Client Importance, Non-Audit Services, and Abnormal Accruals

Hyeesoo Chung
(765) 496-6308
(hyeesoo_chung@mgmt.purdue.edu)

and

Sanjay Kallapur
(765) 494-4516
(kallapur@mgmt.purdue.edu)

Krannert School of Management

Purdue University
West Lafayette, IN 47907

First draft: September 2001

Current draft: December 2001

We thank Ashiq Ali, Mark Bagnoli, Kevin Den Adel, Bob Eskew, Bill Felix, Frank
Heflin, Mark Penno, Susan Watts, and seminar participants at Purdue University and
Southern Methodist University for many helpful comments on this paper.
 Client Importance, Non-Audit Services, and Abnormal Accruals

Abstract

We use data on audit and non-audit fees paid to auditors, disclosed in proxy
statements under SEC auditor independence rules, to assess whether client importance to
auditor is associated with abnormal accruals. Our sample consists of 1,864 clients of Big-
5 audit firms that revealed audit and non-audit fees in proxy statements filed between
February 5, 2001 and June 30, 2001, and had data in Compustat to estimate abnormal
accruals using the modified-Jones model. We measure client importance by the ratio of
client fees to the audit firm’s total US revenues. In view of anecdotal evidence suggesting
that non-audit services are more profitable to the auditor than are audit services, we use
the ratio of non-audit fees from a client to the audit firm’s US revenues as a second
measure of client importance. We do not find a statistically significant evidence of an
association between abnormal accruals and either of the client importance measures. We
also do not find evidence of a statistically significant association between abnormal
accruals and client importance in subsets of the sample partitioned by the extent of client
incentives to manipulate earnings (proxied by high leverage and high market to book
ratio), auditor expertise (proxied by whether the auditor is the industry specialist), and
insider influence on the board of directors (proxied by whether the CEO is the chairman,
and by the percentage of employee directors). Thus we are unable to find evidence of any
association between client importance and abnormal accruals. However, it is possible that
client importance is significantly related to abnormal accruals in other cases which we
have not considered, and our results should be interpreted subject to this caveat.

Keywords: Auditor independence, earnings management, client importance, non-audit

services.

2
 Client Importance, Non-Audit Services, and Abnormal Accruals

1. Introduction

Accounting manipulation has received a lot of attention from regulators recently.

In a widely-quoted speech, SEC Chairman Levitt (1998) expresses concern about erosion

of the quality of accounting earnings, and its ultimate effect on investor trust and market

volatility. In another speech, Levitt (2000) points to the substantial growth in consulting

services provided by auditors. This, he argues, compromises their independence, at least

in appearance if not in fact, because they are serving two masters—investors for audit,

and management for consulting. An article in Forbes (MacDonald, 2001) asserts that in

five of the greatest audit failures of the past two decades (involving Just For Feet, Oxford

Health Plans, Colonial Realty, BCCI, and DeLorean Motor Company), the pursuit of

consulting contracts compromised auditors’ independence.

In response to regulators’ concerns, the Independence Standards Board (ISB) was

established in 1997 as a private-sector body blessed by the SEC (Financial Reporting

Release, FRR, 50: Commission Statement of Policy on the Establishment and

Improvement of Standards Related to Auditor Independence) to make standards for

auditor independence. However, the SEC issued its own rules regarding auditor

independence in November 2000, and the ISB was dissolved as of July 31, 2001. The

new rules issued in November 2000 require companies to disclose separately in proxy

statements the amount of audit fees, information technology (IT) consulting fees related

to financial systems design and implementation, and fees paid to auditors for all other

services. SEC considered banning non-audit services altogether (SEC Proposed Rule:

Revision of the Commission’s Auditor Independence Requirements, file no. S7-13-00),

but eventually only required disclosure; the rationale was that disclosure would enable

investors and the board of directors to judge for themselves whether auditor

independence is impaired.

In previous evidence on the issue of non-audit services and auditor independence

impairment, Palmrose (1999) finds that non-audit services are not the basis for lawsuits

against any auditor, thus casting doubt on the proposition that the provision of non-audit

services leads to audit failures. However, in an op-ed article in the Washington Post

(Breeden et al., 2000), four former SEC chairmen argued that the danger was not so much

from black-and-white issues that auditors faced; rather, it was from the gray areas where

the pressure on auditors to bend to client interests was “subtle, but no less deleterious.”

This concern echoes that of Chairman Levitt referred to above, about erosion in the

quality of accounting earnings, implying a gradual process rather than a dramatic event.

In this paper we seek to provide evidence on the association between client

importance in terms of fees, especially those from non-audit services, and abnormal

accruals. We obtain audit and non-audit fee information from the disclosures newly

available in proxy statements as a result of the SEC rule. Consistent with prior research

(Becker et al., 1998, DeFond and Subramanyam, 1998, and Francis et al., 1999) we use

abnormal accruals estimated using the (modified) Jones model as our measure of earnings

management. Following DeAngelo (1981), our measure of client importance is the ratio

of total fees (audit and non-audit) from a client to the total US revenues of the audit firm.

Because anecdotal evidence suggests that non-audit services have a higher margin than

do audit services, we also use a second measure of client importance, the ratio of non-

audit fees from a client to the audit firm’s total US revenues. Our sample consists of all

2
companies that reveal audit and non-audit fees in proxy statements filed between

February 5 and June 30, 2001, have data on Compustat, and meet certain criteria (such as

being clients of Big 5 audit firms, and not having changed auditors during 2000,

N=1,882). By using all firms rather than restricting our sample to cases where audit

failures are a priori likely (e.g., firms that restate their financials or are subject to SEC

enforcement actions), we provide evidence germane to the concerns about “subtle, but no

less deleterious” manipulation.

After controlling for industry and determinants of abnormal accruals based on

previous studies, we do not find a significant cross-sectional association between the

absolute value of abnormal accruals and either of the client importance ratios (the ratio of

total fees received from the client to the total revenues of the audit firm, or the ratio of

non-audit fees to total auditor revenues). The lack of significance is unlikely to have

resulted from low power of the tests because previous studies have found, for example,

significant differences between abnormal accruals of firms audited by Big 6 and non-Big

6 auditors, unconditionally. Given these findings, it is unlikely that the Jones model, used

in the previous studies as well as in ours, lacks power to detect abnormal accruals. Our

findings are of interest because another contemporaneous study, Frankel et al. (2001),

concludes that there is an association between auditor independence and earnings

quality.1

1
The measure of auditor independence we use is consistent with theoretical arguments in DeAngelo (1981)
and differs from the measure used by Frankel et al. (2001). In section 4 we describe the results of our
replication of one part of Frankel et al.’s study involving abnormal accruals. We find that their results hold
only for the smallest firms when we divide our sample into three groups based on total assets; if auditors
compromise their independence, it is more likely for the largest firms where more substantial amounts of
fees are at stake. Our paper also goes beyond Frankel et al. (2001) by investigating auditor independence
for sub-samples of firms formed by client incentives to manipulate earnings, auditor industry specialization,
and board independence from managers.

3
 To test whether there is an association between client importance and abnormal

accruals for certain groups of firms, we condition on client incentives to manipulate

earnings, auditor expertise, and insider influence on the board of directors. If auditors

compromise their independence, it is more likely to be detectable in samples of clients

that have higher incentives to manipulate earnings. Auditor expertise may create client

dependence on auditor as opposed to auditor dependence on client. Finally, independent

boards might be able to mitigate the effects of impaired auditor independence. We use

leverage and market-to-book ratios as proxies for client incentives. Auditor expertise is

proxied by whether they are the industry specialist. Insider influence is proxied by

whether the CEO is the chairman of the board of directors, and the percentage of

employee directors. We do not find a significant association between the client

importance ratios and abnormal accruals for any of the partitions we consider.

Overall, our results do not enable us to conclude that client importance is

associated with abnormal accruals, either for the full sample or for the sample partitions

we consider. However, it is possible that auditor dependence is significantly related to

abnormal accruals in other cases which we have not considered, and our results should be

interpreted subject to this caveat. Also, our study is designed to present evidence on

subtle manipulation, not on audit failures. It would be worthwhile for future research to

examine whether there are statistical differences in client importance between cases

where audit failures are highly likely (e.g., in firms that restate their financials or are

subject to SEC enforcement actions), and corresponding control firms. Despite these

limitations, our findings add to the evidence on the scope of services and auditor

independence, an issue of recent public policy concern.

4
 We organize the rest of the paper as follows. In section 2 we describe the

institutional background regarding the auditor independence rules. In section 3 we

describe our data, and in section 4, our research design and findings. Because we do not

find a significant association between abnormal accruals and client importance, we

partition the sample according to client incentives to manipulate, auditor expertise, and

insider influence on the board of directors; these results are presented in section 5.

