British Journal of Management, Vol.

21, 607626 (2010)

DOI: 10.1111/j.1467-8551.2010.00708.x

The Impact of Board Independence and

CEO Duality on Firm Performance: A
Quantile Regression Analysis for
Indonesia, Malaysia, South Korea and
Thailand
Dendi Ramdani and Arjen van Witteloostuijn
University of Antwerp, Faculty of Applied Economics, Antwerp Centre of Evolutionary Demography (ACED),
Prinsstraat 13, 2000 Antwerp, Belgium
Email: dendi.ramdani@ua.ac.be; arjen.vanwitteloostuijn@ua.ac.be
We study the eect of board independence and CEO duality on rm performance for a
sample of stock-listed enterprises from Indonesia, Malaysia, South Korea and
Thailand, applying quantile regression. Quantile regression is more powerful than
classical linear regression since quantile regression can produce estimates for all
conditional quantiles of the distribution of a response variable, whereas classical linear
regression only estimates the conditional mean eects of a response variable. Moreover,
quantile regression is better able to handle violations of the basic assumptions in
classical linear regression. Our empirical evidence shows that the eect of board
independence and CEO duality on rm performance is dierent across the conditional
quantiles of the distribution of rm performance, something classical linear regression
would leave unidentied. This nding suggests that estimating the quantile eect of a
response variable can well be more insightful than estimating only the mean eect of the
response variable. Additionally, we nd a negative moderating eect of board size on
the positive relationship between CEO duality and rm performance.

Introduction
In the organization sciences literature, many
issues remain unsolved empirically because
dierent studies report dierent results. In many
topical domains, some studies report signicantly negative relationships, others signicantly
positive ones, and yet others insignicant ones.
We gratefully thank Sang-Woo Nam for providing the
data collected when he worked as a Research Fellow at
the Asian Development Bank Institute, Tokyo, Japan, in
20002005. Arjen van Witteloostuijn gratefully acknowledges the nancial support through the Odysseus
Programme of the Flemish Science Foundation (FWO).

Yet others complicate matters by adding nonlinearities, by nding U-shaped or reversed Ushaped linkages with or without the reection
point being located within the observed range, or
by including all kinds of intermediating or
moderating eects. A well-known example of a
literature revealing such a state of the art is the
one on the corporate diversicationperformance link in strategic management. Of course,
many reasons may explain such a mixed bag of
ndings, obvious candidates being dierent
samples and dierent specications across studies. However, we believe that often such
empirical inconclusiveness may, in part, be
explained by the dominant estimation method

r 2010 British Academy of Management. Published by Blackwell Publishing Ltd, 9600 Garsington Road, Oxford
OX4 2DQ, UK and 350 Main Street, Malden, MA, 02148, USA.

608
applied, being classical linear regression1 or
osprings thereof.
In the board composition and board leadership
literature, the current state of the art is not very
dierent. These two corporate governance mechanisms and their links to rm performance are
hotly debated in the economics, nance and
organization sciences literatures, at both the
theoretical and empirical level, with the evidence
reecting a mixed bag of ndings. On the
theoretical level, two main theories, agency
theory and stewardship theory, have their own
arguments for explaining the link between board
features and rm performance. From an agency
theory perspective, on the one hand, a supervisory board should be dominated by independent non-executive members in order to generate
eective monitoring of executives. Moreover, the
CEO and board chairperson should be dierent
people in order to clearly separate operational
from control responsibilities. From a stewardship
theory perspective, on the other hand, the nonexecutive board should be dominated by inside
members in order to make eective decisions,
since insiders are better informed about the rm
than outside directors. Additionally, this perspective argues that the CEO and board chair
position should be in one hand (CEO duality),
rather than be separated into two positions (CEO
non-duality), because this facilitates clear and
strong leadership. The contradictory predictions
from dierent theories are mirrored in the
available evidence. Some studies found evidence
in line with agency theory, and some others
revealed empirical support for stewardship theory. So, no convincing conclusion can be drawn
from prior studies on the impact of board
composition and board leadership on rm
performance (see, for example, Dalton et al.
(1998) for a meta-analytic review).
In theoretical and empirical work, several
eorts have been carried out to consolidate the

Classical linear regression refers to a conditional mean

linear model of the relationship between independent
and dependent variables which can be solved using
either the least squares or the maximum likelihood
method (see, for example, Johnston and DiNardo,
1997). However, the least squares method is particularly
popular in practice. In this study, we use the terms
ordinary least squares (OLS) regression and classical
linear regression interchangeably.

D. Ramdani and A. van Witteloostuijn

debate on the eectiveness of these two corporate
governance mechanisms, and to reconcile seemingly conicting evidence. On the theoretical
front, Aguilera et al. (2008) argue that the
eectiveness of corporate governance mechanisms depends on critical environmental variables.
Implementation of these mechanisms not only
produces benets, but also implies that costs have
to be incurred. Both the benets and costs tend to
vary with specic environmental variables. Consequently, the mixed evidence mirrors the variety
of external conditions that can be found in the
world of businesses. An earlier study of Demsetz
(1983) provides another theoretical reason for the
observed ambiguity. He argues that the adoption
of corporate governance devices is an endogenous outcome of a maximizing process, as rms
search for an optimal equilibrium in the face of
the advantages and disadvantages of such devices. Therefore, rm heterogeneity may imply
dierent corporate governance choices, which
explains why the impact of corporate governance
variables on rm performance varies from one
study to the other. In a similar vein, Hermalin
and Weisbach (2003), focusing on the board of
directors, argue that corporate governance mechanisms are endogenously determined in the
context of the system of board characteristics,
board action and rm performance. As a result,
they argue, the direct relationship between any
board characteristic and rm performance may
be spurious. Davis, Schoorman and Donaldson
(1997) reconcile the dierent predictions of
agency and stewardship theory by identifying
psychological attributes and situational characteristics that indicate under what conditions
agency theory or stewardship theory is likely to
hold.
On the empirical front, Boyd (1995) develops a
contingency model to explain the mixed evidence.
He argues that the sign and magnitude of the
CEO dualityrm performance relationship vary
systematically across the environmental conditions of municence, complexity and dynamism.
Elsayed (2007) explains the mixed results in the
CEO dualityperformance literature by arguing
that the direction and magnitude of this link are
dierent across industries. Black, Jang and Kim
(2006) refer to the logic that corporate governance implementation can be a signal of the
quality of the rm to then argue why regression
with instrumental variables is needed to address
r 2010 British Academy of Management.

Board Independence, CEO Duality and Firm Performance

this endogeneity issue. After all, high-performing
stock-listed rms may adopt good corporate
governance practices to signal that their rms
insiders behave well i.e. in the interest of
shareholders.
The current study is aimed at empirically
exploring an alternative way to reconcile the
mixed bag of ndings as to the rm performance
eect of CEO duality and board independence by
applying the method of quantile regression. We
believe this method oers an alternative approach to identify the mixture of relationships
that cannot be identied through classical linear
regression, as employed in almost all prior studies
on corporate governance. By using the quantile
regression method, a complete picture can be
derived of how our pair of corporate governance
mechanisms relates to rm performance at
dierent conditional quantiles, whereas classical
linear regression only estimates the conditional
mean eects of a response variable. Therefore, we
can address the question as to whether the sign
and/or magnitude of the corporate governance
rm performance relation are dierent for
dierent levels of rm performance: is this
relationship dierent for, say, high-performing
vis-a`-vis low-performing enterprises?
So, the current study contributes to the debate
by suggesting quantile regression as an advanced
estimation method new to the corporate governance literature, arguing that this can be instrumental in reconciling the seemingly conicting
ndings from studies applying ordinary least
squares (OLS) regression and its many osprings.
Specically, we expect that the relationship
between board independence and CEO duality,
on the one hand, and rm performance, on the
other hand, may well be dierent across performance quantiles.2 Moreover, we explore the
interaction eect of both board features with
board size, arguing that the latter may moderate
both board features performance impact. Ad-

A study closely related to our work is Elsayed (2007),

who employed least absolute value (LAV) regression.
The LAV regression produces estimates for the conditional median of the relationship between dependent and
independent variables. However, the LAV method is
more restrictive than quantile regression. That is,
quantile regression not only estimates at the conditional
median, but also generates estimates for all conditional
quantiles.
r 2010 British Academy of Management.

609

ditionally, we oer fresh evidence for four understudied countries: Indonesia, Malaysia, South
Korea and Thailand. Finally, in the conclusion,
we will argue that quantile regression is widely
applicable in the organization sciences.
Below, we rst oer a brief overview of the
state of the art in corporate governance research
on board independence and CEO duality. Next,
we describe what quantile regression is, in theory,
and how the method works, in practice. Subsequently, we introduce our data. After that, we
demonstrate the application of this regression
method in the context of the relationship between
board characteristics and rm performance.
Finally, we conclude with a discussion. Note that
we decided to devote a larger than usual number
of pages to the introduction of our method as
quantile regression is new to the organization
sciences.

