Dey (2008)

DOI: 10.1111/j.1475-679X.2008.00301.
x
Journal of Accounting Research
Vol. 46 No. 5 December 2008
Printed in U.S.A.
Corporate Governance
and Agency Conicts
A I Y E S H A D E Y
Received 30 May 2006; accepted 3 March 2008

ABSTRACT
I investigate whether corporate governance is associated with the level of
agency conicts in rms. I employ exploratory principal components analysis
on 22 individual governance variables to obtain seven factors that represent
the different dimensions of governance for a rm. I measure the level of
agency conicts in rms based on seven proxies for agency conicts used
in the literature. I nd that rms with greater agency conicts have better
governance mechanisms inplace, particularly those related to the board, audit
committee, andauditor. I alsondthat the compositionandfunctioning of the
board, the independence of the auditor, and the equity-based compensation
of directors are signicantly associated with rm performance, but primarily
for rms with high agency conicts. Overall, the results support the theory
that the existence and role of various governance mechanisms in a rm are a
function of the level of agency conicts in the rm.
Graduate School of Business, University of Chicago. This paper is based on my dissertation

and I would like to thank my committee members Ravi Jagannathan, Thomas Z. Lys, Robert P.
Magee, and Beverly Walther for their valuable suggestions and guidance. I would also like to
thank an anonymous referee, Brian Bushee, Marcus Caylor, Daniel Cohen, Ellen Engel, Tom
Fields, Xiaohui Liu, N. V. Ramanan, Scott Richardson, Jonathan Rogers, Douglas Skinner (the
editor), Ewa Sletten, Abbie Smith, Suraj Srinivasan, Jayanthi Sunder, Shyam V. Sunder, and
the seminar participants at the 2005 AAA annual meeting, University of California at Berkeley,
University of Chicago, Cornell University, Emory University, Georgia State University, Harvard
Business School, University of Houston, University of Illinois, University of Michigan, MIT,
Northwestern University, University of Pennsylvania, University of Southern California, Uni-
versity of Texas at Dallas, Tuck School of Business at Dartmouth, and University of Washington
for very useful comments. I am very grateful to the Zell Center for Risk Research and the
Accounting Research Center at the Kellogg School of Management, Northwestern University
for nancial support. All errors are my own.
1143
Copyright
C
, University of Chicago on behalf of the Institute of Professional Accounting, 2008
1144 A. DEY
1. Introduction
In the wake of the recent scandals and the passage of the new regulation
in 2002, the Sarbanes Oxley Act (SOX), the link between corporate gov-
ernance and shareholder value has re-emerged as a topic of considerable
importance among academics, practitioners, and regulators. A large body
of research examines the association between various aspects of governance
and different features of organizational performance and decisions (Teoh
and Wong [1993], Vafeas [2000], Felo, Krishnamurthy, and Solieri [2003],
Bushman et al. [2004]). However, there is little evidence on the factors that
determine the endogenous presence or effectiveness of these mechanisms.
The relationship between shareholders and corporate managers is
fraught with conicting interests that arise due to the separation of own-
ership and control, divergent management and shareholder objectives, and
information asymmetry between managers and shareholders. Due to these
conicting interests (collectively referred to as agency conicts), managers
have the incentives and ability to maximize their own utility at the expense
of corporate shareholders. Contracts alone are not always enough to re-
solve these conicts (Hart [1995]). Consequently, the owners (and in some
situations the managers themselves) have reason to establish mechanisms
to monitor managerial activities and limit undesirable managerial behavior
(Jensen and Meckling [1976]). As a result, corporate governance structures
evolve that help in mitigating these agency conicts.
The magnitude of the agency conicts varies cross-sectionally across rms
depending on the ease with which managers can exercise their own pref-
erences as opposed to value maximization, the complexity of the rms
operating environment, the attractiveness of perquisites, etc. (Jensen and
Meckling [1976]). Because agency problems vary across rms, the gover-
nance structures required to address these problems are also likely to vary.
As a result, any associationbetweenvarious governance mechanisms andvar-
ious aspects of organizational performance is unlikely to be uniform across
all rms. Consequently, in order to perform more meaningful analyses on
the role of governance in affecting rm performance and other operating
decisions, we rst need to take a step back and examine how governance
mechanisms arise and vary across rms.
The purpose of this paper is to shed some light on this issue by investigat-
ing how governance structures vary cross-sectionally. Specically, I examine
whether the level of agency conicts in a rm is associated with its gover-
nance structure. Evidence on this relation will enhance our knowledge on
the role of governance in rms and whether improving various aspects of
governance will help in improving organizational performance for all rms
or whether such measures will be more effective for certain types of rms.
Corporate governance has multiple dimensions. However, prior research
on the effect of governance on various organizational outcomes typically fo-
cuses on individual governance mechanisms or constructs one-dimensional
GOVERNANCE AND AGENCY 1145
governance metrics by summing up individual variables.
1
Simply aggregat-
ing variables representing different aspects of boards of directors, auditors,
managers, etc., or focusing on some of these variables in isolation is not
likely to be an appropriate approach to measuring the corporate gover-
nance structure of a rm, and generally creates measurement error (Lar-
cker, Richardson, and Tuna [2007]). Further, several individual governance
variables are likely to be interrelated, and ignoring such correlations can
lead to spurious inferences (Agrawal and Knoeber [1996], Bowen, Rajgopal,
and Venkatachalam [2005]).
I consider a large set of individual governance variables and rely on a vari-
ety of traditional measures of governance used by academics and regulators
in order to derive the various dimensions of a rms governance structure. I
apply exploratory principal components analysis (PCA) to 22 individual gov-
ernance variables and obtain seven distinct governance factors representing
the composition and functioning of the board of directors, executive com-
pensation, equity-based compensation of directors, independence of the
auditor, structure and functioning of the audit committee, and the boards
control over nancial reporting quality. I use these factors as representa-
tions of the different underlying dimensions of governance. Given the lack
of a theory on governance structures, this technique is more appropriate
because the process identies various factor structures with the individual
governance indicators that are highly correlated. These factors can then be
used to explain the underlying dimensions of governance. This method is
also employed in a recent study by Larcker, Richardson, and Tuna [2007]
to measure governance. Larcker, Richardson, and Tuna [2007] consider 39
individual governance indicators and use PCA to form 14 governance fac-
tors. They document that these governance indices have mixed association
with abnormal accruals, little relation to accounting restatements, and some
ability to explain future operating performance and future excess stock
returns.
Self-interested managerial behavior resulting from agency conicts can
comprise a range of activities that are not optimal for shareholders. Some
examples of these include empire building, the consumption of corporate
resources as perquisites, the avoidance of optimal risk investments, and ma-
nipulating nancial gures to optimize compensation. I thus consider var-
ious rm-specic attributes that are indicative of the existence of agency
conicts. Specically I include rm size, organizational complexity, own-
ership structure, growth, leverage, operating risk, and free cash ows as
1
Some examples include Wareld, Wild, and Wild [1995] ongreater managerial ownership,
Frankel, Johnson, and Nelson [2002] on the mix between consulting and audit fees paid to
auditors, and Gompers, Ishii, and Metrick [2003] and Brown and Caylor [2004] who consider
several governance variables to construct one-dimensional governance scores. While all the
above studies provide interesting evidence, a more meaningful analysis can be conducted
by considering a broad set of governance variables and the interaction among them while
determining the governance structure of a rm.
1146 A. DEY
measures of the level of agency conicts in rms. I nd evidence consistent
with the theory that governance structures, particularly those related to the
board of directors, the audit committee, and the auditor, are positively re-
lated to the level of agency conicts in rms. In particular, I nd that larger
rms, rms with more diffuse ownership structures, rms with higher lever-
age, and rms withmore operating risk have better governance mechanisms
in place.
Given this result, a natural hypothesis that emerges is that the relation
between overall rm performance and governance is also unlikely to be
uniform across rms. Failing to control for this cross-sectional variation is
likely to be one reason for the mixed evidence on the relation between
overall rm performance and governance in prior studies (Hermalin and
Weisbach [1991], Mehran [1995], Bhagat and Black [2001], among oth-
ers). I examine whether the relation between the governance structure and
overall rm performance varies with the level of agency conicts, and nd
results consistent with this claim. This supports the conclusion that gov-
ernance structures arise endogenously to rms competitive and business
environments, and as a result the role of governance in affecting various as-
pects of organizational performance and decisions is likely to differ across
rms.
My primary contribution is to document factors that are associated with
the determination of governance structures in rms, and to provide evi-
dence that the role of governance is not uniform across rms. I use a large
set of governance and agency variables to showthat rms with higher agency
conicts have stronger governance structures inplace. Consequently, the re-
lation between governance and rm performance is also a function of the
level of agency conicts. These results are consistent with a fundamental hy-
pothesis: The demand for higher quality governance is greater in rms with
greater need for oversight. The evidence supports claims that a uniform set
of governance rules is unlikely to be efcient for all rms (Klein [2002a],
Romano [2005], Ribstein [2005]).
The remainder of the paper is divided into the following sections. Sec-
tion 2 discusses the theoretical background and the research objective, and
provides a brief overview of related research, followed by section 3 which
discusses the research design. Section 4 describes the tests and the results,
section 5 discusses the analyses on the relation between governance and
rm performance, and section 6 concludes.
2. Theoretical Background, Research Objective, and Prior Research
Agency conicts in organizations result from the separation of owner-
ship and control, the conicting objectives of owners and managers, and
the information asymmetry between owners and managers (Coase [1937],
Jensen and Meckling [1976], Fama and Jensen [1983a, b]). As a result of
these agency conicts, and given that managers have sufcient latitude in
applying accepted accounting procedures, they are likely to have incentives
to take actions that maximize their utility, even when those actions do not
maximize shareholder wealth (Watts and Zimmerman [1986]).
The governance structure of a rm involves mechanisms to minimize
these agency conicts. Ceteris paribus, the demand for these control mech-
anisms is likely to be higher for rms with greater need for oversight, or
higher degrees of agency conicts. In other words, agency conicts and
governance mechanisms in a rm are likely to be complementaryhigher
levels of agency conicts will result in stronger governance structures. The
purpose of this paper is to test the hypothesis that the governance struc-
tures in a rm are a function of the agency conicts in the rm. Specically,
I test the theory that rms with higher levels of agency conicts have more
efcient governance mechanisms in place.
Prior research related in spirit to this study includes Lasfer [2002], who
examines the association between board structure and rm value in the
United Kingdom and nds that this relation is a function of a rms growth
opportunities. He nds that while low growth rms are less likely to have an
independent board, their value is positively related to these board structure
variables. In contrast, for high growth rms, the relationship between board
structure and rm value is weak, suggesting that board structure does not
always mitigate agency conicts for these rms. He concludes that imposing
the same board structures for all rms is likely to reduce the value of rms
that are forced to depart from their optimal board structures. Klapper and
Love [2004] and Durnev and Kim [2003] examine the association between
the Credit Lyonnais Securities Asia (CLSA) index of corporate governance
and rm-level characteristics in emerging markets. These studies document
that rms with greater past growth, lower proportions of xed assets, shares
traded in the United States, better investment opportunities, higher owner-
ship concentration, and greater needs for external nancing have higher
governance rankings.
Related studies on U.S. rms include Gillan, Hartzell, and Starks [2003],
who examine the extent to which industry characteristics are associated
with rms governance structures, namely, board structure, charter provi-
sions (i.e., antitakeover devices), andstate of incorporation. They document
that industry factors contribute most of the explainable variation in their
governance indices, and dominate rmfactors. Smith and Watts [1992] also
argue that industry factors and the investment opportunity set determine
rms governance policies. Hermalin and Weisbach [1988, 2003] suggest
that boards of directors are an endogenous response to agency problems.
Bushman et al. [2004] document that board structures, director equity in-
centives, executive compensation, and ownership concentrations vary with
rms accounting systems and organizational complexity.
Klein [2002a] tests for the economic determinants of audit committee in-
dependence, and documents that audit committee independence increases
with board size and the percentage of outsiders on the board. Klein [2002a]
also nds that audit committee independence decreases with the rms
growth opportunities, when the rm experiences two or more consecu-
tive losses and in the presence of alternate monitoring mechanisms. In the
context of audit committees, her ndings also support the argument that
one size doesnt t all and suggest that it is desirable to allow rms some
exibility in deciding on the composition of their audit committees. These
1148 A. DEY
studies provide motivation for further research on how governance mecha-
nisms vary across rms, and how this affects the role of governance in rm
performance and other rm decisions. This study contributes to the evi-
dence in the prior literature by documenting that rms with higher agency
conicts have better governance mechanisms in place, and the relation be-
tween governance mechanisms and overall rm performance also varies
with the level of agency conicts.
One recent study that is relatedtothis paper, particularly fromthe point of
view of the measurement of governance dimensions, is by Larcker, Richard-
son, and Tuna [2007] (henceforth, LRT). LRT develop reliable indices of
various dimensions of governance using a large number of individual gov-
ernance indicators. They argue that using a single indicator or summing up
several indicators to form a one-dimensional index to measure a multidi-
mensional construct such as governance is likely to introduce considerable
measurement error in the analyses. They consider 39 individual governance
indicators andusePCAtoform14governancefactors. I alsoemploy thesame
method, PCA, to construct my governance dimensions. I consider 22 indi-
vidual governance variables and obtain seven factors that represent various
dimensions of governance.
A brief discussion of the main differences in the individual governance
variables in the two studies is as follows. LRT consider stock ownership by
blockholders and activist institutions, debt-to-equity, and preferred stock-to-
equity as governance indicators. I focus oninternal governance mechanisms
in my study and do not include these variables as governance indicators. I
consider institutional ownership and leverage as inputs in determining the
level of agency conicts in the rm.
LRT also include antitakeover devices as additional governance indica-
tors. Although I do not include antitakeover provisions in my main tests, in
additional analyses, I consider the 28 state, federal, and rm-specic pro-
visions included in the G-score computed by Gompers, Ishii, and Metrick
[2003].
2
For these antitakeover variables, the PCA mechanism reveals an
additional ve factors representing the protection of ofcers and directors
from lawsuits, delay mechanisms management can use to prevent hostile
takeovers, protection available to management during a change in control,
protective mechanisms available to management during business combi-
nations, and other state laws. The results indicate that two of the factors,
namely, the protection of directors and ofcers from legal expenses and
protective mechanisms during a change in control, are signicantly asso-
ciated with the level of agency conicts.
3
Finally, I also include various at-
tributes of the auditor among my individual governance variables, and LRT
2
I remove these variables as measures of governance from my main analyses based on the
referees advice that these variables are likely to measure takeover defenses rather than the
internal governance of a rm.
3
These results are not tabulated but are available on request.
include some additional board-related variables, such as the existence of a
lead director and the fraction of busy directors.
LRT focus on examining the association between their indices of gov-
ernance and various accounting outcomes, namely, abnormal accruals, ac-
counting restatements, and future operating performance. They nd that
these governance indices exhibit mixed association with abnormal accruals
and have little relation to restatements, but some ability to explain future
operating performance. In contrast, my primary focus is in understanding
how governance structures vary across forms. Specically, I hypothesize and
test whether differences in governance structures are related to differences
in levels of agency conicts across rms. Evidence on whether different
dimensions of governance vary with agency conicts will provide an em-
pirical basis for the theoretical predictions that governance structures arise
endogenously in response to specic problems faced by rms.
3. Research Design
3.1 DATA
The analysis includes the years 2000 and 2001. I begin from year 2000 be-
cause data on audit fees are unavailable prior to this period. I hand-collect
data on boards and auditors from proxy statements (schedule 14A). Sched-
ule 14A requires rms to disclose all existing and nominated directors
names, ages, all family relationships between directors, nominees, or execu-
tive ofcers, any signicant current or proposed transactions with manage-
