92

CHAPTER - V

METHODOLOGY OF THE STUDY

This chapter explains the methodology adopted for the study. The

size of population, methods used for collection of data, and

statistical methods used for testing of hypotheses has been

discussed exhaustively.

As it is clear in chart (5-1), this chapter has divided into two parts

which are Research Design and Data Analysis. Research design

explains the selection of the sample, the kinds of the data which

have been collected and finally the different utilized variables for

analyzing of the data.

In second part, data analysis is also segmented into three

sections which are Trend Analysis of the Capital Structure and EVA

for the period of the study, analysis of the effect of Capital

Structure on Economic Value Added and last one, Analysis of EVA

ability to serve as an effective proxy for Market Value Added.

In trend analysis, the statistical methods which have been used

to test the hypotheses are described. Second and third sections of

data analysis include furthermore some explanation about the

variables, dependent as well as independent variables which have

been used for testing the hypotheses.

93

Chart 5-1: A summary of methodology details which has followed in the research.

METHODOLOGY

Research

Design

Data

Analysis

Sample Selection

Data Collection Variables Definition Trend Analysis

Analysis of the

effect of Capital

Structure on EVA

Analysis of EVA

ability to present

firm’s value

Automobile

Industry

information

Financial Reports

of Auto Firms

Necessary data

for calculation of

EVA & MVA

Trend of Capital

Structure

Trend of EVA

Dependent

Variables

Independent

Variables

94

5-1) POPULATION

Indian Automobile Industry has been considered as the

population under study. There is several reasons that why this

industry was chosen as population of ongoing research, which has

been explained as follows:

The automobile industry is one of the core industries in the Indian

economy. With the liberalization of the economy, India has

become the playground of major global automobile majors. The

Automotive Industry in India is now working in terms of the

dynamics of an open market. Many joint ventures have been set

up in India with foreign collaboration, both technical and financial

with leading global manufacturers. The Government of India is

keen to provide a suitable economic and business environment

conducive to the success of the established and prospective foreign

partnership ventures.

Far reaching economic reforms aimed at deregulation and

attracting foreign investment have moved India firmly into the

front ranks of rapidly growing Asia-Pacific Region in the automobile

industry. In the dynamics of transition of the Indian economy, the

automobile industry is emerging as a leading industry.

As per the statistical reports of Centre for Monitoring Indian

Economy (CMIE), Automobile Industry has the highest growth rate

of Capital Employed, Net Worth, and Gross Fixed Assets among

manufacturing industries in India for the period of 1996-2004. This

is the rational reason to select Automobile Industry in order to

examine implications of EVA in Indian Industries. These ratios

have been presented in Figures 5-1 to 5-3.

95

Figure 5-1): Average Growth Rate of Capital Employed:

1996-2004 (%) by Industry

13.97

11.3610.71

9.67 9.39

7.53 7.506.88 6.70

5.86

3.03

1.23

0

2

4

6

8

10

12

14

16

Autom

obile

Beverage &

tobacoo

Mashinery

Auto A

ncillaries

Chem

icals

Non-M

etalic Mineral

Food

Paper

Non-Ferrous

Diversified

Ferrous Metals

Textiles

Figure 5-2): Average Growth Rate of Net Worth:

1996-2004(%) by Industry

16.2

13.212.2

9.79.1

7.16.6

5.5 5.4

0.4

17.1

(19.2)

(20)

(15)

(10)

(5)

0

5

10

15

20

Au

tom

obile

Beverage &

tob

acoo

Mash

inery

Au

to A

ncillaries

Ch

emica

lsP

aper

No

n-F

errou

sF

ood

No

n-M

etalic M

ineral

Diversified

Ferro

us M

etalsT

extiles

96

Figure 5-3): Average Growth Rate of Gross Fixed Assets:

1996-2004 (%) by Industry

15.6 15.414.9

13.5 13.412.5 12.5

12.0

10.910.2

9.89.4

0

3

6

9

12

15

18

Autom

obile

Chemicals

Auto A

ncillaries

Beverage &

tobacoo

Non-M

etalic Mineral

Food

Mashinery

Paper

Diversified

Textiles

Ferrous M

etals

Non-F

errous

Source: Corporate Sector, Centre for Monitoring Indian Economy (CMIE)

5-2) Sample of the Study

The study covers Automobile Industry Firms which are listed in

Bombay Stock Exchange. The firms in the population were

selected, based on the following criteria:

1) Automobile Industry firms which have been listed on Bombay

Stock Exchange (BSE) in or before 2001;

2) They must be existing in BSE till the financial year 2005;

3) They should not have negative values for average operating

income during the period of the study.

