Chinese Foreign Direct Investment in Sub-Saharan Africa: A “win-win” situation?

(1)

Bachelorscriptie Max Garcia Hoogland

S1801805

Chinese Foreign Direct Investment in Sub-Saharan Africa: a “win-win” situation? Word count: 7827

Bachelorproject 12: Practicing Democracy in Contemporary Africa Docent: Leila Demarest

Abstract

China’s role on the world stage has become ever more prominent. Since the beginning of this century, China’s focus has shifted to countries abroad. By making more Foreign Direct Investments (FDIs), China has tried to satisfy its growing hunger for more energy and natural resources. It is therefore not surprising that Chinese extractive multinational corporations (MNCs) settle in countries with an abundance of natural resources. However, Chinese FDIs into Africa have also caused concern. It has been argued that China’s extractive behaviour looks a lot like neo-colonialism, as it seizes a new sphere of influence, and it extracts oil and other natural resources. Furthermore, Chinese investments have been argued to undermine democratic institutions and foster corruption. This paper will focus on Chinese FDI, and more specifically, how these FDIs influence the level of corruption in Sub-Saharan African countries. To test if a higher amount of Chinese FDIs increase the level of corruption in Sub-Saharan African countries, I make use of regression analysis. Results suggest that FDIs decrease the level of corruption in Sub-Saharan African countries. This applies to FDIs in general as well as Chinese FDIs specifically. The paper hence does not find evidence for the view that China’s presence in Africa fosters corruption.

(2)

2

1

Introduction

In the last couple of decades, China has become one of the biggest political and economic actors on the world stage. This is demonstrated by an increase of Chinese outward Foreign Direct Investments (FDIs) and the fact that more and more Chinese multinational

corporations (MNCs) set up shop abroad. These FDIs are flows of investment capital across national borders, in which the investor (usually a multinational corporation) retains a

controlling stake over an affiliate established in a different country.

Ever since the start of the 21st century, China has become more ‘global’. This may be best illustrated by China’s ‘going out’ policy that was introduced at the beginning of this century. One of the reasons why China is going global has to do with the fact that if China wants to keep growing economically, measures have to be taken. The country will need to satisfy its growing hunger for more energy and natural resources, and it is therefore not surprising that Chinese extractive MNCs settle in countries with an abundance of natural resources. Africa provides for a large share of the world’s natural resources, and it is obvious that Chinese extractive MNCs choose to open affiliates on this continent.

This is not without consequences however. The Chinese move into Africa has not gone unnoticed and has received a substantial amount of criticism. It is argued that China’s extractive behaviour looks a lot like neo-colonialism, as it seizes a new sphere of influence, and it extracts oil and other natural resources (Jiang, 2009). Furthermore, the Chinese influence in Africa, in the form of Chinese FDI, is argued to have a link with corruption (Pinto & Zhu, 2016).

This link between FDIs and corruption will be the main focus of this work. I specifically investigate whether a higher amount of Chinese FDIs is associated with a higher level of corruption in Sub-Saharan African countries. A large body of literature has focused on the question whether corruption influences FDI flows, but the reverse relation has so far not received sufficient attention.

I focus on Sub-Saharan Africa specifically since most natural resource abundant countries are located in this region. Another reason for focussing on this region specifically is the fact that Chinese FDI flows are targeted at countries lying in this region in particular.

I will use OLS regression to estimate the model, however, I will include standard errors clustered by country to account for non-independence of the observations. In my analysis I find a significantly positive effect between the level of Chinese FDI inflows as percentage of

(3)

3

a country’s GDP and the level of corruption in that country. This goes against the hypothesis and indicates that Chinese FDI fosters a lower level of corruption perception.

In the following section I first sketch the academic and policy debate on the rise of China in Africa. Then I discuss the literature on the links between FDIs and corruption. Next, I elaborate on the data used and the variables selected to conduct this research. This followed by the presentation and discussion of the analysis results. The paper ends with a brief discussion and a conclusion of the research done.

2 The rise of China

In the final decade of the 20th century more attention has been paid to the Chinese influence in Africa. Scholars have also increasingly focused on Chinese resource extraction on the African continent (Kaplinsky, McCormick & Morris, 2007; Chen, Goldstein, Pinaud & Reisen, 2005; Broadman, 2006). This is due to the fact that the Chinese move into Africa is strongly motivated by economic interests, namely resource extraction (Large, 2008, p. 55). To keep growing economically like they have done the past three decades, China will need to satisfy its growing hunger for more and more energy and natural resources. In their view, African resources will sustain the resource consumer trend that is going on in China’s modernization drive (Jiang, 2009, p. 587).

China’s focus on Africa may be best symbolised by its ‘going out’ policy adopted at the end of the 20th century. This new strategy of the Chinese government wanted Chinese companies to go around the world to explore and extract additional energy and resources (Jiang, 2009, p. 588). Not only is China expanding its development assistance to Africa with massive

infrastructure projects, debt forgiveness, new loans, and increased resources for public health, education, and training, but it also stimulates state-owned enterprises (SOEs) to increase their investment and trade with Africa (Gill & Reilly, 2007, pp. 37-38). It is important to look at the going out strategy more in depth to sketch current Chinese presence in Africa.

2.1 China’s ‘going out’ strategy

China’s ‘going out’ strategy has as main focus to increase Foreign Direct investments (FDIs), to stimulate product diversification, to establish and improve risk assessment and prevention mechanisms for overseas investment and transnational operations, and to strengthen the construction of foreign investment promotion and service systems (General Office of the State Council, 2006). Especially Chinese FDIs have increased a lot due to the ‘going out’

(4)

4

strategy. FDIs are flows of investment capital across national borders, in which the investor (usually a multinational corporation) retains a controlling stake over an affiliate established in a different country (Caves, 1996). Where China’s total outward direct investment in 2003 was only $33 billion, this grew to a staggering $230 billion in 2009 (Cheung, de Haan, Qiang & Yu, 2012, p. 201). The most notable increase is arguably Africa’s share in the total Chinese outward investments. In recent years, Africa has become the third largest recipient of foreign direct investments coming from China, after Europe and Asia itself (Besada, Wang &

Whalley, 2011).

In the chart below, the Chinese FDI in million USD is illustrated. This data is derived from The China-Africa Research Initiative. In the period 2003-2017 China’s total FDI inflows to Sub-Saharan Africa accounted for a grand total of 29414.16 million USD.