ment, any signicant business relationships with the rm, and the number
of shares held by the directors. Schedule 14A also requires rms to state
whether they have standing audit, nominating, compensation, and gover-
nance committees. If such committees exist, rms are required to disclose
their functions andresponsibilities, their members, andthe number of times
the committees met during the last scal year. Details regarding the audit
rm and audit and non-audit fees are also provided. Schedule 14A is typi-
cally led by rms in March of every year and reports on the preceding year
(it also provides the names of the directors nominated and elected for the
following year). Accordingly, I code the information released by a rm in
March of year t+1 as the governance characteristics of the rm for year t.
I obtain the rest of the data fromCompustat, Spectrum, and ExecuComp.
The initial sample consists of 1,887 rms with data available in the proxy
statements. Merging these rms with Spectrum and ExecuComp reduces
the sample to 493 rms. After merging with Compustat, the nal sample
comprises 371 rms.
4
4
My sample only covers two years and has relatively large rms. This limits my ability to
generalize the results. However, given that there is considerable cross-sectional variation in the
governance measures and the agency measures during the period I examine, the statistical
analyses are likely to have sufcient power to detect the association between measures of
corporate governance and agency conicts.
1150 A. DEY
3.2 EMPIRICAL METHODOLOGY AND CONSTRUCTS
3.2.1. Corporate Governance. I consider 22 governance variables represent-
ing various features of the board of directors, auditors, and executives. Ta-
ble 1 summarizes the descriptions and measurements of these variables.
5
For the board of directors, I consider the following variables representing
different aspects of the whole board as well as of the audit committee in
particular. I measure the composition of the board as the percentage of
outside directors on the board (%OUTDIR). Inside directors are directors
who are also employees of the rm, and outside directors are nonexecutive
directors, who are considered to be independent from management and
free from any business or other relationship that could materially interfere
with the exercise of their independent judgment. Typically a board with
more outside directors is considered to be more effective in monitoring
management.
The next variable, INDP COMMITTEE, measures the existence of inde-
pendent board committees, namely, the audit, compensation, nominating,
and governance committees. This variable takes the value of two if the board
has separate audit, compensation, nominating, andgovernance committees,
one if either the nominating or governance committee is missing, and zero
if both the nominating and governance committees are missing. Typically
all boards have standing audit and compensation committees, although not
all boards have nominating and governance committees. It has been argued
that nominating committees can improve the quality of appointments. They
are likely to increase the independence of the board, in part by reducing the
control exercised by the CEO in appointments, and result in nominations
more consistent with shareholder interests. The primary tasks performed by
governance committees are to identify and recommend to the board appro-
priate candidates who could serve as director nominees for the next annual
meeting of shareholders; to advise the board with respect to the board com-
position, procedures, and committees; and to develop and recommend to
the CEO and the board a set of corporate governance guidelines applicable
to the company and monitor such governance guidelines. When the nom-
inating and governance committees are absent, these functions are either
performed by the whole board or not performed at all (particularly in the
case of the governance functions). The performance of nominating func-
tions by the whole board, particularly if the CEO and other executives are
members of the board (which is usually the case), could affect the quality
of board appointments.
The variable CEO COMMITTEE is set to zero if the CEO is a member of
the nominating committee and/or the compensation committee, and one
5
Although the inclusion of governance variables and the coding procedure in constructing
the governance indexare subjective toanextent, they are guidedby the dictates of the corporate
governance norms specied in SOX and other Securities and Exchange Commission (SEC)
rulings. Also, I exclude certain governance criteria specied in SOX due to the lack of such
activity and/or disclosure prior to SOX (such as the requirement of auditor rotation).
otherwise. If the CEO is a member of the nominating committee, then that
means that the CEO has control over the selection of the other members
of the board. As a result, the CEO might choose more sympathetic board
members if he has too much inuence on the nominating committee. The
membership of the CEO in the compensation committee indicates that
the CEO makes his own compensation decisions. This may compromise
the incentives and the actions of the CEO. Several studies examine the
separation of the CEO and chairman of the board (COB) positions positing
that agency problems are higher when the same person holds both positions
and documenting evidence consistent with this theory (Fama and Jensen
[1983a], Berle and Means [1933], Yermack [1996]). The CEOs role as the
chairman of the board of directors implies that the CEO has the nal word
in many of the decisions made by the board. Moreover, to the extent that the
other members take decisions that do not antagonize the chair, the role of
the CEO as the COB compromises the independence of the board. If these
roles of the CEOandthe COBare combined, it is necessary to publicly justify
this combination. To capture the role of the CEO as the COB, I include the
variable CEO COB, which is set to zero if the CEO is also the chairman of
the board of directors, and one otherwise.
The variable INDP AUDITCOM takes the value of one if all audit commit-
tee directors are independent, and zero otherwise. The audit committee
is primarily responsible for overseeing the nancial reporting process on
behalf of the board of directors, reviewing the nancial disclosures, and
meeting privately, outside the presence of management, with the rms au-
ditors to discuss the internal accounting control policies and procedures.
Thus, the independent assessment of the audit committee is crucial for the
effective monitoring of a rms nancial reporting process, and there is ev-
idence of a negative relation between audit committee independence and
earnings management (Klein [2002b]). The existence of a nancially liter-
ate member on the audit committee facilitates the proper oversight of the
nancial reporting process. It also ensures proper review and oversight of
the audit functions, rather thancomplete reliance onthe auditors represen-
tation of the rms policies. I code the variable FIN EXPERT as being equal
to one if there exists at least one nancial expert in the audit committee,
and zero otherwise.
6
Nonexecutive directors may have family and/or business relationships
with the company. They may also be associated with not-for-prot orga-
nizations that receive support from the company. Such relations with the
6
The denition of nancial expert I use is as per SOX: A nancial expert is any member
who has the education or experience of a public accountant, auditor, principal nancial ofcer,
comptroller, or principal accounting ofcer of an issuer, or has been in a position requiring the
understanding of generally accepted accounting principles and nancial statements; experi-
ence in the preparation and auditing of nancial statements of comparable issuers; experience
in the application of such principles in connection with the accounting for estimates, accruals,
and reserves; experience with internal accounting controls; and an understanding of audit
committee functions.
1152 A. DEY
company may provide incentives to these directors to endorse certain de-
cisions that they otherwise may object to. Thus, I include the variable
%NONAFFIL OUTDIR as the percentage of outside directors that do not
have any relationship with the rm (family, business, not-for-prot organi-
zations that receive support from the rm). The number of meetings held
during the year provides one indication of how effectively the board func-
tions to the extent that this represents the regular attendance by the board
to rm issues. I include the variables #BD MEET, which is the number of
meetings held by the board during the year, and #AUDIT MEET, which is
the number of meetings held by the audit committee during the year, to
capture this aspect of the functioning of the board.
Smaller boardsizes are consideredtobe more effective inattaining higher
monitoring. Smaller boards have lesser disagreements among board mem-
bers, and are likely to be more efcient and organized in carrying out board
functions, than larger boards (Lipton and Lorsch [1992], Jensen [1993],
Yermack [1996]). To capture the effectiveness of the board based on its
size, I include the variable BD SIZE, which is dened as the negative of the
number of members on the board.
7
Drawing on prior studies, I include two variables to capture the age of
the board members: DIR AGE, which is the negative of the average age of
the directors, and DIR %UNDER70, which is the percentage of directors
under 70 years (LRT).
8
The presence of some form of evaluation of board
performance is likely to enhance the quality of board functioning. In most
rms, boards undertake self-evaluations on an annual basis, although in
some cases the audit committee is delegated the responsibility for evaluating
the performance of the board. I set the variable BD EVAL to take the value
one if there exists some form of evaluations for the performance of the
board, and zero otherwise.
Bhagat, Carey, and Elson [1999] document that directors with substantial
stock ownership act more quickly to replace the CEO. Bhagat and Black
[2001] also nd evidence consistent with independent directors being more
effective if motivated by signicant stock ownership. To measure the equity
ownership of directors, I include the following stock-based compensation
variables: DIR STKCOMP, which is the number of shares of stock (including
restricted stock) received by a nonemployee director divided by the total
number of shares outstanding of the rm, and DIR OPTION, which is the
7
There is some evidence in the literature that board effectiveness declines as board size
increases above a moderate number, an optimal number being about seven to nine directors
(Yermack [1996], Bhagat and Black [1999]). Given this prior evidence, I repeat my analyses
by redening BD SIZE as a variable that takes the value one if the board has between four
and nine members, and zero otherwise. The board size in my sample of rms ranges from
4 to 20 directors, with the mean (median) number being 8.85 (9). Thus, the coding of the
dummy variable B SIZE translates to being equal to one if the number is less than nine and
zero otherwise. However, this does not materially alter any of the results.
8
In companies where a retirement age of a director is specied, this variable equals the
percentage of directors who are under this retirement age.
number of options received by a nonemployee director divided by the total
number of shares outstanding of the rm.
Managerial ownership is an important mechanism to align managers in-
centives to shareholders (Jensen and Meckling [1976], Morck, Shleifer, and
Vishny [1988]). One way to increase ownership is through stock-based com-
pensation. As in Cheng and Wareld [2005], I include the following owner-
shipvariables of the topve executives: the average value of stock awards and
restricted stock grants divided by the total compensation, EXEC STKCOMP,
and the average value of option grants divided by the total compensation,
EXEC OPTION. These variables measure the proportion of the executives
compensations that is sensitive to stock prices. I also include the bonus com-
pensation of the top ve executives of the rm, EXEC BONUS, dened as
the average ratio of bonuses to total compensation.
Provision of nonaudit related services may seriously compromise the in-
dependence of the rms auditor. The empirical evidence on this relation,
however, is mixed(Frankel, Johnson, andNelson[2002], Ashbaugh, Lafond,
and Mayhew [2003], Larcker and Richardson [2004]). The non-audit ser-
vices currently prohibited include: bookkeeping or other services related
to accounting records or nancial statements of the client; nancial infor-
mation systems design and implementation; appraisal or valuation services,
fairness opinions, or contribution-in-kind reports; actuarial services; inter-
nal audit outsourcing services; management functions or human resources;
broker or dealer investment advisor, or investment banking services; and
legal services and expert services unrelated to the audit. I use the follow-
ing variables to represent the independence and functioning efciency of
the rms auditor. The variable NO NONAUDIT takes the value one if the
auditor does not provide any of the non-audit services that are prohibited
under SOX, and zero otherwise. The auditors independence may be fur-
ther compromised if, in addition to the provision of non-audit services, the
non-audit fees exceed the audit fees received by the auditor for the year. I
thus include the variable %AUDIT FEES, which is the ratio of audit fees to
non-audit fees paid to the auditor for the year.
The SEC has declared it unlawful for a company to employ an audit rm
when the CEO and/or CFO and/or controller or an equivalent person was
an employee of that audit rm in the recent past (in the preceding year)
and participated in the audit of the company. The previous employment
of a director (particularly if the director is in the audit committee) by the
companys audit rm can also compromise the audit quality. Recent studies
indicate a threat to audit quality if ofcers previously worked for their com-
panies audit rms (Menon and Williams [2004], Lennox [2005]). There
is also evidence suggesting that companies are more likely to appoint an
audit rm if the company has an ofcer who was an alumnus of that rm;
however, the presence of an independent audit committee reduces the in-
cidence of such ofcerauditor afliations (Lennox and Park [2007]). To
capture this conict of interest, I set the variable EXEC NONAUDITOR to
one if the CEO/CFO/any other top management personnel/any director
1154 A. DEY
TA B L E 1
Individual Governance Variables
Area Variable Description (Name) Measurement
Board of
directors
Percentage of the board
comprising outside
directors (%OUTDIR)
This variable equals the percentage of outside
directors on the board
Existence of independent
audit, compensation,
nominating, and
governance committees
(INDP COMMITTEE)
This variable takes the value +2 if the board has
separate audit, compensation, nominating,
and governance committees; +1 if either the
nominating or governance committees are
missing; and 0 if both the nominating and
governance committees are missing
CEO membership on
nominating and/or
compensation committee
(CEO COMMITTEE)
This variable takes the value +1 if the CEO is not
a member of the nominating committee nor
the compensation committee, and 0
otherwise
The combination of the roles
of the CEO and the
chairman of the board
(CEO COB)
This variable takes the value +1 if the CEO is not
the chairman of the board of directors, and 0
otherwise
Independence of directors in
the audit committee (as
per NYSE/NASDAQ/SEC
rules) (INDP AUDITCOM)
This variable takes the value +1 if all audit
committee directors are independent, and 0
if one or more of them are not independent
Existence of a nancial expert
in the audit committee
(FIN EXPERT)
This variable equals +1 if there exists at least one
nancial expert in the audit committee, and
0 otherwise; a nancial expert is dened as
any member who is nancially literate as per
the denition in SOX
Outside directors with no
personal or business
relationship/family
relationship/relationship
with not-for-prot
organizations that receive
support from the
corporation
(%NONAFFIL OUTDIR)
This variable is the percentage of outside
directors that do not have any family
relationship/business relationship/
relationship with not-for-prot organizations
that receive support from the rm
Frequency of board meetings
(#BD MEET)
This variable equals the number of meetings
held by the board during the year
Frequency of audit committee
meetings (#AUDIT MEET)
This variable equals the number of meetings
held by the audit committee during the year
Size of the board (BD SIZE) This variable equals the negative of the number
of members in the board
Average age of directors
(DIR AGE)
This variable is the negative of the average age of
the directors on the board
The presence of directors
under 70 years of age
(DIR %UNDER70)
This variable equals the percentage of directors
who are under 70 years old; in companies
where a retirement age of a director is
specied, this variable equals the percentage
of directors who are under this retirement
age
Existence of board member
evaluations (BD EVAL)
This variable takes the value of +1 if there exists
some form of evaluations for the
performance of the board, and 0 otherwise
(Continued)
TA B L E 1 Continued
Area Variable Description (Name) Measurement
Director
ownership
Stock compensation for
directors (DIR STKCOMP)
This variable is the number of shares of stock
(including restricted stock) received by
nonemployee directors divided by the total
number of shares outstanding of the rm
Option-based compensation
for directors
(DIR OPTION)
This variable is the number of options received
by nonemployee directors divided by the total
number of shares outstanding of the rm
Executive
compensa-
tion
Stock awards (including
restricted stock) as a
proportion of total
compensation for the top
managers of the rm
(EXEC STKCOMP)
This variable equals the mean of the value of
stock awards and restricted stock grants
divided by the total compensation for the top
ve executives of the rm
The average value of options
grants as a proportion of
total compensation for the
top managers
(EXEC OPTION)
This variable equals the mean of the value of the
options granted divided by the total
compensation for the top ve executives of
the rm
The average value of bonus
compensation as a
proportion of total
compensation for the top
managers of the rm
(EXEC BONUS)
This variable is the mean of the bonus
compensation divided by the total
compensation for the top ve executives of
the rm
Auditor Non-audit services provided by
auditor (NO NONAUDIT)
This variable takes the value +1 if the auditor
does not provide any of the non-audit
services that are prohibited under SOX, and
0 otherwise
Proportion of fees received by
auditor from audit services
(%AUDIT FEES)
This is the ratio of the audit fees to the non-audit
fees paid to the auditor for the year; the
items categorized as non-audit services
include those that are prohibited by SOX
Whether the CEO, CFO,
controller, or equivalent
person was employed by
auditor and participated in
the audit of the issuer in
one year preceding
initiation of audit
(EXEC NONAUDITOR)
This variable takes the value +1 if the
CEO/CFO/any other top management
personnel/any director was not employed by
the auditor and participated in the audit of
the issuer in one year preceding initiation of
audit, and 0 otherwise
If the auditor belongs to a
Big-5 audit rm (BIG5)
This variable equals 1 if the rm is audited by a
Big-5 audit rm, and 0 otherwise
This table summarizes the descriptions and measurements of the 22 individual corporate governance variables
used in this study. These individual variables are used in the principal components analysis procedure to obtain
governance factors.