In short this means that we have a tendency towards relatively

profitable firms. Further (1) Multinational companies that their

parents companies are in other countries; or (2) Assembling

companies that are not manufacturer; or (3) The companies that

have come under the Sick Industrial Companies and they have

97

been referred to the Board of Industrial and Financial

Reconstruction (BIFR) were excluded for the sake of comparability

and consistency.

In order to recognize population under study in all respects,

Bombay Stock Exchange Official Directory, and Bombay Stock

Exchange Corporate compendium reports have been observed. So

according to mentioned survey, the above population that consists

of 17 companies in Automobile Industry has been used for the

study.

5-3) DATA COLLECTION

The study has been based on secondary data and as it is clear in

chart (5-1), three kinds of data and information have been

collected which are as follows:

I) Historical Information of Automobile Industry

II) Financial Reports of Automobile Industry Firms

III) Some economic data for the calculation of EVA and MVA.

Historical data of Automobile Industry as overall and selected

companies particularly have been collected from “Research,

Statistics & Publication Department” and also “Library” of Bombay

Stock Exchange (BSE).

Published annual financial reports of the companies including:

1. balance sheets,

2. profit & loss accounts,

3. schedules related to balance sheet & profit and loss account,

4. note to accounts and Accounting policies;

have been taken into consideration for second part.

Annual reports of a few companies have been requested by

sending email to their email addresses and collected from

companies’ websites or other websites such as

98

“Equitymaster.com”, “Indianinfoline.com”, “Valuenotes.com”,

“BSE.com”, “Myiris.com” and ”Searchindia.com”.

In order to find out whatever financial reports still remained to be

collected, the researcher has visited Pune Stock Exchange and also

observed Prowess database of Centre for Monitoring Indian

Economy (CMIE).

For last part of data collection some libraries of different

Institutes and Colleges such as National Institute of Banking and

Management (NIBM), Symbiosis Institute of Business Management

(SIBM), Symbiosis Center for Management & HRD (SCMHRD),

Symbiosis Institute of International Business (SIIB), Gokhale

Institute of Politics and Economics, Vaikunth Mehta National

Institute of Co-operative Management (VAMNICOM), Institute of

Management Development and Research (IMDR) and Maharatta

Chamber of Commerce Industries & Agriculture have been visited

and the researcher could find some port of data there. Of course it

should be mentioned in this part of data collection, CMIE products

and Prowess database which have been explained before were

much more useful.

However all data needed for the study was collected for the

companies after hard trying for at least one year.

5-4) DATA ANALYSIS

In order to meet the objectives of the study, data analysis has

divided in three parts: (1) Trend analysis of Capital Structure and

Economic Value Added, (2) Analysis of the effect of capital

structure on EVA; and (3) Analysis of whether EVA is the best

performance measure to present firm value and is more relative to

MVA in comparison with the other traditional measures.

The analyses have been done by Panel Data Analysis Model

(which is explained later) and using S-Plus Software. In this

99

research two methods of estimation have been used for panel data

analysis; (1) Generalized Estimating Equations (GEE) and (2)

Feasible Generalized Least Squares (FGLS). They consider within

individual correlations in response.

5-4-1) Trend Analysis:

The objective of this section is to view the capital structure

pattern followed by the Indian Automobile companies and also to

recognize the created amount of economic value added by those

companies in the period of 2001 to 2005.

The analysis is carried out in terms of financial indicators which

are four in number:

(i) Total Debt Ratio

(ii) Debt-Equity Ratio

(iii) Economic Value Added

(iv) EVA to Capital Employed Ratio

The ratios (i) and (ii) indicate the direction of changes in capital

structure practices and the last two indicators present a picture of

the corporate performance to create wealth. All these ratios are

calculated on a year to year basis for the companies.

