3i .. MM. . a“.
. A ‘L
"#94 a. 39..
1. V I.

nilm

.905...

:2...
2:-

22.... .. 1.5L
:32? ....a
fa}.

5.94

do

 

.1232: ..
.57 . x
1 glit,

2’
s . 9
1 tr. 4?»...

. ~
.51- {7‘an)
.I. Ell-p 3| «1
5| italylvlﬁy‘lan

4. I24- \Jx\!
21 Tit...‘
\-

, 2
, , . :1... 4o:
.1115; 1 . 27:. 3.2.7

 

 

 

"le

l
900’

 

‘ LIBRARY
michtgan Sta
University

w

 

 

This is to certify that the

dissertation entitled

Differences Across Audit Firm Types in Assessments
of Non-Profit Organizations' Federal Compliance
presented by

Stefanie Lara Tate

has been accepted towards fulﬁllment
of the requirements for

th . degree in Jamming

 

Mum WM

jor professor

Date 7/)3/01

I I

MS U i: an Afﬁrmative Action/Equal Opportunity Institution 0-12771

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE

DATE DUE

DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 c:lClRCJDateDue.p65-p15

 

 

DIFFERENCES ACROSS AUDIT FIRM TYPES IN ASSESSMENTS OF NON-
PROFIT ORGANIZATIONS FEDERAL COMPLIANCE
By

Stefanie Lara Tate

A DISSERTATION
Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
Department of Accounting

2001

ABSTRACT
DIFFERENCES ACROSS AUDIT FIRM TYPES IN ASSESSMENTS OF NON-
PROFIT ORGANIZATIONS' FEDERAL COMPLIANCE
By
Stefanie L. Tate
This paper investigates whether different types of auditors provide different

assessments of noncompliance with federal regulations and deﬁciencies in internal
control structures in a ﬁnancial statement and compliance audit setting. Audit ﬁrms are
separated into three types - market leaders, non-market leader brand name, and nOn-brand
name - with the expectation that market leaders provide the best assessments, and non-
market leader brand name ﬁrms provide better assessments than non-brand name ﬁrms.
While prior studies consistently indicate that users of ﬁnancial statements and companies
hiring auditors believe there are quality differences between auditors, there is little
research on how these audit quality differences translate into speciﬁc differences in the
auditors’ assessments and judgments. Using a database of almost 77,000 compliance and
ﬁnancial statement audit results for non-proﬁt organizations for the ﬁscal years 1997
through 2000, I ﬁnd market leader and non-market leader brand name auditors report
more noncompliance with federal regulations in the form of questioned costs and ﬁndings
than non-brand name auditors, consistent with audit quality theory. I also ﬁnd that the
market leader brand name auditor reports more ﬁndings than non-market leader brand
name auditors. However, in contrast to expectations, and after controlling for the number
and extent of errors identiﬁed by the auditor, I ﬁnd the market leader brand name auditor

is marginally less likely (rather than more likely) than non-brand name auditors to qualify

their report on an organization’s compliance with federal regulations. I ﬁnd no other
differences between auditor types in their likelihood of qualifying their opinion on
compliance with federal regulations. Also, in contrast to expectations, I ﬁnd the market
leader brand name auditor is marginally less likely to report deﬁciencies in internal
controls than non-market leader brand name auditors who are signiﬁcantly less likely to
report internal control deﬁciencies than non-brand name auditors. Overall, these results

provide mixed results on the quality effects of market leader and brand name auditors.

Dedicated to my parents, Jack and Sarah, for their unending love and support.

iv

ACKNOWLEDGEMENTS

I would like to express my deepest gratitude to the members of my dissertation
committee - Dr. Kathy Petroni (chairperson), Dr. Joseph Anthony, Dr. Alvin Arens, and
Dr. Jeffrey Wooldridge - for all of their time, effort, and most importantly, their
invaluable guidance throughout this process. I would also like to thank the other faculty
members in the Department of Accounting who help provide a positive and open
environment in which doctoral students can thrive and grow.

I would like to thank all of the doctoral students, and a special thank you to Annie
Farrell who was always there to discuss ideas, provide guidance, lean on for support, and
most importantly, to celebrate the successes.

Finally, I would like to thank my parents and my brother for giving me the
courage, the encouragement, and the constant support that has helped to make this dream

come true.

TABLE OF CONTENTS

LIST OF TABLES ............................................................................ vii
CHAPTER 1

INTRODUCTION ............................................................................ 1
CHAPTER 2

THEORY DEVELOPMENT ................................................................ 6
CHAPTER 3

A-133 AUDIT BACKGROUND ........................................................... 11
CHAPTER 4

HYPOTHESES ................................................................................ 14
CHAPTER 5 -
METHODOLOGY ............................................................................ 17
CHAPTER 6

RESULTS AND SUMMARY .............................................................. 25
REFERENCES ................................................................................ 44

vi

Table

10

LIST OF TABLES

Descriptive Statistics - All Data ................................................. 47
Descriptive Statistics - Comparisons Between Auditor Types ............... 49

Descriptive Statistics - Comparisons Between Changes in Auditor Types 51

Pearson Correlations ............................................................... 54
Hypothesis 1 - Questioned Costs ................................................ 55
Hypothesis 2 - Findings ........................................................... 57
Hypothesis 3 - Material Noncompliance with Federal Regulations ........ 59
Hypothesis 4 - Reportable Conditions in Internal Controls ................. 61
Tobit Estimation of Hypotheses One and Two ................................ 63
Logit Estimation of Hypotheses Three and Four .............................. 65

vii

Chapter One
INTRODUCTION

DeAngelo (1981) hypothesizes that audit ﬁrms differentiate themselves based on
quality and that larger audit ﬁrms are expected to provide higher quality audits. Much of
the audit quality literature indicates that quality differences are inferred by both users of
ﬁnancial statements and companies selecting an auditor. These studies have used the
audit ﬁrm type - specialist/non-specialist or brand name/non-brand name - as a proxy for
quality. However, few studies directly test for quality differences between auditor types,
and the results from those studies have been mixed.

This study will test whether there are quality differences in the assessments made
by different types of auditors. Consistent with the prior literature, I separate auditors into
three groups - market leader brand name, non-market leader brand name, and non-brand

name1 - and test for differences in the assessments made by these three types of auditors.

1.1 Motivation

Theory predicts that there are quality differences in auditors, and users have
inferred these differences consistent with the theory. However, without understanding
how those quality differences present themselves in either the auditors' reports or the
related ﬁnancial statements, the true beneﬁts of hiring different types of auditors cannot

be determined. Prior research indicates that brand name auditors reduce management's

 

' Brand name auditors consist of the ﬁve largest auditors in the United States which are commonly referred
to as the Big Five. The market leader auditor is the auditor that audits the largest percentage of total federal
expenditures in this data set.

ability to manipulate earnings, as measured by discretionary accruals or differences
between forecasted and audited earnings (see Becker, et. a1. [1998], Francis, et. al. [1999],
Davidson and Neu [1993], and Clarkson [2000]). Concerns regarding the reliability of
these indirect measures of earnings management (discretionary accruals as estimated by
the Jones model or forecast errors) limit the interpretability of the results. In addition,
outside of earnings management, there has been little research on differences in auditors'
other assessments and judgments. Without a clearer understanding of the actual
differences in more diverse auditor assessments and judgments, clients, users of ﬁnancial
statements, and regulators cannot properly assess what effect audit quality should have on
their decisions.

This study uses a database maintained by the United States Federal Clearinghouse
of the results of ﬁnancial statement and compliance audits performed in accordance with
the Ofﬁce of Management and Budget’s A-133 requirements (A-l 33 audit). These audits
are required for all non-proﬁt organizations receiving signiﬁcant funding from federal
agencies, and require the auditor to issue an opinion on the organization’s ﬁnancial
statements, schedule of federal awards, and compliance with federal requirements, in
addition to providing a report on the organization’s internal control structure and
identiﬁed questioned costs and ﬁndings. Therefore, in addition to the audit opinion on
the ﬁnancial statements, this database allows an investigation of other decisions made by
the auditor including an assessment of the organization’s internal control structure,
compliance with federal regulations, and compliance with speciﬁc grant requirements.

The majority of organizations in the database are not audited by large, brand name

auditors, making this a very different setting from many prior audit quality studies, where

over 90% of the organizations (publicly traded companies) are audited by brand name
auditors. By investigating the effects of different types of auditors on the reporting of
questioned costs, I am also able to estimate possible ﬁnancial implications to an
organization of hiring different types of auditors. In addition, because the database
includes the same organizations over time, unobservable organizational factors that stay
constant over time can be controlled for, including the speciﬁc qualities of the client that

may inﬂuence auditor selection and which have plagued prior studies.

1.2 Overview of Hypotheses

Non-proﬁt organizations that receive federal funds are required to maintain
compliance with all federal regulations stipulated within their federal grant documents.
Auditors of these federal grants must perform tests to determine whether the organization
has maintained compliance with the regulations, and must document all instances of
noncompliance as well as express an opinion on whether the organization has maintained
material compliance with the regulations.

Using DeAngelo’s (1981) deﬁnition of audit quality - the joint probability that an
auditor will both discover an error and report that error - I hypothesize that a higher
quality auditor will identify and report more noncompliance with federal regulations, and
will identify and report more deﬁciencies in an organization’s internal control structure.
DeAngelo (1981) hypothesizes that larger audit ﬁrms provide higher audit quality. In
addition, the prior literature has theorized that specialist ﬁrms will provide higher quality
than non-specialist ﬁrms as they are more likely to invest signiﬁcant resources in

becoming a specialist. The prior literature has used market share as one measure of

specialists. I use this same measure and identify one brand name audit ﬁrm that has a
market share signiﬁcantly higher than any other audit ﬁrm in the sample, and call this
ﬁrm the market leader. I test the effect of both audit ﬁrm size and extent of specialization
together by investigating whether the market leader brand name auditor reports more
noncompliance with federal regulations and more deﬁciencies in internal controls than
non-market leader brand name auditors, and whether non-market leader brand name
auditors report more noncompliance with federal regulations and more deﬁciencies in

internal controls than non-brand name auditors.

1.3 Overview of Research Design

Four hypotheses are tested to determine the effects of the three different types of
auditors (market leader brand name - further referred to as market leader, non-market
leader brand name - further referred to as other brand name, and non-brand name
auditors) on the levels of reported questioned costs, ﬁndings, material noncompliance
with federal regulations, and reportable conditions. The four models are estimated using
multivariate ﬁxed effects and ﬁrst differencing techniques, and additional alternate tests
are performed to ensure the robustness of the results.

The hypotheses are tested using almost 77,000 observations from 35,850 different
non-proﬁt organizations over a four-year period from 1997-2000. All of the data for the
main tests of the hypotheses was obtained from the United States Federal Clearinghouse’s
website. Additional ﬁnancial data was obtained from the non-proﬁt organizations’

Internal Revenue Service’s Form 990’s maintained on Guidestar’s website.

1.4 Overview of Results

Overall, the empirical results are mixed. I ﬁnd the market leader and other brand
name auditors report more errors in the form of questioned costs and ﬁndings than non-
brand name auditors. I also ﬁnd the market leader auditor reports more questioned costs
and ﬁndings than other brand name auditors in a levels but not changes model. However,
in contrast to expectations, and after controlling for the number and extent of errors
identiﬁed by the auditor, I ﬁnd the market leader auditor is marginally less likely (rather
than more likely) than non-brand name auditors to qualify their report on an
organization’s compliance with federal regulations. I ﬁnd no difference between market
leader and other brand name auditors, or between other brand name and non-brand name
auditors, in their likelihood of qualifying their opinion on compliance with federal
regulations. Also, in contrast to expectations, I ﬁnd the market leader other brand name
auditors are less likely to report deﬁciencies in internal controls than non-brand name
auditors, and the market leader auditor is marginally less likely to report internal control
deﬁciencies than other brand name auditors.

The remaining chapters are laid out as follows: Chapter 2 includes a discussion of
the prior literature, Chapter 3 provides detailed background on the A-l33 audits which
are the basis for the dataset, and Chapter 4 presents the hypotheses for the study. Chapter
5 includes a discussion of the methodology used in the paper and the paper concludes

with the results in Chapter 6.

Chapter Two
THEORY DEVELOPMENT
2.] Size as Quality Measure

DeAngelo (1981) deﬁnes audit quality as the “market-assessed joint probability
that a given auditor will both (a) discover a breach in the client’s accounting system, and
(b) report the breach,” (p. 186). DeAngelo hypothesizes that larger audit ﬁrms have more
ability to specialize and innovate through technology, thereby increasing the likelihood
that they will discover a breach in the accounting system. She also contends that larger
audit ﬁrms are more independent from their clients, thereby increasing the likelihood that
they will report an identiﬁed breach. Using these assumptions, DeAngelo hypothesizes
that larger ﬁrms provide higher quality audits.