Prior to SOX, companies were not required to disclose information regarding the prior employment of an
executive in the companys audit rm(EXEC NONAUDITOR) and the presence of a nancial expert (FIN EXPERT).
Thus, if a rm does not disclose anything regarding these two items, then I assume that there are no executives who
were employees of the companys audit rm (i.e., the variable EXEC NONAUDITOR is coded as 1) and the company
has no nancial expert in the audit committee (i.e., the variable FIN EXPERT is coded as 0).
was not employed by the auditor and participated in the audit of the issuer
in one year preceding the initiation of the audit, and to zero otherwise.
Finally, the reputationof anaudit rmcould affect the quality of its audits.
Although it is not clear whether the quality of audits of Big-5 auditors is
1156 A. DEY
always superior, the brand value of the audit rm could be associated with
the monitoring effectiveness of a rm. I set the variable BIG5 to equal one
if the rm is audited by a Big-5 audit rm, and zero otherwise.
In the above variable construction process, other than the equity
ownership and compensation variables for directors and executives
(DIR STKCOMP, DIR OPTION, EXEC STKCOMP, EXEC OPTION, and
EXEC BONUS), higher values of the variables indicate better governance.
Predictions for the ownership and compensation variables are difcult, be-
cause any interpretation depends on whether the existing levels are optimal,
too high, or too low.
9
Table 2 presents descriptive statistics of these individual governance vari-
ables for the sample rms. The median rm has a majority of outside
directors on the board (the variable%OUTDIR has a mean of 74% and a
median of 78%) and has at least one of the nominating or the governance
committees missing. The CEO of the median rm is not on the nominat-
ing or the compensation committees nor is the COB. The median rm
has a fully independent audit committee but does not have any nancial
experts (as dened by SOX) in the committee. Most of the nonexecutive
directors have no family or business relationship with the rm (the variable
%NONAFFIL OUTDIR has a mean of 91%and a median of 92%). The mean
(median) board meets seven(six) times a year and the mean(median) audit
committee meets ve (ve) times a year. Both the mean and median board
has about nine members, and most boards do not have any formal evalua-
tions. The mean and median age of a director is 59 years and about 90% of
directors are under 70 years or the specied retirement age. Thus, from the
point of view of the composition and functioning of the board, the average
sample rm appears to have reasonably good governance as measured by
these traditional governance variables.
The average outside director owns 0.01% of the rm, when considering
both stock-based and option-based compensation. However, a signicant
percentage of the compensation of the top ve executives is comprised of
equity-based compensation. On average, the top ve executives receive 47%
of their compensation in stock awards (including restricted stock) and 49%
of their compensation in option grants. Bonus compensation comprises
13% of the total compensation package.
The median rm in the sample hires its audit rm to provide the var-
ious non-audit services prohibited by SOX. The median of the variable
%AUDIT FEES is less than the mean, indicating positive skewness in the
9
The topic of the level of compensation is a controversial one, with two opposing main
theories regarding the efciency of contracts. One view asserts that compensation contracts
are on average efcient (e.g., Lazear [1995]), and the other view contends that compensation
levels are either too high or too low (e.g., Morck, Shleifer, and Vishny [1988]). If existing
levels are on average efcient, then any change in those levels lowers the quality of governance.
Onthe other hand, if existing levels of compensationare inefcient, thenwhether higher levels
of compensation improve the quality of governance depends on whether the existing levels
are too high or too low.
TA B L E 2
Governance Variables: Descriptive Statistics
Variable Mean Median Std. Deviation
%OUTDIR 0.7401 0.7778 0.1421
INDP COMMITTEE 1.0768 1.0000 0.8138
CEO COMMITTEE 0.5722 1.0000 0.4951
CEO COB 0.3222 1.0000 0.4677
INDP AUDITCOM 0.9518 1.0000 0.4602
FIN EXPERT 0.2244 0.0000 0.4175
%NONAFFIL OUTDIR 0.9081 0.9177 0.1406
#BD MEET 7.2891 6.0000 3.5349
#AUDIT MEET 5.1144 5.0000 2.3442
BD SIZE 8.8491 9.0000 0.4572
DIR AGE 58.6585 59.0000 4.3185
DIR %UNDER70 0.8928 0.9091 0.1318
BD EVAL 0.2665 0.0000 0.4425
DIR STKCOMP 0.00006 0.0000 0.00002
DIR OPTION 0.0001 0.00006 0.0002
EXEC STOCK 0.4678 0.3522 0.3188
EXEC OPTION 0.4906 0.4973 0.3972
EXEC BONUS 0.1317 0.1119 0.1190
NO NONAUDIT 0.0150 0.0000 0.1219
%AUDIT FEES 1.2976 0.5521 2.7652
EXEC NONAUDITOR 0.9849 1.0000 0.1219
BIG5 0.9714 1.0000 0.1668
This table reports the descriptive statistics of the 22 individual corporate governance variables that are
used in the principal components analysis procedure to obtain governance factors.
The variables BD SIZE and DIR AGE are used with a negative sign in the principal components anal-
ysis but the descriptive statistics are presented without the negative sign for more meaningful interpretations.
proportion of fees received by the auditor from audit and non-audit ser-
vices. While the mean of %AUDIT FEES indicates that the proportion of
fees from audit services is approximately 30% higher than that from non-
audit services, the median value indicates that the audit fees are about 60%
of the fees received from non-audit services. Most rms in the sample do
not have any executives who were employed by the auditor. Finally, most
rms employ a Big-5 auditor, which is not very surprising given the sample
comprises relatively large rms.
3.2.1.1. The Governance Factors. I use PCA to form factors that capture
different dimensions of governance and determine which of the above gov-
ernance indicators are associated with each factor. In this procedure the
individual variables are reduced into a smaller number of principal compo-
nents (or articial variables) that account for most of the variance in the
observed variables. I use a combination of the eigenvalue method and the
scree test inorder to determine the number of factors to retain. Inthe eigen-
value method, all factors with an eigenvalue greater than unity are retained.
Inthe scree test, the eigenvalues associatedwitheachcomponent are plotted
and breaks between the components with relatively large eigenvalues and
those with small eigenvalues are identied. The components that appear be-
fore the break are assumed to be meaningful and are retained for rotation,
1158 A. DEY
and those appearing after the break are assumed to be unimportant and
are not retained (Jolliffe [2002]).
This procedure results in seven factors that capture 62.4% of the total
variance of the original data. These factors are rotated using an orthogonal
rotation in order to better interpret the components. I perform an orthog-
onal rotation instead of an oblique rotation, since the former is easier to
interpret in this case. In particular, since these governance components are
usedina multiple regression, anorthogonal rotation, inwhichthe factors re-
main uncorrelated, avoids the complication of multicollinearity. Moreover,
on examining the factor intercorrelations after performing an oblique ro-
tation, I nd that most of the factors are pairwise uncorrelated (0.3 or lower
in magnitude), which indicates that an orthogonal rotation over an oblique
one does not compromise the relation between the variables.
For eachgovernance factor, I examine whichof the individual governance
variables have high factor loadings (bivariate correlations between the vari-
ables and the components). As is common in the PCAprocedure, I consider
a factor loading to be large if its absolute value exceeds 0.40. The retained
factors with the variables that are substantially associated with each factor
and the corresponding factor loadings are summarized in table 3.
Eachfactor represents anunderlying governance dimension, andthe vari-
ables that loadmeasure the same underlying construct. The variables that do
not load in any factor are measuring different constructs, and are unrelated
to the governance dimension measured by this factor. Consequently the
seven factors obtained measure seven distinct dimensions of governance. I
assign a name to each governance factor based on the individual variables
that load for an easier read and more meaningful interpretation of results.
The rst two factors include variables related to the composition and func-
tioning of the board, and I call them Board I and Board II. The next factor,
Exec Comp, appears to capture the bonus and equity-based compensation
of the top ve executives of a rm. The factor Dir Comp appears to cap-
ture the equity-based compensation provided to directors. The next factor,
Auditor, appears to capture the independence of the auditor, and the fac-
tor Audit Comm appears to capture the composition and functioning of
the audit committee of the board of directors. Finally, the factor Fin Rep
appears to capture the boards control over the nancial reporting quality,
that is, whether there exists a nancial expert on the audit committee and
whether any top management or director has any association with the rms
auditor.
10
10
In PCA, it is desirable to have at least three variables loading on each retained component
when the PCA procedure is complete (Rummel [1970]). Two of my components, Dir Comp
andFin Rep, have only twovariables that loadsignicantly. Onmeasuring the reliability of these
components using Cronbachs alpha (discussed below), I nd that the Cronbachs alphas for
these two components are 0.59 and 0.56, respectively, which are somewhat lower than the
benchmark of 0.70 suggested by Nunnally [1967]. However, these levels of reliability are not
uncommon (LRT).
TA B L E 3
Principal Component Analysis: Factors and Factor Loadings
Principal Factor Signicant Components Factor Loadings Cronbachs Alpha
Board I %OUTDIR 0.77 0.85
INDP COMMITTEE 0.82
CEO COMMITTEE 0.72
BD EVAL 0.64
BD SIZE 0.68
Board II DIR %UNDER70 0.87 0.79
DIR AGE 0.84
#BD MEET 0.73
CEO COB 0.43
Exec Comp EXEC STOCK 0.67 0.77
EXEC OPTION 0.75
EXEC BONUS 0.78
Dir Comp DIR STKCOMP 0.47 0.59
DIR OPTION 0.63
Auditor NO NONAUDIT 0.60 0.54
%AUDIT FEES 0.75
BIG5 0.43
Audit Comm INDP AUDITCOM 0.56 0.50
#AUDIT MEET 0.42
%NONAFFIL OUTDIR 0.43
Fin Rep FIN EXPERT 0.58 0.59
EXEC NONAUDITOR 0.61
This table reports the factor loadings (absolute values) on each of the individual corporate governance
variables and the raw Cronbachs alphas for each factor. Factors are computed using exploratory principal
components analysis. For each factor, individual variables with absolute values of the loadings exceeding
0.40 are reported. The reported factor loadings and the Cronbachs alphas are rounded off to two places
after the decimal.
The method used to compute the individual governance variables ensures
that a higher value of each of the governance factors, Board I, Board II,
Auditor, Audit Comm, and Fin Rep, indicates higher quality of the corre-
sponding dimension of governance. However, in the case of the director
and executive compensation factors, namely, Dir Comp and Exec Comp, I
do not make any directional predictions.
To assess the reliability of the factors, I compute Cronbachs alpha, which
is a coefcient of reliability.
11
The raw Cronbachs alpha coefcients are
reported in table 3. Nunnally [1967] suggests that an alpha greater than
0.70 is an acceptable reliability coefcient. The mean (median) value of
the raw Cronbachs alpha is 0.65 (0.59). The alphas for the factors Board I,
Board II, and Exec Comp are over 0.70, but those for the remaining factors,
Dir Comp, Auditor, Audit Comm, and Fin Rep, are a bit lower than the
11
The formula for the standardized Cronbachs alpha is given by = N r /[1 + (N 1)
r ], where N is the number of items and r is the average interitem correlation among the items.
Thus, if the interitemcorrelations are high, then there is evidence that the items are measuring
the same underlying construct, that is, these items are measuring a single unidimensional latent
construct.
1160 A. DEY
benchmark of 0.70. However, as discussed in LRT, low reliability values are
common in early measurement stages, and these factors are very likely to
have higher reliability thanthe single indicators inmeasuringvarious aspects
of governance. Thus, although not perfect measures of governance, these
factors are likely to better capture the underlying governance dimensions
than individual variables or indices compiled by summing up individual
variables.
3.2.2. Agency Variables. Measuring the agency conicts in a rm is chal-
lenging because there exists a variety of rm-specic situations where the
managers have the incentives and the ability to engage in maximizing their
own utilities at the cost of shareholders. Moreover, there is no widely ac-
cepted measure of agency conicts. Thus, I use several variables to proxy
for the scope of agency conicts existing in a rm based on evidence in
prior studies. A brief discussion of each of these variables along with related
references is provided below.
3.2.2.1. Firm Size (SIZE). Size is measured as the natural logarithm of
sales. Large corporations are more likely to have highly diffuse ownership
structures that effectively separate ownershipof residual claims fromcontrol
of corporate decisions. Larger rms also have a greater scale of operations,
which provides greater incentive and opportunities for managers to shirk
(Demsetz andLehn[1985]). Moreover, larger rms are more likely to be un-
der greater political scrutiny, which provides managers of such rms greater
incentives to exercise discretion to minimize political costs (Watts and Zim-
merman [1990]). As a result, I expect the level of agency conicts to be
higher in larger rms.
3.2.2.2. Organizational Complexity (COMPLEX). Organizational com-
plexity is measured as the number of industries the rm operates in, where
the industries are measured by two-digit Standard Industrial Classication
(SIC) codes. Organizational complexity is an important component of the
scope for moral hazard (Bushman et al. [2004]). Multi-industry rms com-
bine diverse operations. The resulting information aggregation problems
can lead to substantial information asymmetries within the rm, or between
rm insiders and outside investors by suppressing the activities of informa-
tionintermediaries (Habib, Johnson, and Naik [1997], Gilsonet al. [2001]).
Capital allocationinsuchrms may also be inefcient, andthe CEOs of such
diverse rms may have reduced focus (Stein [2000]). As Givoly, Hayn, and
DSouza [1999] point out, though rms are required to disclose segment
data, this information suffers from imprecisions of segment identication,
cost allocations, and transfer pricing schemes. Accordingly, I expect agency
conicts to be higher in rms that are more complex.
3.2.2.3. Volatility in Operating Environment (GROWTH; RISK).
Demsetz and Lehn [1985] conjecture that the scope for moral hazard is
greater for managers of rms with more volatile operating environments.
They argue that more volatile environments result in greater difculty in
monitoring managers. I use two variables to proxy for the volatility in the
operating environment of a rm: growth opportunities, as measured by the
book-to-market ratio, and operating risk, as measured by the standard de-
viation of quarterly operating cash ows deated by total assets computed
over the immediately preceding four quarters (measuring operating risk as
the standard deviation of sales deated by total assets over the immediately
preceding four quarters does not materially alter any results).
High growth rms have higher levels of information asymmetry and man-
agers in these rms are likely to have more power due to a greater amount
of resources under their control (Smith and Watts [1992], Jensen [1986]).
Such rms are also expected to have greater incentives to meet earnings
targets, since prior research has shown that the market severely penalizes
growth rms for negative earnings surprises (Skinner and Sloan [2002]).
Riskier rms are expected to have greater incentives to indulge in activities
in order to reduce the perception of risk and thus lower the costs of equity
and debt capital. Accordingly, agency conicts are likely to be higher in
rms with greater operating risk and growth opportunities.
3.2.2.4. Ownership Structure (OWNERSHIP). Ownership structure is
measured as the percentage of capital owned by individual shareholders.
It is calculated as one minus the value of shares held by executives, direc-
tors, and institutional investors divided by the total market capitalization of
the rm. This is a measure of how diffuse the ownership structure of the
rm is, or, how much management control is present in the rm. As argued
by Berle and Means [1932], when managers hold little equity in the rm
and shareholders are too dispersed to enforce value maximization, corpo-
rate assets may be deployed to benet managers rather than shareholders.
In other words, management-controlled rms have considerable discretion
in guiding the affairs of corporations and this discretion could be used to
divert some resources from corporate shareholders (Morck, Shleifer, and
Vishny [1988]). Jensen and Meckling [1976] argue that owner-controlled
rms do not have the same incentives to divert resources, since owner
managers would suffer directly from reduced share values. I expect rms
with more diffuse ownership or more management control to have higher
levels of agency conicts.
3.2.2.5. Leverage (LEV ). Leverage is measured as the ratio of long-term
debt to total assets. The agency costs related to debt are likely to be higher
in rms with greater leverage. Ownermanagers have an incentive to accept
high-risk projects to transfer wealth from creditors to shareholders, and are
also likely to forego positive net present value projects because most of their
created value is captured by bondholders (Myers [1977]). These activities
are likely to reduce rm value. Furthermore, rms with higher leverage
ratios have greater incentives to manage earnings in order to avoid covenant
violations and/or to prevent adverse effects on their debt ratings (Bowen,
Rajgopal, and Venkatachalam [2005], DeFond and Jiambalvo [1994], Watts
1162 A. DEY
TA B L E 4
Correlation Table for the Agency Variables
SIZE COMPLEX OWNERSHIP GROWTH LEV RISK FCF
SIZE 0.0032 0.1765
0.0250 0.0202 0.5563
0.0302
COMPLEX 0.0017 0.0109 0.0032 0.1052 0.1765 0.0021
OWNERSHIP 0.1446
0.0074 0.0205 0.0153 0.1697
0.0509
GROWTH 0.0628 0.0012 0.0488 0.0045 0.0427 0.0026
LEV 0.1992
0.1276 0.0347 0.0257 0.3796
0.0006
RISK 0.3012
0.1864 0.1362
0.0784 0.1343
0.0195
FCF 0.0415 0.0185 0.0414 0.0913
0.1457
0.0433
This table presents the Pearson and Spearman correlation coefcients between the seven agency
variables. Pearson correlations are reported above the diagonal and Spearman correlations are reported
below the diagonal. SIZE is measured as the natural logarithm of sales; COMPLEX is measured as the
number of industries the rm operates in, where the industries are measured by two-digit SIC code;
OWNERSHIP is calculated as one minus the value of shares held by executives, directors, and institutional
investors divided by the total market capitalization of the rm; GROWTH is measured by the book-to-market
ratio; LEV is measured as the ratio of long-term debt to total assets; RISK is measured as the standard
deviation of quarterly operating cash ows divided by total assets computed over the immediately preceding
four quarters; FCF is measured as the free cash ow of the rm scaled by current assets (free cash ows are
dened as the difference between the cash ow from operations of the previous quarter and the preceding
three quarter average of the rms capital expenditures scaled by the current assets of the previous quarter).
indicates signicance at the 1% level.

