education and economic growth a causal analysis

RICHARD GRABOWSKI This dissertation is an empirical exercise in a time-series framework, which seeks to analyze the causal relation between different levels of education and economic gro

This manuscript has been reproduced from the microfilm master UMI films

the text directly from the original or copy submitted Thus, some thesis and

dissertation copies are in typewriter face, while others may be from any type of computer printer

The quality of this reproduction is dependent upon the quality of the copy submitted Broken or indistinct print, colored or poor quality illustrations and photographs, print bleedthrough, substandard margins, and improper alignment can adversely affect reproduction

in the unlikely event that the author did not send UMI a complete manuscript and there are missing pages, these will be noted Also, if unauthorized copyright material had to be removed, a note will indicate the deletion

Oversize materials (e.g., maps, drawings, charts) are reproduced by sectioning the original, beginning at the upper left-hand comer and continuing from left to right in equal sections with small overlaps

ProQuest Information and Learning

300 North Zeeb Road, Ann Arbor, MI 48106-1346 USA

800-521-0600

® UMI

by Sharmistha Self,

B.A., Jadavpur University, 1984

B.Ed., Annamalai University, 1994

M.A., Jadavpur University, 1986

A Dissertation Submitted in Partial Fulfillment of the Requirements for the Doctorate of

Philosophy Degree

Department of Economics

in the Graduate School

Southern [Illinois University

at Carbondale April, 2002

ARCHIVES

UMI Number: 3065385

Copyright 2002 by Seif, Sharmistha

All rights reserved

®

UMI

UMI Microform 3065385 Copyright 2002 by ProQuest information and Learning Company All rights reserved This microform edition is protected against unauthorized copying under Title 17, United States Code

ProQuest Information and Leaming Company

300 North Zeeb Road P.O Box 1346 Ann Arbor, MI 48106-1346

Ì

Dissertation Approval The Graduate School Southem Mlinois University

April 5, 2002

T hereby recommend that the dissertation prepared under my supervision by

Sharmistha Self

Entitled Education and Economic Growth: A Causal Analysis

be accepted in partial fulfillment of the requirements for the

DOCTOR OF PHILOSOPHY degree

Head of Depariment Recommendation concurred in

SHARMISTHA SELF, for the Doctorate of Philosophy degree in ECONOMICS,

presented on 12/07/2001, at Southern Illinois University at Carbondale

TITLE: EDUCATION AND ECONOMIC GROWTH: A CAUSAL ANALYSIS

MAJOR PROFESSOR: DR RICHARD GRABOWSKI

This dissertation is an empirical exercise in a time-series framework, which seeks

to analyze the causal relation between different levels of education and economic growth This study targets twelve Asian countries categorized by geographic region and state of economic growth

The importance and positive influence of human capital on growth has long been established, though, the acceptance of this premise has not been unanimous There is even less agreement among the limited number of studies which have targeted the

different levels of education and/or the role of gender Levels of education, by definition, determine the depth of knowledge and hence are important tools for determining the extent of human capital formation in a country This study takes an in-depth look not just

at correlations but also tests the causal relationship between education at different levels and growth and vice-versa

Once education is segregated by levels, this study finds the correlations and causal relations to vary between levels of education and levels of growth of different countries Thus, in spite of education being a consumer good, different levels appear to have different appeals for a country, depending perhaps upon its economic needs, and, consequently, to the policy maker The results show that economic growth is caused by primary education of the labor force for the relatively poorer countries of South Asia

while the relatively wealthier economies of South East and East Asia are driven by tertiary education

Model specification with the inclusion of gender produces robust results, which show that gender participation at different education levels plays a causal role in growth, though the direction of impact varies between genders The results show that the role of gender at different education levels is linked to the level of development of a particular country or region rather than to the degree of gender participation in education

I dedicate this dissertation to my mother and father whose dreams I helped fulfill by embarking on this journey and whose faith in me prevented me from ever giving up.