Section 6 provides a summary and concluding comments.

2. Institutional Background

The Big 5 audit firms today are multi-disciplinary firms offering many services in

addition to audit services.2 Non-audit service fees have recently been growing faster than

fees from audit services, and a majority of the Big 5 firms’ revenues are now from non-

audit services (SEC Final Rule: Revision of the Commission's Auditor Independence

Requirements; File No. S7-13-00). The concern about non-audit services to audit clients

is that the audit firm is serving two masters--shareholders and the board of directors for

audit services, and management for non-audit services--and this may compromise the

independence of the audits.

The Panel on Audit Effectiveness Report (2000) notes that concern about the

scope of audit services and its effect on auditor independence was first expressed in

SEC’s 1957 annual report. However, a more strident criticism of non-audit services was

2
ISB Discussion Memorandum 99-2, Evolving Forms of Firm Structure and Organization, identifies the
following categories of services provided by CPA firms: Non-audit attest services such as CPA WebTrust,
evaluation, design, and implementation of business controls, tax services, financial services such as credit
or treasury management, information systems technology services, designing employee compensation and
benefit programs, business re-engineering, performing information systems, tax, internal audit or other
functions outsourced by the client, corporate finance services such as advice on mergers and acquisitions,
evaluation of marketing or distribution channels, legal services, and litigation support.

5
made in the Metcalf Committee Report (1977). Following the publication of this report

the AICPA established the SEC Practice Section (SECPS), required peer reviews of audit

firms every three years, and set up the Public Oversight Board (POB) to oversee the peer

reviews and the SECPS. The SECPS adopted criteria that prevented CPA firms from

providing certain services such as executive recruitment. It also required member firms to

file annual returns providing information on the number of clients for whom management

advisory services fees as a percent of auditing fees fall in ranges 0%, 1% - 25%, 26% -

50%, 51% - 100%, and above 100%.

In 1978 the SEC issued ASR 250, Disclosure of Relationships with Independent

Public Accountants, requiring companies to disclose the ratio of fees for non-audit

services to those for audit services. Furthermore, each service for which the client was

billed at least 3 percent of audit fees was to be identified separately. POB studied the

independence issue and issued a report in 1979. The study did not find that non-audit

services compromised independence, although it found that such services were perceived

to impair independence. SEC countered that the POB report did not sufficiently sensitize

the audit profession to independence issues, and responded by issuing ASR 264, Scope of

Services by Independent Accountants, which laid down that audit firms could only

provide those non-audit services that are related to auditing. ASRs 250 and 264 were

however rescinded in 1982 and 1981 respectively; in the case of ASR 250 because “the

detailed non-audit services disclosure may not be of sufficient utility to investors to

justify continuation of the disclosure requirement,” (ASR 296: Relationships Between

Registrants and Independent Accountants) and because in the meanwhile SECPS had

6
instituted the requirement for the disclosures of the number of clients with percentages of

non-audit to audit fees falling in the limits mentioned above.3

Although non-audit services continued to increase throughout the 1980s and

1990s, it was not until the late 1990s that further regulatory action was undertaken. In

1997 the ISB was established as a private sector body to make standards on auditor

independence. POB formed the Panel on Audit Effectiveness, which issued a report in

August 2000 making many recommendations for improving audits. The Panel conducted

quasi-peer reviews, that is, in-depth reviews of the quality of 126 audits of SEC

registrants in 28 different offices of the eight largest audit firms. Regarding non-audit

services, the Panel identified 37 of the 126 reviews as ones in which the auditor had

provided services other than audit and tax. The quasi-peer reviews did not find any case

in which the provision of non-audit services had adversely affected audit effectiveness. In

a quarter of the cases where auditors had provided non-audit services, the Panel’s

reviewers concluded that the non-audit services had a positive effect on audit

effectiveness. Nevertheless, the Panel was split on the question of whether auditors

should provide non-audit services to audit clients. Some Panel members believed that

there was a fundamental conflict of interest between providing audit and non-audit

services (but these members did not object to the provision of tax services), and a total

ban on such services to audit clients would best ensure auditor independence. Other

members of the Panel felt that the long history of the audit profession did not provide a

single case of audit impairment due to non-audit services, and such a ban might even end

up reducing the effectiveness of audits. The Panel report (2000) states both these

3
The SECPS disclosures, however, are at the aggregate audit-firm level; ASR 250 disclosures were at the
individual-company level.

7
positions and the supporting reasoning, without making a recommendation one way or

the other.

In addition to the audit profession’s initiatives, in 1998 NYSE and NASDAQ co-

sponsored a Blue Ribbon Committee on Improving the Effectiveness of Corporate Audit

Committees. Many of the committee’s recommendations were incorporated into SEC

rules.4 Finally, SEC issued its auditor independence rules in November 2000. In addition

to requiring disclosure of audit, IT consulting, and other non-audit fees, the new rules

prohibit auditors from performing more than 40 percent of the internal audit work of

large clients (those having more than $200 million in total assets). Regarding disclosure

of audit and non-audit fees, the new disclosures include dollar amounts; ASR 250

disclosures were of only ratios of non-audit to audit fees. However, ASR 250 disclosures

were disaggregated by type of service (each service for which fees exceeded 3 percent of

audit fees); the new disclosures of non-audit services are disaggregated only into IT

consulting and other services.

3. Data and Descriptive Statistics

We collected from proxy statements information on audit, IT, and other fees paid

by companies to their auditors. Disclosure was required in proxy statements beginning

February 5, 2001. We collected data from proxy statements filed between February 5 and

June 30, 2001. For inter-firm comparisons of the extent of purchase of non-audit services,

4
The rules include requirements that audit committees consist solely of independent directors, have at least
three financially literate directors and at least one having accounting or related financial management
expertise. The definition of independence was made stricter. Audit committees were required to have
formal charters which were to be disclosed in proxy statements, and audit committees had to be informed
about non-audit services and to review auditor independence in light of these disclosures.

8
we scale non-audit fees by the amount of audit fees.5 The proxy statements include those

filed by mutual funds, which under SEC rules have to disclose fees for non-audit services

provided by the auditor not only to the fund, but also its adviser, and entities in a control

relationship with the adviser that provide services to the fund; because the audit fees are

for the fund only, mutual funds tend to have a high ratio of non-audit to audit fees.6 Nine

of the ten companies with the highest ratio are mutual funds. To exclude these cases, we

matched the firms with those in the 2000 Compustat database and present descriptive

statistics on the resulting sample of 2,875 firms. We obtain financial data on these firms

from Compustat. To calculate the ratio of fees received from a given client to the total

revenues of the audit firm, we require data on audit firms’ US revenues. We obtain this

information from the annual data on the top 100 CPA firms published in the Accounting

Today magazine (and its web site www.electronicaccountant.com) in 2001. The

magazine obtained this data by surveying the audit firms.

For the 2,875 sample firms, Table 1 presents descriptive statistics on total assets,

audit and non-audit fees, ratios of IT fees, other fees, and total non-audit to audit fees,

and ratios of client revenues and non-audit client revenues to audit firms’ total revenues.

Mean (median) total assets are $5,979 million ($411 million) and audit fees, $0.646

million ($0.207 million). The mean (median) total fees are $2.236 million ($0.509

million). IT fees are non-zero for approximately 10 percent of the sample (untabulated),

the mean (median) ratio of IT to audit fees is 0.19 (0.00). Non-audit fees to audit fees

5
We could have scaled by sales or total assets, but the non-audit to audit fee ratio is consistent with data
formerly disclosed under ASR 250, and is also of interest in its own right because of the perception of
auditor independence impairment (Earnscliffe 1999, 2000) when the ratio is high.
6
For example, the Irish Investment Fund, Inc. disclosed $24,500 in audit fees to PricewaterhouseCoopers,
and $3.8 million in other fees, for a non-audit to audit fee ratio exceeding 155.

9
ratio is 1.89 (1.16) at mean (median), indicating that in the mean and median case, non-

audit fees exceed audit fees. The mean ratio of client to total audit firm revenues is 0.055

percent (0.012 percent). Even at the third quartile the ratio is only 0.035 percent.7

Of the 2,875 sample firms, 2,604 are audited by Big 5 audit firms, and 271 by

others. The non-audit to audit fee ratio of Big 5 firms’ auditees is much higher than that

of the other firms, 2.01 compared to 0.74 (1.27 v. 0.48) at mean (median). In view of the

disparity, we conduct further tests on Big 5 firms’ auditees only. (Data requirements for

estimating abnormal accruals constrain our final sample to only 110 non-Big 5 audit

firms’ auditees, so we are unable to test for differences among Big 5 versus non-Big 5).