and Zimmerman [1990]). Such rms are thus likely to have higher agency
conicts.
3.2.2.6. Free CashFlow(FCF ). Free cashows are measuredas the differ-
ence between the cash ow from operations of the previous quarter and the
preceding three quarter average of the rms capital expenditures, scaled by
the current assets of the previous quarter.
12
Jensen [1986] argues that con-
icts of interests between shareholders and managers are especially severe
when there are substantial free cash ows generated in the organization.
Thus, I expect that rms with greater levels of free cash ows will have
higher agency conicts.
Table 4 presents the correlations (Pearson and Spearman) between the
above agency variables. The following correlations are statistically signi-
cant (at the 1%level). The variables SIZE and OWNERSHIP are signicantly
correlated (Pearson, 0.18; Spearman, 0.14), suggesting that for this sample
managers of larger rms have more control. Larger rms also have higher
leverage (only the Pearson correlation between SIZE and LEV is signi-
cant, 0.20). Both the Pearson (0.56) and Spearman (0.30) correlations for
the variables SIZE and RISK are signicant, indicating that for this sample
larger rms have greater operating risk. More risky rms also have greater
management control as indicated by the signicant correlations between
the variables OWNERSHIP and RISK (Pearson, 0.17; Spearman, 0.14). Only
the Spearman correlation between GROWTH and FCF is negative and statis-
tically signicant (0.09). As expected, higher growth rms have lesser free
12
I also use an alternative measure representing the cash on hand, dened as (cash +
short-term investments)/total assets. This does not materially alter any results.
cash ows. Sample rms with higher leverage are also more risky (Pearson,
0.38; Spearman, 0.13). Finally, only the Spearman correlation between LEV
and FCF is negative and signicant (0.15), indicating that rms in the
sample with higher leverage also have lesser free cash ows.
3.2.3. Agency Conict Groups. In order to examine how governance struc-
tures of rms vary as a function of the level of agency conicts in the rm,
I rst classify rms into homogeneous groups using cluster analysis based
on the above seven variables. Cluster analysis is an exploratory data analysis
tool which sorts different objects into groups by maximizing the degree of
association between two objects in the same group and minimizing the de-
gree of association between two objects across groups. Using a data analysis
tool, such as cluster analysis, for separating the rms into groups is more ef-
cient (than, for example, ranking the rms and forming deciles) because
the process itself identies the optimal rm clusters for the sample based
on the specied variables.
There are, however, no completely satisfactory methods for determining
the optimal number of clusters for any type of cluster analysis. Based on
simulation studies by Cooper and Milligan [1985, 1988] three criteria that
perform best to determine the optimal number of clusters are the cubic
clustering criterion (CCC) developed by Sarle [1983], a pseudo-F-statistic
(PSF) developed by Calinski and Harabasz [1974], and a pseudo-t
2
-statistic
(PST2) based on a statistic developed by Duda and Hart [1973]. A good
estimate for the optimal number of clusters is obtained by using a consensus
among these three statistics, that is, by determining local peaks of the CCC
and PSF combined with a small value of the PST2 followed by a signicantly
larger PST2 value for the next cluster fusion. A brief overview of the cluster
analysis procedure is provided in the appendix (see Everitt [1980], Massart
and Kaufman [1983], Anderberg [1973] for greater detail).
Table 5, panel A reports the values of these statistics for the sample.
13
The CCC and PSF statistics identify two possible numbers of clusters for the
sample, three and seven. PST2 reveals possible clustering levels at three,
seven, and nine. Thus, three and seven are the possible number of clusters
for this data set. I form three clusters due to limitations on the number of
rms in the sample. Next, I use Wards method of clustering, which forms
clusters by minimizing the variance within each cluster. Wards method of
clustering is the most commonly used method among hierarchical cluster-
ing algorithms, and produces clusters with roughly the same number of
observations (Everitt [1980, 1993]).
In order to distinguish among the groups in terms of the level of agency
conicts, I perform a principal factor analysis of the seven agency variables
that are usedtosegregate the rms intoclusters, andderive anoverall agency
score for each rm. This serves as an overall measure of the level of agency
13
I report the values for 10 clusters, but an analysis of the corresponding statistics for 20
clusters does not reveal any additional clustering possibilities for the sample.
1164 A. DEY
conicts in a rm. The mean value of this agency score indicates that agency
conicts are highest for the rms belonging to cluster 3 (mean agency
score = 1.12), followed by rms in cluster 2 (mean agency score = 0.26),
and are lowest for rms in cluster 1 (mean agency score = 0.82). Based
on this score, cluster 3 represents the high agency conict group (group
HIGH), cluster 2 represents the medium agency conict group (group
MEDIUM), and cluster 1 represents the low agency conict group
(group LOW). The overall agency score is signicantly different across
the three groups (t-statistic for the difference between groups HIGH and
MEDIUM = 21.35; t-statistic for the difference between groups HIGH and
LOW = 30.27; t-statistic for the difference between groups MEDIUM
and LOW = 31.52).
Table 5, panel B presents the mean and median values of the seven indi-
vidual agency variables across the three agency groups. It is interesting to see
which of these seven variables are signicantly different across the groups.
Compared to rms in groups MEDIUM and LOW, rms in group HIGH are
largerthe mean (median) SIZE is 8.77 (8.95) for group HIGH, followed
by 7.14 (7.08) for group MEDIUM, and 5.80 (5.81) for group LOW. There
is not much variation in organizational complexity across the rms, as is evi-
dencedby the identical means andmedians of the variable COMPLEX across
all three agency groups. Firms in the higher agency groups also have a more
diffuse ownershipstructure. The mean(median) values of the variable OWN-
ERSHIP are 0.84 (0.66) for group HIGH, 0.75 (0.55) for group MEDIUM,
and 0.53 (0.41) for group LOW. The values for GROWTH, although higher
in the highest agency group than in the other two lower groups, are more
or less comparable across the MEDIUM and LOW agency groups. The mean
(median) values for GROWTH are 0.05 (0.05) for group HIGH, 0.04 (0.03)
for group MEDIUM, and 0.04 [0.02] for group LOW. Firms in group HIGH
TA B L E 5
Agency Conict Groups
Panel A: Optimal number of clusters
No. of Cubic Clustering Pseudo-F Pseudo-t
2
Clusters Criterion (CCC) (PSF) (PST2)
10 7.2 76 10
3
205
9 12 74 10
3
822
8 15 73 10
3
168
7 29 57 10
4
1,051
6 19 37 10
3
476
5 18 31 10
3
354
4 14 30 10
3
294
3 30 27 10
4
4,655
2 28.4 9 10
2
17 10
2
1 0.0 NA 10 10
2
This table presents the values for the three statistics used to determine the optimal number of clusters
for the sample, namely, the cubic clustering criterion (CCC), the pseudo-F (PSF), and the pseudo-t
2
(PST2). For each criterion, the rows in bold indicate the clustering possibilities identied by the criterion.
(Continued)
Panel B: Cluster-wise means/medians
Cluster 1 Cluster 2 Cluster 3
(Low Agency (Medium Agency (High Agency
Group) Group) Group)
Variable Mean Median Mean Median Mean Median
SIZE
5.7978 5.8119 7.1426 7.0779 8.7658 8.9484

