Urbanization and Economic Growth: Panel Data Evidence from Africa

Urbanization and Economic Growth: Panel Data Evidence from Africa Prepared By: Arega Getaneh October 2018 2 Abstract This paper focuses on Urbanization and Economics growth in Africa. The method applied is panel data investigation to justify the endogeneity problem related to such variables, GMM. We included variables important for economic growth and found that urbanization has both pros and cons for the economic growth in the region which are supported by extended neoclassical economic models. Some variables like government expenditure and Capital formation has to modify to manage the dynamic estimation techniques. The finding that urbanization impacted the economics growth of the region may be explained by the slum and unorganized urban policy in Africa. i Contents 1 2 3 Introduction 1 1.1 Some Backgrounds . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Questions to be answered . . . . . . . . . . . . . . . . . . . . 2 1.3 Endogenous Growth Theory . . . . . . . . . . . . . . . . . . . 3 1.4 Growth at cross Countries level . . . . . . . . . . . . . . . . . 4 1.5 Variables in the Model . . . . . . . . . . . . . . . . . . . . . . 4 1.5.1 6 Related Literature 7 2.1 Urbanization . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 2.2 Urbanization issues in Africa . . . . . . . . . . . . . . . . . . . 8 9 Model Specification 3.1 Estimation Techniques . . . . . . . . . . . . . . . . . . . . . . 10 3.2 Fixed Effect (FE) Estimation . . . . . . . . . . . . . . . . . . 10 3.3 GMM as a Method of Estimation . . . . . . . . . . . . . . . . 11 3.4 Why Dynamic Panel Analysis? 3.5 4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . Testing for Serial Correlation . . . . . . . . . . . . . . . . . . 17 19 Data Presentation and Analysis 4.1 . . . . . . . . . . . . . . . . . 13 Descriptive Analysis . . . . . . . . . . . . . . . . . . . . . . . 19 4.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 19 4.1.2 Summary statistics . . . . . . . . . . . . . . . . . . . . 19 ii 4.1.3 4.2 5 Measurement of Multicollinearity . . . . . . . . . . . . 21 Econometric Estimation Results . . . . . . . . . . . . . . . . 24 4.2.1 Pooled-OLS and Fixed Effects Result . . . . . . . . . . 24 4.2.2 GMM Estimation Results . . . . . . . . . . . . . . . . 27 30 Concluding Remarks iii List of Tables 4.1 4.2 4.3 4.4 4.5 Summary Statistics . . . . . . . . . . . . . . . . . . . . . Missing values summary statistics . . . . . . . . . . . . . The Correlation between Variables . . . . . . . . . . . . Pooled-OLS and Fixed Effects Result with logarithm and One step system GMM and One step difference GMM . . 5.1 5.2 Dynamic panel-data estimation with levels . . . . . . . . . . . 35 Missing summary statistics for Dynamic panel-data estimation 36 iv . . . . . . . . . levels . . . 20 21 22 26 28 List of Figures 4.1 4.2 Scatter plot for per capita GDP . . . . . . . . . . . . . . . . . 23 Scatter plot for Urbanization . . . . . . . . . . . . . . . . . . . 24 v vi Chapter 1 Introduction 1.1 Some Backgrounds Urbanization is closely linked to economic development. As economies develop, relative and absolute changes in demand increase the relative and absolute importance of the manufacturing and service sectors. The relationship between urbanization and economic growth across various countries in Africa is unclear. As these economies urbanize and grow, it is difficult to interpret what one can expect from such growth of urbanization. Whether economic growth stimulates urbanization or vice-versa, or they move together is debatable issue. Indeed, this project tries to investigate the impact of urbanization on economic growth of African countries using panel data and Generalized Method of Moments (GMM) as a method of estimation. Urbanization in Africa has different trends and contentious in its contributions for economic growth. Currently, urbanization is in its high growth rate and indeed, most of African countries economies grow as well. Therefore, it is timely to investigate if urbanization can be important variable as evidences show they grow simultaneously. Cities encompass enormous control over national economies as they provide jobs, access to the best cultural, educational and health facilities and also act as hubs for communication and transports which are necessary conditions for 1 economic development of any nation (Polèse, 2005) On the other hand, cities also cluster massive demand for energy, generate large quantities of waste and concentrate pollution as well as social hardship. According to Sarker et al. (2016), the economic and social crises that have enveloped in most of the developing countries are a result of urban growth without proportionate economic development. Some others also agree that the continuous increase in the proportion of people living in cities as compared to rural areas in the developing countries has resulted to large number of slums and deplorable living conditions in the cities and in most cases worsening the economic circumstances of urban migrants of the countries. 1.2 Questions to be answered The questions asked are; why urbanization which was a necessary condition for economic development of developed countries has underestimated element to the development process of African countries? What explains the relationship between urbanization and economic development in third world nations? Should developing countries encourage urbanization as part of economic development strategy? Or is high rate of urbanization just a necessary condition for economic growth? These and others are frequently asked by scholars and researchers in the area. In this project, we try to answer the last two questions. Some try to answer these questions by applying different methods, for instance, McCoskey et al. (1998) employed OLS however, did not properly address the problem of endogenety.Barrios et al. (2003) employed semi-parametric estimation techniques in their qualitative analysis, as a measure of urbanization and economic growth; their study suffered from the problem of perception bias, not to mention, measurement error. Other studies present cross-sectional estimates, but did not adequately control the problem of endogeneity. Some findings reflect unobserved characteristics which do not vary over time instead of being the consequences of urbanization or might reflect reverse causality. By considering these, in this project we try to answer some of the questions 2 and mainly we examine African data to answer: • What looks like the relationship between urbanization and economic growth in Africa? • Do African countries experience an increasing economic growth and urbanization growth simultaneously? • Should African countries encourage urbanization as part of economic development strategy? Therefore, we revisit urbanization and economic growth in Africa by employing panel data and panel data models. The study avoids problems committed by most of the studies reviewed using IV strategy during estimation of the impact of urbanization on economic growth and we recommend and implicate policy issues regarding urbanization in the continent. 