Board independence and CEO duality

The benchmark
In the corporate governance literature, theoretical and empirical debates abound on the eectiveness of specic structures of (non-)executive
boards. The conicting predictions of agency
theory vis-a`-vis stewardship theory as to the
eectiveness of independent directors or board
independence reect one of these debates. On
the one hand, agency theory argues that a larger
proportion of independent directors will promote
better rm performance. This theory assumes
that managers are individualistic, opportunistic
and self-serving. Then, eective monitoring by
independent boards is a key to making executives
eectively pursue shareholder rather than selfinterests. The (often implicit) assumption is that
independent directors are not hindered by tendencies to pursue private interests. Consequently,
boards with more independent directors can
perform managerial monitoring tasks more eectively (Eisenhardt, 1989; Fama, 1980; Fama and
Jensen, 1983; Jensen and Meckling, 1976). On the
other hand, stewardship theory argues that
boards dominated by insiders are to be preferred
to boards dominated by outsiders as managers
are assumed to be collectivistically and proorganization oriented, as well as trustworthy.
An additional argument is that inside directors
are better informed about their rms, which

610
makes them better able to support eective
decision-making than independent directors.
Stewardship theory assumes that managers are
good stewards of corporations by acting in the
best interests of their principals. As a result, this
theory predicts that insider-dominated boards
will boost rm performance (Davis, Schoorman
and Donaldson, 1997; Donaldson, 1990; Donaldson and Davis, 1991, 1994).
This state of the art in the theoretical arena is
mirrored in empirical work on the board
independencerm performance link. The metaanalytic study of Dalton et al. (1998) clearly
reveals the mixed ndings on the relationship
between board independence (and CEO duality)
and rm performance. They conclude that
neither board composition [proportion of independent directors] nor board leadership structure [CEO duality] has been consistently linked to
rm performance (1998, p. 269). Table 1 presents
a summary of the studies on the relationship
between board independence and rm performance.
We can see that the evidence on this relationship is very mixed indeed. Some studies support
agency theory (Baysinger and Butler, 1985; Cho
and Kim, 2007; Coles, Daniel and Naveen, 2008;
Daily and Dalton, 1992, 1993; Ezzamel and
Watson, 1993; Kesner, 1987; Klein, 1998; Pearce
and Zahra, 1992; Rosenstein and Wyatt, 1990;
Schellenger, Wood and Tashakori, 1989), others
provide evidence for stewardship theory (Agrawal and Knoeber, 1996; Bhagat and Black, 2002;
Cornett, Marcus and Tehranian, 2008; Kesner,
1987; Kiel and Nicholson, 2003; Klein, 1998;
Yermack, 1996) and yet others go against both
theories (Al Farooque et al., 2007; Chaganti,
Mahajan and Sharma, 1985; Cheung, Raub and
Stouraitis, 2006; Ghosh, 2006; Kesner, Victor
and Lamont, 1986).
So, the literature review indicates that, both on
theoretical and empirical grounds, a clear and
unambiguous prediction as to the eect of the
proportion of independent board members on
rm performance is dicult to make. For that
reason, we derive two opposing hypotheses as
our benchmarks Hypothesis 1 on the basis of
agency theory, and Hypothesis 1Alt from stewardship theory.
H1: Board independence is positively associated with rm performance.

D. Ramdani and A. van Witteloostuijn

H1Alt: Board independence is negatively associated with rm performance.
A similar theoretical debate between agency
theory and stewardship theory revolves around
the issue of the eectiveness of CEO duality
(whether or not the CEO and chairman of the
board is the same person) as a governance
mechanism. As with board independence, agency
theory and stewardship theory are associated
with opposite predictions as to the eectiveness
of CEO duality as a governance mechanism. On
the one hand, agency theory argues that personal
separation of the CEO and chairperson roles
(CEO non-duality) is important to develop
eective monitoring by the board. If the CEO
and chair of the board are the same person,
agency theory argues that this is likely to create
abuse of power, since this person will be very
powerful without eective checks and balances to
control her or him. Consequently, agency theory
predicts that rms with separation of the CEO
and the chair of the board i.e. CEO non-duality
perform better than their counterparts without
separation i.e. CEO duality (Fama and Jensen,
1983). On the other hand, stewardship theory
argues that putting the roles of CEO and chair of
the board in a single hand (CEO duality) is
essential to unify and to remove ambiguity from
rm leadership. According to stewardship theory,
when the roles of CEO and chair of the board are
performed by dierent people, they often have
contrary objectives (see, for example, Dalton
et al. (1998) for a meta-analysis). So, stewardship
theory predicts that rms with CEO duality
perform better than rms without such duality.
Again, this theoretical state of the art is
reected in empirical work. Jointly, previous
studies have produced mixed results as to the
sign of the relationship between CEO duality and
rm performance. Table 2 provides a summary of
prior studies on this relationship. As above, we
can see that mixed evidence has been found on
the relationship. Some studies support agency
theory (Kiel and Nicholson, 2003; Pi and Timme,
1993; Rechner and Dalton, 1991), others provide
evidence for stewardship theory (Cornett et al.,
2008; Donaldson and Davis, 1991) and yet others
fail to support either theory (Al Farooque et al.,
2007; Baliga, Moyer and Rao, 1996; Chaganti,
Mahajan and Sharma, 1985; Cheung, Raub and
Stouraitis, 2006; Daily and Dalton, 1992, 1993;
r 2010 British Academy of Management.

Board Independence, CEO Duality and Firm Performance

Elsayed, 2007; Kesner, Victor and Lamont,
1986).
Again, from the above, we derive two opposing
hypotheses as our benchmarks in this study
Hypothesis 2 on the basis of agency theory, and
Hypothesis 2Alt from stewardship theory.
H2: CEO duality is negatively associated with
rm performance.
H2Alt: CEO duality is positively associated
with rm performance.

Conditional rm performance
Some possible explanations of the mixed evidence
as to the relationship between board characteristics, here board independence and CEO duality,
and rm performance might be related to samples
that come from dierent institutional environments (Aguilera et al., 2008) and dierent
industries (Elsayed, 2007), as well as the dierent
psychological attributes of the sampled managers
and/or the dierent characteristics of the sampled
organizations (Davis, Schoorman and Donaldson, 1997). Additionally, Tables 1 and 2 reveal
that prior studies employed standard linear
regression methods such as OLS, weighted least
squares, two-stage least squares (2SLS) or panel
data regression, with the exception of Elsayed
(2007), who used least absolute value (median)
regression. These standard linear regression
methods implicitly estimate the conditional mean
of the relationship between the variables of
interest and the dependent variable. So, the
interpretation of the relationship between the
variables of interest in this prior work should be
limited to this relationship at the conditional
mean. This is not dierent for studies using
ANOVA or MANOVA techniques, which only
imply a comparison at the mean, too.
As prior studies estimate only relationships at
the conditional mean, these studies are silent
about the relationship for low- or high-performing rms. In the current study, we postulate that
the relationships between board characteristics
and rm performance may well vary conditional
on the level of rm performance. Finkelstein and
DAveni (1994) argue that, when rm performance is high, CEO entrenchment is likely to
occur. In such circumstances, CEO status and
r 2010 British Academy of Management.

611

power may well be very strong (Harrison, Torres

and Kukalis, 1988), organizational slack is likely
to be large (Cyert and March, 1963), and perks
and bonuses might be granted to the CEO
(Jensen and Meckling, 1976). So, when rm
performance is outstanding, strong vigilance is
required to monitor the CEO in order to avoid
entrenchment.
In contrast, when rm performance is low, CEO
entrenchment is less likely to occur. Additionally,
low-performing rms need strong unity of command to recover the performance, as is known
from the organizational decline and turnaround
literatures (van Witteloostuijn, 1998). So now
board independence is expected not to be a very
potent governance device, as the likelihood of
entrenchments is small. Yet, CEO duality is
eective to accelerate rm performance for lowperformance rms by providing strong leadership
and unity of command. In all, this logic implies
that board independence can be expected to be
more potent in promoting rm performance in
high- vis-a`-vis low-performance rms, and that
CEO duality can be expected to be more potent in
promoting rm performance in low- vis-a`-vis highperformance rms. This gives Hypotheses 3 and 4.
H3: Board independence is particularly
eective to promote rm performance for
high-performance rms, but not so for lowperformance rms.
H4: CEO duality is particularly eective to
promote rm performance for low-performance rms, but not so for high-performance
rms.