LIST OF TABLES

Table 2.1: Per Capita Output (South Asia)

Table 2.2: Per Capita Output (South East Asia)

Table 2.3: Per Capita Output (East Asia)

Table 2.4: Growth Rates (South Asia)

Table 2.5: Growth Rates (South East Asia)

Table 2.6: Growth Rates (East Asia)

Table 2.7 Correlation with per capita growth (South Asia)

Table 2.8 Correlation with per capita growth (South East Asia)

Table 2.9 Correlation with per capita growth (East Asia)

Impact of Primary Education on Economic Growth in South Asia

Impact of Primary Education on Economic Growth in South East Asia Impact of Primary Education on Economic Growth in East Asia

Impact of Secondary Education on Economic Growth in South Asia

Impact of Secondary Education on Economic Growth in South East Asia Impact of Secondary Education on Economic Growth in East Asia

Impact of Tertiary Education on Economic Growth in South East Asia

Impact of Tertiary Education on Economic Growth in East Asia

Impact of Economic Growth Ratio on Primary Education in South Asia Impact of Economic Growth on Primary Education in South East Asia

Impact of Economic Growth on Primary Education in East Asia

Impact of Economic Growth on Secondary Education in South Asia

: Impact of Economic Growth on Secondary Education in South East Asia Impact of Economic Growth on Secondary Education in East Asia

Table 2.25: Impact of Economic Growth on Tertiary Education in East Asia

Table 2.26: Impact of Different Education Levels on Economic Growth

Table 2.27: Impact of Economic Growth on Education Levels

Table 2.28: Impact of Primary Education on Growth (Panel method)

Table 2.29: Impact of Secondary Education on Growth (Panel method)

Table 2.30: Impact of Tertiary Education on Growth (Panel method)

Table 2.31: Impact of Economic Growth on Primary Education (Panel method)

Table 2.32: Impact of Economic Growth on Secondary Education (Panel method)

Table 3.1 Partial Correlations with per capita growth (South Asia)

Table 3.2 Partial Correlations with per capita growth (South East Asia)

Table 3.3 Partial Correlations with per capita growth (East Asia)

Table 3.4: Impact of gender-based primary education on growth (South Asia)

Table 3.5: Impact of gender-based primary education on growth (South East Asia)

Table 3.6: Impact of gender-based primary education on growth (East Asia)

Table 3.7: Impact of gender-based secondary education on growth (South Asia)

77

78 8l

Table 3.10: Impact of growth on male primary education (South Asia)

Table 3.11: Impact of growth on female primary education (South Asia)

Table 3.12: Impact of growth male primary education (South East Asia)

Table 3.13: Impact of growth on female primary education (South East Asia)

Table 3.14: Impact of growth on male primary education (East Asia)

Table 3.15: Impact of growth on female primary education (South East Asia)

Table 3.16: Impact of growth on male secondary education (South Asia)

Table 3.17: Impact of growth on female secondary education (South Asia)

Table 3.18: Impact of growth on male secondary education (South East Asia)

Table 3.19: Impact of growth on female secondary education (South East Asia)

Table 3.20: Impact of growth on male secondary education (East Asia)

Table 3.21: Impact of growth on female secondary education (East Asia)

Table 3.22: Impact of primary education by gender on growth — Panel Method

Table 3.23: Impact of secondary education by gender on growth — Panel method

Table 3.24: Impact of growth on male primary education — Panel method

Table 3.25: Impact of growth on female primary education ~ Panel method

Table 3.26: Impact of growth on male secondary education ~ Panel method

Table 3.27: Impact of growth on female secondary education — Pane! method

Table 4.1: Gender-gap(%) in enrollment rates at primary level

Table 4.2: Gender-gap (%) in enrollment rates at secondary level

Table 4.3: Average Participation of Children in the labor force (%)

Table 4.4: Average Participation of Females in the labor force (%)

Table 4.5: Simple correlations between male and female enrollment rates and log

(per capita real gdp) in South Asia

Table 4.6: Simple correlations between male and female enrollment rates and log

(per capita real gdp) in South East Asia

Table 4.7 : Simple correlations between male and female enrollment rates and log

(per capita gdp) in East Asia

Table 4.9: Number of Co-integrating equations in models | and model 2

(primary level)

Table 4.10: Number of Co-integrating equations in models | and model 2

(secondary level!)

Table 4.11: Direction of Long-run Impact on log(per capita output)

Table 4.12: Short-run Causal Relations: South Asia (Model 1!)

Table 4.13: Short-run Causal Relations: South Asia (Model! 2)

Table 4.14: Short-run Causal Relations: South East Asia (Mode! 1)

Table 4.15: Short-run Causal Relations: South East Asia (Model 2)

Table 4.16: Short-run Causal Relations: East Asia (Model 1)

Table 4.17: Short-run Causal Relations: East Asia (Model 2)

LIST OF FIGURES Figure 2a: Primary Enrollment Rate in South Asia

Figure 2b: Primary Enrollment Rate in South East Asia

Figure 2c: Primary Enrollment Rate in East Asia

Figure 2c: Pupil Teacher Ratio at Primary Level in South Asia

Figure 2d: Pupil Teacher Ratio at Primary Level in South East Asia

Figure 2e: Pupil Teacher Ratio at Primary Level in East Asia

Figure 2f: Expenditure Per Student in South Asia (% of GNP)