1.3 Endogenous Growth Theory In the 1960s, growth theory consisted mainly of the neoclassical model, as developed by (Solow, 1956; Swan, 1956; Koopmans et al., 1965). Endogenous growth theories that include the discovery of new ideas and methods of production are important for providing possible explanations for long-term growth. Yet the recent cross-country empirical work on growth has received more inspiration from the older neoclassical model, as extended to include government policies, human capital, and the diffusion of technology. Theories of basic technological change seem most important for understanding why the world can continue to grow indefinitely in per capita terms. But these theories have less to do with the determination of relative rates of growth across countries, the key element studied in cross-country statistical analyses. Recent extensions of the model suggest the inclusion of additional sources of cross-country variation, especially government policies with respect to levels of consumption spending, protection of property rights, and distortions of domestic and international markets. 3 1.4 Growth at cross Countries level The Neo classical model starts explaining economic growth as: Y = f (K, L, A) (1.4.1) where Y is output, K is capital formation and L labor and A is technology. For example, Barro and Lee (1993) and Barro and Sala-i Martin (1997) include additional variables to determine the economic growth. With that inspirations, our model incorporates urbanization among the explanatory variables. lnYit = lnGit +lnGEit +lnDGCit +lnOkit +lnUit +lnTit +lnEC +εit (1.4.2) where lnYit economic growth (in logarithmic terms) of country i at time t, GEit gross expenditure of country i at time t, CFit the difference between gross expenditure and gross capital formation of country i, at time t, Okit trade openness (degree of openness) of country i at time t, Uit urbanization of country i at time t, Tit international tourism, receipts of country i at time ECit and εit iid error terms. 1.5 Variables in the Model GDP Per Capita (constant 2010 US$) is used as the dependent variable and we use the logarithm of this variable as we can explain the elasticities of some changes in explanatory variables Gross Capital Formation (constant 2010 US$) Gross capital formation (formerly gross domestic investment) consists of outlays on additions to the fixed assets of the economy plus net changes in the level of inventories. Fixed assets include land improvements (fences, ditches, drains, and so on); plant, machinery, and equipment purchases; and the construction of roads, railways, and the like, including schools, offices, hospitals, private residential dwellings, and commercial and industrial buildings. In4 ventories are stocks of goods held by firms to meet temporary or unexpected fluctuations in production or sales, and ”work in progress.” According to the 1993 SNA, net acquisitions of valuables are also considered capital formation. Data are in constant 2010 U.S. dollars. Gross National Expenditure (constant 2010 US$) Gross national expenditure (formerly domestic absorption) is the sum of household final consumption expenditure (formerly private consumption), general government final consumption expenditure (formerly general government consumption), and gross capital formation (formerly gross domestic investment). Data are in constant 2010 U.S. dollars. Openness or Terms of Trade (% of GDP) The relationship between openness and economic growth has long been a subject of much interest and controversy in international trade literature. Regarding a theoretical relationship between openness and growth most of the studies provide support for the proposition that openness affects growth positively. Romer (1986) and Barro and Sala-i Martin (1997) among others, argue that countries that are more open have a greater ability to catch up with leading technologies of the rest of the world. We will measure openness with trade share ( (Import+Export) ) is used in this analysis ( Trade is the sum GDP of exports and imports of goods and services measured as a share of gross domestic product.) Urban Population (% age of national population) Urban population refers to people living in urban areas as defined by national statistical offices. The data are collected and smoothed by United Nations Population Division. International Tourism,Rreceipts (% of total exports) International tourism receipts are expenditures by international inbound visitors, including payments to national carriers for international transport. These receipts include any other prepayment made for goods or services re5 ceived in the destination country. They also may include receipts from sameday visitors, except when these are important enough to justify separate classification. For some countries they do not include receipts for passenger transport items. Their share in exports is calculated as a ratio to exports of goods and services, which comprise all transactions between residents of a country and the rest of the world involving a change of ownership from residents to nonresidents of general merchandise, goods sent for processing and repairs, nonmonetary gold, and services. Renewable Energy Consumption (% of total final energy consumption) Renewable energy consumption is the share of renewable energy in total final energy consumption. Unemployment with advanced education (% of total labor force with advanced education). The percentage of the labor force with an advanced level of education who are unemployed. Advanced education comprises short-cycle tertiary education, a bachelor’s degree or equivalent education level, a master’s degree or equivalent education level, or doctoral degree or equivalent education level according to the International Standard Classification of Education 2011 (ISCED 2011). 1.5.1 Data Data for this project is extracted from World Development Indicators. We include data of 40 African countries1 for 27 years (from 1990-2016). The starting year 1990 is chosen due the the fact that most African countries have some policy and regime changes in the 1990s. 1 Algeria, Angola, Benin, Botswana, Burkina Faso, Burundi, Cape Verde, Djibouti, Cameroon, Central African Republic, Congo, Dem. Rep., Comoros, Congo, Republic of, Cote d‘Ivoire, Egypt, Equatorial Guinea, Eritrea, Gabon, Gambia, Ghana, Kenya, Lesotho, Liberia, Libya, Malawi, Mali, Morocco, Namibia, Niger, Rwanda, Senegal, Seychelles, Sierra Leone, South Africa, Sudan, Tanzania, Togo, Uganda, Zambia and Zimbabwe 6 Chapter 2 Related Literature 2.1 Urbanization Demographically, the term urbanization denotes the redistribution of population from rural to urban settlement over time. However, it is important to acknowledge that the criteria for defining what is urban may vary from country to country which cautions one against a strict composition of urbanization across countries. The fundamental difference between urban and rural is that urban populations live in larger, denser, and more heterogeneous places as opposed to small, sparse, and less differentiated rural places. (Henderson, 2003). Economically, it is a process that considers human, economic, social, amenities, etc. agglomeration, which translates an area of country side or a village, into a town or part of one or further growth and expansion of already existing urban centers. Urbanization occurs as countries switch sectoral composition away from agriculture into industry and as technological advances in domestic agriculture release labor from agriculture to migrate to cities (Moomaw and Shatter, 1993). That said about the meaning of urbanization, it is worth noting the impacts and/or determinants of urbanization and its impact on an economy. Result of studies of both impacts/determinants and implications of urbanization at cross country level have long been in the literature. 