Auditing: A Journal of Practice & Theory Vol. 11, No.

1 Spring 1992

16

The Association Between Changes in Client Firm Agency Costs and Auditor Switching
Mark L. DeFond
SUMMARY Researchers have posited that auditors speciaiize in the ievel of audit "quality" provided to clients. "Quality" is defined as the probability that the auditor will both detect and report material breaches in the accounting system. Differing quality levels are demanded by clients based upon how closely management's incentives align with those of the company's owners. The disparity between management and owner incentives results in agency conflicts. The extent of these agency conflicts determines the degree of auditing needed to make management credible to current and potential investors. Specifically, the higher (lower) the extent of the agency conflicts, the higher (lower) the demand for audit quality. This study explores this relationship by examining 131 auditor changes for an association between changes in auditor quality and changes in agency conflicts around the time of the auditor change. The statistical technique of principal components analysis is used to model changes in auditor quality as a combination of auditor size, name-brand, industry expertise, and independence. Agency conflicts are proxied by leverage, management ownership, and size of short-term accruals. Changes in agency conflicts are measured over a period spanning two years before the auditor change through two years after, whereas previous research focused on a single point in time. The results provide support for the hypothesis that changes in management ownership and leverage are associated with changes in audit quality.

Mark L. DeFond, Assistant Professor, University of Southern California, School of Accounting, Los Angeles, CA 90089-1421.
The author is indebted to his dissertation committee members Dave Burgstahler, Gary Sundem and especially Jim Jiambalvo for their help and encouragement. In addition, the paper ben-

efitted greatly from the comments of Walt Blacconiere, Ces Jackson, Jim Manegold, Terry Shevlin, D. Shores and workshop participants at the Universities of Arizona and California at Berkeley, Columbia University, the Universities of Iowa, Oregon, Minnesota, Pittsburgh and Southern California. Two anonymous referees also provided suggestions that improved ihe paper. Partial funding for this project was provided by the Deloitte, Haskins and Sells Foundation.

DeFond

17

independently of changes in firm growth and securities issues. The associations also reveal that managers seem to change auditors in anticipation of some agency conflicts and in reaction to others. In addition, the namebrand surrogate yields qualitatively identical results to more complex measures that combine multiple surrogates.

'EVERAL studies have looked for an association between client firm characteristics and auditor switches.' While recognizing that there are many possible reasons for auditor switches, this paper focuses on the branch of this research that has attempted to link the degree of agency conflicts within a firm to the firm's choice of extemal auditor. Previous research is extended in two ways. First, several surrogates are considered for the auditor's ability to alleviate agency conflicts. Surrogates include "independence," constructed as the ratio of the client firm revenues to the total revenues of the auditor's client base (DeAngelo 1981b),^ and a factor composed of auditor size, industry expertise, namebrand, and independence. Second, this paper examines changes in agency conflict variables over time, whereas previous research has generally focused on a single point in time.'' By examining changes over time, this paper considers that managers may change auditors in anticipation of some agency conflict changes and in reaction to others.''

THEORY Demand for Audit Quality Jensen and Meckling (1976) suggest that the demand for auditing results from a desire to reduce the management shirking that results from information asymmetries between managers and owners. They demonstrate that managers will voluntarily increase the observability of their actions by hiring independent auditors to monitor their behavior.^ Expanding on this, DeAngelo (1981a, 1981b) develops a demand and supply rationale for what she terms "audit quality." Using the work of Watts and Zimmerman (1980), she defines audit quality as the probability that an auditor will both (1) discover a breach in the accounting system, and (2) report the breach. Quality, then, captures the attribute of the audit service that helps alleviate the agency conflicts between the manager and equity holders. DeAngelo further argues that auditors will specialize in supplying varying levels of quality, which means that if a firm wishes to change audit quality it must also change auditors. DeAngelo asserts that audit firm size is an effective surrogate for audit quality because it proxies for the magnitude of the audit firm's client-specific quasi-rents. Quasi-rents equal the excess of audit fees over the avoidable costs of performing the audit.

