34
The Effect of Fee Dependence on Non-Big 5 Clients’ Accruals Ken Reichelt PhD Student School of Accountancy University of Missouri - Columbia Columbia, MO 65211 Tel: 573-884-2488 Email: [email protected] and Jere R. Francis School of Accountancy University of Missouri - Columbia Columbia, MO 65211 Tel: 573-882-5156 Fax: 573-882-2437 Email: [email protected] December, 2002 Comments are welcome. Please do not quote without permission

How to Write Results

Embed Size (px)

Citation preview

Page 1: How to Write Results

The Effect of Fee Dependence on Non-Big 5 Clients’ Accruals

Ken Reichelt PhD Student

School of Accountancy University of Missouri - Columbia

Columbia, MO 65211 Tel: 573-884-2488

Email: [email protected]

and

Jere R. Francis School of Accountancy

University of Missouri - Columbia Columbia, MO 65211

Tel: 573-882-5156 Fax: 573-882-2437

Email: [email protected]

December, 2002

Comments are welcome. Please do not quote without permission

Page 2: How to Write Results

ii

The Effect of Fee Dependence on Non-Big 5 Clients’ Accruals

Abstract

Prior research has investigated the effect of fees on economic dependence for Big 5 clients. However, fee dependence is potentially a greater problem for smaller non-Big 5 clients as individual clients are potentially more important. We examine whether non-Big 5 auditor fees influence abnormal accruals reported by audit clients. We separately regress discretionary accruals and total accruals on the proportion of client fees to office and firm level fees, the non-audit fee to total fee ratio, and interaction terms. We find that there is no significant economic dependence resulting from fees either at the office or firm level and no evidence that higher proportions of non-audit fees (to total fees) are significantly associated with higher discretionary and total accruals except for extreme outliers. We do find that single office firms appear to be more conservative than multi-office firms. This suggests that the influence of client size and non-audit fees proportion may not impair non-Big 5 auditor judgment surrounding their clients’ abnormal accruals. Compared to Big-5 clients, non-Big 5 clients are less conservatively biased and have larger abnormal accruals suggesting lower audit quality. Key Words: Auditor independence; Non-audit services; Earnings Management; Non-Big 5 accounting firms; Earnings Management

Page 3: How to Write Results

1

The Effect of Fee Dependence on Non-Big 5 Clients’ Accruals

Introduction We examine whether non-Big 5 auditor fees influence audit quality by examining whether

client size or the proportion non-audit service fees effects abnormal accruals reported by audit clients.

DeAngelo's (1981) seminal paper, Auditor Size and Audit Quality, argues that larger audit firms

supply a higher quality audit because they have 'more to lose' in reputation despite the quasi-rents that

are incurred through audit fee dependence. Non-Big 5 audit firms have been losing SEC client

market share and may be more willing to compromise audit quality. At the audit firm’s office level,

SEC clients make up a larger proportion of revenues, suggesting greater economic dependence than

Big 5 SEC audit clients. Reynolds and Francis (2000) find no evidence that economic dependence

causes Big 5 auditors to report more favorably for larger clients in their offices; however, they do not

examine clients of non-Big 5 firms.

The existence of audit firms providing non-audit services to their audit clients has raised

concerns whether audit firms more favorably report for clients who pay larger non-audit fees (relative

to total audit firm fees), as evidenced by an SEC ruling and the recent signing of the Sarbanes Oxley

Act that both restrict auditors providing certain non-audit services. As well, auditor fees are now

required to be reported by publicly traded companies in the definitive proxy statement for filings after

February 5, 2001 according to SEC Rule 2-01 of Regulation S-X, paragraph (c)(4)(iii) (SEC, 2001).

Previous research of non-audit service fees, examining clients of both Big 5 and non-Big 5 audit

firms, have different results (Frankel, Johnson and Nelson, 2002; Reynolds and Ke, 2001a;

Ashbraugh, LaFond and Maydew, 2002), leaving the issue unresolved.

This study of non-Big 5 audit firms is important since the client size and non-audit fee

arguments would seem to be more apparent. The market share for non-Big 5 audit firms is small,

Page 4: How to Write Results

2

approximately 5% of all SEC client revenues which has been declining for the past decade, thus

increasing competitive pressure and possibly impairing their independence. The non-Big 5 audit firms

have fewer SEC audit clients per office and per firm, possibly inducing greater economic dependence

on these clients for both audit fees and non-audit service fees. The non-Big 5 audit market share can

be divided into 2nd tier firms (approximately 65% of non-Big 5 SEC client revenues consisting of the

3 largest firms who serve the "middle market", and 48% of all non-Big 5 audit clients) and 3rd tier

firms(35% of non-Big 5 SEC client revenues and 52% of non-Big 5 audit clients) who serve smaller

SEC audit clients. A surprisingly large number of 3rd tier firms have only one SEC audit client (more

than one half of the firms) which may impose a greater risk of impaired independence since there is

less diversification of audit risk.

This paper examines the effect of audit fee economic dependence and the provision of non-

audit service fees on audit quality for non-Big 5 audit firms, using abnormal accruals (discretionary

accruals and total accruals) as a measure of auditor independence. We use a sample of 344 non-Big 5

audited firms (that excludes extreme outliers from the 5th and 95th percentile of total accruals) while

using actual fee data to perform centered regression model tests, tests of abnormal accrual variances,

and paired mean tests.

There are essentially three findings. First, we find no evidence that economic dependence

causes non-Big 5 auditors to report abnormal accruals more favorably for larger clients at either the

audit office level or audit firm level. There does appear to be some conservative bias for larger clients,

particularly for single office firms, while multi-office firms appear somewhat neutral. However, this

conservative bias is weaker than what Reynolds and Francis (2000) found for Big 5 firms. We also

find that very large clients (greater than the 3rd quartile) are conservatively biased, as well. Second,

there does not appear to be strong evidence that non-Big 5 audit firms more favorably report abnormal

Page 5: How to Write Results

3

accruals for clients with higher proportions of non-audit service fees. However, we do find that a small

number of firms (the outliers, n=31,that were excluded from the sample) with extremely high or low

total accruals have some evidence that larger proportions of non-audit fees are associated with more

favorable abnormal accruals. Third, we find that the magnitude of abnormal accruals is higher on

average for non-Big 5 audited clients than for Big-5 clients of similar size and industry, suggesting

there is lower audit quality for non-Big 5 audited clients since they are given greater discretion for

abnormal accruals.

This paper contributes to the auditor independence literature by providing evidence that non-

Big 5 firms may not be effected by client size and non-audit service fees, and that auditor

conservatism is greater for single office firms and less so for multi-office firms.