Figure 1: Chinese FDI inflows to Sub-Saharan Africa between 2003-2017 in mln $

Note. Reprinted from China Statistical Yearbook: "Overseas Direct Investment by Countries or Regions", by UNCTAD Bilateral FDI Statistics, retrieved from http://unctad.org

2.2 Is Chinese FDI a “win-win” situation?

FDI is often seen as beneficial to economic growth, especially when FDI is aimed at

developing countries. That FDI has a positive effect on the economic growth of a country is an idea that is agreed upon by several scholars. Ajayi (2006) argues that FDIs augment low

0 1000 2000 3000 4000 5000 6000 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

(5)

5

domestic savings in the process of capital accumulation. The author argues that this

stimulates domestic investments; this then leads to an overall increase in the total investments in the country (Ajayi, 2006, p. 12). According to Carkovic and Levine (2002), FDIs provide a transfer of technical knowledge and make sure there is a spill-over effect (technological knowledge gets transferred to African countries). Foreign investors can also help to reduce ‘idea gaps’ and ‘object gaps’ between developed and underdeveloped countries (Adams, 2009, p. 180). This means that foreign investors can help countries that ‘are lacking physical objects like factories and roads’ (object gaps), or are lacking ‘the knowledge used to create value in a modern economy’ (idea gaps). Literature suggests that these phenomena foster economic growth (Carkovic and Levine, 2002; Greenaway, 1998; Romer, 1993).

Even though the Chinese move into Africa is argued to be a “win-win” situation for both sides, this statement has received a lot of criticism (Alves, 2013). There are scholars who argue that China’s extractive behaviour looks a lot like neo-colonialism, as it seizes a new sphere of influence, and it extracts oil and other natural resources. Scholars even go so far to argue that China props up repressive regimes (Jiang, 2009, p. 585).

Besides scholars, prominent figures in world politics have also openly addressed concerns regarding China’s growing presence and influence in Africa.

John R. Bolton, President Trump’s national security adviser, addressed the issue of the increasing Chinese influence in Africa (Landler & Wong, 2018), stating that ‘China uses bribes, opaque agreements and the strategic use of debt to hold states in Africa captive to Beijing’s wishes and demands’.

Former U.S. president Obama also criticized China’s actions in Africa, even though he did not mention China in particular. At a 2014 summit in Washington he stated that America does not simply extracts minerals for own growth. He sees that ‘partnerships create jobs and opportunities for all our peoples, that unleash the next era of African growth’.

In 2011, back when she was Secretary of State, Hillary Clinton also raised concerns about the increase of Chinese investments into Africa (Quinn, 2011). She stated that America does not want to see new colonialism in Africa. She argued that she did not want China to undermine good governance in Africa, arguing that it was easy for China to come in, take out natural resources, pay off leaders and leave.

In an article written by the RAND Corporation (Morris & Hanauer, 2013), this problem is further highlighted. It is argued that besides poor labor conditions, unsustainable

(6)

6

that take advantage of African governments' relative weaknesses and that foster corruption and wasteful decisionmaking [sic]’.

3 Foreign Direct Investment and corruption

The Africa-China relation gained an unprecedented amount of attention thanks to China’s ‘year in Africa’ in 2006 (Kaplinsky & Morris, 2009, p. 551) - which led to Saharan Africa economically growing rapidly in recent years (Large, 2008, p. 45) - but has not remained without critics. In this paper I focus specifically on the argued relationship between Chinese FDI inflows and corruption in a country. Before doing so, I first review the literature on the relations between FDI inflows, development, and corruption in a country (Wei, 1997, 2000a).

While FDIs have in general been associated with economic growth (Carkovic and Levine, 2002; Greenaway, 1998; Romer, 1993), several scholars also place constraints when it comes to a possible effect of FDI on economic growth. Trevino and Upadhyaya (2003), for instance, argue that FDI only stimulates economic growth when the country has a more open economy. Other scholars state that FDI only works in countries with sufficiently developed financial systems (Alfaro et al, 2004).

A recent study done by Pinto and Zhu (2016) suggests that the link between foreign investment and good governance depends on the effect of FDI on market dynamics of host countries. The authors argue that political and economic conditions create the incentives that “shape opportunities for corrupt behaviour” (Pinto & Zhu, 2016, p. 694). It has also been said that the effects of FDI on corruption are dependent on the level of democratization and development of a country (Pinto & Zhu, 2016). The effects of FDI on corruption are

conditional on host countries’ political and economic environments. Corruption will increase due to FDIs in less developed non-democratic countries. In advanced democracies, however, FDIs decrease the level of corruption (Pinto & Zhu, 2016, p. 694). Other work (Ades & Di Tella, 1999) suggests that incentives to engage in corruption increase when rents associated with the exploitation of natural resources become available.

Establishing an affiliate abroad mainly derives from the existence of assets such as brand names, managerial skills and production technology (Pinto & Zhu, 2016, p. 695). The presence of these multinational corporations (MNCs) can change market dynamics, which in turn shape opportunities for corrupt behaviour. The MNCs coming from more modernised

(7)

7

countries have a productivity advantage over the local businesses. This may also be the case in Africa where Chinese FDIs establish Chinese MNCs. These Chinese MNCs can crowd out African businesses, because of the technological advantage these Chinese MNCs have over the African ones. Lesser productive companies will thus be pushed out of the market, mainly because these domestic companies see a decline in their profits (Aitken and Harrison, 1999). Simultaneously, rents will also rise. This is mainly because the most productive companies will increase their mark-ups over marginal costs (Pinto & Zhu, 2016, p. 695).

Increasing rents tend to foster corruption since these rents will be shared between investors who help create them and government officials who are in a position to regulate investors‘ presence and activity, grant or deny licenses and permits, and uphold restrictive market conditions (Pinto & Zhu, 2016, p. 695).

Another good example of FDIs changing market dynamics is the case of Sinopec (a Chinese oil company) in Angola in 2004. In this year the Export–Import Bank of China, Chexim, provided Angola with a soft loan. This loan was used by Angola to develop things such as infrastructure projects and to construct new government offices. The timing of the provision of this loan could not have been better: “When France-based Total applied to renew its license on an offshore oil block, Angola refused, handing it instead to China’s Sinopec” (Chan-Fishel & Lawson, 2007, p. 65). The fact that the Angolan government provided a Chinese oil company with a license was reprimanded, especially since the Chinese

government had provided Angola with a soft loan just before the license was provided. This was a clear example of how government officials favoured a party due to the inward

investments they received from the country in question.

Following Pinto and Zhu (2016), I expect inward FDI to foster higher levels of corruption in Sub-Saharan Africa, since government officials will engage in corruption when they see good opportunities for rent extraction. I can derive the following hypothesis:

Hypothesis: A higher amount of Chinese FDIs will increase the level of corruption in Sub-Saharan African countries.

Previous work has been done on the relation between FDI and corruption, but this study differs from previous works in several ways. Larraín and Tavares (2004) have looked at the

(8)

8

impact of openness to FDI on corruption. In this research the openness of a country towards FDIs does not play a role, and the research question thus fundamentally differs from theirs. Sandholtz and Gray (2003), research over 150 country cases and argue that greater degrees of international integration lead to lower levels of corruption. International integration is

measured by using multiple variables including GFDI (gross foreign direct investment). Their work also differs from mine, since they mainly look at (political) integration, using variables such as ‘being a member of the UN as a country’. They find that more international

integration, political as well as economically, leads to less corruption.

In this work attention will mainly be paid to Chinese FDI inflows specifically and their effect on corruption.

