Electronic copy available at: />
1

AUDITOR INDEPENDENCE : EVIDENCE ON THE JOINT EFFECTS OF
AUDITOR TENURE AND NON-AUDIT FEES Ferdinand A Gul
The Hong Kong Polytechnic University
School of Accounting & Finance
Hung Hom
Hong Kong
Tel: 852 2766 7771
Fax: 852 2365 9303
Email: &

Bikki Jaggi*
Rutgers University
School of Business
Levin Building
Piscataway-NJ 08850
Tel : 732-445-3539
Fax: 732-445-3201
Email:


Abstract

This study examines whether the impact of non-audit fees on auditor independence is
contingent on auditor tenure. The results, based on a sample of 4,720 U.S. firms for the years
2000 and 2001, show that there is a positive association between non-audit fees and positive
discretionary current accruals, a proxy for auditor independence, for firms with short auditor
tenure of not more than three years. These findings suggest that non-audit fees may impair
auditor independence when auditor tenure is short and not when auditor tenure is long.
Furthermore, exploratory analyses show that the positive association between non-audit fees
and earnings management for firms with short auditor tenure is significant for small clients
but not for large clients. Taken together, these results suggest that the association between
non-audit fees and auditor independence is contingent upon auditor tenure, and that high non-
audit fees have a negative impact on auditor independence when audit tenure is short and
client firm size is small.

Keywords: Auditor independence, non-audit fees, discretionary accruals, auditor tenure.

Data availability: Data are available from sources identified in the paper.

magnitudes of these fees. The SOX and the SEC’s regulations are apparently based on the
assumption that certain types of non-audit services provided by audit firms are likely to
impair auditor independence, thereby reducing audit quality.
Several studies have examined whether there is an association between non-audit fees
paid to audit firms and auditor independence.
1
Of particular interest to us in this paper are the
studies that examine the impact of non-audit fees on discretionary accruals, which are used as
a proxy for earnings management (e.g., Francis and Ke 2002; Frankel et al. 2002; DeFond et
al. 2002; Ashbaugh et al. 2003; Chung and Kallapur 2003; Larcker and Richardson 2004).
Findings of these studies are mixed. Frankel et al. (2002) (hereafter FJN) document that non-
audit fees are positively associated with the magnitude of discretionary accruals, suggesting
that high non-audit fees are likely to impair auditor independence. However, Ashbaugh et al.
(2003) (hereafter ALM) use a more refined measure of discretionary accruals and do not find

4
a statistically significant association between non-audit fees and positive discretionary
accruals. They conclude that there is no systematic evidence to support FJN’s findings that
higher non-audit fees impair auditor independence. Chung and Kallapur (2003) use a
different non-audit fees metric and also find no support for FJN’s findings. Thus, despite
concerns by regulators and the financial press, there is no clear evidence that higher non-audit
fees will negatively affect auditor independence.
Another stream of research in the auditing literature focuses on the association
between auditor independence and auditor tenure (e.g., Dopuch et al. 2001; Geiger and
Raghunandan 2002; Johnson et al. 2002; Myers et al. 2003). The importance of auditor
tenure is also reflected in Section 203 of the SOX, which requires a five-year rotation cycle
for the external lead or reviewing audit partner. A recent report of the Commission on Public
Trust and Private Enterprise appointed by the Conference Board also supports the idea of
auditor rotation (Conference Board 2005). However, empirical evidence to date shows that
longer auditor tenure rather than shorter auditor tenure is associated with higher quality

positive REDCA, i.e., income-increasing discretionary current accruals, is generally of major
concern to investors because managers are more likely to adjust earnings upwards to meet
market expectations or convert losses into small positive earnings (e.g. Burgstahler and
Dichev 1997). The use of negative REDCA (downward adjustment of reported earnings) is,
however, considered to be conservative accounting, and investors are not likely to be too
concerned about its use (see also Butler et al. 2004). Thus, we focus on positive REDCA, but
also conduct tests on the total sample as well as on the sub-sample of firms with negative
discretionary accruals.
Following ALM, we use the log of audit fees, the log of non-audit fees, the log of
total fees and the ratio of non-audit fees to audit fees as the metrics. Consistent with the