Incentives of Non-Big 5 Audit Firms Previous research suggests that non-Big 5 auditors may be compromised by the size of the audit

client. Non-Big 5 audit firms have lower audit quality (DeAngelo 1981), less brand name image

(Francis and Wilson, 1988), higher litigation activity (Palmose, 1988), lower comparable fees

(Craswell, Francis and Taylor, 1995; Francis and Wilson, 1998; Ashbaugh, LaFond and Maydew,

2002), smaller clients with greater business risk (Schwartz and Soo, 1996; Francis and Reynolds,

2001b), are less industry specialized (Hogan and Jeter, 1999), and are losing market share (Hogan and

Jeter, 1999). In 2001, US second-tier firms (Grant Thornton, BDO Seidman, and McGladrey and

Pullen), who have approximately one half of all non-Big 5 firm SEC audit clients, together accounted

for only 5.9 percent of all SEC audit clients, while the Big 5 firms held the remaining 94.1 percent of

the market share, according to the Public Accounting Report (Firms, 2001). Given these facts, there

may be economic fee dependence, especially “where an office or partner was receiving a material

percentage of revenues from a single client or group of clients" (Wallman, 1996). Reynolds and

Page 6: How to Write Results

4

Francis (2000) argue that at the office level, there is a greater likelihood that auditors compromise

independence because the number of clients are fewer while the amount of revenue from large

publicly traded clients is a greater proportion of total office fees. Table 1 shows that second tier firms

not only have a lower market share of SEC audit clients (5.9 percent vs. 94.1 percent) but fewer SEC

clients per office on average (18 vs. 23). This suggests that they may be more willing to compromise

audit quality not only to gain back market share but also the loss of an SEC client may result in a

larger loss of office fees than by a Big 5 firm office.

[INSERT TABLE 1 HERE]

The issue of whether non-audit service fees compromise auditor independence has been

popular in recent years. Consequently, the SEC issued a ruling effective February 5, 2001 limiting

certain non-audit services1 performed by the audit firm, and requiring all listed companies to report in

the definitive proxy filing three categories of fees paid to auditors: 1) audit fees, 2) information

technology services consulting fees, and 3) other non-audit fees charged by the auditor (SEC, 2001).

Empirical results are mixed for whether audit fees, non-audit fees and total audit fees effect audit

quality. Ashbaugh et al. (2002) find that higher proportions of non-audit fees result in more

conservative reporting (higher income decreasing accruals), and are less likely with firms meeting

benchmark earnings. Francis and Ke (2001a) find no evidence that higher levels of non-audit fees

cause Big 5 auditors to manage earnings to meet benchmark earnings. Frankel, et al. (2002) find the

opposite where firms purchasing more non-audit fees report larger absolute discretionary accruals and

are more likely to just meet or beat analyst forecasts. Ashbaugh et al. (2002) find that non-Big 5

auditors permit higher positive discretionary accruals than Big 5 auditors. Frankel et al. (2002) does

not find any significant difference between Big 5 and non-Big 5 auditors.

1 The SEC limits the auditor performing bookkeeping, certain valuation services, and directly operating or supervising a client's information system (SEC, 2001).

Page 7: How to Write Results

5

Hypothesis development

Previous studies have shown that audit quality is reflected by the extent of earnings

management. Becker, Defond, Jiambalvo and Subramanyam (1998) find that non-Big 6 auditors

report discretionary accruals that increase income relatively more than the discretionary accruals

reported by Big 6 auditors. Defond and Subramanyam (1998) find some evidence that audit clients

who switch from a Big 6 to a non-Big 6 auditor have larger discretionary accruals that are more

negative in the final year of the predecessor auditor than with the successor auditor, suggesting that

non-Big 6 auditors are less conservative with discretionary accruals. It would seem reasonable that

non-Big 6 auditors are more tolerant of earnings management, a sign of lower audit quality. More

tolerance for discretionary accruals may be caused by economic dependence at both the office level

and at the firm level.

H1: Economic dependence causes auditors to more favorably report abnormal accruals for

larger clients who are audited by non-Big 5 auditors, relative to the size of the office and/or

firm.

The size of non-audit service fees is a growing and controversial issue as to whether the size of

non-audit fees impairs auditor independence by creating economic dependence. Beck, Frecka and

Solomon (1988) extends the DeAngelo (1981) model and demonstrate that non-audit services provide

client-specific quasi rents to the incumbent auditor. In the sample used for this study (n=344), non-

audit fees comprise 27 percent of total fees on average. Non-audit fees, like audit fees, incur start-up

costs (caused by learning curves) that may influence an auditor's decision to detect and report an error

in the financial statements, if it could mean the loss of future non-audit fees.

H2: Larger non-audit service fees, proportionate to total fees, cause auditors to more favorably

report abnormal accruals for clients who are audited by non-Big 5 auditors.

Page 8: How to Write Results

6

Sample and measures of economic dependence

Sample - There are three types of data used in this study: fee data, auditor data, and company

financial data. The starting point for collecting these data types was to identify all companies audited

by a non-Big 5 auditor (all auditors except Arthur Anderson, Deloitte and Touche, Ernst and Young,

KPMG, and PriceWaterhouseCoopers), which we obtained from 1998, 1999 and 2000 Compustat

data files. Fee data are mainly collected from proxy statements, that are made public on the website,

www.freeedgar.com, and a smaller number from Emerson Research’s fee database. Auditor data,

such as the name of the client's auditor and their office location, were obtained from the company’s

annual 10-K filing from the same website. Company financial data was collected primarily from

Compustat, with a smaller portion from Compact Disclosure, both from the 2000 Compustat year

(June 1, 2000 to May 31, 2001 fiscal year-ends), with selected 1999 Compustat data required for

lagged variables.

The study uses a refined sample (n = 344) that excludes extreme outliers in the 5th and 95th

percentiles of total accruals (net income before extraordinary items less cash flow from operations), in

order to compute discretionary and total accruals that are comparable with previous studies. Initially

2,358 non-Big 5 audited companies were selected from the 1998, 1999, and 2000 Compustat files and

Emerson database, but after eliminating companies which either did not report fees on the SEC

website (942) or had switched to a Big 5 auditor (238), there were 1,178 companies remaining with

actual fee data and auditor data. This sample of 1,178 firms was used in computing the economic

dependence variables, discussed in the next heading, in order to obtain a more complete and accurate

measure. Table 2 provides a calculation of the final sample (n=344) and the data attrition from

eliminating financial companies, companies with missing/incomplete data, and observations with less

Page 9: How to Write Results

7

than four companies per two-digit SIC code. Firms in the financial industry (SIC codes 6000 to 6999)

were excluded because data are not complete for some variables included in the OLS regression tests.

Observations with less than four companies per two-digit SIC code are eliminated from the sample

since OLS regression cannot be performed. Table 11 compares Big 5 audited clients with non-Big 5

audited clients and with the final sample (n=344).

[INSERT TABLE 2 HERE]

Measures of Economic Dependence - We measure economic dependence, the test variable,

by four measures: fee dependence at the office level, fee dependence at the firm level, proportion of

non-audit fees to total fees, and the interaction of these variables. The variable INFLUENCE captures

fee dependence, arising from client size at the office level, and is measured as the proportion of client

total fees (audit and non-audit fees) to total audit firm fees (audit and non-audit fees) at the office

level. The variable TINFLUENCE captures fee dependence, arising from client size at the firm level,

and is measured as the proportion of client total fees (audit and non-audit fees) to total audit firm fees

(audit and non-audit fees) at the firm level. In order to obtain the most complete and accurate

measure, these are calculated on the basis of the total 1,178 non-Big 5 firms that reported fee data. As

a sensitivity measure, we also estimate economic fee dependence using only total client audit fees

(instead of total client fees) with the variables INFLUENCE2, the proportion of client audit fees to

total office level audit fees and TINFLUENCE2, the proportion of client audit fees to total audit firm

fees. To capture the effect that non-audit service fees may induce economic fee dependence, we

introduce the variable FEERATIO, which is the ratio of client non-audit fees to client total fees. To

capture the interaction effect of these variables, we also introduce the following four interaction

variables which are the product of these variables: INFLUENCE*FEERATIO,

TINFLUENCE*FEERATIO, INFLUENCE2*FEERATIO, and TINFLUENCE2*FEERATIO.