Notable differences within the Big 5 firms are that Ernst and Young has the

smallest IT to audit fee ratio, 0.08 at mean, possibly because of the divestiture of its

consulting unit to Cap Gemini in May 2000.8 Nevertheless, its non-audit to audit ratio,

2.07 at mean, was not the smallest among the Big 5 firms. Despite its separation from

Andersen Consulting as of August 7, 2000, Arthur Andersen had an IT to audit fee ratio

of 0.39 at mean, the highest among the Big 5 firms. Andersen’s non-audit to audit fee

ratio, 2.18 at mean, was also close to the largest among the Big 5 (the largest being that

for PricewaterhouseCoopers, with a mean of 2.22).

Panel D of Table 1 presents selected data for the sample partitioned according to

whether or not the firm changed auditors in 2000. Data on auditor change was obtained

7
These ratios probably overstate the true magnitude of client importance, because the proxy disclosures
include fees paid by non-US subsidiaries of US clients to foreign practices of the audit firm. The total
revenues in the denominator of the ratio are for the US practice only.
8
Non-audit fees paid to the divested consulting unit are included in the disclosed amounts if the payment
was for services rendered before May 2000, and not otherwise.

10
from Compustat.9 Consistent with DeBerg et al. (1991) the non-audit to audit fee ratio is

significantly smaller for new auditors (1.52 compared to 1.91 at mean and 0.77 compared

to 1.21 at median). Because of the differences between new auditors and those that have

audited the firm for a longer period of time, in further tests we omit firms switching

auditors.

4. Research Design and Findings

Following previous studies (Becker et al., 1998; Francis et al., 1999; DeFond and

Subramanyam, 1998) we use abnormal accruals estimated using the cross-sectional

modified-Jones model as our measure of earnings management. Bartov et al. (2000) find

that the Jones model abnormal accruals are associated with audit qualifications, and

Heninger (2001) finds that abnormal accruals are associated with litigation against

auditors, further validating the use of the Jones model in our context.10 For each two-digit

SIC code we estimate the following model:

ACC = a0 1/ TA-1 + a1 (∆SALES - ∆AR)/ TA-1 + a2 PPE/ TA-1 + e, (1)

where

9
If the auditor data was missing for either 1999 or 2000 then the firm was deemed not to have changed
auditors. Descriptive statistics are similar if such firms are omitted from the sample.
10
An alternative used by Frankel et al. (2001) is the extent to which firms just meet quarterly earnings
forecasts (or other benchmarks such as earnings in the same quarter of the previous year). In the Earnscliffe
surveys (1999, 2000) however, CEOs, CFOs, and auditors point out that there is limited involvement of
auditors in quarterly reporting. Moreover, earnings forecasts can be met by managing the forecasts or
“street” earnings, which have no audit implications, rather than GAAP earnings. Finally, although AICPA
Statement on Auditing Standards 89 (Audit Adjustments) and SEC Staff Accounting Bulletin 99
(Materiality) clarify that departures from GAAP which are small in relation to net income might still be
material (and thus forbidden) if they enable companies to meet analysts forecasts, Kinney et al. (1999)
point out that companies can meet analysts’ forecasts in other ways such as through biased provisions for
bad debts.

11
ACC = Total accruals deflated by TA-1. Accruals are defined as the difference between

earnings (before extraordinary items and discontinued operations) and cash

from operations.

TA-1 = Total assets at the beginning of the year,

SALES = Sales,

AR = Accounts receivable, and

PPE = Property, Plant, and Equipment.

Abnormal accruals, dacc, are calculated as the residual from the above model.

Model (1) is estimated using all observations on Compustat for the year 2000. Because

Levitt (1998) expresses concerns about ‘cookie-jar accounting,’ implying that firms

manage earnings upwards as well as downwards, we use the absolute value of abnormal

accruals in subsequent tests. Becker et al. (1998), however, argue, that audit firms are

never sued for permitting accruals that decrease earnings, and accordingly signed

abnormal accruals are the correct measure. We also report the results of tests using this

alternative measure.

DeAngelo (1981) argues that auditor independence from a client, the probability

of reporting a discovered breach, is a function of the percentage of the auditor’s total

quasi-rents dependent on retaining the client. She further argues that the percent of total

fees from a client is a good proxy for the ratio of quasi-rents. Accordingly, we use the

ratio of total (audit + non-audit) fees from a given client to the audit firm’s total US

revenues as a measure of client importance. Larger the fee revenues from a client, the

more it is in the auditor’s interest to retain that client. If the economic interest created by

the fee income leads auditors to approve of questionable earnings management practices

12
(Levitt, 2000; MacDonald, 2001), then we should observe a positive association between

the absolute value of abnormal accruals and the client fees to total revenues ratio.

Previous studies that use a similar ratio as a measure of client importance include Stice

(1991) and Lys and Watts (1994).11

Anecdotal evidence suggests that non-audit services have higher margins than do

audit services.12 Therefore we also use a second measure of client importance, the ratio of

non-audit fees from a client to the auditor’s total US revenues. We are unable to use the

auditor’s total non-audit revenues as the denominator because this information is not

available on the same basis as the proxy disclosures: revenue data disclosed to the SECPS

and in the Accounting Today surveys is broken down into those from audit and assurance,

tax, and management advisory services. Any audit or assurance services (e.g., WebTrust

attest services or audit of the pension plan) not directly related to the audit of the financial

statements, however, are classified as non-audit services for purposes of the proxy

disclosure. Also, if the ratio of non-audit to audit revenues is similar across the Big 5

firms,13 regression R2s and t-statistics will be unaffected.

To determine whether |dacc| is associated with auditor dependence, we estimate

the following regressions:

11
Lacking actual fee data, both studies use client sales to total sales of all auditees of the auditor as a proxy
for client importance and examine whether client importance is associated with litigation against auditors.
Lys and Watts (1994) find a positive association; Stice (1991) finds an association when his control firms
are matched by size, but not when they are matched by industry and size.
12
For example, the SEC Independence Rule states: “The increasing importance of non-audit services to
accounting firms is further evidenced by suggestions that the audit has become merely a "commodity" and
that the greater profit opportunities for auditors come from using audits as a platform from which to sell
more lucrative non-audit services.” In their analysis of the data requested by the ISB from Big 5 firms,
Antle and Gitenstein (2000) find that gross profit before partner compensation is higher for audit services,
but consulting partners are better able to participate in multiple projects and delegate responsibility to less
senior professional staff. They also argue that audit contracts last much longer than do consulting contracts
and therefore have much higher present values of future gross margins.
13
Table 1 shows that the ratio of non-audit fees from audit clients to audit fees ranges from 1.63 to 2.22 at
mean for our sample.

13
 dacc = ∑ j a j D j + b1 log(TA) + b2 OCF + b3OCF + + b4 ACC −1 + b5 ACC −+1 + b6 ROA−1 +
b7 ROA−+1 + b8 ACQ + b9 ISSUE + b10 client / rev + u ,
(2)

and

dacc = ∑ j a j D j + b1 log(TA) + b2 OCF + b3 OCF + + b4 ACC −1 + b5 ACC −+1 + b6 ROA−1 +

b7 ROA−+1 + b8 ACQ + b9 ISSUE + b10 non _ aud / rev + u ,
(3)

where

|dacc| = absolute value of abnormal accruals, the residual from model (1),

Dj = 1 if firm belongs to jth industry, and 0 otherwise,

TA = total assets,

OCF = operating cash flows ÷ TA in year -1,

OCF+ = 0 if OCF < 0, and = OCF if OCF ≥ 0,

ACC-1 = total accruals in year -1 ÷ TA in year -2,

ROA-1 = net income in year –1 ÷ TA in year –2,

ACQ = 1 if firm acquired another firm during the year (if Compustat data item 129 > 0),

and 0 otherwise,

ISSUE = 1 if the number of shares outstanding, adjusted for splits and dividends,

increases by more than 10 percent over the previous year, and 0 otherwise.

client/rev = ratio of total client fees (audit + non-audit) to audit firm’s total revenue, and

non_aud/rev = ratio of non-audit fees from the client to audit firm’s total revenue.