Figure 2g: Expenditure Per Student in South East Asia (% of GNP)

Figure 2h: Expenditure Per Student in East Asia (% of GNP)

Figure 2i: Secondary Enrollment Rate in South Asia

Figure 2j: Secondary Enrollment Rate in South East Asia

Figure 2k: Secondary Enrollment Rate in East Asia

Figure 21: Pupil Teacher Ratio at Secondary Level in South Asia

Figure 2m: Pupil Teacher Ratio at Secondary Level in South East Asia

Figure 2n: Pupil Teacher Ratio at Secondary Level in East Asia

Figure 20: Enrollment Rate at Tertiary Level in South East Asia

Figure 2p: Enrollment Rate at Tertiary Level in East Asia

Figure 3a: Female Primary Enrollment Rate in South Asia

Figure 3b: Male Primary Enrollment Rate in South Asta

Figure 3c: Female Primary Enrollment Rate in South East Asia

Figure 3d: Male Primary Enrollment Rate in South East Asia

Figure 3e: Female Primary Enrollment Rate in East Asia

Figure 3g: Female Secondary Enrollment Rate in South Asia

Figure 3h: Male Secondary Enrollment Rate in South Asia

Figure 3i: Female Secondary Enrollment Rate in South East Asia

Figure 3j: Male Secondary Enrollment Rate in South East Asia

Figure 3k: Female Secondary Enrollment Rate in East Asia

Figure 31: Male Secondary Enrollment Rate in East Asia

Figure 4a: Per Capita Real GDP

1.4 Education Variables and Data Sources

1.5 Precedence of Education Level for Growth in Developing Countries

1.6 The Role of Gender in Explaining the Relation between Education

2.2 Data and Methodology

2.3 Education Patterns n Asia

3.2 Data and Methodology

3.3 Male and Female Enrollment Rates - some background

165

179 184

5.3 Individual Chapter Results — a summary

194

196 197

CHAPTER 1: INTRODUCTION

1.1 BRIEF SURVEY OF LITERATURE

A sea of literature exists explaining the relationship between human capital and growth The importance of human capital in promoting growth received a lot of attention

in the 1960s ( Bowman, 1966; Schultz, 1961; Denison, 1967) A survey of the literature

involving growth research reveals that education has been one of the primary components

of human capital as it has the additional quality over other traditional human capital measures of being able to increase the ability and willingness of people to live healthier

lives and be better on-the-job learners after entering the labor force.'

Most New Growth theories, which are primarily responsible for emphasizing the

positive relation between human capital and growth, have suggested that education has a

direct and positive effect on the economic growth of a country (Azariadis and Drazen, 1990; Mankiw, Romer and Weil, 1992) These theories have highlighted the positive externalities and related endogenous growth possibilities associated with education

These are a result of long run increasing returns induced by endogenous technological!

' Nehru, Swanson and Dubey, ~A New Database on Human Capital Stock in Developing Countries and Industrial Countries : Sources, Methodology and Results”, 1995, Journal of Development Economics, vol

46(2), p.379-401

educated workers in society Long run growth is achieved primarily by the knowledge accumulated through education and training (Romer, 1986) Some theories emphasize how convenient it becomes to accumulate further knowledge once a certain critical or threshold level of knowledge has been achieved and even calculate it to be a literacy rate

of forty percent (Azariadis and Drazen, 1990) Thus, following this train of thought, developing countries may experience an initial lull in growth until that critical threshold

is attained (Tallman and Wang, 1992) However, even though all authors on this topic

generally accept the positive impact of education on growth, not all of them agree on the significance of education for a country’s growth rate

Most of the empirical work supporting New Growth literature has been based on

evidence provided by cross-country studies (Romer, 1989; 1990; Barro, 1991; Lau,

Jamison and Louat, 1991; Scott, 1991, Benhabib and Spiegel, 1992; Barro and Lee, 1993: Gundlach, 1993; Baffes and Shah, 1993; Ahuja and Filmer, 1995; Ruth Judson, 1996)

All these studies point to the importance of investment in human capital in general and

education in particular as a means to attain economic growth However when one breaks down education into levels, such as primary, secondary and higher education, there is a lack of research and even among existing literature, there does not seem to be any

consensus regarding the benefits of education in levels for growth While Barro (1997) concludes that primary and college level education has a negative and insi gnificant impact on growth, Sachs and Warner (1995) find a positive, but still insignificant, impact

of both primary and secondary education on growth For this research project, the focus is

not on education as a general concept but education pertaining to different levels, since these determine the degree of knowledge and skill being imparted and thereby account for the extent of human capital formation of the population

