review journal auditing

1

ACCOUNTING HORIZONSSupplement2003pp. 1�16

Does Big 6 Auditor Industry ExpertiseConstrain Earnings Management?

Gopal V. Krishnan

SYNOPSIS: Earnings management remains a popular topic of debate and discussionamong investors, regulators, analysts, and the public. One mechanism that might miti-gate earnings management is auditors� industry expertise. Using a large sample ofclients of Big 6 auditors, this research examines the association between auditor indus-try expertise, measured in terms of both auditor market share in an industry and anindustry�s share in the auditor�s portfolio of client industries, and a client�s level ofabsolute discretionary accruals, a common proxy for earnings management. Clients ofnonspecialist auditors report absolute discretionary accruals that are, on average, 1.2percent of total assets higher than the discretionary accruals reported by clients ofspecialist auditors. This finding is consistent with the notion that specialist auditorsmitigate accruals-based earnings management more than nonspecialist auditors and,therefore, influence the quality of earnings.

Keywords: industry specialization; Big 6; specialist firms; earnings management; dis-cretionary accruals; audit quality.

Data Availability: The data used in this study are publicly available from the sourcesindicated in the text.

INTRODUCTION

Earnings management is a concern for investors, regulators, analysts, and the public. In areview of the earnings management literature, Healy and Wahlen (1999) call for research onfactors that limit earnings management. This study is a response that provides empirical

evidence on one mitigating factor: auditors� industry expertise. Specifically, I examine the associa-tion between Big 6 auditor industry expertise and the level of firms� absolute discretionary accru-als�a common proxy for earnings management.

Bonner and Lewis (1990) find that, on average, more experienced auditors outperform lessexperienced auditors. Similarly, Bedard and Biggs (1991) observe that auditors with more manufac-turing experience are better able to identify errors in a manufacturing client�s data than auditors withless manufacturing experience. This is consistent with the findings of Johnson et al. (1991) thatindustry experience is associated with enhanced ability to detect fraud. Wright and Wright (1997)conclude that significant experience in the retailing industry enhances hypothesis generation inidentifying material errors.

Gopal V. Krishnan is an Associate Professor at the City University of Hong Kong.

I appreciate helpful comments from Patricia Dechow, James Largay, and two anonymous reviewers.

2 Krishnan

Accounting Horizons, Supplement 2003

Specialist auditors are likely to invest more in staff recruitment and training, information tech-nology, and state-of-the art audit technologies than nonspecialist auditors (Dopuch and Simunic1982). Solomon et al. (1999) find that specialist auditors have more accurate nonerror frequencyknowledge than nonspecialists. This finding is important because it is not unusual that client firmssuggest nonerror explanations for ratio fluctuations. Audit effectiveness thus depends on the accu-racy of auditors� nonerror frequency knowledge. All these findings support the conclusion thatauditors� industry-specific knowledge is associated with audit effectiveness.

How does an auditor�s specialized industry knowledge help in detecting earnings management?Maletta and Wright (1996) observe fundamental differences in error characteristics and methods ofdetection across industries. This suggests that auditors who have a more comprehensive understand-ing of an industry�s characteristics and trends will be more effective in auditing than auditors withoutsuch industry knowledge.

Auditors who specialize in the banking industry can assess the adequacy of loan loss provisionsbetter than nonspecialist auditors and, therefore, can improve the credibility of reported earnings.Auditors with expertise in manufacturing can evaluate whether the client�s provision for warrantyobligations is in line with industry standards better than an auditor without this expertise. Specialistauditors are also likely to develop databases detailing industry-specific best practices, industry-specific risks and errors, and unusual transactions, all of which serve to enhance overall auditeffectiveness.

Besides having the resources and the expertise to detect earnings management, specialist audi-tors who enjoy a brand-name reputation have particular incentives to deter and report questionableor aggressive accounting practices. It has been accepted that Big 6 auditors are of higher quality thannon-Big 6 auditors (DeAngelo 1981).1 Given their larger client base, Big 6 auditors have more tolose than non-Big 6 auditors in the event of a loss of reputation. Thus, Big 6 auditors should havemore incentive to protect their reputation than non-Big 6 auditors. MacDonald (1997), for example,reports that between 1994 and 1997, Big 6 auditors dropped 275 publicly traded audit clientsbecause of concerns about harm to their reputations or litigation risk. In short, Big 6 auditors who arealso specialist auditors have both the expertise to detect earnings management and the incentives toreport it. This argument is consistent with the findings of O�Keefe et al. (1994) that specialistauditors exhibit greater compliance with auditing standards (GAAS) than nonspecialist auditors.

RESEARCH QUESTION AND CONTRIBUTIONSTaken together, findings of the above studies support the notion that specialist auditors have the

resources, the industry-specific expertise, and the incentives to detect and constrain earnings man-agement, and therefore enhance the quality of earnings. This leads to the research question:

RQ: The absolute value of discretionary accruals of firms audited by specialist auditorsis lower than the absolute value of discretionary accruals of firms audited by non-specialist auditors.

Consistent with prior research, I use the absolute value of discretionary accruals to proxy foraccruals-based earnings management (Francis et al. 1999). The sample consists of 4,422 firmsaudited by Big 6 auditors from 1989 through 1998. I focus on clients of Big 6 auditors because theyaudit more than 80 percent of firms in the Compustat database. Confining the sample firms to thoseaudited by the Big 6 makes it possible to isolate differences due to industry expertise rather thandifferences in audit quality between Big 6 auditors and other auditors.

1 Consistent with the prior research, I refer to the original Big 8 auditors (Big 5 after 1998 and now Big 4) as Big 6 auditors.

Does Big 6 Auditor Industry Expertise Constrain Earnings Management? 3


Using sales as a measure of client size, I estimate auditors� industry expertise using two commonproxies: the audit fees an auditor earns in an industry relative to the total audit fees earned by allauditors serving that particular industry, and an auditor�s audit fees earned from an industry relativeto fees earned from clients across all industries served (Gramling and Stone 2001). I find that theclients of nonspecialist auditors report higher absolute discretionary accruals than the clients ofspecialist auditors. This finding is consistent with the notion that auditors� industry expertise moder-ates earnings management.

