Human Capital
*
Educational Attainment versus Work
Experience
Marijn G.J. van Essen
**
August 2008
Under the supervision of
dr. M. Koetter
*
I would like to thank my supervisor, Michael Koetter, for his efforts and time spent at guiding me through the obscure depths better known as quantitative analysis, and my family and friends, in particular Maike Bastiaans and my parents, for supporting me all the way. I know quite certainly that this thesis is not written making use of a special talent or experience; curiosity, obsession and dogged endurance, combined with self-criticism, have brought me to my ideas.
**
Human Capital
Abstract
Theory suggests that improvements in human capital improve economic growth. However, the results from previous research present mixed evidence. In part this may be explained by a lack of accurate data and an unclear idea about the channels through which human capital may influence growth. This thesis aims at reinvestigating the relationship between human capital and economic growth using a more comprehensive definition of human capital and comparing the influence of its components; the level of human capital obtained at school and the level of human capital obtained while on the job. The results provide evidence for a potential positive influence of human capital in general, in particular educational attainment. Contrary to expectations, additional work experience has a strong negative influence on growth in more developed countries.
Keywords: Human capital; Solow-Swan growth model; Educational attainment; Work
experience.
Table of Contents
ABSTRACT………..……….1 1 INTRODUCTION...3 2 LITERATURE REVIEW...6 2.1 Human Capital ... 6 2.2 Hypotheses ... 123 DATA & METHODOLOGY ...14
3.1 Data ... 14
3.2 The Economic Model... 15
3.3 Panel Data... 18
3.4 Estimation... 19
4 RESULTS...21
4.1 OLS Estimation ... 21
4.2 GMM Estimation ... 22
5 SUMMARY & CONCLUSIONS...28
1 Introduction
In 1943 Winston Churchill prophesised: “The empires of the future will be empires of
the mind” 1. Has he proven to be right? This paper emphasizes the role of human capital on economic growth. Since the pioneering work of Mincer (1974) this subject gained increasing attention. Today, it is generally reasoned that quality improvements in human capital will accelerate economic growth. Supposing this to be true, what are the potential benefits of a more knowledgeable and better skilled workforce? For one, more human capital facilitates the absorption of superior technologies from leading countries. This channel is expected to be especially important for schooling at the secondary and higher levels (Barro 2001, Nelson & Phelps 1966). Second, human capital tends to be more difficult to adjust than physical capital. Therefore, a country that starts of with high levels of human capital may be able to grow rapidly just by adjusting upward the quantity of physical capital, whereas the reverse tends to be far more difficult (Barro, 2001).
Nonetheless, much disagreement still stems from mixed research results. Past research in general has focused on simple definitions of human capital. Initially, many studies have tried to relate average years of schooling with economic growth or individual earnings (Li & Florida 2006, Stevens & Weale 2004, Barro & Lee 2000, Becker & Chiswick 1966, Florida 2002). A second group has focussed on the relationship between work experience and economic performance (Cook 2004, Reimers & Garvey 1979, Mincer 1974). In general these studies conclude that human capital has a significant positive effect on economic growth. However, among others, Islam (1995) finds that any positive impact of human capital that may have existed in cross-country analysis disappears in a panel data sample.
Despite these mixed results, there still seems to be a run on highly educated employees. In the literature often referred to as the ‘war for talent’ (a term coined by
McKinsey & Company2). This war for talent proves that considering human capital as
the key business resource and the most important source of competitive advantage is a widespread phenomenon. In part, this is caused by the expectation that the supply of
1
Winston Churchill in his speech of September 5, 1943 at Harvard University after being awarded an honorary degree.
2
educated employees will increasingly lag behind demand (Valetta & Hodges, 2005). Some studies even conclude that today already 75 percent of the executives indicate that their companies either do not have enough educated employees sometimes or are always short of them (Fishman, 1998). See for example also figure 13. The result is that in the OECD a considerable earnings premium (found to lie between 25 and 100 percent) is associated with attaining tertiary education as compared to even upper-secondary education (Ischinger, 2007).
Figure 1: Unemployment rates by education4
This marked interest in especially (young) highly educated employees implies that there are significant differences in the return of education and experience. However, does education indeed have a more beneficial effect on the creation of knowledge and competencies and, consequently, the ability to improve performance than experience can have? Some authors seriously doubt whether education really has that potential. Ericsson, Prietula & Cokely (2007), for example, argue that it takes a performer (whether it is in music or in business) between 10 and 25 years of intense training to become a top performer. On the other hand, it seems difficult to believe that such a strong focus on educational attainment above work experience exists in the job market without a reason. Something must have urged human resource managers to start
3
Valetta, R., Hodges, J. “Age and Education Effects on the Unemployment Rate” (2005), FRBSF
Economic Letter, 2005-15; July 15, 2005. 4
looking at a different pool of potential employees. In either case, these issues raise an important question: Does educational attainment have a more beneficial influence on performance than work experience? Following recent trends, it is expected that this is indeed true. Although, this influence may perhaps, considering the potential presence of diminishing returns, be particularly pronounced for lower income countries. This thesis thus investigates, using an extensive sample of countries and a maximum of 8 time period of 5 years, the relationship between the qualitative factors of labour and economic growth. Economic performance is measured as the growth in real GDP per capita (PPP, international dollars). To measure experience a proxy will be used.
This thesis contributes to the existing literature in the following ways. Although, there is quite a considerable amount of literature studying the effects of either educational attainment or work experience on personal earnings, performance, or the return of investment, many of these studies used simple cross-country data. Furthermore, almost all studies fail to study both educational attainment and work experience simultaneously in a single model. Taking advantage of substantial improvements in data availability, this thesis approaches human capital, composed of educational attainment and work experience, as a set of complementary and substitutionary types of knowledge and compares the distinct impact of these components on performance. Considering that human capital is often considered to be the most important business resource of which the importance is expected to increase even more in the future, the results of such a study may have important implications for human resource policy and economic performance in general.
This thesis continues in the following manner. Section two reviews the existing literature concerned with the link between human capital and economic growth and derives the hypotheses of interest. Section three discusses the data and methodology used for estimation. Section four discusses the results and section five concludes.
2 Literature Review
Next, earlier work in the field of human capital and its impact on growth will be discussed. It must be noted here that in this discussion the term human capital can refer to educational attainment, work experience, or both. Which meaning is applicable will follow logically from the text. Afterwards, the hypotheses of importance to this thesis are derived.