7 In terms of development and growth theory, urbanization occupies a puzzling position. On the one hand, it is recognized as fundamental to the multidimensional structural transformation that low-income rural societies undergo to modernize and to join the ranks of middle- and high-income countries. Some models, such as Lucas (2004), explicitly consider how urbanization affects the growth process (primarily through the enhanced flow of ideas and knowledge attributable to agglomeration in cities. 2.2 Urbanization issues in Africa A growing literature argues that, in the period since independence, Africa’s urbanization process has differed fundamentally from the rest of the developing world. This complements a large literature arguing that many economic processes in Africa are different Binswanger and Townsend (2000); Acemoglu and Robinson (2010). The urbanization literature suggests that two stylized facts that hold for the rest of the world do not hold for Africa. First, while rapid urbanization in other developing regions has been accompanied by fast macro-economic growth, Africa has seen urbanization without growth (Gollin et al., 2016). Second, while, in the rest of the world, urbanization has generally been accompanied by a sector transformation from agriculture to manufacturing, Africa is urbanizing without industrialization. Empirical evaluation of both stylized facts is complicated by relatively low quality data for Africa and by the issue of how to evaluate causal relationships. Despite the region’s clear urbanization trend, national governments and development groups continue to direct their energies toward rural economic development. Economists attribute this focus to a widespread belief in “urban bias,” the notion that urban groups receive preferential economic treatment because, by virtue of their location, they are able to pressure the government more effectively than rural groups. 8 Chapter 3 Model Specification To meet the main objective of this project, urbanization is included as one of the explanatory variables for economic growth. More specifically, the project has the following model, adapted from the extended version of neo-classical growth model: Yit = αt + θ′ Xit + β ′ Uit + εit (3.0.1) where Yit measures the log of annual growth rate in real GDP per capita in country i at time t; Xit is n × k vector of variables identified as the important determinants of economic growth in country i and at time t supposed to have full column rank and E(x′it εit ) = 0; Uit is the interest variable which is urbanization rate in country i and at time t with E(Uit′ εit ) 6= 0 hence, endogenous; α θ and, β are θ k ×n vector of coefficients; and εit is the iid error term. Xit are extended Barro’s regression economic growth determinants which include: gross capital formation, gross government expenditure, percentage of total trade, FDI, international tourism and renewable energy consumption. In addition, urbanization is defined as ratio of urban population to total population. More interestingly, the project have the following dynamic regression model: Yit = αi + γ ′ Yit−1 + θ′ Xit + β ′ Uit + εit 9 (3.0.2) where everything is the same as in (3.0.1), but in (3.0.2), we include the lag of the dependent variable as explanatory. 3.1 Estimation Techniques Pooled-OLS To estimate the above econometric model, we apply Pooled-OlS as a benchmark to see the signs and magnitude of the estimations. Since in panel data analysis, the assumptions underlying pooled-OLS are unlikely to be maintained (it will be heavily biased because of unobserved heterogeneity (Uit and Xit would be correlated)). This is due to the fact that pooled-OLS also relies on a between comparison (variation). Compared with the cross-sectional OLS the bias is lower, because pooled-OLS also takes in-to account the within variation. 3.2 Fixed Effect (FE) Estimation The number of years (T = 27). If T were larger than number of countries (N = 40), one year shocks impact on the country’s apparent fixed effect would dwindle and so would the endogeneity problem raise here. There are two ways to deal with this endogeneity. One way to solve this problem is using Difference Generalized Methods of Moments (DGMM) since it transforms the data to remove the fixed effects and the other is, to instrument (Yit ), and any other similarly endogenous variables which have reverse causality using System GMM (Greene, 2008). An intuitive first attack on the fixed effects is to draw them out of the error term by entering dummies for each individual-the so-called Least Squares Dummy Variables (LSDV) estimator. To show how it works mathematically consider the following: If zi is unobserved, but correlated with Xit and Uit then the least squares estimator of γ, β and θ would be biased and inconsistent because of an omitted 10 variable. In this instance, the model: Yit = ψi + γ ′ Yit−1 + θ′ Xit + β ′ Uit + εit (3.2.1) where ψi = zt′ αi , embodies all effects and specifies an estimable conditional mean. This fixed effects approach takes αi to be a group-specific constant term in the regression model. It should be noted that the term “fixed” as used here signifies the correlation of αi and Xit , that αi is non-stochastic. In applying Fixed Effects (FE) estimation, the important thing is that the country specific errors have disappeared. There is no longer need of the assumption that country specific error is uncorrelated with explanatory variables. i.e time-constant unobserved heterogeneity is no longer a problem. Time-invariant country characteristics (fixed effects), such as geography and demographics, may be correlated with the explanatory variables. The fixed effects are contained in the error term in equation (3.2.1), which consists of the unobserved country-specific effects, αi , and the observation-specific errors, εit . 