To study the capital structure and EVA trends, the researcher has

computed the descriptive statistical values such as mean, median,

variance, standard deviation, minimum, maximum and range of

each ratio, for each firm and also for each year, by using SPSS

package. Various charts show the above descriptions of capital

structure and economic value added ratios. Moreover, in order to

review of Capital Structure and EVA changes over the time,

generalized estimating equations regression model has been fitted.

100

In this case Generalized Estimating Equations (GEE)1

population-averaged model was done by using statistical software

S-plus.

Measures of Capital Structure and Firm Value:

As mentioned before, four financial measures have been used in

this section. Two of them present an insight into the financial

pattern of the Automobile Industry firms, and next two indicate

ability of firms to enhance the firm’s value and create capital

employees wealth. They are computed as bellow:

I. Total Debt Ratio: AssetsTotal

DebtTotal

The total debt ratio measures the extent to which borrowed funds

support the firm’s assets. The numerator of this ratio includes all

liabilities, short-term as well as long-term, and the denominate of

this ratio is total assets.

II. Debt-Equity Ratio: Equity

Debt

The debt-equity ratio shows the relative contributions of creditors

and owners. The numerator of this ratio includes all loans,

short-term as well as long-term, secured as well as unsecured

loans and the denominate of this ratio consists of net worth which

includes share holder equity, reserve and surplus.

III. Economic Value Added:

NOPAT–(Cost of Capital x Capital Employed)

EVA is the financial performance measure that captures the true

economic profit of a company. It is defined as the excess of a

company’s after tax net operating profit over the required

minimum rate of return that investors and lenders could get by

1 Kung-Yee Liang, Scott L. Zeger, “Longitudinal Data Analysis Using Generalized

Linear Models”, Biometrika, Vol. 73, No. 1, Apr 1986, pp.13-22.

101

investing in other securities of comparable risk. Economic Value

Added gives managers superior information and superior

motivation to make decisions that will create the greatest

shareholder wealth in any publicly owned or private enterprise.

IV. EVA to Capital Employed Ratio: EmployedCapital

EVA

EVA is an absolute value. There is no minimum value or specified

value. When it is expressed as a ratio to capital employed, one will

be able to compare the performance of a company over the years

and also to make comparisons between companies in an industry.

So the researcher has used the ratio of EVA to capital employed

instead of EVA.

The EVA to Capital Employed Ratio describes the relative wealth

creation by total capital employed. The numerator of this ratio

includes the amount of economic value added (EVA) and the

denominate of that is capital employed which consists of equity as

well as loans.

5-4-2) Analysis of the effect of capital structure on EVA

In this section the researcher tries to find out whether any

relationship exists between the capital structure - as a sources of

finance- and economic value added -as a tool to recognize how

those sources has been utilized in the company- on the other

hand. For this purpose all variable measures have been calculated

for each firm and each year of the period under study.

In order to find out whether any relationship exists between

dependent and independent variables, simple linear regression for

panel data analysis that takes into account the correlation between

those variables was used. The estimation of the parameters ()

was based on a technique called cross-sectional time-series

102

feasible generalized least squares (FGLS) regression. 2 Statistical

analysis was done by using statistical software S-plus.

Measures for the Variables

For this part of the study two independent variables have been

chosen to analysis the impact on two dependent variables. The

measures for the variables are given in detail as follows:

a. Dependent Variables:

The measure of created wealth in the firms has been considered

as dependent variable. It is presented by EVA and also the ratio of

EVA to Capital Employed, which have been explained in trend

analysis section in details:

I) Economic Value Added:

NOPAT–(Cost of Capital x Capital Employed)

II) EVA to Capital Employed Ratio: EmployedCapital

EVA

b. Independent Variables:

Measures of independent variables also have been explained

before in this chapter in details. Here they are presented in

summary.

Total Debt Ratio (TDRATIO) = AssetsTotal

DebtTotal

Debt-Equity Ratio (DERATIO) = Equity

Debt

2 Jerry Hausman, Guido Kuersteiner, “Difference in Difference Meets Generalized

Least Squares: Higher Order Properties of Hypotheses Tests”, MIT,

Department of Economics, Cambridge, June2004.

103

5-4-3) Analysis of EVA versus traditional performance

measures:

In third part of the analysis, the relationship of different

performance measures with market value added is studied. The

researcher tries to find out whether EVA is more associated with

stock returns.