My study contributes to two aspects of the literature. The first is the literature on audit quality.My contribution is to demonstrate that audit quality varies even among Big 6 auditors. Research likeBecker et al. (1998) tends to focus on audit-quality differences between Big 6 and non-Big 6 auditorsand implicitly treats the Big 6 auditors as a homogeneous group in terms of audit quality. Myextension treats auditor industry expertise as a dimension of audit quality. Second is the literature onauditors� industry specialization. Here I contribute by providing an empirical link between auditors�industry expertise and audit quality. In their review of audit firm industry expertise literature,Gramling and Stone (2001) observe that there is limited examination of whether industry specializa-tion is associated with audit quality. They see the dearth of research as surprising, given the impor-tance that client firms and audit standards setters place on industry expertise.

My findings suggest that one reason specialty auditors charge a premium over nonspecialtyauditors is because they are able to constrain accruals-based earnings management better thannonspecialty auditors and thus add credibility to the quality of reported earnings.

SAMPLE SELECTION AND MEASURES OF AUDITORS� INDUSTRY EXPERTISEI searched the 2000 version of Compustat PC Plus to identify the sample firms and their

auditors. My sample period covers a ten-year period from 1989 through 1998. The selection criteriaare as follows. I exclude financial institutions (SICs between 6000 and 6999) because auditorinformation for these firms is unavailable on Compustat. I also eliminate firms that changed fiscalyear-ends during the period of analysis. As previously stated, I restrict my sample to firms audited bythe Big 6 auditors.

This sample selection procedure yields 24,114 firm-year observations representing 4,422 firms.Yearly firm-year observations are 1,481, 1,746, 1,806, 1,857, 2,046, 2,302, 2,624, 3,009, 3,510, and3,733 for 1989 through 1998, respectively.

Big 6 Auditor Portfolio SharesAuditor industry expertise is unobservable, so researchers must rely on proxies to estimate it.

Yardley et al. (1992) develop a measure of auditor industry expertise that estimates industry special-ization by the proportion of an auditor�s audit fees earned from one industry of all those served. Asaudit fee information has been unavailable until recently, researchers have used sales or assets or thesquare root of assets as the base to estimate the proportion of audit fees received from a particularindustry.

Similar to Kwon (1996), I estimate auditor portfolio shares as follows (hereafter, Big 6 auditorportfolio shares, or Big 6 PS):

1

1 1

(

ik

ik

J

ijkj

ik JK

ijkk j

SALESBig 6 PS

SALES

=

= =

=∑

∑∑(1)

where SALES is sales revenue, and the numerator is the sum of the sales of all Jik clients of audit firmi in industry k. The value of i ranges from 1 to 6, representing the Big 6 auditors. Two-digit SIC

4 Krishnan


codes identify industry categories. The denominator in Equation (1) is the sales of all clients of auditfirm i summed over all K industries.

An example will demonstrate the computation of Big 6 PS for Coopers & Lybrand (CL) for thetransportation equipment industry (two-digit SIC code 37). In 1989, CL had seven clients inthe transportation equipment industry. For 1989, the total sales of these seven clients were $97,094.54million (a large portion generated by a single client, Ford). During 1989, CL served 33 industries andaudited 202 clients. The total sales of these clients amounted to $395,525.09 million. Big 6 PS forCoopers & Lybrand for the transportation equipment industry for 1989 is thus:

Big 6 PSCL,Transportation equipment,1989 = $97,094.54/$395,525.09 = 0.2455.This suggests that the transportation equipment industry alone accounted for about a quarter of CL�saudit fees in 1989. As an auditor�s industry expertise may change over time, I repeat this step foreach year and then aggregate CL�s portfolio shares over the years 1989 through 1998 for eachindustry.

During 1989 through 1998, the total sales of all clients of CL in the transportation equipmentindustry amounted to $1,094,132.01 million. During the period, CL served 49 industries and audited2,647 client-years. The combined sales of clients representing those 49 industries amounted to$6,350,583.13 million.

Big 6 PS for CL using the aggregate data is computed as follows:Big 6 PSCL,Transportation equipment,1989�1998 = $1,094,132.01/$6,350,583.13 = 0.1723.

The final step involves identifying industries in which CL is considered a specialist auditor. Icode a firm�s top-three portfolio shares as the auditor�s specialty and the remaining industries asnonspecialty. For CL, industries that represent the top three portfolio shares are communications,transportation equipment, and food stores. To put it differently, these three industries are the topthree moneymakers for CL during the period 1989 through 1998. Thus, CL is defined as a specialistauditor only for these three industries over the sample period. I repeat these steps for each Big 6auditor to identify industry specializations.

Table 1 reports industry specialization and portfolio shares for each Big 6 auditor for the pooledsample (year-by-year portfolio shares are not reported). With the exception of Arthur Andersen, theBig 6 firms appear to show an expertise in the transportation equipment industry, although theportfolio shares range from 29.6 percent for Deloitte & Touche to 12.4 percent for Ernst & Young.

Big 6 Auditor Industry Market SharesI use an alternative measure of auditors� industry expertise in order to minimize measurement

error associated with estimation of auditors� industry specialization and to enhance the reliability ofthe findings. Following Gramling and Stone (2001), I calculate an auditor�s industry market share toproxy for audit fees earned by an auditor in an industry as a proportion of the total audit fees earnedby all auditors that serve that particular industry (hereafter, Big 6 auditor industry market share, orBig 6 IMS):

1

1 1

ik

ikK

J

ijkj

ik JI

ijki j

SALESBig 6 IMS

SALES

=

= =

=∑

∑∑(2)

where SALES is sales revenue, and the numerator is the sum of sales of all Jik clients of audit firm i inindustry k. The denominator is the sales of Jik clients in industry k summed over all Ik audit firms inthe sample with clients (Jik) in industry k. The denominator does not include the sales of clients ofnon-Big 6 auditors as the sample includes only clients of Big 6 auditors. To estimate industry market



TABLE 1Big 6 Auditor Portfolio Shares for 1989�1998

Portfolio SharesAuditor Industry (SIC code) (in %)Arthur Andersen Petroleum refining (29) 12.85

Chemicals and pharmaceuticals (28) 8.05Air transportation (45) 6.88

Coopers & Lybrand Communications (48) 17.34Transportation equipment (37) 17.23Food stores (54) 8.50

Ernst & Young Petroleum refining (29) 16.44General merchandise stores (53) 15.55Transportation equipment (37) 12.40

Deloitte & Touche Transportation equipment (37) 29.56Durable goods�wholesale (50) 23.89Chemicals and pharmaceuticals (28) 11.85

KPMG Peat Marwick Electrical and electronic equipment (36) 20.38Machinery and computer equipment (35) 14.07Transportation equipment (37) 13.49

Price Waterhouse Petroleum refining (29) 25.83Machinery and computer equipment (35) 18.83Transportation equipment (37) 13.68

Sales is used as the base in calculating portfolio share. The following example illustrates the calculation of portfolio sharesfor Coopers & Lybrand (CL). For the period 1989 through 1998, the total sales of all clients of CL in the transportationequipment industry (two-digit SIC code 37) amounted to $1,094,132.01 million. During the same period, the combinedsales of all clients across all industries served by CL amounted to $6,350,583.13 million. Thus, CL�s portfolio share forthe transportation equipment industry = ($1,094,132.01/$6,350,583.13) × 100 = 17.23%. Portfolio shares for other Big 6auditors are calculated in a similar manner.For each auditor, top-three portfolio shares are coded as industries where the auditor is considered a specialist. Totalnumber of firm-year observations equals 24,114. The sample consists of 2,782 firm-year observations for specialist audi-tors and 21,332 observations for nonspecialist auditors.

share for each auditor, I require a minimum of ten observations for each pair of two-digit SIC codesand calendar years.