2.1 Human Capital
Human capital is an increasingly important issue on the economic and social agenda. The Lisbon Summit, for example, has as its goal to make Europe the most competitive
and most dynamic knowledge based economy in the world5. In a knowledge based
economy, human capital is the most important competitive advantage and employment for people with no or low levels of education is low. In the discussion on how to promote economic growth, investment in human capital has thus been given high priority. The belief is that an educated labour force has higher productivity levels and is better at creating, implementing, and adopting new technologies, thereby generating growth.
Consequently, measures of human capital have found their way to growth theory and several studies have tried to estimate the relationship between human capital and economic growth. However, the literature has not yet been able to establish a unanimous standpoint regarding this relationship. Especially, much debate still surrounds the discussion on what methods best proxy the level of human capital actually obtained by an individual. The most common proxies are discussed next (see also Teixeira & Fortuna, 2004).
For one, especially earlier work used data on enrolment rates (most commonly at the secondary level) as a proxy of human capital. The major benefit of using this data is that it is widely available. However, clearly it also exhibits some crucial shortcomings. Enrolment ratios are not very likely to accurately represent any changes in the human capital stock, since the students currently enrolled are not yet part of the
5
labour force. Similarly, but more importantly, considering differences in survival and repetition rates (progression and completion rate) or differences in income (affordability of education at subsequent levels) and the educational system itself (quality of education) no generalisations whatsoever can be made using enrolment rates as a proxy of human capital.
Another proxy of human capital is found in the form of adult literacy rates. Again, this data is widely available, which is certainly a great benefit. However, such rates ignore any investment in and the attainment of skills and abilities, since it does not include qualifications above the basic levels of education (apart from many developing countries literacy rates close to 100% are usually obtained at an early age). It must be noted that the OECD is attempting to develop a new, augmented measure of international adult literacy (IALS6). It takes into account many aspects of human capital and combines it into a single measure and value. This measure looks very promising indeed, but only in the future as data currently is still very restricted. Third, international test scores are sometimes used. Indeed this measure seems to be able to quite accurately represent the potential abilities of the workforce. However, tests are usually only taken in earlier grades and as such may not precisely approximate the level of human capital eventually attained by the population. Greatest shortcoming, however, is again that data on international test scores is still limited to only a number of developed countries.
Measures of the average educational attainment, finally, which have become available through new data sets such as that of Barro & Lee (2000)7, are currently most commonly used. Educational attainment is a stock variable and it takes into account the total amount of formal education received and completed by the labour force. Often this data is used to specify human capital by average years of schooling, which gives the same weight to any year of schooling acquired and ignores the fact that one additional year of schooling at different levels does not raise human capital stock by an equal amount regardless of the level of schooling and quality of the education system that provided it. Still, setting aside the specification of the average number of years of schooling, educational attainment, as the percentage of the population that
6
Organisation for Economic Co-operation and Development (OECD): “Adult Literacy”, Directorate
for Education. Available from OECD. 7
has attained some threshold level of education, may be better able to measure the qualitative aspect of human capital (Florida 2002, Barro 2001).
Considering that data on educational attainment still provides the best available information about the stock and flow of human capital and that it may help to provide an answer to the question whether the war for talent is indeed justified, this thesis uses data on educational attainment to proxy the level of human capital obtained at schools.
Educational Attainment
Educational attainment is a measure of the highest level of education an individual has enjoyed and completed. Human capital, particularly that attained through education, has been emphasized as a critical determinant of economic progress (Barro & Lee, 2000). The logic behind this is quite straightforward. A higher rate of employees who have enjoyed increasingly advanced levels of education signifies a more skilled and more productive labour force. As such, it is not that strange that firms will even base their location decisions on labour availability and quality (Florida 2002, Hall & Jones 1999). The measurement of educational attainment is quite complex, as national education systems are often very different. To be able to compare the different national education systems, the International Standard Classification of Education
(ISCED978) is designed by UNESCO. However, still categorization problems may
occur. For example, although ISCED97 provides guidance on which qualifications and stages of education should be assigned to specific ISCED97 categories, the classifications do not fully reflect the heterogeneity of national education systems. As a consequence, in the presence of conflicting qualification interpretations, national governments can reallocate national education levels to a different ISCED97 classification for political considerations. For example, Steedman (2000) points out that the United Kingdom chooses such a definition of ISCED97 levels which gives the impression that the education system produces a larger supply of human capital. Easterlin (1981) further suggests that the method of schooling is very important and he argues that the quality of education can as such differ greatly despite the level of
8
school attendance. Finally, it is worth mentioning that education systems may also differ within countries over time. During the last 30 years, 32 countries have changed at least once the typical duration of schooling (UNESCO, Statistical Yearbook). The Barro & Lee (2000) dataset, however, controls for these changes.
Evidence from the Literature
Results of previous research on the effects of human capital have lead to a somewhat miscellaneous perception. Estimates of Barro & Lee (2000) show that one additional year of schooling (roughly a one standard deviation change) raises the growth rate by 0.44 percent per year. Also Becker & Chiswick (1966) find that high attainment rates of adult males at the secondary and higher levels are positively related to subsequent growth. Stevens & Weale (2003) find that a one percent increase in the enrolment rate raises GDP by 0.35 percent. In the classic study of Mincer (1974), it is concluded that for white males not working on farms, an extra year of education raises individual earnings by about 7 percent, suggesting a performance premium associated with higher levels of education. Barro (1997) suggests that one extra year of education (for men) raises the growth rate by 1.2 percent per year. Moreover, he argues that this percentage is likely to be even higher over time when the better educated slowly start to replace the less educated. The results of Matthews et al imply that between 1856 and 1973 an improved level of education contributed 0.3 percent per year to the growth of output in the United Kingdom, with an overall growth of 1.9 percent per year (Stevens & Weale, 2003). Finally, research conducted by the OECD concluded that countries expanding tertiary education attainment more, experienced lower rates of unemployment and even lesser qualified employees participants on the job market appeared to have better employment opportunities as a result (Ischinger, 2007).