Many studies following DeAngelo (1981) assume DeAngelo's hypothesis that
larger ﬁrms provide higher quality audits. These studies investigate both the relationship
between audit ﬁrm size and organizational (client) factors, and the effects of audit ﬁrm
size on the decisions of users of ﬁnancial statements. The results are generally consistent
with the theory that organizations select larger audit ﬁrms when agency costs are high (for
example, see DeFond [1992], Firth and Smith [1992], and Simunic and Stein [l987]2).
Studies on risk and leverage indicate that higher risk ﬁrms tend to select larger audit ﬁrms
(see Datar, F eltharn and Hughes [1991], Firth and Smith [1992], Clarkson and Simunic

[1994], Copley, Gaver, and Gaver [1995], Firth and Liau-Tan [1998], and Lee, et. a1.

 

2 Beasley and Petroni (2001) investigate the relationship between board composition and audit quality.
They include variables to control for agency costs, and ﬁnd results on these control variables consistent
with the theory that ﬁrms with higher agency costs select higher quality auditors.

[1999])3. There are also generally consistent results which indicate that ﬁnancial
statements audited by larger audit ﬁrms are relied upon more heavily by decision makers
than ﬁnancial statements audited by smaller audit ﬁrms (for example, see Allen [1994],
Balvers, et. a1. [1988], Beatty [1989], and Teoh and Wong [1993]).

Two previous studies investigate the effect of brand name on an organization’s
audit fee and estimate the fee premium paid to the brand name auditors. Craswell, et. al.
(1995) ﬁnd that Australian publicly traded companies are willing to pay approximately
30% more to obtain a brand name auditor and Beattie, et. al. (2000) in a current working
paper ﬁnd English fund raising charities are willing to pay a fee premium averaging
18.5% for a brand name auditor. Assuming that organizations will pay more for a higher
quality audit, both of these studies provide results that are consistent with DeAngelo’s
hypothesis that larger audit ﬁrms provide higher quality audits.

A small number of studies directly test DeAngelo's theory with mixed results,
depending on how quality is actually measured. Two studies using an external evaluation
of compliance with speciﬁc auditing standards as their measure of quality ﬁnd results
consistent with DeAngelo. Among smaller, non-Big Six, audit ﬁrms, Colbert and Murray
(1998) ﬁnd larger ﬁrms receive more unqualiﬁed AICPA Private Companies Practice
Section Peer Review reports than smaller ﬁrms. In the non-proﬁt setting, O’Keefe, et. al.

(1994) ﬁnd that Big Eight audit ﬁrms have fewer violations of generally accepted

 

3 Titrnan and Trueman (1986) and Thornton and Moore ( 1993) both provide analytical models that suggest
higher risk ﬁrms would select lower quality auditors, as opposed to higher quality auditors. Simunic and
Stein (1987) ﬁnd empirical results consistent with this hypothesis, and F eltham, Hughes and Simunic (1991)
ﬁnd no consistent signiﬁcant results. However, these earlier studies do not control for audit fee, or audit
supply, which can have a signiﬁcant impact on the results. The later studies which ﬁnd a positive
relationship control for the audit supply effect.

auditing standards on California school district audits as determined by California State
Controller reviews.

When qualities of the actual ﬁnancial statements are used as a measure of quality,
the results become less consistent. Krishnan and Schauer (2000) investigate the ﬁnancial
statement disclosures of voluntary health and welfare non-proﬁt organizations. Using the
Statements of Financial Accounting Standards Nos. 93 and 99, and the AICPA guide
Audits of Voluntary Health and Welfare Organizations, they determine what ﬁnancial
statement disclosures are required. These requirements include disclosures on accounts
such as investments, depreciation, pledges, and donated services. They review individual
ﬁnancial statements and calculate the number of correct required disclosures made. A
disclosure is coded as "correct" if it is required and it has been properly disclosed. A
disclosure is also coded as "correct" if the disclosure was not required, and it was
properly not disclosed. For instance, if the organization does not have investments, they
do not need an investment disclosure. Krishnan and Schauer ﬁnd that the ﬁnancial
statements of voluntary health and welfare non-proﬁt organizations audited by larger
audit ﬁrms have more correct ﬁnancial statement disclosures than those audited by
smaller audit ﬁrms.

Becker, et. al. (1998), Francis, et. al. (1999), and Davidson and Neu (1993) ﬁnd
that larger audit ﬁrms are associated with a reduction in the extent of earnings
management in the ﬁnancial statements4. These three studies use indirect measures of
earnings management as measures of ﬁnancial statement quality. Becker, et. al. and

Francis, et. al. use the level of discretionary accruals as estimated from a modiﬁed Jones

model and Davidson and Neu use the difference between audited and forecasted earnings.
However, when more direct measures of ﬁnancial statement quality are used, the results
are not consistent. Clarkson (2000) investigates the relationship between audit ﬁrm type
and ﬁnancial forecast accurancy. Forecast accuracy is measured as the absolute
difference between the forecasted earnings disclosed in an IPO prospectus and the actual
realized earnings for the period forecasted. Clarkson ﬁnds results consistent with larger
audit ﬁrms being associated with more accurate ﬁnancial forecasts than smaller audit
ﬁrms, although Firth and Smith (1992) do not. In addition, Petroni and Beasley (1996)
are unable to ﬁnd any differences in the estimation errors of claim loss reserves of
property casualty insurers between ﬁnancial statements audited by Big Eight and non-Big

Eight audit ﬁrms.

2.2 Market Share as Quality Measure

Audit ﬁrms may choose to specialize in a speciﬁc industry, and would be
expected to provide a higher quality audit than a ﬁrm which has not specialized in that
industry. A number of studies have investigated the relationship between specialization
and client factors. DeFond (1992) uses specialization as well as size as a measure of
audit quality and ﬁnds organizations with higher agency costs are more likely to select
specialist auditors. In direct tests of quality, Brown and Raghunandan (1995) and
O’Keefe, et. al. (1994) ﬁnd that specialist auditors had fewer audit deﬁciencies than non-
specialist auditors as determined by quality controls reviews performed by federal agency

Inspector Generals and the California State Controller’s Ofﬁce, respectively.

 

4 Clarkson (2000) is unable to reproduce Davidson and Neu's results when additional controls for risk are

2.3 Summary

In summary, prior research consistently shows that organizations and decision
makers believe that larger ﬁrms provide higher quality audits than smaller ﬁrms.
However, research investigating actual differences in the auditors' products has resulted
in inconsistent results. Studies that look at these differences have focused primarily on
broad measures of eamings management. There is a lack of a clear understanding of the
differences in auditors’ judgments and assessments made throughout the audit between
larger and smaller ﬁrms, or between audit ﬁrms considered to be high and low quality.

The remainder of this study seeks to provide a better understanding of these differences.

 

included in the regression models.

10

Chapter Three

A-133 AUDIT BACKGROUND

Non-proﬁt organizations that receive greater than $300,000 in federal funding
(i.e., grants from federal agencies) are required to have an annual or biennial audit
performed in accordance with the Ofﬁce of Management and Budget’s Circular A-l33
(A-133 audit). This circular requires the auditor to provide opinions on the organizations’
ﬁnancial statements, schedule of federal awards, and compliance with federal regulations.
The auditor must also provide a report on the organizations’ internal control system and a
schedule of ﬁndings and questioned costs.

The organizations’ ﬁnancial statements consist of the statement of ﬁnancial
position, statement of activities, and statement of cash flows prepared in accordance with
generally accepted accounting principles (GAAP). These statements correlate directly
with the balance sheet, income statement, and statement of cash ﬂows of a for-proﬁt
organization. The schedule of federal awards provides a detail of all federal money
received by the organization, indicating the grant number and name, the total grant
amount, the amount received to date, and the federal agency from which the money was
received. This schedule is comparable to schedules included in a for-proﬁt organization’s
notes to the ﬁnancial statement that provide detail for speciﬁc combined numbers on the
ﬁnancial statements (i.e., a detailed schedule of ﬁxed assets or debt).

When an organization receives federal funds, it must agree to adhere to speciﬁc
government regulations governing the use of those funds and must maintain an adequate

system of internal controls. For example, government regulations require non-proﬁts to

11

make attempts to use minority ﬁrms when seeking subcontractors, adequately monitor
subrecipients, and maintain sound intemal control policies. In an A-133 audit, the auditor
must assess the organization’s compliance with these requirements and provide an
opinion as to whether or not there have been material instances of noncompliance by the
organization, as well as provide a report on the internal control system, documenting
whether or not there were reportable conditions and/or material weaknesses in the internal
control structure. In a for-proﬁt audit, there is no comparable report to the opinion on
compliance, but auditors are required to provide a report to the audit committee or board
of directors when there are reportable conditions or material weaknesses of intemal
control similar to the one required for non-proﬁt organizations. (AICPA Professional
Standards, AU Section 325.16).

The ﬁnal report provided by the auditor is the schedule of ﬁndings and questioned
costs, and is especially relevant for this study. This report details all instances of
noncompliance with federal regulations. Findings are any instances of noncompliance
that do not have a direct ﬁnancial impact, while questioned costs have a direct ﬁnancial
impact on the grants. For instance, most organizations are required to submit ﬁnancial
and performance reports regularly to the granting agencies. If an organization fails to
submit these reports or submits them late, this would be considered a ﬁnding but not a
questioned cost as the failure to remit reports on time would have no impact on the grant
funds expended. In contrast, if an auditor found that the organization charged one grant
with costs that were incurred under another grant, those costs would be considered
questioned. Costs charged to grants selected for substantive audit tests that cannot be

supported by underlying documentation would also be considered questioned costs. In

12

addition, federal regulations limit the amounts and types of certain expenses. For
example, reimbursable salaries are limited to a government maximum, and costs such as
alcohol are speciﬁcally excluded from reimbursement. If an auditor ﬁnds that
organizations have charged these types of expenses to a grant, they will be included as
questioned costs. In addition, an auditor will also require adjustments to the ﬁnancial
statements for any questioned costs or ﬁndings that have a direct effect on the
presentation of the ﬁnancial statements.

The auditor is required to report all identiﬁed instances of noncompliance and
questioned costs, regardless of materiality, in the schedule of ﬁndings and questioned
costs. Although there is no direct corollary of this schedule in a for-proﬁt audit, the
schedule of ﬁndings and questioned costs is similar to a listing of audit differences and

adjustments.

l3

Chapter Four
HYPOTHESES

In the A-l33 audit setting, auditors must have an understanding of not only GAAP
requirements for non-proﬁt organizations, but must also have specialized knowledge of
grant requirements and federal regulations to adequately perform their job. In addition to
the normal state education requirements, all auditors providing A-l 33 audits must meet a
minimum standard of 24 hours of continuing professional education speciﬁcally in
government and A-1 33 courses. F inns can also choose to provide more education and
training to their employees to increase their efﬁciency and effectiveness on these audits,
and therefore differentiate themselves based on quality.

Organizations that receive federal funds are often dependent on the continuation
of those funds to remain solvent. The identiﬁcation of questioned costs or ﬁndings,
reports of material noncompliance with federal regulations, or internal control
deﬁciencies can result in a reduction or discontinuance of federal funding, and therefore
could have a signiﬁcant negative impact on the organization’s ability to continue to
operate. As such, organizations may have an incentive to attempt to dissuade an auditor
from reporting any noncompliance. Therefore, the auditors’ independence is clearly
important in these audits.

If audit quality is deﬁned as done in Chapter 2 as the “market-assessed joint
probability that a given auditor will both (a) discover a breach in the client’s accounting
system, and (b) report the breach,” (DeAngelo [1981], p. 186) higher quality auditors are
expected to ﬁnd and report more compliance and internal control problems than lower

quality auditors, assuming the same rate of problems across the auditors’ clients. In an A-

14

133 audit, higher quality auditors are expected to have a better understanding of all
government regulations and the requirements for an adequate system of internal controls.
A higher quality auditor will design and perform more effective audit tests to identify
costs that are questionable, policies that are not in compliance with federal regulations,
and signiﬁcant deﬁciencies in the internal control structure. In addition, higher quality
auditors are less likely than lower quality auditors to be inﬂuenced by clients’ desires not
to include these issues in their A-133 audit reports.