Another interesting issue concerns the role of gender in the relationship between education and economic development Cross-country data shows that overall private retums to investment in education is marginally higher for women than men

(Psacharopoulos, 1994) while social returns to investment are significantly higher for

women than men (United Nations, Fertility Behavior in the Context of Development,

1987) This information seems to imply a stronger link between women’s education and economic development as compared to that of men Results of studies that have

researched the impact of female education on level of output or growth are mixed According to Barro and Lee (1994) and Forbes (1997), this impact is negative and

significant, while Casselli, Esquivel and Lefort (1996) find the impact of female

education to be positive and significant According to Barro and Lee (1994) ˆ high spread between male and female (education) attainment is a good measure of backwardness” Some authors, however, do not share this view Thus, a country-by-country study of the impact of different levels of education (based on gender) on growth will be able to clarify this relationship and consequently, be able to address issues relating to the role of gender

in policy pertaining to educational spending

What seems amply clear and obvious from the above discussion is that there is enough evidence, based on cross-country regression analysis, to support some association between education and economic growth However in spite of these earlier studies, there

or that changes in education “cause” changes in growth Though education has

undoubtedly been an important contributing factor of economic development, the

literature does not support a cause-and-effect relationship running from education to

economic growth or vice versa It is quite possible that changes in growth could have reacted passively and yet concurrently with advances in education The identification of causal relationships among a set of variables is among the major goals of empirical research, which has recognized that a high correlation among variables does not in any way imply that they are causally related Correlation occurs because of a linear

association while causality is a result of common association of each variable with additional factors.”

The empirical literature seems to assume a causal relation from education to economic growth and rejects a causal relation from growth to education However, there

is some evidence of a feedback from growth to education as well For example, a study

by Pritchett (1996) shows that once a certain level of growth has been achieved,

additional investments in education can reverse the positive effect of education to negative The study by Bils and Klenow (2000), finds that, in an overlapping generations framework, the impact of growth to education is much more significant and evident than the impact of education on growth It is seen that wealthier economies typically exhibit better education, which is made possible through more and better schools One also needs to realize at the same time that the level of growth itself is different between

* Pierce and Haugh (1977)

countries and this gives rise to different needs from the working population and

consequently, perhaps, the level of training or education they require Natural follow-on questions then are, depending upon the economic situation of a country, does education

at any particular level lead to higher growth, does growth lead to better education at a particular level, or whether both occur with or without some difference in timing? Depending on the answer to these questions, decisions, on not only funding education

itself, but also, which level to fund, arise

1.2 OBJECTIVES

The primary objectives of this research are as follows:

(1) To test whether education and growth are correlated over time and, in case this is found to be true, the direction and significance of this correlation

(2) To test whether educational quantity and growth have some “causal” relationship, with or without feedback, in the “Granger sense™

(3) To test whether educational quality and growth have some ‘causal’ relationship, with

or without feedback, in the “Granger sense”

(4) To test whether the relationships tested above also hold when a gender-based

connotation and distinction is introduced into the relationship

(5) To test whether all the results from all the above tests allow a link to be established based on the relationship between the level of education and the stage of economic development Additionally, whether this link, if found, also applies to a gender-based analysis

of growth That is, the quantity of primary, secondary, and/or tertiary education does not

“Granger cause’ economic growth against the alternative of the existence of a causal relation, in the ‘Granger’ sense, between the quantity of primary, secondary, and/or tertiary education and economic growth;

iii) In the presence of information on past levels of growth, additional information on qualitative aspects of past primary, secondary or tertiary level education measured by variables such as pupil teacher ratio and/or expenditure per student, does not improve the predictive powers of growth That is, quality of primary, secondary, and/or tertiary

education does not “Granger cause’ economic growth against the alternative of the

existence of a causal relation, in the “Granger’ sense, between the quality of primary,

secondary, and/or tertiary education and economic growth;