Interestingly, Becker & Chiswick (1966) also find that the individual returns are different between the South and the non-South of the United States, implying that expected personal returns from education are higher when the variance in years of schooling is higher. In other words, education can be over-supplied and will probably exhibit diminishing returns. The same conclusion is drawn by Stevens & Weale (2003) who find that the early stages of education are more valuable than the later stages. One extra year of education raises labour income by about 10 percent, but only by 6.5 percent in advanced countries. Also Krueger & Lindahl (2001) find that for low levels of education, education contributes positively to growth, while for high levels of education it may even depress the rate of growth. The marginal effect of education on economic growth is positive for countries where the average worker spends less than 7.5 years in education (e.g. the OECD average is 8.4 years). Similarly, Wolff & Gittleman (1993) find that among the upper income group, where there is much more variation in tertiary education than in primary or secondary education, that tertiary education is the only statistically significant variable. On the other hand, for the poor countries primary education is statistically significant while differences in tertiary education, which show rather little variation, are not. Islam (1995) and Benhabib & Spiegel (1994) explain these results, suggesting that human capital may affect economic growth through its influence on the catch-up term. The analysis should therefore address the question whether economic growth is influenced by the change in human capital as well as the level. Whereas, changes in human capital correspond to human capital as a factor of production, human capital in levels corresponds to human capital as facilitating the adoption of new technologies.
A last small subset of studies has included work experience as a component of human capital. Both Papageorgiou & Pérez-Sebastián (2002) and Cook (2004) have found that including work experience as a determinant of human capital significantly improves the fit of the model used.
Work Experience
already knows the drill. They simply have experienced many of the situations before and (roughly) know how to deal with them. Rosen (1972) indeed argues that learning does not stop after graduation. Instead, it is economical to shift its location to the market; for after some point learning and work are complementary, and knowledge is more efficiently acquired in combination with work experience rather than in school. Ideally, work experience measures actual work experience from individual to individual, including industry specifications, on-the-job learning, special courses and the like. Unfortunately, this is impossible to realize on a larger scale and without an exact data source. For this reason, many researchers return to measuring not actual, but potential work experience (Cook 2004, Reimers & Garvey 1979, Mincer 1974). By definition, this measure is calculated as follows: potential work experience = age
– time spent in school – non-school, non-work time. The last variable measuring time
actual work experience. Hence, although potential work experience is still merely a proxy, it is the best approximation available and is thus used in this thesis to proxy for actual work experience in the form of: median age of the population – average years
of schooling – entry age.
Expanding the level of focus
In the above presented literature review, some of the reasoning is on an individual basis (e.g. the link between individual earnings and education). However, this thesis relies on country-level data. Hence, the important question is raised whether the same reasoning can be extended to the country-level. The best way to look at this issue is probably to think of a country as an aggregate of individuals. If an individual’s earnings increase as a result of knowledge accumulation, either in school or on the job (Card 2001, Mincer 1989), then if the overall accumulation of knowledge of a group of individuals (e.g. all inhabitants of a country) increases, so must their combined earnings. By definition, their combined earnings, or the total output of the economy, are equal to GDP. In addition much of the variation is likely to be averaged out when focusing on populations as a whole (Stevens & Weale, 2003).
2.2 Hypotheses
role of human capital as a condition for the absorption and the creation of new technologies is more important than its role in production. Finally, Stevens & Weale (2003) and Benhabib & Spiegel (2004) argue that any positive influence on economic growth may be particularly pronounced for developing countries due to the catch-up effect. Meaning that countries with high initial levels of development tend to grow slower than countries with lower initial levels. This also implies that a country may expect growth in a subsequent period to be lower than in the current period. Adding to the literature, the aim of this thesis is to find evidence of the impact of the human capital, covering both educational attainment and work experience, on economic growth. This leads to the following hypotheses:
Hypothesis 1
Levels of human capital have a significant positive effect on economic growth.
Hypothesis 2
An increase in educational attainment has a significant positive effect on economic growth.
Hypothesis 3
An increase in work experience has a significant positive effect on economic growth.
Hypothesis 4
3 Data & Methodology
This section will describe the methodology of earlier studies and the data and methodology used in this thesis.
3.1 Data
The initial sample consists of all Barro & Lee (2000) countries. Similar datasets are available however none is as extensive9. The Barro & Lee dataset totals 142 countries
in the period 1960 to 200010 and, most importantly, not only provides information on
average schooling years, but also includes percentages of those who have attained or completed each level of schooling. Thus, the initial sample consists of all countries in the Barro & Lee (2000) dataset in the period 1960 to 200011. All countries with less than three subsequent observations are removed from the sample. Other data necessary to conduct this research is among others obtained from the Organisation for Economic Co-operation and Development (OECD); the Worldbank; the United Nations Educational, Scientific and Cultural Organization (UNESCO); and Penn World Tables 6.2. These sources provide a consistent and reliable database including various topics and many countries.
Data about educational attainment and schoollife expectancy are extracted from Barro & Lee (2000), which is available through the Worldbank. Demographic data is obtained from the U.S. Census Bureau, the United Nations World Population Prospects and UNESCO. Real GDP per capita (PPP) growth rates are based on constant 2000 International dollars and are extracted from Penn World Tables 6.2. These tables also provided data on government consumption and trade openness. Data on gross fixed capital formation and the rate of inflation are obtained from the
Worldbank12 and the OECD. Appendix 7.1, table 1 provides an overview of the
countries in the sample. Appendix 7.3, table 2 presents the variable definitions and their sources. Table 3 proposes expected coefficient signs. Finally, appendix 7.4, table
9
See for example Kyriacou (1991) and Nehru et al (1995) who have constructed datasets on average schooling years.
10
In their most recent paper Barro & Lee provide an update of educational attainment up to the year 1995. A projection is constructed for the year 2000. Some countries have also data available for the year 1955. Considering their small number, this period is left out of the sample.
11
Considering the use of 5 year periods, the data actually covers the period 1960 to 2004.
12
4a, 4b and 5 present the descriptive statistics of developed and developing countries13 and a correlation matrix, respectively.
3.2 The Economic Model
The economic model that is used in this thesis is based on a standard way to model economic growth, namely the Cobb-Douglas production function, augmented with human capital14, which is given by formula (1). For a thorough derivation of the model please see appendix 7.2.