The research done by Pinto and Zhu (2016) comes closest to this work. Although they test the same relation as I do, my research is focussed on Chinese FDI and this research only targets Sub-Saharan Africa when it comes to the receiving countries of foreign FDIs.

4 Data

I will test if there is a significant relation between the Chinese FDI and the level of corruption in Saharan Africa by using a cross-national research design. When speaking of Sub-Saharan Africa I will select all African countries except for Algeria, Djibouti, Egypt, Libya, Morocco, Somalia, Sudan and Tunisia. I use this classification created by the UNDP in Africa, which states that Sub-Saharan Africa consists of a total of 46 countries.

Our dependent variable is the level of corruption in Sub-Saharan Africa. To get the data needed to measure the level of corruption in Sub-Saharan Africa I use data from the Corruption Perception Index (CPI) provided by Transparency International for the period 2012-2017. The CPI is constructed by using data from 13 different surveys and assessments produced by the 10 independent organisations (Transparency International, 2010).

Using the CPI, countries get assigned a rating ranging from 0 to 100, where 0 means a country is highly corrupt and 100 means a country is free of corruption.

In 2012, Transparency international has revised their methodology used to construct the index of the CPI scores. From that year forward, it is possible to compare scores from one year to the next. Hence the time period chosen from this study in 2012-2017, which is the maximum period possible when using the revised CPI scores.

(9)

9

Our independent variable is the Chinese FDI into Sub-Saharan Africa. To measure the FDI into Sub-Saharan Africa I look at the FDI as a percentage of the GDP of each African country in Sub-Saharan Africa. The data used here is derived fromThe China-Africa Research

Initiative1. In this dataset provided by The China-Africa Research Initiative, the amount of FDIs China has given African countries per year in the years 2011-2016 is illustrated. The only thing we need to add to this data is the annual GDP of these Sub-Saharan Countries, because we can then see what the FDIs are as a percentage of the annual GDP of these

countries. We thus create the independent variable by dividing the Chinese FDI in unadjusted U.S. Dollars every Sub-Saharan country gets by the total GDP per Sub-Saharan African country in unadjusted U.S. Dollars. Our dependent variable is measured from 2012-2017, all the other values are measured from 2011-2016. The data is structured this way, since we can only try to determine the direction of the effect when all the other values except for the dependent variable are from a previous year (t-1).

I will also use several control variables in this research. These control variables will be essential, since they give us a better understanding of the relationship between the main variables of interest. In this research several control variables are used: Natural resources as a percentage of GDP, the total incoming FDIs per country as a percentage of GDP, income distribution (Gini-coefficient), Democracy score, the GDP per capita measured in Purchasing Power Parity (PPP) per Sub-Saharan country, the interaction effect between Chinese FDI and natural resources, and finally, the interaction effect between Chinese FDI and Democracy.

The first control variable is revenues gained from the export of natural resources as a percentage of GDP. Countries that are heavily relying on natural resources as a source of national income tend to be more vulnerable to corruption (Leite & Weidmann, 1999). Melo and Quinn (2015) also see natural resources leading to vulnerabilities and see MNCs play an important role. The authors give the example of a MNC such as Royal Dutch Shell (operating in Nigeria). In the case of these companies which have a specific interest in natural resources, resource rents and the importance of property rights have led to commodity sectors often being the target of corruption scandals (Melo & Quinn, 2015, p. 33). It is argued that especially Chinese mining companies and other SOEs based in Africa focused on natural resources are not as transparent about their operations as they could be. This is particularly

(10)

10

the case in African environments where corruption and instability can be endemic (Chintu & Williamson, 2013). We thus argue that higher revenues gained from the export of natural resources as a percentage of the GDP increase the likeliness of corruption occurring in these Sub-Saharan countries.

The second control variable is the total FDI Sub-Saharan countries acquire as a percentage of their GDP measured in U.S. dollars. If FDIs are a more than substantial amount of the total GDP a Sub-Saharan country has, the dependency on foreign investments in these countries is likely to be higher. As we argued before, following Pinto and Zhu (2016), the presence of these multinational corporations (MNCs) can change market dynamics, which in turns shapes opportunities for corrupt behaviour. We thus argue that a higher amount of FDI as a

percentage of a country’s GDP will be more likely to foster corrupt behaviour. Perhaps not just Chinese FDIs but FDIs as such, regardless of origin.

The third control variable is the Gini coefficient. The Gini coefficient is an important control variable since it gives an insight in how income is distributed within a country. GDP per capita might be a good indicator for measuring the average income of an individual in a country; it doesn’t however say anything about the distribution of wealth within these

countries. Literature has suggested that there is a positive correlation between corruption and the level of income inequality (Gupta, Davoodi & Alonso-Terme, 2002). Therefore, I will use data acquired from the Gini index to check how evenly the wealth is distributed in these countries. In this research the Gini coefficient ranges from 0 to 100. If a country scores low on the Gini coefficient this means that there is a low level of inequality, and if a country has a high Gini coefficient this means that there is a high level of inequality. A higher rate of inequality is the result of uneven income distribution (the elite will keep more of the wealth for themselves).

The fourth control variable is the variable Democracy. Scholars argue that the level of rights that are guaranteed by the political system can be associated with corruption. Democratic political systems tend to have a high level of political and civil rights. In these societies some characteristics such as free press and open elections can increase the likelihood of exposing corruption (Larraín & Tavares, 2004, p. 221). I will therefore expect a lower level of corruption when the level of democracy in an African country is high. The data needed for this control variable is provided by Freedom House. The Sub-Saharan countries get assigned

(11)

11

a score on Political Rights as well as Civil Liberties. These scores are both ranked from 1 to 7, where 1 means most rights/liberties and 7 means least rights/liberties. Countries with a higher level of political rights will thus have a lower score than countries that have low political rights. To create the control variable we take the sum of the score each country gets on Political Rights and Civil Liberties for each year, and then divide this score by 2 to get the mean score of the two values. If the mean score of these two values is low, the country in question has a high level of democracy. This new variable will thus be called Democracy, since this variable is a combination of the civil liberties and the political rights in a country.

The fifth control variable is the GDP per capita measured in Purchasing Power Parity (PPP) of a country. Even if we are using the Gini coefficient as a control variable, the GDP per capita of a country is needed in the light of the institutions in Sub-Saharan African countries. Neeman et al. (2008) argue that a high per capita income is closely associated with the existence of efficient and transparent institutions. I expect that a higher GDP per capita is closely related with ‘good’ institutions which in turn lead to a lower level of corruption. I use the GDP per capita PPP measure because this takes into account how much citizens in each country can actually buy with their income as calculated in US dollars. This is a good way to make the GDP per capita measure more comparable across countries. However, the GDP per capita PPP measure needs to be corrected for skewness. In general, GDP per capita tends to have a right skew, meaning the mass of cases are bunched at lower values (Benoit, 2011, p. 2). I will therefore use the natural logarithm of the GDP per capita PPP, which will turn out to be normally distributed.