Board size
On the one hand, larger board size provides,
potentially, more monitoring resources, which
may enhance rm performance (Alexander,
Fennell and Halpern, 1993; Goodstein, Gautam
and Boeker, 1994; Mintzberg, 1983; Pfeer, 1972,
1973; Pfeer and Salancik, 1978). On the other
hand, however, larger board size makes coordination, communication and decision-making
more troublesome, and larger board size may
trigger free-riding issues among the many board
members. All this may lower rm performance
(Hermalin and Weisbach, 2003; Jensen, 1993;

Proportion of outside directors

Daily and Dalton (1992)

Proportion of outside directors

Proportion of outside directors

Proportion of outside directors

Agrawal and Knoeber (1996)

Yermack (1996)

Klein (1998)

Proportion of inside directors on

nance
Proportion of inside directors on
investment

Proportion of outside directors

Proportion of independent directors

of outside directors

Daily and Dalton (1993)

Ezzamel and Watson (1993)

Pearce and Zahra (1992)

of neutral outside

of corporate outside

of nancial outside

Proportion
directors
Proportion
directors
Proportion
directors
Proportion

ROA

ROA
Productivity
Market return
Productivity

Tobins Q
Tobins Q

ROA, ROE, PER

Average prot to capital ratio
Change in prot to capital ratio

ROA
ROE
EPS
Net prot margin
ROA
ROE
Priceearnings ratio (PER)

Prot margin
Return on equity (ROE)
Return on assets (ROA)
Earnings per share (EPS)
Stock market performance
Total return to investment
(ROI)
ROA
ROE
ROI
Abnornal market return

Proportion of inside directors

Rosenstein and Wyatt (1990)

Illegal activities

Proportion of outside directors

Proportion of outside directors

Relative return on equity

Firm failure

Dependent variables

Proportion of independent directors

Proportion of outside directors

Independent variables

Schellenger, Wood and

Tashakori (1989)

Baysinger and Butler (1985)

Chaganti, Mahajan and Sharma
(1985)
Kesner, Victor and Lamont
(1986)
Kesner (1987)

Author(s)

Table 1. Summary of the studies on board independence

186 small rms listed in the USA

184 UK companies from Exstat
database and Hambro Company
Guide
400 large US rms
452 large US industrial
corporations
641 rms in listed S&P 500

100 US rms listed in Inc.

magazine

450 rms from Fortune 500

1251 observations of director

announcements in NYSE and
AMEX corporations

526 random rms from

Compustat database

205 rms of Fortune 500

266 major US corps from Forbes

21 pairs of retailing rms in the
USA
384 rms of Fortune 500

Data

Signicantly positive

Not signicant
Signicantly negative
Not signicant
Signicantly positive

Signicantly negative
Signicantly negative

Signicantly positive
Signicantly positive
Signicantly positive
Not signicant
Not signicant
Not signicant
Outside signicantly
higher
Signicantly positive
Not signicant
Signicantly positive

Signicantly positive

Not signicant

Signicantly positive
Not signicant
Not signicant
Signicantly positive

Signicantly positive
Not signicant
Signicantly positive
Not signicant
Signicantly negative
Not signicant

Not signicant

Signicantly positive
Not signicant

Results

OLS and 2SLS regression

OLS regression and panel data
regression with xed eects
OLS regression

MANOVA
OLS regression

ANOVA and MANOVA

MANOVA

Weighted least squares

Correlation analysis

Correlation analysis

ANOVA and OLS regression

Simultaneous equation regression

ANOVA

Methods

612
D. Ramdani and A. van Witteloostuijn

r 2010 British Academy of Management.

r 2010 British Academy of Management.

Proportion of non-executive
directors

Proportion of independent nonexecutive directors

Outside directors participation rate
Proportion of non-executive
directors
Proportion of outside directors

Kiel and Nicholson (2003)

Ghosh (2006)

Cheung, Raub and Stouraitis

(2006)
Cho and Kim (2007)
Al Farooque et al. (2007)
Discretionary accruals
Adjusted EBIT/assets
Tobins Q

347 rms listed in Korea

723 rms in Bangladesh

ROA
Market to book value equity

8165 year-rms taken from

Execucomp database

100 rms of S&P Index

127 listed manufacturing rms in

India

348 Australian listed corporations

1338 listed rms in Hong Kong

934 large US corporations

Tobins Q
Operating income to assets ratio
Sales to assets ratio
Stock price return
Assets growth
Operating income growth
Sales growth
Tobins Q
ROA
ROA
Adjusted Tobins Q
Average value of ROA, ROE
and ROS
Market-adjusted CAR

ROA
Productivity
Market return
Productivity

ROS, return on sales; CAR, cumulative abnormal returns; EBIT, earnings before interest and tax.

Proportion of inside directors

Proportion of outside directors

Bhagat and Black (2002)

Cornett, Marcus and Tehranian

(2008)
Coles, Daniel and Naveen
(2008)

Proportion of inside directors on

compensation committee
Board independence (proportion of
independent directors minus
proportion of insiders)

Proportion of inside directors on

audit committee

Signicantly negative
Signicantly positive
Signicantly negative

Signicantly positive
Not signicant

Not signicant

Signicantly negative
Signicantly negative
Not signicant
Not signicant
Not signicant
Not signicant
Not signicant
Signicantly negative
Not signicant
Not signicant
Not signicant
Not signicant

Not signicant
Not signicant
Not signicant
Signicantly negative

3SLS regression

Pooled and panel data regression

OLS regression
OLS and 2SLS regression

OLS regression

OLS regression and correlation

analysis
OLS regression

OLS and 3SLS regression

Board Independence, CEO Duality and Firm Performance

613

CEO duality
CEO duality

CEO duality
CEO duality

CEO non-duality
CEO duality

Donaldson and Davis (1991)

Daily and Dalton (1992)

Pi and Timme (1993)

Daily and Dalton (1993)

Baliga, Moyer and Rao (1996)

Kiel and Nicholson (2003)

Market to book value equity

CEO duality
CEO duality
Lagged CEO duality

Elsayed (2007)

Cornett, Marcus and Tehranian

(2008)

ROA
Tobins Q
Discretionary accruals
Adjusted EBIT/assets

ROA
Market-adjusted CAR

CEO duality

Cheung, Raub and Stouraitis

(2006)
Al Farooque et al. (2007)

ROE
PER
Market value of the rm
Tobins Q

ROE
Priceearnings ratio
Cost eciency
ROA
ROA

ROE
Prot margin
ROE
ROA

Firm failure

Dependent variables

Illegal activities
ROI

Independent
variables

Chaganti, Mahajan and Sharma

CEO duality
(1985)
Kesner, Victor and Lamont (1986) CEO duality
Rechner and Dalton (1991)
CEO duality

Author(s)

Table 2. Summary of the studies on CEO duality

100 rms of S&P Index

357 Egyptian public rms

723 rms in Bangladesh

736 listed rms in Hong Kong

172 rms of Fortune 500

348 Australian listed
corporations

112 publicly traded US

commercial banks
186 small listed corporations in
the USA

337 US corporations
100 US rms listed in Inc.
magazine

21 pairs of retailing rms in the

USA
384 rms of Fortune 500
141 companies of Fortune 500

Data

Not signicant
Not signicant
Signicantly positive
Signicantly negative

Not signicant

Not signicant
Not signicant

Not signicant
Not signicant
Not signicant
Signicantly negative

Not signicant
Not signicant
Signicantly negative
Not signicant
Not signicant

Duality signicantly lower

Duality signicantly lower
Duality signicantly higher
Not signicant

Not signicant
Duality signicantly lower

Not signicant

Results

Panel data regression

OLS and 2SLS

regression
LAV regression

OLS regression

OLS regression
OLS regression

MANOVA

OLS regression

ANOVA
ANOVA and
MANOVA

ANOVA
ANOVA and
MANOVA

ANOVA

Methods

614
D. Ramdani and A. van Witteloostuijn

r 2010 British Academy of Management.

615

Board Independence, CEO Duality and Firm Performance

Lipton and Lorsch, 1992). This implies that
board size might moderate the impact of board
independence on rm performance. Depending
upon whether the eect of board size is positive
or negative, this moderating eect is expected to
be positive or negative accordingly. So, again, we
suggest Hypotheses 5 and 5Alt.
H5: Board size positively moderates the impact
of board independence on promoting rm
performance.
H5Alt: Board size negatively moderates the
impact of board independence on promoting
rm performance.
Similarly, we postulate that board size moderates
the association between CEO duality and rm
performance. As explained above, CEO duality
may impact rm performance positively due to
the advantages of strong leadership and unity of
command (stewardship theory), or negatively due
to the disadvantages of weaker monitoring and
controlling (agency theory). This gives two
possible interaction eects with board size. In
the rst case, the positive impact of CEO duality
interacts with the positive eect of a larger board.
In the second case, the negative impact of CEO
duality interacts with the negative eect of a
larger board. Together, this gives Hypotheses 6
and 6Alt.
H6: Board size positively moderates the impact
of CEO duality on promoting rm performance.
H6Alt: Board size negatively moderates the
impact of CEO duality on promoting rm
performance.

How does quantile regression work?

Quantile regression estimates the conditional
quantiles of a response variable in a linear model,
providing a complete view of the possible
relationships between a response variable and
explanatory variables (Koenker and Bassett,
1978; Koenker and Hallock, 2001). Compared
to OLS regression, which only estimates the
conditional mean of a response variable, quantile
regression is more powerful by producing separ 2010 British Academy of Management.

rate estimates for all conditional quantiles of a

response variables distribution.3 Besides, quantile regression does not require strict assumptions
like classical linear regression as to normality,
homoscedasticity and absence of outliers (Johnston and DiNardo, 1997). Formally, following
Koenker and Bassett (1978), a conditional
quantile function can be expressed as follows:
Qy yi jxi ay x0i by

with y 2 0; 1

where yi is the response variable of observation i,

xi is the vector of covariates representing
individual observation i, y represents the yth
quantile, where quantile refers to a point taken
along the cumulative distribution, and subscript
i 5 1, 2, . . ., n reects an index for individual
observations. Qy(yi|xi) denotes the yth conditional quantile of yi given xi. By way of
comparison, recall that the OLS regression
function is expressed as Eyjx myjx a x0i b,
which is the classical linear function that estimates the conditional mean my|x, namely the
average value of y for a given value of x.
The optimization problem of the conditional
quantile function is
2

min 4

b2RK

X
i2fi:yi x0i by g

yjyi 

x0i by j

1  yjyi 

x0i by j5

i2fi:yi < x0i by g

2
where R indicates the dimensions of the independent variables (K). The optimization problem
of this function is to search for the yth quantile
regression estimators (b(y)) that minimize the
absolute value of a weighted sum of the residuals
between observed values (yi) and tted values
(xi0b). We assign a weight of y to the points lying
below the quantile regression line (the rst term
3
Quantile regression has been widely applied in dierent
literatures outside the organization sciences, generally,
and corporate governance, particularly. Illustrative
examples are Goel and Ram (2004) and Manning,
Blumberg and Moulton (1995) in elasticity of demand
work, Arias, Hallock and Escudero (2001), Buchinsky
(2001) and Eide and Mark (1998) in education
economics, Barreto and Hughes (2004) in economic
growth studies, Buchinsky (1994), Garc a, Hernandez
and Nicolas (2001), Machado and Mata (2001) and
Nielsen and Rosholm (2001) in wage analysis, Ribeiro
(2001) in labour economics, Abrevaya (2001) in population economics, Bassett and Chen (2001) in portfolio
investment research, and Cade, Terrell and Schroeder
(1999) and Knight and Ackerly (2002) in ecology
science.