Page 10: How to Write Results

8

Model specification

Two dependent variable measures are used to test the two hypotheses: total accruals and

discretionary accruals. Total accruals (TACC) are defined as net income before extraordinary items

less cash flow from operations. Discretionary accruals (DACC) are measured using the Jones (1991)

model modified for cross-sectional industry variation (see DeFond and Jiambalvo, 1994; and Defond

and Subramanyam, 1998). For both measures, we examine three aspects of accruals for evidence of

earnings management: positive signs (income increasing), negative signs (income decreasing), and

absolute value (magnitude). Prior research suggests that firms may manage earnings with income

increasing or income decreasing accruals to optimize bonus plan targets (Healy, 1985), avoid debt

covenant violations (DeFond and Jiambalvo, 1994), and seek import protection relief from the

International Trade Commission (ITC) (Jones, 1991).

Discretionary accruals (DACC) are defined as the residual error term (observed less predicted

value) from the following OLS regression model:

TAijt/Aijt-1 = β0jt [1/Aijt-1] + β1jt [∆REVijt/Aijt-1] + β2jt[PPEijt/Aijt-1] + εijt, (1)

where

TAijt/Aijt-1 = total accruals scaled for lagged assets (period t-1) for company i in industry j,

Aijt-1 = lagged total assets for company i in industry j,

∆REVijt/Aijt-1 = the change in revenue scaled for lagged assets for company i in industry j,

PPEijt/Aijt-1 = gross property plant and equipment scaled for lagged assets for company i in

industry j, and

εijt = is the residual error term for company i in industry j in time t.

Page 11: How to Write Results

9

This model is scaled for lagged assets to control for the size of the audit client, and is computed cross-

sectionally by industry using the first two digits of the SIC code based on 639 firm observations as

detailed in table 2.

The following least squares regression models are then used to test the two hypotheses:

DACC = β0 + β1INFLUENCE + β2FEERATIO + β3INFLUENCE*FEERATIO + β4OCF + β5SALES +

β6DEBT + β7PBANK + ε (2)

DACC = β0 + β1TINFLUENCE+ β2FEERATIO + β3TINFLUENCE*FEERATIO + β4OCF + β5SALES +

β6DEBT + β7PBANK + ε (3)

TACC = β0 + β1INFLUENCE + β2FEERATIO + β3INFLUENCE*FEERATIO + β4OCF + β5SALES +

β6DEBT + β7PBANK + ε (4)

TACC = β0 + β1TINFLUENCE+ β2FEERATIO + β3TINFLUENCE*FEERATIO + β4OCF + β5SALES +

β6DEBT + β7PBANK + ε (5)

Where

DACC = discretionary accrual (the residual error term of equation 1)

TACC = total accrual (Net Income before extraordinary items less cash flow

from operations),

INFLUENCE = the proportion of client total fees to the office level total fees,

TINFLUENCE = the proportion of client total fees to total audit firm fees,

FEERATIO = the ratio of client non-audit fees to client total fees,

INFLUENCE*FEERATIO = the product of INFLUENCE and FEERATIO,

TINFLUENCE*FEERATIO = the product of TINFLUENCE and FEERATIO,

Page 12: How to Write Results

10

OCF = operating cash flow scaled for lagged assets,

SALES = the log of client sales ($000),

DEBT = the ratio of total debt to total assets, and

PBANK = the probability of bankruptcy measured by the Altman Z-score2.

We also perform the same four least squares regression models using the alternative measures

of economic dependence: INFLUENCE2 and TINFLUENCE2. By having two measures of economic

dependence at the office level (INFLUENCE and INFLUENCE2) and the firm level (TINFLUENCE

and TINFLUENCE2), and by measuring DACC and TACC with three methods: absolute value,

negative values and positive values, we use twenty four regressions to test the two hypotheses.

To measure income smoothing, we perform four tests comparing variances of partitioned

samples for the two types of influence level (office level and audit firm level) times the two types of

accruals (discretionary and total accruals). Defond and Park (1997) find evidence of income

smoothing, suggesting earnings management may depend on future prospects. In these tests, we

partition the sample into two sub-samples based on the median value of the influence variable (one

sub-sample is above the median and the other is below the median). We then compare the variances

of the two sub-samples for statistical significance. To compare audit quality between Big 5 and non-

Big 5 audited firms, we compare total accruals and discretionary accruals between Big 5 audited firms

and non-Big 5 firms matched by industry and similar sales revenue to determine whether there is any

difference between Big 5 and non-Big 5 discretion for their client’s abnormal accruals.

2 Altman z-score = (0.717 * net working capital/assets + 0.847 * retained earnings/assets + 3.107 * earnings before interest and taxes/assets + 0.42 * book value of equity/liabilities + 0.998 * sales/assets). This is the revised Z’ model that is more suitable for private firms where market value of shares is not readily available (Altman, 1983). Smaller values indicate a lower probability of bankruptcy.

Page 13: How to Write Results

11

Results

Descriptive Statistics

Descriptive statistics of the dependent and independent variables are reported in Table 3.

Compared to the descriptive statistics of Francis and Reynolds (2000), mean discretionary accruals

(0.1423 vs. 0.087) and total accruals (0.174 vs. 0.098) are similar. They are also comparable to

Becker, et al. (1998) who report for non-Big 6 audited clients 0.170 for mean absolute valued

discretionary accruals and 0.221 for mean absolute valued total accruals. Regarding other

independent variables, INFLUENCE is probably higher because Non-big 5 audit firms have fewer

clients than the Big 5 audit firms; while OCF, SALES, DEBT, and PBANK are probably lower

because of the smaller size and higher client risk. Also worth noting, is that 51 firms (14.8 percent)

have a zero FEERATIO where there are no non-audit services fees. Table 11 compares Big 5 firms to

non-big 5 firms and the final sample of 344.

[INSERT TABLE 3 HERE]

Test results

When testing hypothesis 1 for client size fee dependence at the office level, we find some

evidence that there is a conservative bias, rejecting hypothesis 1 at the office level. Referring to tables

4 and 5, we find that there are three cases in six where there is statistical significance3 of the

INFLUENCE variable in the regression models, absolute valued DACC, absolute valued TACC, and

negative TACC. As well, the same six regressions on INFLUENCE2 reveal similar results4.

However the signs of the coefficients suggest that there is conservative bias. Absolute valued DACC

and TACC are negatively signed suggesting that larger clients of audit offices may have smaller

3 For this paper, we define statistical significance were p-values are less than 0.10. 4 Significant regressions results on INFLUENCE2 are as follows: negative TACC (p=0.0418), absolute valued TACC (p = 0.0053), and negative TACC (p = 0.0763).

Page 14: How to Write Results

12

magnitudes of abnormal accruals. Similarly with negative TACC the coefficient is positive,

suggesting that larger clients of audit offices are more likely to have smaller income decreasing

accruals. As well, the variance of accruals is not significant at the office level (table 8), suggesting

that there is no income smoothing. Therefore, hypothesis 1 is rejected at the office level due to

evidence of auditor conservatism.