The last two control variables used are interaction effects. The interaction effects are

beneficial to my research since they can explain how independent variables work together to impact my dependent variable. Besides helping to explain more of the variability in the dependent variable, it also provides me with a better understanding of the relationship between the dependent and independent variables (Lavrakas, 2011).

The first interaction effect (Chinese FDI x Democracy) will measure what happens to the dependent variable if both Chinese FDI and Democracy increase in value. In this case, we would like to know whether Chinese FDIs lead to corruption especially in non-democracies, since an increase on Democracy means a lower level of civil liberties and political freedoms. The second interaction effect (Chinese FDI x Natural resources) will measure what happens to the dependent variable if both Chinese FDI and the revenue deriving from natural

(12)

12

know whether Chinese FDIs lead to corruption especially in countries with an abundance of natural resources.

Most of my data is derived from The World Bank DataBank (Natural resource incomes as a percentage of GDP, total FDIs as a percentage of the GDP, the Gini Coefficient and the GDP per Capita). As mentioned earlier, the CPI data is obtained from Transparency International, the Chinese FDI inflows are obtained from The China-Africa Research Initiative and the data concerning political rights is derived from Freedom House.

Several missing values occur. These and the descriptive statistics for each variable are shown in Table 1.

Table 1: Frequency table of OLS regression

N(Valid) N(Missing) Mean Standard error Minimum Maximum CPI 270 6 33.36 0.709 11 65 Chinese FDI as % of GDP 275 1 0. 395 0.059 -2.822 10.702 NR as Percentage of GDP (2011-2016) 270 6 14.153 0.742 0.001 60.121 Foreign direct

investment, net inflows (% of GDP)

275 1 5.913 0.648 5.386 86.989

GINI index (World Bank estimate) 269 7 44.095 0.499 30.80 65.00 Democracy 275 1 4.249 0.099 1.00 7.00 Purchasing Power Parity 270 6 8.002 0.058 6.42 10.50 ChineseFDIxNR 270 6 6.735 1.201 -38.66 209.50 ChineseFDIxDem 275 1 1.749 0.331 -9.88 74.91

There are missing values for a total of 35 cases.

For the dependent variable missing values are found for 6 cases and for the independent variable missing values are found for 1 case only. For the FDI inflows as a percentage of the

(13)

13

GDP, the interaction effect between the Chinese FDI and Democracy and the political rights (Democracy) only 1 case each is missing. The natural resources as a percentage of the GDP and the interaction between the Chinese FDI and the natural resources both have 6 cases missing. Finally, the Gini coefficient has 7 missing cases and the GDP per capita has 6 missing cases.

In the case of the Gini coefficient the amount of missing cases was previously much higher. To correct for the overabundance of missing cases the Gini coefficient per country is the same for all of the 6 years. Eritrea has no Gini coefficient because it has never been

published. For South Sudan in 2012 the Gini coefficient is missing as well, since this country did not officially exist in 2012 – hence there is no data at all for South Sudan in 2012.

Data was not available for Eritrea when it comes to the GDP per capita (PPP). There were 5 incidents where cases had a value of 0 – these are to be considered a being missing cases. However, this data was obtained via the CIA World Factbook (2015). This data is an estimate, but an improvement to the missing cases because GDP per capita is an important variable in the model

5 Results

I now turn to the model. As stated before, the dependent variable is the CPI per Sub-Saharan African country in the period 2012-2017. The independent variable is the Chinese FDI as a percentage of the GDP of the receiving country per year. As discussed above, I will make use of 7 control variables. I use OLS regression to estimate the model, however, I include

standard errors clustered by country to account for non-independence of the observations. To do this I use SPSS’ (version 23) the Complex Samples General Linear Model (CSGLM) procedure. Assumption violations were checked and do not affect results (see Appendix). Regression results are reported in Table 2.

Table 2: Linear regression model perception of corruption according to respondents in Sub-Saharan Africa in the period 2012-2017

Model 1 Model 2 Model 3 Model 4

(Intercept) 20.667** (9.064) 20.231** (9.025) 21.607** (8.897) 21.173** (8.832) Chinese FDI as a percentage of GDP 0.980 (1.472) -0.258 (0.938) 1.714 (1.245) 0.510 (0.493)

(14)

14 Natural resources as a percentage of GDP -0.255** (0.109) -0.253** (0.108) -0.213** (0.088) -0.212** (0.089) Total FDIs as a percentage of GDP 0.100* (0.057) 0.108* (0.055) 0.102* (0.055) 0.110* (0.055) Gini coefficient -0.057 (0.087) -0.067 (0.087) -0.054 (0.088) -0.063 (0.088) Democracy -3.385*** (0.834) -3.458*** (0.800) -3.484*** (0.811) -3.554*** (0.781) Natural logarithm of GDP per capita (measured in PPP) 4.037*** (1.059) 4.183*** (0.999) 3.888*** (1.048) 4.031*** (0.974)

Chinese FDI * Natural resources 0.058 (0.043) 0.057 (0.046) Chinese FDI * Democracy -0.235 (0.253) -0.226 (0.244) R2 0.599 0.598 0.596 0.595 Adj. R2 0.586 0.587 0.585 0.585 N 263 263 263 263

Note: OLS regression coefficients. Clustered standard errors in parentheses; * p<0.1 ** p<0.05 ***p<0.001.

5.1 Model 1

Looking at the independent variable, the Chinese FDI as a percentage of GDP, we see a value of 0.980 which is not significant with a p-value higher than 0.1. If the independent variable would increase with one, the dependent variable would increase with 0.980. This would mean that if the Chinese FDI as a percentage of the GDP would increase with one percent, it would result in an increase of 0.980 points on the CPI. In theory, an increase of Chinese FDI into Sub-Saharan African countries would thus lead to a minimal increase on the CPI scale, which would mean that Chinese FDI fosters a lower level of corruption perception. This goes in against the theory discussed in the theoretic framework. However, the p-value is higher than 0.1, hence conclusions cannot be drawn from these findings.

(15)

15

of -0.255 on the dependent variable. This means that when the revenues gained from natural resources as a percentage of the GDP increase with one, the dependent variable will decrease with 0.255. In this case it could be argued that a higher share of natural resource revenue of the GDP would lead to an increase of the corruption perception in a Sub-Saharan African country, since a decrease on the CPI means that a country is more corrupt. Our first control variable in the first model is highly significant with a p-value lower than 0.05. Literature (Leite & Weidmann, 1999) suggest that countries that are heavily relying on natural resources as a source of national income tend to be more vulnerable to corruption.

For my second control variable I find a score of 0.100. In this case, when the total FDIs as a percentage of GDP increase with one percentage point, the dependent variable will increase with 0.100. We can thus argue that an increase in FDIs into Sub-Saharan African countries leads to a decrease of the corruption perception (since a higher score means a lower level of corruption perception). The control variable has a p < 0.1, which means that this control variable is significant, however, this is only at the 90% level.

