -.1876 -1.987** -2.206* -2.005** -2.534 -1.99** Observations 165 165 165 165 165 165 R-square 0.2858 0.3538 0.2852 0.3466 0.2780 0.3537 VIF 3.38 3.20 3.18 Chow (Prob>F) 0.000 0.000 0.000 Hausman (Prob>chi2) 0.027 0.077 0.024 Lagrange (Prob>chibar2) 0.000 0.000 0.000 legend: * p<0.1; ** p<0.05; *** p<0.01
18
and poverty rate do not have a significant impact on income inequality, while the previous studies (Dartanto, 2017; Afandi, 2017), demonstrate the importance of financial service share and poverty rate on income inequality. There is no rational explanation for this situation and why we finally consider using random effects that provide more rational results. Likewise, the R-square in random effect offers a higher value compared to the one in fixed effect.
VIF value shows that there is no multi-collinearity issue in this model. The partial models, by relaxing education and manufacturing from the estimation, show there is little difference with the full model. From this situation, we can be sure that there is no multi-collinearity issue in this estimation.
By examining the regression result of the full model, when we include all the variables and measure average ICT adoption as the average adoption in mobile phone, computer and internet adoptions, the adoption rate offers a weakly significant positive impact on income inequality, which means an increase in ICT adoption will increase income inequality in Indonesian provinces. A 1% increase in the average percentage of mobile phone, computer and internet adoption will increase the Gini coefficient by (1.052 + 2*(-0.164)*lnICT) percent. Consequently, until here, we know that the level of current ICT adoption determines the
impact of ICT on income inequality, to be positive or negative. Additionally, this shows the
existence of the inverted U-shape relationship between ICT adoption and income inequality. By assuming that the first difference regarding the Gini coefficient is equal to zero, we can calculate the turning point of the ICT adoption as:
𝜕𝑙𝑛𝐺𝐼𝑁𝐼 𝜕𝑙𝑛𝐼𝐶𝑇 = 1.052 + 2(−0.164)𝑙𝑛𝐼𝐶𝑇 = 0 1.052 − 0.328𝑙𝑛𝐼𝐶𝑇 = 0 1.052 = 0.328𝑙𝑛𝐼𝐶𝑇 1.052 0.328 = 𝑙𝑛𝐼𝐶𝑇 = 3.207 𝐼𝐶𝑇 = 𝑒3.207 = 24.704
Thus, if the average ICT adoption ratio for mobile phone, computer and internet is lower than
24.70%, ICT will have a positive impact on income inequality, while a higher average ICT
19
7.3.Regression Result without Mobile Phone as a Proxy of ICT
There is an issue regarding using mobile phone as an ICT proxy in this study. While a smartphone offers high opportunity to access the benefit ICT through some applications, a conventional mobile phone can only access ICT through calling, texting and limited internet access. As the data available do not differentiate between the smartphone and the conventional mobile phone, we perform further regression by relaxing the mobile phone adoption rate as ICT tools. By using the same approach with the previous estimation, the regression results are shown in Table 8 below.