COMPLEX 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
OWNERSHIP
0.5332 0.4190 0.7510 0.5543 0.8435 0.6634

GROWTH 0.0422 0.0194 0.0392 0.0349 0.0468 0.0499
LEV
0.2986 0.5901 1.2706 0.7638 1.6385 1.3964

RISK
0.0246 0.0209 0.0450 0.0416 0.0664 0.0597

FCF 0.1035 0.0537 0.1967 0.0691 0.2276 0.0713
Overall Agency Score
AGENCY 0.3794 1.4509 1.5913
No. of rms 111 143 117
This table presents the cluster-wise mean and median values of the seven variables that are used to
segregate rms into three clusters and the overall agency scores for each cluster. SIZE is measured as the
natural logarithm of sales; COMPLEX is measured as the number of industries the rm operates in, where
the industries are measured by two-digit SICcode; OWNERSHIP is calculated as one minus the value of shares
held by executives, directors, and institutional investors divided by the total market capitalization of the rm;
GROWTH is measured by the book-to-market ratio; LEV is measured as the ratio of long-term debt to total
assets; RISK is measured as the standard deviation of quarterly operating cash ows divided by total assets
computed over the immediately preceding four quarters; FCF is measured as the free cash ow of the rm
scaled by current assets (free cash ows are dened as the difference between the cash ow from operations
of the previous quarter and the preceding three quarter average of the rms capital expenditures scaled by
the current assets of the previous quarter); the overall agency score, AGENCY, is computed by performing
principal factor analysis of the four variables SIZE, OWNERSHIP, LEV , and RISK.
indicates a variable that has signicant factor loadings when principal factor analysis is performed on
all of the seven agency variables.
The results of Wilcoxon-Mann-Whitney two sample tests for differences in medians of SIZE, OWNERSHIP,
LEV , and RISK and t-tests for differences in means of SIZE, OWNERSHIP, LEV , and RISK and the agency
score, AGENCY, between the three agency groups are reported below (
,

, and

indicate signicance at
less than the 10%, 5%, and 1% levels, respectively):
t-statistic for the differences in means and z-statistic for differences in medians between the HIGH agency
group and the MEDIUM agency group:
SIZE: t = 24.35
; z = 20.82
OWNERSHIP: t = 13.66
; z = 12.34
LEV : t = 1.44; z = 2.33
RISK: t = 1.84
; z = 2.04
AGENCY: t = 26.39
t-statistic for the differences in means and z-statistic for differences in medians between the HIGH agency
group and the LOW agency group:
SIZE: t = 49.61
; z = 27.51
OWNERSHIP: t = 15.34
; z = 17.43
LEV : t = 2.27
; z = 2.13
RISK: t = 2.25
; z = 3.24
AGENCY: t = 43.64
t-statistic for the differences in means and z-statistic for differences in medians between the MEDIUM
agency group and the LOW agency group:
SIZE: t = 36.09
; z = 26.20
OWNERSHIP: t = 6.90
; z = 8.60
LEV : t = 1.98
; z = 0.19
RISK: t = 0.70; z = 0.84
AGENCY: t = 40.82
1166 A. DEY
also have higher leverage than rms in the other two groupsthe mean
(median) LEV is 1.64 (1.40) for group HIGH, followed by 1.27 (0.76) for
group MEDIUM and 0.30 (0.59) for group LOW. The mean and median
values for RISK are also increasing in the level of agency conicts, with
the mean (median) being 0.07 (0.06) for group HIGH, followed by 0.05
(0.04) for group MEDIUM and 0.02 (0.02) for group LOW. Finally, the
mean (median) values for FCF are 0.22 (0.07) for group HIGH, 0.20 (0.07)
for group MEDIUM, and 0.10 (0.05) for group LOW.
Although the individual agency variables are, on average, higher in the
higher agency groups (with the exception of COMPLEX), the overall agency
score indicates that the variables SIZE, OWNERSHIP, LEV , and RISK have
the maximumfactor loadings. This suggests that the overall factor primarily
reects the variability in these three variables. Given this result, I perform a
t-test (Wilcoxon-Mann-Whitney test) of the differences in the means (medi-
ans) of these variables across the groups. I nd that, except for the mean of
LEV between the HIGH and the MEDIUM groups and the median of LEV
and mean and median of RISK between the MEDIUM and LOW groups,
all four variables are signicantly higher in the higher agency groups (see
table 5, panel B).
14
The above result suggests that rms inthe three agency clusters differ from
each other primarily in SIZE, OWNERSHIP, LEV , and RISK. I thus rene
the overall agency score by performing a factor analysis on only these four
variables, SIZE, OWNERSHIP, LEV , and RISK.
15
This rened agency mea-
sure is called AGENCY. This procedure eliminates any noise in the agency
score owing to the other agency measures that are not signicantly different
across the three groups (namely, COMPLEX, GROWTH, and FCF ). For the
groups HIGH, MEDIUM, and LOW the mean values of AGENCY are 1.59,
1.45, and 0.38, respectively. Not surprisingly, AGENCY is also signicantly
different across the three groups. I use this score as a measure of the overall
level of agency conicts in rms in subsequent analyses.
16
One potential probleminusingsize andgrowthopportunities as measures
of agency conicts is that these variables could also be viewed as investment
opportunity variables (Smith and Watts [1992]). To be more condent that
these measures are capturing the level of agency conicts, I conduct the
following robustness check. I repeat my analyses using the variable OWN-
ERSHIP (described above) as my measure of agency conicts. This variable
measures how diffuse the ownership structure of the rmis and the amount
14
One possibility for only four of the seven agency variables to be signicantly different
across the clusters is likely to be due to forming three instead of seven clusters. If I could form
seven clusters with the data, the formation of ner clusters would most probably have rms in
each of the clusters signicantly different in more of the agency variables (except possibly for
the variable COMPLEX, given that there is not much variation in this across the sample rms).
15
I thank the referee for this suggestion.
16
I also repeat my analyses using the original factor obtained by performing factor analysis
on all seven variables and obtain very similar results. Thus, I only report results using the factor
AGENCY, which is a cleaner measure.
of control the manager has, and is a more traditional measure of agency con-
icts (Wareld, Wild, and Wild [1995], Morck, Shleifer, and Vishny [1988]).
As in these studies, I form agency conict groups based on three levels of
managerial control: when OWNERSHIP 95% (highest agency conicts),
when 75% < OWNERSHIP < 95% (medium agency conicts), and when
OWNERSHIP 75% (lowest agency conicts). The results obtained by us-
ing this alternate specication are consistent with those reported in the
paper.
17
4. Tests and Results: Agency Conicts and Governance Quality
I begin by investigating the relation between agency conicts and gover-
nance quality. Table 6 reports the median values of the seven governance
factors for the entire sample, as well as for the three agency conict groups
(the factors have a mean of zero and standard deviation of one by con-
struction). The median values of the factors Board I, Board II, Auditor,
Audit Comm, and Fin Rep are highest in group HIGH and lowest in group
LOW, indicating that rms with greater agency conicts have better quality
of governance in these dimensions. Such rms are likely to have greater
demands for more monitoring mechanisms and thus have stronger gover-
nance systems in place. Interestingly, for Exec Comp, the median value is
the highest for group LOW and the lowest for group HIGH, implying a
negative relation between the percentage of stock and option compensa-
tion of executives and agency conicts. One explanation for this inverse
relation could be the effect of substitution between governance mecha-
nisms. Prior studies document that the composition of the board is inversely
related to managerial equity holdings (Weisbach [1988], Mehran [1992],
Barnhart and Rosenstein [1998]). Given that both the board-related factors,
Board I and Board II, are positively related to agency conicts, this could
explain the negative relation between executive compensation and agency
conicts.
I conduct two-sample Wilcoxon-Mann-Whitney tests to test whether the
medians of these factors are signicantly different across the three groups
(the z-statistics obtainedare presentedintable 6). The three factors Board I,
Board II, and Audit Comm are signicantly higher in the HIGH versus the
MEDIUM, the HIGH versus the LOW, and the MEDIUM versus the LOW
agency groups. Among the other factors, Auditor is signicantly higher in
the HIGH versus the LOW groups andthe MEDIUM versus the LOW groups.
Exec Comp is signicantly lower in the HIGH versus the MEDIUM, the
HIGH versus the LOW, and the MEDIUM versus the LOW agency groups.
Dir Comp is not signicantly different across any of the three groups, and
Fin Rep is signicantly higher in the HIGH versus the LOW and HIGH ver-
sus the MEDIUM groups, andis marginally higher inthe MEDIUM versus the
17
These are not reported for the sake of brevity, but are available on request.
1168 A. DEY
TA B L E 6
Principal Factors: Descriptive Statistics
Agency Group
Principal Factor All HIGH MEDIUM LOW
Board I 0.1563 0.3548 0.1992 0.0359
Board II 0.0077 0.2529 0.1626 0.3859
Exec Comp 0.1927 0.3151 0.2535 0.0227
Dir Comp 0.1688 0.1464 0.2418 0.1616
Auditor 0.0397 0.1210 0.0782 0.0373
Audit Comm 0.2012 0.4061 0.3138 0.1233
Fin Rep 0.0208 0.1343 0.0263 0.0564
This table presents the median values of the seven governance factors for all rms, and for rms in the
three agency conict groups. The seven governance factors are obtained by performing principal compo-
nents analysis on 22 individual corporate governance variables.
The results of Wilcoxon-Mann-Whitney two sample tests for differences in medians between the gover-
nance factors for the three agency groups are reported below (
,

, and

indicate signicance at the less
than 10%, 5%, and 1% levels, respectively):
z-statistic for differences in medians between the HIGH agency group and the MEDIUM agency group:
Board I: z = 5.41
Board II: z = 2.52
Exec Comp: z = 3.58
Dir Comp: z = 0.51

Auditor: z = 1.17
Audit Comm: z = 3.15
Fin Rep: z = 3.88
z-statistic for differences in medians between the HIGH agency group and the LOW agency group:
Board I: z = 6.81
Board II: z = 2.24
Exec Comp: z = 7.33
Dir Comp: z = 1.19

Auditor: z = 5.26
Fin Rep: z = 3.01
z-statistic for differences in medians between the MEDIUM agency group and the LOW agency group:
Board I: z = 5.18
Board II: z = 2.21
Exec Comp: z = 4.92
Dir Comp: z = 0.47