616

D. Ramdani and A. van Witteloostuijn

in equation (2)), and a weight of 1 y to the

points located above the quantile regression line
(the second term in equation (2)). Furthermore,
^ is estimated
the estimated covariance matrix S
through
X
^ y
b

y1  y 1
X 0 X1
n
fey 02

where fey 0 is the probability density of the

error term e(y) evaluated at the yth quantile of the
error distribution (Hao and Naiman, 2007). An
estimated standard error for coecient estimator
^y can be obtained by taking the square root
b

of the corresponding diagonal element of covar^

iance matrix S.
Intuitively, how quantile regression works, visa`-vis OLS regression, is illustrated in Figures 1(a),
1(b) and 1(c) with each gure representing one
particular case. We use these three panels to
clearly distinguish between three cases: with all
OLS assumptions satised (Figure 1(a)), with the
homoscedasticity assumption violated (Figure
1(b)) and with a skewed distribution and outliers
(Figure 1(c)). We present three-dimensional
gures, in which y is a response variable, x
denotes an explanatory variable and the third
axis gives density. Figure 1(a) gives the bench-

Figure 1. (a) Normal distribution with homoscedasticity; (b) normal distribution with heteroscedasticity; (c) skewed distribution with
outliers

r 2010 British Academy of Management.

Board Independence, CEO Duality and Firm Performance

mark case in which the data perfectly full the
normality, homoscedasticity and no-outlier assumptions. In this case, estimation of the conditional mean of the response variable is
representative for the relationship between the
response variable and explanatory variables. If
we perform quantile regression, all the coecients that are estimated for each and every
quantile are exactly equal to the coecient
produced by OLS regression.4 However, if the
assumptions of the OLS regression are violated,
then the mean regression line will not be
representative, and hence this line will not
adequately demonstrate the relationship between
the response and explanatory variables. The case
with a violation of the homoscedasticity assumption is illustrated in Figure 1(b),5 and the case
with skewed data is illustrated in Figure 1(c).
4

We can see that, for a given x1, the value of y1 could be

ya1, yb1 , yc1 or any other value of y along the broken line
across ya1, yb1 and yc1. Similarly, for a given x2, the value
of y could be ya2, yb2 , yc2 or any other value of y along the
broken line across ya2, yb2 and yc2. Applying similar logic,
we could draw the value of y for a given x3. In Figure
1(a), the conditional mean regression line is ^
y for OLS,
where y^ denotes the predicted value of y produced by
OLS regression, in which the line links average values of
y for given values of x. Clearly, this regression line is
representative for the model that species the relationship between y and x if the assumptions to perform OLS
regression are all fullled: normal distribution, equal
variance (homoscedasticity) and absence of outliers.
Moreover, if we perform quantile regression to estimate
values of a conditional quantile of y, the slope of the
quantile regression line should not be dierent from the
OLS regression line. For example, assume that the value
of y at quantile 0.10 given x1 is ya1, given x2 is ya2 and
given x3 is ya3. Then, the quantile 0.10 regression line is ^
y
for y 5 0.10, where y^ denotes the predicted value of y
and y 5 10 is quantile 0.10. If this regression line runs
parallel with ^
y for OLS, then the regression coecients
associated with both lines are equal. Similarly, estimating the quantile 0.90 (y 5 0.90) regression line (^
y for
y 5 0.90) produces the regression line that runs parallel
with the ^
y for OLS and ^
y for y 5 0.10 lines. And so
on.
5
We can see that the conditional variance is not equal at
all values of x. In this case, the OLS regression line is not
adequately representing the relationship between the
response and the explanatory variables for the whole
sample. For example, at the quantiles 0.10 and 0.90, the
slopes of both regression lines (which are at ^
y for
y 5 0.10 and ^
y for y 5 0.90, respectively) are dierent
from ^
y for OLS. Therefore, the coecient estimates
produced by OLS regression when the conditional
variance is unequal cannot adequately capture the shape
r 2010 British Academy of Management.

617

Note that the estimation of quantile regression

coecients is based on the weighted sum of the
residuals for the whole sample, and not just on
the portion of the sample at that quantile.
Therefore, we never lose degrees of freedom,
which is especially important when the number of
observations in the sample is not very large. The
optimization problem of the quantile regression
function can be solved by linear programming
methods (see Hao and Naiman (2007) for
algorithmic details), since the problem is based
on order statistics without having an explicit
form (Buchinskiy, 1994). Additionally, note also
that quantile regression classies the sample into
low- up to high-level groups (quantiles) of the
dependent variable (y), which is dierent from
simple categorization.6 In quantile regression, the
grouping of the dependent variable (y) is conditional on the independent variable (x); in simple
categorization, grouping of the dependent variable is just to sort out the value of this dependent
variable (e.g. Elsayed, 2007).

Data and methodology

Empirical setting
Our empirical setting includes the enterprises
listed on the stock exchanges in four East Asian
countries Indonesia, Malaysia, South Korea
and Thailand in 20012002. This is an
attractive sample to explore the relationship
between board characteristics and rm performance for at least the following reasons. Corof the relationship between the response and the
explanatory variables for the whole sample.
6
To illustrate this, compare the observations at (x1, yb1 )
and (x2, ya2) in Figure 1(a). If we follow the simple
classication method, we could classify these two
observations as one group with low categorization since
y1 and y2 are almost identical in value. However, in
quantile regression, we should distinguish between the
(x1, yb1 ) observation and the (x2, ya2) observation by
looking at the value of the independent variable (x) for
each observation. The (x1, yb1 ) observation reaches y1
given x1; and the (x2, ya2) observation reaches y2 given x2.
We know that both observations are associated with an
almost similar value of y (y1  y2), but x2x1. Therefore,
we could conclude that the (x1, yb1 ) observation has a
better conditional y value than the (x2, ya2) observation,
given the respective values of the independent variable
(x). As a result, in quantile regression, the (x1, yb1 )
observation will be grouped in the higher category (line)
of the quantile regression than the (x2, ya2) observation.

618
porations in these countries have similar characteristics, such as high concentration of family
ownership, often belonging to a business group
with a pyramidal structure and cross-ownership,
low corporate transparency, extensive and diversied business structures, and risky nancial
strategies (Claessens and Fan, 2002; Claessens
et al., 1999). As family ownership is very
common, with family members often sitting in
the top management team, the key agency
relationship in our sample involves the controlling owner and minority shareholders rather than
managers vis-a`-vis outside shareholders (Claessens, Djankov and Lang, 2000).
These characteristics make the study of the
rm performance impact of board independence
and CEO duality important in this sample. We
could expect that both corporate governance
mechanisms work well, as suggested by agency
theory. The literature on corporate governance in
these countries, mostly viewed from a corporate
nance perspective, is heavily biased toward the
agency theory perspective. Then, the argument is
that board independence and CEO non-duality
are suited to minimize agency cost (Claessens
and Fan, 2002; Claessens, Djankov and Lang,
2000; Lemmon and Lins, 2003; Mitton, 2002).
So, this empirical setting oers the opportunity to
test this logic by looking into the roles of board
independence and CEO duality as a governance
mechanism to enhance rm performance conditional on rm performance and in interaction
with board size.
Data and measures
We use a data set from a corporate governance
survey in Indonesia, Malaysia, Thailand and
South Korea conducted by the Asian Development Bank Institute. The questionnaire was
mailed to respondents during JulyOctober
2003 (Nam and Nam, 2004). The survey took a
sample of stock-listed enterprises. The sample
involves about 19.82% (66 rms) of the total
population of rms listed on the stock exchange
in Indonesia, 6.48% (111 rms) in South Korea,
8.28% (75 rms) in Malaysia and 11.21% (61) in
Thailand. When needed, we collected data from
annual reports to add missing information.
Respondents of the survey are corporate secretaries, executives and non-executives. Questions
addressed to corporate secretaries asked about