This result goes against the expectations deriving from the theoretical framework. It has been argued that if FDIs form a substantial amount of the total GDP a Sub-Saharan country has, the dependency on foreign investments in these countries is high. In that case, the presence of multinational corporations (MNCs) can change market dynamics, which in turns shapes opportunities for corrupt behaviour (Pinto & Zhu, 2016). Following these theoretical assumptions, we would expect to see that a higher amount of FDI as a percentage of a country’s GDP will be more likely to foster corrupt behaviour. The findings in the table above, however, suggest otherwise.

The third control variable, the Gini coefficient of a Sub-Saharan African country, has a score of -0.057. An increase of the Gini coefficient with one point will lead to a 0.057 decrease of the dependent variable. If a Sub-Saharan African country scores one point higher on the Gini scale (income inequality increases), the CPI score will decrease with 0.057 and the perceived corruption will thus be higher. A score of 0.057 isn’t particularly strong, and its p-value is higher than 0.1, so we conclude that this effect isn’t significant.

The level of political rights, or Democracy as noted above in the table, is the fourth control variable. The control variable has a coefficient of -3.385, which means that if a country has a lower level of Democracy, because the score increase with one, the CPI score will decrease with 3.385. We thus argue that a lower level of political rights leads to a higher level of corruption perception. This control variable is significant since there is a p-value lower than 1 percent. This result is in line with the expectations deriving from the theoretical framework,

(16)

16

since it is argued that a lower level of political rights that are guaranteed by the political system can be associated with corruption. Characteristics such as free press and open elections can increase the likelihood of exposing corruption (Larraín & Tavares, 2004, p. 221).

The fifth control variable, the natural logarithm of the GDP per capita (measured in PPP) of Sub-Saharan African countries, has a score of 4.037. This would mean that a one percent increase of the natural logarithm of the GDP per capita, the CPI score would increase with 0.04. The CPI increases with 0.04 because I divide 4.037 by 100, since I am dealing with a control variable that is a natural logarithm. A higher GDP per capita would thus lead to a lower perception of corruption. This fifth control variable is highly significant since it has a p-value that is lower than 0.01. This is in line with the expectations deriving from the theoretical framework. As stated before, Neeman et al. (2008) have argued that a high per capita income is closely associated with the existence of efficient and transparent institutions. As expected, the GDP per capita results in a significant negative effect, which means that a higher GDP per capita is closely related with ‘good’ institutions which in turn lead to a lower level of corruption.

Our sixth control variable in the first model, the interaction effect between the natural resource revenues as a percentage of the GDP and the amount of Chinese FDI into Sub-Saharan African countries, has a score of 0.058.

I can argue that Chinese FDIs lead to corruption especially in countries with an abundance of natural resources, since the interaction effect increases if natural resources and FDIs increase, and it has a negative effect on the CPI. I could therefore argue that Chinese FDIs lead to corruption especially in countries with an abundance of natural resources, however, this effect is not significant. I cannot draw any conclusion because of this.

Our seventh and final control variable in the first model, the interaction effect between

democracy and the amount of Chinese FDI into Sub-Saharan African countries, has a score of -0.235. This means that countries that score less on Democracy (because there is an increase of 1) and receive more Chinese FDIs have a positive effect on the corruption perception. I will therefore argue that countries that have lower civil liberties and political rights and receive more Chinese FDIs do not per se have a higher level of corruption perception. This score is not significant, so I cannot draw any conclusions from this finding.

5.2 Model 2

(17)

17

Chinese FDI and Democracy, has been removed. This is mainly due to the fact that interaction effects tend to create multicollineairty. This was also the case for the variables included in Model 1 (see Appendix). In model 2 and model 3, I hence test the interactions separately. In the final model, model 4, all interaction effects are removed.

The removal of the interaction effect has changed the values of all variables listed in the table above; however even though the values have increased or decreased in size, most of the values still have the same (significantly) negative or positive effect on the dependent variable. However, the independent variable has changed. Where in the first model the independent variable had a positive effect on the dependent variable (an increase in Chinese FDI led to a higher CPI score), an increase of the independent variable with one in the second model leads to a decrease of the CPI with 0.258. So, after removing the first interaction effect, the

independent variable now has a negative effect on the CPI. However, this effect still is not significant.

5.3 Model 3

In the third model I have removed the interaction effect between the Chinese FDI and the natural resources. The interaction effect between the Chinese FDI and Democracy is in turn added to the model.

Once again, the removal of the interaction effect has changed the values of all variables listed in the table above. All of the values still have the same (significantly) negative or positive effect on the dependent variable.

However, after the removal of the above mentioned interaction effect the independent variable has substantially increased in size to a value of 1.714 and has become a positive effect. It still is not a significant effect, with a p-value higher than 0.1.

5.4 Model 4

After removing the last interaction effect (the effect between the Chinese FDI and Democracy) from the third model, not much has changed. Even though the values have increased or decreased in size, most of the values still have the same (significantly) negative or positive effect on the dependent variable. I only find a substantial decrease in value of the independent variable compared to the third model: the independent variable now has

(18)

18

Now all the variables in the table above and their effects are explained, I will interpret the R2

and the adjusted R2 _{that belong to the three models. But before interpreting these two R’s I}

will elaborate on their function and why it is important to include them in the model.

The R2 explains shows how close the points (or country cases) are fitted to the regression line. The score that belongs to the R2 is quite simply the percentage of data explained by the model (Field, 2009, p. 179).

There is, however, a fundamental problem when it comes to the R2. The R2 will continue to increase every time an independent variable is added to the analysis. This means that in theory a larger amount of variables leads to higher percentage of data explained by the model. Besides the R2, there is also the adjusted R2. This measure tells us how much variance in Y would be accounted for if the model had been derived from the population from which the sample was taken (Field, 2009, p. 221).

Since this analysis contains a substantial amount of independent variables, it is wise to look at both the adjusted R2 as well as the R2. We will examine both R2 and the adjusted R2 in each of the three different models.

So, as stated before, the R2 is a measure of the amount of variability in one variable that is shared by the other. In the first model, the R2 _{has a score of 0.599, which means that the}

model accounts for 59.9% of the variability of the dependent variable. In the second model, after the interaction effect between the Chinese FDI and Democracy has been removed, the R2_{changes slightly and now has a value of 0.598. After removing the last interaction effect}

we see that the R2 has decreased in size to a value of 0.595 in model 3. Finally, the R2 in all of the three models is highly significant since there is a p-value lower than 0.01. The adjusted R² similarly changes little between Models, indicating that the interaction effects did not add anything to explaining the dependent variable.

6 Discussion

As we have seen above, the independent variable has a positive effect on the dependent variable. If the Chinese FDI as a percentage of GDP increases with one percentage point, the CPI increases in the first, the third and the fourth model. In the second model however, when the first interaction effect is removed, a rise in the Chinese FDI as a percentage of GDP results in a decrease of the CPI. The effect of the independent variable in the three models is however not significant at the 90%-level. Going against my expectations, the independent variable doesn’t cause a lower CPI score, which means that a bigger amount of Chinese FDIs

(19)

19

don’t fuel corruption. This finding is contrary to the expectations we had, and it doesn’t align with my main hypothesis.