6
literature (e.g., Johnson et al. 2002), we use the cut-off point of three years for splitting the
sample into short and long auditor tenure. In the sensitivity tests, we also use different cut-off
points for splitting the sample into long and short auditor tenure.
Similar to ALM’s findings, the results using the total sample show that there is no
significant association between non-audit fees and REDCA. When we split the sample into
positive and negative discretionary accruals and include auditor tenure in the analyses, the
association between non-audit fees and earnings management becomes clearer. For example,
when we include a dummy variable for auditor tenure (1 for short tenure < =3years, 0
otherwise) and interact this variable with non-audit fees in the positive discretionary accruals
sample, we obtain a significant positive coefficient for the interaction term. In other words,
firms with high non-audit fees and short tenure are associated with higher positive
discretionary accruals. Similar results are obtained when separate tests are conducted for
firms with long and short auditor tenure. The results on the firms with negative discretionary
accruals, however, show no significant association. These findings thus indicate that there is
an upward adjustment of reported earnings when non-audit fees are high and auditor tenure is
short. On the other hand, the results for firms with long auditor tenure show that there is no
significant association between positive discretionary accruals and non-audit fees. These
findings suggest that auditor independence may be compromised when non-audit fees are

The remainder of the paper is organized as follows. In part two, we develop a
hypothesis for the study, and part three discusses the research methodology. The results are
discussed in part four, and conclusions are presented in part five. 8
II. HYPOTHESIS
Auditor Tenure, Non-Audit Fees and Auditor Independence

Auditor independence has been the subject of considerable debate for some time, and
it has been examined from the perspectives of both non-audit fees and auditor tenure. The
existing studies on the association between non-audit fees and auditor independence have
evaluated whether higher non-audit fees create an economic bond between the auditor and the
client, which enables the client firm to exert influence over the auditors’ decisions and make
auditors less independent as non-audit fees increase.
3
However, as pointed out earlier, the
findings of empirical studies using discretionary accruals as a proxy for auditor independence
provide conflicting evidence.
The impact of auditor tenure on auditor independence has also attracted attention from
regulatory agencies, congressional bodies, the Conference Board, the popular press as well as
academics (e.g., U.S. Senate 1977; AICPA 1978; Wall Street Journal 1991; SEC 1994;
Conference Board 2005). A recent report of the Commission appointed by the Conference
Board states that “when there is confluence of circumstances that could put in question the
audit firm’s independence, the Commission believes that audit committees should carefully
consider rotation of audit firm (p. 39)”. In addition, the report illustrates the existence of
certain circumstances that merit consideration of rotation, and recommends serious
consideration of auditor rotation when the audit firm “has been employed for a substantial
period of time, e.g. over ten years”. Auditor rotation is supported on the ground that longer
auditor tenure leads to economic bonding between the auditor and client firms, which may

management, we conjecture that there will be a negative association between non-audit fees
and auditor independence for firms with short auditor tenure. The arguments below are
presented in support of our conjecture.
Prior research suggests that auditor incentives are traded off between the auditor’s
desire to protect firm reputation (brand name) and to maintain profits (quasi rents) from the

10
auditor-client relationship (Citron and Taffler 1992; Johnson et al. 2002). The desire to
maintain profits or quasi rents from an existing client relationship in the early years of the
engagement could cause the auditor to be more accommodating to the client. Geiger and
Raghunandan (2002) argue that auditors are willing to be more accommodating to the client
in the early years of an audit, because they are influenced by their desire to limit their losses
on the current engagement as a result of the competitive practice of low-balling and to be able
to ensure continuity of the engagement. Thus, it can be argued that a client would purchase
non-audit services in the early years of an engagement and persuade the auditor not to
conduct in-depth analyses which may uncover earnings management. It is also possible that
the auditor may face special audit problems earlier in the tenure because of unfamiliarity with
the client’s business, and thus may have difficulty in detecting earnings management. Based
on these arguments, we conjecture that large non-audit fees combined with short auditor
tenure are likely to be associated with higher earnings management. On the other hand, an
auditor with a relatively longer tenure and an established relationship with the client over a
period of time is likely to place more emphasis on reputation protection than profit, and is
less likely to be affected by high non-audit fees. As a result of the long tenure, the auditor
will have a better understanding of the client’s accounting system and be in a better position
to detect earnings management. Consequently, we argue that because of a combination of
reputation protection and better understanding of the client’s accounting system, high non-
audit fees are unlikely to be associated with earnings management when auditor tenure is
long. We develop the following hypothesis to test the impact of auditor tenure on the
association between auditor independence and non-audit fees:
H