D. Ramdani and A. van Witteloostuijn

factual information as to corporate governance
implementation. Questions addressed to the
other respondents involved their opinion of
corporate governance implementation. In this
study, we use the factual corporate governance
information as we are not interested in perceptual
issues.
Our dependent variable is rm performance,
measured in terms of the average value of return
on assets (ROA) 20012002, being dened as
earnings before interest and tax (EBIT) divided by
total assets (book value). Our independent variables are (a) the proportion of independent
directors and (b) CEO duality. The proportion
of independent directors is simply the number of
independent directors divided by the total number
of directors. The number of independent directors
is from the question How many independent
directors does your board have?, and the total
number of directors from the question How many
directors does your board have in total? The
latter question also gives the board size measure.
The CEO duality variable is from the question
Does the CEO of your rm also serve as board
chair? We coded CEO duality as 1, and nonduality as 0. Our selection of control variables is
based on the corporate governancerm performance literature (e.g. Black, Jang and Kim, 2006):
assets (book value in million US$), xed assets to
sales ratio (book value), debt to equity ratio (book
value), growth of sales in 19972002, and year of
rst stock exchange listing (up to 2002). Industry
dummies are included to correct for industry
eects, and a country dummy is added to correct
for institutional dierences across our countries.
Table 3 provides the relevant descriptive
statistics for our variables. All variables with
the exception of dummy variables are transformed into natural logarithms in order to reduce
skewness (see Mukherjee, White and Wuyts,
1998, pp. 97105). We also performed data
centring, implying that the mean value of all
variables is zero. Data centring facilitates easy
comparison of measures of central tendency and
dispersion. Data are normally distributed if the
value of skewness is zero and kurtosis is lower
than 3. If skewness is zero, the distribution of the
data is symmetric. If kurtosis is lower than 3, the
tails of the data are thin (Mukherjee, White and
Wuyts, 1998).
We can see that the skewness value for all
variables is not close to zero, indicating that the
r 2010 British Academy of Management.

619

Board Independence, CEO Duality and Firm Performance

Table 3. Descriptive statistics
Variable

Median

Std Dev.

ROA
Proportion of independent directors
CEO duality
Board size
Assets
Fixed assets per sales
Debt to equity ratio
Growth sales
Year listed

0.002
 0.002
 0.392
0.020
 1.069
 0.030
 0.847
0.016
0.095

0.058
0.017
0.488
0.416
4.103
1.285
3.921
1.401
0.826

Min

Max

Skewness

Kurtosis

 0.555
 0.030
 0.392
 1.367
 7.206
 6.736
 6.832
 9.832
 2.390

0.381
0.066
0.608
1.076
9.276
5.081
7.672
6.584
1.460

 2.376
0.486
0.444
 0.293
0.307
 0.463
0.223
 1.761
 1.225

36.893
4.390
1.203
2.988
1.774
7.558
1.729
23.839
4.516

Note: The data are normalized such that the mean values are zero.

variables are not symmetrically distributed.

Additionally, the kurtosis value is above 3 for
four variables i.e. ROA, proportion of independent directors, xed assets per sales and
growth of sales. This signals observations with
extreme values. Moreover, the mean is dierent
from the median for three variables: CEO
duality, assets and debt to equity ratio. This
implies that the distribution of the data is not
bell-shaped. The JarqueBera normality test for
each variable indicates that only for the board
size variable can we not reject the null hypothesis
that the data are normally distributed
(w2(2) 5 4.48 and po0.10)). For the other variables, we reject the null hypothesis that the data
are normally distributed at the 1% level of
condence with two degrees of freedom. In all,
these data diagnostics suggest that quantile
regression may be more appropriate than OLS
regression to investigate if and how the relationship between corporate governance and rm
performance might dier across quantiles.

Evidence
The main results are presented in Table 4,
reporting the OLS regression and quantile
regression outcomes. We also performed a robust
regression7 in order to verify whether our OLS
ndings are aected by outlier observations. The
OLS regression is aimed to test the benchmark
hypotheses, which are Hypotheses 1 and 2. The
results show that in the OLS regression (using the
reg command in STATAs (version 9.2) statis7

In a general sense, robust regression is a method that

attempts to reduce the impact of outliers on regression
estimators by smoothing the eect of outliers on the
coecient regression (see, for example, Rousseeuw and
Leroy, 1987).
r 2010 British Academy of Management.

tical software), we should reject Hypotheses 1

and 1Alt that board independence has either a
positive or negative association with rm performance. However, the robust regression (rreg
command) reveals that we cannot reject Hypothesis 2Alt. That is, we nd evidence to support
stewardship theorys claim that CEO duality is
positively associated with rm performance
(0.014, with po0.05). However, our OLS results
may suer from endogeneity. Endogeneity of
board characteristics through rm performance
(Hermalin and Weisbach, 2003) would imply that
the OLS estimates are biased and inconsistent
(Johnston and DiNardo, 1997).
To address endogeneity, we employed instrumental variables for our two independent variables: board independence and CEO duality.
Finding instrumental variables is not an easy task,
however, since we need to nd variables that are
correlated with our explanatory variables but not
with the error terms. Based on the data we have,
we identied three potential instrumental variables: external debt, a dummy variable indicating
whether the CEO is a professional manager or has
a family relation with the majority owner, and a
dummy variable signalling whether the board of
director includes at least one foreign member.
First, larger external debt may be associated with
a larger proportion of independent directors, since
external parties prefer to have this monitoring
device in place. In fact, simple correlation analysis
supports this argument: the correlation for external debt and proportion of independent directors is 0.27, and the one for external debt and
ROA only 0.002. Second, the professional CEO
dummy might indicate that the corporation seeks
to apply good operating and monitoring practices.
Then, the corporation would be open for a larger
proportion of independent directors and CEO
non-duality. Indeed the correlation of the profes-

620

D. Ramdani and A. van Witteloostuijn

Table 4. OLS and quantile regression results

OLS regression

Corporate governance variables

Proportion of independent directors
CEO duality
Control variables
Board size
Assets
Fixed assets per sales
Debt to equity ratio
Growth sales
Years listed
Country dummy
South Korea
Malaysia
Thailand
Industry dummy
Chemical
Iron and metal
Electronics
Transport equipment
Distribution and trade
Food and beverages
Agribusiness
Construction
Constant
R2 (Pseudo R2 for quantile regression)

Quantile regression

OLS

Robust OLS

Q0.10

0.267
0.015

0.098
0.014**

 0.243
0.006

 0.046
0.009

0.008
0.007***
0.001
 0.008***
0.002
0.000

0.040*
0.014
0.006
 0.018
0.009*
0.001

0.013
0.010
0.010**
0.006***
0.003
0.001
 0.011***  0.007***
0.004*
0.002*
0.001
 0.001

0.011
0.012
0.005
0.001
0.005
0.002
 0.007**  0.005
0.001
 0.002
 0.001
 0.006

0.007
0.004
0.000

 0.016
0.000
 0.022

 0.005
0.013
 0.014

0.003
0.003
 0.005
0.009
0.010
0.009
 0.003
0.003
 0.006
0.079

 0.013
 0.001
 0.021
0.003
 0.006
 0.006
 0.012
 0.012
0.038
0.068

 0.015
0.015
 0.026
0.004
 0.007
 0.008
 0.028
 0.012
0.053
0.112

0.023**
0.012***
0.007**
 0.015***
0.005**
0.002

Q0.25

0.013
 0.001
 0.015

0.000
0.006
0.003

0.053
0.003
 0.009

0.015
 0.001
0.003

 0.003
0.019
 0.016
0.017
0.006
 0.009
0.012
0.010
 0.005
0.139

0.005
0.005
 0.003
0.012*
0.013*
0.011
 0.002
0.006
 0.006
0.136

0.007
0.004
0.005
0.033
 0.001
0.004
0.033
0.028
 0.072
0.160

0.013
0.014
0.005
0.023**
0.024**
0.012
0.022
0.024**
 0.037
0.083

Q0.5 (median)
0.283**
0.010*

Q0.75

Q0.90

0.078
 0.191
0.028**
0.029

Notes: The benchmark for the industry dummy is utilities.

*po0.1; **po0.05; ***po0.01.

sional CEO dummy with the proportion of

independent directors is 0.19 and with CEO
duality  0.15, whilst the correlation with ROA
is a very low at 0.03. Third, a foreign board
member is another potential monitoring-strengthening device. The correlation of the foreign board
member dummy with CEO duality is  0.16, but
only 0.11 with ROA.
This correlation analysis indicates that this set of
three variables may be used as instrumental
variables, since all three are closely correlated with
both independent board and CEO duality variables, but not with our dependent rm performance variable. Our empirical strategy is to run
2SLS to correct for the potential endogeneity of
board independence and CEO duality, using our
three instrumental variables. We employed the
Hausman test to check if the OLS estimates are
consistent (Johnston and DiNardo, 1997, pp. 338
342). The null hypothesis of this test is that the
OLS and 2SLS estimators are similar. We found
that the Hausman test statistic is 0.19, which is
lower than the 5% critical value from the w2(2)
distribution. This result of the Hausman test shows
that we cannot reject the null hypothesis. So, our

OLS estimates are consistent, and endogeneity is

not an issue. Therefore, we can proceed without
the instrumental variables procedure in further
analyses.
The other results in Table 4 are from the
quantile regression for y 5 0.1, y 5 0.25, y 5 0.50,
y 5 0.75 and y 5 0.90. We use the qreg command in STATAs (version 9.2) statistical software. We nd that the proportion of independent
directors has a signicantly positive association
with ROA at y 5 0.50 (0.283, with po0.05), and
that CEO duality has a positively signicant
association with ROA at y 5 0.50 (0.010, with
po0.10) and y 5 0.75 (0.028, with po0.05). The
dierent results from the OLS vis-a`-vis the
quantile regression indicate that estimating only
the conditional mean of the response variable can
be inappropriate when the data fail to meet the
assumptions needed to perform an OLS regression analysis. Checking for normality of the
residuals from the OLS regression, using the
ShapiroWilk W test, shows that we must reject
the null hypothesis that the residuals are normally
distributed (W 5 0.76 and po0.01). Checking for
homoscedasticity of the residuals from the OLS
r 2010 British Academy of Management.