[INSERT TABLES 4 AND 5 HERE]

When testing hypothesis 1 for client size fee dependence at the firm level, we find stronger

evidence of conservatism, rejecting hypothesis 1 at the firm level. Referring to tables 6 and 7, we find

that in four cases out of six there is statistical significance of the TINFLUENCE coefficient in the

regression models: absolute valued DACC, negative DACC, absolute valued TACC, and negative

TACC. As well, the same six regressions on TINFLUENCE2 reveal similar results5. The significant

coefficients for TINFLUENCE AND TINFLUENCE2 are signed the same as with office level results,

suggesting conservative bias. However, variance of accruals tests are only significant for TACC

(Table 8), suggesting some income smoothing. Therefore, hypothesis 1 is rejected at the firm level due

to evidence of auditor conservatism.

[INSERT TABLES 6 TO 8 HERE]

When examining hypothesis 2, non-audit fee dependence, we find no significant evidence

supporting hypothesis 2. Referring to tables 4 to 7, we find that in eleven out of twelve tests there is

no evidence that the test variable, FEERATIO, is significant, except for negative DACC at the office

level. When INFLUENCE2 and TINFLUENCE2 are included as test variables, all 12 tests are not

significant. Therefore, we cannot accept hypothesis 2.

5 Significant regression results on TINFLUENCE2 are as follows: absolute valued DACC (p = 0.0001), negative DACC (p <0.0001), absolute valued TACC (p = 0.0022), and negative DACC (p = 0.0004).

Page 15: How to Write Results

13

Sensitivity Analysis

Heteroskedasticity

All OLS regression results are tested for heteroskedasticity using the test of 1st and 2nd

moments (SPEC option in SAS) whereby the null hypothesis specifies that the results are

homoskedastic. All test results with p-values < 0.10 are adjusted using White’s standard errors and

normal z-scores, according to Hayashi (2000).

Multicollinearity and Centering of Test Variables

All 24 regressions discussed in the results section are centered to reduce multicollinearity,

arising from the use of interaction terms. In most cases, after centering, there is no significant

multicollinarity since all VIF (Variance Inflation Factors) are less than 10 and all Pearson correlation

coefficients are less than 0.306.

Centering is performed on all test variables and involves reducing the observed value by the

sample mean for all test variables ( xxX ii −= ). In most cases, after centering, there are no significant

VIF values (> 10) or Pearson correlation coefficients (> 0.30) for all test variables in all 24 regressions

discussed in the results section. According to Neter, et al (1996), high multicollinearities may exist

between predictor variables when interaction terms are added, and centering may reduce this

multicollinearity. Non-centered regressions have significant Pearson coefficients between interaction

terms and the other test variables. For example, a non-centered regression of equation 2 (absolute

valued DACC regressed at the office level), results in Pearson coefficients of

INFLUENCE*FEERATIO with INFLUENCE and FEERATIO of 0.65 and 0.63, respectively. For

6 except in 3 out of 24 regressions (between FEERATIO and INFLUENCE*FEERATIO (r = 0.33) from positive TACC regressed on INFLUENCE, between TINFLUENCE and TINFLUENCE*FEERATIO (r = 0.37) from negative TACC

Page 16: How to Write Results

14

the same centered regression, Pearson coefficients are -0.08 and 0.08, respectively. Because of the

increased power of centered regressions, we center the remainder of the sensitivity tests.

Signed Results

We also performed four centered regressions using signed values of TACC and DACC as

dependent variables and found contradictory results for hypotheses 1 and 2. We found significant p-

values for INFLUENCE, INFLUENCE2, TINFLUENCE and TINFLUENCE27 with positive

coefficients, supporting hypotheses 1. We do not find any support for hypothesis 2 since FEERATIO

is not significant in all four tests.

Combined Financial Data from 1999

Because of the small sample (n = 344), a new sample was added by matching 1999 Compustat

company financial data to the 2000 fee and auditor data. Similar but weaker centered regression

results occur with the use of the combined 1999/2000 financial data set, suggesting that 1999

company financial data are not similar to 2000 financial data for this sample.

Multi and Single Office Firms

We further test whether the number of offices in an audit firm effect the results, by partitioning

the sample into single office firms (n = 117) and multi office firms (n = 227), and we find evidence

suggesting conservative bias for single office firms but not for multi office firms. For single office

firm tests, INFLUENCE is significant in four out of six tests8, and INFLUENCE2 is significant in

regressed on TINFLUENCE, and between TINFLUENCE2 and TINFLUENCE2*FEERATIO (r = 0.36) from negative TACC regressed on TINFLUENCE2) 7 Significant regression results are as follows: INFLUENCE: DACC (p = 0.0019) and TACC (p = 0.0048); INFLUENCE2: DACC (p = 0.0040) and TACC (p = 0.0081); TINFLUENCE: DACC (p = 0.0016) and TACC (p = 0.0004); and TINFLUENCE2: DACC (p = 0.0004) and TACC (p = 0.0003). 8 Significant regression results are as follows: absolute valued DACC (p = 0.0068), negative valued DACC (p = 0.0073), absolute valued TACC (p =0.0004), and negative TACC (p = 0.0007).

Page 17: How to Write Results

15

three tests out of six9, while the signs of the significant coefficients are the same as in tables 4 to 5,

suggesting some conservative bias and no support for hypothesis 1. For single office firms,

FEERATIO is only significant in four out of twelve tests10 with negative coefficient signs suggesting

that hypothesis 2 is not supported due to lack of evidence. FEERATIO is slightly higher for multi-

office firms (0.299) vs. single office firms (0.176). Multi office firm test results show no significant

test variables, rejecting hypothesis 1 and 2 and suggesting neutral bias.

Size Indicator Variable

To test for whether the size of the firm (number of offices) is significant, we add a

dichotomous variable SIZE, which is given the value of 1 for firms with three or more offices, and 0

otherwise, to the twenty-four centered regressions, and find no conclusive results. The size variable is

not significant in all twenty-four centered regressions, possibly due to high multicollinearity between

the variables INFLUENCE, TINFLUENCE, INFLUENCE2, and TINFLUENCE2. The same results

occur when SIZE is defined as 1 for audit firms with four or more offices and 0 otherwise.

Dichotomous Influence Variable

In order to test for whether very large clients effect the results, INFLUENCE and

TINFLUENCE are assigned a 1 if greater than the 3rd quartile and 0 otherwise, and we find evidence

suggesting conservative bias for very large clients thus rejecting hypothesis 1 and no evidence to

support hypothesis 2. TINFLUENCE is significant in 4 out of 6 tests11, and with the same coefficient

signs as in the main results. Similar results are found when INFLUENCE2 and TINFLUENCE2 are

9 Significant regression results are as follows: absolute valued DACC (p = 0.0075), negative DACC (p = 0.0028) and absolute valued TACC (p = 0.0894), with the same coefficient signs found in tables 4 to 5. 10 Significant regressions results when FEERATIO is regressed with INFLUENCE are as follows: positive valued DACC (p = 0.0455), and negative TACC (p = 0.0737), and when regressed with INFLUENCE2: positive DACC (p = 0.0458), and negative TACC (p = 0.0840). Significant FEERATIO coefficients are negatively signed.

