Education and economic growth in emerging and developing countries : educational attainment and GDP

(1)

Education and Economic Growth in Emerging and

Developing Countries

Educational attainment and GDP

___________________________________________________________________________

Bachelor Thesis

Written by: Koen Maas (10665951) Supervisor: Oana Fortuna

Date: June 2016

Abstract

The goal of this research paper is to investigate the possibility of a significant relationship between educational attainment and economic growth, by doing a cross-section data analysis. The dataset consists of 55 emerging or developing countries as defined by the International Monetary FUND (IMF). From the regression analysis it can be concluded that for this group of countries there is no significant effect of educational attainment on economic growth.

(2)

1. Introduction

It is often assumed in economics that education is a positive driver of economic growth in countries in the long run (Lucas, 1988). Improving the education of people increases their knowledge and skills. These skills and knowledge can be seen as a part of human capital, besides health for example. And since increasing human capital has a direct effect on

economic growth, improving education can increase economic growth. Growth in countries is mostly measured using Gross Domestic Product or GDP in both the short-run as well as in the long-run. In the short-run growth can only increase aggregate demand which will cause a higher level of GDP if there is spare capacity to produce more in the economy.The focus in this research paper however is growth in the long-run since it encompasses a time period of 30 years in which the aggregate supply can increase as well as the aggregate demand. In the long run the productive capacity can be altered in reaction to a possible rise in demand and therefore increase GDP levels in countries.

In developing and emerging economies education levels differ enormously from advanced economies. Therefore the United Nations (2016) made it one of their Millennium Development Goals (MDGs) to close this gap as much as possible, aiming to provide universal primary education. The MDGs were set for the year 2015 and the results are that enrolment in primary school has increased from 83 to 91 percent. But there are still 57 million children of primary school age not in school (United Nations, 2016). More educated children should ultimately lead to a more educated workforce. So this means a higher quality of human capital and in turn increase the GDP per capita, which ultimately could lead to better living conditions. Barro (2001) for example, points out that human capital

accumulation is an important part of the development process. It is therefore interesting to see if education has actually had a significant positive effect on GDP growth in these emerging economies over the past few decades. If this is the case then the investment in education can more easily be justified by governments as a sound economic policy. There are however two ways to improve education.

Improvement in education can be quantitative, by getting more children to school for example, and qualitative by increasing test scores. Quantitative educational improvement can be measured by years of attainment (Barro, 2013). Hanushek & Wößmann (2007) stress the importance of improving the quality of education and point out that only spending more money on education may not be enough. Hanushek & Wößmann (2007) also point out however that qualitative education data for developing countries is only available for a small

(4)

selection of countries. This means that for this research paper the qualitative component will only be discussed in the literature and not in the data analysis. The focus of this research paper lies on quantity of education and its effect on growth.

Summing up the above, the research question of this paper will be: Does educational attainment have a significant effect on economic growth in developing countries? To address this objective a cross section analysis will be done, estimating the effect of educational attainment on average long-run growth over the period 1980-2010 for 55 developing and emerging countries as defined by the IMF (2015). Furthermore, a number of control variables that affect economic growth will be looked at when explaining the effect of educational attainment in these countries. These include GDP at market prices in 1980, fertility rates and investment ratio.

This paper is set out as follows. Section 2 will provide a review of the relevant literature on this topic. In section 3 the data and their respective sources will be discussed. Also, a detailed explanation of the methodology will be provided. Subsequently, the result will be analysed in section 4. The last part, section 5, provides the conclusion

2. Literature review

Before doing a cross country analysis of the effect of educational attainment on growth, it is important to understand how education and human capital affect growth and how different economic theories and previous literature model economic growth in general. A review of the existing literature on measuring educational improvement is also necessary to contextualise the findings of this research paper.

2.1 Economic models on growth and human capital

Barro (2013) explains that different growth theories complement each other in explaining growth of countries, rather than compete. Endogenous growth models help to understand why countries can continue to grow in the long run in spite of diminishing returns on physical and human capital. Endogenous growth models are relatively new growth theories, first thought up in the 1980’s. These models differed from the previous exogenous models used by Solow (1957) for example, in the sense that they tried to define the growth variable more clearly and remove any obscurity. Exogenous models used the effects of growth but didn’t clearly

explain the determinants of economic growth. So endogenous models looked at the determinants of the savings rates and technological progress, rather than just using those

(5)

variables and giving no explanation where they came from. This uncertainty or vagueness of variables was often the point exogenous models were criticized on. One of the major

contributors of the endogenous growth models is Romer (1990), says Barro (2013). Romer rejects the assumption of exogenous models that countries’ growth rates converge to a steady state (1986, 1994). Romer (1986) attributes the difference between the possible growth countries to the difference in knowledge. This accumulated knowledge can be seen as capital (human) and therefore explain increased production in countries with higher knowledge growth. Besides this explanation for increased production, Romer (1986) also mentions decreasing marginal returns of knowledge and externalities, external effects created by this new acquired knowledge of individuals.

While Romer(1986) only mentions knowledge, which can be inferred as a component of human capital, Lucas (1988) specifically focusses on the whole concept of human capital. Contrarily to Romer (1986), Lucas (1988) uses the neoclassical model, also known as the Solow-Swan model, and not Endogenous growth theory. Lucas (1988) effectively extends the neoclassical model by a human capital component. He argues that human capital affects both labour and physical capital and therefore combines both measures of capital into one in his model. But as mentioned previously by Barro (2013), this doesn't necessarily have to lead to conflicting conclusions. Barro (2013) points out that neoclassical growth theory for example can help explain why one country can grow faster than the other.