An example will demonstrate the computation of Big 6 IMS for CL for the transportationequipment industry. The numerator is the same for both Equations (1) and (2). In 1989, there were 47firms (including firms audited by other Big 6 auditors) in the transportation equipment industry; theirtotal sales amounted to $355,634.61 million. Big 6 IMS for CL for 1989 is calculated as follows:

Big 6 IMSCL,Transportation equipment, 1989 = $97,094.54/$355,634.61 = 0.2730.This indicates that CL�s share of the transportation equipment industry is 27.3 percent. Repeat-

ing the steps for each of the other nine years, I aggregate CL�s industry market shares over the years1989 through 1998. For the period 1989�1998, the total sales of all clients of CL in the transporta-tion equipment industry amounted to $1,094,132.01 million. There were 629 firm-years for thetransportation equipment industry, with combined sales amounting to $5,724,707.32 million. Big 6IMS for CL for the period is:

Big 6 IMSCL,Transportation equipment,1989�1998 = $1,094,132.01/$5,724,707.32 = 0.1911.

6 Krishnan


Shares for other industries served and for each Big 6 auditors are computed the same way to identifyindustry specializations.

Table 2 reports industry market shares of Big 6 auditors for the pooled sample for selected indus-tries (year-by-year industry market shares are not reported). A specialty is defined as any industry bytwo-digit SIC code where the auditor�s market share exceeds 15 percent. For example, in the period1989�1998, CL�s market share exceeded 15 percent in the following industries: metal mining, oil andgas, lumber and wood products, chemicals and pharmaceuticals, primary metal, fabricated metal, trans-portation equipment, communications, and food stores. According to this alternative measure, CL isdefined as a specialist auditor for these nine industries for the period 1989�1998.

TABLE 2Big 6 Auditor Industry Market Shares for Selected Industries for 1989�1998

Two-Digit Industry Market Shares (in %)

SIC Industry AA CL EY DT KPMG PW

10 Metal mining 41.35 30.38 10.24 2.39 3.98 11.6613 Oil and gas 50.84 18.96 3.29 3.49 5.38 18.0420 Food and kindred products 14.26 14.46 18.62 12.13 19.34 21.1822 Textile mill products 7.11 4.37 54.83 19.06 13.14 1.4923 Apparel and other finished products 19.82 6.05 43.64 13.95 2.41 14.1324 Lumber and wood products 61.67 17.99 1.74 5.49 0.86 12.2525 Furniture and fixtures 25.93 0.46 3.77 3.79 13.31 52.7426 Paper and allied products 44.72 6.64 9.97 18.28 10.08 10.3127 Printing and publishing 16.78 6.67 22.20 21.27 10.78 22.2928 Chemicals and pharmaceuticals 15.05 15.28 5.55 22.16 16.88 25.0829 Petroleum refining 18.63 2.66 23.85 0.28 17.13 37.4630 Rubber and plastics 9.25 4.11 21.74 1.33 7.71 55.8632 Stone, clay, glass, and concrete 43.95 0.52 31.05 11.43 8.28 4.7733 Primary metal 9.57 20.42 14.81 14.13 5.72 35.3534 Fabricated metal 8.38 23.72 19.74 5.78 20.99 21.3835 Machinery and computer equipment 8.43 6.26 10.92 6.27 29.14 38.9836 Electrical and electronic equipment 6.83 4.12 16.66 8.12 43.28 20.9937 Transportation equipment 4.20 19.11 13.75 32.79 14.97 15.1738 Scientific instruments 13.96 6.59 11.56 4.57 24.66 38.6645 Air transportation 44.59 0.00 35.22 2.18 16.75 1.2648 Communications 7.47 50.26 14.29 10.67 4.37 12.9450 Durable goods�wholesale 3.48 2.37 10.02 71.73 4.62 7.7753 General merchandise stores 10.02 1.24 44.03 23.35 10.86 10.5054 Food stores 2.23 45.26 2.53 37.57 5.32 7.09

AA: Arthur Andersen; CL: Coopers & Lybrand; EY: Ernst & Young; DL: Deloitte & Touche; KPMG: KPMG PeatMarwick; and PW: Price Waterhouse.Sales is used as the base in calculating industry market share. The following example illustrates the calculation of industrymarket shares for Coopers & Lybrand (CL). For the period 1989 through 1998, the total sales of all clients of CL in thetransportation equipment industry (two-digit SIC code 37) amounted to $1,094,132.01 million. During the same period,the combined sales of all clients in the transportation equipment industry amounted to $5,724,707.32 million. Thus, CL�smarket share in the transportation equipment industry = ($1,094,132.01/$5,724,707.32) × 100 = 19.11%. Industry marketshares for other industries are calculated in a similar manner.An auditor is coded as a specialist in industries where the auditor�s market share exceeds 15 percent. Bold indicates thatauditor is a specialist. Total number of firm-year observations equals 24,114. The sample consists of 12,221 firm-yearobservations for specialist auditors and 11,893 observations for nonspecialist auditors.



Portfolio Share versus Industry Market ShareBoth portfolio share and industry market share indicate that CL has a specialty in the following

industries: communications, transportation equipment, and food stores. In the other six industries,CL has a specialty based on industry market share but not portfolio share. A comparison of Tables 1and 2 shows that this result is even more marked for the other firms.

The industry market share measure tends to classify more firms as clients of specialist auditors thanthe portfolio share measure. About 51 percent of sample firms are classified as clients of specialistauditors according to industry market share, compared to about 12 percent according to portfolio share.An examination of year-by-year values of the portfolio shares and the industry market shares indicatesthat the top-three portfolios are fairly stable over the ten-year period for each Big 6 auditor (results notreported). The industry market shares exhibit more variation, suggesting that this may be a noisiermeasure of auditors� industry expertise. This is consistent with Krishnan�s (2001) argument that portfo-lio share captures the efforts of auditors to differentiate their products better than industry market share,and may be a better proxy for auditor industry expertise than industry market share.