Table 8 Regression Result with Internet and Computer as ICT
Variable
Full Model Model Without Education Model Without Manufacture share Fixed Effect Random Effect Fixed Effect Random Effect Fixed Effect Random Effect lnICT .586** .645** .579** .643** .624** .647** lnICT2 -.103** -.115*** -.104** -.116*** -.111** -.116*** lnRGDP .054 -.055 .047 -.060 .058 -.056 lnAGRI .286* -.064*** .292* -.065*** .324** -.065*** lnMANU -.053 .002 -.052 -.000 lnFIN -.001 .116** -.007 .113** .031 .114** lnGOV -.236* -.075** -.240** -.080** -.241** -.077*** lnTRADE -.362* -.184*** -.360** -.188*** -.326* -.184*** lnEDU -.150 -.034 -.075 -.032 lnFDI -.003 -.002 -.003 -.003 -.003 -.002 lnPOV .021 .091*** .027 .091*** .026 .091*** _cons -1.091 -1.170** -1.375 -1.179** -1.661 -1.161** Observations 165 165 165 165 165 165 R-square 0.2937 0.3634 0.2933 0.3549 0.2865 0.3616 VIF 3.29 3.15 3.11 Chow (Prob>F) 0.000 0.000 0.000 Hausman (Prob>chi2) 0.011 0.043 0.015 Lagrange (Prob>chibar2) 0.000 0.000 0.000 legend: * p<0.1; ** p<0.05; *** p<0.01
After excluding the mobile phone adoption ratio from the average ICT adoption, we observed that the impact of ICT adoption became stronger, but with a lower coefficient. It might explain the importance of distinguishing the adoption of smartphone and conventional mobile phone on measuring the impact of ICT on income inequality. Again, by considering the same factors as before, we use random effect as our preference model. An 1% increase in computer and internet adoption ratio will increase the Gini coefficient by (0.586 + 2*(-0.103)*lnICT) percent. Applying the first difference relating to the Gini coefficient formula, with respect to ICT, equal to zero, we determine the turning point of ICT adoption as:
𝜕𝑙𝑛𝐺𝐼𝑁𝐼
𝜕𝑙𝑛𝐼𝐶𝑇 = 0.586 + 2(−0.103)𝑙𝑛𝐼𝐶𝑇 = 0
20
0.586
0.206 = 𝑙𝑛𝐼𝐶𝑇 = 2.844
𝐼𝐶𝑇 = 𝑒3.207 = 17.184
Therefore, when the average ICT adoption ratio for computer and internet is lower than
17.18%, ICT will have a positive impact on income inequality, while a higher average ICT
adoption will lower the income inequality level.
The insignificant impact of real GDP per capita on income inequality might show the balance part of the shift from agriculture to manufacturing sectors and from agriculture to services sectors. While the shift from agriculture to manufacturing sectors lower income inequality, the shift from agriculture to services sectors increase income inequality. This study rejects the previous study by Dartanto et al. (2017), that exhibits a weakly positive significant effect of economic growth on income inequality. By examining the inverted U-shape relationship between economic growth and income inequality by Kusnetz (1955), we might conclude that Indonesia was in a stable period. Moreover, a higher real GDP per capita might lower the income inequality in the next periods.
Economic structure, which has a critical function on settling income inequality in Indonesia as studied by Novalia (2014), Dartanto et al. (2017), and Afandi et al. (2017), also demonstrates an essential part in this study. An increase in agriculture will lower income inequality in society. As explained in a previous study (Dartanto et al., 2017), agriculture is a labour-intensive sector with low-income workers. An increase in this sector will increase workers’ wages and reduce income inequality. While manufacturing does not display a significant impact on income inequality, the financial sector which is considered a service sector, exhibits
a positive significant effect on income inequality. Similarly, as explained in the study by
Dartanto et al. (2017), the service sector is a capital-intensive sector that hires highly-skilled workers earning high salaries. Consequently, an increasing share in this sector will increase the demand for this type of labour and increase their income, which will finally widen the gap with the low-skilled workers.
The role of the government sector on controlling inequality is clearly seen in this study. An
increase in government shares on GDP will reduce inequality among society. This finding
supports the research by Miroro and Adera (2014), that the government has a significant part to play in providing ICT for the poor in order to reduce poverty. Nevertheless, ICT is not the only tool the government can employ to reduce poverty and increase inequality. Dartanto et al. (2017), show that government infrastructure expenditure has an important role in determining income inequality in society. Another economic structure variable, share of trade on GDP, also participates in specifying the inequality level. The increasing portion of trade on GDP will
reduce income inequality within society. This might relate to the high number of small and
medium enterprises (SMEs) in Indonesia, hence an increasing portion of trade means that more poor people engage in trade that may possibly increase their quality of life and reduce income inequality. This finding is supported by Afandi et al. (2017) and explains the important role of informal trade on the Indonesian economy.