2.2 Measures of education

A number of different researches stress the importance of differentiating education into quantity and quality measures (Barro 2013, Hanushek & Wößmann 2007, Lee and Barro 2001). Education can be measured by school attainment rates, so how many children go to and stay in school (quantitative). But education can also be measured by recording the school

test results of children on different subjects/areas and comparing those (qualitative). As

previously mentioned this paper will focus on the quantitative measure of education and only discuss the qualitative measure here in the review of literature.

A researcher that has done a lot of research on the relation between school attainment rates and growth is Robert Barro. He, together with his colleague Jong-Wha Lee, constructed his own database with educational quantity measures for over 140 countries (Barro & Lee, 2013). Barro & Lee (2013) explain that earlier empirical studies used literacy rates or school enrolment ratios to explain to measure the relationship between educational attainment across

(6)

countries. Previous researchers mainly used these measurements because the data on literacy rates and school enrolment ratios is widely available for a range of different countries. Someone is defined as being literate by the World Bank (2016) if they can write a short statement about every day and can understand what they wrote. The World Bank also includes numeracy in their literacy ratio calculations, so if a person can make simple calculations. School enrolment ratios are simply the percentage of people enrolled in a particular level of education relative to the total population. Barro and Lee (2013) explain that they constructed this database partly because measuring educational attainment with these two variables does not give a good representation of human capital stock available for production. Barro (2013) uses this new database to do a panel data analysis of a group of countries and finds that for males, increases in educational attainment have a significant effect on growth. For females this effect is found not be significant. Barro (2013) therefore concludes that female potential is not fully utilized. In the next section of this paper a more in depth look will be given into the construction of this relatively new database from Barro and Lee (2013) and how it will be implemented in this research.

Hanushek & Kimko (2000) say that this new way of measuring human capital by Barro and Lee (2013) is an improvement from using the simple measures of literacy rates or school enrolment ratios as measure for educational attainment. However, Hanushek & Kimko (2000) also explain that most of the previous studies they have looked at, only used some form of quantitative measure. It is important to note here that Barro and Lee have begun constructing the database before the year 2000 and keep updating it. Hanushek & Kimko (2000) find this measure of human capital to be inadequate and point out that when analysing growth across countries the main components of human capital are math and science skills, so qualitative measures. If you can better identify the cause of the differences in the quality of the work force in a country then the differences in growth can be better explained. Hanushek & Wößmann (2007) said the following about educational quality: “We have come to

conclude that educational quality – particularly in assessing policies related to developing countries – is THE key issue.” Educational quality is measured by gathering internationally comparable test scores from children on a number of different subject such as math, science and reading (Barro and Lee, 1996). An example of these standardized test scores are the PISA scores. PISA stands for Programme for International Student Assessment and is a worldwide study from theOrganisation for Economic Co-operation and Development (OECD) that tests and records math, reading and science performance scores from 15 year olds across different countries, mainly member states (OECD, 2016). Hanushek & Wößmann

(7)

(2007) used these scores to compare educational quality and education spending across a number of countries and found that there is no clear relationship between more educational spending and improving PISA scores.The reason researchers that do cross country studies on education and economic growth do not use qualitative education data more often is that they are constrained in the information that they can use. The data must be accurate, consistent and available for a considerable period of time for a large group of countries (Barro and Lee, 1996).

3. Data & Methodology

3.1 Variables and Data

The goal of this paper is to estimate whether there is a relationship between educational attainment in 1980, and average economic growth over the period 1980-2010. The region of interest are emerging and developing economies. Excluding countries for which data is incomplete (most commonly GDP and educational attainment), the sample consists of the following 55 countries: Algeria, Argentina, Bahrain, Barbados, Bolivia, Botswana, Burundi, Cameroon, Chile, Colombia, Democratic Republic of the Congo, Costa Rica, Côte d'Ivoire, Dominican Republic, Ecuador, Egypt, El Salvador, Fiji, Gabón, The Gambia, Ghana, Guatemala, Honduras, India, Indonesia, Iran, Jamaica, Jordan, Kenya, Lesotho, Malaysia, Mauritius, Mexico, Morocco, Nepal, Nicaragua, Niger, Pakistan, Panama, Papua New Guinea, Peru, Philippines, Rwanda, Saudi Arabia, Senegal, South Africa, Sudan, Swaziland, Thailand, Togo, Trinidad and Tobago, Turkey, Uruguay, Venezuela, and Zimbabwe.

Now follows a discussion of all variables and their respective data sources. All variables are subject to logarithmic transformation such that all estimated coefficients can be interpreted as elasticities, i.e a 1% change in an independent variable will cause a change in the dependent variable, which is average growth over the period 1980-2010, calculated as follows: (𝑙𝑛(𝐺𝐷𝑃2010) − 𝑙𝑛(𝐺𝐷𝑃1980))/31

Each value is multiplied by a hundred, such that all coefficients are interpreted as elasticities, given in a basepoint value. This does not affect results, but does allow for a clear interpretation of the coefficients.

(8)

3.1.1 Educational attainment

The central independent variable in this study will be educational attainment as a percentage of the total population of 25 years and over from the database of Barro and Lee (2013). Barro and Lee (2013) used survey observations as a starting point for their educational attainment estimations. They then used school enrolment ratios and the population structure of the country to estimate the flow of new school entrants. Data from this database is available every five years from 1950 till 2010. Due to the availability for the other variables used in this research, the choice was made to take 1980 as a baseline year and therefore use the data from this year in database. The highest level someone has attained in education in a country is divided into four categories: no schooling, primary, secondary, tertiary. In the database there is also a distinction made between the total percentage of the population having had some level of education and the one actually completing the education. In this study the average years of schooling from primary, secondary, tertiary levels of education will be looked at in the following chapter. Also the average total years will be used to look if they have a significant effect on growth in the period 1980-2010. The database also provides the total population in every country at 5 year intervals and provides the average years of schooling in total and per attainment level. Barro (2013) found the secondary and tertiary level values to be significant in relation to growth, so it interesting to see if for this sample similar results will be found.