Page 18: How to Write Results

16

centered regressed. No significant results are found to support hypothesis 2. In another test we assign

the INFLUENCE and TINFLUENCE variables a value of 1 if above the median and 0 otherwise and

find no conclusive results.

Paired means test

A paired means test is performed to test whether there are significant differences in

discretionary and total accruals between non-Big 5 audited firms and Big-5 audited firms in the same

industry with similar sales revenue. This test finds that DACC and TACC are significantly higher for

non-Big 5 audited firms than Big-5 audited firms in the same industry with similar sales revenue.

Table 9, shows that p-values for the paired t-tests for DACC and TACC are very significant. The

results are similar to Becker, et al (1998) and the test matches firms by industry and similar sales

revenue.

[INSERT TABLE 9 HERE]

Outlier Sample

We performed 24 separate regressions with only the outliers (n = 31 after removing missing

values, the 5th and 95 percentile of TACC), that were originally excluded from the other tests in this

study, and we found some weak support for hypothesis 2 for firms with extreme total accruals. When

INFLUENCE and TINFLUENCE are included in the centered regressions, FEERATIO is significant

11 Significant regression results are as follows: absolute valued DACC (p = 0.0020), negative DACC (p = 0.0082), absolute valued TACC (p = 0.0009), and negative TACC (p = 0.0017).

Page 19: How to Write Results

17

in five out of twelve tests12. When INFLUENCE2 and TINFLUENCE2 are included in the centered

regressions, FEERATIO is significant in two out of twelve tests13.

Discussion

There are essentially three findings in this study. First, non-Big 5 auditors do not seem to

favorably report abnormal accruals for larger clients either at the office level or at the firm level.

There does appear to be some conservative bias with single office firms for larger clients, which is

unexpected, while multi-office firms appear somewhat neutral. we also find that very large clients

(greater than the 3rd quartile) are conservatively biased, as well. Second, there does not appear to be

strong evidence that non-Big 5 audit firms more favorably report abnormal accruals for clients with

higher proportions of non-audit service fees, except for some outlier firms. Tests of outlier firms (a

separate sample of 31 firms with extremely high or low total accruals) suggest some evidence that

larger proportions of non-audit fees result in more favorable abnormal accruals. Third, we find that

the magnitude of abnormal accruals is higher for non-Big 5 audited clients than for Big-5 clients.

The first finding of conservative bias for non-Big 5 audited firms is consistent with findings

by Reynolds and Francis (2000) for Big 5 audited firms. However the findings of this study are

weaker because p-values are weaker for absolute valued and negative accruals while positive valued

accruals do not appear to be significant in most cases. This appears to be consistent with DeAngelo’s

(1981) argument that larger audit firms supply a higher audit quality. However, single office firms

appear to be conservatively biased, while multi-office firms appear neutral, contradicting DeAngelo’s

argument. One likely explanation is that clients audited by single office audit firms are more likely to

12 Significant regression results are as follows: absolute valued DACC at the office level (p = 0.0506), positive DACC at the office level (p = 0.0864), absolute DACC at the firm level (p = 0.0942), positive DACC at the firm level (p = 0.0982), and positive TACC at the firm level (p =0.0434). 13 Significant regression results are as follows: absolute valued DACC at the office level (p = 0.0157), and absolute valued DACC at the firm level (p = 0.0509).

Page 20: How to Write Results

18

result in business failure, since the mean PBANK is lower for single office firms vs. multi office firms

(-1.5801 vs. -1.3167, respectively), so single office firms may be more conservative to avoid litigation.

As well, single office firms are smaller and may have more engagement partners who are also the

principal partners of the firm, thus fewer principal-agency conflicts. Another possible explanation is

that the multi-office firms are very competitive and are more likely to be less conservative in order to

gain market share. As discussed earlier, non-Big 5 audit firms have been losing market share to big-5

audit firms. As well, there are fewer competitors that make up a larger portion of the non-big 5 client

market. Table 10 shows that the top 10 non-big 5 firms (all multi-office firms) account for 80.6% of

total fee revenue of all non-Big 5 firms in the sample, suggesting that single office firms may position

themselves into a separate market niche.

[INSERT TABLE 10 HERE]

A second finding is that there is little evidence of non-audit fee dependence. This suggests

that non-Big 5 auditor’s decisions for management’s abnormal accruals may not be effected by the

proportion of non-audit fees. We do find that there is some evidence that non-audit fees are significant

with a small portion of outlier firms (< 10 percent of the sample) that have extremely high or low total

accruals. Ashbaugh, et. al (2002) find a negative and marginally significant relation between

discretionary accruals and fee ratio for all SEC audit clients tested (Big 5 and non-Big 5), when using

the industry cross-sectional Jones model. This suggests that for non-Big 5 audited firms, a larger

proportion of non-audit fees (relative to total fees) may not effect the magnitude of abnormal accruals

reported by management.

The third finding is that absolute values of discretionary accruals and total accruals, scaled for

lagged assets, are significantly higher for non-Big 5 audited firms than for Big-5 audited firms, based

on matched pair t-tests. This corroborates with Becker, et al, who find that non-Big 6 auditors report

Page 21: How to Write Results

19

discretionary accruals that increase income relatively more than those reported by Big 6 auditors

(Becker, DeFond, Jiambalvo, and Subramanyam, 1998). This provides evidence that non-Big 5 firms

may be less objective than big-five firms, since they permit greater discretion to their client’s

abnormal accruals.

One of the limitations of this study is the low power of the tests; consequently the results may

be more significant with more data. We have a small sample (n =344) for the 2000 Compustat year,

which is only a fraction of the total data set for all non-Big 5 audited firms (due to data attrition) and

this may reduce the power of our tests due to the smaller sample size. Secondly, there are fewer

clients at the office level for non-Big 5 firms than Big 5 firms, which reduce the power of tests at the

office level. Test variables at the office level (INFLUENCE and INFLUENCE2) were often less

significant than test variables at the firm level (TINFLUENCE and TINFLUENCE2).

This paper contributes to the steam of literature surrounding auditor independence and auditor

fees by providing evidence that single office non-Big 5 audited firms are conservative in total fee

dependence based on abnormal accruals at the office and firm level, that multi-office non-Big 5

audited firms appear neutral in bias, and by providing evidence suggesting the ratio of non-audit fees

to total fees may not influence abnormal accruals. These findings also suggest that non-Big 5 auditors

are less independent than Big 5 auditors, in this respect, since they seem less conservative and their

clients have higher abnormal accruals.