β α β α − − = ( ) ( ) [ ( ) ( )]1 ) (t K t H t At L t Y , (1)
where Y is output, K is physical capital, L is labour and A represents a technology factor. H represents human capital and t denotes time. The above implies that the production function exhibits constant returns to scale. Since it is non-linear, logs of both sides are taken (log-log model), resulting in the following equation15:
+ + + = −ln( − )] ln( − ) ln( ) ) [ln(GDPCi,t GDPCi,t 1 α ϕ1 GDPCi,t 1 ϕ2 GFCFi,t ϕ3ln[n+g]+ϕ4PROFi,t−1+ϕ5EXPi,t−1 +ϕ6[ln(PROFi,t)−ln(PROFi,t−1)]+ ϕ7[ln(EXPi,t)−ln(EXPi,t−1)]+ϕ8Xi,t +ϕ9DVi,t +εi,t (20)
where i is the country index, t is the year index and εi,t denotes the error term. Xi,t
denotes the control variables. The overall model is based on that of Barro (2001), who
13
The tables display the descriptive statistics of the variables after the removal of outliers. Outliers are removed according to a simple percentile rule, i.e. all entries below the first and above the 99th percentile are removed from the sample.
14
The Cobb-Douglas functional form of production functions is used to represent the relationship of an output to inputs. It was proposed by Knut Wicksell (1851-1926) and tested against statistical evidence by Paul Douglas and Charles Cobb in 1928. For production, the function is: Y = ALαKβ. Where: Y = total production (the monetary value of all goods produced in a year); L = labour input; K = capital input; A, α and β are the output elasticities of labour and capital, respectively. These values are constants determined by available technology. Source: Wikipedia.
15
seems to be able to explain much of the variation in the observations with his model. The different variables of the model are discussed hereafter, however, for a short overview, including their definitions, sources and expected coefficient signs please see appendix 7.3, table 2.
The main dependent variable of this study is economic growth, as measured by the difference in logs16 of real gross domestic product (GDP) per capita (PPP, 2000 international $). Real terms assure that the effect of inflation is controlled for. Per
capita measures remove the direct effect of population growth17. GDP measures based
on Purchasing Power Parity (PPP) are considered to be better measures of overall well-being, since it also takes account of non-traded goods which tend to be relatively cheap in lower-income countries and hence may be especially preferred when the sample consists of both developed and developing countries. Moreover, PPP exchange rates are usually relatively stable over time which will allow for a better estimation of the relationship between the dependent variable and the independent variables.
The difference in log PROFi,t captures the effect of a change in the educational
attainment of the population. Educational attainment in this respect refers to the percentage of the population which has completed tertiary education. The threshold of the level of education is of course somewhat arbitrary (earlier studies have usually considered the secondary level of education, either on the basis of enrolment or completion rates). However, by choosing to focus on only the completion rate of the highest level, this thesis may be better able to capture the qualitative effect of education on the accumulation of human capital. Following the literature, it is expected that the change in educational attainment positively affects economic growth.
The difference in log EXPi,t captures the effect of a change in work experience on
economic growth. Observations are derived using the common definition of potential work experience (Reimers & Garvey, 1979). Work experience is expected to have a positive effect on the dependent variable.
16
Note that the dependent variable thus takes the form of log(GDPCi,t) - log(GDPCi,t-1) = log(GDPCi,t /
GDPCi,t-1). The log of such a ratio approximates the percentage change over time. 17
The log of gross fixed capital formation (GFCFi,t) equals s, or investment, in the
above discussion. However, since it also includes depreciation the separate term δ is removed in the final model. GFCF is measured as the average rate during the period and it is expected to have a positive effect on growth.
Initial conditions are controlled for using the initial level of real GDP per capita. Benhabib & Spiegel (1994) argue that this allows the model to capture the effects of technology and the catch-up effect. Barro (2001) shows that for poorer countries the marginal effect of income on the growth rate tends to be small but may be positive, whereas for richer countries this effect is strongly negative. This last result leads to the transformation to squared initial GDP levels for developed countries.
Benhabib & Spiegel (1994) find that initial (i.e. stock) levels of human capital enter significantly in the regression, which they contribute to its relation with technology. Similarly, Florida (2002) finds that a high level of tertiary graduates is a key variable in attracting high-technology industries. Since human capital in levels may affect the speed of adoption of technology and thus that the stock of human capital is related to the level of technology accumulated in the past, high initial levels of human capital suggest a higher level of technology and consequently higher growth rates.
Next, the average rate of government consumption as a percentage of GDP is included. Barro (2001) finds that an increase of 10 percent in government consumption reduces growth by 1.6 percent per year. Indeed, Barro and Sala-i-Martin (1995), argue that government consumption may proxy for political corruption, bad government and the direct effects of non-productive public expenditures. Furthermore, a large component of government expenditures are wages and salaries and they have been showed to be unambiguously associated with lower growth. Finally, it is argued that private investments are crowded out by government expenditures, further reducing growth.
In addition, the inflation rate is included to control for macroeconomic instability. Higher inflation tends to reduce growth.
(2001), however, argues that the effect of trade openness diminishes as a country becomes richer, therefore, trade openness is interacted with the log of initial GDP for developed countries.
Finally, a development dummy, DVi,t is included to differentiate between developed
and developing countries. Instead of the usual Worldbank classification18, countries are categorized according to the results of Krueger & Lindahl (2001), meaning that countries are categorised according to average schooling years. This may help to provide even further insights in the role of human capital on growth. The development dummy is expected to show a negative relationship with the dependent variable.
3.3 Panel Data
Panel data analysis may produce considerably different results than simple cross-sectional analyses. There are several benefits associated with the use of panel data (see also Brüderl, 2005). For one, using panel data allowance can be made for differences in the aggregate production function of countries when national but unobservable country effects exist. Second, panel data can significantly improve the power and efficiency of the estimates by increasing the number of observations and, consequently, the number of degrees of freedom, which may reduce issues concerning collinearity and variability among the explanatory variables. In addition, panel data may provide information about the time-ordering of events and the speed of adjustment to economic policy changes. Finally, panel data controls for unobserved heterogeneity and is better able to control for the effects of missing or unobserved variables.
A word of caution is in order when short time series are used as they may severely mitigate the gain from using panel data. Still, it is clear that panel data is preferred above cross-sectional data. Whereas cross-sectional data only provides a true causal effect when no differences between cross-sections exists, panel data uses within estimation and possibly between estimation. As such, panel data analysis can control for omitted or unobserved variables.