3.1.2 Control variables

Other variables that were found to have significant effects on economic growth will be discussed here. These variables are all extracted from the World Bank database and all of these control variables are converted into logarithmic values and measured in the baseline year 1980, unless stated otherwise. Firstly, the initial level of GDP at market prices in 1980 should have a negative effect on growth (Sala-i-Martin, 1997). Higher initial income implies lower marginal output, thus a path of slower growth. This relation comes from the the neoclassical model which contains the convergence principle. There is a difference to be made between absolute and conditional convergence. Absolute convergence means that in the long run, GDP per capita converges to the same balanced growth in all countries (Sorensen and Whitta-Jacobsen, 2005). This means that it possible for developing countries to

experience higher growth than advanced economies because they have a lower level of GDP initially. Conditional convergence only means that GDP per capita converges to a country

(9)

specific growth path, which is determined by the characteristics of that country. (Sorensen and Whitta-Jacobsen, 2005). Mankiw, Romer and Weil (1992) find evidence of conditional convergence in a cross country study they did using data from 121 countries between 1960 and 1985. The existence of conditional convergence means that every country has different growth potential and that developing countries may not be able to reach higher levels of GDP because their balanced growth path is not as high as for advanced economies. This could imply that improving human capital to raise growth levels may not be possible to the extent of that of the advanced economies. Besides this first control variable, Barro (2013) lists a number of different control variables that affect economic growth:

Firstly, International Openness is thought to be significant factor in economic growth (Sachs and Warner, 1995). It increases competition and efficiency which in turn increases growth. International openness is measured here by dividing total trade (imports plus exports) by GDP. This is simply defined by the world bank as: Trade as a percentage of GDP.

The next control variable to be included in the model estimation is the Fertility Rate. A higher fertility rate means more children per woman and this could affect economic growth in the long run. Research by de la Croix and Doepke (2003) for example, indicates that poor parents may have more children and invest little in education. This in turn lowers average education and subsequently growth. Although, a higher fertility rate may lead to more school going children and therefore a higher average years of schooling, which can cause

endogeneity in the model.

Rule of Law is included because it is an institutional quality measure that captures the

perception of people to which extent rule of the society are followed. This is a good

measurement for the quality the police courts and crime and violence levels. High levels of crime and a police force that is not capable of dealing with this, is for example a way for Rule of law to negatively affect economic growth. The earliest accessible year is 1996 from the World Governance Indicators (WGI) of Kaufmann et al. (2010). This 1996 variable is used since it is the earliest publicly available measure and moves slowly over time. Meaning that there is very little evidence that the WGI indicators averages change over time (Kaufmann et al., 2011). The 1996 values can therefore safely be used as a control variable to compare economic growth across countries over time. All these variables measure on a scale from -2.5 to 2.5. Hence, to remove negative values we simply rescale by adding 2.5 to every

observation, resulting in a possible range of 0 to 5. No original observation is exactly -2.5, so no new value will be zero. Now the natural logarithms can be calculated. For all these

(10)

Investment Ratio is included because securing growth in the long run requires that

governments and businesses in countries keep investing in new technologies and human capital to keep improving productivity and competition of the economy. This way the GDP can continue to grow. Sala-i-Martin et al. (2004) for example, finds that the investment price is a significant determinant in long term economic growth. The investment ratio is estimated using the gross capital formation as a share of GDP. The gross capital formation is the term used by the world bank and was previously called gross domestic investment. So this includes private and public investment.

Lastly the growth of Terms of Trade is included as a control variable. The value of traded goods determines how many goods are traded, and thus have an impact on GDP in all countries. This means that when Terms of Trade improves the economic growth also likely to rise (Bleaney and Greenway, 2001). Terms of Trade growth is calculated by the following formula: (export values/import values) *100.

One variable that is included by Barro (2013) is real GDP per capita in 1980. This variable will also be included in the OLS regression, however when this variable is added, the variable GDP at market prices in 1980 and the fertitlity rate will be dropped. This is done since Fertlity is a measure of population growth and real per capita GDP would otherwise not add a new explanatory control variable to the regression and cause multicollinearity problems. This also opens the possibility to see which GDP control variable has highest significance in explaning GDP growth.The partial relationship that Barro (2013) finds is therefore first explained by two variables and then one in the OLS regression.

Barro (2013) also included the Government consumption, and the Inflation Rate, as control variables but these are not included in this model. Government consumption data is available but this variable was then modified by Barro (2013) by subtracting government military and education spending. Data is not available for enough countries to make these modifications and include this variable in the control model. The Inflation Rate (based on consumer prices) is also not included in this model. Sala-i-Martin et al., 2004) find the inflation rate to have a weak relation to growth in the long run. Real per Capita GDP is excluded from the regression since initial GDP at market prices in 1980 and Fertility Rate are included already.With all the variable that are used in the regression now explained, it is useful to look at the summary of statistics of those variables.

(11)

3.2 Summary of statistics

Table 3.2 shows the summary statistics for all variables including the dependent variable growth.