Page 22: How to Write Results

20

Table 1 - Analysis of Audit fees by Big 5 vs. Second Tier Non-Big 5 firms

Revenues reported in $ millions Source: Annual Survey of National Accounting Firms - 2001

No. of Partners

No. of Offices

No. of SEC Audit Clients

FY 2000 Net Revenue

Revenue Per Office

SEC Clients Per Office

Big 5 firms PricewaterhouseCoopers 2,794 156 2,975 $ 8,299 $ 53 19 Deloitte and Touche 2,155 103 2,763 5,838 57 27 KPMG 1,928 140 1,802 4,724 34 13 Ernst and Young 1,946 82 2,922 4,271 52 36 Arthur Anderson 1,313 80 2,407 3,600 45 30

sub-total 10,136 561 12,869 26,732 48 * 23 Second Tier firms Grant Thornton 256 47 380 416 9 8 BDO Seidman 288 37 321 412 11 9 McGladrey and Pullen 311 100 107 127 1 1

sub-total 855 184 808 955 5 * 4 Total 10,991 745 13,677 $ 27,687 $ 37 * 18 Percentages Big 5 firms 92.2% 75.3% 94.1% 96.6% Second tier firms 7.8% 24.7% 5.9% 3.4%

* Average per office

Page 23: How to Write Results

21

Table 2 – Calculation of Sample Size Non-Big 5 firms with audit fee data 1,178 Less Emerson records without CUSIPS (99) Less financial institution companies (SIC 6000 to 6999) (433) Less records not located in either Compustat or Compact Disclosure (7) Records available for calculation of discretionary accrual model (1) by industry

639

Less missing values required for calculating discretionary accruals (total accruals, lagged assets, revenues, and property plant and equipment)

(155)

Less 5th and 95th percentile of total accruals1 (48) Less records with < 4 companies per industry (2 digit SIC)2 (41) Less missing values required for hypotheses testing (OCF, Sales, Debt, and PBank)

(51)

Final sample size 344 1. Sample excludes the 5th and 95th percentile of Total Accruals (Net Income before Extraordinary

Items less Cash Flow from Operations) to minimize extreme outliers. 2. Records with less than 4 companies per industry are excluded for calculating industry cross

sectional discretionary accruals.

Page 24: How to Write Results

22

Table 3 – 2000 Descriptive Statistics for 344 firms Used for the Client Size and Non-Audit Service Fee Tests (U.S. Companies on Compustat and Compact Disclosure with Non-Big 5 Auditors) Variable Mean Std. Dev. Median Lower

Quartile Upper

QuartileDACC

0.143 0.164 0.090 0.035 0.196

TACC

0.174 0.210 0.098 0.044 0.218

INFLUENCE

0.416 0.379 0.259 0.093 0.843

TINFLUENCE

0.237 0.363 0.035 0.035 0.296

FEERATIO

0.270 0.205 0.259 0.103 0.395

INFLUENCE*FEERATIO

0.117 0.156 0.043 0.009 0.180

TINFLUENCE*FEERATIO

0.052 0.106 0.004 0.000 0.042

OCF

-0.122 0.472 0.007 -0.172 0.091

SALES

9.658 2.087 9.691 8.632 10.977

DEBT

0.330 0.612 0.204 0.039 0.409

PBANK

-1.416 15.927 1.477 -0.611 2.910

Variable definitions: DACC = absolute value of discretionary accruals (the residual term of the cross sectional industry cross-secitonal Jones model referred to in equation 1), scaled for lagged assets. TACC = absolute value of total accruals (net income before extraordinary items less cash flow from operations), scaled for lagged assets. INFLUENCE = total fees of client/total fees of all public clients of the office issuing the audit report. TINFLUENCE = total fees of client/total fees of all public clients of the firm issuing the audit report. FEERATIO = non-audit fees/total fees. INFLUENCE*FEERATIO = the product of INFLUENCE and FEERATIO. TINFLUENCE*FEERATIO = the product of TINFLUENCE and FEERATIO. OCF = operating cash flows, scaled by lagged assets. SALES = log of client sales ($000). DEBT = ratio of total debt to total assets. PBANK = probability of bankruptcy measured by the Altman Z-score (0.717 * net working capital/assets + 0.847 * retained earnings/assets + 3.107 * earnings before interest and taxes/assets + 0.42 * book value of equity/liabilities + 0.998 * sales/assets).

Page 25: How to Write Results

23

Table 4 –2000 OLS Regressions of Discretionary Accruals (DACC) at Audit Office Level

Variable Absolute Value of DACC

n= 344 R2 = 0.150; p <0.0001

Negative DACC

n = 172 R2 = 0.189; p <0.0001

Positive DACC

n = 172 R2 = 0.082; p = 0.0099

Estimate p-value* Estimate

p-value* Estimate

p-value*

Intercept

0.235 <0.0001 -0.221 0.0112 0.211 <0.0001

INFLUENCE+

-0.037 0.0755 0.057 0.1697 0.002 0.9133

FEERATIO+

0.058 0.1807 -0.155 0.0375 -0.004 0.9291

INFLUENCE* FEERATIO+

0.091 0.4024 -0.172 0.4210 0.044 0.6779

OCF

-0.033 0.3872 0.082 0.1127 -0.005 0.8722

SALES

-0.010 0.0392 0.006 0.4710 -0.011 0.0057

DEBT

-0.019 0.3304 0.018 0.3368 -0.004 0.9260

PBANK

-0.003 0.0514 0.003 0.0586 -0.001 0.8561

Variable definitions: INFLUENCE = total fees of client/total fees of all public clients of the office issuing the audit report. FEERATIO = non-audit fees/total fees. INFLUENCE*FEERATIO = the product of INFLUENCE and FEERATIO. OCF = operating cash flows, scaled by lagged assets. SALES = log of client sales ($000). DEBT = ratio of total debt to total assets. PBANK = probability of bankruptcy measured by the Altman Z-score (0.717 * net working capital/assets + 0.847 * retained earnings/assets + 3.107 * earnings before interest and taxes/assets + 0.42 * book value of equity/liabilities + 0.998 * sales/assets). *Indicates regression results are not homoskedastic and white’s standard errors are used for calculating p-values. +Indicates variables are centered (observed value less sample arithmetic mean) to reduce multicollinearity.

Page 26: How to Write Results

24

Table 5 –2000 OLS Regressions of Total Accruals (TACC) at Audit Office Level Variable Absolute Value of TACC

n = 344

R2 = 0.272; p <0.0001

Negative TACC

n = 248 R2 = 0.275; p <0.0001

Positive TACC

n = 96 R2 = 0.120 p = 0.1169

Estimate

p-value* Estimate

p-value* Estimate

p-value

Intercept

0.323 <0.0001 -0.376 <0.0001 0.106 0.0179

INFLUENCE+

-0.075 0.0026 0.069 0.0506 -0.027 0.1113

FEERATIO+

0.024 0.6386 -0.071 0.2690 -0.027 0.4855

INFLUENCE*FEERATIO+

0.157 0.1931 -0.199 0.2406 0.017 0.8536

OCF

-0.048 0.2448 0.044 0.3206 -0.078 0.0236

SALES

-0.017 0.0017 0.020 0.0025 -0.002 0.6100

DEBT

0.006 0.7791 -0.008 0.6953 -0.031 0.1733

PBANK

-0.005 <0.0001 0.004 <0.0001 0.001 0.4810

Variable definitions: INFLUENCE = total fees of client/total fees of all public clients of the office issuing the audit report. FEERATIO = non-audit fees/total fees. INFLUENCE*FEERATIO = the product of INFLUENCE and FEERATIO. OCF = operating cash flows, scaled by lagged assets. SALES = log of client sales ($000). DEBT = ratio of total debt to total assets. PBANK = probability of bankruptcy measured by the Altman Z-score (0.717 * net working capital/assets + 0.847 * retained earnings/assets + 3.107 * earnings before interest and taxes/assets + 0.42 * book value of equity/liabilities + 0.998 * sales/assets). *Indicates regression results are not homoskedastic and white’s standard errors are used for calculating p-values. +Indicates variables are centered (observed value less sample arithmetic mean) to reduce multicollinearity.