Finally, the Pearson correlation coefficient (not reported) between portfolio share and industrymarket share is 0.45, significant at the 0.01 level. Pearson correlation coefficients between the twomeasures for each Big 6 auditor are positive and also significant at the 0.01 level for each auditor(results not reported).

RESEARCH METHODI estimate discretionary accruals using a cross-sectional variation of the Jones (1991) accruals

estimation model also used by DeFond and Jiambalvo (1994).2 I use the absolute value of discretion-ary accruals (ABDAC) as a proxy for accruals-based earnings management (Becker et al. 1998;Francis et al. 1999). In other words, all else equal, higher ABDAC is consistent with a conclusion thatauditors allow their clients to exercise greater accounting flexibility.

Several variables have been identified that are correlated with discretionary accruals (Becker etal. 1998; Bartov et al. 2000): size (SIZE) defined as log of total assets, and leverage (LEV) defined aslong-term debt divided by total assets. Since firms with higher absolute values of total accruals arelikely to have greater discretionary accruals, I include the absolute value of total accruals divided bytotal assets at the beginning of the year (ABACCR) as a control variable. DeFond and Subramanyam(1998) find that discretionary accruals are related to auditor changes. I include a dummy variable(NEWAUD) equal to 1 if the first sample year is the first year with a new auditor, and 0 otherwise.Similarly, another dummy variable (OLDAUD) is set equal to 1 if the last sample year is followed byan auditor change, and 0 otherwise. Further, following Bartov et al. (2000), to control for growth, Iinclude market-to-book (MKBK) ratio as a control variable.

2 Estimation of discretionary accruals involves two steps. First, nondiscretionary accruals are estimated using the cross-sectional version of the Jones (1991) model. This model estimates nondiscretionary accruals as a function of the levelof property, plant, and equipment, and changes in revenue:

where ACCRj,t is total accruals for firm j in year t; TA is total assets; ∆REV is change in net revenue; and PPE is property,plant, and equipment. Total accruals are calculated as the difference between net income before extraordinary items anddiscontinued operations and cash flows from operations. Consistent with prior research, this model is estimated sepa-rately for each combination of two-digit SIC codes and calendar years. Fitted values are defined as nondiscretionary(expected) accruals.

Second, the error term in the model (the difference between total accruals and nondiscretionary accruals) represents theunexplained or discretionary component of accruals.

The median estimates of α 1, α2, and α3 are �0.100, 0.049, and �0.071, respectively. The percentages of regressioncoefficients that are in the predicted direction are 68 percent and 96 percent for α2 and α3, respectively. The medianadjusted R2 is 0.24.

tjtj

tj

tj

tj

tjtj

tj eTAPPE

TAREV

TATAACCR

,1,

,3

1,

,2

1,1

1,

, αα1α ++∆

+=−−−−

8 Krishnan


Finally, I include two performance-related controls. First is earnings persistence (PERSIST).Following Ali (1994), observations in each year are partitioned into ten groups according to theabsolute value of change in income before extraordinary items. Observations in the four extremedeciles (top-two deciles and bottom-two deciles) are classified as low-persistence firms, and obser-vations in the middle six deciles are classified as high-persistence firms. Second, observations withnegative earnings (LOSS) are coded as 1; observations of profitable firms are coded as 0.

In all, eight control variables are included:

The variable of interest, SPECLST, is defined in four ways. As a continuous measure it equalsthe auditor�s portfolio share or its industry market share. As a dichotomous measure, SPECLSTequals 1 for specialist auditors and 0 for nonspecialist auditors, depending on the auditor�s portfolioshare or industry market share. An observation of ß9 < 0 is consistent with the notion that specialistauditors are able to constrain accruals-based earnings management.

RESULTSPearson correlation coefficients for the variables in Equation (3) are reported in Table 3.Correlations for the portfolio share measure appear above the diagonal, and correlations for the

market share measure below the diagonal. The correlation between SPECLST and ABDAC is negative aspredicted and statistically significant at the 0.01 level for both measures of auditors� industry expertise.This suggests that, other things equal, auditor industry expertise is associated with less accruals-basedearnings management.

Discretionary Accruals: Specialist versus Nonspecialist AuditorsPanel A of Table 4 reports descriptive statistics separately for clients of specialist and nonspe-

cialist auditors depending on the portfolio share measure, for income before extraordinary itemsover total assets at the beginning of the year, log of total assets, long-term debt over total assets, totalaccruals, absolute value of total accruals, absolute value of discretionary accruals, income-increas-ing discretionary accruals, and income-decreasing discretionary accruals.3

The results indicate that clients of specialist auditors are slightly more profitable, are larger, andcarry less debt than clients of nonspecialist auditors. These differences are significant at the 0.01level. Differences in mean and median values of measures of accruals-based earnings managementfor clients of specialist and nonspecialist auditors are significant at the 0.01 level for all the fivemeasures. More important, clients of nonspecialist auditors report higher discretionary accruals thanclients of specialist auditors.

Panel B of Table 4 presents statistics for clients of specialist and nonspecialist auditors depending onindustry market share. The differences in mean and median values of earnings management measures forclients of specialist and nonspecialist auditors are significant at the 0.01 level for four of five measures.

In summary, the level of the absolute value of discretionary accruals is lower for clients ofspecialist auditors under both measures of auditor industry expertise. I also calculate mean andmedian values of ABDAC by year, and compare the ten annual means and medians of the clients ofspecialist auditors to the ten annual means and medians of the clients of nonspecialist auditors. Thedifferences in mean and median are significant at the 0.01 level (results not reported).

Table 5 highlights the differences in mean value of ABDAC between clients of specialist andnonspecialist auditors for each Big 6 auditor. The histogram in Panel A is based on portfolio shares.Clients of specialist auditors have lower discretionary accruals than clients of nonspecialist auditors3 Mean, median, upper quartile, and lower quartile of discretionary accruals for the pooled sample are �0.006, �0.001,

0.044, and �0.051, respectively.

(3)µββββ

ββββββ

9876

543210

tttt

tttttt

SPECLSTLOSSPERSISTOLDAUDNEWAUDABACCRMKBKLEVSIZEABDAC

++++++++++=

.