The data has been classified and tabulated in such forms and

ways, which enabled to generate substantial evidence to test the

hypotheses. All variable measures have been calculated for each

firm and each year of the period under study. SPSS package and

Microsoft Excel Software have been utilized for calculating the

descriptive statistical values such as mean, median, variance,

standard deviation, minimum, maximum and range of each ratio,

for each year. Various charts show the above statistical values.

Moreover, a generalized least square model for panel data using

correlation was fitted to find out whether any relationship exists

between dependent and independent variables. The estimation of

the parameters was based on a technique called

cross-sectional time-series feasible generalized least squares

(FGLS) regression. Statistical analysis was done by using statistical

software S-plus.

Measures for the Variables

For this part of the study five independent variables have been

chosen to analysis the impact on one dependent variable. The

measures for the variables are given in detail as follows:

I) Dependent Variable:

Market Value Added which is the indicator of firm value is

considered as dependent variable:

Market Value Added = ValueBookValueMarket

104

The MVA explains the value added to a particular equity share over

its book value. It informs how much has been added in the

economic value of shareholders. So Market Value Added can be

estimated by subtracting the book value of shares from the market

value of shares.

II) Independent Variables:

Five performance Measures are considered as independent

variables. They are as follows.

1. Return on Assets (ROA) = AssetsTotal

EBIT

Return on Assets measures a company’s earnings in relation to all

of the resources it has employed. ROA tells us what earnings were

generated from invested capital (assets).

The assets of the company are comprised of both debt and

equity. Both of these types of financing are used to fund the

operations of the company. The ROA figure gives investors an idea

as to how effectively the company is converting the money it has

into net income.

2. Return on Capital Employed (ROCE)

= DebttermLongWorthNet

EBIT

Return on Capital Employed (ROCE) is a measure of the returns

that a company is realizing from its capital. It calculates as profit

before interest and tax divided by the difference between total

assets and current liabilities. The resulting ratio represents the

efficiency with which capital is being utilized to generate revenue.

3. Earning Per Share (EPS) = SharesEquityofNumberThe

PAT

Earning Per Share is the portion of a company's profit allocated to

each outstanding share of common stock.

105

Earning Per Share as the name indicates, is the “per share

earning” of a company in a reported period. This is the most

important factor in fundamental analysis of a stock. This coupled

with a few of the related ratios should give you a fair idea about

the worth of a stock.

4. Economic Value Added (EVA)=

NOPAT–(Cost of Capital x Capital Employed)

EVA attempts to measure how much 'value' was created by an

organization during an accounting period for its shareholders. it is

defined as the excess of a company’s after tax net operating profit

over the required minimum rate of return that investors and

lenders could get by investing in other securities of comparable

risk. So when a company's net operating profit after tax exceeds

its capital employed charge, its EVA is positive and value has been

created. If the result is negative, the firm's management could not

meet the expected returns of investors.

5. Net Operating Profit After Tax (NOPAT):

Net Operating Profit after Tax is an alternative indicator for

measuring operating efficiency for leveraged companies. It is an

estimate of what a company would earn if it didn't have any debt,

equal to operating income times (1 minus the tax rate). NOPAT is

frequently used for calculating Economic Value Added (EVA).

5-4-4) Notes

Variable names used in correlation and regression have been

given in parentheses.

EBIT indicates earning before interest and tax and PAT means

profit after tax.

106

5-5) A BRIEF INTRODUCTION TO PANEL DATA MODEL

ANALYSIS:

5-5-1) Panel Data

Panel data, also called longitudinal data or cross-sectional time

series data, are data where multiple cases (people, firms,

countries etc) were observed at two or more time periods.

There are two kinds of information in cross-sectional time-series

data: the cross-sectional information reflected in the differences

between subjects, and the time-series or within-subject

information reflected in the changes within subjects over time.

Panel data regression techniques allow you to take advantage of

these different types of information.3

The descriptor panel data comes from surveys of individuals. In

this context, a “panel” is a group of individuals surveyed

repeatedly over time. Historically, panel data methodology within

economic had been largely developed through labor economics

applications. Now, economic applications of panel data methods

are not confined to survey or labor economic problems and the

interpretation of the descriptor “panel analysis” is much broader.