18
In the Worldbank classification, economies are divided according to GNI per capita, calculated using the World Bank Atlas method. The groups are: low income, $935 or less; lower middle income, $936 - $3,705; upper middle income, $3,706 - $11,455; and high income, $11,456 or more. Source:
3.4 Estimation
The statistical software used to produce the estimations is EViews 5.0. Reference is made to appendix 7.4. Note that the main discussion draws from Bond, Hoeffler & Temple (2001).
A few problems have been of most concern in previous research. One issue is simultaneity. Especially educational attainment is expected to improve economic growth but is in turn also expected to be positively influenced by economic growth. The rationale behind this is that part of the earnings of the growing economy will be spend on education, inherently increasing average educational attainment. In addition, in growth theory in general the believe exists that the growth rates of nations tend to convergence, meaning that less developed countries tend to catch-up with leading countries. One common way to deal with this factor is to control for initial conditions, i.e. the use of a lagged dependent variable, making the model dynamic19. It is important that the chosen methodology is able to deal with these issues.
Hence, in line with many of the more recent studies estimation is done by means of the Generalised Methods of Moments (GMM) method. Indeed, also according to Judson & Owen (1999) GMM estimation is the preferred estimation procedure for the type of panel data used in this thesis (i.e. short, unbalanced time dimension and fairly large cross-sectional dimension) because it creates the least biased estimates. The advantages of GMM estimation are that it is able to control for country-specific effects and simultaneity as a consequence of possible endogeneity between the dependent and independent variables and is also able to handle dynamic models. Moreover, it is robust to violations of homoskedasticity and normality and it has no problems with the dynamic nature of the regression equation.
To begin with, first-differences are taken to remove the unobserved time-invariant country-specific effects and then to instrument the right-hand-side variables in the first differenced equations using levels of the series lagged two periods or more (it is assumed that the error term is not serially correlated). However, if the instrumental variables are weak, as in the case of persistent time series but only few observations, large finite sample biases can arise20. Unfortunately, output is known to be a highly
19
Note that this is already discussed in section 3.2: The Economic Model.
20
persistent time series. Hence, it is advocated to use a system GMM in stead. The system consists of equations in both differences with lagged levels as instruments and levels with lagged first-differences as instruments. Based on the above, the following system is estimated: + + = −ln( − ) ln( − ) ln( ) ) ln(GDPCi,t GDCPCi,t 1 ϕ1 GDPCi,t 1 ϕ2 GFCFi,t 1 , 5 1 , 4 3ln[n+g]+ϕ PROFit− +ϕ EXPit− ϕ +ϕ6ln(PROFi,t)−ln(PROFi,t−1)+ t i t i t i EXP X EXP, , 1 8 , 7ln( ) ln( ) ϕ ϕ − − + +ϕ9DVi,t +εi,t + ∆ + ∆ = − ∆ln(GDPCi,t) ln(GDCPCi,t−1) ϕ1ln(GDPCi,t−1) ϕ2 ln(GFCFi,t) 1 , 5 1 , 4 3∆ln[n+g]+ϕ ∆PROFit− +ϕ ∆EXPit− ϕ +ϕ6∆ln(PROFi,t)−ln(PROFi,t−1)+ t i t i t i EXP X EXP, , 1 8 , 7∆ln( )−ln( − )+ϕ ∆ ϕ +ϕ9DVi,t +εi,t
The instrumental variables for the first-differences equation are lagged twice and the instrumental variables of the levels equation are the most recent lagged differences of the variables.
A Sargan test, in EViews represented by the J-Statistic, is used to test the validity of the additional instruments and the second-order serial correlation test to test for the existence of autocorrelation in the residuals21. Robustness of the results is tested using different numbers of control variables. Multicollinearity is checked for using a correlation matrix. Finally, as already mentioned, GMM is able to deal with endogeneity and stationarity so no allowance for this is needed.
21
4 Results
The regression outputs are shown in table 6abc and 7 of appendix 7.4. The remainder of this section will further discuss potential causes and implications of the results obtained.
4.1 OLS Estimation
Appendix 7.4, table 6abc list the results obtained using OLS estimation procedures. It is worthy to note that OLS estimation will produce unbiased estimations, however, considering the inclusion of a lagged dependent variable, the estimations are not BLUE22. For the same reason, the Durbin-Watson statistic does not hold. However, results are provided for illustrative purposes. The Fixed effects OLS estimator using the Seemingly Unrelated Regression (SUR) model corrects for both period heteroskedasticity and general correlation of observations within a given cross-section. As expected the rate of gross fixed capital formation is significant and with the expected positive sign in every regression. Its coefficient seems to be consistently higher in the developed sample, implying a greater importance of physical capital in more developed countries. Also, government consumption as a percentage of GDP is estimated to have its expected negative sign and is significant. This suggests that government consumption indeed has a negative influence on economic growth. The inflation rate has its expected negative sign, however, fails to be significant. Similarly, the extent of trade openness has its expected positive sign for developing countries, however fails to be significant in this case. On the other hand, it has an unexpected significant positive sign for developed countries and a significant positive sign in the complete sample. Initial real GDP per capita is significant for all samples, always with a negative sign. Finally, the estimated effects of human capital on economic growth show that work experience consistently performs better in the regressions. Whereas the stock of the number of years of work experience is significant in all samples (although with a negative sign for developed countries) and the differences in logs of years of work experience for developed and developing countries separately, only the
22
stock of professionals is significant for developing countries, however here it has the wrong sign. The overall better performance of stock levels was already somewhat highlighted by Benhabib & Spiegel (1994). That differences mostly fail to enter significantly, may also be due to longer implementation periods. It may simple last longer than the chosen period of 5 years for the economy to adjust and fully benefit from an improvement in human capital.
4.2 GMM Estimation
Table 7 of appendix 7.4 reports the results of the GMM estimation procedure. The GMM estimates are robust to heteroskedasticity, contemporaneous correlation and
autocorrelation of unknown form23. The Durbin-Watson statistic and R² are not
reported since they do not hold in a dynamic model.