Auditor: z = 4.10
Fin Rep: z = 1.37
LOW groups. These results support the theory that rms with higher levels
of agency conicts have higher quality of governance structures in place,
particularly those related to the board of directors, the audit committee, the
boards control over the nancial reporting process, and the auditor.
In order to perform a more formal test of the relation between agency
conicts and the governance factors, I estimate the following regression:
AGENCY
j q
= +
7
i =1
i
GOV FACTOR
i
+ (1)
In the above equation AGENCY corresponds to the principal factor analysis
of the four signicant agency variables, SIZE, OWNERSHIP, LEV , and RISK,
TA B L E 7
Agency Conicts and Governance Quality
AGENCY
jq
= +
7
i =1
i
GOV FACTOR
i
+
Coefcient
(t-Statistic)
Intercept 0.1768
(7.01
)
Board I
j
0.2967
(11.53
)
Board II
j
0.0587
(2.38
)
Exec Comp
j
0.3652
(16.92
)
Dir Comp
j
0.0026
(0.09)
Auditor
j
0.1468
(5.63
)
Audit Comm
j
0.2452
(10.76
)
Fin Rep
j
0.0680
(1.45)
Adjusted R
2
0.3602
F -value (Pr > F) 78.58 (<0.0001)
No. of rms 371
This table analyses the association between the level of agency conicts, as represented by the agency
score, AGENCY, and the various dimensions of governance, as represented by the seven governance factors.
t-statistics and the levels of signicance are reported in parentheses. For each rm j and quarter q, the
dependent variable, AGENCY, is computed by performing principal factor analysis of the four variables
SIZE, OWNERSHIP, LEV , and RISK; Board I, Board II, Exec Comp, Dir Comp, Auditor, Audit Comm,
and Fin Rep are the seven governance factors obtained by performing exploratory principal components
analysis on 22 individual governance variables.
and

indicate signicance at the less than 5% and 1% levels, respectively.
and GOV FACTOR variables correspond to the seven governance factors,
namely, Board I, Board II, Exec Comp, Dir Comp, Auditor, Audit Comm,
and Fin Rep.
The results of this regression are reported in table 7. The results of the
regression correspond to the univariate results discussed earlier. The factors
Board I, Board II, Auditor, and Audit Commare positive and highly signi-
cant, indicating that rms with higher levels of agency conicts have higher
qualities of governance structures in place with respect to the board and the
auditor. As in the univariate tests, Exec Comp is negative and signicant,
indicating an inverse relation between agency conicts and the stock and
options owned by executives. Finally, the factors Dir Comp and Fin Rep are
not signicantly related to the level of agency conicts.
The above results support the hypothesis that governance structures arise
in response to the agency conicts in rms. Firms with higher levels of
agency conicts have more efcient governance structures in place, par-
ticularly, more independent and better functioning boards and audit com-
mittees, and a better quality auditor. A natural extension of the result that
1170 A. DEY
governance structures vary as a function of the agency conicts in rms is
that the role played by these governance mechanisms in affecting overall
rm performance is also likely to vary as a function of the agency conicts.
In the next section I investigate this conjecture.
5. Additional Analyses: Firm Performance and Governance
One important question that has been rigorously studied in the gover-
nance literature is whether governance, or various aspects of governance,
affects overall rm performance. The empirical evidence on this relation,
however, is mixed. For instance, Hermalin and Weisbach [1991] and Bhagat
and Black [2001] nd no relation between the proportion of outsider direc-
tors and various performance measures. In contrast, Baysinger and Butler
[1985] and Rosenstein and Wyatt [1990] showthat the market rewards rms
for appointing outside directors.
A few studies also suggest that rms with a high percentage of indepen-
dent directors may perform worse. For instance, Yermack [1996] reports a
signicant negative correlation between the proportion of independent di-
rectors and contemporaneous Tobins Q, but no signicant correlation for
several other performance variables (sales/assets, operating income/assets,
operating income/sales). Yermack [1996] also documents an inverse rela-
tion between board size and protability, asset utilization, and Tobins Q.
Agrawal and Knoeber [1996] report a negative relation between the pro-
portion of outside directors and Tobins Q. Klein [1998] does not nd
a signicant relation between rm performance and board structure as
a whole, but documents that inside director representation on a boards
nance and investment committees correlates with improved rm perfor-
mance. She nds little evidence that the audit, compensation, and nomi-
nating committees, which are usually dominated by independent directors,
affect performance.
18
More recently, LRT document that some of their governance indices are
associated with future return on assets (ROA). They nd that institutional
ownership, long-term and bonus compensation of the CEO, and certain an-
titakeover measures have a positive relationwithfuture ROA, andboardsize,
audit committee size and compensation committee size, and the busyness
of directors have a negative relation with future ROA.
I conjecture that one explanation for the mixed association between gov-
ernance and performance is that governance mechanisms are likely to pos-
itively affect overall rm performance only under certain circumstances.
Specically, I test whether the relation between governance mechanisms
and rm performance varies as a function of the level of agency conicts
in rms. The governance mechanisms in place are important in monitor-
ing the actions of managers in rms with high agency conicts. In contrast,
18
The literature on governance and rm performance is vast, and I discuss only some of
the important and relevant studies. See Shleifer and Vishny [1997], John and Senbet [1998],
and Hermalin and Weisbach [2003] for more detailed literature reviews.
managers in low agency conict rms are not likely to require as much mon-
itoring, and so these governance mechanisms are not likely to be important
in affecting future performance.
I expect the performance of high agency rms to be positively related to
the factors Board I, Board II, Auditor, Audit Comm, and Fin Rep. As be-
fore, the associationwithExec Comp and Dir Comp is not clear. Incontrast,
for the lower agency companies the governance factors are not expected to
be important determinants of their performance. Based on prior research
I use two proxies for rm performance: Tobins Q (Q) dened as (market
value of equity + total debt)/total assets, and return on assets (ROA), de-
ned as (net income before extraordinary items)/total assets (Agrawal and
Knoeber [1996], Yermack [1996], Bhagat and Black [1999], among others).
I use the one-year-ahead values of the above variables to control for potential
endogeneity issues.
The mean (median) values of the variable Q for the HIGH, MEDIUM,
and LOW agency groups are 1.97 (1.77), 1.61 (1.14), and 1.24 (1.06), re-
spectively. The mean (median) values of the variable ROA for the HIGH,
MEDIUM, and LOW agency groups are 0.04 (0.02), 0.02 (0.01), and 0.01
(0.01), respectively. There is a monotonic relationbetweenthe performance
measures and the level of agency conicts, and as expected, the high agency
group has higher values for both performance measures. Next, I formally
examine whether the relation between governance and rm performance
is a function of agency conicts. I perform the following regression for each
of the three agency groups:
PERFORMANCE MEASURE
j,t +1
= +
7
i =1
i
GOV FACTOR
i,t
+ L PERFORMANCE MEASURE
j,t
+ RD/SALES
j,t
+ (INV +PPE)/TOTALASSETS
j,t
+ (2)
where PERFORMANCE MEASURE represents the variables Q and ROA, and
GOV FACTOR represents the seven governance factors, Board I, Board II,
Exec Comp, Dir Comp, Auditor, Audit Comm, and Fin Rep. Given that I
performthe above regression separately for each of the three agency groups
that have similar characteristics of the various agency measures (particularly
for size, operating risk, and leverage), I do not include these variables as ad-
ditional controls inthe regression. However, basedonprior studies, I include
the lagged values of the dependent variables (L PERFORMANCE MEASURE
in equation (2) is either the lagged value of Q or the lagged value of
ROA), a proxy for the assets in place measured by the sum of inventory
and gross property, plant and equipment divided by total assets (INV +
PPE)/TOTALASSETS, and a proxy for growthopportunities measured as the
research and development expenses divided by sales (RD/SALES) (Mehran
[1995], Klein [1998]).
Table 8 reports the results of this analysis. Panel Areports the results when
Q is the dependent variable, and panel B reports the results when ROA is
1172 A. DEY
used as the dependent variable. The results in panel A indicate that Q is
positively associated with the two factors related to the composition and
functioning of the board, Board I and Board II; director compensation,
Dir Comp; the independence of the auditor, Auditor; and audit committee
effectiveness, Audit Comm, for the HIGH agency rms. The factor Fin Rep
is positive and marginally signicant. Only the factors Board I, Board II,
TA B L E 8
Firm Performance and Governance Quality
Panel A: Tobins Q and governance quality
Q
j,t +1
= +
7
i =1
i
GOV FACTOR
i,t
+ L Q
j,t
+ RD/SALES
j,t
j,t
+
HIGH MEDIUM LOW
Agency Agency Agency
Expected Coefcient Coefcient Coefcient
Sign (t-Statistic) (t-Statistic) (t-Statistic)
Intercept 4.2476 0.9307 0.8996
(4.05
) (2.08
) (3.48
)
Board I
j,t
+ 0.1451 0.4609 0.0117
(3.33
) (3.76
) (1.11)
Board II
j,t
+ 0.1302 0.2857 0.0503
(2.68
) (2.35
) (1.18)
Exec Comp
j,t
? 0.0721 0.0299 0.0203
(1.18) (1.24) (0.51)
Dir Comp
j,t
? 0.2569 0.1352 0.1518
(3.51
) (1.56) (1.89
)
Auditor
j,t
+ 0.2407 0.1747 0.0278
(3.19
) (1.02) (1.19)
Audit Comm
j,t
+ 0.1689 0.5972 0.1174
(3.10
) (3.16
) (1.43)
Fin Rep
j,t
+ 0.1832 0.0906 0.0143
(1.88
) (0.50) (0.26)
L Q
j,t
+ 0.3677 0.4488 0.4266
(4.50
) (7.58
) (4.56
)
RD/SALES
j,t
+ 0.1349 0.1172 0.0904
(1.47) (0.08) (1.09)
(INV +PPE)/ + 0.2029 0.0429 0.1611
TOTALASSETS
j,t
(1.18) (0.48) (2.31
)
Adjusted R
2
0.4135 0.4338 0.2903
F -value (Pr > F) 12.87 (<0.0001) 14.10 (<0.0001) 11.06 (<0.0001)
No. of rms 110 142 111
This table analyses the association between future Tobins Q and the various dimensions of governance
as represented by the seven governance factors. t-statistics and the levels of signicance are reported in
parentheses. For each rm j and year t, the dependent variable, Q, is the one-year-ahead (market value of
equity + total debt)/total assets; Board I, Board II, Exec Comp, Dir Comp, Auditor, Audit Comm, and
Fin Rep are the seven governance factors obtained by performing principal components analysis on 22
individual corporate governance variables; L Q is the lagged value of the dependent variable Q; RD/SALES
is measured as research and development expenses divided by sales; (INV +PPE)/TOTALASSETS is
measured as the sum of inventory and gross property, plant, and equipment divided by total assets.
,

, and

indicate signicance at the less than 10%, 5%, and 1% levels, respectively.
(Continued)
Panel B: Return on assets and governance quality
ROA
j,t +1
= +
7
i =1
i
GOV FACTOR
i,t
+ L ROA
j,t
+ RD/SALES
j,t
j,t
+
HIGH Agency MEDIUM Agency LOW Agency
Expected Coefcient Coefcient Coefcient
Sign (<t-Statistic) (<t-Statistic) (<t-Statistic)
Intercept 0.0511 0.0514 0.0348
(2.83
) (2.54
) (4.05
)
Board I
j,t
+ 0.0169 0.0643 0.0037
(2.18
) (2.40
) (0.70)
Board II
j,t
+ 0.0488 0.0054 0.0012
(2.58
) (0.41) (0.49)
Exec Comp
j,t
? 0.0048 0.0100 0.0075
(0.41) (1.12) (1.54)
Dir Comp
j,t
? 0.0221 0.0045 0.0057
(2.25
) (0.30) (0.61)
Auditor
j,t
+ 0.0023 0.0135 0.0021
(0.49) (1.24) (0.67)
Audit Comm
j,t
+ 0.0705 0.0431 0.0245
(2.13
) (2.38
) (4.02
)
Fin Rep
j,t
+ 0.0041 0.0126 0.0026
(0.64) (0.94) (0.66)
L ROA
j,t
+ 0.2231 0.1139 0.1799
(7.22
) (3.73
) (3.86
)
RD/SALES
j,t
+ 0.0093 0.0312 0.0447
(0.23) (0.59) (0.66)
(INV +PPE)/ + 0.0371 0.0537 0.0325
TOTALASSETS
j,t
(0.98) (0.79) (1.86
)
Adjusted R
2
0.2269 0.1153 0.1542
F -value (Pr > F) 6.98 (<0.0001) 5.63 (<0.0001) 6.88 (<0.0001)
No. of rms 110 142 111
This table analyses the association between future return on assets and the various dimensions of
governance as represented by the governance factors. t-statistics and the levels of signicance are reported
in parentheses. For each rm j and year t, the dependent variable, ROA, is the one-year-ahead (net income
before extraordinary items)/total assets; Board I, Board II, Exec Comp, Dir Comp, Auditor, Audit Comm,
and Fin Rep are the seven governance factors obtained by performing principal components analysis on
22 individual corporate governance variables; L ROA is the lagged value of the dependent variable ROA;
RD/SALES is measured as research and development expenses divided by sales; (INV +PPE)/TOTALASSETS
is measured as the sum of inventory and gross property, plant, and equipment divided by total assets.
,