(iv) In the presence of information on past levels of primary, secondary, or tertiary level education such as enrollment rate, pupil teacher ratio and/or educational expenditure per student, additional information on past rates growth does not improve the predictive powers of education That is, economic growth does not “Granger cause’ primary,

secondary, or tertiary education against the alternative of the existence of a causal relation, in the ‘Granger’ sense, from economic growth to primary, secondary, and/or tertiary education If this holds it implies a feedback from growth to education:

v) In the presence of information on past levels of growth, additional gender-based information on past rates of primary or secondary level education such as enrollment rate does not improve the predictive powers of growth That is, male and/or female primary education or secondary education does not “Granger cause’ economic growth against the

alternative of the existence of a causal relation, in the ‘Granger’ sense, between male

and/or female education at different levels and economic growth:

vi) In the presence of past levels information on gender-based education such as past rates of primary or secondary level enrollment rate, additional information on past rates

of growth does not improve the predictive powers of gender-based education That is, growth does not “Granger cause’ male or female education at primary or secondary level against the alternative of the existence of a causal relation, in the “Granger” sense,

between economic growth and male and/or female education at different levels (if this holds, it implies feedback from growth to male and/or female education at different

income/output;

viti) Any causal relation found between education at different levels and growth is not related to different levels of growth itself against the alternative of the existence of some link between level of education and level of growth:

ix) Any causal relation found between education of males and females at different

levels and growth is not related to different levels of growth or output itself against the alternative of the existence of some link between level of education of males and/or females and level of growth or output

1.3.1 INTRODUCTION TO CAUSALITY

The concept of causality can be found in philosophical writings as early as Aristotle Much of this discussion of causality was deterministic in nature, related to specific examples taken from classical physics or chemistry, and consisting primarily of two cases, necessity (If A occurs B must occur) and sufficiency (If I did observe B

occurring, A must have occurred) These definitions, by their nature, were hard to apply

to statistics or the sciences The basic definition of causality, as found in the Encarta and widely accepted is: The cause of any event is a preceding event without which the event

in question would not have occurred

A good example, to explain the concept of causality, is the age-old debate over the proposition that smoking causes lung cancer In 1964, the Surgeon General had issued

a report, which linked cigarette smoking to cancer, particularly lung cancer and death This report was based on the fact that there existed a strong, established and proven correlation between smoking and lung cancer and it was assumed that this implied

causality between the two This conclusion was severely criticized by the tobacco

industry and their viewpoint was strongly backed by well-known statisticians such as Sir

Ronald Fisher Their argument was that, a model, which had absolutely no connection between lung cancer and smoking, could just as well explain this correlation In fact it could be the case that there was an unknown subspecies or breed that could exist and cause cancer and a craving for smoking simultaneously

1.3.2 GRANGER CAUSALITY

The philosophical concept of causality suffered from problems of application in the mathematical sciences or statistics Thus, several attempts were made to introduce a

probabilistic definition of causality Such a definition would make causality a useful tool

easily adaptable to stochastic events or processes It was important to find a concrete and testable definition of causality, which could be based just on statistical data Such an operational definition of causality was introduced by Granger in 1969 His definition was based on three principal axioms: the past and present may cause the future, but future cannot cause the past; if all the knowledge in the universe available at a given time contains no redundant information such that if some variable Z, is functionally related to one or more variables in a deterministic manner, then Z, should be excluded from that universe of knowledge; all causal relationships are unidirectional throughout time Based

on these assumptions, Granger defined causality as testing whether lagged information on

a variable y provides any statistically significant information about a variable x in the presence of lagged x If this is true then ~y Granger-causes x” Granger showed that the cross-spectrum between two variables can be decomposed into three parts - parts one and two relating to a single arm of a feedback situation and part three representing the

influence of instantaneous causality

Granger’s definition (1969) of causality is explained in terms of predictability Following Granger, let A, be a stationary stochastic process and let {A,, t = 0, td?

denote the information set which includes at least {(X, Yo} Let A = tAls<t}, A, =

{Alsst}, and let Ak) represent the set (A, j = k, k+1, } We similarly define X,

Y X,, V Next we let P(AIB) denote the opuimum, unbiased, least-squares predictor

of A, using the set of values given be B, Also, we assume U, to be the set of information available in the universe since time t-1 and U,- Y, to be all the information apart trom the

specified series Y, Granger (1969) then gave the following four definitions:

Causality ; if o°(X|U) < o°(X| U-Y) then it implies that Y is causing X or Y, is causing X,

where o” is the variance of the series

Feedback < if o°(X| U)<o°(X| U-Y )

o(¥| U)<e(YIU-X)

that is, X, is causing Y, and also Y, is causing X,

Instantaneous Causality : if o"(X| U, ¥)<o°(X| U), that is, when we include the

present value of Y, in the “prediction”, then the current value of X, is better “predicted”

than when it is not

Causality Lag : if Y, causes X, then the causality lag ‘m’ which is an integer can be

defined as the least value of k such that o°(X[U-Y(k)) < o°(X[U-Y(k+1))