Interestingly, once GMM estimation is used the results change drastically in favour of human capital. This in contrast to much of the earlier research done. The results indicate that the rate of investment is highly significant and with the expected positive sign. Note that the estimated effects are greater than the effects of any other variable and that the effect is especially large for developed countries. This last result may be explained by the fact that in developed countries additional physical capital can be operated with enough additional human capital so as to reap its maximum benefit. Indeed, Lucas (1990) suggests that physical capital has trouble flowing to developing countries due to their poor endowment with factors complementary to physical capital. As such the marginal product of physical capital is not that high, despite its scarcity.
Furthermore, the log of population growth added with an assumed technology growth rate enters with a negative sign in the developed sample, enters insignificantly in the developing sample and enters with an unexpected positive sign in the complete sample. However, also here it fails to enter significantly. The results may be explained by the believe that population growth has a dual effect on economic growth. In the short run, population increases are detrimental to a nation’s economy due to the pressure that is put on natural resources24. In the long run, however, population
23
QMS, “EViews 5 User’s Guide”, p. 688.
24
This vision was already advocated by Thomas Malthus in 1798 in his work: “An Essay on the
growth and ultimately an increased labourforce may lead to lower labour costs, shifting the usage of money from wages into advancement through technology, leaving the country better off than if the problems never occurred (Kothare, 1999). Similarly to the results of Barro (2001), government consumption also enters significantly and with the expected negative sign. Government consumption may hinder economic growth by pushing aside more productive private investment. Government consumption may further hinder economic growth by, as argued for by Barro and Sala-i-Martin (1995), serving as a proxy for political corruption and bad government.
The inflation rate does have the correct sign, however, fails to enter significantly. Khan & Senhadji (2001) find that certain thresholds levels of inflation exist. Meaning that inflation only starts to be harmful to an economy at levels above these thresholds. For developed countries, they find, this threshold is estimated to lie between 1%-3%. For developing countries this threshold may lie at rates as high as 11%-12%. Considering these relatively high thresholds, this may help to explain why inflation fails to enter significantly.
Trade openness (interacting with initial GDP per capita levels) shows to have no real effect on growth. It does enter significantly at the 10 percent level for developed countries, however, only with a very modest effect. Most importantly, here it enters with a positive sign which is clearly not in line with expectations based on the findings of Barro (2001).
Initial GDP levels enter highly significant with a negative sign in each sample. In line with Barro (2001) and Benhabib & Spiegel (1994), this result provides evidence for the existence of convergence, meaning that countries may expect lower growth rates in subsequent periods.
The most important results pertain to the effects of human capital on economic growth. The results show that initial stock levels of human capital enter significant and with a positive sign, except for the stock of professionals in developing countries. The reason for this insignificance is probably the very low level of professionals in
Speculations of Mr. Godwin, M. Condorcet, and Other Writers." First published anonymously in 1798
most developing countries. With levels often even below 1 percent25 their influence may indeed be negligible. Recall that a nation’s ability to adopt and implement new technology from leading countries is theorised to be dependent on human capital stock levels. In other words, the positive influence in general of human capital in levels suggests that countries more abundant in human capital are better able to attract additional physical capital, tend to grow faster and close any technology gap faster than others, implying convergence. Finally, it is worthy to note that, except for developed countries, the stock of work experience has a greater influence on growth than the stock of professionals. Perhaps, this results can be explained by considering the advanced level and great importance of physical capital in developed countries. Very specific skills are needed to absorb and implement highly advanced new technologies from leading countries. Assuming that work experience is much more general by nature, work experience may simply not bring forth the desired skills needed.
In conclusion, the results do lend evidence, although no proof of a relationship between the stock of professionals and growth in developing countries is found, to support hypothesis 1; Levels of human capital have (in general) a significant positive
effect on economic growth.
Next, the effects of human capital in differences of logs. In contrast to the results of Islam (1995), differences in human capital are estimated to have a significant effect on economic growth. Differences in educational attainment enter positive and significant in all samples26, implying that educational attainment as a factor of production is positively related to economic growth. Hence, also hypothesis 2 is supported; Increases in educational attainment have a significant positive effect on
economic growth.
Regarding differences in work experience, only differences in the number of years of work experience in developing countries fails to enter significantly. However, the significant but negative relationship between changes in work experience and economic growth in developed countries is perhaps most surprising. The negative sign
25
The average percentage of professionals is only 2.26 percent in developing countries as compared to 3.43 percent in the complete sample and even 7.51 percent in the developed sample.
26
indicates that an additional year of average work experience will actually considerably lower economic growth in developed countries. Lee (2007) and Brunow & Hirte (2006) suggest that this remarkable result is partly due to age structures. More developed countries are characterised by a contracting age pyramid; i.e. low birth rates, low death rates, longer life expectancy and a higher dependency ratio (i.e. ratio of non-working age population to working age population). As such, high observations of work experience are mainly caused by an ageing population. In this context, Lee (2007) separates two periods of ‘demographic dividend’ that are associated with changing age structures. First, as the fertility rate slows down, the working age population will grow faster than the consuming population (hence, a lower dependency ratio), leading to a boost in per capita GDP levels. However, this effect is only temporarily and eventually per capita income will decrease again. De second demographic dividend occurs when the mortality rate falls. Longer life expectancy should urge the need to accumulate more wealth during a lifetime (since longer life requires increased savings for retirement). Whether this wealth accumulation is indeed realised depends to a large extend on the institutional context of a country and to the extent to which it favours people’s reliance on savings for old age support. In so far as older persons depend on family transfers or, in particular, public pensions the second dividend is reduced and may potentially not even lead to an increase in per capita income levels. Considering that many of today’s developed countries have refrained from encouraging asset accumulation rather than transfers for old age support, the second dividend is greatly reduced. Additional years of work experience (implying an older population) will then only put more stress on transfer programmes, leading to lower growth rates. Similar suggestions have been made by Hall & Jones (1999). In conclusion, since differences in work experience fail to enter significantly for developing countries and significant but strongly negative for developed countries, hypothesis 3 is not supported by the results; increases in work
experience do not have a significant positive effect on economic growth.
the coefficient has a negative value. Hence, although its relative effect is quite small, policymakers will always opt for increasing the percentage of professionals in order to boost growth. Results, in this light, support hypothesis 4; The effect of additional
educational attainment on economic growth is more positive than the effect of additional work experience on economic growth.
As a final note, it may be interesting to generalise the results obtained and discuss how they may relate to the already mentioned war for talent. After an initial study of McKinsey & Company, a great run on talent was created with as a result an incredible earnings premium of up to 100 percent for obtaining a tertiary degree. Work experience in this context seems to be neglected.