So why does this effect not turn out significantly negative?

As mentioned earlier in this work, literature (Larraín and Tavares, 2004) has also suggested that (Chinese) FDI can diminish the level of corruption in a country. These two authors agree with Elliot (1997) on the idea that “foreign standards of probity have an impact on local officials and their behaviour” (Larraín and Tavares, 2004, p. 219). In that sense, we could argue, FDIs might diminish the level of corruption because Chinese values are adopted by the local officials.

Another reason why my findings are not in line with my hypothesis might be the fact that literature on Chinese FDI can be based on subjective views. As Prah (2007) argues, the Chinese move into Sub-Saharan Africa “must be understood as western disquiet regarding the expanding profile of China in Africa, which threatens to undermine and overthrow the hegemony that the West has hitherto held in African economies and foreign trade” (p. 70). Because of this western disquiet it might not be surprising that my findings are different than expected.

Even though these arguments above suggest that it might not be surprising that the relation between the dependent and independent is not like what we would’ve expected, research (Pinto & Zhu, 2016) done on this topic has turned out to be significant – in contrary to my findings. Why do those works succeed in proving a relationship between Chinese FDIs and the level of corruption in Sub-Saharan African countries?

A reason might be that the variables used by these two authors differ significantly from the variables used in this research. Their dependent variable is, like in this research, the CPI. However, the CPI used in Pinto and Zhu’s work covers other years, namely from 2000 to 2004. The biggest difference, however, comparing this research to theirs, is the fact that the CPI scores of all countries available are selected in their work. Where this research has been focussed on the CPI score in Sub-Saharan African countries, Pinto and Zhu’s work does not focus on one particular region.

There might be another, more fundamental, reason why my findings are not in line with the hypothesis formulated. My theory suggests that political and economic conditions (the crowding out of domestic corporations by Chinese MNC) create the incentives that “shape opportunities for corrupt behaviour”. My research is limited to only six years, and it would therefore be difficult to detect a significantly negative effect between Chinese FDI and corruption when using such a small timeframe.

(20)

20

Besides the dependent variable used in their work, other variables are used in their work when compared to this one. In their work, a special place has been reserved for ethnic

fractionalization and the religious identity of countries. In this research, the aspect of religion and ethnic fractionalization is not accounted for, however this could be added in further research since Sub-Saharan Africa is rich and diverse when it comes to ethnicities and religions.

Besides using an OLS regression, Pinto and Zhu also make use of the two-stage least squares (2SLS) type of analysis (Pinto & Zhu, 2016, p. 697). In this research the focus mainly lies on an OLS type of regression – this is another reason why the results might differ from theirs. Although we have taken independent variables from the year prior to the CPI score, it is still difficult to establish what is cause and effect. It may still be the case that countries with low levels of corruption in general attract more FDIs.

All in all, looking at other research, it might still possible to prove a significant relationship between FDI and corruption; however, it is still debatable if this relationship is also provable in the case of Sub-Saharan Africa. We could argue that further research on this topic is needed. When specifying on a different region or time period, we see that results can differ dramatically. Further research, using different research methods, could also provide us with different results that could contribute to the general academic knowledge on this topic.

7 Conclusion

This research has focussed on the question whether a higher amount of Chinese FDIs will increase the level of corruption in Sub-Saharan African countries. While much research has been done relationship between FDI and corruption, the reversed relation researched in this work has not gained much attention yet.

Following Pinto and Zhu (2016), I have argued that dependent on the political and economic conditions in a country, Chinese FDIs can “shape opportunities for corrupt behaviour”. Yet even though literature has suggested otherwise, I have not been able to prove a negative relationship between the amount of Chinese FDI and the level of corruption in Sub-Saharan African countries.

A possible explanation for not finding this significant relationship may be due to the fact that this research has specifically been focused on Sub-Saharan Africa only. Another possible explanation might be that the statistical test conducted in this work is too simple to take into account the complex interrelations between FDIs and corruption.

(21)

21

(22)

22

Literature

Adams, S. (2009). Can foreign direct investment (FDI) help to promote growth in Africa? African Journal of Business Management, 3(5), 178-183.

Ades, A., & Di Tella, R. (1999). Rents, competition, and corruption. American economic review, 89(4), 982-993.

Aitken, B. J., & Harrison, A. E. (1999). Do domestic firms benefit from direct foreign investment? Evidence from Venezuela. American economic review, 89(3), 605-618. Ajayi, S. (2006). Paper for presentation at the ADB. In AERC International Conference on

Accelerating Africa’s Development five years into the twenty-first century, Tunis, Tunisia.

Alfaro, L., Areendam, C., Kalemli-Ozcan, S., & Sayek, S. (2004). FDI and economic growth: The role of local financial markets. Journal of International Economics, 64(1), 89-112.

Arezki, R., & Gylfason, T. (2013). Resource rents, democracy, corruption and conflict: Evidence from sub-Saharan Africa. Journal of African Economies, 22(4), 552-569. Benoit, K. (2011). Linear regression models with logarithmic transformations. London School

of Economics, London, 22(1), 23-36.

Besada, H., Wang, Y., & Whalley, J. (2011). China's growing economic activity in Africa. In China's Integration Into The World Economy (pp. 221-254).

Broadman, H. G. (2006). Africa's silk road: China and India's new economic frontier. The World Bank.

Carkovic, M., & Levine, R. (2002). Does foreign investment accelerate economic growth? Finance Department, University of Minnesota.

Caves, R. E., & Caves, R. E. (1996). Multinational enterprise and economic analysis. Cambridge university press.

Chan-Fishel, M., & Lawson, R. (2007). Quid pro quo? China's investment-for-resource swaps in Africa. Development, 50(3), 63-68.

Chen, M. X., Goldstein, A., Pinaud, N., & Reisen, H. (2005). China and India: What’s in it for Africa. Draft. Paris: OECD Development Centre.

Cheung, Y. W., De Haan, J., Qian, X., & Yu, S. (2012). China's outward direct investment in Africa. Review of International Economics, 20(2), 201-220.

Chintu, N., & Williamson, P. J. (2013). Chinese state-owned enterprises in Africa: Myths and realities. Ivey Business Journal, 2.

(23)

23

CIA World Factbook. (2015). Eritrea: GDP – per capita. Retrieved from

https://www.cia.gov/library/publications/the-world-factbook/geos/er.html Elliott, K. A. (1997). Corruption and the global economy. Peterson Institute. Field, A. (2009). Discovering statistics using SPSS. Sage publications.

General Office of the State Council. (2006). Better implement the”going out” strategy. Retrieved from http://www.gov.cn/node_11140/2006-03/15/content_227686.htm Gill, B., & Reilly, J. (2007). The tenuous hold of China Inc. in Africa. Washington

Quarterly, 30(3), 37-52.