All the above definitions are based on the fact that A, is a stationary series In case of a nonstationary series, o(X| U) etc will depend on time t and the very existence of

causality may vary over time

The above definitions of causality can be illustrated easily - we choose a simple bi-variate model to do so here If X, and Y, are two non-stationary series with zero means, then a simple causal mode! was defined by Granger (1969) as

X, = > aX, + > bYu, + &, pst ze!

where ¢, and n, are two uncorrelated white-noise series Technically, the value of m can

be infinity but practically data streams being of finite lengths limits the value of m to be finite and shorter than the given time series Following Granger’s definition of causality, the above equations imply that Y, is causing X, provided b, is non-zero and, similarly, X,

is causing Y, if some c, is not zero If both occur then the definition implies that the

relationship involves a feedback Similarly, the definition of instantaneous causality can

be given by,

X+ bY = 2 aX, + DY, + & pet jal

given that the variables require the above representation [n the above situation

instantaneous causality is occurring and the knowledge of Y, improves the “prediction”

or goodness of fit of the first equation for X,

In 1983, Granger introduced the concept of error-correcting methodology to be applicable in situations where the variables are co-integrated The conditions under which a finite-order time series vector

X& = UX J=l, Nj is said to be co-integrated of orders d,b and represented as x, ~

Cl(d,b) is

(i) each x, ~ I(d), that is, each X,, Is Integrated of order d, and

(11) a, is a set of constants such that

“Granger sense” This method has been widely accepted and applied to empirical work dealing with Granger causality In the bi-variate case, a pair of stochastic series given by

x, and y, will be considered to be in equilibrium when

C(1,1), then there will be a possibility of equilibrium to occur and if x,, y, are both [(0)

and also C(0,1), then there will be a number of equilibria making this error-correction

methodology relevant in the presence of co-integratedness

* C.W.J Granger, ‘Co-integrated variables and error-correcting models’ 1983 Department of economics

discussion paper no 83-13a (University of California San Diego).

in 1995, Toda and Yamamoto introduced a method that could be used to estimate vector auto-regression models (VAR) in levels and test the general restrictions on the parameter matrices even though the processes may be integrated or co-integrated of an arbitrary order This procedure can be applied to test for Granger causality in the case

where the null hypothesis is formulated as zero restrictions on the coefficients of the lag

of a subset of the variables The usual Wald test statistic for Granger non-causality based

on levels estimation has a nonstandard asymptotic distribution and depends on nuisance parameters if the process is integrated of order 1 However, this model intentionally over-

fits VAR’s and thus suffers from a lack of efficiency.‘

Before conducting any of the above tests, the (non) stationarity of the series gdp, capital labor ratio, pupil-teacher ratio are ascertained by testing the null hypothesis of a unit root and for co-integration where the null hypothesis of a unit root cannot be

rejected The enroliment rate variables are not tested similarly (even though unit-root tests indicate non-stationarity over the period under consideration) because they are reported in percent form and, intuitively speaking, they are 1(0) and have an upper bound even if the upper bound may not have been reached within the testable sample giving the appearance of non-stationarity

There are certain difficulties associated with Granger causality as there invariably are with any other sophisticated statistical procedure Granger causality test suffers from arbitrariness in lag choice and level of significance Other problems can arise from

measurement errors (Newbold 1978) or from temporal aggregation (Granger and

Newboid 1986) For example, some difficulties are related to data inadequacies such as insufficient frequency of data, some arise as a consequence of missing variables leading toa misinterpretation of results where relationships between variables are being

considered; sometimes there is a discrepancy between the recording of a variable and the time when it occurred resulting in incorrect interpretation: sometimes it is difficult to decide which variable should be excluded from the information set; interpretation problems can also arise from the idea of a leading indicator What is important for making a correct interpretation is the existence of some convincing theory for the causal mechanism under consideration However, most of the drawbacks of Granger causality relate to its capability of being operational and not to its basic definition.’