Table 3.2 Summary Statistics

The 55 observations are the countries in this dataset. Firstly, GDP growth shows an evenly spread distribution with a minimum growth of -13% and a maximum growth of 13%. The country with the highest average long run growth is Colombia with 13% growth. However Mexico has seen a negative growth of -13% over the same time period. This shows that the sample of countries is diverse in growth rates, but the mean of 2.2% shows an overall growth of GDP in these countries. These differing growth rates however do seem to point out that there is an absence of absolute convergence. This is because all these countries are

considered to be emerging or developing countries, so if absolute convergence would be

Variable Obs Mean Std. Dev. Min Max

Totalschool 55 3.362364 1.75843 .46 6.98 Primary 55 2.470364 1.362173 .4 5.87 Secondary 55 .7781818 .4578172 .05 1.69 Tertiary 55 .1147273 .1029357 .01 .5 TotalschoolFemale 55 2.799091 1.970071 .09 6.9 PrimaryFemale 55 2.117455 1.527587 .06 5.86 SecondaryFemale 55 .6023636 .4754295 .02 1.63 TertiaryFemale 55 .0792727 .0930721 0 .51 TotalschoolMale 55 3.910182 1.699991 .83 7.05 PrimaryMale 55 2.820364 1.275706 .59 5.88 SecondaryMale 55 .9401818 .5032064 .08 1.92 TertiaryMale 55 .1503636 .1289123 .01 .52

GDP1980 55 4.11e+10 7.85e+10 2.73e+08 4.70e+11

GDPcap1980 55 2929.049 4002.197 178.8503 21320.7 Fertility 55 5.424927 1.539375 2.004 8.448 Ruleoflaw 55 2.118957 .6929011 .5681931 3.548292 Openness 55 70.63206 45.31722 11.54567 239.3489 Investmentratio 55 24.66951 8.600409 5.624008 47.86489 TermsOfTradegrowth 55 -.2744435 1.778 -6.121131 3.849286 GDPgrowth 55 2.280216 4.768086 -13.20698 13.11355

(12)

present for this group of countries it is expected to see high growth rates across the whole group and certainly not see negative growth rates over this long time period.

The summary of the educational attainment variables is also given in table 3.2, this includes the average years in primary, secondary and tertiary education plus the total average years of education. What can be concluded from these statistics is that not many people have had an education on a high level for many years. Which leads to most of the people in these developing countries having primary education as their highest attained level. This can be seen in table 3.2 by looking at the means for all separate education levels and the total. The mean of total (male and female) average years of primary education is 2.47 years and the mean of the total average years of education is 3.36 years, while the means of secondary and tertiary are below one with 0.78 and 0.11 average years respectively. This means the average of higher levels of education is low compared to the primary education statistic and that people in these developing countries often only have had primary education and nothing more. When the education statistics are seperated into male and female caterories, it becomes clear that females have less average years of schooling then males. For females the mean of average years of education in total is 2.799091 years and for males this 3.910182 years. This means that males have on average one more year of schooling compared to females in these 55 countries. This is mainly due to the differences in higher levels of education. Males have .9401818 average years in secondary education and .1503636 average years in tertiary education, while for females these numbers are .6023636 and .0792727 average years respectively. This means that males have a higher average education level than females in these countries.

It also very useful to look at some simple graphs like scatter plots before doing a complete regression analysis. These graphs can sometimes already show a relation between the variables that are being used in the regression. The four scatterplots of GDP growth and average years of education of the total population on the following pages in figures 3.1-3.4, reveal that there is no quickly observable positive or negative relation between growth and educational attainment. That is, when a country has higher average in years of education in 1980 than another country, the GDP growth over the next 30 years is not necessarily higher as well. Colombia (COL) and Mexico (MEX) are a good example for the apparent absence of this relation bewteen average years of schooling and GDP growth.

(13)

Figure 3.1 Scatterplot GDP growth & Avg. Years of Total Schooling

Figure 3.2 Scatterplot GDP growth & Avg. Years of Primary Schooling

Colombia and Mexico have the nearly the same average years of education on all levels, but Mexico has negative GDP growth of -13% whereas Colombia has had a positive GDP growth of 13%. In general the countries with the most average years of education are the South-American and Middle-South-American countries. These countries are most often situated to right of the middle in the scatter plots. Example of these countries are Colombia (COL), Mexico (MEX), Barbados (BRB), Chili (CHL), Argentina (ARG), Uruguay(URY), Trinidad and Tobago (TTO), and Panama (PAN). If these scatter plots indeed shows an accurate

ALG _ARG BHR BRB BOL BWA BDI CMR CHL COL ZAR CRI CIV DOM ECU EGY SLV FJI GAB GMB GHA GTM HND IND IDN IRN JAM JOR KEN LSO MYS MUS MEX MAR NPL NIC NER PAK PAN PNG _PERPHL RWA SAU SEN ZAF SDN THA SWZ TGO TTO TUR URY VEN ZWE -1 5 -1 0 -5 0 5 10 15 G D Pg ro w th 1.00 2.00 3.00 4.00 5.00 6.00 7.00 Avg. Years of Total Schooling

ALG _ARG BHR BRB BOL BWA BDI CMR CHL COL ZAR CRI CIV DOM ECU EGY SLV FJI GAB GMB GHA GTM HND IND IDN IRN JAM JOR KEN LSO MYS MUS MEX MAR NPL NIC NER PAK PAN PNGRWA PHL_PER SAU SEN ZAF SDN THA SWZ TGO TTO TUR URY VEN ZWE -1 5 -1 0 -5 0 5 10 15 G D Pg ro w th 1.00 2.00 3.00 4.00 5.00 6.00 7.00 Avg. Years of Primary Schooling

(14)

relationship between the two variables, the regression in the following section should show no significant relation as well between GDP growth and average years of education.