The industry dummies control for fixed industry effects. Kothari et al. (2001)

argue that the best way to control for the effect of performance on abnormal accruals is to

match firms by ROA in previous year. However, there are hardly any firms with no non-

14
audit services with which to match; because of this difficulty in matching we control for

performance using past ROA and current operating cash flows. Because our dependent

variable is the absolute value of abnormal accruals, we interact each of these variables

with a dummy indicating whether the variable is positive. For example, coefficient b2

represents the relationship between |dacc| and OCF for firms with negative OCF, and b2 +

b3, that for positive OCF. Thus the specification allows for different coefficients for

negative and positive OCF. We include past accruals in the regression to control for the

normal relation between accruals of successive years. We do not include current-year

ROA or current-year accruals because any earnings manipulation would affect abnormal

accruals as well as current-year ROA and accruals; using current-year accruals as an

independent variable could therefore remove the effect of interest.14 Consistent with prior

research (Firth, 1997; DeFond and Subramanyam, 1998) we control for ACQ and ISSUE

because events like acquisitions and stock issuance might be correlated with the fee

ratios, and associated with the amount of abnormal accruals.

To mitigate the effects of outliers we delete the top 1 percent of observations that

are most influential for the coefficients on client/rev and non_audit/rev. Influential

observations are identified using the |DFBETAS| statistic proposed by Belsley et al.

(1980). Because descriptive statistics in Table 1 indicate that the non-audit to audit ratios

of non-Big 5 auditees is systematically lower than that of the Big 5 auditees, we estimate

regressions using only the Big 5 auditees. For the same reason we also delete firms that

changed auditors in the year 2000. We also delete firms in the financial industry (two-

digit SIC codes 60 – 69) because abnormal accruals estimated using the Jones model are

14
As expected, the t-statistics for the coefficients on client/rev and non_aud/rev are even weaker when we
include current-year accruals as an independent variable.

15
not meaningful for these firms. These criteria, and the requirement of the availability of

data to estimate abnormal accruals using equation (1), result in a sample of 1,864 firms

(1,882 before removal of outliers).

Table 2 presents results of estimating regression equations 2 and 3. The

association between the absolute value of abnormal accruals and either of the client

importance ratios (client fees to auditor’s total revenues or client non-audit fees to

auditor’s total revenue ratio) is insignificant. In fact, the coefficients are negative, the

opposite of what would be predicted if concern about fees compromised auditor

independence. Among the control variables, OCF, OCF+, and ISSUE are associated with

|dacc| at the 5 percent level of significance or better. The regression R2s are 26 and 28

percent in regressions 2 and 3.

Sensitivity Analysis

We perform several additional tests for sensitivity analysis and summarize the

results below. First, the correlations between client/rev (non_aud/rev) and log of total

assets is 0.50 (0.46). The results are qualitatively unchanged when we include TA instead

of log(TA), when we do not include either TA or log(TA), and when we include

dummies for all but one of 10 groups of TA (to control for a non-linear relation between

size and |dacc|). Second, our results are qualitatively unchanged when we use the absolute

value of abnormal current accruals. Current accruals are defined as change in non-cash

working capital. The abnormal component is the residual from the regression of current

accruals deflated by lagged total assets on the reciprocal of total assets, and change in

sales minus change in accounts receivable, deflated by lagged total assets (i.e., PPE is

dropped from regression equation 1). Third, all results are qualitatively unchanged when

16
we use the Jones model, rather than the modified-Jones model, to estimate abnormal

accruals.

Fourth, the results are qualitatively similar when we use 1999 accruals (and audit

and non-audit fees for 2000). If auditors had been permitting earnings management

before 1999 depending on the magnitude of fees, but changed their strategy for the 2000

financials in view of the impending disclosure of audit and non-audit fees (the SEC rule

was issued in November 2000, before the end of December fiscal years), then we should

have found a relationship between abnormal accruals and client dependence ratios using

1999 data. The audit and non-audit fees pertain to 2000, not 1999, but it is likely that

auditors could have formed reasonable expectations about 2000 fee income when

attesting 1999 financials.

Fifth, results are qualitatively unchanged when the dependent variable is dacc, not

|dacc|. As described before, Becker et al., 1998, argue that dacc is the correct dependent

variable because auditors are never sued for high amounts of income-decreasing accruals.

Sixth, our results are unchanged when we measure client importance by the ratios of

client fees and non-audit fees to audit firms’ average US revenues per office. Information

on the number of offices of each audit firm is obtained from Accounting Today.

Unfortunately, information on the revenues generated by each office of the audit firms is

not separately available, so we have to use the average revenue per office as the

denominator for our ratios.

Non-audit to audit fee ratio as an auditor independence measure

The Earnscliffe survey (1999) finds that there is a perception of auditor

independence impairment when the amount of non-audit fees is large in relation to audit

17
fees. This measure has also been used in previous research by Glezen and Millar (1985),

Parkash and Venable (1993), and Firth (1997) (the first two papers used ASR 250

disclosures which did not contain dollar amounts). Use of the ratio as a measure of

auditor independence impairment would imply that, e.g., the extent of independence

impairment with respect to a company that pays audit fees of $100,000 and non-audit

fees of $200,000 is the same as that with respect to another company that pays audit and

non-audit fees of $10 million and $20 million respectively, which is unlikely. Burton

(1980) also forcefully argues that the amount of non-audit fees in relation to audit fees is

perceived to impair independence only by those who do not understand the process of

auditing. Nevertheless, in this sub-section we describe results using this alternative

measure of auditor independence.

Using a regression model similar to (2) we fail to find a significant association

between |dacc| and the non-audit to audit fee ratio (t = +0.71). These results contrast with

those of Frankel et al. who find a positive and significant coefficient. There are several

differences between their research design and ours. They do not control for fixed industry

effects, and they use a transformation of the non-audit to audit fee ratio, namely, the non-

audit to total fee ratio (= (non-audit/audit)/(1+non-audit/audit)). When we change these

two aspects of our research design, we obtain a coefficient and t-statistic similar to theirs:

coefficient = 0.12, t-statistic = 3.98. When we introduce industry dummies, the t-statistic

drops substantially (to 1.86), but is nevertheless significant at the 5 percent level (one-

tailed test). Further analysis, however, indicates that the coefficient is significant for only

the smallest group of firms, when we divide the sample into three equal-sized groups

according to total assets. The coefficient estimates (t-statistics) for small, medium, and

18
large firms are 0.09 (2.48), 0.05 (1.47), and 0.01 (0.17) respectively. As mentioned

before, if independence impairment is a function of this ratio, it is more likely to occur

with the largest firms, rather than the smallest firms. Accordingly, we do not believe that

these findings suggest independence impairment. In any case, our evidence is incremental

to that in Frankel et al. (2001), and should help readers form their own conclusions.

Discussion

An insignificant coefficient implies an inability to reject the null hypothesis of no

association between client importance and the absolute value of abnormal accruals.

However, an inability to reject the null is not the same as its acceptance: the

insignificance could be caused either by the lack of an association, or by noise resulting

in low power of the tests. The latter possibility is less likely, however, because previous

studies (Becker et al., 1998; Francis et al., 1999) are able to detect significant differences

between abnormal accruals of firms audited by Big 6 versus non-Big 6 auditors, and we

use the same method to estimate abnormal accruals. This point can be stated differently

as follows. Our coefficient on client/rev (-2.161) and its associated standard error (6.75 =

2.161/0.32) imply that the upper 95% confidence limit for the coefficient is 11.069 (=

-2.161 + 6.75*1.96). The inter-quartile range for client/rev is 0.00025 (= third quartile,

0.000298 – first quartile, 0.000048).15 Hence, the upper 95% confidence limit for the

difference in the absolute value of abnormal accruals between firms whose client/rev fall

at the third and first quartiles respectively is 0.28 percent (= 11.069*0.00025) of total

assets. Whether this difference is material or not could depend on the circumstances;

however, it is an order-of-magnitude smaller than the 3.6 percent difference (= 0.173 –

15
These values are for the regression sample. Thus they differ slightly from those presented in Table 1 for
the full sample.

19
0.137, from their Table 3) found by Francis et al. (1999) between absolute abnormal

accruals of auditees of Big 6 versus non-Big 6 firms.

Even if there is no association between client importance and abnormal accruals

unconditionally, it is possible that there is such an association under specific conditions.16

We consider the following conditioning factors that could affect the association between

client importance and abnormal accruals. First, it is possible that not all firms have

sufficiently high incentives for earnings management. Thus client importance is

irrelevant if the firm did not attempt earnings management in the first place, and

inclusion of such firms weakens our tests. Second, it is possible that in some cases, as

Arruñada (1999) argues, non-audit services can create client dependence on the auditor.

Third, it is possible that even though auditors are willing to compromise their

independence, monitoring by the board of directors prevents any effects on financial

statements. We partition the sample into sub-samples based on the above criteria in an

attempt to evaluate these explanations for the lack of association.