3.3 GMM as a Method of Estimation As to Baltagi (2008), which this project employs, gives more information and more variability in data since large numbers of data points are available. These special features lead to less collinearity among variables, more degrees of freedom and more efficiency. Furthermore, panel data method better controls heterogeneity among economic variables and allows one to construct and test more complicated behavioral models than pure cross-sectional and time series data. However, looking at our models in equation (3.0.2) it includes the lagged dependent variable as a regressor. From this growth equation, the unobservable country specific effect (αi ) affects the dependent variable lnYit . Due to this relationship, lnYit−1 is a function of αi i.e (Yit−1 ǫit ) 6= 0. This econometric problem accompanied with the omitted variable bias, measurement error, and endogeneity of other regressors, which are common problems in 11 panel data models, leads to inefficient estimates by standard econometric techniques like OLS and FE. The GMM method provides a solution to these problems as well as yielding estimates of unobserved country-specific effects and dummy coefficients for which the usual methods would be inappropriate given the dynamic nature of the regression Bond (2002). There are two types of GMM estimators based on the assumption made and resulting moment restriction, and thus variables used as instruments in the estimation process: the difference GMM and the system GMM. The difference GMM estimator (DGMM) for dynamic panels was introduced by Holtz-Eakin et al. (1988), Arellano and Bond (1991), and Arellano and Bover (1995). It is based on differencing the series to eliminate unobserved country specific effects and use lagged explanatory and dependent variables as instruments, called-internal instruments to avoid autocorrelation problem. So, let’s show how GMM estimation technique alleviates the problems faced in the above standard estimation techniques like OLS and FE. The difference and system GMM estimators can be seen as part of broader historical trend in econometric practice toward estimators that make fewer assumptions about the underlying data-generating process and use more complex techniques to isolate useful information. The difference and system GMM estimators are designed for panel analysis, and embody the following assumptions about the data-generating process: 1) There may be arbitrarily distributed fixed individual effects. This argues against cross-section regressions, which must essentially assume fixed effects are not problems, and in favor of a panel set-up, where variation over time can be used to identify parameters. 2) The process may be dynamic, with current realizations of the dependent variable influenced by past ones. 3) Some regressors may be endogenous. 4) The idiosyncratic disturbances (those apart from the fixed effects) may have individual-specific patterns of heteroskedasticity and serial correlation. 5) The idiosyncratic disturbances are uncorrelated across individuals. In addition, some secondary worries shape the design: 6) Some regressors may 12 be predetermined but not strictly exogenous: even if independent of current disturbances, still influenced by past ones. The lagged dependent variable is an example. The number of time periods of available data, T, may be small (panel is “small T, large N.”) 7) Finally, since the estimators are designed for general use, they do not assume that good instruments are available outside the immediate data set. In effect, it is assumed that: The only available instruments are “internal”based on lags of the instrumented variables. However, the estimators do allow inclusion of external instruments. 3.4 Why Dynamic Panel Analysis? 1) Static panel estimates, as do the OLS models, omit dynamics causing the problem of dynamic panel bias Roodman (2015) and as such do not allow one to study the dynamics of adjustment (Baltagi, 2008). Omitted dynamics means that such models are wrongly specified, because they omit the entire history of the right-hand-side variables (Bond, 2002). 2) In this panel, there are 40 countries (N) that are analyzed over a period of 27 years (T). Accordingly, there are more countries (N) than years (T). Many authors argue that the dynamic panel model is specially designed for a situation where “T” is smaller than “N” to control for dynamic panel bias (Bond, 2002; Baum and Christopher, 2006; Roodman, 2006). 3) The problem of potential endogeneity is also much easier to address in the dynamic panel models than in the static and OLS models that do not allow the use of internally generating instruments. An underlying advantage of the dynamic GMM estimation is that all variables from the regression that are not correlated with the error term (including lagged and differenced variables) can be potentially used as valid instruments (Greene, 2008). 4) Finally, the OLS and static panel estimates do not allow a separate analysis of the short and long-run effects of urbanization on economic growth; hence, an additional advantage of the dynamic panel model is its ability to identify both short run impact and long-run urbanization effects (Baltagi, 2008; Roodman, 2006), which is particularly important for this project. 13 After identifying the dynamic panel model as the most appropriate econometric technique for the estimation, the we had to decide which dynamic panel approach to apply. Notwithstanding that the General Method of Moments (GMM) is the method of estimation of dynamic panel models that provides consistent estimates, one still must decide whether to use: differenceGMM or, system-GMM estimation. Before deciding which GMM method is appropriate to our model, it is rational to show how the GMM works. To begin with, one step GMM estimator was used as it has been shown to result in more reliable inferences Baltagi (2008). The selection of one-step GMM was based on the fact that it ensures consistency and efficiency while dealing with heteroscedasticity and serial correlation. Also, to avoid the potential endogeneity among variables, one-step difference GMM technique is used. Having introduced the various panel data tests and estimation techniques to calculate growth of real GDP per capita and urbanization, we now able to investigate the impact of urbanization on the economic performance. As urbanization, which is the variable of interest in this project, is among the explanatory variables in the theoretical growth regression equation. Yit = αi + γ ′ Yit−1 + θ′ Xit + β ′ Uit + ηi + εit (3.4.1) εit = ηi + vit E(µi ) = E(vit ) = (ηi vit ) = 0 (3.4.2) Where ηi represents unobserved country-specific factors and εit is the error term of the dynamic panel regression model. Turning to the growth determinants, the model includes one period lagged value of economic growth variable, urbanization which is the interest variable and all the other determinants that are represented by a vector of Xit (i.e., gross capital formation, gross expenditure, interest payments, terms of trade, FDI, and some institutional variables). Recall that the one period lagged value of economic growth is also good explanatory variable for conditional convergence. 