Figure 3.3 Scatterplot GDP growth & Avg. Years of Secondary Schooling

Figure 3.4 Scatterplot GDP growth & Avg. Years of Tertiary Schooling

ALG _ARG BHR BRB BOL BWA BDI CMR CHL COL ZAR CRI CIV DOM ECU EGY SLV FJI GAB GMB GHA GTM HND IND IDN IRN JAM JOR KEN LSO MYS MUS MEX MAR NPL NIC NER PAK PAN PNG _PERPHL RWA SAU SEN ZAF SDN THA SWZ TGO TTO TUR URY VEN ZWE -1 5 -1 0 -5 0 5 10 15 G D Pg ro w th 1.00 2.00

Avg. Years of Secondary Schooling

ALG _ARG BHR BRB BOL BWA BDI CMR CHL COL ZAR CRI CIV DOM ECU EGY SLV FJI GAB GMB GHA GTM HND IND IDN IRN JAM JOR KEN LSO MYS MUS MEX MAR NPL NIC NER PAK PAN PNG _PER PHL RWA SAU SEN ZAF SDN SWZTHA TGOTTO TUR URY VEN ZWE -1 5 -1 0 -5 0 5 10 15 G D Pg ro w th 1.00 Avg. Years of Tertiary Schooling

(15)

3.3 Empirical Design

In this paper, a maximum sample of 55 emerging and developing countries is used to do a cross-section economic-growth analysis using standard OLS estimation. The model equation becomes as follows:

𝐺𝑟𝑜𝑤𝑡ℎ = 𝛽₀+ 𝛽₁𝑋_𝑖 + 𝛽₂𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛_𝑖 + 𝜀_𝑖

In this equation X indicates the vector of control variables, education indicates the measure for educational attainment, is the constant, and are the estimated coefficients given in base point elasticities, and is the error term.

To determine whether the disturbances are heteroscedastic, two tests are run for a model that includes all control variables (with GDP at market prices and Fertility rate) and the educational variables for the total population, so no seperation between male and female. For this model specification, both the Breusch-Pagan test and the White test for

heteroscedasticity do not reject the null hypothesis of constant error-term variance. These tests produced p-values of 0.3156 and 0.4365, respectively. Hence, using heteroscedasticity-robust standard errors is not advised. Woolridge (2009) explains that using

heteroscedasticity-robust t-statistics can have a distribution that is not very close to the normal t-distribution when the sample size is small. Using heteroscedasticity-robust standard errors would therefore lead to possible false conclusions of the regression later on. This research paper has a sample size of 55 which can be considerd a small sample for a cross-sectional regression and this explanation is therefore a valid reason not to use

heteroscedasticity-robust standard errors.

In the next section the model specification is estimated. This will include all the control variables and education variables. Besides this OLS regression, correlations between the education and control variables and growth will be looked at.

4. Analysis

This section will provide the results of the data analysis aimed at answering the following research question: Does educational attainment have a significant effect on economic growth in developing countries?

4.1 Correlations

To start off this analysis, the simple correlations among variables will be discussed. These correlations do not show causal relations between the variables but do simply show

(16)

associations between the variables. The correlations here are Pearson correlation coefficients and have a value between -1 and 1. The correlation coefficient fits a staight line trough all data points of that variable. The slope of that line determines if the coefficient is positive or negative. If the line through the data point has a positive slope the correlation coefficient will be between 0 and 1. And when there is a level trend in the data for example the correlation will be 0 and the fitted line will be straight. When the data points are close to the fitted line that means that the correlation coefficient is high, close 1 or -1. If the data points are far from the fitted line then the correlation coefficient will be lower, and close to 0. When calculating these correlation statistics, the variables are assumed to be continuous and interval. A correlation is usefull to see the linear relationship between variables. It shows how stark a variable will react positively or negatively if another variable changes.

Table 4.1 shows the correlation table with GDP growth, the education variables and all the control variables. What can be seen in table 4.1 is that there is almost no correlation between the average years in total and primary, education and GDP growth. The correlations being -0.0695, -0.0264, respectively. So there is no positive or negative linear relationship between GDP growth and these two educational variables. Also a negative relationship between average years in secondary and tertiary education GDP growth is seen. These correlations are also higher than the other levels of education, namely -0.1994, and -0.1811 respectively. This means that when there is a positive change in average years in secondary education GDP growth is likely to also be lower.

The correlations between the different measures of educational attainment are unsurprisingly high. It is expected that if the number of average years of primary schooling rises, the other education levels also show a similar rise, since people with primary education are likely to continue with secondary education and then tertiary education.

Looking at the control variables initial GDP in 1980 and Openness have the largest correlations with GDP growth at respectively -0.3125 and 0.2581. It was already expected in the previous section that initial GDP would have a negative relation with GDP growth since higher initial income implies lower marginal output, and thus a path of slower growth. The positive correlation of the variable Openness is a logical one, since it essentially shows how much trade is possible with a country. So a higher degree of Openness would allow for more trade and therefore contribute to higher GDP growth in a country. This is not a surprising result because previous literature such as Harrison (1996) also found there to be a generally positive accosiation between growth and various measures of Openness.

(17)

Table 4.1 Correlations

4.2 OLS Estimation

In this section the results of OLS regression will be presented. Firstly, table 4.2 shows the regression results with only GDP growth as the dependant variable and the education variables. Column 1 shows a model that simply consists GDP growth and average years of total schooling. What can be seen here is that there is a small negative and insignificant relationship between these two variables. The coefficient here is -0.00514, which means that a 1% rise in average years of total schooling actually slightly decreases the growth of GDP by -0.00514. All the coefficients in this table and the next table 4.3 which will include the

control variables can be interperted in this way. Deininger, K. and P. Olinto. (2000) find that this relation between education and GDP growth may be negative due to the inequality that exists in countries. They point out that in countries were assets are distributed unequally, expanding education policies will have little impact. Considering the fact that all the

cpuntries in this regression are developing and emerging countries, inequality is often higher compared to advanced economies and this explanation may therefore be valid one in this analysis. Variables GD P g ro wth ln To talsc h oo l ln P rima ry ln S ec o n d ary ln Tertiary ln F erti li ty ln GD P 1 9 8 0 ln Op en n ess TOTg ro wt h ln In v estra ti o ln Ru leo flaw GDPgrowth 1.000 lnTotalschool -0.070 1.000 lnPrimary -0.026 0.980 1.000 lnSecondary -0.199 0.841 0.721 1.000 lnTertiary -0.181 0.685 0.578 0.798 1.000 lnFertility 0.067 -0.662 -0.650 -0.584 -0.426 1.000 lnGDP1980 -0.313 0.318 0.244 0.469 0.459 -0.231 1.000 lnOpenness 0.258 0.093 0.115 0.002 -0.083 -0.010 -0.517 1.000 TOTgrowth 0.062 0.023 -0.021 0.130 0.104 -0.053 0.086 0.081 1.000 lnInvestratio 0.093 0.102 0.095 0.064 0.091 -0.056 0.051 0.520 0.327 1.000 lnRuleoflaw -0.180 0.457 0.409 0.462 0.403 -0.587 0.129 0.214 0.250 0.366 1.000