, and

indicate signicance at the less than 10%, 5%, and 1% levels, respectively.
and Audit Commare positive and signicantly related to Q for the MEDIUM
group. However, for the LOW agency group, only the factor representing
director compensation, Dir Comp, is positive and marginally signicant.
The factor Exec Comp is not signicant for any of the agency groups.
Among the control variables, as expected, the lagged dependant variable
L Q is positive and signicant for all three agency groups. The variable
RD/SALES is not signicant for any of the three agency groups. This result is
not surprising giventhat this variable is a proxy for growthopportunities and
growth is one of the variables used to separate rms into the various agency
1174 A. DEY
groups. Thus, there is unlikely to be much variation in growth opportunities
across rms in each agency group. The proxy for assets in place, (INV +
PPE)/TOTALASSETS, is positive and signicant only for the LOW agency
group and insignicant for the other two groups.
The results for the performance measure ROAhave a similar avor as well.
The factors representing the composition and functioning of the board,
Board I and Board II, director stock and option compensation, Dir Comp,
and audit committee effectiveness, Audit Comm, are positive and signi-
cant for the HIGH agency group, while one of the board-related factors,
Board I, and audit committee effectiveness, Audit Comm, is positive and
signicant for the MEDIUM agency group. For the LOW agency group only
audit committee effectiveness, Audit Comm, is positive and signicant. The
factors Exec Comp and Fin Rep are not signicant for any of the agency
groups.
The results for the control variables are similar to those obtained earlier.
The lagged variable L ROA is positive and signicant for all three agency
groups. The variable RD/SALES is not signicant for any of the three agency
groups and (INV + PPE)/TOTALASSETS is positive and marginally signi-
cant only for the LOW agency group.
The above evidence suggests that the composition and functioning of
the board, an effective audit committee, and the stock and option com-
pensation provided to directors are signicantly associated with future rm
performance. However, this relation holds primarily for rms in the high-
est agency group for both the performance variables. Interestingly, for the
performance variable ROA, the audit committee factor is signicant for all
agency groups, indicating that a strong audit committee is associated with
higher ROA for all rms. These results support the conjecture that the re-
lation between various aspects of governance and rm performance is a
function of the rms level of agency conicts. In other words, the result in
the prior literature that there is no noticeable relation between the propor-
tion of outside directors (which is part of the factor Board I in my analysis)
and rm performance is probably true only for rms where the level of
agency conicts is low (Hermalin and Weisbach [1991], Bhagat and Black
[2001]). Overall, these results provide greater reinforcement for the theory
that the presence and role of governance structures vary across rms de-
pending on various rm-specic characteristics, one of which is the level of
agency conicts present in rms.
6. Conclusion
I examine the relation between the level of agency conicts in a rm
and its governance structure. I nd that rms with higher levels of agency
conicts also have better governance mechanisms in place, particularly re-
lated to the composition and functioning of the board of directors, the
audit committee, and the independence of the auditor. These results sup-
port the theory on corporate governance that governance mechanisms are
an endogenous response to a rms business and economic environment.
I also examine whether the relation between governance and overall rm
performance varies as a function of the agency conicts, and nd evidence
consistence with this conjecture. The evidence documented in the paper
provides support for the argument raised by several researchers on gover-
nance that one size does not t all.
These results are also important from a regulatory point of view. The
objective of improving shareholder value and investor condence in corpo-
rate disclosures and other operating decisions by introducing governance
reforms will be better met if there is evidence on the aspects of governance
that need to be modied, and the types of rms for which such measures
will be more effective. Finally, this analysis reiterates the importance of fur-
ther research on how governance structures develop, and the relations be-
tween the different dimensions of governance. This will not only help in
increasing our knowledge on why rms have different governance struc-
tures, but also enable more sophisticated examinations of the importance
of governance mechanisms in affecting organizational performance and
shareholder value.
APPENDIX
Cluster Analysis
Cluster analysis (CA) is a multivariate procedure for detecting natural
groupings in data. CA classication is based upon the placing of objects into
more or less homogeneous groups, in a manner such that the relationship
between groups is revealed. CA lacks an underlyingbody of statistical theory
and is heuristic in nature. It requires decisions to be made by the user
relating to the calculation of clusters, decisions that have a strong inuence
onthe results of the classication. CAis useful for the classicationof groups
or objects and is more objective than subjective.
In the classication literature, two kinds of clustering algorithms are re-
ported: partitioning and hierarchical methods. A partitioning method clas-
sies objects into a speciednumber of groups (say k), whichtogether satisfy
the following criteria:
1) Each group must contain at least one object
2) Each object must belong to one group
These conditions imply that k n, where n is the total number of objects.
Hierarchical algorithms do not construct a single partition with k clusters,
but they deal with all values of k in the interval [1, n]. Thus in hierarchical
clustering the data are not partitioned into a particular cluster in a sin-
gle step. Instead, a series of partitions takes place, which may run from a
single cluster containing all objects to n clusters each containing a single
object. Hierarchical clustering is subdivided into agglomerative methods,
which proceed by series of fusions of the n objects into groups, and divisive
methods, whichseparate n objects successively intoner groupings. Since all
1176 A. DEY
agglomerative hierarchical techniques ultimately reduce the data to a single
cluster containing all the individuals, and the divisive techniques will nally
split the entire set of data into n groups each containing a single individual,
the researcher needs to determine an optimal number of clusters and de-
cide on a particular stage to stop (the determination of the optimal number
of clusters is discussed below). Hierarchical agglomerative techniques are
more commonly used.
An agglomerative hierarchical clustering procedure produces a series of
partitions of the data, P
n
, P
n1
, . . . . . . . , P
1
. The rst P
n
consists of n sin-
gle object clusters, the last P
1
, consists of a single group containing all n
cases. At each particular stage the method joins together the two clusters
that are closest together (most similar). (At the rst stage, of course, this
amounts to joining together the two objects that are closest together, since
at the initial stage each cluster has one object.) Differences between meth-
ods arise because of the different ways of dening distance (or similarity)
between clusters. Some important agglomerative techniques include single
linkage clustering, complete linkage clustering, average linkage clustering,
average group linkage, and Wards hierarchical clustering method.
19
I use
Wards hierarchical clustering method in this paper, which is also the most
commonly used method (see Everitt [1993]). It produces spherical clusters
that are roughly the same size. The aim is to join objects together into ever
increasing sizes of clusters using a measure of similarity of distance.
THE OPTIMAL NUMBER OF CLUSTERS
An important issue in the CA procedure is determining the optimal num-
ber of clusters. Unfortunately, there are no completely satisfactory methods
for determining the number of population clusters for any type of cluster
analysis (Everitt [1979, 1980], Hartigan [1985]). If the purpose in clustering
is dissection, that is, to summarize the data without trying to uncover real
clusters, it may sufce to look at the R
2
for each variable and pooled over all
variables. Plots of R
2
against the number of clusters are also useful. Ordinary
signicance tests, such as analysis of variance F -tests, are not valid for testing
differences between clusters. Since clustering methods attempt to maximize
the separation between clusters, the assumptions of the usual signicance
tests, parametric or nonparametric, are drastically violated. For the same
reason, methods that purport to test for clusters against the null hypothesis
that objects are assigned randomly to clusters (Klastorin [1983]) are useless.
Most valid tests for clusters either have intractable sampling distributions or
involve null hypotheses for which rejection is uninformative.
For clustering methods based on distance matrices, a popular null hy-
pothesis is that all permutations of the values in the distance matrix are
equally likely (Ling [1973]). Using this null hypothesis, one can do a per-
mutation test or a rank test. The trouble with the permutation hypothesis
19
Other possible hierarchical methods are described in Rohlf [1970], Lukasova [1979], and
Hansen and Tukey [1992].
is that with any real data, the null hypothesis is implausible even if the data
do not contain clusters. Rejecting the null hypothesis does not provide any
useful information (Hubert and Baker [1977]). Another common null hy-
pothesis is that the data are a random sample from a multivariate normal
distribution (Wolfe [1978], Duda and Hart [1973]). The multivariate nor-
mal null hypothesis is better than the permutation null hypothesis, but it is
not satisfactory because there is typically a high probability of rejection if
the data are sampled from a distribution with lower kurtosis than a normal
distribution, such as a uniform distribution. Hawkins, Muller, and Krooden
[1982] discuss a highly conservative Bonferroni method for hypothesis test-
ing. The conservativeness of this approach may compensate to some extent
for the liberalness exhibited by tests based on normal distributions when
the population is uniform.
A number of more formal techniques have also been suggested. Sarle
[1983] uses extensive simulations to develop the cubic clustering criterion
(CCC), which can be used for crude hypothesis testing and estimating the
number of population clusters. The CCC is based on the assumption that
a uniform distribution on a hyperrectangle will be divided into clusters
shaped roughly like hypercubes. In large samples that can be divided into
the appropriate number of hypercubes, this assumption gives very accurate
results. In other cases the approximation is generally conservative.
Beale [1969] gives an F -test, which may be used to test whether a sub-
division into c
2
clusters is signicantly better than a subdivision into some
smaller number of clusters, c
1
. The test statistic is dened as follows:
F (c
1
, c
2
) =
R
c
1
R
c
2
R
c
2
_
_
_
n c
1
n c
2
__
c
2
c
1
_
2/p
1
_
where R
c
= (n c)S
2
c
, and S
2
c
is the mean square deviation from cluster
centers in the sample, n is the number of objects, and p is the number of
variables. This statistic is compared with F , with p(c
2
c
1
) and p(n c
2
)
degrees of freedom. A signicant result indicates that a subdivision into c
2
clusters is an improvement over a subdivision into the smaller number, c
1
.
This procedure is more useful when the clusters are fairly well separated
and approximately spherical in shape.
Another method suggested by Calinski and Harabasz [1974] is to take the
value of c that corresponds to the maximum value of C, where C is given by:
C =
trace(B)
c 1
_
trace(W)
n c
where B and W are the between and within cluster sum of squares
and cross-product matrices, i.e., B =
c
i =1
n
i
( x
i
x)( x
i
x)
and W =
1
nc
c
i =1
n
i
j =1
(x
i j
x
i
)(x
i j
x
i
)
, and c is the number of clusters. This

statistic is sometimes referred to as the pseudo-F statistic.
Marriott [1971] suggests another possible procedure for assessing the
number of clusters, where the value of c for which c
2
det(W) is minimum
1178 A. DEY
is taken. For unimodal distributions, Marriott [1971] shows that this pro-
cess is likely to lead to accepting the value of c such that c = 1, and for
strongly groupeddata it will leadtothe appropriate value for c. The sampling
properties of the associated statistic c
2
det(W)
det(T)
, where T =
1
n
c
i =1
n
i
j =1
(x
i j