Greenaway, D. S. (1998). Does trade liberalisation promote economic development? Scottish Journal of Political Economy, 45(5), 491-511.

Gupta, S., Davoodi, H., & Alonso-Terme, R. (2002). Does corruption affect income inequality and poverty? Economics of governance, 3(1), 23-45.

Kaplinsky, R., McCormick, D., & Morris, M. (2007). The impact of China on sub-Saharan Africa.

Kaplinsky, R., & Morris, M. (2009). Chinese FDI in Sub-Saharan Africa: engaging with large dragons. The European Journal of Development Research, 21(4), 551-569.

Kolstad, I., & Wiig, A. (2012). What determines Chinese outward FDI?. Journal of World Business, 47(1), 26-34.

Landler, M., & Wong, E. (2018). Bolton Outlines a Strategy for Africa That’s Really About Countering China. Retrieved from

https://www.nytimes.com/2018/12/13/us/politics/john-bolton-africa-china.html Large, D. (2008). Beyond ‘dragon in the bush’: the study of China–Africa relations. African

affairs, 107(426), 45-61.

Larraín, B., & Tavares, J. (2004). Does foreign direct investment decrease corruption? Cuadernos de economía, 41(123), 199-215.

Lavrakas, P. J. (2008). Encyclopedia of survey research methods. Sage Publications. Le Billon, P. (2003). Buying peace or fuelling war: the role of corruption in armed

conflicts. Journal of International Development: The Journal of the Development Studies Association, 15(4), 413-426.

Leite, Carlos, & Weidmann, Jens. (1999). Does Mother Nature Corrupt - Natural Resources, Corruption, and Economic Growth.

Melo, L., & Quinn, M. A. (2015). Oil, foreign direct investment and corruption. The International Journal of Business and Finance Research, 9(1), 33-49.

(24)

24

relationship. Santa Monica, CA: RAND Corporation.

Neeman, Z., Paserman, M. D., & Simhon, A. (2008). Corruption and openness. The BE Journal of Economic Analysis & Policy, 8(1).

Pinto, P. M., & Zhu, B. (2016). Fortune or evil? The effect of inward foreign direct investment on corruption. International Studies Quarterly, 60(4), 693-705.

Prah, K. K. (2007). China and Africa: Defining a relationship. Development, 50(3), 69-75. Quinn, A. (2011). Clinton warns against "new colonialism" in Africa. Retrieved from

https://www.reuters.com/article/us-clinton-africa/clinton-warns-against-new-colonialism-in-africa-idUSTRE75A0RI20110611

Romer, P. (1993). Idea gaps and object gaps in economic development. Journal of monetary economics, 32(3), 543-573.

Sachs, J. D., & Warner, A. M. (1995). Natural resource abundance and economic growth (No. w5398). National Bureau of Economic Research.

Sandholtz, W., & Gray, M. M. (2003). International integration and national corruption. International Organization, 57(4), 761-800.

Transparency International. (2010). Corruption Perceptions Index 2010: Long Methodological Brief. Retrieved from

http://transparency.ee/cm/files/cpi2010_long_methodology_en.pdf UNDP in Africa. (n.d.). About Sub-Saharan Africa. Retrieved from

http://www.africa.undp.org/content/rba/en/home/regioninfo.html

Wei, S. J. (1997). Why is corruption so much more taxing than tax? Arbitrariness kills (No. w6255). National bureau of economic research.

Wei, S. J. (2000). How taxing is corruption on international investors? Review of economics and statistics, 82(1), 1-11.

(25)

25

Appendix

In the appendix I will discuss whether the OLS assumptions have not been violated in the regression analyses. If one of the assumptions is violated, this means I have a problem concerning the validity of the linear regression results.

I will check five different assumptions:

- No non-linear relations - No multicollinearity - No heteroskedasticity

- Normal distribution of the errors - No outliers/influential cases

No non-linear relations

To test the first assumption, the absence of a non-linear relationship, I will create a scatterplot for each of the four models in which the standardized residuals are portrayed on the y-axis and the standardized predicted values are portrayed on the x-axis.

(26)

26

Model 2

(27)

27

Model 4

When controlling for the absence of a non-linear relationship, I want to avoid any shape that could look like it is non-linear (e.g. quadratic or exponential). A random cloud of

observations indicates that there is a linear relationship. This is also the pattern shown in the four scatterplots above. Each of the four models indicates that there is a linear relationship; hence the linearity assumption has not been violated.

No multicollinearity

If we want to control if multiple independent variables heavily influence each other we want to check for multicollinearity. In general, we don’t want to encounter high inter-correlation between independent variables as this increases standard errors and increases the risk of dismissing results which would otherwise be significant

To check for multicollinearity we will make use of collinearity statistics. These statistics will show us to which extent multicollinearity is present. To interpret the data, we will look at the VIF-value of each independent variable. If we find a Tolerance value lower than 0.2 this is cause for concern. At the same time, finding a VIF-Value higher than 5 and especially higher than 10 is problematic. In the table below all four models are depicted with their respective VIF-value and the tolerance level.

I find a VIF-value of 10.245 with a tolerance level of 0.098 for the interaction effect between the amount of Chinese FDI as a percentage of the GDP and Democracy in model 1. The

(28)

VIF-28

value of the independent variable also raises concerns with a value of 11.521 and a tolerance level of 0.087.

I will run different models where the interaction effects are included separately to avoid multicollinearity. Even though the data suggests otherwise, I will run a third model were the second interaction effect (the interaction effect between the amount of Chinese FDI as a percentage of the GDP and natural resources) is also excluded from the model. The reason why I will exclude the interaction effects is the fact that the two interaction effects are both a multiplication of the independent variable with another control variable. It would therefore not be surprising to find the second interaction effect influence the independent variable (even though the VIF-value and the tolerance level do not have values that would immediately raise concern).

Table 3: Tolerance level and VIF-value of the linear regression model

Model 1 Model 2 Model 3 Model 4

Tole-rance VIF Tole-rance VIF Tole-rance VIF Tole-rance VIF (Constant) Chinese FDI as % of GDP (Unadjusted) 0.087 11.521 0.406 2.463 0.097 10.275 0.942 1.062 NR as Percentage of GDP (2011-2016) 0.496 2.017 0.497 2.014 0.717 1.395 0.718 1.393 Foreign direct investment, net inflows (% of GDP) 0.782 1.280 0.815 1.228 0.782 1.279 0.815 1.227 GINI index (World Bank estimate) 0.831 1.203 0.854 1.170 0.832 1.202 0.856 1.169 Democracy 0.653 1.532 0.689 1.451 0.678 1.474 0.716 1.396 PPP 0.703 1.423 0.762 1.313 0.726 1.378 0.791 1.265 ChineseFDIxNR 0.318 3.149 0.318 3.147

(29)

29

ChineseFDIxDem 0.098 10.245 0.098 10.240

In the second model I find that after removing the control variable between Chinese FDI and Democracy the problem of multicollinearity is solved, since there is no VIF-value higher than 5 or 10. There is no tolerance level lower than 0.2 in the second model, which shows that the removal of the above named interaction effect solves a big part of the problem concerning multicollinearity.