Does Big 6 Auditor Industry Expertise C

onstrain Earnings Managem

ent?9

Accounting Horizons, Supplem

ent 2003

TABLE 3Pearson Correlation Coefficients for 1989�1998

ABDAC SIZE LEV MKBK ABACCR NEWAUD OLDAUD PERSIST LOSS SPECLSTABDAC 1.000 �0.293* �0.048* 0.005 0.661* 0.034 0.050 �0.038* 0.238* �0.094*

SIZE �0.293* 1.000 0.247* 0.100 �0.236* �0.075* �0.085* 0.332* �0.342* 0.125*

LEV �0.048* 0.247* 1.000 0.004 0.007 0.006 �0.001 0.071* 0.050* �0.037*

MKBK 0.005* 0.010* 0.004 1.000 0.008 0.002 0.001 0.003 �0.010 0.001ABACCR 0.661* �0.236* 0.007 0.008* 1.000 0.035* 0.066* �0.017* 0.285* �0.077*

NEWAUD 0.035* �0.075* 0.006 0.002 0.035* 1.000 0.012*** 0.001 0.050* �0.020*

OLDAUD 0.050* �0.085* �0.001 0.001 0.066* 0.012*** 1.000 �0.016** 0.086* 0.007PERSIST �0.038* 0.332* 0.071* 0.003 �0.017* 0.001 �0.016** 1.000 �0.026* 0.067*

LOSS 0.238* �0.342* 0.050* �0.010 0.285* 0.050* 0.086* �0.026* 1.000 �0.026*

SPECLST �0.088* 0.193* 0.023* �0.005 �0.062* �0.027* 0.016** 0.081* �0.064* 1.000

*, **, *** Represent statistical significance at the 0.01, 0.05, and 0.10 levels, respectively, for a two-tailed testCorrelation coefficients for the portfolio share measure appear above the diagonal; correlations for the market share measure appear below the diagonal.All firms were audited by Big 6 auditors. Total number of firm-year observations is 24,114. When auditors� industry expertise is measured based on the portfolio share measure, thesample consists of 2,782 and 21,332 observations for specialist and nonspecialist auditors, respectively. For the market share measure, the sample consists of 12,221 and 11,893observations for specialist and nonspecialist auditors, respectively. See Tables 1 and 2 for definitions of auditors� portfolio shares and industry market shares, respectively.ABDAC is absolute value of discretionary accruals where discretionary accruals are determined using the cross-sectional version of the Jones (1991) model (see footnote 2). SIZE isthe log of total assets. LEV is leverage calculated as long-term debt divided by total assets. MKBK is market-to-book ratio. ABACCR is absolute value of total accruals divided bytotal assets at the beginning of the year. NEWAUD equals 1 if first sample year is the first year with a new auditor, and 0 otherwise. OLDAUD equals 1 if the last sample year isfollowed by an auditor change, and 0 otherwise. PERSIST equals 1 for firms with low earnings persistence and 0 for firm with high earnings persistence. Persistence is measured asfollows: observations in each year are partitioned into ten groups based on the absolute value of change in income before extraordinary items. Observations in the four extremedeciles (top two deciles and bottom two deciles) are classified as low-persistence firms and observations in the middle six deciles are classified as high-persistence firms. LOSSequals 1 if income before extraordinary items is negative, and 0 otherwise. SPECLST equals 1 for clients of specialist auditors and 0 for clients of nonspecialist auditors.

10K

rishnan

Accounting Horizons, Supplem

ent 2003

TABLE 4Descriptive Statistics: Clients of Specialist versus Nonspecialist Auditors for 1989�1998

Panel A: Big 6 Auditor Portfolio SharesMean Median

Type of Accrual Specialist Nonspecialist t�statistic Specialist Nonspecialist z�statistic

PROFITABILITY 0.008 �0.009 3.94 * 0.048 0.040 4.09 *

SIZE 5.908 5.048 16.71* 5.538 4.915 14.61*

LEVERAGE 0.163 0.186 �6.56* 0.128 0.136 �3.20*

ACCR �0.044 �0.055 5.71 * �0.046 �0.050 4.27 *

ABACCR 0.073 0.099 �17.77 * 0.059 0.069 �9.74*

ABDAC 0.054 0.080 �22.44 * 0.039 0.049 �11.62 *

Income-increasing DAC 0.054 0.074 �13.27 * 0.039 0.047 �6.59*

Income-decreasing DAC �0.054 �0.085 18.14* �0.038 �0.052 9.70 *

Panel B: Big 6 Auditor Industry Market SharesMean Median

Type of Accrual Specialist Nonspecialist t�statistic Specialist Nonspecialist z�statistic

PROFITABILITY 0.003 �0.018 5.33 * 0.045 0.037 7.23 *

SIZE 5.565 4.717 30.67* 5.412 4.611 28.00*

LEVERAGE 0.188 0.178 3.53 * 0.151 0.114 8.67 *

ACCR �0.053 �0.054 0.36 �0.050 �0.048 �1.09ABACCR 0.089 0.103 �9.67* 0.064 0.070 �7.32*

ABDAC 0.069 0.085 �13.73 * 0.043 0.054 �13.47 *

Income-increasing DAC 0.065 0.079 �9.00* 0.042 0.051 �9.00*

Income-decreasing DAC �0.073 �0.090 10.25* �0.044 �0.056 9.97 *

* Represents statistical significance at the 0.01 level.Auditors are classified into specialists and nonspecialists based on their portfolio shares and industry market shares. See Tables 1 and 2 for definitions of portfolio shares andindustry market shares, respectively.

PROFITABILITY is income before extraordinary items over total assets at the beginning of the year; SIZE is log of total assets; LEVERAGE is long-term debt over total assets.ACCR is total accruals divided by total assets at the beginning of the year. ABACCR is absolute value of ACCR. ABDAC is absolute value of discretionary accruals. Discretion-ary accruals (DAC) are computed as the error term from the Jones (1991) model (see footnote 2). Income-increasing discretionary accruals are positive DAC. Income-decreasingdiscretionary accruals are negative DAC.Total number of firm-year observations equals 24,114 representing years 1989 through 1998. When auditors� industry expertise is measured based on the portfolio share mea-sure, the sample consists of 2,782 and 21,332 observations for specialist and nonspecialist auditors, respectively. For the industry market share measure, the sample consists of12,221 and 11,893 observations for specialist and nonspecialist auditors, respectively.

Tests are two-tailed. t-statistics are from t-tests of the differences in the means and z-statistics are from Wilcoxon two-sample tests.