14 The GMM method provides a solution to these problems using OLS and FE estimation techniques as well as yielding estimates of unobserved country specific effects and dummy coefficients for which the usual methods would be inappropriate given the dynamic nature of the regression. From (3.4.1), ηi is country specific effect. To eliminate the country-specific effect, let’s take the first-differences of equation (3.4.3) below, (Yit −Yit−1 ) = (αi −αi )+γ ′ (Yit−1 −Yit−2 )+θ′ (Xit −Xit−1 )+β ′ (Uit −Uit−1 )+(εit −εit−1 ) (3.4.3) In the presence of the country specific effect αi , it is well known that the OLS estimate of the coefficient on the lagged dependent variable γ is likely to be biased upward since the lagged dependent variable is positively correlated with αi (Blundell and Bond, 1998). By transformation process, the country specific effects αi can be removed as one can see from (3.4.3). A disadvantage of using the fixed effects model is that it uses only the variation within countries and the cross-sectional variation is discarded. In addition, equation (3.4.3) contains a lagged endogenous variable, namely the economic growth. Thus, with a small number of time series periods, the model provides biased and inconsistent estimates even if data from many countries are considered. In contrast to the OLS estimate, the fixed effects estimate of the coefficient on the lagged dependent variable γ is likely to be biased down ward (Arellano and Bond, 1991). They suggest an alternative estimation technique that addresses the presence of the lagged endogenous variable and permits a certain degree of endogeneity in the other explanatory variables. Arellano and Bond (1991) use difference GMM estimator to eliminate the country-specific effect, and then used all possible lagged levels as instruments. Accordingly, we apply Arellano and Bond’s difference equation and the model becomes: ∆Yit = γ ′ ∆Yit−1 + θ′ ∆Xit + β ′ ∆Uit + ∆εit E(∆Yit−1 ∆εit ) 6= 0 15 (3.4.4) where ∆ is a difference operator and the others as defined before. In the first equation above, we got rid of αi , which is correlated with our regressors, but we generated a new endogeneity problem. The second equation above illustrates one of our regressors is related to our unobservables. The solution is instrumental variables. Which instrumental variables? Arellano–Bond suggest the second lags of the dependent variable and all the feasible lags thereafter. This generates the set of moment conditions defined by E(Yi,(t−2) ∆εit ) = 0 E(Xi(t−3) ∆εit ) = 0 ··· E(Ui(t−j) ∆εit ) = 0 We have 27 time period which yield the following set of instruments: t = 27 Yt−25 , Yt−24 , Yt−23 , Yt−22 , Yt−21 , Yt−20 , · · · , Yt−3 , Yt−2 , Yt−1 t = 26 Yt−24 , Yt−23 , Yt−22 , Yt−21 , Yt−20 , · · · , Yt−3 , Yt−2 , Yt−1 t = 25 Yt−23 , Yt−22 , Yt−21 , Yt−20 , · · · , Yt−3 , Yt−2 , Yt−1 .. . ··· t = 10 Yt−8 , Yt−7 , Yt−6 , Yt−5 , Yt−4 , Yt−3 , Yt−2 , Yt−1 t = 9 Yt−7 , Yt−6 , Yt−5 , Yt−4 , Yt−3 , Yt−2 , Yt−1 t = 8 Yt−6 , Yt−5 , Yt−4 , Yt−3 , Yt−2 , Yt−1 t = 7 Yt−5 , Yt−4 , Yt−3 , Yt−2 , Yt−1 t = 6 Yt−4 , Yt−3 , Yt−2 , Yt−1 t = 5 Yt−3 , Yt−2 , Yt−1 t = 4 Yt−2 , Yt−1 t = 3 Yt−1 Assuming that Xit are predetermined in the sense that Xit and εit are uncorrelated, but Xit may be correlated with εi,t−1 . and earlier errors, Xit lagged one period or more are also used as valid instruments. Thus, the relevant 16 moment conditions are: E(Yit−j ∆vit ) = 0 for j ≥ 2t; t = 3, . . . ., T E(Xit−j ∆vit ) = 0 for j ≥ 2t; t = 3, . . . ., T E(Uit−j ∆vit ) = 0 for j ≥ 2t; t = 3, . . . ., T where vit = αi + εit The system GMM estimator combines the standard set of moment conditions in first differences with lagged levels as instruments, with an additional set of moment conditions derived from the equation in levels. The availability of additional moment conditions depends on assumptions made about the correlation between Xit and the country-specific effect αi . Arellano and Bover (1995), it is assumed that the difference of Xit is uncorrelated with the individual effects although Xit and αi can be correlated. Thus, the additional moment conditions for the equation in levels are: E(∆Yit−1 vit ) = 0 (3.4.5) E(∆Xit−1 vit ) = 0 (3.4.6) E(∆Uit−1 vit ) = 0 (3.4.7) the difference GMM dynamic panel estimator uses the following moment conditions: E(Yit−j εit − εi,t−1 ) = 0 forj ≥ 2; t = j. . . T (3.4.8) E(Xit−j εit − εi,t−1 ) = 0forj ≥ 2; t = j. . . T 3.5 (3.4.9) Testing for Serial Correlation We can test these conditions in STATA. In essence, the differenced unobserved time-invariant component should be unrelated to the second lag of the dependent variable and the lags thereafter. If this is not the case, we are back to the initial problem, endogeneity. Again, a bit of math will help us 17 understand what is going on. All is well if ∆ǫit = ∆vit (3.5.1) The unobservable is serially correlated of order 1 but not serially correlated of orders 2 or beyond. But we are in trouble if ∆ǫit = ∆vit + ∆vi(t−1) (3.5.2) The second lag of the dependent variable will be related to the differenced time-varying component ∆ǫit . Another way of saying this is that the differenced time-varying unobserved component is serially correlated with an order greater than 1. 18 Chapter 4 Data Presentation and Analysis 4.1 4.1.1 Descriptive Analysis Introduction Under this section, which is the heart of the investigation, we present and discuss the main results in accordance with answering the objectives of the project. Before we conduct the GMM regressions of economic growth, we examine the summary statistics, intensity of collinearity, the degree of graphical association, and the extent of normality for the economic growth variables and urbanization. Table (4.1) below reports the summary statistics for the dependent and independent variables of the specified growth model. 4.1.2 Summary statistics As the summary statistics show, some of the variables have too many missing values. Per capita GDP, openness, urbanization and energy consumption are the variable with less missing values. As far as economic growth is concerned, the real GDP per capita in PPP (2011 US$) of African countries has on average grown by 4679.43 percent under the period taken into consideration. Compared to their economic status and living condition, this per capita GDP could be taken as typical. When 19 Table 4.1: Summary Statistics Variable Mean Std. Dev. Min Max Obs. PCI overall between within 4679.43 6079.24 5979.606 2536.167 247.4365 642.2806 -13944.28 40015.82 24008.61 25113.19 Capital For overall between within 7.81E+09 1.54E+10 1.23E+10 7.01E+09 -1.35E+07 5.83E+07 -2.09E+10 9.65E+10 5.51E+10 5.80E+10 N = 740 n = 37 bar = 20 Gross Exp overall between within 3.47E+10 6.72E+10 5.68E+10 2.24E+10 2.92E+08 6.60E+08 -6.49E+10 4.26E+11 2.95E+11 1.65E+11 N = 739 n = 37 bar = 19.973 Openness overall between within 74.52915 47.29362 38.14408 29.11954 11.08746 26.89246 -57.91608 531.7374 226.7788 379.4878 N = 998 n = 39 bar = 25.5897 Urban overall between within 40.26608 17.74929 17.60246 3.711669 6.271 9.054963 28.40712 87.366 80.78674 51.92075 N = 1048 n = 39 T = 26.8718 Tourism overall between within 13.69709 13.92816 13.62536 4.65538 0.0009562 0.1245983 -3.797377 67.43046 51.99502 41.66949 N = 686 n = 38 bar = 18.0526 Energy cons overall between within 60.23018 30.2252 29.4626 7.919134 0.0589587 0.3537298 28.42465 98.3426 96.46798 107.7834 N = 1001 n = 39 T = 25.6667 20 N = 1031 n = 39 T = 26.4359 one explores the smallest and the greatest GDP per capita realized in Liberia (247.44) in 1995 and Equatorial Guinea 40015.82 in 2008 respectively. Since 1995 most African countries began a major program of economic reform and liberalization which helped the country to design favorable political, social and economic conditions that enabled the nation to significantly increase and achieve the highest level of GDP growth not only in Africa but also in the world (Wiarda, 2018). Urban Population (% of total) was 6.271% in Burundi in 1990 and still small in 2016 (i.e. 11.2%). However, in 2016, 87.4% population of Gabon live in cities which is by far large number compared to the whole average (40.3). Table (4.2) shows the missing value summary statistics. From the Table, we Table 4.2: Missing values summary statistics Obs < . Obs = . Variable PCI Capital Formation Gross Expenditure Openness Tourism Energy Consumption Urbanization Obs > . 