Hence, we will use the terms “longitudinal data” and “panel data”

interchangeably although, for simplicity, we often use only the

former term.

Longitudinal and panel databases and models have taken on

important roles in the literature. They are widely used in the social

science literature, where panel data are also known as pooled

cross-sectional time series, and in the natural science, where panel

data are referred to as longitudinal data.4

3 James H. Stock, Mark W. Watson, “Introduction to Econometrics”, 2003,

Chapter 8, Regression with Panel Data.

4 Edward W. Frees, “Longitudinal and Panel Data: Analysis and Applications in

the Social Sciences”, CAMBRIDGE UNIVERSITY PRESS, 2004.

107

There are several advantages of longitudinal data compared with

either purely cross-sectional or purely time-series data. Here we

focus on two important advantages: the ability to study dynamic

relationships and to model the differences, or heterogeneity,

among subjects. Ofcourse, longitudinal data are more complex

than purely cross-sectional or times-series data and so there is a

price to pay in working with them. The most important drawback is

the difficulty in designing the sampling scheme to reduce the

problem of subjects leaving the study prior to its completion,

known as attrition.5

5-5-2) Panel Data Analysis

Panel data analysis is a method of studying a particular subject

within multiple sites, periodically observed over a defined time

frame. Within the social sciences, panel analysis has enabled

researchers to undertake longitudinal analyses in a wide variety of

fields. In economics, panel data analysis is used to study the

behavior of firms and wages of people over time. In political

science, it is used to study political behavior of parties and

organizations over time. It is used in psychology, sociology, and

health research to study characteristics of groups of people

followed over time. In educational research, researchers study

classes of students or graduates over time.

With repeated observations of enough cross-sections, panel

analysis permits the researcher to study the dynamics of change

with short time series. The combination of time series with

cross-sections can enhance the quality and quantity of data in

ways that would be impossible using only one of these two

dimensions.6 Panel analysis can provide a rich and powerful study

5 Ibid, 2004.

6 D. Gujarati, “Basic Econometrics”, 4th ed., 2003, New York: McGraw Hill, pp.

638-640.

108

of a set of people, if one is willing to consider both the space and

time dimension of the data.7

Panel data analysis endows regression analysis with both a spatial

and temporal dimension. The spatial dimension pertains to a set of

cross-sectional units of observation. These could be countries,

states, counties, firms, commodities, groups of people, or even

individuals. The temporal dimension pertains to periodic

observations of a set of variables characterizing these

cross-sectional units over a particular time span.

In other word longitudinal data analysis represents a marriage of

regression and time-series analysis. As with many regression data

sets, longitudinal data are composed of a cross section of subjects.

Unlike regression data, with longitudinal data we observe subject

over time. Unlike time-series data, with longitudinal data we

observe many subjects. Observing a broad cross section of

subjects over time allow us to study dynamic, as well as

cross-sectional, aspects of a problem.

Panel data require special statistical methods of analysis because

the responses at different time points on the same individual may

not be independent even after conditioning on the covariates. For

a linear regression model this means that the residuals for the

same individual are correlated. in other word Data measured

repeatedly on the same units over time violate the usual

regression model assumption of independent observations,

successive measures on the same unit are correlated and require

special statistical methods to account for the correlation. There is a

large body of methods that can be used to analyze panel data,

ranging from the simple to the complex, some useful references

7 Robert A. Yaffee, “A Primer for Panel Data Analysis”, Sep 2003, Updated April

2005.

109

are Diggle et al. (2002)8, Everitt (1995)9, and Hand and Crowder

(1996).10

5-5-3) Types of Panel Analytic Models

There are several types of panel data analytic models. There are

constant coefficients models, fixed effects models, and random

effects models. Among these types of models are dynamic panel,

robust, and covariance structure models. Solutions to problems of

heteroskedasticity and autocorrelation are of interest here.

5-5-4) Model Estimation 11

Models have to be estimated by methods that handle the

problems afflicting them. A constant coefficients model with

residual homogeneity and normality can be estimated with

ordinary least squares estimation (OLS). As long as there is no

group wise or other heteroskedastic effects on the dependent

variable, OLS may be used for fixed effects model estimation as

well (Sayrs, 1989). For OLS to be properly applied, the errors have

to be independent and homoskedastic. Those conditions are so

rare that is often unrealistic to expect that OLS will suffice for such

models (Davidson and MacKinnon, 1993).