TABLE 5Mean Values of Absolute Discretionary Accruals between Clients of Specialist

and Nonspecialist Big 6 Auditors for 1989�1998

Auditors are classified into specialists and nonspecialists based on portfolio share and industry market share. AA is ArthurAndersen; CL is Coopers & Lybrand; EY is Ernst & Young; DL is Deloitte & Touche; KPMG is KPMG Peat Marwick;and PW is Price Waterhouse. See Tables 1 and 2 for definitions of portfolio shares and industry market shares, respec-tively.Discretionary accruals are computed as the error term from the Jones (1991) model (see footnote 2). Total number of firm-year observations equals 24,114 representing years 1989 through 1998. When auditors� industry expertise is measuredbased on the portfolio share measure, the sample consists of 2,782 and 21,332 observations for specialist and nonspecial-ist auditors, respectively. For the industry market share measure, the sample consists of 12,221 and 11,893 observationsfor specialist and nonspecialist auditors, respectively.

0

0 .01

0 .02

0 .03

0 .04

0 .05

0 .06

0 .07

0 .08

0 .09

AA CL EY DT KPMG PW

B ig 6 Au d ito rs

Mea

n A

bsol

ute

Dis

c A

ccru

als

S pecialistN on-specialist

0

0 . 0 1

0 . 0 2

0 . 0 3

0 . 0 4

0 . 0 5

0 . 0 6

0 . 0 7

0 . 0 8

0 . 0 9

0 .1

AA C L E Y D T K P M G P W

B i g 6 A u d i t o r s

Me

an

Ab

so

lute

Dis

c A

S p e c ia l is tN o n -s p e c ia l i s t

Panel A: Auditor Industry Expertise Based on Portfolio Shares

Panel B: Auditor Industry Expertise Based on Industry Market Shares

Mea

n A

bsol

ute

Dis

c A

ccru

als

12 Krishnan


in every case. The differences in mean value of discretionary accruals between clients of specialistand nonspecialist auditors are significant at the 0.01 level in a two-tailed test (results not reported).

The histogram in Panel B is based on auditor industry market shares. Once again, clients ofspecialist auditors have lower discretionary accruals than clients of nonspecialist auditors, except inthe case of Price Waterhouse. Overall, these findings are consistent with the notion that specialistauditors mitigate accruals-based earnings management more than nonspecialist auditors.

Multivariate AnalysisResults in Tables 4 and 5 do not control for research confounds that might be associated with

discretionary accruals. Estimation of Equation (3) controls for variables correlated with discretion-ary accruals. I estimate Equation (3) year-by-year as well as for the pooled sample. Pooling dataallows for more powerful tests because of the larger sample size. An upward bias in t-statistics due tocross-sectional correlation in regression residuals is, however, a concern with models using annualdata (Bernard 1987).

I address cross-sectional correlation in two ways. In the pooled model, I include nine year-dummy variables DY to indicate fiscal years 1989 through 1997 and 14 industry-dummy variables DIrepresenting two-digit SIC code numbers 13, 20, 28, 33, 34, 35, 36, 37, 38, 48, 49, 50, 58, and 73.Each two-digit SIC code represents at least 2 percent of the total sample. The objective is to capturetime- and industry-specific commonalities in the dummy variable coefficients and thus reduce corre-lations among regression residuals.

The second procedure involves estimating cross-sectional regressions for each year. FollowingAli (1994), I test significance of parameter estimates using t-statistics for the cross-temporal distri-butions of the year-by-year estimates. I also examine intertemporal independence by analyzingcorrelations between residuals across years.

Table 6 presents descriptive statistics for the pooled and year-by-year samples for variables inEquation (3) based on portfolio share. Results using both the dichotomous variable specification and thecontinuous variable specification are reported. SPECLST is negative and statistically significant at the0.01 level for the pooled sample. Results for the year-by-year samples indicate that SPECLST is negativeas expected in each of the ten years, and the mean coefficient is significant at the 0.01 level. The year-by-year results also indicate that the pooled results are not just due to use of a large sample.

Table 7 reports the same descriptive statistics for the pooled and year-by-year samples forvariables in Equation (3) based on industry market share. SPECLST is negative and significant in thiscase at the 0.05 level. Overall, the results indicate strongly that the level of absolute discretionaryaccruals is negatively associated with auditors� industry expertise.

In summary, the results hold under both measures of auditors� industry expertise for both dichoto-mous and continuous variable specifications. Results are consistent with the notion that specialist audi-tors serve to mitigate accruals-based earnings management more than nonspecialist auditors.

Additional Tests for Robustness of FindingsI apply some additional tests to examine the sensitivity of my results to alternative variable

definitions and model specifications. I use the cross-sectional variation of the modified Jones (1991)model to estimate discretionary accruals. The mean values of absolute discretionary accruals underthe modified Jones model for clients of specialist and nonspecialist auditors are 0.053 and 0.076,respectively. The median values are 0.037 and 0.047, respectively. The differences in mean andmedian values are statistically significant at the 0.01 level. I re-estimate Equation (3) using discre-tionary accruals obtained from the modified Jones model, and the results are consistent with thosealready reported. For the market share measure, I increase the cutoff rate to 25 percent to identifyspecialist auditors and re-estimate Equation (3). The results are comparable to results reported inTable 7; SPECLST is negative and significant at the 0.01 level.



I re-estimate Equation (3) using an alternate measure of auditors� industry expertise�industrymarket share calculated using square root of total assets as the base instead of sales (Krishnan 2001).Once again, SPECLST is negative and significant at the 0.05 level. Next, I exclude two-digit SICcategories not represented by specialist auditors and estimate the equation using the remaining obser-vations. The results indicate that absolute discretionary accruals are lower for clients of specialistauditors, and SPECLST is negative and significant at the 0.01 level. My results are not sensitive to theinclusion of utilities (SICs from 4000 and 4999).