621

Board Independence, CEO Duality and Firm Performance

(a)

(b)

0.10

0.05
0.00

CEO Duality

proportion of indep. directors

1.00

0.00

1.00

0.05
2.00
0

.2

.4
.6
Quantile

.8

.2

.4
.6
Quantile

.8

Figure 2. Estimates for (a) independent directors and (b) CEO duality
Note: The 90% condence intervals are superimposed over the shaded areas.

regression, using the BreuschPagan test, indicates that we have to reject the null hypothesis
that the variance of the residuals is homogeneous
(w2(1) 5 62.42 and po0.01). The dierent results
for OLS vis-a`-vis quantile regression are therefore
not surprising. In our sample, estimating the eect
of corporate governance variables on rm performance at dierent points of the rm performance conditional distribution using quantile
regression is clearly warranted, since each quantile
may be associated with dierent eects. Note that
although the robust OLS regression (column 3 in
Table 4) gives better results than the OLS
regression (column 2 in Table 4) by showing that
CEO duality is signicantly and positively associated with ROA, the robust OLS estimates must
be interpreted as the relationship at the conditional mean.
The quantile regression results show, indeed,
that the eects of corporate governance variables
dier across the quantiles in the conditional
distribution of rm performance. To reveal this,
the eects for all quantiles are visualized in Figures
2(a) and 2(b) for the proportion of independent
directors and CEO duality, respectively. These
gures are produced in STATA using the grqreg
command after running the qreg command. We
are particularly interested in how the eect of a
corporate governance mechanism varies with the
quantiles. Note that we only report the ndings for
r 2010 British Academy of Management.

our corporate governance variables, for parsimony

and given our theoretical focus, and not for the
control variables. We plot the coecients of
corporate governance variables along the vertical
axis and the quantiles along the horizontal axis.
The line in the middle of the shaded area reects
the coecient estimates of the quantile regression
in dierent quantiles. The broken line in each
gure gives the standard OLS estimate of the
conditional mean eect. The shaded grey area
depicts a 90% pointwise condence band for the
quantile regression estimates.
Figure 2(a) shows that the proportion of
independent directors has the largest positive eect
around y 5 0.50, being smaller in all quantiles
above and below 0.50. Almost all coecients of
the quantile regression are lower than the estimate
from the OLS regression, except at y 5 0.50. This
eect is signicantly positive around y 5 0.50, but
it is not signicant in the quantile lower than 0.30
and in the quantile higher than 0.70, which is
indicated by the very broad band of the interval of
condence. Here, the quantile regression analysis
reveals that the eect of the proportion of
independent directors is dierent across quantiles
indeed. In a standard OLS regression, this cannot
be revealed as only a single estimate is produced,
which is conditional on the mean. These results of
quantile regression imply that the proportion of
independent directors is an eective governance

622
mechanism in mediocre performing rms, but not
for under-performing and top-performing rms.
The results partly support Hypothesis 3, arguing
that board independence is not an eective way to
advance rm performance for low-performance
rms. However, we failed to nd evidence that
board independence is eective for high-performance enterprises.
Figure 2(b) shows that CEO duality is signicantly and positively associated with rm performance in the range from quantiles 0.30 to 0.75
with the eect being larger in the higher quantiles.
The results also show that CEO duality is not
signicant in the quantiles lower than 0.30 and in
the quantiles larger than 0.75, which is indicated
by the very broad band of the interval of
condence. This implies that CEO duality as a
corporate governance mechanism is not functioning well for under-performing and top-performing
rms. In this case, these ndings show that the
relationship between CEO duality and rm
performance is dierent across quantiles, which is
again something that cannot be captured by
applying the classical linear regression. The ndings on the relationship between CEO duality and
rm performance using quantile regression partly
support Hypothesis 4, indicating that CEO duality
is not an eective way to promote rm performance for high-performance corporations. However, we nd no evidence to support the argument
that CEO duality is an eective way to promote
rm performance of low-performance enterprises.
The other additional results are provided in
Table 5, reporting the estimates of the interaction
eects as to board characteristics and board size.
In Model 1, the interaction of the proportion of
independent directors and board size is not
signicant. This goes against Hypotheses 5 and
5Alt: evidence for a moderating eect of board
size on the relationship between the proportion of
independent directors and rm performance is
missing. In contrast, in Model 2, the interaction
of CEO duality and board size is negatively
signicant (  0.036, with po0.05). In Model 3,
with all interaction eects included, similar
results are found for the CEO dualityboard size
interaction variable (  0.036, with po0.05), with
the interaction eect of the proportion of
independent directors with board size being
insignicant. Based on this regression analysis,
we reject Hypothesis 6. However, we cannot
reject Hypothesis 6Alt, as we found evidence that

D. Ramdani and A. van Witteloostuijn

Table 5. Interaction of board size with board characteristics
Model 1
Corporate governance
variables
Proportion of
0.168
independent directors
CEO duality
0.015
Control variables
Board size
0.024**
Assets
0.013***
Fixed assets per sales
0.007**
Debt to equity ratio
 0.015***
Growth sales
0.005**
Years listed
0.002
Country dummy
South Korea
0.014
Malaysia
0.001
Thailand
 0.015
Industry dummy
Chemical
 0.003
Iron and metal
0.020
Electronics
 0.015
Transport equipment
0.017
Distribution and trade 0.006
Food and beverages  0.009
Agribusiness
0.015
Construction
0.011
Interaction eects
Proportion of
 0.487
independent directors *
board size
CEO duality *
board size
Constant
 0.007
2
R
0.143

Model 2

Model 3

0.397

0.298

0.018

0.018

0.026**
0.027**
0.013*** 0.014***
0.007**
0.008**
 0.016***  0.016***
0.005**
0.005**
0.002
0.002
0.004
 0.007
 0.022

0.006
 0.005
 0.022

 0.004
0.020
 0.019
0.014
0.004
 0.011
0.010
0.009

 0.004
0.020
 0.018
0.013
0.004
 0.011
0.013
0.010
 0.499

 0.036**  0.036**
0.000
0.151

 0.002
0.155

Notes: The benchmark for the industry dummy is utilities.

po0.1; **po0.05; ***po0.01.

board size negatively moderates the relationship

between CEO duality and rm performance.

Discussion and conclusion

The general purpose of this study is to explore the
impact of board independence and CEO duality
on rm performance by applying the quantile
regression method. Our study is dierent from
prior work that analyses the relationship between
board characteristics and rm performance at the
mean value, using either conditional mean
regression (OLS or 2SLS) or ANOVA/MANOVA methods. In contrast, the quantile regression
method generates dierent estimates at conditional quantiles. This implies that we can explore
the impact of board characteristics on rm
performance at dierent levels of rm performance i.e. for low, mediocre and high levels of
r 2010 British Academy of Management.

Board Independence, CEO Duality and Firm Performance

rm performance. We believe that investigating
the eects of board characteristics on rm
performance at dierent levels of rm performance provides a better understanding of the
conditionality of this eect, since the empirical
literature, to date, does not dierentiate these
eects across dierently performing rms.
So, this study goes beyond the current state of
the art in the empirical literature on the eects of
board characteristics on rm performance. We
hypothesize that low-performing enterprises may
require dierent types of board characteristics
compared to high-performing rms. On the one
hand, we suggest that low-performing rms need
strong leadership and unity of command to
advance their performance. This is in line with
stewardship theory. On the other hand, in contrast,
high-performing rms need good surveillance to
avoid CEO entrenchment. This is in accordance
with agency theory. From this it follows that lowperforming enterprises will benet from CEO
duality, and high-performing corporations from
board independence. So, our argument reconciles
agency and stewardship theories by suggesting that
the eect of board composition and leadership is
conditional on initial rm performance.
Our application of quantile regression in the
context of the relationship between corporate
governance and rm performance in Indonesia,
Malaysia, South Korea and Thailand revealed
that the impact of the pair of prominent
corporate governance mechanisms focused upon
here proportion of independent directors and
CEO duality varies according to (conditional)
level of rm performance. Our main results show
that board independence and CEO duality are
eective in average-performing rms, but not
eective for their low- and high-performing
counterparts. This set of ndings suggests that
the performance impact of board independence
and CEO duality are conditioned non-linearly by
initial performance. This pair of results may
reconcile the contrasting insights from agency
and stewardship theory: predictions of either
theory are supported or rejected, depending upon
the level of rm performance.
On the one hand, independent directors can
operate either as a watchdog to control the
managers, as suggested by agency theory, or as a
good steward and mediator to bring external
resources to the rm, as claimed by stewardship
theory. The reason why independent directors
r 2010 British Academy of Management.