As discussed in Chapter 2, DeAngelo proposes that larger audit ﬁrms are more
likely to provide higher quality audits due to their greater resources for increased training
and their larger client base that allows them more independence from individual clients.
In prior audit quality studies using DeAngelo’s hypothesis, the largest audit ﬁrms with a
brand name have been grouped together and compared to all other audit ﬁrms.

Prior audit quality studies have also investigated whether specialist audit ﬁrms
provide a higher quality audit than non-specialist audit ﬁrms. One measure of
specialization used in these studies is the extent of market share an auditor has in a
particular market. An audit ﬁrm is considered a specialist in these studies if the ﬁrm
audits greater than 10% of the available audit market. This study uses the same deﬁnition
of specialist or market leaders.

I combine the two streams of audit quality research and test whether specialist or
market leader auditors. provide better quality audits than other brand name auditors, who

are expected to provide higher quality audits than non-brand name auditors. Based on the

 

5 Only one furn - a brand name auditor - met this deﬁnition of specialization and therefore there are no non-
brand name auditors included as a market leader.

15

prior research and the expectations related to auditors’ performance in A-133 audits, the

following hypotheses are tested:

H1: The market leader auditor will report more questioned costs than other brand
name auditors who will report more questioned costs than non-brand name
auditors, ceteris paribus.

H2: The market leader auditor will report more ﬁndings than other brand name
auditors who will report more ﬁndings than non-brand name auditors, ceteris
paribus.

H3: The market leader auditor will issue more qualiﬁed opinions on an
organization's compliance with federal regulations than other brand name auditors
who will report more qualiﬁed opinions on an organization’s compliance than
non-brand name auditors, ceteris paribus.

H4: The market leader auditor will identify and report more reportable conditions

in internal control structures than other brand name auditors who will report more
reportable conditions than non-brand name auditors, ceteris paribus.

l6

Chapter Five
METHODOLOGY
5.1 Data
The entire contents of the Federal Clearinghouse database were downloaded in
November 2000. This database consists of all A-133 audit reports and summary reports
submitted to the Federal Clearinghouse, and are available to the public through the
Internet at http://harvester.census.gov/sac. The database was started in 1997 and
continues through 2000. Therefore an organization could be included up to four times.
There were a total of 87,567 organization-years included in the database. One thousand
four hundred eighty-nine (1,489) observations were excluded because the organization
did not receive a complete ﬁnancial statement audit (i.e., received a program audit). An
additional 1,239 observations were excluded because they were for biennial rather than
annual audits. Organizations are identiﬁed in the database by their employer
identiﬁcation number (EIN - similar to the CUSIP for public companies), provided to
them by the Internal Revenue Service. It is possible for organizations to ﬁle more than
one report for each employer identiﬁcation number in a given year", and it is possible for
these reports to be received by the Federal Clearinghouse at different times. Therefore, to
reduce the possibility that year-to-year comparisons do not contain the same reports for
each EIN, these observations were excluded from the database, thereby decreasing the
sample by an additional 3,118 observations. The ﬁnancial health of an organization may

have some impact on the regressions included in this study. However, there is no

 

6 The same EIN can be used for different divisions, departments (in the case of governmental
organizations), or locations (in the case of nursing homes and hospitals) that are under the same

17

independent measure of ﬁnancial health included within the database or available
externally7. Therefore, 660 observations with going concern opinions, and 221 with no
data on the going concern opinion, were excluded to eliminate potential effects on the
regression models of poor ﬁnancial health, as measured by the auditors. Finally, 3,883
observations that were audited by state auditors were excluded as the focus of this study
is on CPA ﬁrms. The remaining sample includes 76,957 organization-years of data.
Because each regression includes different variables, and certain data may be missing for
individual observations, the sample size for each regression is different, and is less than
the total observations in the sample. The number of observations used for each

regression is included in the tables.

5.2 Models
5.2.1 Hypotheses One and Two - Questioned Costs and Findings

Actual questioned costs and ﬁndings are all instances of noncompliance with
federal regulations. Identiﬁed questioned costs and ﬁndings include only those costs or
client policies that the auditor ﬁnds during testwork and reports as not being in
compliance with federal regulations. Because auditors do not test 100 percent of all costs
charged to every grant, identiﬁed questioned costs and ﬁndings are not equivalent to
actual questioned costs and ﬁndings. Identiﬁed questioned costs and ﬁndings are a
function of the extent of testwork performed by the auditor, the auditor's understanding of

the federal regulations, the effectiveness of the auditor's tests, and the willingness of the

 

management. If each of these divisions, departments or locations requires a separate audit report, the same
EIN will be used for each.

7 The IRS database that includes organizations’ 990 ﬁlings are not available for the time period in this
study.

18

auditor to report those questioned costs. All of these factors together represent the
auditor's quality. Therefore, identiﬁed questioned costs and ﬁndings, which are what is
included in the auditors' reports, and hence the dataset for this study, are expected to be a
function of actual questioned costs and ﬁndings and auditor quality.

Actual questioned costs and ﬁndings are hypothesized to increase for
organizations that are less familiar with federal regulations, i.e., organizations in their
ﬁrst year of receiving federal funds. As the complexity and number of federal regulations
increase, management’s ability to monitor all costs charged to each program and all
policies required for each program decreases. Although all federal agencies must use the
same general standards that are included in the OMB A-133 regulations, each agency may
also institute additional regulations. As the number of federal agencies from which an
organization receives funding increases, the complexity of the federal requirements
increases, and therefore the expected extent of questioned costs and ﬁndings also
increases.

Management may be able to mitigate the difﬁculties in monitoring numerous
programs by instituting effective internal controls. Strong internal controls help to ensure
that all costs charged to individual grants and all organizational policies are accurate and
meet federal regulations, and therefore indications of weak internal controls may also be
indications of environments where higher questioned costs and ﬁndings are expected.
External factors, including regulatory and economic factors, may also have some impact
on the occurrence of questioned costs or ﬁndings.

The theoretical models for hypotheses one and two combine those factors which

affect the occurrence of actual questioned costs and ﬁndings as discussed above, with the

19

quality of the auditor, which impacts the identiﬁcation and reporting of questioned costs

and ﬁndings. The model used is:

Reported Q.C. or Reported Findings = f(Audit Quality, Familiarity, Complexity, Internal

Controls, External Factors)

This equation can be estimated by replacing the individual constructs included in
the equation with observable variables. The actual equations used to test hypotheses one
and two are included below with a discussion following for each of the variables:

QC%n = on + BlMLit + BZOBNH + B3NEW“ + B4PROGn + [35LFEDit + BbRCit + B7Y98t +
[38Y99t + B9Y00t + an
and
FIND" = or. + BIMLH + BZOBNH + B3NEW“ + B4PROG,-t + BSLFEDH + BbRCn + B7Y98t +

B8Y99t + BoYOOt + 8n

Because questioned costs are generally limited by an organization’s total federal
expenditures, questioned costs reported by the auditor as a percent of total federal
expenditures (QC%) is used as the dependent variable8 for hypothesis one, while total
ﬁndings reported (FIND) is used as the dependent variable for hypothesis two.

To test each of the hypotheses directly, audit quality is separated into three

measures - market leader, other brand name, and non-brand name. Each regression

 

8 The absolute value of total questioned costs is used as there are a few observations where total questioned
costs are negative. Negative questioned costs can result from differences between total federal expenditures
from federal agency documentation and total federal expenditures from the organization’s documentation.
This usually occurs when there is a federal reimbursement check in transit. Results do not change when
these negative observations are eliminated.

20

includes a dummy variable for the market leader auditor (ML) and other brand name
auditors (OBN)9. ML (OBN) is coded “1” if the audit ﬁrm is a market leader auditor
(other brand name auditor) and “0” otherwise. Consistent with prior studies that
investigate the effects of specialization or market leadership as a measure of audit quality
(see Craswell, et. al. [1995], DeFond [1992], and Beattie, et. al [2000]), I designate all
auditors auditing greater than 10% of total federal expenditures in the sample as a market
leader. Only one audit ﬁrm, KPMG, meets this deﬁnition of a market leader, auditing
approximately 22% of the total federal expenditures in the database. I expect a positive
coefﬁcient on both ML and OBN, and I expect the coefﬁcient on ML to be greater than
the coefﬁcient on OBN.

Familiarity with federal regulations is measured using the variable called NEW.
Fiscal year 1997 is the ﬁrst year the database was maintained and therefore is the ﬁrst
year any organization can be included in the database. For any organization whose ﬁrst
A-l33 report is after 1997, NEW is coded as “1” in the ﬁrst year that the information is
included, and is “0” otherwise”).

The complexity of federal programs is measured by two variables - the total
number of programs for which an organization receives funding (PROG - a proxy for the

total number of agencies from which an organization receives funding)11 and the natural

 

9 There is no dummy variable explicitly for non-brand name auditors as these ﬁrms are identiﬁed by being
coded “O” in both the market leader and other brand name auditor dummies.

'0 Using this measure, any organization whose ﬁrst year of receiving federal funding was 1997 will be
miscoded. The results of additional tests to determine the effects of this miscoding are reported in Chapter
6.

” The total number of agencies is not included in the database. However, this information can be
determined manually by reviewing each report individually and counting the number of different agencies
represented as evidenced by the individual programs listed on an organizations’ schedule of federal awards.
Based on a random sample of 100 organization years included in the dataset, the correlation between the
number of federal programs and the number of different agencies from which an organization receives

21

log of total federal expenditures (LF ED). The indication of reportable conditions (RC) is
included to control for the quality of the internal control structure and year dummies
(Y98, Y99, and Y00) are included to control for general regulatory and economic

environmental factors that can change from year to year.

5.2.2 Hypothesis Three - Compliance with Federal Regulations

Material noncompliance with federal regulations occurs when an organization
does not comply with those regulations that have a material effect on the federal programs
under which they are operating. Auditors form their opinion on whether or not an
organization has material instances of noncompliance by reviewing the extent and nature
of ﬁndings and questioned costs identiﬁed throughout the audit. Given this, actual
material noncompliance is expected to be affected by many of the same factors that affect
an organization’s actual questioned costs and ﬁndings - management’s familiarity with
federal programs, complexity of regulations, and the strength of the internal control
structure, as well as regulatory and economic factors.

Similar to the previous discussion on questioned costs and ﬁndings, actual
material noncompliance does not equal reported material noncompliance, since the latter
is dependent on the quality of the auditor. I estimate the effects of auditor type on the
likelihood of an organization receiving a qualiﬁed opinion on its federal regulation
compliance by controlling for those organizational factors that are expected to impact

actual material noncompliance. I also control for the level and extent of reported

 

funding is approximately 72%. This correlation appears to be sufﬁciently high to support using the number
of federal programs as a proxy for the number of different agencies, as well as a control for the number and
complexity of federal programs under which an organization operates.

22

questioned costs and ﬁndings as these should have a direct impact on the opinion. The
equation estimated is:
MNCit = Oti + BlMLit + BZOBNII + B3NEWII + B4PROGit + BSLFEDII + BORCit + B7FrNDir
+ BsQCO/on + B9Y98r + B10Y99t+ BllYOOt + 8i:
where:
MNCn = dummy variable indicating whether there was a qualiﬁed opinion on
compliance with federal regulations at time t (0 indicates an

unqualiﬁed opinion, 1 otherwise)

All other variables are deﬁned as above in hypotheses one and two.

5.2.3 Hypothesis Four - Reportable Conditions in Internal Controls
Reportable conditions in internal controls are “matters that. . .represent signiﬁcant

deﬁciencies in the design or operation of the internal control structure, which could
adversely affect the organization’s ability to record, process, summarize, and report
ﬁnancial data...” (AICPA Professional Standards, AU Section 325.02). They are
expected to be affected by the organization’s familiarity with federal programs as well as
the complexity and extent of the federal regulations, and external regulatory and
economic factors. Higher quality auditors are expected to be more likely to identify and
report reportable conditions in internal controls because of their more effective testwork
and their greater independence from their clients. The equation used to test this
hypothesis is:

RC“ = or: + BlMLn + BZOBNR + B3NEWR + B4PROGn + 05LFED" + B6Y98‘ + B7Y99t +

BgYOOt + an

where all variables are deﬁned as above under hypotheses one and two.