(18)

In column 2 of table 4.2 the average years of schooling is seperated into three levels of education: primary, secondary and tertiary education. The coefficients for these variables are: 0.0176, -0.0205 and -0.00278 respectively. This means that a 1% rise in average years of primary education slightly increases GDP growth, but an increase in average years of

secondary and tertiary education slightly decreases GDP growth. This is a similar effect as seen with the total average years of education. However all these variables appear not be significant and these three variables are also jointly insignificant. The F-statistic that proves this is 0.2915, which means the null hypothesis that they are all jointly significant is rejected.

Table 4.2 - OLS Estimation including GDPgrowth and Education

Notes: estimated using log-log OLS regression; dependent variable is average GDP growth over the period 1980-2010; all coefficients can be interpreted as elasticities; standard errors in parentheses; significance is indicated by * for p<0.05, ** for p<0.01, and *** for p<0.001

In columns 3 till 5 of table 4.2 the three levels of education are each added seperately. The coefficients are now -0.00189, -0.0125 and -0.00804 for primary, secondary and tertiary education. There is no real change in the magnitude of the coefficients only than the change in sign for average years of primary education. When added seperately in column 3 this

(1) (2) (3) (4) (5)

Estimation GDPgrowth GDPgrowth GDPgrowth GDPgrowth GDPgrowth

lnTotalschool -0.00514 (0.0101) lnPrimary 0.0176 -0.00189 (0.0140) (0.00983) lnSecondary -0.0205 -0.0125 (0.0166) (0.00846) lnTertiary -0.00278 -0.00804 (0.00994) (0.00600) Constant 2.816* -0.690 2.416* 1.685* 0.167 (1.238) (2.639) (0.961) (0.752) (1.701) N 55 55 55 55 55 adj. R-sq -0.014 0.015 -0.018 0.022 0.015

(19)

coeffient is now negative instead of positive like in column 2. So this means that the sole effect of a 1% increase in average years of primary education now causes a slight decrease in GDP growth of -0.00189. This negative relationship between education and GDP growth contradicts the results found by Barro (2013), but are not completely new. For example Li, H., and H. Zou. (1998) also find a negative relation between primary enrolment and growth. An explanation for a negative result of the education coefficients in the columns in table 4.2 could be that if children stay in school longer they cannot also work on the land for example. This lost productivity for families will cause income to be lower. And the costs of schooling become higher if the children stay in school longer. These families will have less money to spend and contribute to growth of the economy, which in turn lower GDP growth. This explanation is not supported by previous literature but could still be an interesting way to think about this result.

After looking at a simple model with only the education as an explanatory variable, the control variables are added to the regression one by one in table 4.3. In column 1 the three educational variables are added and so is intial GDP in 1980. The coefficient of initial GDP in 1980 is negative at -0.726 which means that a high initial GDP in 1980 decreases GDP growth from 1980-2010. This result is in line with expectations formed in the previous sections, derived by (Sala-i-Martin, 1997).

The coefficients of the educational variables show small decreases in magnitude compared to the coefficients in column 2 of table 4.2, but the sign stays the same. This means that the effect of increases in average years of education on GDP growth decreases with the addition of intial GDP as an explanatory variable. In the rest of the model specifications in columns 2 to 7 the coefficients of educational variables show almost no change in magnitude. They all stay close to zero. The change in sign for average years of tertiary education to a negative one from column 3 onward is not an important change since in none of the column is the coefficient significant. The results are different from the results of Barro (2013) who found, as previously mentioned, secondary and tertiary schooling to have a significant effect on growth. However Barro did a panel study over three short periods in time. In this research paper a cross section was done over a longer period of time, namely 30 years.

(20)

Table 4.3 - OLS Estimation including GDPgrowth, Education and Control variables

(1) (2) (3) (4) (5) (6) (7)

Estimation GDPgrowth GDPgrowth GDPgrowth GDPgrowth GDPgrowth GDPgrowth GDPgrowth

lnPrimary 0.0139 0.0143 0.0189 0.0126 0.0148 0.0148 0.0119 (0.0139) (0.0154) (0.0148) (0.0155) (0.0158) (0.0160) (0.0158) lnSecondary -0.0137 -0.0134 -0.0196 -0.0166 -0.0181 -0.0186 -0.0154 (0.0168) (0.0174) (0.0170) (0.0177) (0.0179) (0.0188) (0.0185) lnTertiary -0.000108 -0.000166 -0.00223 0.000488 0.000370 0.000504 0.00271 (0.00990) (0.0100) (0.0102) (0.0101) (0.0101) (0.0103) (0.0102) lnGDP1980 -0.726 -0.725 -0.438 -0.471 -0.445 -0.553 (0.436) (0.440) (0.534) (0.538) (0.616) (0.606) lnFertility 0.00169 -0.000376 0.000778 0.000640 -0.0251 (0.0271) (0.0272) (0.0273) (0.0277) (0.0308) lnGDPcap1980 -0.00245 (0.00813) lnOpenness 0.0119 0.0105 0.0115 0.0119 (0.0125) (0.0126) (0.0171) (0.0167) TOTgrowth 0.285 0.297 0.384 (0.374) (0.401) (0.395) lnInvestratio -0.00220 0.00986 (0.0244) (0.0248) lnRuleoflaw -0.0406 (0.0231) Constant 17.40 17.08 1.199 6.081 7.046 6.749 13.28 (11.16) (12.40) (6.813) (16.95) (17.07) (17.57) (17.58) N 55 55 55 55 55 55 55 adj. R-sq 0.048 0.029 -0.003 0.027 0.019 -0.002 0.041