623

contribute so little, if anything, in very lowperforming enterprises may be that these rms do
not need good surveillance, but rather a mediator
to obtain external resources. Well-informed inside directors, being well aware of the internal
conditions of the rm, are better positioned to
help to make the right decisions as to the
appropriate business strategy (Baysinger and
Hoskisson, 1990). As rm performance increases,
though, the likelihood of CEO entrenchment
goes up: then, board independence is required to
discipline the CEO (Finkelstein and DAveni,
1994). Our result that board independence is
signicant in the 0.30.7 quantiles range supports
this entrenchment argument, so being in line with
agency theory (Fama, 1980; Fama and Jensen,
1983; Jensen and Meckling, 1976). Conversely, as
the rm moves into the top rank, board
independence will turn ineective, again. This
may be caused by the fact that the adoption of
these practices by excellent rms operates as a
signal that they behave well, which is in line with
the expectations of the outside world anyway
(Black, Jang and Kim, 2006).
On the other hand, the result that CEO duality
is not signicant to enforce the performance of
under-performing corporations may be because
such rms are so engaged with all kinds of serious
challenges that the marginal contribution of the
CEO duality governance mechanism is close to
zero. The nding that CEO duality does enhance
rm performance in mediocre enterprises oers
support for stewardship theory, which argues
that duality creates unambiguous leadership and
unity of command (Fayol, 1949; Massie, 1965).
The nding goes against popular beliefs in East
Asia that the role of the chair of the board should
be separated from the position of the CEO
(Claessens and Fan, 2002; Claessens, Djankov
and Lang, 2000). Perhaps, average-performing
rms face complicated operational and managerial issues that demand strong and solid leadership. For the top-performing enterprises,
insignicance of CEO duality may be due to the
fact that the likelihood of entrenchment is larger
in such rms (Finkelstein and DAveni, 1994).
Conversely, these rms need good surveillance
mechanisms by implementing CEO non-duality,
an argument in line with agency theory. Additionally, we found a negative moderating eect of
board size on the positive relationship between
CEO duality and rm performance. This implies

624
that the positive eect of CEO duality does
diminish as the number of board members
increases. The reason is that, as the number of
board members increases, decision-making processes, coordination and communication become
more problematic (Lin, 1996; Yermack, 1996),
which reduces the benets of CEO duality.
The ndings in this study have implications for
both theoretical understanding and corporate
governance practices. The result that the eectiveness of board characteristics depends on the
level of rm performance underscores the contingency theory of management. Contingency
theory proposes that the optimal structure of
any organization is dependent upon a wide
variety of contingencies (Donaldson, 2001; Drazin and Van de Ven, 1985; Pugh et al., 1969). Our
ndings suggest that the optimal design of board
independence and CEO duality is conditional on
the level of initial rm performance. This is a
result that is highly relevant from a regulatory
perspective as well. A single one size ts all
design of corporate governance, as reected in
many corporate governance codes, that is applied
to a wide variety of enterprises is likely to be
inappropriate, as a specic characteristic of a rm
may t with a dierent best practice than
suggested. This implies that regulators should be
careful not to design strict one size ts all
corporate governance rules, but should rather
specify conditional best practices. As an alternative, the often introduced comply or explain
clause may oer a second-best option.
Our study, as any, features a number of
limitations. A key one is the cross-section nature
of our data. This means that we cannot explore
causality and time lag issues. For instance, it may
be that the eects of an independent board
membership and CEO duality take several years
to materialize. Apart from collecting panel data,
this issue can perhaps be resolved by weighting
board independence and CEO duality by the
tenure of the (non)executive ocers involved.
Another important research opportunity is to
apply quantile regression widely in the corporate
governance research domain and organization
sciences at large, for that matter. This study
demonstrates that quantile regression can provide
additional insights in the understanding of the
relationship between board characteristics and
rm performance. We believe that the method can
be extensively utilized to evaluate corporate

D. Ramdani and A. van Witteloostuijn

governance eectiveness in promoting rm performance and to reveal unexplored relationships
between corporate governance variables and rm
performance. In addition, institutional contexts
may moderate the relationship between board
characteristics and rm performance, an issue
which we leave for future research.

References
Abrevaya, J. (2001). The eects of demographics and maternal
behavior on the distribution of birth outcomes, Empirical
Economics, 26, pp. 247257.
Agrawal, A. and C. R. Knoeber (1996). Firm performance and
mechanisms to control agency problems between managers
and shareholders, Journal of Financial and Quantitative
Analysis, 31, pp. 377397.
Aguilera, R. V., I. Filatotchev, H. Gospel and G. Jackson
(2008). An organizational approach to comparative corporate governance: costs, contingencies, and complementarities, Organization Science, 19, pp. 475492.
Alexander, J. A., M. L. Fennell and M. T. Halpern (1993).
Leadership instability in hospitals: the inuence of board
CEO relations and organization growth and decline,
Administrative Science Quarterly, 38, pp. 7499.
Al Farooque, O., T. van Zijl, K. Dunstan and A. K. M. W.
Karim (2007). Corporate governance in Bangladesh: link
between ownership and nancial performance, Corporate
Governance: An International Review, 15, pp. 14531468.
Arias, O., K. F. Hallock and W. S. Escudero (2001). Individual
heterogeneity in the returns to schooling: instrumental
variables quantile regression using twins data, Empirical
Economics, 26, pp. 740.
Baliga, B. R., R. C. Moyer and R. S. Rao (1996). CEO duality
and rm performance: whats the fuss?, Strategic Management Journal, 17, pp. 4153.
Barreto, R. A. and A. W. Hughes (2004). Under performers
and over achievers: a quantile regression analysis of growth,
Economic Record, 80, pp. 1735.
Bassett Jr, G. W. and H. L. Chen (2001). Portfolio style:
return-based attribution using quantile regression, Empirical
Economics, 26, pp. 293305.
Baysinger, B. D. and H. N. Butler (1985). Corporate
governance and the board of directors: performance eects
of changes in board composition, Journal of Law, Economics
and Organization, 1, pp. 101124.
Baysinger, B. D. and R. E. Hoskisson (1990). The composition of
boards of directors and strategic control: eects on corporate
strategy, Academy of Management Review, 15, pp. 7187.
Bhagat, S. and B. Black (2002). The non-correlation between
board independence and long-term rm performance,
Journal of Corporation Law, 27, pp. 231273.
Black, B. S., H. Jang and W. Kim (2006). Does corporate
governance predict rms market values? Evidence from
Korea, Journal of Law, Economics and Organization, 22,
pp. 366413.
Boyd, B. K. (1995). CEO duality and rm performance: a
contingency model, Strategic Management Journal, 16, pp.
301312.

r 2010 British Academy of Management.

Board Independence, CEO Duality and Firm Performance

Buchinsky, M. (1994). Changes in the U.S. wage structure
19631987: application of quantile regression, Econometrica,
62, pp. 405458.
Buchinsky, M. (2001). Quantile regression with sample
selection: estimating womens return to education in the
U.S, Empirical Economics, 26, pp. 87113.
Cade, B. S., J. W. Terrell and R. L. Schroeder (1999).
Estimating eects of limiting factors with regression
quantiles, Ecology, 80, pp. 311323.
Chaganti, R. S., V. Mahajan and S. Sharma (1985). Corporate
board size, composition and corporate failure in retailing
industry, Journal of Management Studies, 22, pp. 400417.
Cheung, Y.-L., P. R. Raub and A. Stouraitis (2006). Tunneling, propping, and expropriation: evidence from connected
party transactions in Hong Kong, Journal of Financial
Economics, 82, pp. 343386.
Cho, D. S. and J. Kim (2007). Outside directors, ownership
structure and rm protability in Korea, Corporate Governance: An International Review, 15, pp. 739752.
Claessens, S. and J. P. H. Fan (2002). Corporate governance in Asia:
a survey, International Review of Finance, 3, pp. 71103.
Claessens, S., S. Djankov and L. H. P. Lang (2000). The
separation of ownership and control in East Asian corporations, Journal of Financial Economics, 58, pp. 81112.
Claessens, S., S. Djankov, J. P. H. Fan and L. H. P. Lang
(1999). Expropriation of minority shareholders: evidence
from East Asia, World Bank Policy Research Working
Paper 2088.
Coles, J. L., N. D. Daniel and L. Naveen (2008). Boards: does one
size t all?, Journal of Financial Economics, 87, pp. 329356.
Cornett, M. M., A. J. Marcus and H. Tehranian (2008).
Corporate governance and pay-for-performance: the impact
of earnings management, Journal of Financial Economics, 87,
pp. 357373.
Cyert, R. M. and J. G. March (1963). A Behavioral Theory of
the Firm. Englewood Clis, NJ: Prentice Hall.
Daily, C. M. and D. R. Dalton (1992). The relationship
between governance structure and corporate performance in
entrepreneurial rms, Journal of Business Venturing, 7, pp.
375386.
Daily, C. M. and D. R. Dalton (1993). Board of directors
leadership and structure: control and performance implications, Entrepreneurship Theory and Practice, 17, pp. 6581.
Dalton, D.R, C. M. Daily, A. E. Ellstrand and J. L. Johnson
(1998). Meta-analytic reviews of board composition, leadership structure and nancial performance, Strategic Management Journal, 19, pp. 269290.
Davis, J. H., F. D. Schoorman and L. Donaldson (1997).
Toward a stewardship theory of management, Academy of
Management Review, 22, pp. 2047.
Demsetz, H. (1983). The structure of ownership and the theory
of the rm, Journal of Law and Economics, 26, pp. 373390.
Donaldson, L. (1990). The ethereal hand: organizational
economics and management theory, Academy of Management Review, 15, pp. 369381.
Donaldson, L. (2001). The Contingency Theory of Organizations. Thousand Oaks, CA: Sage.
Donaldson, L. and J. H. Davis (1991). Stewardship theory or
agency theory: CEO governance and shareholder returns,
Australian Journal of Management, 16, pp. 4964.
Donaldson, L. and J. H. Davis (1994). Boards and company
performance: research challenges the conventional wisdom,
Corporate Governance: An International Review, 2, pp. 151
160.

r 2010 British Academy of Management.