23

5.3 Controls for Organizational Factors

Organizations select their own auditor. Prior audit quality literature indicates that
certain organizational factors such as risk and extent of agency costs are associated with
the choice of the speciﬁc quality level of the auditor (see discussion in Section 2.1 for
details on the prior literature). These organizational factors may correlate with the
dependent variables included in this study, and it is therefore important to control for
these in order to obtain interpretable results. The panel data nature of the dataset, with
the same organizations included for a number of years, is used to control for these
organizational factors, with each organization acting as its own control. Using ﬁxed
effects and ﬁrst differencing estimation techniques, all organizational factors that do not
change over time are controlled for, and therefore should have no effect on the results of

the study.

24

Chapter Six
RESULTS AND SUMMARY
6.1 Descriptive Statistics

Panel A of Table 1 provides descriptive statistics on the continuous variables
included in this study. The organizations’ federal expenditures vary considerably,
ranging from $27,055 to more than $16.7 billion. Questioned costs range from $0 to more
than $34.1 million, and total ﬁndings reported by the auditor range from O to 105. The
low means for questioned costs and ﬁndings of $9,200 and .49, respectively, are due
mainly to the signiﬁcant number of observations with zero questioned costs and ﬁndings.
Based on a comparison of the means and medians for total federal expenditures and
questioned costs, it is apparent that the data is skewed toward smaller organizations and
organizations with fewer questioned costs. The mean number of programs organizations
manage is 10, with a range from 1 to 1,605.

(Insert Table 1 about here)

Panel B of Table 1 provides the descriptive statistics for the binary variables
included in the regressions. There are 2,241 observations audited by the market leader,
KPMG, and 4,872 observations audited by one of the other brand name audit ﬁrms”.
The remaining 69,844 observations were audited by a non-brand name ﬁrm.
Approximately 4% of all observations received a report indicating material
noncompliance with federal regulations, while almost 23% of the observations had
indications of signiﬁcant deﬁciencies in their internal control systems as evidenced by

reportable conditions. These statistics indicate that many of these organizations have

25

considerable problems with their internal control structures, and therefore may be able to
beneﬁt greatly from external audits.

Tables 2 and 3 provide descriptive statistics on individual sub-groups of
observations within the dataset.

(Insert Table 2 about here)
It is evident from Table 2, Panel A that organizations with a market leader or an other
brand name auditor have signiﬁcantly more federal expenditures, questioned costs,
numbers of programs, and numbers of ﬁndings than organizations with a non-brand name
auditor. Within brand name audited organizations, market leader audited organizations
have signiﬁcantly more in federal expenditures and ﬁndings than organizations audited
by other brand name auditors.

Based on the univariate comparisons in Panel B of Table 2, market leader and
other brand name audited organizations are less likely to receive a qualiﬁed opinion on
their compliance with federal regulations or receive a report of reportable conditions than
non-brand name audited organizations. These comparisons, however, do not control for
differences in the sizes of each organization’s federal program activities and other
organizational factors that will be used in the multiple regression analyses.

(Insert Table 3 about here)

Panels A and B of Table 3 provide a comparison of the regression variables for
organizations that switched among the three auditor types - market leader, other brand
name, and non-brand name. Panel A provides the statistics for organizations that

decreased their expected level of audit quality, while Panel B presents the statistics for

 

'2 Given that the data spanned 1997 through 2000, the period in which Price Waterhouse merged with

26

those organizations that increased their expected level of audit quality. There are a total
of 514 organization-years in which an organization changed their type of auditor. There
were 121 changes from a market leader auditor to a non-brand name auditor (a decrease
of two levels of expected audit quality - further referred to as a two-level decrease), and
285 changes from either a market leader auditor to an other brand name auditor, or from
an other brand name auditor to a non-brand name auditor (a decrease of one level of
expected quality - further referred to as a one-level decrease). There were 77 changes
from a non-brand name auditor to a market leader auditor (an increase of two levels of
expected audit quality - further referred to as a two-level increase), and 31 changes from
either a non-brand name auditor to an other brand name auditor or from an other brand
name auditor to a market leader auditor (an increase of one level of expected audit quality
- further referred to as a one-level increase).

The results in Panel A indicate that organizations that had both one-level and two-
level decreases in the type of auditor had large decreases in federal expenditures in the
year of the change. The decrease for one-level changes is statistically different than the
increase seen in federal expenditures for organizations that did not shift their auditor type.
Organizations that had a two-level decrease in auditor had a large decrease in the number
of programs they managed in the year of the change, which is statistically different from
the increase in the number of programs for organizations that did not change auditor type.
Contrary to expectations, organizations that decreased their level of expected audit quality
had more reportable conditions noted in the year of the change than organizations that did

not change auditors. In addition, organizations with a two-level decrease had a larger

 

Coopers and Lybrand, ﬁrms with either of these two auditors were coded as “PWC”.

27

increase in reportable conditions than organizations with a one-level decrease in auditor
type.

The only signiﬁcant difference among the organizations reported in Panel B of
Table 3 is the difference in federal expenditures. Organizations that had a one-level
increase in the type of auditor had a signiﬁcantly larger increase in federal expenditures
than organizations that did not change their auditor type. Although the difference in
questioned costs between organizations increasing their auditor level and organizations
not changing their auditor type is not statistically signiﬁcant at a 5% or lower level, it is in
the expected direction. This is also true with the difference in ﬁndings. There was an
increase in the number of reported ﬁndings for organizations that increased their expected
audit quality level, although it is not statistically different than the no shiﬁ organizations.
Contrary to expectations, but consistent with the univariate results in Panel A,
organizations that increased their expected level of quality show a decrease in reportable
conditions as compared to no shift organizations, although, again, the difference is not
statistically signiﬁcant.

Table 4 provides simple Pearson correlations among the variables included in the
regressions. Most of the correlations are signiﬁcant and in the expected direction. For
instance, as the number of programs and level of federal expenditures increase,
organizations have more questioned costs, ﬁndings, and material noncompliance with
federal regulations. In addition, organizations that have deﬁciencies in their internal
control systems (as evidenced by reportable conditions) are also more likely to have
questioned costs and ﬁndings. Organizations with market leader or other brand name

auditors are associated with larger questioned costs and ﬁndings, but contrary to

28

expectations, they have fewer reportings of material noncompliance and reportable

conditions than organizations with non-brand name auditors.

6.2 Hypotheses Tests
6.2.1 Hypothesis One - Questioned Costs

The results for hypothesis one using both a ﬁxed effects levels and ﬁrst
differencing changes model are included in Table 5. Both models were estimated using
OLS. The coefﬁcients on both ML and OBN are consistent across the levels and changes
regressions, and each are signiﬁcant at a 10% or better level, using a one-tailed test. The
results indicate that both the market leader and other brand name auditors report more
questioned costs than non-brand name auditors. In addition, the larger coefﬁcient on ML
as compared to OBN in the levels model indicates that the market leader auditor reports
more questioned costs than other brand name auditors, as predicted by hypothesis one
(one tailed p-value<0.01). The coefﬁcient estimate for ML is not signiﬁcantly greater
than that for OBN in the changes model, although the difference is in the expected
direction (one tailed p-value<0.14). Organizations that select the market leader auditor
can expect, on average, $61,600 more in questioned costs than organizations with a non-
market leader auditor, and organizations with an other brand name auditor can expect
$23,100 more in questioned costs than organizations with a non-brand name auditor
(based on average federal expenditures of approximately $7,700,000).

(Include Table 5 about here)
The results on the control variables are generally consistent between the levels and

changes model, although many of them are not statistically or economically signiﬁcant.

29

The existence of reportable conditions in an organization’s internal control system,
however, is associated with a 0.004 increase in questioned costs as a percent of total
federal expenditures, or approximately $30,800. The number of programs an
organization manages does not impact questioned costs, although as federal expenditures
increase, questioned costs as a percent of federal expenditures decrease, indicating that

questioned costs do not increase ratably with federal expenditures.

6.2.2 Hypothesis Two - Findings

The effects of different types of auditors on the number of ﬁndings reported are
included in Table 6. Both the levels and changes models were estimated using OLS.
Consistent with hypothesis two, the market leader and other brand name auditors report
signiﬁcantly more ﬁndings than non-brand name auditors (one tailed p-value<0.01 and
0.05 for the levels and changes models, respectively). Determining whether market
leaders report more ﬁndings than other brand name auditors depends on the model used.
The levels model indicates a signiﬁcant difference at a p-value<0.05, but the changes
model does not indicate a signiﬁcant difference between the two designations within
brand name auditors (both using a one tailed test).

(Insert Table 6 about here)

Most of the control variables are signiﬁcant, and are in the expected direction in
both the changes and the levels models. As the number of programs and the level of
federal expenditures increases, so do the number of ﬁndings. The results of the model
also indicate that organizations less familiar with federal programs are more likely to

have more ﬁndings, as are organizations that have poor internal controls.

30

6.2.3 Hypothesis Three - Material Noncompliance

The results of the effect of auditor type on the issuance of a report indicating
material noncompliance with federal regulations are included in Table 7. Both the levels
and changes models were estimated using OLS. Contrary to expectations, after
controlling for the number of ﬁndings and extent of questioned costs, both the levels and
changes models indicate that the market leader auditor is less likely, rather than more
likely than non-brand name auditors, to issue a qualiﬁed opinion on compliance with
federal regulations (signiﬁcant at two tailed p-value<0.07). Therefore, given an
organization with the same number of ﬁndings and the same amount of questioned costs,
a market leader auditor will be less likely than a non-brand name auditor to issue a
qualiﬁed opinion on that organization’s compliance. This appears to be an indication that
the market leader auditor is providing lower quality than non-brand name auditors. Also
contrary to expectations, other brand name auditors are not signiﬁcantly more likely than
non-brand name auditors to issue a qualiﬁed opinion on compliance, and may actually be
less likely to report material noncompliance (two tailed p-value near 0.20). Again, this
indicates that other brand name auditors are not providing higher quality assessments as
expected. There is no statistical difference in the likelihood of a qualiﬁed opinion on
compliance among the brand name ﬁrms, regardless of being a market leader or not.

(Insert Table 7 about here)

As expected, as the number of ﬁndings and the extent of questioned costs
increase, organizations are more likely to have material noncompliance reports issued. In
addition, organizations with indications of reportable conditions in their internal control

systems are more likely to have qualiﬁed opinions on their federal program compliance.

31

The number of programs does not appear to have any effect on an organization’s
compliance, although an increase in total federal expenditures does tend to increase the

likelihood of receiving a qualiﬁed opinion on compliance (one tailed p-value<0.10).

6.2.4 Hypothesis Four - Reportable Conditions

The effect of audit quality on the reporting of reportable conditions is included in
Table 8. Both the levels and changes models were estimated using OLS. Market leaders
and other brand name auditors are signiﬁcantly less likely to report reportable conditions
than non-brand name auditors. In addition, market leaders are even less likely to issue
reportable conditions letters than other brand name auditors (two tailed p-value<0.08).
Both of these results are in direct contradiction to those proposed in hypothesis four.
Holding other organizational factors constant, auditors that are deemed “higher quality”
by users of ﬁnancial statements are less likely to report signiﬁcant deﬁciencies in an
organization’s internal control structure. Assuming that all the proper controls have been
included in the model, the brand name auditors are identifying, reporting, or both
identifying and reporting less instances of reportable conditions than non-brand name
auditors.

(Insert Table 8 about here)

As expected, the results on the control variables indicate organizations in their
ﬁrst year of receiving federal funds are more likely to have reportable conditions. In
addition, as the size of an organization's federal programs increase, so does the

organization's likelihood of having reportable conditions.

32

6.3 Additional Estimation Methods
6.3.1 Hypotheses One and Two - Questioned Costs and Findings

More than 90% of all observations in the dataset have zero questioned costs or
zero ﬁndings. Although the linear estimation models used earlier provide a reasonable
and easily interpretable estimation of the effects of different types of auditors on the
measures of audit quality, the Tobit model provides another method to estimate the
coefﬁcients while controlling for the large number of zeroes in the population
(Wooldridge [2000])”. Therefore, a Tobit model was estimated using maximum
likelihood estimation to test hypotheses one and two. The results of this estimation
technique for both hypotheses are included in Table 9.

(Insert Table 9 about here)

The magnitudes of the coefﬁcients from the Tobit estimation cannot be directly
compared to those from the ﬁxed effects model. However, the Tobit coefﬁcient estimates
for the market leader and other brand name auditors for the questioned costs model are
positive and have similar signiﬁcance levels to those estimated using ﬁxed effects
estimation. Similar to the results found in the OLS estimated changes model, market
leader auditors do not report statistically more questioned costs than other brand name
auditors (one tailed p-value<0.26), although the difference is in the expected direction.
Combining the results of the ﬁxed effects, ﬁrst differencing and Tobit models, it appears

that market leader and other brand name auditors report more questioned costs than non-

 

” The Tobit model provides an opportunity to test both the direct effects of the independent variables on
the dependent variable, as well as the effects of the independent variables on the dependent variable given
that the dependent variable is greater than zero. This additional estimation is not relevant to the current
study, and therefore the results of this estimation are not provided.