(Insert Table 2 here)

The mean (median) of audit fees is $0.59 ($0.22) million, whereas the mean (median) of non-
audit fees is $1.34 ($0.23) million respectively. These means (and medians) are generally

12
similar to those reported by ALM. There are no significant differences in the fee metrics for
the years 2000 and 2001.
Additionally, we examine the fee metrics for firms audited by Big 5 and non-Big 5
firms, where Big 5 audit firms refer to Arthur Andersen, Deloitte & Touche, Ernst and Young,
KPMG, and PricewaterhouseCoopers.
7
The results (untabulated) show that the mean
(median) of audit fees for clients of Big 5 is $0.62 ($0.24) million, and the mean (median) of
non-audit fees is $1.43 ($0.25) million respectively. As for clients of non-Big 5, the mean
(median) of audit fees is $0.15 ($0.1) million, and the mean (median) of non-audit fees is
$0.14 ($0.05) million, respectively. The means and medians suggest that audit as well as
non-audit fees are significantly higher for Big 5 clients.
As pointed out earlier, we use the log of audit fees, the log of non-audit fees, the log
of total fees and the ratio of non-audit fees to total fees (FEERATIO) as fee metrics. The use
of these metrics allows us to compare the results of our study with those obtained from the
ALM study.
Validity of Audit Fee Metrics
We test the validity of audit fee data by evaluating the association between important
variables identified in the audit fee literature and audit fees (e.g., ALM). The model, based
on ALM, includes the variables of a client firm’s auditor choice, audit complexity, audit risk,
and demand for consulting services, as well as the variable of auditor tenure. The audit fee
model is used for testing the validity of audit fee data:
LEVMVBVFINANCINGMERGEACQLNMVEBIGFEE
654321

i
ii
MMIESINDUSTRYDU
LOGTENSPECIALROANEGATIVEINVRECROA
(1)
where,
FEE = FEERATIO, AUDIT, NON-AUDIT, or TOTAL fees. With the
exception of FEERATIO, we take the natural log of each fee
variable to normalize the distributions of these variables, allowing

13
the cross-sectional aggregation of observations
BIG5 = 1 if the firm is audited by Big 5, and 0 otherwise
LNMVE = the natural log of the firm’s market value of equity
MERGEACQ
= 1 if the firm is engaged in a merger or acquisition during the study
period, and 0 otherwise
FINANCING = 1 if MERGEACQ is not equal to 1 and any of the following
conditions apply: long-term debt increased by 20 percent or more,
number of shares outstanding increased by 10 percent or more after
controlling for stock splits, and 0 otherwise
MVBV = the firm’s market-to-book ratio defined as its market value of equity
divided by book values
LEV = A firm’s total assets less its book value divided by its total assets
ROA =
the firm’s return-on-asset ratio calculated as net income before
extraordinary items divided by beginning of the year total assets
INVREC = the sum of the firm’s receivables and inventory divided by its total
assets
NEGATIVE_ROA = 1 if the firm’s ROA is negative, and 0 otherwise


Long versus Short Auditor Tenure
Geiger and Raghunandan (2002) in their analyses show that auditors are likely to be
more accommodating in the initial three years of auditor tenure. This effect is mitigated after
six or more years. They thus conclude that the threat to auditor independence is the greatest
during the first few years of the auditor/client relationship. Consistent with their analyses, we
set the cut-off point at three years for differentiating between short and long auditor tenure.
We, however, also conduct analyses on different cut-off points.
Association between Discretionary Accruals and Non-audit Fees

We evaluate the association between discretionary accruals and audit fee metrics
using an OLS regression model, similar to that used by ALM, except for the variables
LOGTEN (tenure) and YEARDUM (year dummy).