625

Drazin, R. and A. H. Van de Ven (1985). Alternative forms of

t in contingency theory, Administrative Science Quarterly,
30, pp. 514539.
Eide, E. and H. S. Mark (1998). The eect of school quality on
student performance: a quantile regression approach,
Economics Letters, 58, pp. 345350.
Eisenhardt, K. M. (1989). Agency theory: an assessment and
review, Academy Management Review, 14, pp. 5774.
Elsayed, K. (2007). Does CEO duality really aect corporate
performance?, Corporate Governance: An International Review, 15, pp. 12031214.
Ezzamel, M. A. and R. Watson (1993). Organizational form,
ownership structure, and corporate performance: a contextual empirical analysis of UK companies, British Journal of
Management, 4, pp. 161176.
Fama, E. (1980). Agency problems and the theory of the rm,
Journal of Political Economy, 88, pp. 288307.
Fama, E. F. and M. C. Jensen (1983). Separation of ownership
and control, Journal of Law and Economics, 26, pp. 301325.
Fayol, H. (1949). General and Industrial Management (trans. by
C. Stours). London: Pitman.
Finkelstein, S. and R. A. DAveni (1994). CEO duality as a
double-edged sword: how boards of directors balance
entrenchment avoidance and unity of command, Academy
of Management Journal, 37, pp. 10791108.
Garc a, J., P. J. Hernandez and A. L. Nicolas (2001). How wide is
the gap? An investigation of gender wage dierences using
quantile regression, Empirical Economics, 26, pp. 149167.
Ghosh, S. (2006). Do board characteristics aect corporate
performance? Firm-level evidence for India, Applied Economics Letters, 13, pp. 435443.
Goel, R. K. and R. Ram (2004). Quantile regression estimates
of cigarette demand elasticities for the United States, Journal
of Economics and Finance, 28, pp. 413421.
Goodstein, J., K. Gautam and W. Boeker (1994). The eects of
board size and diversity on strategic change, Strategic
Management Journal, 15, pp. 241250.
Hao, L. and D. Q. Naiman (2007). Quantile Regression.
London: Sage.
Harrison, J. R., D. L. Torres and S. Kukalis (1988). The
changing of the guard: turnover and structural change in the
top-management positions, Administrative Science Quarterly, 33, pp. 211232.
Hermalin, B. E. and M. S. Weisbach (2003). Boards of directors
as an endogenously determined institution: a survey
of the economic literature, Economic Policy Review, 9, pp.
726.
Jensen, M. C. (1993). The modern industrial revolution, exit,
and the failure of internal control systems, Journal of
Finance, 48, pp. 831880.
Jensen, M. C. and W. H. Meckling (1976). Theory of the rm:
managerial behavior, agency costs, and ownership structure,
Journal of Financial Economics, 3, pp. 305360.
Johnston, J. and J. DiNardo (1997). Econometrics Methods, 4th
edn. New York: McGraw-Hill.
Kesner, I. F. (1987). Directors stock ownership and organizational performance: an investigation of Fortune 500 companies, Journal of Management, 13, pp. 499508.
Kesner, I. F., B. Victor and B. Lamont (1986). Board
composition and the commission of illegal acts: an investigation of Fortune 500 companies, Academy of Management
Journal, 29, pp. 789799.
Kiel, G. C. and G. J. Nicholson (2003). Board composition and
corporate performance: how the Australian experience informs

626
contrasting theories of corporate governance, Corporate
Governance: An International Review, 11, pp. 189205.
Klein, A. (1998). Firm performance and board committee structure, Journal of Law and Economics, 41, pp.
275303.
Knight, C. A. and D. D. Ackerly (2002). Variation in nuclear
DNA content across environmental gradients: a quantile
regression analysis, Ecology Letters, 5, pp. 6676.
Koenker, R. and G. S. Bassett (1978). Regression quantiles,
Econometrica, 46, pp. 3350.
Koenker, R. and K. F. Hallock (2001). Quantile regression,
Journal of Economic Perspectives, 15, pp. 143156.
Lemmon, M. L. and K. V. Lins (2003). Ownership structure,
corporate governance and rm value: evidence from the
East Asian nancial crisis, Journal of Finance, 58, pp. 1445
1468.
Lin, L. (1996). The eectiveness of outside directors as a
corporate governance mechanism: theories and evidence,
Northwestern University Law Review, 90, pp. 898976.
Lipton, M. and J. W. Lorsch (1992). A modest proposal for improved corporate governance, Business Lawyer, 1, pp. 5977.
Machado, J. A. F. and J. Mata (2001). Earning functions in
Portugal 19821994: evidence from quantile regressions,
Empirical Economics, 26, pp. 115134.
Manning, W. G., L. Blumberg and L. H. Moulton (1995). The
demand for alcohol: the dierential response to price,
Journal of Health Economics, 14, pp. 123148.
Massie, J. L. (1965). Management theory. In J. G. March
(ed.), Handbook of Organizations, pp. 387422. Chicago, IL:
Rand-McNally.
Mintzberg, H. (1983). Power In and Around Organizations.
Englewood Clis, NJ: Prentice Hall.
Mitton, T. (2002). A cross-rm analysis of the impact of
corporate governance on the East Asian nancial crisis,
Journal of Financial Economics, 64, pp. 215241.
Mukherjee, C., H. White and M. Wuyts (1998). Econometrics and
Data Analysis for Developing Countries. New York: Routledge.
Nam, S. W. and I. C. Nam (2004). Corporate Governance in Asia:
Recent Evidence from Indonesia, Republic Korea, Malaysia and
Thailand. Tokyo: Asian Development Bank Institute.

D. Ramdani and A. van Witteloostuijn

Nielsen, H. S. and M. Rosholm (2001). The publicprivate
sector wage gap in Zambia in the 1990s: a quantile regression
approach, Empirical Economics, 26, pp. 169182.
Pearce, J. A. I. and S. A. Zahra (1992). Board compensation
from a strategic contingency perspective, Journal of Management Studies, 29, pp. 411438.
Pfeer, J. (1972). Size and composition of corporate boards of
directors: the organization and its environment, Administrative Science Quarterly, 17, pp. 218229.
Pfeer, J. (1973). Size, composition, and function of hospital
boards of directors: the organization and its environment,
Administrative Science Quarterly, 18, pp. 349364.
Pfeer, J. and G. R. Salancik (1978). The External Control of
Organizations: A Resource-dependence Perspective. New
York: Harper and Row.
Pi, L. and S. G. Timme (1993). Corporate control and bank
eciency, Journal of Banking and Finance, 17, pp. 515530.
Pugh, D. S., D. Hickson, C. R. Hinings and C. Turner (1969).
The context of organization structures, Administrative
Science Quarterly, 14, pp. 91114.
Rechner, L. and D. Dalton (1991). CEO duality and
organizational performance: a longitudinal study, Strategic
Management Journal, 12, pp. 155160.
Ribeiro, E. P. (2001). Asymmetric labor supply, Empirical
Economics, 26, pp. 183197.
Rosenstein, S. and J. G. Wyatt (1990). Outside directors, board
independence, and shareholder wealth, Journal of Financial
Economics, 26, pp. 175191.
Rousseeuw, R. J. and A. M. Leroy (1987). Robust Regression
and Outlier Detection. New York: Wiley.
Schellenger, M. H., D. D. Wood and A. Tashakori (1989).
Board of director composition, shareholder wealth, and
dividend policy, Journal of Management, 15, pp. 457467.
van Witteloostuijn, A. (1998). Bridging behavioral and
economic theories of decline: organizational inertia, strategic
competition, and chronic failure, Management Science, 44,
pp. 501519.
Yermack, D. (1996). Higher market valuation for rms with a
small board of directors, Journal of Financial Economic, 40,
pp. 185211.

Dendi Ramdani is a PhD student at the University of Antwerp, Belgium. He obtained his MPhil
(Research Master) in economics and econometrics from the University of Groningen, the
Netherlands, and his MSc in economics from the University of Indonesia, Indonesia. His PhD
research project focuses on the impact of corporate governance on rm performance in dierent
institutional contexts.
Arjen van Witteloostuijn is a Research Professor of economics and management at the University of
Antwerp, Belgium, and Professor of Institutional Economics at Utrecht University, the Netherlands.
He has published widely in such international journals as the Academy of Management Journal,
Academy of Management Review, American Sociological Review, Economica, Journal of Economic
Behavior and Organization, Journal of Economic Psychology, Journal of International Business
Studies, Journal of Management Studies, Management Science, Organization Studies, Organization
Science and Strategic Management Journal. His current research interests cross-border
macroeconomics, industrial organization, organizational ecology and social psychology.

r 2010 British Academy of Management.