Our controls for client incentives to manage earnings are leverage and market-to-

book ratio. DeFond and Jiambalvo (1994) find that the Jones model abnormal accruals

are positive for firms about to violate covenants, and Press and Weintrop (1990) find that

leverage is associated with the probability of violating covenants. Thus high-leverage

firms are more likely to manage earnings. Skinner and Sloan (1999) show that the

consequences of not meeting earnings forecasts are severe for growth firms. Thus growth

16
Even in this case our findings are of interest because, as mentioned above, previous papers have found a
difference between abnormal accruals of Big 6 and non-Big 6 auditees unconditionally. Also, Frankel et al.
(2001) conclude that there is an unconditional association between earnings quality (proxied by whether or
not firms meet or beat benchmarks, and by the absolute value of Jones-model abnormal accruals), and
auditor independence (the non-audit to total fees ratio).

20
firms, proxied by high market to book ratio, are a priori more likely to engage in earnings

management. Given the benefits of a smooth pattern of earnings, they have an incentive

to manage both upwards, during bad times, and downwards, during good times.

To investigate possible client dependence caused by non-audit services, we

partition the firms into auditees of industry-specialist and non-industry-specialist

auditors. Clients of industry-specialist auditors might be less likely to replace them

because of access to their greater expertise. We designate the firm with the largest

number of auditees in each two-digit SIC code as the industry specialist, consistent with

Craswell et al. (1995).17

Effective board monitoring is more likely when insiders have less influence on the

board. Our proxies for insider influence on the board of directors are (1) whether the

CEO is the chairman, and (2) the percentage of non-employee directors. Previous studies

have found that these factors also proxy for the independence and activity of audit

committees, which are responsible for determining audit fee and independence

arrangements. Collier and Gregory (1999) find that audit committee activity is lower in

UK firms when the CEO is the chairman. Menon and Williams (1994) and Beasley and

Salterio (2001) find that audit committee independence and activity is positively

associated with the percentage of non-employee directors for US and Canadian firms

respectively. They argue that audit committee and board independence are related

because the audit committee reports to the board.

There are, however, other possibilities that we are unable to control for. For

example, it is possible that not all non-audit services have the effect of compromising

17
Craswell et al. (1995) also determine the auditors market share weighted by sales. Our results using the
alternative method are similar.

21
independence. ASR 264 mentions tax advice as an example of a service that has obvious

economies of scope with the audit function (for auditing the deferred tax accrual). If there

are economies of scope and some of the resulting cost savings are passed on to the client,

then a client would have to pay more to purchase such services from a party other than

the auditor; a threat to stop purchasing such services is therefore less credible than a

threat to stop purchasing other kinds of services. Also an expectation of recurring

purchase of a service creates more incentives for the auditor, than does a one-time

purchase. However the proxy disclosures do not separate the different kinds of services

so we are unable to control for the possibility that the inclusion of fees for such services

in the ratios weakens the power of our tests.

5. Results of Partitioning the Sample

To determine whether the client importance ratios (client/rev or non_aud/rev) are

associated with |dacc| for partitions of the sample, we form dummy variables representing

the partition. We include all but one of the dummies as independent variables in the

regression (if we included all the dummies then their sum would be collinear with the

sum of the industry dummies). We also interact the client importance ratio with the

dummy variables for the partitions. Thus, to investigate the association between |dacc|

and client/rev for firms partitioned into three groups by leverage, we estimate the

following regression:

dacc = ∑ j a j D j + b1 log(TA) + b2 OCF + b3OCF + + b4 ACC −1 + b5 ACC −+1 + b6 ROA−1 +

b7 ROA−+1 + b8 ACQ + b9 ISSUE + b10 I levg =low + b11 I levg = mid + b12 client / rev × I levg =low +
b13 client / rev × I levg = mid + b14 client / rev × I levg = high + u ,

(4)

22
where

Ilevg=low = 1 if the firm falls in the lowest group when the sample is divided into 3 groups

by levg, and 0 otherwise,

Ilevg=mid = 1 if the firm falls in the middle group when the sample is divided into 3 groups

by levg, and 0 otherwise,

Ilevg=high = 1 if the firm falls in the highest group when the sample is divided into 3 groups

by levg, and 0 otherwise, and

levg = total liabilities divided by stockholders’ equity.

Coefficients b12, b13, and b14 measure the association between client/rev and |dacc|

for firms with low, mid, and high leverage, respectively. This specification constrains the

coefficients on the control variables (but not the intercepts) to be equal for the three

groups of firms. We also relax this constraint by estimating the regressions separately for

the partitions. The results are similar; for brevity we do not present the results of separate

estimation. To estimate the relationship between |dacc| and the other client importance

ratio, non_aud/rev, we replace client/rev in regression equation (4) with non_aud/rev.

Panel A of Table 3 presents descriptive statistics on non-audit to audit fee ratio,

client/rev, and non_aud/rev for firms partitioned into three groups by leverage and by the

market-to-book ratio. The mean and median non-audit to audit fee ratio are lowest for the

middle group of firms partitioned by leverage, but mean and median client/rev and

non_aud/rev increase with leverage. All ratios increase with the market to book ratio.

Panel B presents coefficient estimates and t-statistics for regression equation 4. The

coefficients on client/rev and non_aud/rev interacted with dummy variables for each of

23
leverage partitions are insignificant; thus there is no evidence of an association between

|dacc| and client/rev or non_aud/rev even for firms with high leverage. Similarly the

coefficients on client/rev and non_aud/rev interacted with dummies for each of the

market-to-book partitions are insignificant. The regression R2s and coefficients on the

control variables and their associated t-statistics in panels B and C are similar to those in

Table 2.

Panel A of Table 4 presents statistics indicating that industry-specialist auditors

enjoy a higher mean and median ratio of non-audit to audit fees, consistent with their

greater expertise. The regression coefficients on client/rev and non_aud/rev interacted

with dummies for auditees of industry-specialist and non-specialist auditors are not

significant. The results are similar (not tabulated) when the specialist auditor is defined as

the one whose clients have the largest combined sales in the industry. Thus there is no

evidence of an association between |dacc| and client/rev or non_aud/rev even for auditees

of non-industry-specialist auditors.

Results presented in panel A of Table 5 indicate that the mean non-audit to audit

fee ratio does not differ much between firms whose CEO is the chairman and firms

whose CEO is not the chairman (2.01 compared to 2.05). The non-audit to audit fee ratio

increases as the percentage of outside directors increases. This finding is consistent with

that of Parkash and Venable (1993), in that firms better able to cope with the resulting

agency problems have higher non-audit to audit fee ratios. The coefficients on client/rev

or non_aud/rev are also insignificant for firms whose CEO is the chairman (panel B), and

for firms with low percentage of outside directors (panel C). Thus we fail to find a

24
significant association between abnormal accruals and auditors’ client dependence for

any of the partitions we consider.

6. Conclusions

In this paper we test whether the Jones-model abnormal accruals of audit clients

are associated with the client’s importance to the auditor. Our measures of client

importance are the ratios of total fees (audit and non-audit) from a client to the total US

revenues of the audit firm, and of non-audit fees from the client to the audit firm’s total

US revenues. After controlling for industry and determinants of abnormal accruals based

on previous studies, we do not find a significant cross-sectional relationship between the

absolute value of abnormal accruals and either of the client importance ratios. Our

findings of lack of significance are unlikely to be attributable to low power of the tests,

given previous studies’ findings that the Jones-model abnormal accruals differ

significantly for firms audited by Big 6 and non-Big 6 auditors. We also do not find a

significant association between the client importance ratios and abnormal accruals for

partitions of the sample based on client incentives to manipulate earnings, auditor

expertise, and insider influence on the board of directors.

Overall, our results do not enable us to conclude that client importance is

associated with abnormal accruals, either for the full sample or for the sample partitions

we consider. However, it is possible that auditor dependence is significantly related to

abnormal accruals in other cases which we have not considered, and our results should be

interpreted subject to this caveat. Data limitations prevent us from distinguishing between

recurring and non-recurring non-audit services, and between services that enjoy

significant economies of scope with the audit function, and those that do not. Also, our

25
study is designed to address the issue of subtle manipulation of earnings, and not of audit

failures. Nevertheless, our findings add to the evidence on the scope of audit firm

services and auditor independence, an issue of recent public policy concern.

26
 References

Antle, R., and M. Gitenstein (2000). Analysis of Data Requested by the Independence

Standards Board from the Five largest Accounting Firms. Independence Standards

Board.

Arruñada, B. (1999). The Economics of Audit Quality (Boston: Kluwer Academic

Publishers).

Bartov, E., F. A. Gul and J. S. L. Tsui (2000). Discretionary-Accruals Models and Audit

Qualifications. Journal of Accounting and Economics, 30 (3): 421-52.