LOGREDCA = α + β
1
FEE+ β
2
BIG5+ β
3
PRECURRACCL+ β
4
LNMVE

εββ
ββββ
β
β
β
β

YEARDUM = 1 if the fiscal year is 2001, 0 if fiscal year is 2000 Other variables have been defined earlier. We use the variable YEARDUM to control for the
effect of time period. Consistent with ALM and Warfield et al. (1995), the dependent
variable REDCA is transformed by taking the natural logarithm of absolute values to correct
for possible violations of normality.
We use the natural log of tenure in our analysis to overcome the problem that
observations are not normally distributed. We conduct tests on the total sample and also
separately on the sub-samples of positive and negative discretionary accruals. The impact of
audit tenure is evaluated by including an interaction term for audit fees and auditor tenure in
the analyses, and also by conducting separate tests on firms with short and long auditor tenure.

IV. RESULTS AND DISCUSSION
Calculation and Descriptive Statistics of Auditor Tenure
Auditor tenure is calculated on the basis of information about the year when the audit
firm first started audit work with the client firm, as provided in the Compustat database. As
the Compustat database used in this study dates back to 1984, there may be a measurement
error for auditor tenure for some firms. This measurement error is, however, unlikely to
seriously affect our results because tenure of nine years or more is considered as long tenure.
Furthermore, our sub-samples of short and long tenure are based on the three year cut-off
point.

16
Descriptive statistics on auditor tenure are provided in Table 2. The results show that
mean (median) auditor tenure for the total sample is 8.6 (7) years. The mean and median of
the sub-samples of short and long auditor tenure are also provided in the table.
Descriptive Statistics on Variables
We calculate REDCA based on all firms included in Compustat. Descriptive statistics
for different variables used in the analyses are also provided in Table 2 for the total sample as


The results reported in Panel A of Table 4 show that the coefficient for LNAF is
negative, but insignificant, and that the coefficient for LOGTEN is significantly negative for
all models. The results on the tenure variable are consistent with the earlier findings (e.g.,
Myers et al. 2003), showing that longer tenure is associated with lower earnings
management.
12
With the exception of the following control variable on the LNAF regression,
the results on other control variables are consistent with ALM’s results: BIG5, MERGEACQ,
MVBV, and DLOSS. The coefficient for BIG5 is significantly negative in ALM’s results,
whereas it is insignificantly positive in our results. The variable MERGEACQ is positive in
ALM’s as well as in our results, but the coefficient is insignificant in our results and it is
significant in ALM’s results. The variables of MVBV and DLOSS are insignificantly
negative in our results, whereas they are significantly positive in ALM’s results.
In order to mitigate the econometric problems associated with the OLS regression test
on the positive discretionary accruals, whose distribution is truncated at zero, we conduct a
Tobit test on this sub-sample by setting the lower bound to zero. Results of the Tobit test
(untabulated) are similar to those of the OLS regression test, suggesting that the results
presented in the table are robust.
We focus on the impact of auditor tenure on the association between non-audit fees
and auditor independence by including an interaction term for each fee metric and auditor
tenure. Consistent with the literature (e.g., Johnson et al. 2002), we split the sample into long
and short tenure based on the three-year cut-off point and auditor tenure is coded as 1 when it

18
is three years or less. The results reported in Panel B of Table 4 show that the coefficient for
LNAF is negative and insignificant, whereas the coefficient for the interaction term between
non-audit fees and auditor tenure (LNAF*TENDUM) is positive and significant (p<0.05, 2-
tailed).
13

that there is a trend of lower discretionary accruals when non-audit fees are high and auditor
tenure is long. Overall, these results support our earlier findings that higher non-audit fees
are significantly associated with higher positive discretionary accruals when auditor tenure is
short and there is no significant association between non-audit fees and positive discretionary
accruals when auditor tenure is long.
We also conduct tests on the firms with negative discretionary accruals. The results
(untabulated) for the total sample with an interaction variable as well as on the sub-samples
of firms with short and long tenure indicate significantly negative coefficient for LAF for
long tenure only, whereas the coefficients for all other fee metrics are insignificant.
The above results support our hypothesis that higher non-audit fees are associated
with higher discretionary accruals only when the auditor tenure is short. These results thus
indicate that the association between non-audit fees and auditor independence is contingent
upon auditor tenure.
Sensitivity Test Results Based on Different Tenure Cut-off Points for Auditor Tenure
We also use the cut-off points of two, four and five years for identifying the firms
with short and long auditor tenure. The results of the tests (untabulated) on short and long
auditor tenure based on the cut-off point of two years are similar to the results for the three-
year cut- off. The coefficient of non-audit fees for the short tenure sub-sample is positive and
significant (p < 0.01, 2-tailed). These results provide additional support for our findings that
there is higher earnings management when non-audit fees are high and auditor tenure is short.
The test results on the sub-samples of short and long auditor tenure based on the cut-off