TABLE 6Regression of Absolute Discretionary Accruals on Control Variables

and Big 6 Auditor Portfolio Shares for 1989�1998

Pooled Sample Year-by-Year Samples

Independent Dichotomous Variable Continuous Variable Mean NumberVariables Coefficients t-statistic Coefficients t-statistic Coefficients t-statistic Positive

Intercept 0.058 30.97* 0.058 31.49* 0.058 13.57* 10/10SIZE �0.005 �23.06 * �0.005 �24.04 * �0.006 �13.70 * 0/10LEV �0.001 �0.48 �0.003 �1.33 �0.012 �3.83* 2/10MKBK 0.000 0.42 0.000 0.18 0.000 1.00 6/10ABACCR 0.495 122.97 * 0.537 131.28 * 0.497 17.89* 10/10NEWAUD �0.001 �0.72 0.002 1.37 0.000 0.16 4/10OLDAUD �0.002 �1.28 0.000 0.15 �0.004 �2.59** 2/10PERSIST 0.004 3.77 * 0.004 4.07 * 0.003 2.97 ** 9/10LOSS �0.000 �0.38 �0.003 �2.46** 0.002 0.97 7/10SPECLST �0.012 �8.47* �0.042 �5.01* �0.007 �9.66* 0/10Adjusted R2 0.479 0.501 0.467

*, ** Indicate two-tailed significance at the 0.01and 0.05 levels, respectively.Big 6 auditors are classified into specialists and nonspecialists based on their portfolio shares. Sales is used as the base incalculating the portfolio share (see Table 1 for more information on the calculation of portfolio share).ABDAC is absolute value of discretionary accruals where discretionary accruals are determined using the cross-sectionalversion of the Jones (1991) model (see footnote 2). SIZE is the log of total assets. LEV is long-term debt divided by totalassets. MKBK is market-to-book ratio. ABACCR is absolute value of total accruals divided by total assets at the beginningof the year. NEWAUD equals 1 if first sample year is the first year with a new auditor, and 0 otherwise. OLDAUD equals 1if the last sample year is followed by an auditor change, and 0 otherwise. PERSIST equals 1 for firms with low earningspersistence, and 0 for firm with high earnings persistence. Persistence is measured as follows: observations in each yearare partitioned into ten groups based on the absolute value of change in income before extraordinary items. Observationsin the four extreme deciles (top two deciles and bottom two deciles) are classified as low-persistence firms and observa-tions in the middle six deciles are classified as high-persistence firms. LOSS equals 1 if income before extraordinary itemsis negative, and 0 otherwise.SPECLST equals portfolio shares for the continuous variable specification. For the dichotomous variable specification,SPECLST equals 1 for industries representing the top three portfolio shares and 0 for the remaining industries. Year-by-year results are based on the dichotomous specification.The specification for the pooled sample include nine year-dummy variables DY for 1989 through 1997, and 14 industry-dummy variables DI representing two-digit SIC code numbers 13, 20, 28, 33, 34, 35, 36, 37, 38, 48, 49, 50, 58, and 73.Total number of observations equals 24,114 consisting of 2,782 and 21,332 observations audited by specialist and non-specialist auditors, respectively. Yearly firm-year observations are 1,481, 1,746, 1,806, 1,857, 2,046, 2,302, 2,624, 3,009,3,510, and 3,733 for 1989 through 1998, respectively.The pooled sample aggregates the individual year samples. Means of individual-year parameter estimates are reported forthe year-by-year models with t-values (with 9 degrees of freedom) estimated from the cross-temporal distribution of theseestimates. Reported adjusted R2s for the year-by-year samples are cross-temporal means.

ttt

ttttttt

SPECLSTLOSSPERSIST

OLDAUDNEWAUDABACCRMKBKLEVSIZEABDAC

µ+β+β+β+

β+β+β+β+β+β+β=

987

6543210

14 Krishnan


TABLE 7Regression of Absolute Discretionary Accruals on Control Variables

and Big 6 Auditor Industry Market Shares for 1989�1998

Pooled Sample Year-by-Year Samples

Independent Dichotomous Variable Continuous Variable Mean NumberVariables Coefficients t-statistic Coefficients t-statistic Coefficients t-statistic Positive

Intercept 0.058 30.41* 0.058 30.08* 0.057 11.81* 10/10SIZE �0.005 �23.08* �0.005 �23.16* �0.005 �11.77* 0/10LEV �0.001 �0.49 �0.001 �0.45 �0.012 �3.84* 2/10MKBK 0.000 0.38 0.000 0.38 0.000 0.64 8/10ABACCR 0.513 126.48* 0.513 126.49* 0.517 23.24* 10/10NEWAUD �0.001 �0.69 �0.001 �0.69 0.001 0.27 4/10OLDAUD �0.003 �1.76*** �0.003 �1.76*** �0.005 �3.18** 1/10PERSIST 0.003 3.44* 0.003 3.45* 0.002 2.50** 8/10LOSS �0.001 �1.05 �0.001 �1.05 0.001 0.46 5/10SPECLST �0.002 �2.45** �0.005 �2.08** �0.005 �2.34** 1/10Adjusted R2 0.488 0.488 0.483

*, **, *** Indicate two-tailed significance at the 0.01, 0.05, and 0.10 levels, respectively.Big 6 auditors are classified into specialists and nonspecialists based on their industry market shares. Sales is used as thebase in calculating the industry market share (see Table 2 for more information on the calculation of industry marketshare).ABDAC is absolute value of discretionary accruals where discretionary accruals are determined using the cross-sectionalversion of the Jones (1991) model (see footnote 2). SIZE is the log of total assets. LEV is long-term debt divided by totalassets. MKBK is market-to-book ratio. ABACCR is absolute value of total accruals divided by total assets at the beginningof the year. NEWAUD equals 1 if first sample year is the first year with a new auditor, and 0 otherwise. OLDAUD equals 1if the last sample year is followed by an auditor change, and 0 otherwise. PERSIST equals 1 for firms with low earningspersistence and 0 for firm with high earnings persistence. Persistence is measured as follows: observations in each year arepartitioned into ten groups based on the absolute value of change in income before extraordinary items. Observations inthe four extreme deciles (top two deciles and bottom two deciles) are classified as low-persistence firms and observationsin the middle six deciles are classified as high-persistence firms. LOSS equals 1 if income before extraordinary items isnegative, and 0 otherwise. SPECLST equals industry market shares for the continuous variable specification. For thedichotomous variable specification, SPECLST equals 1 for industries where the auditor�s market share exceeds 15 per-cent, and 0 for the remaining industries. Year-by-year results are based on the dichotomous specification.The specification for the pooled sample include nine year-dummy variables DY for 1989 through 1997, and 14 industry-dummy variables DI representing two-digit SIC code numbers 13, 20, 28, 33, 34, 35, 36, 37, 38, 48, 49, 50, 58, and 73.Total number of observations equals 24,114 consisting of 12,221 and 11,893 observations audited by specialist andnonspecialist auditors, respectively. Yearly firm-year observations are 1,481, 1,746, 1,806, 1,857, 2,046, 2,302, 2,624,3,009, 3,510, and 3,733 for 1989 through 1998, respectively.The pooled sample aggregates the individual year samples. Means of individual-year parameter estimates are reported forthe year-by-year models with t-values (with 9 degrees of freedom) estimated from the cross-temporal distribution of theseestimates. Reported adjusted R2s for the year-by-year samples are cross-temporal means.

ttt

ttttttt

SPECLSTLOSSPERSIST

OLDAUDNEWAUDABACCRMKBKLEVSIZEABDAC

µ+β+β+β+

β+β+β+β+β+β+β=

987

6543210

As an additional diagnostic check, I compare the nondiscretionary accruals of clients of specialistauditors and nonspecialist auditors. The ability of the specialist auditors in constraining earnings man-agement should be more evident in the level of discretionary rather than nondiscretionary accruals.Consistent with this expectation, I find that the differences in mean and median values of nondiscretionaryaccruals between the clients of specialist and nonspecialist auditors are not significant at the 0.10 level.