x)(x
i j
x)
, under the assumption that that the population has a uniform

distribution, are investigated by Monte-Carlo methods, and the use of these
results may be helpful in deciding the appropriate number of clusters for
the data.
Duda and Hart [1973] propose a ratio criterion,
J e (2)
J e (1)
, where J e (2) is
the sum of squared errors within clusters when the data are partitioned into
two clusters, and J e (1) gives the squared errors when only one cluster is
present. The hypothesis of a single cluster is rejected if the ratio is smaller
than a specied critical value.
Cooper andMilligan[1985] compare 30 methods for estimating the num-
ber of population clusters using four hierarchical clustering methods. The
three criteria that perform best in these simulation studies are the pseudo-
F -statistic developed by Calinski and Harabasz [1974], the
J e (2)
J e (1)
statistic
developed by Duda and Hart [1973] that can be transformed into a pseudo-
t
2
-statistic, and the cubic clustering criterion, CCC. The best approach is to
look for consensus among the three statistics, that is, local peaks of the CCC
and pseudo-F statistic combined with a small value of the pseudo-t
2
statistic
anda larger pseudo-t
2
for the next cluster fusion. It must be emphasizedthat
these criteria are appropriate only for compact or slightly elongated clus-
ters, preferably clusters that are roughly multivariate normal. In this study I
use these three statistics, and determine the optimal number of clusters for
the sample by looking for a consensus among the three statistics.
REFERENCES
AGRAWAL, A., AND C. KNOEBER. Firm Performance and Mechanisms to Control Agency Prob-
lems Between Managers and Shareholders. Journal of Financial and Quantitative Analysis 31
(1996): 37797.
ANDERBERG, M. R. Clustering Analysis for Applications. New York, NY: Academic Press, 1973.
ASHBAUGH, H.; R. LAFOND; AND B. MAYHEW. Do Non-Audit Services Compromise Auditor
Independence? Further Evidence. The Accounting Review 78 (2003): 61139.
BARNHART, S., AND S. ROSENSTEIN. Board Composition, Managerial Ownership, and Firm
Performance: An Empirical Analysis. The Financial Review 33 (1998): 116.
BAYSINGER, B., AND H. BUTLER. Corporate Governance and Board of Directors: Performance
Effects of Changes inBoardComposition. Journal of Law, Economics andOrganization 1(1985):
10124.
BEALE, E. Euclidean Cluster Analysis. Bulletin of the International Statistical Institute 43 (1969):
924.
BERLE, A., AND G. MEANS. The Modern Corporation and Private Property. NewYork: Harcourt, Brace
& World, 1932.
BHAGAT, S., AND B. BLACK. The Uncertain Relationship between Board Composition and Firm
Performance. The Business Lawyer 54 (1999): 92163.
BHAGAT, S., AND B. BLACK. The Non-correlation Between Board Independence and Long-
Term Firm Performance. Journal of Corporation Law 27 (2001): 23174.
BHAGAT, S.; D. CAREY; AND C. ELSON. Director Ownership, Corporate Performance, and Man-
agement Turnover. The Business Lawyer 54 (1999): 885920.
BOWEN, R.; S. RAJGOPAL; AND M. VENKATACHALAM. Accounting Discretion, Corporate Gover-
nance and Firm Performance. Working paper, University of Washington and Duke Univer-
sity, 2005. Web site, http://papers.ssrn.com/sol3/papers.cfm?abstract id=367940.
BROWN, L., AND M. CAYLOR. Corporate Governance and Firm Performance. Work-
ing paper, Georgia State University and University of South Carolina, 2004. Web site,
http://papers.ssrn.com/sol3/papers.cfm?abstract id=586423.
BUSHMAN, R.; Q. CHEN; E. ENGEL; AND A. SMITH. Financial Accounting Information, Organi-
zational Complexity and Corporate Governance Systems. Journal of Accounting & Economics
37 (2004): 167201.
CALINSKI, T., AND J. HARABASZ. A Dendrite Method for Cluster Analysis. Communications in
Statistics 3 (1974): 127.
CHENG, Q., AND T. WARFIELD. Equity Incentives and Earnings Management. The Accounting
Review 80 (2005): 44176.
COASE, R. The Nature of the Firm. Economica 4 (1937): 386405.
COOPER, M., AND G. MILLIGAN. An Examination of Procedures for Determining the Number
of Clusters in a Data Set. Psychometrika 50 (1985): 15979.
COOPER, M., ANDG. MILLIGAN. The Effect of Measurement Error onDetermining the Number
of Clusters in Cluster Analysis. In W. Gaul and M. Schader (eds.), Data Analysis, Decision
Support and Expert Knowledge Representation in Marketing and Related Areas of Research. Berlin:
Springer-Verlag, 1988: 31928.
DEFOND, M., AND J. JIAMBALVO. Debt Covenant Effects and the Manipulation of Accruals.
Journal of Accounting & Economics 17 (1994): 14576.
DEMSETZ, H., AND K. LEHN. The Structure of Corporate Ownership: Causes and Conse-
quences. Journal of Political Economy 93 (1985): 115577.
DUDA, R., AND P. HART. Pattern Classication and Scene Analysis. New York: John Wiley and Sons,
Inc., 1973.
DURNEV, A., AND E. KIM. To Steal or Not to Steal: Firm Attributes, Legal Environment and
Valuation. Working Paper, McGill University and University of Michigan, 2003. Web site,
EVERITT, B. Unresolved Problems in Cluster Analysis. Biometrics 35 (1979): 16981.
EVERITT, B. Cluster Analysis. London, UK: Heinemann Educational Books, 1980.
EVERITT, B. Cluster Analysis. London, UK: Edward Arnold, 1993.
FAMA, E., AND M. JENSEN. Separation of Ownership and Control. Journal of Law and Economics
26 (1983a): 30125.
FAMA, E., AND M. JENSEN. Agency Problems and Residual Claims. Journal of Law and Economics
26 (1983b): 32749.
FELO, A.; S. KRISHNAMURTHY; AND S. SOLIERI. Audit Committee Characteristics and the Per-
ceived Quality of Financial Reporting: An Empirical Analysis. Working paper, Pennsylvania
State University, SUNY - Binghamton University and University of Scranton, 2003. Web site,
FRANKEL, R.; M. JOHNSON; AND K. NELSON. The Relation Between Auditors Fees for Non-audit
Services and Earnings Management. The Accounting Review 77 (2002): 71105.
GILLAN, S.; J. HARTZELL; AND L. STARKS. Explaining Corporate Governance: Boards, Bylaws
and Charter Provisions. Working paper, University of Texas at Austin, 2003. Web site,
GILSON, S.; P. HEALY; C. NOE; AND K. PALEPU. Analyst Specialization and Conglomerate
Breakups. Journal of Accounting Research 39 (2001): 56582.
GIVOLY, D.; C. HAYN; AND J. DSOUZA. Measurement Errors and Information Content of Seg-
ment Reporting. Review of Accounting Studies 4 (1999): 513.
GOMPERS, P.; J. ISHII; AND A. METRICK. Corporate Governance and Equity Prices. Quarterly
Journal of Economics 118 (2003): 10755.
HABIB, M.; D. JOHNSON; AND N. NAIK. Spinoffs and Information. Journal of Financial Interme-
diation 6 (1997): 15376.
1180 A. DEY
HANSEN, K., AND J. TUKEY. Tuning a Major Part of a Clustering Algorithm. International Statis-
tical Review 60 (1992): 2144.
HART, O. Firms, Contracts and Financial Structure. London, UK: Oxford University Press, 1995.
HARTIGAN, P. Computation of the Dip Statistic to Test for Unimodality. Applied Statistics 3
(1985): 3205.
HAWKINS, D.; M. MULLER; AND J. KROODEN. Cluster Analysis. In Topics in Applied Multivariate
Analysis. Cambridge, England: Cambridge University Press, 1982: 30356.
HERMALIN, B., AND M. WEISBACH. The Determinants of Board Composition. Rand Journal of
Economics 19 (1988): 589606.
HERMALIN, B., AND M. WEISBACH. The Effects of Board Composition and Director Incentives
on Firm Performance. Financial Management 20 (1991): 10112.
HERMALIN, B., AND M. WEISBACH. Boards of Directors as an Endogenously Determined Insti-
tution: A Survey of the Economic Literature. Economic Policy Review 9 (2003): 726.
HUBERT, L., AND F. BAKER. An Empirical Comparison of Baseline Models for Goodness-of-Fit
in r -Diameter Hierarchical Clustering. J. Van Ryzin (ed.), In Classication and Clustering.
New York: Academic Press, Inc., 1977.
JENSEN. 1976.
JENSEN, M. Agency Costs of Free Cash Flow, Corporate Finance and Takeovers. American
Economic Review 76 (1986): 3239.
JENSEN, M. The Modern Industrial Revolution, Exit, and the Failure of Internal Control Sys-
tems. Journal of Finance 48 (1993): 83180.
JENSEN, M., AND W. MECKLING. Theory of the Firm: Managerial Behavior, Agency Costs and
Ownership Structure. Journal of Financial Economics 3 (1976): 30560.
JOLLIFFE, I. Principal Components Analysis. Springer Series in Statistics, Second edition. New York:
Springer, 2002.
JOHN, K., AND L. SENBET. Corporate Governance and Board Effectiveness. Journal of Banking
and Finance 22 (1998): 371403.
KLAPPER, L., AND I. LOVE. Corporate Governance, Investor Protection, and Performance in
Emerging Markets. Journal of Corporate Finance 10 (2004): 70328.
KLASTORIN, T. Assessing Cluster Analysis Results. Journal of Marketing Research 20 (1983): 928.
KLEIN, A. Firm Performance and Board Committee Structure. Journal of Law and Economics
41 (1998): 275303.
KLEIN, A. Economic Determinants of Audit Committee Independence. The Accounting Review
77 (2002a): 43552.
KLEIN, A. Audit Committee, Board of Director Characteristics, and Earnings Management.
Journal of Accounting & Economics 33 (2002b): 375400.
LARCKER, D., AND S. RICHARDSON. Fees Paid to Audit Firms, Accrual Choices, and Corporate
Governance. Journal of Accounting Research 42 (2004): 62558.
LARCKER, D.; S. RICHARDSON; AND I. TUNA. Corporate Governance, Accounting Outcomes and
Organizational Performance. The Accounting Review 82 (2007): 9631008.
LASFER, M. Board Structure and Agency Costs. Working paper, City University, 2002. Web
site, http://papers.ssrn.com/sol3/papers.cfm?abstract id=314619.
LAZEAR, E. Personnel Economics. Cambridge, MA: MIT Press, 1995.
LENNOX, C. Audit Quality and Executive Ofcers Afliations with CPA Firms. Journal of
Accounting & Economics 39 (2005): 20131.
LENNOX, C., ANDC. PARK. Audit FirmAppointments, Audit FirmAlumni, andAudit Committee
Independence. Contemporary Accounting Research 24 (2007): 23558.
LING, R. A Probability Theory of Cluster Analysis. Journal of the American Statistical Association
68 (1973): 15969.
LIPTON, M., AND J. LORSCH. A Modest Proposal for Improved Corporate Governance. The
Business Lawyer 59 (1992): 5977.
LUKASOVA, A. Hierarchical Agglomerative Clustering Procedure. Pattern Recognition 11
(1979): 36581.
MARRIOTT, F. Practical Problems ina Method of Cluster Analysis. Biometrics 27 (1971): 50114.
MASSART, D., AND L. KAUFMAN. The Interpretation of Analytical Chemical Data by the Use of Cluster
Analysis. New York: John Wiley & Sons, 1983.
MEHRAN, H. Executive Incentive Plans, Corporate Control and Capital Structure. Journal of
Financial and Quantitative Analysis 27 (1992): 53960.
MEHRAN, H. Executive Compensation Structure, Ownership and Firm Performance. Journal
of Financial Economics 38 (1995): 16384.
MENON, K., AND D. WILLIAMS. Former Audit Partners and Abnormal Accruals. The Accounting
Review 79 (2004): 1095118.
MORCK, R.; A. SHLEIFER; AND R. VISHNY. Management Ownership and Market Valuation: An
Empirical Analysis. Journal of Financial Economics 20 (1988): 293315.
MYERS, S. Determinants of Corporate Borrowing. Journal of Financial Economics 5 (1977):
14775.
NUNNALLY, J. Psychometric Theory. New York: McGraw Hill, 1967.
RIBSTEIN, L. Sizing Up SOX. Working paper, University of Illinois College of Law, 2005.
ROHLF, F. Adaptive Hierarchical Clustering Schemes. Systematic Zoology (1970): 5882.
ROMANO, R. The Sarbanes Oxley Act and the Making of Quack Corporate Governance. Yale
Law Journal May (2005): 1521611.
ROSENSTEIN, S., AND J. WYATT. Outside Directors: Board Independence and Shareholder
Wealth. Journal of Financial Economics 26 (1990): 17591.
RUMMEL, R. Applied Factor Analysis. Evanston, IL: Northwestern University Press, 1970.
SARLE, W. Cubic Clustering Criterion, SAS Technical Report A-108. Cary, NC: SAS Institute Inc.,
1983.
SHLEIFER, A., AND R. VISHNY. A Survey of Corporate Governance. Journal of Finance 52 (1997):
73783.
SKINNER, D., AND R. SLOAN. Earnings Surprises, Growth Expectations, and Stock Returns or
Dont Let an Earnings Torpedo Sink Your Portfolio. Review of Accounting Studies 7 (2002):
287312.
SMITH, C., AND R. WATTS. The Investment Opportunity Set and Corporate Financing, Divi-
dend, and Compensation Policies. Journal of Financial Economics 32 (1992): 26392.
STEIN, J. Information Production and Capital Allocation: Decentralized Versus Hierarchi-
cal Firms. Working paper, Harvard University, Department of Economics, 2000. Web site,
TEOH, S., AND T. WONG. Perceived Auditor Quality and the Earnings Response Coefcient.
The Accounting Review 68 (1993): 34666.
VAFEAS, N. Board Structure and the Informativeness of Earnings. Journal of Accounting and
Public Policy 19 (2000): 13960.
WARFIELD, T.; J. WILD; AND K. WILD. Managerial Ownership and the Informativeness of Earn-
ings. Journal of Accounting & Economics 20 (1995): 6192.
WATTS, R., AND J. ZIMMERMAN. Positive Accounting Theory. Englewood Cliffs, NJ: Prentice Hall,
1986.
WATTS, R., AND J. ZIMMERMAN. Positive Accounting Theory: A Ten Year Perspective. The
Accounting Review 65 (1990): 13156.
WEISBACH, M. Outside Directors and CEO Turnover. Journal of Financial Economics 20 (1988):
43160.
WOLFE, J. Comparative Cluster Analysis of Patterns of Vocational Interest. Multivariate Behav-
ioral Research 13 (1978): 3344.
YERMACK, D. Higher Market Valuation of Companies With a Small Board of Directors. Journal
of Financial Economics 40 (1996): 185202.

Dey (2008)

Uploaded by

Copyright:

Available Formats

Dey (2008)

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Dey (2008)

Uploaded by

Copyright:

Available Formats

DOI: 10.1111/j.1475-679X.2008.00301.

Received 30 May 2006; accepted 3 March 2008

Graduate School of Business, University of Chicago. This paper is based on my dissertation

0.0250 0.0202 0.5563

0.0074 0.0205 0.0153 0.1697

0.1276 0.0347 0.0257 0.3796

indicates signicance at the 1% level.

5.7978 5.8119 7.1426 7.0779 8.7658 8.9484

0.5332 0.4190 0.7510 0.5543 0.8435 0.6634

0.2986 0.5901 1.2706 0.7638 1.6385 1.3964

0.0246 0.0209 0.0450 0.0416 0.0664 0.0597

LEV : t = 1.44; z = 2.33

Board II: z = 2.52

Exec Comp: z = 3.58

Dir Comp: z = 0.51

Fin Rep: z = 3.88

Board II: z = 2.24

Exec Comp: z = 7.33

Dir Comp: z = 1.19

Audit Comm: z = 7.58

Fin Rep: z = 3.01

Board II: z = 2.21

Exec Comp: z = 4.92

Dir Comp: z = 0.47

Audit Comm: z = 8.24

Fin Rep: z = 1.37

, and c is the number of clusters. This

, under the assumption that that the population has a uniform

You might also like