33

brand name auditors, but there may be no difference between reported questioned costs
for market leader and other brand name audited organizations.

The Tobit results for ﬁndings indicate results consistent with those found from the
estimates of the coefﬁcients in the ﬁxed effects model. The market leader auditor reports
statistically more ﬁndings than both the other brand name auditors (p-value<0.04) and the
non-brand name auditors (p-value<0.001). In addition, the other brand name auditors
reports statistically more ﬁndings than the non-brand name auditors (p—value<0.01).
Therefore, hypothesis two is supported using both a ﬁxed effects linear estimation as well
as the Tobit estimation which better controls for the signiﬁcant incidences of zero

ﬁndings in the population.

6.3.2 Hypotheses Three and Four - Material Noncompliance with Federal
Regulations and Reportable Conditions

Hypotheses three and four are tested using levels models with dependent variables
that are discrete, taking on values of zero or one. Each of these models was re-estimated
using a maximum likelihood estimate of a conditional logit model which better controls
for the discrete nature of the dependent variables, while also controlling for those
organizational factors that do not change over time (Greene [1997]). The results of these
estimations are included in Table 10.

(Insert Table 10 about here)

The results of the logit models can be compared on direction and signiﬁcance to

the estimates obtained from the ﬁxed effects estimates included in Tables 7 and 8. For

both hypotheses, the coefﬁcients on other brand name auditors are negative and highly

34

signiﬁcant. In addition, the market leader auditor is signiﬁcantly less likely than both
other brand name and non-brand name auditors to report reportable conditions
(hypothesis four). These results are generally consistent with the results discussed in
Section 6.2, although the results from the logit model on the material noncompliance
hypothesis are more signiﬁcant than those reported earlier. Together, these results
continue to suggest that market leaders and other brand name auditors report fewer
qualiﬁed opinions on compliance with federal regulations and fewer reportable conditions
than do non-brand name auditors. Because the magnitudes of the coefﬁcients have no
economic meaning, no translation was performed to compare the magnitudes of the logit

estimates with those reported in Tables 7 and 8.

6.4 Sensitivity Tests
6.4.1 Miscoding of “NEW”

The variable NEW is used to proxy for the familiarity of the organization with
federal grant requirements, and is coded as “1” in the ﬁrst year that an organization is
included in the database if that year is not 1997. NEW is coded as zero for all
organizations in 1997 because that is the ﬁrst year of the database, and therefore is the
ﬁrst year that all organizations are included in the database. If an organization received
federal funds for the ﬁrst time in 1997, although they are “new” to federal programs,
NEW would still be coded as zero. To determine if this miscoding affects the results, all
1997 data was eliminated and the same equations were estimated for each hypothesis

using both the ﬁxed effects and changes models.

35

The results from these tests are mixed. In the questioned costs regression (testing
hypothesis one), estimates of the two coefﬁcients on ML (market leader) and OBN (other
brand name) are slightly larger than those with all of the data included, and are still
signiﬁcant at p—value<0.10. In the ﬁndings regression, the estimates of the coefﬁcients
on ML and OBN were not signiﬁcantly different than zero. The results for the material
noncompliance model were similar to the results found with all of the data included
(reported in Table 7), in that the estimates of the coefﬁcients on ML and OBN were
negative, but not consistently signiﬁcant across both the ﬁxed effects and changes
estimations. The results on hypothesis four, reportable conditions, are generally
consistent with those reported in Table 8, however the market leader coefﬁcient was not
statistically greater than the other brand name coefﬁcient.

When the 1997 data is eliminated from the population, the number of observations
drops from 76,957 to 54,379. In addition, the average number of years that an
organization is included in the database drops from 2.15 with the 1997 data included, to
1.66 without the 1997 data”. Both of these decreases could account for the slightly
inconsistent results found with the smaller dataset. Much of the power for the tests of the
hypotheses comes from the large number of observations included in the estimation. In
addition, both the ﬁxed effects and changes models rely on organizations remaining in the

sample for at least two years; any reduction in the average number of years that an

 

'4 When the 1997 data is deleted, approximately 12,000 of the 54,300 observations represent organizations
with only one year of data in the remaining data set. The ﬁxed effects estimation technique time-demeans
each observation, meaning that the organization average for each variable is subtracted from the actual
variable value. For organizations with only one year of data, the average and the actual value for each
variable is the same, and the time demeaned values (or values used in the regression estimation) are zero.
Therefore, although these organizations still remain in the dataset and in the regression, they have no effect
on the regression results. [Wooldridge, 2000]

36

organization is included in the dataset will have detrimental effects on the power of the
tests. Even with this reduction in power, the results of this additional test indicate that the

miscoding of the variable “NEW” does not greatly affect the results reported earlier.

6.4.2 Additional Measure of Size

Prior audit quality literature has found that the size of an organization is
signiﬁcantly correlated with the type of auditor selected. In this study, I hypothesize that
the size or complexity of an organization may also correlate with each of the dependent
variables used in this study. Therefore, in order to obtain interpretable results, it is
important that a measure of size be included in the models. I have used the log of total
federal expenditures as a measure of size. However, it is possible for organizations to be
quite large with few federal program expenditures, and therefore this measure of Size may
not adequately control for the real size of the organization.

The database used for this study does not include any additional ﬁnancial
variables that can be used as another measure of size. Therefore, additional ﬁnancial
information was obtained from another source — the Guidestar website (available to the
public on the Internet at www.Guidestar.org). This website maintains copies of Internal
Revenue Service Form 990’s (the tax return for a non-proﬁt organization), as well as
summary ﬁnancial data for over 700,000 non-proﬁt organizations.

I selected a sample of organizations from the original database used for this study,
attempting to obtain a sample with similar characteristics on the dependent variables as
those in the original database. The sample was selected using a stratiﬁed random

sampling technique. I ﬁrst stratiﬁed the original database into two populations —

37

organizations with zero questioned costs and organizations with positive questioned
costs. I then randomly selected a representative sample from each population subgroup.
For instance, I selected approximately 95% of the sample from the subgroup with zero
questioned costs, and 5% of the sample from the subgroup with positive questioned costs,
similar to the rates found in the entire population. I also compared the occurrence of
positive ﬁndings, a qualiﬁed opinion on compliance with federal regulations, and the
reporting of reportable conditions in the sample obtained to the rates in the entire
population to ensure that the sample selected had similar occurrences of these conditions
as those found in the total database. In order to increase the power of the tests, I also
selected additional observations in which the organization changed their auditor type,
which will assist in increasing the variance on the independent variables of interest.

For each of the organizations selected, I obtained the total assets, total revenues
and total expenditures for each organization as reported on and included in Guidestar’s
database. The organization had to have at least two concurrent years of Form 990’s in
Guidestar’s database in order to be included in the sample. Without two concurrent
years, the ﬁxed effects and ﬁrst difference estimations cannot be performed.

The ﬁnal sample included 494 observations. The results of the regression
analyses are not reported as they generally report insigniﬁcant results, with or without the
additional ﬁnancial measures. This is because of the very small number of observations.
The additional data can be used to determine if total federal expenditures is an adequate
measure of size. Therefore, I calculated the correlation between total federal expenditures
and the newly obtained size measures of total assets, total revenues, and total

expenditures to determine if federal expenditures is a reasonable proxy for these other

38

measures of size. The correlations range from 0.917 (federal expenditures and total
assets) to 0.961 (federal expenditures and total expenses). All of the correlations are
highly signiﬁcant (two tailed p-value<0.00), thereby providing support for the use of total

federal expenditures as a reasonable measure of size.

6.4.3 Another Test for the Effects of Size

As discussed above in Section 6.4.2, it is important in this study to control for the
size of an organization as size may correlate with both the type of auditor selected and the
dependent variables in the study. It is evident from Table 2 that the federal expenditures
of organizations audited by market leaders and other brand name auditors are very
different than the federal expenditures of organizations audited by non-brand name
auditors. To provide an additional robustness check that size is not driving the results
previously reported, I ran the regressions on a subset of the data where organizations were
matched on size.

To obtain the matched dataset, ﬁrst, average federal expenditures was calculated
for each organization. Then, for each organization audited by the market leader, I found a
similar sized organization audited by an other brand name ﬁrm, and a similar sized
organization audited by a non-brand name ﬁrm. An organization was considered to be
similar in size if its average federal expenditures was within 10% of the average federal
expenditures of the market leader audited organization. A match had to be obtained in
both the other brand name and the non-brand name subsets of data or the original

organization was eliminated.

39

There were 1,057 organizations audited for at least one year by the market leader
auditor. This represents 2,239 total observations audited by the market leader. A match
was found for all but ﬁve organizations (representing 12 observations). The ﬁnal
matched sample included 2,226 observations audited by the market leader, 1,685
observations audited by an other brand name auditor, and 2,196 observations audited by a
non-brand name auditor. Within the matched sample, mean (median) federal
expenditures are $30,900,000 ($5,950,663) for market leader audited organizations,
$28,900,000 ($5,842,887) for other brand name audited organizations, and $22,100,000
($4,758,965) for non-brand name audited organizations”.

Each of the four regressions was re-estimated with the matched data using an OLS
estimate of the ﬁxed effects model. The results are generally consistent with those
obtained in the original estimations reported in Section 6.3, and are therefore not reported
in a separate table. In both the questioned costs and ﬁndings regressions, I ﬁnd the
market leader and other brand name auditors report more questioned costs and more
ﬁndings than non-brand name auditors. The market leader coefﬁcient is signiﬁcant using
a one tailed p-value<0.001 and the other brand name coefﬁcient is signiﬁcant using a one
tailed p-value<0.08 in both regressions. In both regressions, I do not ﬁnd any statistical
difference between the market leader and other brand name auditors. The results for the
material noncompliance regression (hypothesis three) indicate no statistically signiﬁcant

differences among auditor types, although the coefﬁcients on both ML and OBN are

 

'5 These means are statistically different using a two tailed p-value<0.001. However, given the large
number of observations in the sample, obtaining three subsets of matched data with statistically equivalent
means is virtually impossible. Even dropping observations where a match within 5% could not be identiﬁed
results in means that are statistically different using a two tailed p-value<0.05. The means and medians are

40

negative - similar to the results reported in Table 7. Similar to the results previously
reported on reportable conditions (hypothesis four), within the matched dataset, market
leader and other brand name auditors report less reportable conditions than non-brand
name auditors (two tailed p-value<0.001) and the market leader reports marginally less
reportable conditions than other brand name auditors (two tailed p-value<0.11). These
results are generally consistent with those reported in Table 8, although they are in direct

contrast to the hypothesized results.

6.4.4 Extreme Observations

Included in the database were ten observations in which total questioned costs
exceeded total federal expenditures. For each observation, I reviewed the copy of the
schedule of questioned costs and ﬁndings in the Federal Clearinghouse’s database to
ensure that there was not a data entry error. I then re-estimated the questioned costs
model for hypothesis one without these large observations included. The results were

unchanged from those reported in Table Five.

6.4.5 Summary of Additional Robustness Tests

The results on the robustness tests discussed above are generally consistent with
the results obtained using the original models and estimation methods. Although some of
the additional tests result in less signiﬁcance than was found with the original models,
this appears to be due to the reduction in the amount of data used for estimation. The

results provided in the original analyses are generally signiﬁcant; however, much of the

 

not economically very different. Therefore, this sample of the data is reasonable to determine the effects of

41

power for the tests comes from the large number of observations in the data set.
Therefore, it is not unusual to expect insigniﬁcant results when the data set is reduced in

size.

6.5 Summary

In summary, the results on the hypotheses are mixed. Based on the results in this
study, users of ﬁnancial statements can expect that market leaders and other brand name
auditors will ﬁnd and report more ﬁnancial errors (questioned costs) and speciﬁc
instances of noncompliance (ﬁndings) than non—brand name auditors, holding all other
factors constant. This result is consistent with the proposal that market leaders and other
brand name auditors provide higher quality audits than non-brand name auditors.
However, market leader auditors are less likely that non-brand name auditors to issue
qualiﬁed opinions on compliance with federal regulations, and both market leaders and
other brand name auditors are less likely to report reportable conditions in internal
controls than non-brand name auditors. Assuming that all the proper control variables
have been included in the regression, these results are consistent with market leaders and

other brand name auditors providing lower quality than non-brand name auditors.