Notes: estimated using log-log OLS regression; dependent variable is average GDP growth over the period 1980-2010; all coefficients can be interpreted as elasticities; standard errors in parentheses; significance is indicated by * for p<0.05, ** for p<0.01, and *** for p<0.001

In Column 2 Fertility rate is added to regression analysis. This variable has a small positive coefficient of 0.00169, meaning that a 1% increase in the fertility rate will increase GDP growth by 0.00169 basis-points. The coefficients for the fertility rate in the rest of the columns in table 4.3 stays close to zero. The small positive or negative sign close to zero is

(21)

different from the conclusion of de la Croix and Doepke (2003) who found a substantial negative effect between the fertility rate and GDP growth. However, in their study de la Croix and Doepke (2003) made a distinction between the fertility rate as an endogenous variable in a inequality measure and fertility as an exogenous variable. The exogenous fertility variable which is also used in this regression has a smaller effect on growth than the endogenous variable. The effect of initial GDP is stable in column 2 compared to column 1 at around -0.725.

Column 3 shows a model specification with GDP per Capita instead of initial GDP at market prices and the fertility rate. This is done, as explained in the previous section, to see which GDP control variable has highest significance in explaning GDP growth. The

coefficient for GDP per capita is -0.00245 which is smaller than the coefficient for intial GDP. However the sign is the same, namely negative. So a higher GDP per capita also has a negative effect on GDP growth. Both initial GDP in column 2 and GDP per capita in column 3 are not significant, which means that none of the two GDP control variables is better than the other as an explanator for GDP growth. Furthermore, Barro (2013) also says that it is commonly known that there is no simple relation between GDP per capita and growth across countries. Therefore the choice was made to not use GDP per capita in the following columns 4 to 7. In these columns the other control variables will be added to the model which includes initial GDP and Fertility rate.

Column 4 adds the control variable Openness to the regression estimation. The coefficient is 0.0119, which means that a rise in level of openness of a country contributes positively to GDP growth. This positive relationship between a greater level of openness and more GDP growth is in line with the results found by Sachs and Warner (1995). Only here, again, the variable turn out to be not significant. Adding the variable openness decreases the magnitude of the variable for initial GDP to -0.438. In all the following columns the

coefficient of intial GDP lies between -0.438 and -0.553 and doesnt go back to the initial figure of around -0.725. So the negative impact of initial GDP on growth decreases when openness is also considered as a variable.

In Column 5 the variable Terms of Trade growth is added to the regression. The coeffiicient here is 0.285, so a 1% increase in terms of trade growth will cause a 0.285% increase in GDP growth. The magintude of the coefficient of terms of trade growth increases in columns 6 and 7, with 0.297 and 0.384 respectively. This positive effect is in line with the results from Bleaney and Greenway (2001), but the results of here are not significant whereas

(22)

in their study the variable was found be a significant explanatory variable in the GDP growth regression.

Column 6 shows the addition of the Invesment ratio to the model and has a regression coefficient of -0.00220. So a 1% increase in the investmentratio will decrease GDP growth by -0.00220%. In column 7 the sign changes to a positive sign, but the coefficient stays close to zero, at 0.00986. In both column the effect of the investment ratio is not significant. This doesn’t necessarily contradicts the results from Sala-i-Martin et al. (2004) who found that the investment price is a significant determinant in long term economic growth. The investment ratio is not equal to the investment price, but both are meassures for the impact of investment on economic growth. Therefore one measure being significant and the other one not

significant is at the very least an unexpected result.

Lastly column 7 adds the rule of law to the regression. The coeffient for the rule of law is -0.0406. This means that a 1% increase in the rule of law in a country would decrease the GDP growth by -0.0406 basis points. This result is different from Barro (2001) who concludes that rule of law has a positive and statistically significant effect on growth. However Barro (2001) uses data from Knack, S. and Keefer, P. (1995) whereas this paper uses data gathered from the World Governance Indicators (WGI) of Kaufmann et al. (2010). This difference in data source maybe causing the difference in results as well. Especially because data on rule of law and other institutional variables is subjective data and could therefore show variation in numbers and measurements depending on the researchers that gather the data.

In none of the model specifications any variable, control or educational, turns out to be statistically significant, which is contradictory to the results of Barro (2013). So this leads to the believe that this model has little explanatory power. This is confirmed by looking at the adjusted R-squared at the bottom of table 4.3. The adjusted R-squared ranges from -0.002 to 0.048. A value close to 0 means the model has little to no explanatory power. The meaning that the model has little explanatory power. For initial GDP this may indicate an absence of unconditional or absolute convergence. If countries were to grow to the same level of

economic output, lower-income countries are expected to grow more rapidly, as explained in the literature part of this research paper.

(23)

5. Conclusion

The objective of this paper was to research to which extent educational attainment can be considered a significant determinant to GDP growth. A review of the existing literature studies focussing on educational attainment and economic theory formed the basis for the analysis in this research paper. It is often assumed that education is an impotant factor in increasing economic growth in the long run. The dataset used in the analysis consists of data on 55 developing and emerging countries across the world, during a period of 30 years for 10 different variables and GDP growth.