Page 27: How to Write Results

25

Table 6 –2000 OLS Regressions of Discretionary Accruals (DACC) at Audit Firm Level

Variable Absolute Value of DACC

n = 344

R2 = 0.159; p <0.0001

Negative DACC

n = 172 R2 = 0.211; p <0.0001

Positive DACC

n = 172 R2 = 0.070; p <0.0001

Estimate

p-value* Estimate

p-value* Estimate

p-value*

Intercept

0.251 <0.0001 -0.243 0.0058 0.214 <0.0001

TINFLUENCE+

-0.066 0.0002 0.128 <0.0001 -0.005 0.7643

FEERATIO+

0.038 0.3637 -0.108 0.1378 0.001 0.9790

TINFLUENCE*FEERATIO+

-0.069 0.4833 0.048 0.7865 -0.105 0.3722

OCF

-0.031 0.4129 0.077 0.1292 -0.006 0.8391

SALES

-0.011 0.0188 0.008 0.3189 -0.011 0.0051

DEBT

-0.021 0.2959 0.021 0.2527 -0.005 0.8955

PBANK

-0.003 0.0526 0.003 0.0455 0.000 0.8650

Variable definitions: TINFLUENCE = total fees of client/total fees of all public clients of the firm issuing the audit report. FEERATIO = non-audit fees/total fees. TINFLUENCE*FEERATIO = the product of TINFLUENCE and FEERATIO. OCF = operating cash flows, scaled by lagged assets. SALES = log of client sales ($000). DEBT = ratio of total debt to total assets. PBANK = probability of bankruptcy measured by the Altman Z-score (0.717 * net working capital/assets + 0.847 * retained earnings/assets + 3.107 * earnings before interest and taxes/assets + 0.42 * book value of equity/liabilities + 0.998 * sales/assets). *Indicates regression results are not homoskedastic and white’s standard errors are used for calculating p-values. +Indicates variables are centered (observed value less sample arithmetic mean) to reduce multicollinearity.

Page 28: How to Write Results

26

Table 7 –2000 OLS Regressions of Total Accruals (TACC) at Audit Firm Level

Variable Absolute Value of TACC

n = 344

R2 = 0.273; p <0.0001

Negative TACC

n = 248 R2 = 0.281; p <0.0001

Positive TACC

n = 96 R2 = 0.106; p = 0.1785

Estimate

p-value Estimate

p-value* Estimate

p-value

Intercept

0.346 <0.0001 -0.402 <0.0001 0.109 0.0168

TINFLUENCE+

-0.088 0.0023 0.104 0.0006 -0.020 0.2689

FEERATIO+

-0.003 0.9472 -0.035 0.5878 -0.033 0.3716

TINFLUENCE*FEERATIO+

0.015 0.9194 -0.007 0.9623 -0.010 0.9169

OCF

-0.046 0.0595 0.040 0.3632 -0.082 0.0100

SALES

-0.019 0.0008 0.022 0.0009 -0.003 0.5534

DEBT

0.004 0.7953 -0.006 0.7784 -0.033 0.1601

PBANK

-0.005 <0.0001 0.004 <0.0001 0.001 0.4907

Variable definitions: TINFLUENCE = total fees of client/total fees of all public clients of the firm issuing the audit report. FEERATIO = non-audit fees/total fees. TINFLUENCE*FEERATIO = the product of TINFLUENCE and FEERATIO. OCF = operating cash flows, scaled by lagged assets. SALES = log of client sales ($000). DEBT = ratio of total debt to total assets. PBANK = probability of bankruptcy measured by the Altman Z-score (0.717 * net working capital/assets + 0.847 * retained earnings/assets + 3.107 * earnings before interest and taxes/assets + 0.42 * book value of equity/liabilities + 0.998 * sales/assets). *Indicates regression results are not homoskedastic and white’s standard errors are used for calculating p-values. +Indicates variables are centered (observed value less sample arithmetic mean) to reduce multicollinearity.

Page 29: How to Write Results

27

Table 8 – Tests of Differences in the Variance of Signed Accruals for Low Influence versus High Influence Clients of the Non-Big 5 Auditors Sample Size Mean Std. Dev. p-value

Office Level – Discretionary Accruals Low influence clients

172 -0.0720 0.2246

High influence clients

172 -0.0100 0.1987

0.1101

Office Level – Total Accruals Low influence clients

172 -0.1710 0.2490

High influence clients

172 -0.0900 0.2234

0.1562

Audit Firm Level – Discretionary Accruals Low influence clients

172 -0.0660 0.2270

High influence clients

172 -0.0160 0.1977

0.0724+

Audit Firm Level – Total Accruals Low influence clients

172 -0.1450 0.2471

High influence clients

172 -0.1160 0.2318

0.4037

Low influence clients are defined as those in which the variable INFLUENCE (office level) or TINFLUENCE (firm level) is below the sample median. High influence clients are those in which the variable INFLUENCE is above the sample median value. The sample is the same use for the accruals tests in Tables 4 to 7. +Indicates variance in the low influence client segment is statistically different from the variance in the high influence client segment based on the F-test.

Page 30: How to Write Results

28

Table 9 – Paired Tests of Mean Absolute Valued Accruals between Non-Big 5 Audited Firms and Big-5 Audited Firms Sample (n = 344) Non-Big 5 Mean

Big 5 Mean Mean Difference

Standard Error

T- value P - value

Absolute Valued Discretionary Accruals 0.1428 0.1081 0.0347 0.0100 3.46 0.0006

Absolute Valued Total Accruals 0.1743 0.1317 0.0426 0.0126 3.39 0.0008

This test compares a sample of non-Big 5 audited clients matched with a Big 5 audited clients by 2-digit SIC code and comparable sales revenue. We find significant differences for both mean absolute valued discretionary accruals and mean absolute total accruals in a two-tailed test.

Page 31: How to Write Results

29

Table 10 - Descriptive Statistics of Office and Firm Level Client Count and Top 10 Non-Big 5 firms by office/firm levels Descriptive Statistics of Number of Clients at Office and Firm Level; n = 344 clients Office Level – clients per

office n = 189 offices

Firm Level- clients per firm

n = 107 firms Mean 1.83 3.22Median 1 1Standard Deviation 1.71 10.11Upper Quartile 2 2Lower Quartile 1 1Maximum 14 81Minimum 1 1

Top 10 Non-Big 5 firms by number of clients from sample of 344 clients Auditor Clients Total audit fees Total fees Grant Thornton 81 $10,699,772 $24,248,453BDO Seidman 67 13,671,229 22,885,770McGladrey and Pullen 19 2,617,548 5,790,117Moss Adams 9 563,731 970,625Richard A. Eisner 7 560,905 915,805Hein and Assoc 7 446,388 866,274Goldstein Golub Kessler 6 674,119 923,475Pannell Kerr Forster 5 1,198,986 1,649,829Tanner and Company 5 239,000 247,000Moore Stephens 4 316,200 363,300Total – top 10 firms 210 $30,987,878 $58,860,348 Total Sample 344 $41,082,996 $73,062,829% Top 10 firms to total sample 61.0% 75.4% 80.6%