There are several approaches to extend generalized linear models

to clustered data. Mixed effect models and transition models

(Diggle, Liang, and Zeger 1994, Chapter 7, 9-10) fully specify the

joint distribution within clusters via latent variables or conditional

dynamics.

8 P. J. Diggle, P. Heagerty, K. Y. Liang & S. L. Zeger, “Analysis of longitudinal

data”, 2nd Edition, 2002, Oxford, UK: Oxford University Press.

9 B. S. Everitt, “the analysis of repeated measure: A practical review with

example”, The Statistician, 44, pp. 113-135.

10 D. J. Hand, M. Crowder, “practical longitudinal data analysis” London:

chapman@hall.

11 Robert A. Yaffee, “A Primer for Panel Data Analysis”, Sep 2003, Updated April

2005.

110

With the presence of random effects, likelihood estimation

necessitates the integration over the random effects distributions,

which may be numerically intractable. Lee and Nelder (1996, 2 The

R Package geepack for Generalized Estimating Equations 2001)

introduced hierarchical generalized linear models and showed that

the integration may be avoided by working on the h-likelihood.

Compared to these approaches, the method of GEE fits marginal

mean models with the advantage that only correct specification of

marginal means is needed for the parameter estimator to be

consistent and asymptotically normal. This approach has become

an important tool in analyzing longitudinal data or repeated

measures arising in a wide variety of applications. For a discussion

on the relation between marginal and mixed effects models, see

Heagerty and Zeger (2000) and Nelder and Lee (2004).

Heteroskedastic models are usually fitted with estimated or

feasible generalized least squares (EGLS or FGLS).

Heteroskedasticity can be assessed with a White or a

Breusch-Pagan test. For the most part, fixed effects models with

groupwise heteroskedasticity cannot be efficiently estimated with

OLS. If the sample size is large enough and autocorrelation

plagues the errors, FGLS can be used. Random sampling and

maximum likelihood iterated by generalized least squares have

also been used (Greene, 2002). Beck and Katz (1995) reportedly

found that if the sample size is finite or small, the total number of

temporal observations must be as large as the number of panels;

moreover they reportedly found that OLS with panel corrected

errors provided more efficient estimation than FGLS (Greenberg,

2003; STATA, 2003).

If the model exhibits autocorrelation and/or moving average

errors, first differences (Wooldridge, 2002) or GLS corrected for

ARMA errors can be used (Sayrs, 1989). Hausman and Taylor

111

(1981) have used weighted instrumental variables, based only on

the information within the model, for random effects estimation to

be used when there are enough instruments for the modeling. The

instrumental variables, which are proxy variables uncorrelated with

the errors, are based on the group means. The use of these

instrumental variables allows researchers to circumvent the

inconsistency and inefficiency problems following from correlation

of the individual variables with the errors.

For dynamic panels with lagged dependent variables, Arellano,

Bond, and Bover have used general methods of moments, which

are asymptotically normal (Wooldridge, 2002). With greater

numbers of moment conditions, they are able to handle some

missing data and they can attain gains in efficiency as long as

there are three or four periods of data (Greene, 2002).

Another estimation procedure was developed by Arnold Zellner,

called seemingly unrelated regression (SUR) requires that the

number of explanatory variables in each cross-section is the same.

In the SUR approach, variables are transformed with a form of

Cochrane-Orchutt correction to model the autocorrelation. Feasible

generalized least squares are used to estimate a covariance

matrix. The parameter estimates are also modeled. The process is

iterated until the errors are minimized.

LIMDEP 8 (Greene, 2002) has its own protocol for estimating

random parameter models, including the limited dependent

variable models. The limited dependent variable models are

population averaged models. In LIMDEP, the estimation for such

models begins with an OLS estimation of starting values and then

proceeds to simulation with Halton draws. This procedure, Greene

maintains, is generally faster than the quadrature estimation used

by STATA. When the panels are large in number and size, it may

be the only timely method for estimation.