20
points of four and five years show no significant results for the non-audit fee coefficients.
These findings thus indicate that short auditor tenure of up to three years is likely to be
associated with lower auditor independence.
Results on Individual Years
We repeat all the tests on data for positive REDCA separately for the years 2000 and
2001 separately. The untabulated regression results for the years 2000 and 2001 indicate that
they are similar to the results for the pooled sample for two years; the significance levels of

understand the weaknesses in the client’s accounting system. Further, the client firm may
purchase more non-audit services to improve their accounting system, which could mean
higher non-audit fees for the auditor. The weaker accounting system may thus result in
higher discretionary accruals and additional non-audit services are likely to result in higher
non-audit fees. This means that the positive association between discretionary accruals and
higher non-audit fees for small clients may not necessarily reflect the lack of auditor
independence.
We evaluate whether there is a positive association between discretionary accruals
and non-audit fees for small client firms when auditor tenure is short. In order to explore the
role of client size in the relationship between non-audit fees and positive discretionary
accruals, we conduct tests on the sub-samples of small and large clients firms separately. We
use the median asset size to split the sample into subgroups of large and small clients. The
results on the sub-samples of small and large clients with short auditor tenure are provided in
Table 6.
16

(Insert Table 6 here) 22
The results on the sub-sample of large client firms (Panel B) show that the coefficient
of LNAF is insignificant, whereas the results on the small client firms (Panel A) show that
the coefficient of LNAF is positive and significant (p<0.05, 2-tailed). These results thus
indicate that discretionary accruals are higher for small client firms when auditor tenure is
short and non-audit fees are high.
We also conduct tests on the role client size separately for the years 2000 and 2001.
The results (untabulated) for year 2000 indicate that the coefficient of non-audit fees for the
short tenure sub-group is positive and significant (p<0.01, 2-tailed) for small firms (N=95)
but insignificant for large firms (N=96). The results for the long tenure sub-group for large
(N=380) and small (N=379) firms show that none of the non-audit fee coefficients are

Limitations of this study include issues related to the sample selection, and
measurement-related problems in estimating discretionary accruals and auditor tenure. The
results of this study are based on data for two years, and thus the findings may not be
generalizable to other years. Further, the study uses US data and these results may not be
applicable in other countries where institutional settings and regulatory requirements are
different. It is possible that high positive discretionary accruals may be correlated with the
firm’s perceived riskiness or other underlying events that may have caused auditor change
during the study. Because the Compustat data base used in this study goes back to 1984 only,
our measurement of auditor tenure is constrained by this limitation. Additionally, as pointed
out earlier, the findings may have been influenced by the start-up firms in the sample.
Therefore, the findings need to be interpreted with caution.

24
It is widely recognized that there are measurement errors in calculating discretionary
accruals. We have attempted to mitigate this problem by using REDCA, which is considered
to be a better proxy for earnings management. Additionally, we have focused on positive
discretionary accruals, which are considered to be a better proxy for earnings management
and auditor independence. It is possible that detailed analyses of negative discretionary
accruals, which reflect the accounting “big-bath”, conservative accounting and/or the creation
of “cookie jar” reserves, may provide some additional insight into the association of non-
audit fees with auditor independence. Unraveling these issues in the context of the linkage
between non-audit fees and earnings management is left for future research.
As discussed in the paper, we have used a log of fee metrics as the dependent variable
in the analyses on the assumption that log transformation of fee metrics is more suitable to
the focus of our study. The validity of our findings may, therefore, be influenced by the
appropriateness of the fee metrics used in the study.
Finally, it is possible that the results may be interpreted without any reference to the
“independence story”.
17
The positive association between non-audit fees and positive