Finally, I examine in a two-stage analysis whether self-selection of clients of specialist andnonspecialist auditors is driving the observed differences in discretionary accruals. In the first stage,



I run a logistic model of auditor choice similar to the one used by Francis et al. (1999) and computethe inverse Mills ratio (IMR) (see Berndt 1991, Chapter 11). In the second stage, I estimate Equation(3) after including the IMR as an additional independent variable. The results indicate that SPECLSTis negative and significant at the 0.01 level. This finding mitigates concern that sample self-selectionis driving the reported results.

The results of all these additional tests confirm the basic finding that clients of nonspecialistauditors consistently exhibit higher levels of absolute discretionary accruals than clients of specialistauditors.

CONCLUDING REMARKSThe rise of accruals-based earnings management in recent years has prompted calls for reforms

to restore confidence in reported accounting information. Specialist auditors have the expertise, theresources, and the incentive to constrain opportunistic reporting of accruals and thereby enhance thequality of earnings. This study examines whether auditors� industry expertise mitigates the tendencyof managers to engage in accruals-based earnings management.

When Big 6 auditors are partitioned into specialists and nonspecialists, I find that clients ofnonspecialist auditors exhibit higher levels of discretionary accruals than clients of specialist audi-tors. This finding persists after controlling for firm size, industry effects, and other factors that areknown to affect discretionary accruals. In summary, the finding is consistent with the notion thatauditors� industry expertise moderates accruals-based earnings management.

One implication is employment of a Big 6 auditor that is also an industry specialist can furtherenhance the credibility of accounting information. While audit firms have undertaken significantrestructuring efforts along industry lines, empirical evidence on the payoff from these efforts islimited. The findings of this research support the notion that there are likely returns to investing inspecialization in the form of increased audit effectiveness and associated credibility.

I should note one caveat. One cannot rule out the possibility that audit clients with lowerdiscretionary accruals tend to self-select specialist auditors, even though a sensitivity test suggeststhat the results are not driven by sample self-selection.

REFERENCESAli, A. 1994. The incremental information content of earnings, working capital from operations, and cash

flows. Journal of Accounting Research (Spring): 61�74.Bartov, E., F. Gul, and J. Tsui. 2000. Discretionary-accruals models and audit qualifications. Journal of

Accounting and Economics (December): 421�452.Becker, C., M. DeFond, J. Jiambalvo, and K. Subramanyam. 1998. The effect of audit quality on earnings

management. Contemporary Accounting Research 15 (Spring): 1�24.Bedard, J., and S. Biggs. 1991. The effect of domain-specific experience on evaluation of management

representation in analytical procedures. Auditing: A Journal of Practice & Theory (Supplement): 77�95.Bernard, V. 1987. Cross-sectional dependence and problems in inference in market-based accounting research.

Journal of Accounting Research 25 (Spring): 1�48.Berndt, E. 1991. The Practice of Econometrics: Classic and Contemporary. New York, NY: Addison-Wesley.Bonner, S., and B. Lewis. 1990. Determinants of auditor expertise. Journal of Accounting Research (28): 1�28.DeAngelo, L. 1981. Auditor size and auditor quality. Journal of Accounting and Economics 3 (December):

183�199.DeFond, M., and J. Jiambalvo. 1994. Debt covenant violations and manipulation of accruals. Journal of

Accounting and Economics 17 (January): 145�176.��, and K. Subramanyam. 1998. Auditor changes and discretionary accruals. Journal of Accounting and

Economics 25 (February): 35�68.

16 Krishnan


Dopuch, N., and D. Simunic. 1982. The competition in auditing: An assessment. In Fourth Symposium onAuditing Research. Urbana, IL: University of Illinois at Urbana�Champaign.

Francis, J., E. Maydew, and H. Sparks. 1999. The role of Big 6 auditors in the credible reporting of accruals.Auditing: A Journal of Practice & Theory 18 (Fall): 17�34.

Gramling, A., and D. Stone. 2001. Audit firm industry expertise: A review and synthesis of the archivalliterature. Working paper, Georgia State University.

Healy, P., and J. Wahlen. 1999. A review of the earnings management literature and its implications forstandard setting. Accounting Horizons (December): 365�383.

Johnson, P., K. Jamal, and R. Berryman. 1991. Effects of framing on auditor decisions. Organization Behaviorand Human Decision Processes (50): 75�105.

Jones, J. 1991. Earnings management during import relief investigations. Journal of Accounting Research 29(Autumn): 193�228.

Krishnan, J. 2001. A comparison of auditors� self-reported industry expertise and alternative measures ofindustry specialization. Asia-Pacific Journal of Accounting & Economics (8): 127�142.

Kwon, S. 1996. The impact of competition within the client�s industry specialization on the auditor selectiondecision. Auditing: A Journal of Practice & Theory (Spring): 53�70.

MacDonald, E. 1997. More accounting firms are dumping risky clients. Wall Street Journal (April 25).Maletta, M., and A. Wright. 1996. Audit evidence planning: An examination of industry error characteristics.

Auditing: A Journal of Practice & Theory (Spring): 71�86.O�Keefe, T., R. King, and K. Gaver. 1994. Audit fees, industry specialization, and compliance with GAAS

reporting standards. Auditing: A Journal of Practice &Theory (Fall): 41�55.Solomon, I., M. Shields, and R. Whittington. 1999. What do industry-specialist auditors know? Journal of

Accounting Research (Spring): 191�208.Wright, S., and A. Wright. 1997. The effect of industry specialization on hypothesis generation and audit

planning decisions. Behavioral Research in Accounting (9): 273�294.Yardley, J., N. Kauffman, T. Cairney, and D. Albrecht. 1992. Supplier behavior in the U.S. audit market.

Journal of Accounting Literature (11): 151�184.

review journal auditing

Documents