1.3.3 BRIEF REVIEW OF DEVELOPMENTS IN CAUSALITY TESTING

There have been a number of developments in methods to test for causality since Granger's work (1969) Hsiao suggested a procedure that combined Granger’s causality test with Akaike’s (1969, 1970) Final Prediction Error (FPE) criterion to overcome the shortcoming associated with the Granger and Sims test This method was criticized by Kang( 1989), who argued that the FPE procedure is based on the assumption of a purely autoregressive (AR) process with no window for any intervening AR parameter or

moving average (MA) component His method was a combination of Box-Jenkins

* Toda and Yamamoto, “Statistical Inference in Vector Autoregression With Possibly Integrated Processes”

1995, Journal of Econometrics 66, p 225-250

* For detailed discussion see Granger( 1979)

ARIMA modeling However, if his model yielded a pure AR process, then it was

equivalent to the FPE criterion.®

Some of the other tests of causality that were introduced after the Granger's 1969

paper, were, Priestley (1971), Granger (1973), Haugh (1972,1976), Pierce and Haugh

(1977), Pierce (1979) and Stokes and Neuburger (1979) However, none of these has had

a wide application in empirical work Priestley (1971) introduced the model with two- sided transfer functions or lag distributions Though his work was not primarily

developed for the purpose of testing causality, it had a striking resemblance to the tests proposed by Haugh (1972) and Granger (1973) Haugh (1972), developed an approach, which could identify the degree and direction of association between two time series that were covariance stationary.’ Granger (1973) proposed a stepwise regression procedure in

a bi-variate autoregressive model Haugh (1976) and Pierce and Haugh (1977) suggested

a method that would pre-whiten the two series first to obtain the residuals and then detect causality by lead and lag cross-correlations in these residuals The method of transfer function analysis to test for causality was introduced by Pierce (1979) and Stokes and Neuburger (1979) The more recent developments in causality testing came from Granger

(1980, 1983), Granger and Weiss (1983), Davidson (1986), Stock (1987), Engle and Granger (1987), Johansen (1988), Granger (1988) and Toda and Yamamoto (1995)

among others

1.4 EDUCATION VARIABLES AND DATA SOURCES

Various studies on human capital have used different variables as proxies for education The main reason for this has been the lack of reliable and relevant data,

especially for developing countries Here, a combination of variables will be used to

represent education Given that reliable data is available only on a few select variables, they have been chosen to accomplish certain specific objectives Six variables have been chosen to serve as proxies for education and these can be broadly categorized as three quantitative variables, meaning that they characterize the number of people receiving education, and three qualitative variable, meaning that they explain the quality of

education being imparted Some of these variables have been used in some of the seminal Papers on this topic and utilizing them will enable a verification of the validity of their conclusions, in terms of a causal relationship

For variables to serve as relevant proxies for educational quantity, primary,

secondary, tertiary enroliment rates (Mankiw, Romer and Weil, 1992) have been

selected The variables chosen to serve as proxy for educational quality are per capita spending on primary education, and pupil to teacher ratio at primary and secondary level For variables to serve as gender-based measures of education, female and male primary enrollment rates and female and male secondary enrollment rates have been chosen Additionally, the total fertility rate has been included in the gender-based discussion, to

act as a distinction between the genders Thus, for the analysis based on gender, the

education variables will only describe the quantitative aspects of education

* Bahmani-Oskooee, Mohtadi and Shabsigh, Exports growth and Causality in LDCs - A re-examination”,

Development Indicators (WDI), 1998, will be primarily used Additionally, in some cases where data is not available in the WDI database, the Nehru and Dhareshwar data set,

1993, will be used These will provide the necessary education statistics segregated by level and gender from 1966 to 1996 for most countries A total of twelve countries, comprising Asia, have been selected for the analysis These countries include poor or

low-income countries such as Bangladesh, Nepal, Pakistan, India, and Sri Lanka, low-to-

middle income countries such as Indonesia, Philippines, and Thailand, middle-to-high income countries such as Malaysia and high-income countries such as South Korea, Singapore, and Japan For simplicity, these countries have been categorized by

geographic location within Asia Thus, Bangladesh, India, Nepal, Pakistan, and Sri Lanka fall under the classification of South Asia, Indonesia, Philippines, Thailand, Malaysia, and Thailand form South-East Asia while Japan and South Korea are included in East Asia

The rest of the dissertation is organized as follows: Chapter 2 conducts detailed Granger causality and cross-correlation tests between education variables and economic growth The education variables, in this chapter, are categorized by level, and by

qualitative or quantitative aspect within level The analysis is based on growth rates in this chapter In Chapter 3, the focus turns to gender and the education data is segregated accordingly Additionally, a new variable is introduced to allow for a distinction between genders Chapter 4, in philosophy, continues the analysis started in the previous chapter

1991, Journal of Development Economics 36 p.405-415,

but introduces some changes, both in methodology and data Chapter 5 is the concluding chapter, which brings together the results from all the other chapters and ends with a brief discussion on policy implications