6.6 Implications and Future Research
As with prior audit quality literature, the results of this study are mixed. Whether
market leader and other brand name auditors provide better quality depends on the

measure of quality used in the study. The results are generally consistent with

 

size.

42

expectations when questioned costs and ﬁndings are used as the quality measure, but they
are not consistent with expectations when the auditors' reporting of reportable conditions
or opinion on material compliance are used as the quality measure. These inconsistent
results indicate that additional research needs to be conducted to determine the true nature
of the relationship between auditor type and audit quality.

Future research related to this study might investigate why market leader and
other brand name auditors report more questioned costs and ﬁndings, but report less
material noncompliance with federal regulations. It is possible that larger and more
specialized auditors have different decision making processes than smaller and less
specialized auditors in terms of assessing the materiality of these errors. Future research
might investigate the decision processes of the different audit ﬁrm types.

The data used in this study also provides a rich environment for additional
research. A large number of the organizations in the dataset have poor internal control
structures, and may therefore provide an excellent setting to test the effects of the
adequacy of internal controls on organizational performance. In addition, future research
might investigate whether organizations compensate for their poor internal control
structures by controlling agency problems with other techniques, such as having a more

active board of directors or selecting higher quality auditors.

43

REFERENCES

Allen, Arthur C., 1994, The effect of large-ﬁnn audits on municipal bond rating
decisions, Auditing: A Journal of Practice and Theory, 13, Spring, 115-125.

American Institute of Certiﬁed Public Accountants Professional Standards, Volume 1,
Commerce Clearing House, Inc., Chicago, Illinois.

Balvers, R.J., B. McDonald, and RE. Miller, 1988, Underpricing of new issues and the
choice of auditor as a signal of investment banker reputation, The Accounting Review,
63, October, 605-622.

Beasley, Mark and K. Petroni, 2001, Board independence and audit ﬁrm type, Auditing:
A Journal of Practice and Theory, 20, March, 97-114.

Beattie, Vivien, A. Goodacre, K. Pratt, and J. Stevenson, 2000, The determinants of audit
fees - Evidence from the voluntary sector, Working paper.

Beatty, Randolph P., 1989, Auditor reputation and the pricing of initial public offerings,
The Accounting Review 64, October, 693-709.

Becker, Connie L., M. L. Defond, J. J iarnbalvo, and KR. Subramanyam, 1998, The effect
of audit quality on earnings management, Contemporary Accounting Research, 15,
Spring, 1-24.

Brown, Clifford D. and K. Raghunandan, 1995, Audit quality in audits of federal
programs by non-federal auditors, Accounting Horizons, 9, September, l-lO.

Clarkson, Peter M., 2000, Auditor quality and the accuracy of management earnings
forecasts, Contemporary Accounting Research, 17, Winter, 595-622.

Clarkson, Peter M. and Dan A. Simunic, 1994, The association between audit quality,
retained ownership, and ﬁrm-speciﬁc risk in US. vs. Canadian [PO markets, Journal of
Accounting and Economics, 17, 207-228.

Colbert, Gary and D. Murray, 1998, The association between auditor quality and auditor
size: An analysis of small CPA ﬁrms, Journal of Accounting, Auditing, and Finance, 13,
Spring, 135-150.

Copley, P.A, J .J . Gaver, and KM. Gaver, 1995, Simultaneous estimation of the supply

and demand of differentiated audits: Evidence from the municipal audit market, Journal
of Accounting Research, 33, Spring, 137-155.

44

Craswell, Allen T., J. R. Francis, and S. L. Taylor, 1995, Auditor brand name reputations
and industry specializations, Journal of Accounting and Economics, December, 20, 297-
322.

Datar, S.M., G.A. Feltham, and J .S. Hughes, 1991, The role of audits and audit quality in
valuing new issues, Journal of Accounting and Economics, 14, 3-49.

Davidson, Ronald A. and D. Neu, 1993, A note on the association between audit ﬁrm size
and audit quality, Contemporary Accounting Research, 9, Spring, 479-488.

DeAngelo, Linda E., 1981, Auditor size and audit quality, Journal of Accounting and
Economics, 3, December, 183-199.

DeFond, Mark L., 1992, The association between changes in client ﬁrm agency costs and
auditor switching, Auditing: A Journal of Practice and Theory, 11, Spring, 16-31.

F eltham, G.A., J. S. Hughes, and DA. Simunic, 1991, Empirical assessment of the
impact of auditor quality on the valuation of new issues, Journal of Accounting and
Economics, 14, 375-399.

Firth, Michael and CK. Liau-Tan, 1998, Auditor quality, signaling, and the valuation of
initial public offerings, Journal of Business Finance and Accounting, 25, J anuary/March,
145-165.

Firth, Michael and A. Smith, 1992, Selection of auditor ﬁrms by companies in the new
issue market, Applied Economics, 24, 247-255.

Francis, Jere R., E.L. Maydew, and H. C. Sparks, 1999, The role of big 6 auditors in the
credible reporting of accruals, Auditing: A Journal of Practice and Theory, 18, Fall, 17-
34.

Greene, William H., 1997, Econometric Analysis, Third Edition, Prentice-Hall, Inc.

Hogan, CE, 1997, Costs and beneﬁts of audit quality in the [PC market: A self-
selection analysis, The Accounting Review, 72, January, 67-86.

Lee, Philip, D. Stokes, S. Taylor, T. Walter, 1999, The association between audit quality,
accounting disclosures and ﬁrm-speciﬁc risk: Evidence from the Australian [PC market,
Working paper.

Krishnan, Jagan and RC. Schauer, 2000, The differentiation of quality among auditors:

Evidence from the not-for—proﬁt sector, Auditing: A Journal of Practice and Theory, 19,
Fall, 9-25.

45

O’ Keefe, T.B., R.D. King, and KM. Gaver, 1994, Audit fees, industry specialization,
and compliance with GAAS reporting standards, Auditing: A Journal of Practice and
Theory, 13, Fall, 41-53.

Petroni, Kathy and M. Beasley, 1996, Errors in accounting estimates and their relation to
audit ﬁrm type, Journal of Accounting Research, 34, Spring, 151-171.

Simunic, Dan and M. Stein, 1987, Product differentiation in auditing: Auditor choice in
the market for unseasoned new issues, Research Monograph No. 13,Vancouver, B.C.:

Canadian Certiﬁed General Accountants' Research Foundation.

Teoh, Siew H. and T]. Wong, 1993, Perceived auditor quality and the earnings response
coefﬁcient, The Accounting Review, 68, April, 346-366.

Thornton, DB. and G. Moore, 1993, Auditor choice and audit fee determinants, Journal
of Business Finance and Accounting, 20, April, 333-349.

Titman, S. and B. Trueman, 1986, Information quality and the valuation of new issues,
Journal of Accounting and Economics, 8, 159-172.

Wooldridge, Jeffrey M., 2000, Introductory Econometrics: A Modem Approach, South-
Westem College Publishing - Thomson Learning.

46

Table 1
Descriptive Statistics - All Data

Panel A — Continuous Variables (All Data)

 

 

 

 

 

 

 

 

 

 

 

Mean Standard
Variable (Median) Deviation Minimum Maximum
Federal Expenditures 7,701,029 145,000,000 27,055 16,701 ,635,000
(FED) (1,167,189)
Questioned Costs 9,200 265,679 0 34,183,000
(QC) (0)
Questioned Cost 0.20% 3.3% 0 471.66%
Percent (QC%) (0)
Number of Programs 10 27.4 1 1,605
(PROG) (6)
Findings 0.49 1 .65 0 105
IFIND) (0)

 

Panel B — Binary Variables @11 Data) - Percenta e of Occurrences of a l

 

 

 

 

 

 

 

 

 

 

All Data

(N = 76,957)
Market Leader (ML) 2.91%
Other Brand Name (OBN) 6.33%
Material Noncompliance (MNC) 3.77%
Reportable Conditions (RC) 22.83%
New (N EW) 16.90%
LEGEND

Federal Expenditures (FED)

Questioned Costs (QC)

Questioned Cost Percent (QC%)

Number of Programs (PROG)

Findings (FIND)

Market Leader (ML)

Other Brand Name (OBN)

total federal expenditures

absolute value of total questioned costs
identiﬁed by the auditor

absolute value of questioned costs
identiﬁed by the auditor as a percent of
total federal expenditures

= number of different federal programs

total ﬁndings identiﬁed by the auditor

= dummy variable indicating market leader

47

auditor (1 if market leader, 0 otherwise)
dummy variable indicating other brand
name auditor (1 if other brand name, 0
otherwise)

 

Table 1
(continued)

Material Noncompliance (MNC)

1 if other than unqualiﬁed opinion issued by
auditor on compliance with federal regulations, 0
otherwise

1 if reportable condition identiﬁed by auditor, 0
otherwise

1 if ﬁrst year of data included in database is other
than 1997, 0 otherwise

Reportable Conditions (RC)

New (NEW)

48

Panel A - Comparison of Auditor Types

Table 2
Descriptive Statistics - Comparisons Between Auditor Types

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Means (Medians)
Market Other Brand Non-Brand
Leader Name Name
(N = 2,241) (N = 4,872) (N = 69,844L *

Federal Expenditures 61,100,000 28,900,000 4,505,830

FED) (6,018,362) (3,725,993) (1,074,990) a,b,c
Questioned Cost 58,413 49,994 4,791
QC) (0) (0) (0) b,c
Questioned Cost 0.005 0.002 0.002
Percent (QC‘VQ (0) (0) (0) a,c
Number of Programs 26.66 24.92 8.48
(PROG) (9) (8) (6) b,c
Findings 1.13 0.79 0.45
(FIND) (0) (0) (0) a,b,c

Panel B - Comparison of Auditor Types
Binary Variables - Percengge of Occurrences of a 1
Market Other Brand Non-Brand
Leader Name Name
(N = 2,241) (N = 4,872) (N = 69,844) *

Material Noncompliance

MNC) 1.78% 1.64% 3.99% b,c
Reportable Conditions (RC) 7.36% 6.22% 24.48% a,b,c
New (NEW) 11.2% 13.75% 17.31% a,b,c

 

 

 

 

 

 

LEGEND
Federal Expenditures (FED)
Questioned Costs (QC)

Questioned Cost Percent (QC%)

Number of Programs (PROG)
Findings (FIND)

= total federal expenditures

by the auditor

the auditor as a percent of total federal

expenditures
= number of different federal programs
= total ﬁndings identiﬁed by the auditor

49

absolute value of total questioned costs identiﬁed

absolute value of questioned costs identiﬁed by

 

Material Noncompliance (MNC)

Reportable Conditions (RC)

New (NEW)

US

Table 2
(continued)

1 if other than unqualiﬁed opinion issued by
auditor on compliance with federal regulations, 0
otherwise

1 if reportable condition identiﬁed by auditor, 0
otherwise

1 if ﬁrst year of data included in database is other
than 1997, 0 otherwise

Market leader audited organizations are statistically different than other brand
name audited organizations at a p-value<0.05.

Other brand name audited organizations are statistically different than non-
brand name audited organizations at a p-value<0.05.

Market leader audited organizations are statistically different than non-brand
name audited organizations at a p-value<0.05.