22 313 314 55 367 52 5 Obs < . 1,031 740 739 998 686 1,001 1,048 Unique values Min Max > 500 > 500 > 500 > 500 > 500 > 500 > 500 247.4365 -1.35e+07 2.92e+08 11.08746 .0009562 .0589587 6.271 40015.82 9.65e+10 4.26e+11 531.7374 67.43046 98.3426 87.366 can see that urbanization has more observation than other variables. Data for International Tourism, Receipts (% of total exports) is the least available. International tourism receipts are expenditures by international inbound visitors, including payments to national carriers for international transport and this requires careful national records which are 4.1.3 Measurement of Multicollinearity The dynamic GMM estimation method can help solve the problems of endogeneity, omitted variable bias, serial correlation and heteroskedasticity. 21 However, one crucial pre-estimation test that should be done is the test of the approximate linear relationship among the explanatory variables (i.e. multicollinearity). Pair-wise correlation between regressors is one of the most commonly employed detection method to measure the problem of multicollinearity. The correlation matrix (left for the sake of space) measures the severity of linear relationship among the economic growth explanatory variables. In a nutshell, the outcome reveals the absence of the problem of multicollinearity. The only highest pair-wise correlation occurred between Gross expenditure and capital formation, 0.9466, which is large compared to correlation values of 0.8 which are commonly taken as a Rule of Thumb to conclude the presence of multicollinearity problem. Although the above descriptive analysis portrays the individual association between growth rate of real GDP per capita and its determinants, it neither exposes the statistical significance of the relationship nor guarantees whether this relationship can be maintained when other explanatory variables are simultaneously included. The most dependable and sophisticated econometric analysis will be undertaken in the following subsection. In addition to the pairwise correlation of variables, we can see the ecoTable 4.3: The Correlation between Variables ln ln ln ln ln ln ln PCI CF GE Open Ur Tour EC ln PCI ln CF ln GE ln Open 1.0000 0.6019 0.4734 0.3881 0.5978 0.0719 -0.7606 1.0000 0.9461 -0.1184 0.4012 -0.0595 -0.5279 1.0000 -0.2731 0.322 -0.0197 -0.4767 ln Ur 1.0000 0.4064 1.0000 -0.0594 -0.0745 -0.2879 -0.5552 ln Tour ln EC 1.0000 -0.1785 1.0000 nomic growth and urbanization trend during the period we have taken into consideration. The scatter plots (4.1 ) and (4.2) do not show whether there exist relationship between the two variables though both of them are grow22 40015.819 40015.819247.43654 40015.819247.43654 40015.819247.43654 Benin Botswana Burkina Faso Burundi Cabo Verde Cameroon Central African R. Comoros Congo, Dem. Rep. Congo, Rep. Cote d'Ivoire Djibouti Egypt, Arab Rep. Equatorial Guinea Eritrea Gabon Gambia, The Ghana Kenya Lesotho Liberia Libya Malawi Mali Morocco Namibia Niger Senegal Seychelles Sierra Leone South Africa Sudan Tanzania 1990 40015.819 247.43654 Pre Capita GDP 40015.819247.43654 Angola 247.43654 Algeria Togo 1990 2000 2010 Uganda 2020 1990 2000 2010 Zambia 2020 1990 2000 2010 2000 2010 2020 1990 2000 2010 2020 1990 Zimbabwe 2020 1990 2000 2010 2020 Year Figure 4.1: Scatter plot for per capita GDP 23 2000 2010 2020 100 Angola Benin Botswana Burkina Faso Burundi Cabo Verde Cameroon Central African R. Comoros Congo, Dem. Rep. Congo, Rep. Cote d'Ivoire Djibouti Egypt, Arab Rep. Equatorial Guinea Eritrea Gabon Gambia, The Ghana Kenya Lesotho Liberia Libya Malawi Mali Morocco Namibia Niger Senegal Seychelles Sierra Leone South Africa Sudan Tanzania 50 0 100 50 0 50 100 0 Urabanization 100 0 50 100 0 50 Algeria 1990 Uganda Zambia 2010 2020 1990 2000 2010 2020 1990 2000 2010 2020 Zimbabwe 0 50 100 Togo 2000 1990 2000 2010 2020 1990 2000 2010 2020 1990 2000 2010 2020 1990 2000 2010 2020 Year Figure 4.2: Scatter plot for Urbanization ing during the period. Therefore, we can carry out the major econometric estimation as most of the variables are normally distributed. 4.2 Econometric Estimation Results 4.2.1 Pooled-OLS and Fixed Effects Result In this part of the analysis we discuss the determinants of Per capita GDP (measure of economic growth). The output in Table (4.4) below will be reviewed a bit more carefully. First, one can see that the F-test is statistically significant, which means, the model is working. The adjusted R2 of 24 0.937 means that more 90% of the variance of economic growth is accounted for by the model and in this case, the adjusted R2 indicates that approximately about 0.927 of the variability of economic growth is accounted for by the model, even after considering the number of predictor variables in the model. The coefficient for each of the variable indicates the amount of change one could expect in economic growth given a unit change in the value of that variable, given that all other variables in the model are held constant. Urbanization, the interest variable, (β = −0.005) in OLS and −0.008 in FE estimations and significant in FE estimation. However, urbanization is significant in level with (−0.826). The striking result is that the coefficient of urbanization is negative which is unexpected. The lagged dependent variable is significant and its sign is expected. The magnitude and the sign shows the conditional convergence of poor countries have the tendency to grow faster than rich countries(Barro and Sala-i Martin, 1997). As it is mentioned in the methodology section, the fact that urbanization is measured with error and urbanization and economic growth are measured endogenously creates attenuation and may bias the Pooled-OLS estimates downward. However, one can solve these problems using instrumental variable method of estimation. These instruments must be important factors in accounting for the variation in urbanization rates that one observes, but have no direct effect on economic growth. However, before considering the instrumental variable method of estimation,the fixed effects results will be discussed. As far as the fixed effect result is concerned, the model controls for all timeinvariant differences between the individuals, so the estimated coefficients of the fixed effects models cannot be biased because of omitted time-invariant characteristics. One side effect of the features of fixed effects models is that they cannot be used to investigate time-invariant causes of the dependent variables. Technically, time-invariant characteristics of the countries are perfectly collinear with countries dummies. Substantively, fixed-effects models are designed to study the causes of changes within a person (or countries). A time-invariant characteristic cannot cause such a change, because it is constant for each person. 