This war for talent would lead us to think that (young) professionals have higher productivity levels or other profit-generating capabilities than people with an
abundance of work experience27. One rationale is that professionals contribute greatly
to the innovative powers of a firm or country, simply just because of their lack of work experience. This lack of experience has not yet bounded their minds to the common ways of thought that have shaped themselves within a certain group and therefore makes them able to think ‘out-of-the-box’. On the other hand, some authors seriously doubt whether the war is justified. Still, it seems difficult to believe that the current war for talent exists without a reason. Something must have urged human resource managers to start looking at a different pool of potential employees. Perhaps, this trend started with Enron, which wanted to be the ultimate talent company. This company decided to bring in only the best graduates in order to endow Enron with as much talent as possible, with self-proclaimed success (Gladwell, 2002). Eventually this company completely collapsed because of far more complex reasons than its human resources policy. Nevertheless, it remains interesting to find out whether its human resource policy may have had perhaps a slight influence on the failure of this company. In other words, what if talent is overrated?
Since, a common proxy for talent in the literature is educational attainment (Li & Florida 2006, Stevens & Weale 2004, Florida 2002, Barro & Lee 2000, Becker & Chiswick 1966), the results obtained through this study, especially those concerning hypothesis 4, may give an answer to that question.
27
For developing countries, the fact that a change in the number of years of work experience fails to enter significant leads to the conclusion that indeed professionals contribute more to a company than experienced employees. For developed countries the same conclusion is reached. However, the main reason here is the considerable negative impact of a change in the number of years of work experience. Nationwide this negative effect is somewhat retarded in the long run given that the stock of work experience will boost growth. Although it can be assumed that individual companies will also profit from the technological advancements made, this effect is still not large enough to compensate for initial losses.
5 Summary & Conclusions
Empirical research has failed to uniformly prove the relationship between human capital and economic growth. However, as Stevens & Weale (2003) point out, the fact remains that people with higher individual levels of human capital are paid more than
people with lower human capital levels28. With the reasonable assumption that a wage
rate is based on productivity, or someone’s marginal product, even the simplest model implies a growth of output following an increase in the overall level of human capital. This observation has such generality, that any conclusion of an absence of such an effect has to be considered to be a fundamental flaw of the way empirical work is carried out. Indeed, the dissent in earlier research is likely to stem from the many difficulties in estimating the relationship between human capital and economic growth. Research has often had to deal with the non-availability of data or imperfect proxies. With such detailed datasets, as for example the Barro & Lee dataset, approximations have improved considerably. However, still it is by no means perfect and much discord still exist about how human capital ought to enter the production function.
Fortunately, making use of the advancement in data and theory to derive a more complete definition of human capital, the results reported in this thesis do provide some evidence for the natural perception of a positive relationship between human capital and growth. Except for the level of professionals in developing countries, stock levels of human capital show to be positively related to growth, generally supporting hypothesis 1. Moreover, the results show indisputably that an increase in the percentage of professionals will boost growth, supporting also hypothesis 2. Hypothesis 3 is not supported by the data, particularly due to a strong negative relationship between additional years of work experience and growth in developed countries. Finally, based on the above, also hypothesis 4 is supported. A first generalisation of these results may lead to a justification of the current war for talent.
Limitations
Most of the hypotheses tested in this thesis are supported to some extent, however, some limitations of this thesis should be taken into account. For one, the dataset
28
should be extended to include more observations. In particular the sample of developed countries is by nature very small, a larger time period should be able to improve estimations. Second, although much improvement is achieved, perfect proxies for human capital are still not formulated. The definition of educational attainment used in this thesis, as measured by the percentage of employees who have enjoyed and completed tertiary education, may not perfectly measure the actual set of skills and knowledge and the aptitude that individuals have effectively attained at school. Similarly, potential work experience is expected to differ from actual work experience. Applying an even more advanced measure of human capital, such as the IALS dataset, in a similar regression may considerably improve estimation.
Brunow & Hirte (2006) further found that different age structures have different effects on per capita growth and that the fit of the regression model is considerably improved after including age structures. Also Lee (2007) argues that the age structure of a country may be able to positively influence growth under certain circumstances and Reimers & Garvey (1979) state that the extent of deviation between actual and potential work experience depends on demographic characteristics, among others age structures. Manuelli & Seshadri (2005) find in fact that if a country in the lowest decile of the world income distribution was endowed with the demographic characteristics of the representative country in the top decile, output per worker would
double29. Hence, this model may be limited by the lack of age structures included.
Recommendations
Research in this field is still very much in development. Future research must try to make use of more reliable and internationally comparable proxies that lie closer to reality and specifically take into account the quality of human capital. Age structures should be included, making it possible to test what effect age structures (in particular in interaction with human capital) have on economic growth. In addition, it should be interesting to see how differences in institutional environment and policies effect growth. Finally, the number of observations should be increased.
29
6 References
• Barro, R.J., “Human Capital and Growth” (2001), The American Economic
Review, Vol. 91, No. 2, Papers and Proceedings of the Hundred Thirteenth
Annual Meeting of the American Economic Association, (May, 2001), pp. 12-17.
• Barro, R.J., Lee, J-W. “International Data on Educational Attainment:
Updates and Implications”, CID Working Paper, No. 42, April 2000.
• Barro, R., and Sala-i-Martin, X. “Economic Growth” (1995), McGraw-Hill, New York.
• Becker, G.S., Chiswick, B.R. “Education and the Distribution of Earnings” (1966), American. Economic Review, May 1966, Vol. 56, No. 1/2, pp 358-69. • Benhabib, J., Spiegel, M.M., “The Role of Human Capital in Economic
Development: Evidence from Aggregate Cross-country Data” (1994), Department of Economics, New York University, New York, NY 10003,
USA.
• Bijlsma, E. “Oudere werknemer vitaler”, De Pers, March 14, 2008.
• Bond, S., Hoeffler, A., Temple, J. “GMM Estimation of Empirical Growth
Models” (2001), Economics Papers, Economics Group, Nuffield College,
University of Oxford, No. 2001-W21.