No signiﬁcant differences

50

Table 3
Descriptive Statistics - Comparisons Between Changes in Auditor Types

Panel A - Expected Quality Decreases
Continuous Variables - Means (Medians)
Binary Variables - Percentage of Occurrences of “1”

 

 

 

 

 

 

 

 

 

Decrease Two Decrease One
Levels of Quality1 Level of Quality2 No Shift
(N = 121) (N = 285) (N = 40,857) *

AF ederal -549,027 -1,199,273 598,027 c
Expenditures (AFED) (139,978) (46,842) (45,984)
AQuestioned Costs -28,691 -47,722 -1,050 ns
(AQC) (0) (0) (0)
AQuestioned Cost -0.006 -0.000 -0.000 ns
Percent (AQC%) (0) (0) (0)
ANumber of -2. 149 -0.702 0.205 b
Programs (APROG) (0) (0) (0)
AF indings (AFIND) -0.339 -0.091 0071 ns

(0) (0) (0)
AMaterial Non-
compliance (AMNC) 1.65% 1.05% -0.42% ns
AReportable
Conditions (ARC) 15.70% 5.26% -2.04% a,b,c

 

 

 

 

 

51

 

Table 3
(continued)

Panel B - Expected Quality Increases
Continuous Variables - Means (Medians)
Binary Variables - Percentage of Occurrences of “1”

 

 

 

 

 

 

 

 

 

 

 

 

 

 

AFederal Expenditures (FED)
AQuestioned Costs (QC)

AQuestioned Cost Percent (QC%)

ANumber of Programs (PROG)

AF indings (FIND)

AMaterial Noncompliance (MNC)

Increase Two Increase One
Levels of Quality1 Level of Quality2 No Shift
(N = 77) (N = 31L (N = 40,857) *

AFederal 6,413,512 8,115,603 598,027 (1
Expenditures (AFED) (770.433) (294,644) (45,984)
AQuestioned Costs 38,946 274,171 -1,050 ns

AQC) (0) (0) (0)
AQuestioned Cost 0.003 0.036 -0.000 ns
Percent (AQC%) (0) (0) (0)
ANumber of 2.286 2.129 0.205 ns
Programs (APROG) (0) (1) (0)
AF indings (AF IND) 0.442 0.548 -0.071 ns

(0) (0) (0)
Material Non-
compliance (AMNC) -l .30% 0% -0.42% ns
AReportable
Conditions LARC) -9.09% -6.45% -2.04% ns
LEGEND

change in total federal expenditures in year of
auditor change ~

change in absolute value of total questioned costs
identiﬁed by the auditor in year of auditor change
change in absolute value of questioned costs
identiﬁed by the auditor as a percent of total
federal expenditures in year of auditor change
change in number of different federal programs
in year of auditor change

change in total ﬁndings identiﬁed by the auditor
in year of auditor change

change in opinion on material compliance in year
of auditor change (-1 indicates change from
unqualiﬁed opinion to other-than-unqualiﬁed
opinion, 0 indicates no change, and 1 indicates
change from other-than-unqualiﬁed to
unqualiﬁed)

52

 

Table 3
(continued)

AReportable Conditions (RC) = change in reportable conditions in year of auditor

change (-1 indicates a change from no reportable
conditions to at least one reportable condition, 0
indicates no change, and 1 indicates a change
from at least one reportable condition to no
reportable conditions)

' This column provides data for organizations that changed between a market leader
auditor and a non-brand name auditor. These are organizations that changed their
expected level of quality by two levels.

2 This column provides data for organizations that changed between a market leader
auditor and an other brand name auditor or between an other brand name auditor and a
non-brand name auditor. These are organizations that changed their expected level of
audit quality by one level.

* a

US

Organizations decreasing their expected audit quality by two levels (from
market leader to non-brand name) are statistically different than organizations
decreasing their expected audit quality by one level (from market leader to
other brand name or from other brand name to non-brand name) at a two-sided
p-value<0.05.

Organizations decreasing their expected audit quality by two levels (from
market leader to non-brand name) are statistically different than organizations
not changing their auditor type at a two-sided p-value<0.05.

Organizations decreasing their expected audit quality by one level (from
market leader to other brand name or from other brand name to non-brand
name) are statistically different than organizations not changing their auditor
type at a two-sided p-value<0.05.

Organizations increasing their expected audit quality by one level (from other
brand name to market leader or from non-brand name to other brand name) are
statistically different than organizations not changing their auditor type at a
two-sided p-value<0.05.

No signiﬁcant differences

53

Table 4
Pearson Correlations

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

QC QC% MNC RC FIND ML OBN PROG LFED NEw
QC 1.000
QC% .220 1.000
MNC .063 .104 1.000
RC .033 .055 .245 1.000
FIND .267 .122 .245 .317 1.000
ML .032 .016 -.018 -.064 .067 1.000
OBN .040 .003 -.029 -.103 .047 -.045 1.000
PROG .099 -.004 .009 .017 .127 .105 .141 1.000
LFED .100 -.013 .015 -.031 .172 .209 .230 .343 1.000
NEW -.006 .004 .011 .005 -.000 -.026 -.022 -.079 -.131 1.000

 

Items in bold are signiﬁcant at the 5% or less level.

LEGEND

QC
QC %

MNC
RC
FIND
ML
OBN
PROG

LFED
NEW

questioned costs identiﬁed by the auditor at time t
questioned costs identiﬁed by the auditor as a percent of total federal

expenditures at time t

dummy variable indicating whether there was a qualiﬁed opinion on
compliance with federal regulations at time t (0 indicates an
unqualiﬁed opinion, 1 otherwise)
dummy variable indicating whether there was at least one reportable
condition in internal controls identiﬁed at time t (1 indicates at least

one internal control reportable condition, 0 otherwise)

total ﬁndings identiﬁed by the auditor at time t
dummy variable indicating auditor market leader at time t (1 if market
leader brand name, 0 otherwise)
dummy variable indicating other brand name auditor at time t (1 if
other brand name, 0 otherwise)

number of different federal programs at time t

natural log of total federal expenditures at time t
dummy variable indicating whether the current year is the ﬁrst year
that the organization is receiving federal funds (1 if the ﬁrst year

included in the database is not 1997, 0 otherwise)

54

 

Table 5

Hypothesis 1 — Questioned Costs

QC%it = Oti + BIMLit + BzOBNii + B3NEW11+ B4PROGII + BSLFEDn + BoRCii + [37Y981 +

B8Y991 + BgYOOt + 8n

 

 

 

 

 

 

 

 

 

 

 

 

 

 

N = 76,836 N = 41,269
Levels”l Changs'
Expected Std. Std.
Sign Coefﬁcient Error Coefﬁcient Error
ML + .008 *** .0019 .009 * .0071
OBN + .003 ** .0016 .003 ** .0019
NEW + -.000 .0005 -.000 .0004
PROG + .000 .0000 .000 .0000
LFED + -.001 *** .0003 -.001 .0006
RC + .004 *** .0004 .004 *** .0007
Y98 n/a -.000 .0002 n/a n/a
Y99 n/a -.000 .0002 .000 .0004
Y00 n/a -.001 .0008 -.000 .0005
constant n/ a .01 7 ** * .0049 -.000 .0003

 

 

 

 

 

 

 

R2 = .003 and .002 for levels and changes models, respectively.

ML > OBN at p-value < .005 and .133 for levels and changes model, respectively (one-

sided test)

LEGEND

*, **, *** Signiﬁcant at 10%, 5%, and 1% levels, respectively, based on one-sided t-
tests for all variables with predicted signs equal to actual signs. For variables
without predicted signs, and for those whose actual sign are different than
predicted, p-values are based on two-sided tests. P-values for the changes
model are based on standard errors robust to general serial correlation and
heteroskedasticity.

QC%it
MLit
OBN“

NEW]!

= absolute value of total questioned costs identiﬁed by the auditor as a
percent of total federal expenditures at time t
= dummy variable indicating auditor market leader at time t (1 if market
leader brand name, 0 otherwise)
= dummy variable indicating other brand name auditor at time t ( 1 if
other brand name, 0 otherwise)
= dummy variable indicating whether the current year is the ﬁrst year
that the organization is receiving federal funds (1 if the ﬁrst year
included in the database is not 1997, 0 otherwise)

55

 

PROG]:
LFED“
RC1:

Y98t-Y001

Table 5
(continued)

number of different federal programs at time t

natural log of total federal expenditures at time t

dummy variable indicating whether reportable conditions were
reported by the auditor at time t (1 if reportable condition reported, 0
otherwise)

dummy variables to indicate year at time t

8 Both models are estimated using OLS estimation. The levels model is estimated using
ﬁxed effects estimation. In the changes model, the variables in the regression are the
changes from time t-l to t for each variable as deﬁned above. The changes in dummy
variables are speciﬁed as:

AML“

AOBN“

ARC“

dummy variable indicating the change in the type of market leader
auditor from time t-l to t (-1 indicates change from market leader to
non-market leader, 0 indicates no change, and 1 indicates change from
non-market leader to market leader)

dummy variable indicating the change in the type of non-market leader
auditor from time t-l to t (-1 indicates change from other brand name
to non-brand name, 0 indicates no change, and 1 indicates change from
non-brand name to other brand name)

dummy variable indicating the change from time t-l to t in the
organization's report on internal controls (-1 indicates a change from
no reportable conditions to at least one reportable condition, 0
indicates no change, and 1 indicates a change from at least one
reportable condition to no reportable conditions)

56

Table 6
Hypothesis 2 — Findings

FIND“ = (1] + BIMLit '1' BZOBNR ‘1' B3NEWn + B4PROG“ + BsLFEDit + BoRCn + B7Y981 +

B8Y99t + B9Y00t + 8n

 

 

 

 

 

 

 

 

 

 

 

 

 

N = 76,945 N = 41,365
Levels‘l Charges”
Expected Std. Std.
Sign Coefficient Error Coefficient Error
ML + .404 *** .0856 .348 ** .1756
OBN + .237 *** .0688 .325 ** .1415
NEW + .026 .0203 .026 * .0166
PROG + .002 *** .0008 .003 ** .0016
LFED + .067 *** .0154 .096 *** .0205
RC + .907 *** .0188 .874 *** .0255
Y98 n/a -.064 *** .0100 n/a n/a
Y99 n/a -.105 *** .0107 .020 .0151
Y00 n/a -.160 *** .0352 -.016 .0246
constant n/a -.666 *** .2178 -.065 *** .0109

 

 

 

 

 

 

 

 

R2 = .06 and .05 for levels and changes models, respectively.

ML > OBN at p-value < .035 and .453 for levels and changes model, respectively (one-

sided test)

LEGEND

*, **, *** Signiﬁcant at 10%, 5%, and 1% levels, respectively, based on one-sided t-
tests for all variables with predicted signs equal to actual signs. For variables
without predicted signs, and for those whose actual sign are different than
predicted, p-values are based on two-sided tests. P-values for the changes
model are based on standard errors robust to general serial correlation and

heteroskedasticity.
FIND“ = total number of ﬁndings identiﬁed by the auditor at time t
ML“ = dummy variable indicating auditor market leader at time t (1 if market
leader brand name, 0 otherwise)
OBN,. = dummy variable indicating other brand name auditor at time t (1 if
other brand name, 0 otherwise)
NEW“ = dummy variable indicating whether the current year is the ﬁrst year

that the organization is receiving federal funds (1 if the ﬁrst year
included in the database is not 1997, 0 otherwise)
PROG“ = number of different federal programs at time t

57

 

LFED"
RC1:

Y981-Y001

Table 6
(continued)

natural log of total federal expenditures at time t

dummy variable indicating whether reportable conditions were
reported by the auditor at time t (1 if reportable condition reported, 0
otherwise)

dummy variables to indicate year at time t

3 Both models are estimated using OLS estimation. The levels model is estimated using
ﬁxed effects estimation. In the changes model, the variables in the regression are the
changes from time t-l to t for each variable as deﬁned above. The changes in dummy
variables are speciﬁed as:

AMI-4'1

AOBN]: 1

ARC“

dummy variable indicating the change in the type of market leader
auditor from time t-l to t (-1 indicates change from market leader to
non-market leader, 0 indicates no change, and 1 indicates change from
non-market leader to market leader)

dummy variable indicating the change in the type of non-market leader
auditor from time t-l to t (-1 indicates change from other brand name
to non-brand name, 0 indicates no change, and 1 indicates change from
non-brand name to other brand name)

dummy variable indicating the change from time t-l to t in the
organization's report on internal controls (-1 indicates a change from
no reportable conditions to at least one reportable condition, 0
indicates no change, and 1 indicates a change from at least one
reportable condition to no reportable conditions)

58

Hypothesis 3 — Material Noncompliance with Federal Regulations

Table 7

MNCit = 011 + BiMLii + BzOBNii + 133NEW11+ [34PROGir + 1551-1: EDir + B6RCi1 + B7FINDii

+ BBQCo/Oit + B9Y981 + BIOY991 + Bl lYOOt + 8it

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

N = 76,846 N = 41,269
Levels' Changes'
Expected Std. Std.
Sign Coefficient Error Coefﬁcient Error
ML + -.020 * .0127 -.018 * .0138
OBN + -.020 .0102 -.018 .0098
NEW + .004 .0030 .004 * .0028
PROG + .000 .0001 .000 .0001
LFED + .004 * .0023 .003 .0025
RC + .070 *** .0029 .065 *** .0055
FIND + .019 *** .0007 .019 *** .0027
QC% + .520 *** .0327 .555 *** .0987
Y98 n/a -.001 .0015 n/a n/a
Y99 n/a -.002 .0016 .001 .0024
Y00 n/a .001 .0052 .004 .0044
constant n/a -.038 .0322 -.002 .0015

 

 

 

 

 

 

 

R2 = .095 and .046 for levels and changes models, respectively.