112

If there are enough temporal observations, they can use either

the lagged levels or lagged differences as instruments, while the

other variables serve as their own instruments in an extension. If

group sizes are larger than 20 and the autocorrelation is higher

than 0.4, the random effects quadrature algorithms can bog down

or even fail to converge (STATA, 2003).

Robust estimation, when one has heteroskedasticity,

autocorrelation, or outliers to contend with, may be performed

with the general methods of moments (GMM) and combination of

White and Newey-West estimators to obtain robust panel standard

errors. Arellano, Bond, and Bover have used GMM in their models

and these are incorporated into LIMDEP version 8 and Stata

version 8 special edition. GMM models tend to be robust with

respect to heteroskedasticity and nonnormality.

In this research two methods of estimation have been used for

panel data analysis; (1) Generalized Estimating Equations (GEE) 12

and (2) Feasible Generalized Least Squares (FGLS)13. They

consider within individual correlations in response.

The generalized estimating equation (GEE) approach of Zeger and

Liang14 facilitates analysis of data collected in longitudinal, nested,

or repeated measures designs. GEEs use the generalized linear

model to estimate more efficient and unbiased regression

parameters relative to ordinary least squares regression in part

because they permit specification of a working correlation matrix

that accounts for the form of within-subject correlation of

12GARY A. BALLINGER, “Using Generalized Estimating Equations for Longitudinal

Data Analysis”, Purdue University, Organizational Research Methods, Vol. 7,

No. 2, 2004, pp.127-150.

13Jerry Hausman and Guido Kuersteiner, “Difference in Difference Meets

Generalized Least, Squares: Higher Order Properties of Hypotheses Tests”,

June 2004.

14Liang, K.-Y.,&Zeger, S. L., “Longitudinal data analysis using generalized linear

models”, Biometrika, 73, 1986, pp.13-22.

113

responses on dependent variables of many different distributions,

including normal, binomial, and Poisson.

Generalized Estimating Equations are a general method for

analyzing data collected in clusters where 1) observations within a

cluster may be correlated, 2) observations in separate clusters are

independent, 3) a monotone transformation of the expectation is

linearly related to the explanatory variables and 4) the variance is

a function of the expectation. It is essential to note that the

expectation and the variance referred to in points (3) and (4) are

conditional given cluster-level or individual-level covariates.

In addition to a fixed or random effects specification, panel data

may suffer from serially correlated errors if the time span is

sufficiently long and from heteroskedastic errors if the cross-

section units have different scales. It was likely that the data set

for this analysis includes both problems.To account simultaneously

for serial correlation and heteroskedasticity, a feasible generalized

least squares (FGLS) approach provides an alternative estimator.

That is, the generalized least squares estimation procedure

(GLS)15, but with an estimated covariance matrix, not an assumed

one.

Because the statistical method which the researcher has used is a

new method so it is necessary to introduce that.

Regression analysis and time-series analysis are two important

applied statistical methods used to analyze data. Regression

analysis is a special type of multivariate analysis in which several

measurements are taken from each subject. We identify one

measurement as a response, or dependent variable; our interest is

in making statements about this measurement, controlling for the

other variables. 15 Franco Peracchi, “Methods for Panel Data”, University of Rome, Tor Vergata,

Spring 2004.

114

With regression analysis, it is customary to analyze data from a

cross section of subjects. In contrast, with time-series analysis, we

identify one or more subjects and observe them over time. This

allows us to study relationships over time, the dynamic aspect of a

problem. To employ time-series methods, we generally restrict

ourselves to a limited number of subjects that have many

observations over time.

Longitudinal data analysis represents a marriage of regression

and time-series analysis. As with many regression data sets,

longitudinal data are composed of a cross section of subjects.

Unlike regression data, with longitudinal data we observe subject

over time. Unlike time-series data, with longitudinal data we

observe many subjects. Observing a broad cross section of

subjects over time allow us to study dynamic, as well as

cross-sectional, aspects of a problem.

Longitudinal and panel databases and models have taken on

important roles in the literature. They are widely used in the social

science literature, where panel data are also known as pooled

cross-sectional time series, and in the natural science, where panel

data are referred to as longitudinal data.16

16 Edward W. Frees, “Longitudinal and Panel Data: Analysis and Applications in

the Social Sciences”, CAMBRIDGE UNIVERSITY PRESS, 2004.