Beasley, M. and S. Salterio (2001). The Relationship Between Board Characteristics and

Voluntary Improvements in Audit Committee Composition and Experience.

Contemporary Accounting Research, forthcoming.

Becker, Connie, Mark DeFond, James Jiambalvo and K.R. Subramanyam (1998). The

Effect of Audit Quality on Earnings Management. Contemporary Accounting

Research, 15(1): 1-24.

Belsley, D., E. Kuh, and R. Welsch (1980). Regression Diagnostics (New York: John

Wiley & Sons).

Breeden, R., R. Hills, D. Ruder, and H. Williams, 2000. Accounting for Conflicts. The

Washington Post (7/21/2000).

Burton, J. (1980). A Critical Look at Professionalism and Scope of Services. Journal of

Accountancy, 149(4): 48-56.

Collier, P. and A. Gregory (1999). Audit Committee Activity and Agency Costs. Journal

of Accounting and Public Policy, 18: 311-32.

Craswell, A. T., J. R. Francis and S. L. Taylor (1995). Auditor Brand Name Reputations

27
 and Industry Specializations. Journal of Accounting & Economics, 20(3): 297-

322.

DeAngelo, L. E. (1981). Auditor Size and Audit Quality. Journal of Accounting and

Economics, 3: 183-99.

DeBerg, C. L., S. E. Kaplan and K. Pany (1991). An Examination of Some Relationships

Between Non-Audit Services and Auditor Change. Accounting Horizons, 5(1):

17-28.

DeFond, M. L. and J. Jiambalvo (1994). Debt Covenant Violation and Manipulation of

Accruals. Journal of Accounting and Economics, 17 (1-2): 145-76.

DeFond, M. L. and K. R. Subramanyam (1998). Auditor Changes and Discretionary

Accruals. Journal of Accounting and Economics, 25(1): 35-67.

Earnscliffe Communications and Research (November, 1999). Report to the United

States Independence Standards Board: Research into Perceptions of Auditor

Independence and Objectivity.

Earnscliffe Communications and Research (July, 2000). Report to the United States

Independence Standards Board: Research into Perceptions of Auditor

Independence and Objectivity Phase II.

Firth, M. (1997). The Provision of Non-Audit Services by Accounting Firms to Their

Audit Clients. Contemporary Accounting Research, 14(2): 1-21.

Francis, J., E. Maydew and H. C. Sparks (1999). The Role of Big 6 Auditors in the

Credible Reporting of Accruals. Auditing, 18(2): 17-34.

Frankel, R., M. Johnson and K. Nelson (2001). Auditor Independence and Earnings

Quality. Working paper.

28
Glezen, G. W. and J. A. Millar (1985). An Empirical Investigation of Stockholder

Reaction to Disclosures Required by ASR No. 250. Journal of Accounting

Research, 23(2): 859-70.

Heninger, W. (2001). The Association Between Auditor Litigation and Abnormal

Accruals. Accounting Review, 76(1): 111-26.

Kinney, W., D. Burgstahler and R. Martin (1999). The Materiality of Earnings Surprise.

Working paper.

Kothari, S. P., A. Leone and C. Wasley (2001). Performance Matched Discretionary

Accruals. Working paper.

Levitt, A. (1998). The Numbers Game: Speech at the NYU Center for Law and Business.

Levitt, A. (2000). Renewing the Covenant with Investors: Speech at the NYU Center for

Law and Business.

Lys, T. and R. L. Watts (1994). Lawsuits Against Auditors. Journal of Accounting

Research, 32 (Supplement), 65-93.

MacDonald, E. (2001). Was that Auditor’s Opinion Really Independent? Forbes

(3/19/01).

Menon. K. and J. D. Williams (1994). The Use of Audit Committees for Monitoring.

Journal of Accounting and Public Policy, 13(2): 121-39.

Panel on Audit Effectiveness (2000). Report and Recommendations.

Palmrose, Z-V. (1999). Empirical Research on Auditor Litigation: Considerations and

Data. Sarasota, FL: AAA.

Parkash, M. and C. F. Venable (1993). Auditee Incentives for Auditor Independence:

The Case of Nonaudit Services. The Accounting Review, 68(1): 113-33.

29
Press, E. G. and J. B. Weintrop (1990). Accounting-Based Constraints in Public and

Private Debt Agreements: Their Association with Leverage and Impact on

Accounting Choice. Journal of Accounting and Economics, 12 (January): 65-95.

Skinner, D. and R.G. Sloan (1999). Earnings Surprises, Growth Expectations and Stock

Returns. Working Paper.

Stice, J. D. (1991). Using Financial and Market Information to Identify Pre-Engagement

Factors Associated with Lawsuits Against Auditors. The Accounting Review,

66(3): 516-33.

30
 Table 1
Descriptive statistics

Panel A: All Firms

Variable Mean SD Q1 Median Q3
Total Assets ($ M) 5979.30 34963.40 108.34 410.61 1837.50
Audit Fees ($ '000) 646.39 1864.84 111.00 206.57 492.00
Total Fees ($ '000) 2236.49 7035.73 213.28 509.00 1426.47
All Firms (N=2,875) IT/Audit Ratio 0.19 1.39 0.00 0.00 0.00
Other/Audit Ratio 1.70 2.05 0.48 1.10 2.21
Non/Audit Ratio 1.89 2.55 0.49 1.16 2.37
Client/rev x 1000 (N=2,785) 0.55 1.64 0.05 0.12 0.35

Panel B: Big 5 versus Non-Big 5 Auditors

Total Assets ($ M) 6552.80 36687.28 123.95 471.19 2128.00
Audit Fees ($ '000) 701.13 1950.83 125.00 235.00 544.95
Total Fees ($ '000) 2446.57 7360.47 256.25 587.67 1587.50
Big 5 (N=2604) IT/Audit Ratio 0.21 1.45 0.00 0.00 0.00
Other/Audit Ratio 1.80 2.11 0.54 1.21 2.33
Non/Audit Ratio 2.01 2.64 0.56 1.27 2.51
Client/rev x 1000 (N=2,604) 0.45 1.34 0.05 0.11 0.29
Total Assets ($ M) 468.59 1587.33 28.79 123.24 423.22
Audit Fees ($ '000) 120.38 144.05 51.50 77.72 122.50
Total Fees ($ '000) 217.87 313.70 75.70 124.38 226.00
Non-Big 5 (N=271) IT/Audit Ratio 0.02 0.13 0.00 0.00 0.00
Other/Audit Ratio 0.72 0.82 0.22 0.48 0.90
Non/Audit Ratio 0.74 0.83 0.24 0.48 0.97
Client/rev x 1000 (N=181) 1.95 3.67 0.31 0.65 1.99

Panel C: Breakdown by Big 5 Auditor

IT/Audit Ratio 0.39 2.51 0.00 0.00 0.00
AA (N=545)
Non/Audit Ratio 2.18 3.49 0.57 1.26 2.52

DT (N=388) IT/Audit Ratio 0.16 0.73 0.00 0.00 0.00

Non/Audit Ratio 1.63 1.95 0.46 1.00 1.91

EY (N=566) IT/Audit Ratio 0.08 0.52 0.00 0.00 0.00

Non/Audit Ratio 2.07 2.44 0.66 1.43 2.69

KPMG (N=457) IT/Audit Ratio 0.17 0.96 0.00 0.00 0.00

Non/Audit Ratio 1.74 2.07 0.41 1.07 2.26

PWC (N=648) IT/Audit Ratio 0.22 1.40 0.00 0.00 0.00

Non/Audit Ratio 2.22 2.67 0.62 1.48 2.87

31
 Table 1, contd.

Panel D: Firms that did and did not Change Auditors in Year 2000
Variable Mean SD Q1 Median Q3
Audit Fees ($ '000) 660.08 1908.46 113.10 210.00 498.00
No Auditor Change
(N=2,677) Total Fees ($ '000) 2305.95 7241.54 221.64 520.00 1466.00
Non/Audit ratio 1.91 2.55 0.51 1.21 2.39
Audit Fees ($ '000) 461.25 1105.83 86.00 161.89 397.04
Auditor Change
(N=198) Total Fees ($ '000) 1297.39 2980.96 150.00 357.23 911.00
Non/Audit ratio 1.52 2.53 0.30 0.77 1.85

IT/Audit ratio = IT consulting fees paid to the auditor ÷ audit fees,

Other/Audit ratio = all other fees paid to the auditor ÷ audit fees,
Non/Audit ratio = (IT fees + other fees) ÷ audit fees
Client/rev = total fees received from a particular client ÷ total revenues of the audit firm.
Revenues for some of the auditors in our sample are unavailable.