• Brüderl, J. “Panel Data Analysis” (2005), University of Mannheim, March. Available from the Universität Mannheim site, http://www.sowi.uni-mannheim.de/lehrstuehle/lessm/veranst/Panelanalyse.pdf
• Brunow, S., Hirte, G. “Age Structures and Regional Economic Growth” (2006), Jahrbuch für Regionalwissenshaft, No. 26, pp. 3-23
• Card, D. “Estimating the Return to Schooling: Progress on Some Persistent
Econometric Problems” (2001), Econometrica, Vol. 69, No. 5 (Sep., 2001),
pp. 1127-1160.
• Easterlin, R.A. “Why Isn’t the Whole World Developed?” (1981), The Journal
of Economic History, Vol. 41, No. 1, The Tasks of Economic History (Mar.,
1981), pp. 1-19.
• Ericsson, K.A., Prietula, M.J., Cokely, E.T. “The War for Talent” (2007),
Harvard Business Review, July - August 2007.
• European Union “General Report on the Activities of the European Union”, General Report 2005, chapter II, section 1. Available from the European Union site, http://europa.eu/generalreport/en/2005/rg38.htm.
• EViews 5 User’s Guide, Copyright © 1994–2004 Quantitative Micro Software, LLC.
• Fishman, C. “The War for Talent” (1998), Fast Company Magazine, Issue 16, (July 1998), pp. 104.
• Florida, R. “The Economic Geography of Talent” (2002), Annals of the
Association of American Geographers, Volume 92, Number 4, December
2002, pp. 743-755.
• Gladwell, M. “The Talent Myth” (2002), Department of Human resources,
The New Yorker, July 22, 2002.
• Gundlach, E., “Solow vs. Solow: Notes on Identification and Interpretation in
the Empirics of Growth and Development” (2005), Review of World Economics, Vol. 141, No. 3, Oct. 2005.
• Hall, R.E., Jones, C.I. “Why Do Some Countries Produce So Much More
Output Per Worker Than Others?” (1999), The Quarterly Journal of Economics, Vol. 114, No. 1, (Feb., 1999), pp. 83-116.
• Ischinger, B. “Can the world be over-educated?”(2007), OECD Observer, No 263, October 2007.
• Islam, N. “Growth Empirics: A Panel Data Approach” (1995), The Quarterly
Journal of Economics, Vol. 110, No. 4, (Nov., 1995), pp. 1127-1170.
• Judson, R.A., Owen, A.L. “Estimating Dynamic Panel Models: A Guide For
Macroeconomists” (1999), Economic Letters 65: pp. 9-15.
• Khan, M.S., Senhadji, A.S. “Threshold Effects in the Relationship Between
Inflation and Growth” (2001), IMF Staff Papers, Vol. 48, No. 1.
• Kothare, R. “Does India’s Population Growth Have A Positive Effect on
• Krueger, A. & Lindahl, M. “Education for Growth: Why and for Whom?” (2001), Journal of Economic Literature, 39, pp. 1101-1136.
• Kyriacou, G.A., “Level and Growth Effects of Human Capital: A
Cross-Country Study of the Convergence Hypothesis” (1991), C.V. Star Center for Applied Economics, New York University, May 1991, revised May 1992.
• Lee, R. “The Economic Implications of Changing Age Structures” (2007),
University of California at Berkeley, PPT, based on research supported by the
National Institute of Aging.
• Li, T., Florida, R. “Talent, Technological Innovation and Economic Growth in
China” (2006), Joseph, L. Rotman Institute of Economics, University of
Toronto, The Martin Prosperity Institute, February 2006.
• Lucas, R. E. “Why doesn't capital flow from rich to poor countries?” (1990),
American Economic Review, Papers and Proceedings, p. 92-96.
• Mankiw, N.G., Romer, R., Weil, D.N. “A Contribution to the Empirics of
Economic Growth” (1992), The Quarterly Journal of Economics, Vol. 107,
No. 2, (May, 1992), pp. 407-437.
• Manuelli, R.E., Seshadri, A. “Human Capital and the Wealth of Nations” (2005), Department of Economics, University of Wisconsin-Madison, May 2005.
• McKinsey & Company “Winning the War for Talent”, Wharton Leadership
Digest, McKinsey & Co., 1998.
• Mincer, J. “Human Capital and the Labour Market: A Review of Current
Research” (1989), Educational Researcher, Vol. 18, No. 4. (May, 1989), pp.
27-34.
• Mincer, J. “Schooling, Earnings and Experience” (1974), Columbia
University Press, New York.
• Nelson, R. & Phelps, E. “Investment in Humans, Technical Diffusion and
Economic Growth” (1966), American Economic Review, 56, pp. 69-75.
• Papageorgiou, C., Perez-Sebastián, F. “Matching Up the Data On Education
With Economic Growth Models” (2002), IVIE Working Papers, WP-AD
2002-23.
• Reimers, C., Garvey, N. “Toward a Better Measure of Work Experience” (1979), Industrial Relations Section, Princeton University, Working Paper #119.
• Rosen, S. “Learning and Experience in the Labor Market” (1972), The
Journal of Human Resources, Vol. 7, No. 3. (Summer, 1972), pp. 326-342.
• Stevens, P., Weale, M. “Education and Economic Growth” (2003), National
Institute of Economic and Social Research, August 2003.
• Teixeira, A.A.C., Fortuna, N. “Human capital, innovation capability and
economic growth in Portugal, 1960–2001” (2004), Portuguese Economic Journal, Spring-Verlag 2004.
• Topel, R. “The Labour Market and Economic Growth” (1999), The Handbook
of Labour Economics, North Holland, Amsterdam, Vol.3, P.3, Ch. 44.
• UNESCO, “International Standard Classification of Education” (1997).
Available from ENESCO site, http://www.unesco.org/education/
information/nfsunesco/doc/isced_1997.html.
• Valetta, R., Hodges, J. “Age and Education Effects on the Unemployment
Rate” (2005), FRBSF Economic Letter, 2005-15; July 15, 2005.
• Wolff, E.N., Gittleman, M. “International Comparisons of Inter-industry
Wage Differentials" (1993), Review of Income and Wealth, Blackwell
Publishing, Vol. 39, pp. 295-312, September.
Websites
• Wikipedia: http://en.wikipedia.org/wiki/Main_Page
• OECD: http://www.oecd.org/home/
• UNESCO: http://portal.unesco.org/education/en/
• US Census Bureau: http://www.census.gov/
• Worldbank: http://www.worldbank.org/