Page 32: How to Write Results

30

Table 11 – Descriptive Statistics of Big 5 Clients to non-Big 5 Clients and Sample Big 5 Clients

n = 4,261 Non-Big 5 Clients

n =1007 Sample of Non-Big 5

Clients n = 344 DACC – mean*

0.083 0.201 0.143

TACC – mean*

0.108 0.234 0.174

INFLUENCE – mean+

0.076 0.424 0.416

TINFLUENCE – mean+

0.0003 0.274 0.237

FEERATIO – mean+

0.488 0.270 0.271

Sales Revenue ($M) – mean*

1,890.237 80.934 134.102

Net Income ($M) – mean*

80.351 -0.566 2.629

Net Income > 0 – percent*

59.916 40.616 50.000

Return on Assets – mean*

-0.081 -1.599 -0.241

Return on Assets – std. dev.*

0.571 20.170 1.721

DEBT – mean*

0.555 1.553 0.330

PBANK – mean*

1.542 -12.795 -1.416

This table compares various descriptive statistics for Big 5 SEC audit to non-Big 5 SEC audit clients to the final sample (n=344). *Big 5and non-Big 5 clients are from Compustat 2000 (except for those variables marked with a +) and exclude financial services (SIC code 6000-6999), outliers (5th and 95th percentile) and missing value required for calculating discretionary accruals. +Big 5 results for INFLUENCE are based on Reynolds and Francis (2000), TINFLUENCE is based on Public Accounting Report (2001), and FEERATIO is based on Emerson Research data, and non-Big 5 results are based on non-Big 5 firms with audit fee data (n=1,178).See Table 2 details the calculation of the final sample (n=344). Variable definitions: DACC = absolute value of discretionary accruals (the residual term of the cross sectional

industry cross-sectional Jones model referred to in equation 1), scaled for lagged assets. TACC = absolute value of total accruals (net income before extraordinary items less cash flow from operations), scaled for lagged assets. INFLUENCE = total fees of client/total fees of all public clients of the office issuing the audit report. TINFLUENCE = total fees of client/total fees of all public clients of the firm issuing the audit report. FEERATIO = non-audit fees/total fees. Sales = client sales. Net Income = Net Income of client. Net Income > 0 = percent of clients with positive net income. Return on Assets = Net Income/Assets. DEBT = ratio of total debt to total assets. PBANK = probability of bankruptcy measured by the Altman Z-score (0.717 * net working capital/assets + 0.847 * retained earnings/assets + 3.107 * earnings before interest and taxes/assets + 0.42 * book value of equity/liabilities + 0.998 * sales/assets).

Page 33: How to Write Results

31

References Altman, E. 1983. Corporate Financial Distress: A Complete Guide to Predicting, Avoiding, and Dealing with Bankruptcy. New York: Wiley. pp. 121. Ashbaugh, H., LaFond, R., and Maydew, B. W. 2002. Do Non-Audit Services Compromise Auditor Independence. Working paper. Beck, P., Frecka, T., and Solomon, I. 1988. A model of the market for MAS and audit services: knowledge spillovers and auditor-auditee bonding. Journal of Accounting Literature, 7: 50-64. Becker, C. L., DeFond, M. L., Jiambalvo, J., and Subramanyam, K. R. 1998. The Effect of Audit Quality on Earnings Management. Contemporary Accounting Research, 15(1): 1-24. Craswell, A. T., Francis, J. R., and Taylor, S. L. 1995. Auditor brand name reputations and industry specializations. Journal of Accounting and Economics, 20: 297-322. DeAngelo, L. E. 1981. Auditor Size and Audit Quality. Journal of Accounting and Economics, 3: 183-199. Dechow, P. M., Sloan, R. G., and Sweeney, A. P. 1995. Detecting earnings management. The Accounting Review, 70(2): 193-225. DeFond, M. L. and Jiambalvo, J. 1994. Debt covenant violation and manipulation of accruals. Journal of Accounting and Economics, 17: 145-176. DeFond, M. L. and Park, C. W. 1997. Smoothing income in anticipation of future earnings. Journal of Accounting and Economics, 23: 115-139. DeFond, M. L. and Subramanyam, K. R. 1998. Auditor changes and discretionary accruals. Journal of Accounting and Economics, 25: 35-67. Firms, P. A. R. A. S. o. N. A. 2001. PwC Takes Over As Biggest Firm, But Here Comes D&T, Public Accounting Report - The Independent Newsletter of the Profession since 1978, Vol. Special Supplement, Annual Survey of National Accounting Firms - 2001: S1-S2. Francis, J. R. and Simon, D. T. 1987. A Test of Audit Pricing in the Small-Client Segment of the U.S. Audit Market. The Accounting Review, 62(1): 145-157. Francis, J. R. and Wilson, E. R. 1988. Auditor Changes: A Joint Test of Theories Relating to Agency Costs and Auditor Differentiation. The Accounting Review, 63(4): 663-682. Francis, J. R., Maydew, E. L., and Sparks, H. C. 1999. The Role of Big 6 Auditors in the Credible Reporting of Accruals. Auditing: A Journal of Practice and Theory, 18(2): 17-34.

Page 34: How to Write Results

32

Francis, J. R. and Ke, B. 2001a. Do Non-Audit Services Compromise Auditor Independence? Working paper. Francis, J. R. and Reynolds, K. J. 2001b. Do Large Accounting Firms Screen Out Risky Audit Clients? Working paper. Frankel, R. M., Johnson, M. F., and Nelson, K. K. 2002. The Relation Between Auditor's Fees for Non-Audit Services and Earnings Quality. Working paper. Hayashi, F. (2000). Econometrics. Princeton University Press. Princeton. pg. 118. Healy, P. M. 1985. The effect of bonus schemes on accounting decisions. Journal of Accounting and Economics, 7: 85-107. Hogan, C. E. and Jeter, D. C. 1999. Industry Specialization by Auditors. Auditing: A Journal of Practice and Theory, 18(1): 1-17. Jones, J. J. 1991. Earnings management during import relief investigations. Journal of Accounting Research, 29(2): 193-228. Neter, J., Kutner, M.H., Nachtsheim, and C.J., Wasserman, W. 1996. Applied Linear Regression Models. 3Rd Ed. Irwin. Chicago. pp 313. Palmrose, Z.-V. 1988. An Analysis of Auditor Litigation and Audit Service Quality. The Accounting Review, 63(1): 55-73. Reynolds, K. J. and Francis, J. R. 2000. Does size matter? The influence of large clients on office-level auditor reporting decisions. Journal of Accounting and Economics, 30: 375-400. Schwartz, K. B. and Soo, B. S. 1996. Evidence of Regulatory Noncompliance with SEC Disclosure Rules on Auditor Changes. The Accounting Review, 71(4). SEC, U. S.; Final Rule: Revision of the Commission's Auditor Independence Requirements; http://www.sec.gov/rules/final/33-7919.htm; March 25, 2002, 2002. Wallman, S. 1996. The future of accounting , part III: reliability and auditor independence. Accounting Horizons, 10: 76-97. White Paper by the “Big 6” Accounting Firms, 1992, The Liability Crisis in the United States. Journal of Accountancy. November 1992: (19-23).