32
 Table 2
Parameter estimates and t-statistics for regression equations:

dacc = ∑ j a j D j + b1 log(TA) + b2 OCF + b3 OCF + + b4 ACC −1 + b5 ACC −+1 + b6 ROA−1 +

b7 ROA−+1 + b8 ACQ + b9 ISSUE + b10 client / rev + u ,
and

dacc = ∑ j a j D j + b1 log(TA) + b2 OCF + b3OCF + + b4 ACC −1 + b5 ACC −+1 + b6 ROA−1 +

b7 ROA−+1 + b8 ACQ + b9 ISSUE + b10 non _ aud / rev + u,

Model 1 Model 2
Parameter Parameter
Estimate t-statistic Estimate t-statistic
log(TA) -0.004 -0.81 -0.003 -0.63
OCF -0.188 -11.16 -0.192 -12.09
+
OCF 0.457 6.70 0.479 7.48
ACC-1 -0.001 -0.27 -0.001 -0.28
+
ACC-1 0.021 0.34 0.030 0.54
ROA-1 0.001 0.33 0.000 0.28
ROA-1+ -0.034 -0.57 -0.041 -0.73
ACQ 0.013 0.85 0.007 0.49
ISSUE 0.112 7.11 0.105 7.11
Client/rev -2.161 -0.32
Non_aud/rev -3.148 -0.42
2
R 0.263 0.280
N 1864 1864

|dacc| = absolute value of abnormal accruals = ê the estimate of the residual from the model:
ACC = a0 1/ TA-1 + a1 (∆SALES - ∆AR)/ TA-1 + a2 PPE/ TA-1 + e.
ACC = Total accruals ÷ TA-1. Accruals are defined as net income before extraordinary items and
discontinued operations minus cash from operations.
TA-1 = Total assets at the beginning of the year,
SALES = Sales,
AR = Accounts receivable,
PPE = Property, Plant, and Equipment,
Dj = 1 if firm belongs to jth industry, and 0 otherwise,
TA = total assets,
OCF = operating cash flows ÷ TA-1,
OCF+ = 0 if OCF < 0, and = OCF if OCF ≥ 0,
ACC-1 = total accruals in year -1 ÷ TA in year -2, and

33
ROA-1 = net income in year –1 ÷ TA in year –2.
ACC+ and ROA+ are defined similar to OCF+.
ACQ = 1 if firm acquired another firm during the year (if Compustat data item 129 > 0), and 0 otherwise,
ISSUE = 1 if the number of shares outstanding, adjusted for splits and dividends, increases by more than 10
percent over the previous year, and 0 otherwise.
Client/rev = ratio of total client fees (audit + non-audit) to audit firm’s total revenue, and
Non_aud/rev = ratio of non-audit fees from the client to audit firm’s total revenue.

34
 Table 3
Parameter estimates and t-statistics for regression of |dacc| on control variables and
client importance ratios: firms grouped by leverage and market-to-book ratio.

Panel A: Descriptive Statistics for Firms Grouped by Leverage and by Market-to-Book Ratio
By Leverage:
Variable Mean SD Q1 Median Q3
Non/Audit ratio 2.11 2.12 0.66 1.54 2.84
Leverage = Low Client/rev x 1000 0.14 0.30 0.04 0.08 0.14
(N=627) Non_aud/rev x 1000 0.09 0.22 0.02 0.04 0.09
Non/Audit ratio 1.87 2.52 0.53 1.12 2.24
Leverage = Mid Client/rev x 1000 0.46 1.41 0.05 0.12 0.31
(N=628) Non_aud/rev x 1000 0.34 1.24 0.02 0.06 0.20
Non/Audit ratio 2.16 3.20 0.58 1.29 2.53
Leverage = High Client/rev x 1000 0.79 1.90 0.07 0.21 0.73
(N=627) Non_aud/rev x 1000 0.56 1.39 0.03 0.12 0.48
By Market-to-Book Ratio:
Non/Audit ratio 1.71 2.37 0.44 1.05 2.07
MB = Low Client/rev x 1000 0.27 0.82 0.04 0.09 0.21
(N=625) Non_aud/rev x 1000 0.19 0.68 0.01 0.04 0.14
Non/Audit ratio 1.98 2.25 0.63 1.28 2.53
MB = Mid Client/rev x 1000 0.48 1.26 0.06 0.13 0.36
(N=626) Non_aud/rev x 1000 0.35 1.10 0.02 0.07 0.23
Non/Audit ratio 2.45 3.18 0.76 1.68 2.88
MB = High Client/rev x 1000 0.64 1.89 0.05 0.13 0.39
(N=625) Non_aud/rev x 1000 0.46 1.40 0.02 0.08 0.26

35
 Table 3, contd.
Panel B: Parameter estimates and t-statistics for the regressions:
dacc = ∑ j a j D j + b1 log(TA) + b2 OCF + b3OCF + + b4 ACC −1 + b5 ACC −+1 + b6 ROA−1 +
b7 ROA−+1 + b8 ACQ + b9 ISSUE + b10 I levg =low + b11 I levg = mid + b12 imp × I levg =low +
b13 imp × I levg = mid + b14 imp × I levg = high + u ,

Model 1 Model 2
Parameter Parameter
Estimate t-statistic Estimate t-statistic
log(TA) -0.003 -0.56 -0.001 -0.22
OCF -0.186 -11.03 -0.190 -11.95
+
OCF 0.460 6.72 0.478 7.44
ACC-1 -0.001 -0.32 -0.001 -0.34
+
ACC-1 0.018 0.31 0.029 0.52
ROA-1 0.001 0.37 0.001 0.34
+
ROA-1 -0.033 -0.56 -0.045 -0.80
ACQ 0.013 0.89 0.008 0.55
ISSUE 0.111 6.82 0.101 6.62
Ilevg=low 0.006 0.29 0.022 1.11
Ilevg=mid -0.027 -1.46 -0.016 -0.93
Client/rev * Ilevg=low -42.973 -1.07
Client/rev * Ilevg=mid -2.347 -0.25
Client/rev * Ilevg=high -1.392 -0.16
Non_aud/rev * Ilevg=low -51.627 -1.04
Non_aud/rev * Ilevg=mid -5.723 -0.54
Non_aud/rev * Ilevg=high -0.617 -0.06
2
R 0.265 0.282
N 1864 1864

36
 Table 3, contd.
Panel C: Parameter estimates and t-statistics for the regressions:
dacc = ∑ j a j D j + b1 log(TA) + b2 OCF + b3OCF + + b4 ACC −1 + b5 ACC −+1 + b6 ROA−1 +
b7 ROA−+1 + b8 ACQ + b9 ISSUE + b10 I mb =low + b11 I mb = mid + b12 imp × I mb =low +
b13 imp × I mb = mid + b14 imp × I mb = high + u ,

Model 1 Model 2
Parameter Parameter
Estimate t-statistic Estimate t-statistic
log(TA) -0.004 -0.84 -0.003 -0.60
OCF -0.190 -11.06 -0.194 -12.04
+
OCF 0.465 6.73 0.494 7.60
ACC-1 -0.001 -0.28 -0.001 -0.30
+
ACC-1 0.011 0.19 0.018 0.33
ROA-1 0.001 0.35 0.001 0.32
ROA-1+ -0.025 -0.42 -0.029 -0.53
ACQ 0.014 0.93 0.008 0.55
ISSUE 0.112 6.91 0.107 7.05
Imb=low 0.011 0.55 0.021 1.12
Imb=mid 0.014 0.75 0.022 1.27
Client/rev * Imb=low 0.063 0.00
Client/rev * Imb=mid -3.882 -0.36
Client/rev * Imb=high -0.653 -0.07
Non_aud/rev * Imb=low -0.989 -0.06
Non_aud/rev * Imb=mid -7.409 -0.62
Non_aud/rev * Imb=high 0.582 0.06
2
R 0.263 0.281
N 1864 1864

Imp equals Client/rev in model 1, and Non_aud/rev in model 2.

Ilevg=low = 1 if firm falls in the lowest group when the sample is divided into 3 groups by leverage, and 0
otherwise,
Ilevg=mid = 1 if firm falls in the middle group when the sample is divided into 3 groups by leverage, and 0
otherwise,
Ilevg=high = 1 if firm falls in the highest group when the sample is divided into 3 groups by leverage, and 0
otherwise,

37
Imb=low = 1 if firm falls in the lowest group when the sample is divided into 3 groups by market-to-book
ratio, and 0 otherwise,
Imb=mid = 1 if firm falls in the middle group when the sample is divided into 3 groups by market-to-book
ratio, and 0 otherwise,
Imb=high = 1 if firm falls in the highest group when the sample is divided into 3 groups by market-to-book
ratio, and 0 otherwise,
levg = total liabilities divided by stockholders’ equity, and
mb = market-to-book ratio.
See table 2 for definitions of other variables.