21
of income inequality undertaken in Indonesia by Dartanto et al (2017). The low average years of education in Indonesia might cause this unimportant role of human capital in influencing inequality. As described in the data description, average years of schooling in Indonesia is less than eight years, which means that more than half of the young people in Indonesia are not secondary high school graduates.
Incoming investment into Indonesian provinces also does not have a positive impact on income inequality in Indonesia provinces. This fact rejects the study by Lee and Wie (2015) that FDI influences inequality through the transfer of technology that produces skill-biased technological change. Moreover, the role of FDI on increasing income from capital as shown by Dartanto et al. (2017) is also not proven in this study. The low level of transfer of knowledge can be considered the main aspect in the slight impact of FDI on income inequality.
7.4.Issue of Endogeneity
There are some issues in this study that might influence the regression results; specifically, omitted variable and reverse causality. The omitted variable issue is related to the lack of data regarding wage structure in Indonesia, which means we are unable to involve the labour market channel on influencing income inequality in Indonesia. Moreover, a report by UNCTAD (2005), confirms that income inequality between countries might influence the existence of the digital divide. Even though a study by Davis (1989) explains that usefulness and ease of use factors are the main reason for ICT adoption, the impact of income inequality on ICT adoption still needs to be considered. To verify if there is an endogeneity problem in this study, we perform an endogeneity test by adding the ICT adoption level in t-1 as a determinant of ICT adoption level at time t.
We run several regressions to perform the endogeneity test. First we run the ICT as a function of ICT in t-1, then we predict the residual of this estimation (namely ehat). Finally, we run the GINI with the full model except the square of ICT. From the last estimation, we check if the error term has a significant impact on income inequality in the first run. The significant p-value of error shows that there are endogeneity problems in this study that might cause a biased result. The estimation that we performed is shown as below:
𝑙𝑛𝐼𝐶𝑇𝑖𝑡 = 𝛽0+ 𝛽1𝑙𝑛𝐼𝐶𝑇𝑖𝑡−1+𝛽2𝑙𝑛𝑅𝐺𝐷𝑃𝑖𝑡+ 𝛽3𝑙𝑛𝐴𝐺𝑅𝐼𝑖𝑡+ 𝛽4𝑙𝑛𝑀𝐴𝑁𝑈𝑖𝑡+ 𝛽5𝑙𝑛𝐹𝐼𝑁𝑖𝑡
+ 𝛽6𝑙𝑛𝐺𝑂𝑉𝑖𝑡+ 𝛽7𝑙𝑛𝐸𝐷𝑈𝑖𝑡+ 𝛽8𝑙𝑛𝐹𝐷𝐼𝑖𝑡+ 𝛽9𝑙𝑛𝑃𝑂𝑉𝑖𝑡+ 𝜀1
𝑙𝑛𝐺𝐼𝑁𝐼𝑖𝑡 = 𝛼𝑖+ 𝛽1𝑙𝑛𝐼𝐶𝑇𝑖𝑡+ 𝛽2𝑙𝑛𝑅𝐺𝐷𝑃𝑖𝑡+ 𝛽3𝑙𝑛𝐴𝐺𝑅𝐼𝑖𝑡+ 𝛽4𝑙𝑛𝑀𝐴𝑁𝑈𝑖𝑡+ 𝛽5𝑙𝑛𝐹𝐼𝑁𝑖𝑡 + 𝛽6𝑙𝑛𝐺𝑂𝑉𝑖𝑡
+ 𝛽7𝑙𝑛𝐸𝐷𝑈𝑖𝑡+ 𝛽8𝑙𝑛𝐹𝐷𝐼𝑖𝑡+ 𝛽9𝑙𝑛𝑃𝑂𝑉𝑖𝑡+ 𝛽10𝑒ℎ𝑎𝑡 + 𝜀
From the formula above, we obtain the following regression result:
Source SS df MS Number of obs 132
F (10,121) 529.34
Model 8.108 10 .8108 Prob>F 0.0000
Residual .1853 121 .0015 R-squared 0.9777
Adj R-squared 0.9758
22
lnICT Coef. Std.Err. t P>|t| [95% Conf.Interval]
ICT-1 .0958 .028 33.48 0.000 .901 1.01 lnRGDP -.006 .015 -0.39 0.698 -.037 .024 lnAGRI .003 .007 .045 0.653 -.010 .017 lnMANU -.004 .007 -.064 0.523 -.019 .009 lnFIN -.009 .015 -.058 0.561 -.040 .021 lnGOVE .008 .009 0.90 0.372 -.010 .027 lnTRADE .003 .015 0.24 0.814 -.027 .034 lnEDU .0009 .048 0.02 0.984 -.094 .096 lnFDI .0003 .002 0.17 0.869 -.004 .004 lnPOV -.011 .009 -1.20 0.234 -.029 .007 _cons .252 .125 2.01 0.047 .003 .500