25 Table 4.4: Pooled-OLS and Fixed Effects Result with logarithm and levels VARIABLES OLS FE ln PCI(lagged) 0.987*** (0.00383) 0.0160*** (0.00489) -0.00812 (0.00518) 0.0249*** (0.00500) -0.00525 (0.00520) 0.00205 (0.00136) -1.38e-05 (0.000106) 0.792*** (0.0218) 0.0172** (0.00786) 0.0922*** (0.0177) 0.0153* (0.00882) -0.142*** (0.0353) 0.00704** (0.00277) -0.000257 (0.000371) ln Capital Formation ln Gross Expenditure ln Openness ln Urbanization ln Tourism ln Energy Consumption PCI(lagged) 498 0.940 -0.422** (0.198) 498 0.998 498 0.947 498 0.998 Openness Urbanization Tourism Energy Consumption ∗ 0.955*** (0.0170) 1.03e-08** (5.06e-09) -1.83e-09 (1.09e-09) 1.202 (0.860) 1.023 (8.092) -0.0142 (1.855) -3.030 (2.885) 294.3 (505.1) -0.120** (0.0576) Gross Expenditure Observations R2 FE 1.000*** (0.00341) 7.13e-10 (1.58e-09) -1.67e-11 (3.41e-10) 0.963*** (0.238) -2.471*** (0.826) -0.696 (0.811) -2.782*** (0.524) 277.0*** (59.94) Capital Formation Constant OLS t statistics in parentheses , p < 0.05,∗∗ p < 0.01,∗∗∗ p < 0.001 26 As far as our findings are concerned, urbanization and tourism become statistically significant which are not in OLS estimation. σu and σe are .238 and .032 respectively. F test that all ui = 0 : F (35, 455) = 8.01 with P rob > F = 0.0000. Four countries due to country specific problems and and more than 500 observations are dropped due to missing values. The correlation of (Ui , βX) of F E result is 0.5329 and it indicates that the errors are correlated with the regressors in the fixed effects model. 13.6% percent of the variance is due to differences across panels. (ρ = .65) is known as the intra-class correlation. (σu = 262.04) is the standard deviation of residuals within groups and (σe = .032) is the standard deviation of residuals (overall error term) εit . The problem in this fixed effect is that still our interest variable, urbanization is not significant. Finally, To see if time fixed effects are needed when running a FE model, the joint test was computed to see if the dummies for all years are equal to 0. Therefore, we reject the null that all year’s coefficients are jointly equal to zero. Thus, no need of time fixed effects. 4.2.2 GMM Estimation Results In this section we investigate the impact of urbanization on economic growth using more advanced econometric estimation techniques which were provided in the previous sections. To study the link between urbanization and economic growth, we estimate the specified economic growth model. From Table (4.5) regression results, as matter of novelty, the study came up with very interesting evidence that ranges from the presence of conditional convergence in African countries with all other statistically significant regressors. We again confirm that the sign of urbanization is negative for Per capita GDP in Africa. A recent empirical work in urbanization of Africa supports of our findings. By Urban areas, growing both in population and in land cover, pose threats to the integrity of the continent’s ecosystems and biodiversity but their growth also create opportunities for conservation. The burgeoning urban populations, especially in Sub-Saharan Africa, increase the strain on already insufficient infrastructure and bring new governance chal27 Table 4.5: One step system GMM and One step difference GMM VARIABLES ln PCI(lagged) ln Urbanization ln Capital formation ln Gross Expenditure ln Openness ln Tourism ln Energy Consumption SGMM DGMM 0.992*** (0.0141) -0.0972*** (0.0244) 0.0383** (0.0176) -0.0315* (0.0174) 0.0632*** (0.0117) 0.0130** (0.00549) 0.000838** (0.000415) 0.601*** (0.0336) -0.299*** (0.0645) -0.00541 (0.0109) 0.208*** (0.0236) 0.0612*** (0.0128) 0.00296 (0.00418) -0.00123** (0.000561) 498 36 461 33 Observations Number of Country 1 Arellano-Bond test for AR(1) in first differences: z = −5.94 P r > z = 0.000 2 Arellano-Bond test for AR(2) in first differences: z = −0.43 P r > z = 0.664 3 Sargan test of overid. restrictions: χ234 = 93.14 P rob > χ2 = 0.000 4 The notes are for SGMM and we can do the same for the DGMM. 28 lenges (Güneralp et al., 2017). The coefficient of the first lag of per capita GDP is positive and significant at 1%, it confirms that countries with lower per capita income grew faster than countries with higher per capita income. The result is in line with the theoretical prediction that was dealt in the previous chapters as well as the empirical investigations of Barrios et al. (2003) and others. Table (4.5) also shows difference GMM estimation results, where we use the lagged and difference variables as instruments for country level economic growth rates. And, the result shows that a a negative impact of urbanization on economic growth. As model specification part, the Sargan test of over identifying restriction tests whether the instruments applied for the DGMM growth regressions are correlated with the error terms. Hence, the validity of the SGMM and DGMM growth regression results can be measured by the Sargan test value. The Sargan test of over-identifying restriction, as we point in the regression table notes above, supports the correctness of the SGMM and DGMM methods of estimations. For instance, the Wald test for SGMM χ252 = 51.13, ρ > χ252 = 0.508 indicates the regression coefficients are not jointly zero. This is due to the fact that the research failed to reject the null hypothesis which hypothesizes the regression coefficients are not jointly zero. The lagged levels of these variables are weak instruments for equations in difference especially in small samples. However, the sample is not small and all the results are significant using DGMM and SGMM as well. Finally, the Arellano-Bond test for zero autocorrelation in first-differenced error test shows that H0 No autocorrelation. 