38
 Table 4
Parameter estimates and t-statistics for regression of |dacc| on control variables and
client importance: auditees of industry specialist and non-specialist auditors.

Panel A: Descriptive Statistics for Firms Audited by Industry Specialist versus Non-Industry-Specialist
Auditors
Variable Mean SD Q1 Median Q3
Non/Audit ratio 1.91 2.22 0.56 1.25 2.37
Non-Industry-Specialist Client/rev x 1000 0.45 1.32 0.05 0.11 0.30
(N=1,246) Non_aud/rev x 1000 0.32 1.11 0.02 0.06 0.19
Non/Audit ratio 2.32 3.34 0.66 1.47 2.77
Industry-Specialist Client/rev x 1000 0.50 1.55 0.05 0.11 0.30
(N=627) Non_aud/rev x 1000 0.35 1.09 0.02 0.06 0.20
Panel B: Parameter estimates and t-statistics for the regressions:
dacc = ∑ j a j D j + b1 log(TA) + b2 OCF + b3OCF + + b4 ACC −1 + b5 ACC −+1 + b6 ROA−1 +
b7 ROA−+1 + b8 ACQ + b9 ISSUE + b10 I nonspec + b11imp × I nonspec + b12 imp × I spec + u

Model 1 Model 2
Parameter Parameter
Estimate t-statistic Estimate t-statistic
log(TA) -0.004 -0.92 -0.003 -0.75
OCF -0.188 -11.14 -0.191 -12.06
OCF+ 0.452 6.62 0.474 7.39
ACC-1 -0.001 -0.27 -0.001 -0.28
+
ACC-1 0.023 0.38 0.032 0.57
ROA-1 0.001 0.34 0.000 0.28
+
ROA-1 -0.039 -0.67 -0.046 -0.83
ACQ 0.014 0.96 0.008 0.60
ISSUE 0.110 6.99 0.103 6.97
Inonspec -0.002 -0.16 -0.007 -0.50
Client/rev * Inonpec -4.894 -0.58
Client/rev * Ispec 1.873 0.20
Non_aud/rev * Inonpec -6.389 -0.67
Non_aud/rev * Ispec 0.879 0.08
R2 0.264 0.281
N 1864 1864

39
Imp equals Client/rev in model 1, and Non_aud/rev in model 2.
Inonspec = 1 if the firm’s auditor is not the industry specialist (the auditor with the greatest number of clients
in the two-digit SIC code to which the firm belongs), and 0 otherwise, and
Ispec = 1 if the firm’s auditor is the industry specialist, and 0 otherwise.
See table 2 for definitions of other variables.

40
 Table 5
Parameter estimates and t-statistics for regression of |dacc| on control variables and
client importance: firms whose CEO is and is not the chairman, and firms grouped
by percentage of outside directors.

Panel A: Descriptive Statistics for Firms whose CEO is or is not the Chairman, and Firms Grouped by
Percentage of Non-employee Directors.
Firms whose CEO is or is not the Chairman:
Variable Mean SD Q1 Median Q3
Non/Audit ratio 2.01 2.27 0.64 1.35 2.53
CEO is Chairman Client/rev x 1000 0.53 1.45 0.06 0.13 0.38
(N=1,121) Non_aud/rev x 1000 0.37 1.10 0.02 0.08 0.24
Non/Audit ratio 2.05 2.83 0.57 1.28 2.51
CEO is not Chairman Client/rev x 1000 0.35 1.22 0.04 0.09 0.21
(N=721) Non_aud/rev x 1000 0.25 0.97 0.02 0.05 0.14
By Percentage of Non-employee Directors:
Non/Audit ratio 1.80 2.06 0.55 1.23 2.28
%Outdir = Low Client/rev x 1000 0.35 1.21 0.04 0.09 0.22
(N=644) Non_aud/rev x 1000 0.25 0.98 0.01 0.05 0.14
Non/Audit ratio 2.02 2.42 0.56 1.33 2.60
%Outdir = Mid Client/rev x 1000 0.38 1.20 0.05 0.10 0.24
(N=635) Non_aud/rev x 1000 0.28 1.01 0.02 0.05 0.16
Non/Audit ratio 2.28 3.00 0.73 1.45 2.84
%Outdir = High Client/rev x 1000 0.67 1.66 0.07 0.17 0.53
(N=560) Non_aud/rev x 1000 0.47 1.16 0.03 0.09 0.33

41
 Table 5, contd.
Panel B: Parameter estimates and t-statistics for the regressions:
dacc = ∑ j a j D j + b1 log(TA) + b2 OCF + b3 OCF + + b4 ACC −1 + b5 ACC −+1 + b6 ROA−1 +
b7 ROA−+1 + b8 ACQ + b9 ISSUE + b10 I ceo _ not _ chmn + b11imp × I ceo _ not _ chmn + b12 imp × I ceo _ is _ chmn + u

Model 1 Model 2
Parameter Parameter
Estimate t-statistic Estimate t-statistic
log(TA) -0.004 -0.83 -0.003 -0.59
OCF -0.186 -11.12 -0.189 -12.10
+
OCF 0.436 6.41 0.459 7.21
ACC-1 0.000 -0.10 0.000 -0.10
ACC-1+ 0.015 0.25 0.024 0.44
ROA-1 0.000 0.09 0.000 0.03
+
ROA-1 -0.024 -0.41 -0.031 -0.57
ACQ 0.019 1.29 0.013 0.96
ISSUE 0.113 7.17 0.106 7.17
Iceo_not_chmn 0.010 0.71 0.015 1.12
Client/rev * Iceo_not_chmn -4.964 -0.47
Client/rev * Iceo_is_chmn 0.210 0.02
Non_aud/rev * Iceo_not_chmn -6.882 -0.57
Non_aud/rev * Iceo_is_chmn -0.439 -0.05
2
R 0.264 0.282
N 1864 1864

42
 Table 5, contd.
Panel C: Parameter estimates and t-statistics for the regressions:
dacc = ∑ j a j D j + b1 log(TA) + b2 OCF + b3OCF + + b4 ACC −1 + b5 ACC −+1 + b6 ROA−1 +
b7 ROA−+1 + b8 ACQ + b9 ISSUE + b10 I % outdir =low + b11 I % outdir = mid + b12 imp × I % outdir =low +
b13 imp × I % outdir = mid + b14 imp × I % outdir = high + u

Model 1 Model 1

Parameter Parameter
Estimate t-statistic Estimate t-statistic
log(TA) -0.003 -0.54 -0.002 -0.35
OCF -0.188 -11.23 -0.191 -12.22
OCF+ 0.437 6.42 0.460 7.22
ACC-1 0.000 -0.08 0.000 -0.10
+
ACC-1 0.013 0.23 0.023 0.42
ROA-1 0.000 0.05 0.000 -0.02
+
ROA-1 -0.026 -0.44 -0.033 -0.60
ACQ 0.017 1.14 0.011 0.78
ISSUE 0.111 7.02 0.104 7.03
I%utdir=low 0.044 2.39 0.044 2.62
I%outdir=mid 0.021 1.18 0.015 0.93
Client/rev * I%outdir=low -7.960 -0.65
Client/rev * I%outdir=mid -3.574 -0.30
Client/rev * I%outdir=high 3.414 0.35
Non_aud/rev * I%outdir=low -7.230 -0.53
Non_aud/rev * I%outdir=mid -3.971 -0.32
Non_aud/rev * I%outdir=high 2.486 0.21
2
R 0.266 0.285
N 1864 1864

Imp equals Client/rev in model 1, and Non_aud/rev in model 2.

Firms with the same %outdir are classified into the same group, therefore groups of firms with %outdir =
low, mid, and high are not of the same size.
Iceo_not_chmn = 1 if CEO is not the chairman, and 0 otherwise,
Iceo_is_chmn = 1 if CEO is the chairman, and 0 otherwise,
I%outdir=low = 1 if firm falls in the lowest group when the sample is divided into 3 groups by %outdir, and 0
otherwise,
I%outdir=mid = 1 if firm falls in the middle group when the sample is divided into 3 groups by %outdir, and 0
otherwise,

43
I%outdir=high = 1 if firm falls in the highest group when the sample is divided into 3 groups by %outdir, and 0
otherwise, and
%outdir = non-employee directors as a percentage of total directors.
See table 2 for definitions of other variables.

44