After predicting the residual value (ehat), we obtain the following regression result:
Source SS df MS Number of obs 132
F (10,121) 10.24
Model .695 11 .063 Prob>F 0.0000
Residual .741 120 .0061 R-squared 0.4841
Adj R-squared 0.4368
Total 1.437 Root MSE .07861
lnGINI Coef. Std.Err. t P>|t| [95% Conf.Interval]
lnICT .106 .060 1.78 0.078 -.012 .225 lnRGDP .001 .031 0.06 0.953 -.060 .064 lnAGRI -.025 .014 -1.81 0.073 -.054 .0024 lnMANU .005 .014 0.39 0.695 -.023 .034 lnFIN .119 .031 3.81 0.000 .057 .1817 lnGOVE -.044 .018 -2.34 0.021 -.081 -.006 lnTRADE -.070 .031 -2.22 0.028 -.132 -.007 lnEDU -.249 .097 -2.57 0.011 -.442 -.057 lnFDI .005 .004 1.13 0.263 -.003 .014 lnPOV .130 .018 6.92 0.000 .093 .167 ehat -.111 .192 -.058 0.561 -.492 .268 _cons -1.02 .257 -3.98 0.000 -1.532 -.514
By investigating the insignificant p-value of ehat in the last regression, we can conclude that there is no endogeneity problem in this study.
8. Conclusion and Limitations of the Study
23
ICT will increase the number of people who can appreciate the benefits of ICT and increase their income. This study contributes more literature in relation to the determinants of income inequality in Indonesia and also supports the possibility of an inverted U-shape relationship between ICT adoption and income inequality. By identifying the turning point of the inverted U-shape relationship, we can obtain a minimum ICT adoption ratio that must be achieved to reduce income inequality in society, otherwise, ICT will increase income inequality.
There are three significant points that can be taken from this empirical study of 33 provinces in Indonesia from 2012-2016. First, the importance of the ICT adoption rate in determining the level of income inequality in Indonesia. From the regression result, we can see that the impact of ICT adoption on income inequality is dependent on the current ICT adoption level. Second, it is important for the government to achieve approximately 25% of the average adoption rate of mobile phones, computers and the internet or roughly 17% of the average adoption rate of computers and the internet to generate a positive impact of ICT on inequality. Third, the government can support the agriculture and trade sectors to reduce income inequality in society.
As a consequence of the limited time and resources, there are a few limitations as regards this study, which are the lack of a labour channel, limited sample and simple econometric application. Moreover, limited literature pertaining to the relationship between ICT and household income is also a weakness concerning this study. There are some spaces for improvement in future studies in relation to the relationship between ICT adoption rate and inequality, which are:
1. Investigate the impact of ICT on income inequality in Indonesia via labour market channel;
2. Increase the sample size and time period for studies, because in this study, very limited data are available to process;
3. Differentiate the adoption rate of the conventional mobile phone and smartphone because the smartphone has much greater ICT benefits than the conventional;