29 Chapter 5 Concluding Remarks In this project, we explored the impact of urbanization on the economic growth of 40 African countries from the period 1990 to 2016, the impact of urbanization on economic growth of those African countries is examined. To achieve this, we use several statistical/econometric approaches. As far as economic growth is concerned, the average Per capita real GDP in PPP terms is 4679 for those countries in our model under the period taken into consideration. Compared to their economic status and living condition, this per capita GDP could be taken as typical but has to be invoked and stimulated as some of the countries show positive steps (Güneralp et al., 2017). Different theoretical models and empirical investigations have demonstrated how urbanization can generate incorrect/correct signals to economic development and thereby result in distortion which is due to unavailability of data and methodological weaknesses. These and other related issues caught our attention towards studying African urbanization and its effect on the continent’s economic progress. And finally, we draw the following main conclusions. First, the overall growth of urbanization is highly important and has been shown to be quite robust to the inclusion of potentially relevant covariates in regression as well as in different estimation methods though our finding shows that urbanization has negatively correlated with economic growth in 30 the region. Second, there is the presence of conditional convergence in Africa. Therefore, poor countries have the tendency to grow faster than rich countries. Those countries with the lowest growth rates are those who did not urbanize. Therefore, we recommend the following points: • During city planning it should be ensured that adequate infrastructures and other utilities are available to support the urban residences. This eliminates some of the negative aspects of urbanization and possible to make urbanization having great impact on economic growth and industrialization through agglomeration economic. • This finding generally upholds the theoretical assertion of positive relationship between economic performance and openness, International Tourism, Receipts (% of total exports), and grross expenditure (including interest payment). However, urbanization, capital formation and Renewable Energy Consumption (% of total final energy consumption) were found to stimulate economic growth in Africa. • Thus, policy measures that enhance the growth of urban areas over time which avoid the negative effects and promote open trade has the potential of significantly stimulating economic growth in Africa. In other words, sound urban development policies that support economic openness with greater emphasis on liberalization policy since the continent stands to gain from this policy stance better off to be called for. 31 Bibliography Acemoglu, D. and Robinson, J. A. (2010). Why is africa poor? Economic history of developing regions, 25(1):21–50. Arellano, M. and Bond, S. (1991). Some tests of specification for panel data: Monte carlo evidence and an application to employment equations. The review of economic studies, 58(2):277–297. Arellano, M. and Bover, O. (1995). Another look at the instrumental variable estimation of error-components models. Journal of econometrics, 68(1):29– 51. Baltagi, B. (2008). Econometric analysis of panel data. John Wiley & Sons. Barrios, S., Bertinelli, L., Strobl, E., and Teixeira, A. C. (2003). Agglomeration economies and the location of industries: a comparison of three small european countries. Barro, R. J. and Lee, J.-W. (1993). International comparisons of educational attainment. Journal of monetary economics, 32(3):363–394. Barro, R. J. and Sala-i Martin, X. (1997). Technological diffusion, convergence, and growth. Journal of Economic growth, 2(1):1–26. Baum, C. F. and Christopher, F. (2006). An introduction to modern econometrics using Stata. Stata press. Binswanger, H. P. and Townsend, R. F. (2000). The growth performance of agriculture in subsaharan africa. American Journal of Agricultural Economics, 82(5):1075–1086. 32 Blundell, R. and Bond, S. (1998). Initial conditions and moment restrictions in dynamic panel data models. Journal of econometrics, 87(1):115–143. Bond, S. R. (2002). Dynamic panel data models: a guide to micro data methods and practice. Portuguese economic journal, 1(2):141–162. Gollin, D., Jedwab, R., and Vollrath, D. (2016). Urbanization with and without industrialization. Journal of Economic Growth, 21(1):35–70. Greene, W. H. (2008). The econometric approach to efficiency analysis. The measurement of productive efficiency and productivity growth, 1(1):92–250. Güneralp, B., Lwasa, S., Masundire, H., Parnell, S., and Seto, K. C. (2017). Urbanization in africa: challenges and opportunities for conservation. Environmental Research Letters, 13(1):015002. Henderson, V. (2003). The urbanization process and economic growth: The so-what question. Journal of Economic growth, 8(1):47–71. Holtz-Eakin, D., Newey, W., and Rosen, H. S. (1988). Estimating vector autoregressions with panel data. Econometrica: Journal of the Econometric Society, pages 1371–1395. Koopmans, T. C. et al. (1965). On the concept of optimal economic growth. Lucas, Jr, R. E. (2004). Life earnings and rural-urban migration. Journal of Political Economy, 112(S1):S29–S59. McCoskey, S., Kao, C., et al. (1998). A panel data investigation of the relationship between urbanization and growth. United State Naval Academy and Syracuse University, pages 1–24. Moomaw, R. and Shatter, A. M. (1993). Urbanization as a factor in economic growth. Journal of Economics, 19(2):1–6. Polèse, M. (2005). Cities and national economic growth: a reappraisal. urban Studies, 42(8):1429–1451. 33 Romer, P. M. (1986). Increasing returns and long-run growth. Journal of political economy, 94(5):1002–1037. Roodman, D. (2006). How to do xtabond2: An introduction to difference and system gmm in stata. Roodman, D. (2015). xtabond2: Stata module to extend xtabond dynamic panel data estimator. Sarker, S., Khan, A., and Mannan, M. M. (2016). Urban population and economic growth: South asia perspective. European Journal of Government and Economics, 5(1):64–75. Solow, R. M. (1956). A contribution to the theory of economic growth. The quarterly journal of economics, 70(1):65–94. Swan, T. W. (1956). Economic growth and capital accumulation. Economic record, 32(2):334–361. Wiarda, H. (2018). Civil society: The American model and third world development. Routledge. 34 Appendix Table 5.1: Dynamic panel-data estimation with levels Capital Formation Gross Expenditure (1) GMM 2.74e-09 (0.62) 1.69e-08∗∗∗ (14.50) Openness 3.814∗∗∗ (7.87) Urbanization 65.03∗∗∗ (16.46) Tourism 12.83∗∗∗ (6.95) Energy Consumption -32.66∗∗∗ (-19.03) Constant 2389.7∗∗∗ (10.45) 498 Observations t statistics in parentheses ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 35 Table 5.2: Missing summary statistics for Dynamic panel-data estimation country Freq. Algeria Angola Benin Botswana Burkina Faso Burundi Cabo Verde Cameroon Comoros Congo, Dem. Rep. Congo, Rep. Cote d’Ivoire Egypt, Arab Rep. Equatorial Guinea Eritrea Gabon Gambia, The Ghana Kenya Lesotho Liberia Malawi Mali Morocco Namibia Niger Senegal Seychelles Sierra Leone South Africa Sudan Tanzania Togo Uganda Zambia Zimbabwe Total Percent Cum. 11 1 21 21 11 19 7 21 10 11 20 8 21 2 6 11 12 10 21 9 10 14 20 21 21 10 20 1 15 21 21 21 21 21 1 7 2.21 0.2 4.22 4.22 2.21 3.82 1.41 4.22 2.01 2.21 4.02 1.61 4.22 0.4 1.2 2.21 2.41 2.01 4.22 1.81 2.01 2.81 4.02 4.22 4.22 2.01 4.02 0.2 3.01 4.22 4.22 4.22 4.22 4.22 0.2 1.41 2.21 2.41 6.63 10.84 13.05 16.87 18.27 22.49 24.5 26.71 30.72 32.33 36.55 36.95 38.15 40.36 42.77 44.78 49 50.8 52.81 55.62 59.64 63.86 68.07 70.08 74.1 74.3 77.31 81.53 85.74 89.96 94.18 98.39 98.59 100 498 100 36