Controlling for other factors in the regression model, the results show that educational attainment measured by average years of education is not siginificant in explaning changes in GDP growth in the period 1980-2010. The insignificant negative effect of average years of secondary and tertiary schooling in some of the model specifications is remarkable, since it contradicts results found by previous literature. The conclusion from this research paper is that increasing the average years of schooling seems be the wrong policy when solely

focussing on increasing economic growth. Education may still play a part in the development of these 55 countries later on when they have evolved into developed economies. But it is apparently not the most important at this stage of a country's development. In these

developing and emerging countries improvement of physical capital instead of human capital may just be far more important in the growth of GDP. It will be interesting to experiment with different variables than that are used here, to see if they can be a better explanatory variable. This would create more reliable control model in which to test the educational attainment measures.

This analysis contributes to the existing literature on cross-sectional studies focussing on educational attainment. It does so by specifically looking at 55 developing and emerging countries and not at advanced economies. Determining that educational attainment is apparently not a significant contributor to growth allows these 55 developing and emerging economies to focus there efforts on other policies that might actually contribute to GDP growth. By establishing which variables affect growth in these countries, the welfare of the population can be increased quicker and with greater steps. So according to this analysis education would not be a wise policy to increase GDP growth. This is a contradictory conclusion to many other existing literature.

(24)

Bibliography

Barro, R. J. (2001). "Human Capital and Growth." American Economic Review, 91(2): 12–17

Barro, R.J. and J.W. Lee., (1996). "International Measures of Schooling Years and Schooling Quality," American Economic Review, 86, 218-23.

Barro, R.J and J.W. Lee, (2013). "A New Data Set of Educational Attainment in the World, 1950-2010." Journal of Development Economics, vol 104, pp.184-198. Bleaney, M., Greenway, D., (2001). The impact of terms of trade and real exchange rate

volatility on investment and growth in Sub-Saharan Africa. Journal of Development

Economics 65, 491 – 500

Deininger, K. and P. Olinto. (2000). “Asset Distribution, Inequality, and Growth,” World

Bank Policy Research, work. paper 2375.

de la Croix, D., and M. Doepke. (2003). "Inequality and Growth: Why Differential Fertility Matters," American Economic Review, 93(4), 1091-1113.

Hanushek, E.A. and Kimko, D.D. (2000). "Schooling, labor force quality, and the growth of nations", American Economic Review, vol. 90 (5) (December), pp. 1184–208.

Harrison, A. (1996). “Openness and Growth: A TimeSeries, Cross-Country Analysis for Developing Countries.” Journal of Development Economics, March, 48(2), pp. 419–47

IMF. (2015). "Developing & Emerging Countries" International Monetary Fund's World

Economic Outlook Report, April 2015. Retrieved 6 4, 2016, from IMF:

http://www.imf.org/external/pubs/ft/weo/2015/01/weodata/groups.html

Kaufmann, D., Kraay, A., & Mastruzzi, M. (2010). "The Worldwide Governance Indicators: A Summary of Methodology". Data and Analytical Issues, World Bank Policy

Research Working Paper 5430.

Kaufmann, D., Kraay, A., & Mastruzzi, M. (2011). The worldwide governance indicators: Methodology and analytical issues. Hague Journal on the Rule of Law, 3(2), 220–246. Knack, S. and Keefer, P. (1995) “Institutions and Economic Performance: Cross-Country

Tests Using Alternative Institutional Measures.” Economics and Politics, November, 7(3), pp. 207–27.

Lee, J.W., and R.J. Barro (2001). “Schooling Quality in a Cross-Section of Countries”,

(25)

Li, H., and H. Zou. (1998). "Income Inequality is not Harmful for Growth: Theory and Evidence," Review of Development Economics, 2(3), 318-334.

Lucas R.E. (1988). "On the Mechanics of Economic Development," Journal of Monetary

Economics, July, 3-42.

Mankiw, G., D. Romer, and D. Weil. 1992. "A Contribution to the Empirics of Economic Growth," Quarterly Journal of Economics, 107.

OECD (2016). Pisa Scores. Retrieved 6 2, 2016, from OECD: http://www.oecd.org/pisa/aboutpisa/

Romer, P. M. (1986). Increasing Returns and Long-Run Growth. Journal of Political

Economy, 94, 1002-1037.

Romer, Paul M., (1990). Endogenous Technological Change. Journal of Political Economy, 98(5) part II, S71-S102.

Romer, P. M., (1994). The Origins of Endogenous Growth. The Journal of Economic

Perspectives, 8, 3-22.

Sachs, J. D. and Warner, A. M. (1995). “Economic Reform and the Process of Economic Integration.” Brookings Papers on Economic Activity, , (1), pp. 1–95.

Sala-i-Martin, X. (1997) “I Just Ran 2 Million Regressions.” American Economic

Review, 87(2), pp. 178–83.

Sala-i-Martin, X., Doppelhofer, G., & Miller, R., (2004). Determinants of long-term growth: A Bayesian averaging of classical estimates (BACE) approach. American Economic

Review, 94(4), 813-835.

Solow, R. M. (1956). A contribution to the theory of economic growth. The Quarterly

Journal of Economics, 70(1), 65-94.

Sorensen, P.B., Whitta-Jacobsen, H.J., (2005). Introducing Advanced Macroeconomics:

Growth and Business Cycles. Berkshire, UK: McGraw Hill.

United Nations (2016). Millennium Goals. Retrieved 6 2, 2016, from United Nations: http://www.un.org/millenniumgoals/education.shtml

Wooldridge, J.M., (2009). Introductory Econometrics. 4 ed. South-Western. World Bank (2016). Retrieved 6 2, 2016, from World Bank Open Data:

http://data.worldbank.org/

Education and economic growth in emerging and developing countries : educational attainment and GDP