In the third model the interaction effect between Chinese FDI and natural resources is removed. The independent variable as well as the remaining interaction effect has a VIF-value higher than 10, which is cause for concern. The tolerance level has also dropped to a value lower than 0.2 after the above mentioned interaction effect is removed.

In the last model, when all the interaction effects are removed, I find different results. Not one of the independent variables has a VIF-value higher than 5 and no tolerance level is lower than 0.2.

We can conclude that after removing the interaction effect from the model, we’ve dealt with the problem of multicollineairity.

No heteroscedasticity

To check for hetroscedasticity we will make use of the same four scatterplots used to control for non-linearity. When checking for this assumption, we want the points to be distributed evenly over the scatterplot.

(30)

30

Model 2

(31)

31

(32)

32

As we progress on the x-axis, the variation in the residuals should be roughly similar for all of the four models – this isn’t the case here. I will therefore argue that there is indeed some heteroscedasticity in all of the four models in this research.

Normal distribution of errors

We will now test the assumption that is the normal distribution of errors. To control this we make use of a normal P-P plot. We create a normal P-P plot for all of the four models.

(33)

33

Model 2

(34)

34

Model 4

As we can see in the four graphs above, the points of the model follow the diagonal line without any extreme variation. Therefore, we will assume that the residuals in this model are normally distributed.

No outliers/influential cases

To control for outliers and influential cases we will look at two things: Cook’s distance and the standardized residuals. Cook’s distance can show us if there is an influential case in the regression model which might distort the results. Influential cases tend to be cases with an extreme score. The scores diverge so much from the other cases that influential cases can influence the regression line.

The standardized residuals are used to detect outliers. Outliers are cases that lie far away from the regression line.

When we test for both of these values we want a Cook’s distance not higher than a 1. For the standardized residuals we don’t want more than 5% of the cases to have a higher score than 1.96 or a score lower than -1.96, 1% of the cases to have a score higher than 2.58 or lower than -2.58, and we don’t want any cases with a score higher than 3.29 or lower than 3.29.

In the tables below the findings regarding the Cook’s distance (cases with a value higher than 1) and the standardized residuals (cases with a value higher than 1.96 or lower than -1.96, higher than 2.58 or lower than -2.58 and higher than 3 or lower than -3) are illustrated for our

(35)

35

first model.

Table 4: Frequency Table of cases with Standardized residuals with a value higher than 1.96 or lower than -1.96

Frequency Percent Valid Percent

Valid Value between -1.96 and 1.96 251 90.9 95.4

Value higher than 1.96 or lower than -1.96 12 4.3 4.6

Total 263 95.3 100.0

Missing 13 4.7

(36)

36

Value lower than -2.58 or higher than 2.58 6 2.2 2.3

Total 263 95.3 100.0

Missing 13 4.7

Total 276 100.0

Table 6: Frequency Table of cases with Standardized residuals with a value higher than 3 or lower than -3

Valid Value between -3 and 3 257 93.1 97.7

Value lower than -3 or higher than 3 6 2.2 2.3

Total 263 95.3 100.0

Missing 13 4.7

Total 276 100.0

Table 7: Frequency Table of cases with a Cook’s distance higher than 1

Valid Cook’s distance lower than 1 262 94.9 99.6

Cook’s distance higher than 1 1 .4 .4

Total 263 95.3 100.0

Missing 13 4.7

(37)

37

For our first model, we find one case with a Cook’s distance higher than 1. This means that there is one influential case in this model. The influential case in this model is the Central

African Republic in 2015. In 2015 this country has a Cook’s distance of 1.902 and is thus well exceeding the maximum value of 1.

To check if the observation affects the regression results, I will filter this case out of the regression model before running the model again. But before I do this I will first need to control for problems that arise when it comes to standardized residuals.

Even though less than 5% of the cases have a score higher than 1.96 or lower than -1.96, I still find cases exceeding the limits established. There are in fact 6 cases remaining when controlling for standardised residuals with a score higher than 3 or lower than -3. According to the assumptions, finding any standardized residuals with scores exceeding these values should be filtered out. This is problematic since these outliers can influence the findings. These 6 cases all belong to Rwanda in the period 2012-2017. The standardized residuals that linked to these 6 years all have a value greater than 3. I will therefore filter Rwanda out of the model, since a standardized residual with a score higher than 3 might influence the outcome of the model significantly.

Before I present the results of the linear regression models after the cases mentioned above are filtered out, I still need to address the missing cases in the model. In this model 13 cases are labelled as missing, this is also mentioned in the tables above. This is not any different from the first model of linear regression, since that model also excluded a total of 13 cases that were seen as invalid.

I will now discuss the Cook’s distance and the standardized residuals for all of the other 3 models. After I have controlled for the Cook’s distance and the standardized residuals, I will present my findings in one final linear regression table.

Model 2

(38)

38

Missing 13 4.7

Total 276 100.0

Total 263 95.3 100.0

Missing 13 4.7

Total 276 100.0

Table 10: Frequency Table of cases with Standardized residuals with a value higher than 3 or lower than -3

Total 263 95.3 100.0

Missing 13 4.7

Total 276 100.0

Valid Cook’s distance lower than 1 263 100.0 100.0

Cook’s distance higher than 1 0 0 0

(39)

39

Missing 13 4.7

Total 276 100.0

The findings regarding the Cook’s distance and the standardized residuals in the second model are almost identical. The only difference I found was that there is no Cook’s distance with a value higher than 1 in the second model.

I then filter out the cases with standardized residuals higher than the values permitted. I find that there is no significant change in values in the second model compared to when there was not controlled for the assumptions of influential cases and outliers.

Model 3

Table 12: Frequency Table of cases with Standardized residuals with a value higher than 1.96 or lower than -1.96

Total 263 95.3 100.0

Missing 13 4.7

Total 276 100.0

Total 263 95.3 100.0

Missing 13 4.7

(40)

40

Value lower than -3 or higher than 3 ₆ _2.2 _2.3

Total ₂₆₃ _95.3 _100.0

Missing 13 4.7

Total 276 100.0

Valid _{Cook’s distance lower than 1} 263 94.9 99.6

Cook’s distance higher than 1 1 0.4 0.4

Total 263 95.3 100.0

Missing 13 4.7

Total 276 100.0

The findings regarding the Cook’s distance and the standardized residuals in the third model are almost identical to the findings in the first model. Compared to the second model, I now find one case with a Cook’s distance higher than 1, just like in the first model.

The values after checking for assumptions of influential cases and outliers are the same as they were before I checked for assumptions.

Model 4

(41)

41

Total 263 95.3 100.0

Missing 13 4.7

Total 276 100.0

Total 263 95.3 100.0

Missing 13 4.7

Total 276 100.0

Total 263 95.3 100.0

Missing 13 4.7

Total 276 100.0