ML < OBN at p-value < .984 and .981 for levels and changes model, respectively (two-
sided test)

LEGEND
*, **, *** Signiﬁcant at 10%, 5%, and 1% levels, respectively, based on one-sided t-
tests for all variables with predicted signs equal to actual signs. For variables
without predicted signs, and for those whose actu'al sign are different than
predicted, p-values are based on two-sided tests. P-values for the changes
model are based on standard errors robust to general serial correlation and

MNC“

MLit

heteroskedasticity.

= dummy variable indicating whether there was a qualiﬁed opinion on

compliance with federal regulations at time t (0 indicates an

unqualiﬁed opinion, 1 otherwise)
= dummy variable indicating auditor market leader at time t (1 if market
leader brand name, 0 otherwise)

59

 

OBN]!

NEW"

PROG“
LFED"
RC1:

FIND“
QC%it

Y98t-Y001

Table 7
(continued)

dummy variable indicating other brand name auditor at time t (1 if
other brand name, 0 otherwise)

dummy variable indicating whether the current year is the ﬁrst year
that the organization is receiving federal ﬁmds (1 if the ﬁrst year
included in the database is not 1997, 0 otherwise)

number of different federal programs at time t

natural log of total federal expenditures at time t

dummy variable indicating whether reportable conditions were
reported by the auditor at time t (1 if reportable condition reported, 0
otherwise) '

total number of ﬁndings identiﬁed by the auditor at time t

absolute value of total questioned costs identiﬁed by the auditor as a
percent of total federal expenditures at time t

dummy variables to indicate year at time t

3 Both models are estimated using OLS estimation. The levels model is estimated using
ﬁxed effects estimation. In the changes model, the variables in the regression are the
changes from time t-l to t for each variable as deﬁned above. The changes in dummy
variables are speciﬁed as:

AMNC“ =

AM Lil

AOBN“

ARC]!

dummy variable indicating the change in material noncompliance from
time t-l go t (-1 indicates change from unqualiﬁed opinion to other-
than-unqualiﬁed opinion, 0 indicates no change, and 1 indicates
change from other-than-unqualiﬁed to unqualiﬁed)

dummy variable indicating the change in the type of market leader
auditor ﬁ'om time t-l to t (-1 indicates change from market leader to
non-market leader, 0 indicates no change, and 1 indicates change from
non-market leader to market leader)

dummy variable indicating the change in the type of non-market leader
auditor from time t-l to t (-1 indicates change from other brand name
to non-brand name, 0 indicates no change, and 1 indicates change from
non-brand name to other brand name)

dummy variable indicating the change from time t-l to t in the
organization's report on internal controls (-1 indicates a change from
no reportable conditions to at least one reportable condition, 0
indicates no change, and 1 indicates a change from at least one
reportable condition to no reportable conditions)

60

Hypothesis 4 — Reportable Conditions in Internal Controls '

Table 8

RC1: = Oli + BIMLit + BZOBNII + B3NEW11+ B4PROG11 + BSLFEDit + BoY981 + B7Y99r +

 

 

 

 

 

 

 

 

 

 

 

 

BgYOOt '1' Sit
N = 76,945 N = 41 ,365
Levels” Changes‘
Expected Std. Std.
Sig Coefﬁcient Error Coefﬁcient Error
ML + -.136 *** .0225 -.143 *** .0304
OBN + -.091 *** .0181 -.087 *** .0223
NEW + .017 *** .0053 .018 *** .0049
PROG + .000 .0002 -.000 ** .0002
LFED + .012 *** .0041 .009 ** .0046
Y98 n/a -.020 *** .0026 n/a n/a
Y99 n/a -.038 *** .0028 .002 .0042
Y00 n/a -.069 *** .0093 -.012 .0085
constant n/a .092 .0573 -.019 *** .0026

 

 

 

 

 

 

 

 

R2 = .017 and .002 for levels and changes models, respectively.

ML < OBN at p-value < .063 and .071 for levels and changes model, respectively (two-

sided test)

LEGEND

*, **, *** Signiﬁcant at 10%, 5%, and 1% levels, respectively, based on one-sided t-
tests for all variables with predicted signs equal to actual signs. For variables
without predicted signs, and for those whose actual sign are different than
predicted, p-values are based on two-sided tests. P-values for the changes
model are based on standard errors robust to general serial correlation and
heteroskedasticity.

RC1:

MLit
OBN]!

NEWii

= dummy variable indicating whether reportable conditions were
reported by the auditor at time t (1 if reportable condition reported, 0
otherwise)

= dummy variable indicating auditor market leader at time t (1 if market
leader brand name, 0 otherwise)

= dummy variable indicating other brand name auditor at time t (1 if
other brand name, 0 otherwise)

dummy variable indicating whether the current year is the ﬁrst year

that the organization is receiving federal funds (1 if the ﬁrst year
included in the database is not 1997, 0 otherwise)

61

 

Table 8

(continued)
PROG“ = number of different federal programs at time t
LFED" = natural log of total federal expenditures at time t
Y98t-Y00t = dummy variables to indicate year at time t

3 Both models are estimated using OLS estimation. The levels model is estimated using
ﬁxed effects estimation. In the changes model, the variables in the regression are the
changes ﬁ'om time t-l to t for each variable as deﬁned above. The changes in dummy
variables are speciﬁed as:

 

ARC" dummy variable indicating the change from time t-l to t in the
organization's report on internal controls (-1 indicates a change from
no reportable conditions to at least one reportable condition, 0
indicates no change, and 1 indicates a change from at least one
reportable condition to no reportable conditions)

AML" = dummy variable indicating the change in the type of market leader
auditor from time t-l to t (-1 indicates change from market leader to
non-market leader, 0 indicates no change, and 1 indicates change from
non-market leader to market leader)

dummy variable indicating the change in the type of non-market leader
auditor from time t-l to t (-1 indicates change ﬁ'om other brand name
to non-brand name, 0 indicates no change, and 1 indicates change from
non-brand name to other brand name)

62

Table 9

Tobit Estimation of Hypotheses One and Two

Questioned Costs and Findings

QC%1I= (11 + BrMLii + BzOBNii + B3NEW11+ B4PROGit + BsLFEDn + B6RC11 + B7Y981+

B8Y99t + BoYOOt + 8n

FINDit = a: + BIMLir + BZOBNir + [33NEWir ‘1’ B4PROG11+ BsLFEDii + BoRCii + B7Y981 +

B8Y991 + BgYOOt + £11

Panel A - Tobit Estimates

 

 

 

 

 

 

 

 

 

 

 

 

 

N = 76,836 N = 76,945
Questioned Costs'I Findin s'
Expected Std. Std.
Sign Coefﬁcient Error Coefﬁcient Error
ML + .070 ** .032 1.955 *** .472
OBN + .048 ** .028 1.076 *** .395
NEW + .012 .010 .176 * .128
PROG + .000 .000 .003 .003
LFED + .009 * .007 .301 *** .093
RC + .117 *** .008 4.281 *** .101
Y98 n/a -.010 ** .004 -.294 *** .061
Y99 n/a -.015 *** .005 -.477 *** .066
Y00 n/a -.050 ** .021 -1.217 *** .251
constant n/a -.743 *** .022 -15.807 *** .280

 

 

 

 

 

 

 

 

R2 = .15 and .13 for questioned costs and ﬁndings models, respectively.

ML < OBN at p-value < .261 and .039 for questioned costs and ﬁndings models,
respectively (one-sided tests)

LEGEND

*, **, *** Signiﬁcant at 10%, 5%, and 1% levels, respectively, based on one-sided t-
tests for all variables with predicted signs equal to actual signs. For variables
without predicted signs, and for those whose actual sign are different than
predicted, p-values are based on two-sided tests. P-values for the changes
model are based on standard errors robust to general serial correlation and
heteroskedasticity.

QC%n

= absolute value of total questioned costs identiﬁed by the auditor as a
percent of total federal expenditures at time t

63

 

FIND“
MLit

OBN“
NEW“
PROG“

LFED"
RC1:

Y98t-Y001

Table 9
(continued)

total number of ﬁndings identiﬁed by the auditor at time t

dummy variable indicating auditor market leader at time t (1 if market
leader brand name, 0 otherwise)

dummy variable indicating other brand name auditor at time t (1 if
other brand name, 0 otherwise)

dummy variable indicating whether the current year is the ﬁrst year
that the organization is receiving federal funds (1 if the ﬁrst year
included in the database is not 1997, 0 otherwise)

number of different federal programs at time t

natural log of total federal expenditures at time t

dummy variable indicating whether reportable conditions were
reported by the auditor at time t (1 if reportable condition reported, 0
otherwise)

dummy variables to indicate year at time t

3‘ Coefﬁcient estimates were obtained using tobit estimation. The actual models estimated
include each of the independent variables as well as organization time-averages for each
independent variable to account for ﬁrm heterogeneity. The coefﬁcients on the averages
have not been included in this table.

Only the direction and signiﬁcance of the Tobit estimated coefﬁcients can be directly
compared to the ﬁxed effects estimates included in Tables 5 and 6. The magnitudes of
the Tobit coefﬁcients cannot be directly compared to the magnitudes of the estimates
from the ﬁxed effects models.

64

Table 10
Logit Estimation of Hypotheses Three and Four
Material Noncompliance and Reportable Conditions

MNC11= Oti + BrMLit + BzOBNir + BiNEWn + B4PROGII + BSLFEDH + 136RC11 + B7FIND11
'l' B8QC%it + B9Y98t + B10Y99t ‘1‘ BIIYOOr + 811

RC1: 0L1+ BIML11+ 132013Nit + 133NEwit + [3..PROGit + [isLFEDit + 0,1(98. + 07Y99. +

 

 

 

 

 

 

 

 

 

 

 

 

 

 

BSYOOt + 81':
Panel A - Logit Estimates
N = 76,836 N = 76,945
Material Noncom liance‘I Reportable Conditions'
Expected Std. Std.
Sign Coefﬁcient Error Coefﬁcient Error
ML + -.408 .937 -1.637 *** .341
OBN + -2.289 *** .866 -1 .253 *** .287
NEW + .259 * .183 .402 *** .096
PROG + .021 * .013 .001 .003
LFED + .143 .128 .199 *** .063
RC + 1.464 *** .119 n/a n/a
FIND + .337 *** .034 n/a n/a
QC% + 9.356 *** 2.046 n/a n/a
Y98 n/a -.029 .082 -.298 *** .041
Y99 n/a -.129 .091 -.581 *** .045
Y00 n/a -.145 .370 -1.325 *** .186

 

 

 

 

 

 

 

 

 

R2 = .19 and .02 for material noncompliance and reportable conditions models,
respectively.

ML < OBN at p-value < .068 and .318 for the material noncompliance and reportable
conditions models, respectively (two-sided test)

LEGEND

*, **, *** Signiﬁcant at 10%, 5%, and 1% levels, respectively, based on one-sided t-
tests for all variables with predicted signs equal to actual signs. For variables
without predicted signs, and for those whose actual sign are different than
predicted, p-values are based on two-sided tests. P-values for the changes
model are based on standard errors robust to general serial correlation and
heteroskedasticity.

MNC“ = dummy variable indicating whether there was a qualiﬁed opinion on

compliance with federal regulations at time t (0 indicates an

unqualiﬁed opinion, 1 otherwise)

65

MLit
OBN]!
NEW“
PROG“

LFED“
RCir

FIND];
QC%it

Y981-Y001

Table 10
(continued)

dummy variable indicating auditor market leader at time t (1 if market
leader brand name, 0 otherwise)

dummy variable indicating other brand name auditor at time t (1 if
other brand name, 0 otherwise)

dummy variable indicating whether the current year is the ﬁrst year
that the organization is receiving federal funds (1 if the ﬁrst year
included in the database is not 1997, 0 otherwise)

number of different federal programs at time t

natural log of total federal expenditures at time t

dummy variable indicating whether reportable conditions were
reported by the auditor at time t (1 if reportable condition reported, 0
otherwise)

total number of ﬁndings identiﬁed by the auditor at time t

absolute value of total questioned costs identiﬁed by the auditor as a
percent of total federal expenditures at time t I
dummy variables to indicate year at time t

a Coefﬁcient estimates obtained using conditional ﬁxed effects logit estimation. Only the
direction and signiﬁcance of the logit estimated coefﬁcients can be directly compared to
the ﬁxed effects estimates included in Tables 7 and 8. The magnitudes of the logit
coefﬁcients cannot be directly compared to the magnitudes of the estimates from the
ﬁxed effects models.

66

  
   

        

      

  

MHG

Illljtlljtllillli

321

[titty]