{1.5 PRECEDENCE OF EDUCATION LEVEL FOR GROWTH IN DEVELOPING

COUNTRIES

Chapter 2 will test for correlations and causal relationship between the qualitative and quantitative measures of education at different levels and economic growth As mentioned earlier, the main objective behind conducting these tests is to look for a possible link between educational levels and stage of economic development

SIMPLE-CORRELATIONS

In order to check for the existence of correlation between the education and growth variables, the following null hypothesis is tested:

Hy: correlation (Ay,, AX,) = 0

Against the alternative,

H, : correlation (AX,, Ay,) > 0

Here 1, ts the log of Y, (per capita GDP), and the first difference of log is estimated in

order to test for growth rates To test H, against the alternative H,, we check the

correlations between the series given by E[(Ay, — E(Ay, 4X, - E(X))} ,

VVari Ay,) Var(X,)

"David Pierce and Larry Haugh, “Causality in Temporal Systems”, 1977, Journal of Econometrics 5, p 280

These tested the correlation between one particular gender-based education variable and growth while keeping the impact of the other gender-based education variable constant This allowed for the isolation of impact of one specific gender on growth In order to calculate the partial correlation, the gender variable of choice and the growth variable are individually regressed on the gender variable whose impact is meant to be held constant Thereafter simple correlations are conducted between the residuals from the two simple regressions conducted earlier

GRANGER CAUSALITY

In testing for Granger causality the following null hypothesis is tested,

Họ, E(Ayd Ay, A¥e2, Ayes Ayee Xe Xá Xoi Xa, AZ4., A24: AZ43 ÁZ4n)E E(Ay,

Ay.i, AY.2, Ay3 AYe, AZ.t, AZ:, AZ: ÁZ cm)

for all m > 0 — No Granger causality

Against the alternative of,

HE(Ay; Ayer, A¥e2, A¥is Avee Xui, Xói Xói Xe, AZ, AZ AZ 3 AZ,

mÈ£E(AylAy, Ayc:, Ayi3 A¥ia, AZ, AZ2, AZ.3 AZren)

For some m > 0 — Granger causality

Here the Ay represents first difference of the log of per capita GDP, AZ represents first difference of the log of the capital labor ratio, and X represents the education variables Using the above information, the model constitutes the following six 3x3 vectors:

Before conducting any of the above tests, the (non) stationarity of all relevant Senes is tested, and in cases where any number of series is non-stationary, co-integration between the series is tested using Johensen’s co-integration test For equations where there is no evidence of co-integration between the variables, the following equation is

estimated:

Ay, = 8) - > Ôn Ay, * > db: ÂZ„ + > dX, * €ụ (a.)

For the lagged variables appearing on the right-hand-side, the number of lags is

determined using the Akaike Information Criterion (AIC) and Schwartz Criterion (SC) and the lag that gives the lowest AIC and SC and best fit is chosen The information

criterion has been widely used in analysis involving time series in order to determine the appropriate length of the distributed lag The Akaike Information Criterion (AIC) is

Similar testing is conducted to assess the forecasting power of output growth on

education Here the equation used is very similar to (a 1) above It is given by,

X, = a> > a1, Xi + >: G2, AZ, + > a; Ayi,+ ey (b.1)

jel pl pl

If 2, or a3, are found to be statistically significant and different from zero, it is

concluded that the change in capital labor ratio and/or growth has a causal impact on education Where co-integration is present between non-stationary variables, equation (a.1) above is replaced with a version of the vector error-correction model introduced by Granger (1983) In this case, the following equation is used to test for causality:

Ay, = ay + > a;, Ay, + » 8n ÂZm + 2, aX, * SEC + 6, (42), rel ;sỉ pel

where the error-correction term is given by EC,=3,- % Xea- 7 Zo after normalizing the co-integrating vector For equations that measure the impact of growth on education but the series are found to be co-integrated, we replace (b 1) with

The results, from the above tests, are provided for individual! countries and for

regions as a whole, based on panel estimation There is another variable of interest,

namely capital-labor ratio Here the emphasis is on the relationship between education and output and no further investigation is carried out on the relationship between

physical capital and human capital (education) or between physical capital and growth and is left as an avenue for possible further research

1.6 THE ROLE OF GENDER IN EXPLAINING THE RELATION BETWEEN

EDUCATION AND GROWTH

This chapter analyzes the educational impact of each individual gender on growth

in a time-series framework The primary objective is see whether the relation between education level and growth, which was found to be closely linked to the level of

development of a country or region, can be explained further in terms of the

contributions of males and females at different education levels Specifically, the primary objective of this analysis is to conduct a country-by-country and regional time series