'This literature includes Smith and Nichols (1982), Simunic and Stein (1987), Healy and Lys (1986), Eichenseher and Shields (1989), Johnson and Lys (1990), Francis and Wilson (1988), Palmrose (1984), and Smith (1986). ^It is noted, however, that the AICPA Code of Ethics does not recognize this measure as a surrogate for independence. ^The exception is Francis and Wilson (1988), that examined changes in agency conflict variables in the period prior to a change in auditor. By looking at changes prior to the change in auditor there is an implicit assumption that managers react to these changes rather than anticipate them. "It is acknowledged that managers may not be the only parties responsible for auditor selection. Jensen (1982) and Leddy (1982) note that audit committees of the board of directors and shareholders may also be involved in this process.

'It may be argued that extemal auditing is not necessarily the best response to this demand for monitoring. However, if it is assumed that the production of total monitoring is met by a combination of two factors, (1) internal controls (including internal auditing), and (2) external auditing, and that these are both normal goods, then under relatively mild assumptions, as the production of monitoring increases, the demand for extemal auditing will at least not decrease.

18
Audit Credibility Dopuch and Simunic (1982) also develop a demand model for auditing. Using the concept of "credibility," they define "greater credibility" to mean that the audited financial statements are more likely to be free of intentional misrepresentation. While their model differs from DeAngelo's, the implications are similar. That is, both "quality" and "credibility" refer to that characteristic of auditing that reduces the agency conflict between managers and equityholders. Simunic and Stein (1987) discuss the motives for auditing firms to specialize in providing a specific credibility level. They argue that audit firms choose to specialize in specific credibility levels based on a desire for monopoly rents. Initially, auditors choose credibility levels where they perceive a demand but no supply. Hypothesizing that there are large fixed costs to establishing a name-brand reputation at a given credibility level, Simunic and Stein argue that once established, the auditor finds it costly to change credibility levels. Simunic and Stein further maintain that specific audit firms may not provide multiple levels of credibility due to the inflexibility inherent in an organization that specializes. If the organization is "geared up" to provide a specific credibility level {i.e., product), then it is not likely to be cost effective in competing with suppliers who specialize at other levels.

Auditing, Spring 1992

table 1 are discussed, as appropriate, in subsequent sections. Surrogates for Audit Quality

Auditor Size. Francis and Wilson (1988) and Johnson and Lys (1990) use the ratio of the sales of the clients of the new and old auditor to proxy for the change in audit firm size. Size based upon client sales is a good quality surrogate to the extent that client sales revenues are correlated with quasi-rents. The surrogate for change in auditor size used here is similar to the measure used previously. The revenues of client firms audited by the old auditor are subtracted from the revenues of client firms audited by the new auditor and this difference is divided by the larger of the old or new audit firm client revenues. This variable is bounded by minus one and one, with positive numbers indicating a switch to a larger auditor. Client firm sales revenues are obtained from Who Audits America (1979 through 1984). Name-Brand Reputation. Prior studies all used membership in the Big Eight (now Big Six) to characterize name-brand reputation.* While Big Eight firms have an international reputation, second tier firms have a national reputation. In this study, the change in namebrand reputation is computed by assigning a value of two to Big Eight firms, one to second tier firms and zero to local firms (those that do not fit into the other categories). The value of the audit firm prior to the switch then is subtracted from the value HYPOTHESES of the firm subsequent to the switch to yield Previous studies that have looked at the a rank-ordered variable taking on one of five association between agency conflicts and values ( - 2 , - 1 , 0, 1, 2). Positive numbers audit quality are summarized in table 1. indicate an increase in name-brand reputaMethodologically, all of these papers ex- tion while negative numbers indicate a decept Francis and Wilson (1988) have at- crease.^ The criterion for classification as a tempted to relate audit quality level with the level of agency conflicts measured at a point "Simunic and Stein (1987), however, excluded from in time. Francis and Wilson (1988) looked their tests another commonly recognized group of auat both "levels" and "changes" in agency ditors known as "second tier" firms. 'This ranking imposes a specific interval ordering conflicts over the period before the switch. on the changes between these groups of auditors when Johnson and Lys (1990), though not testing the theory only supports an ordinal ordering. Comagency conflict hypotheses, are included in bining this variable with the other measures of audit quality using principal components analysis (which is table 1 because they looked at the control discussed in a following section) is expected to mitivariables used in this study. The papers in gate this problem to some degree.

DeFond TABLE 1 Summary of Multivariate Results of Prior Studies with Respect to the Agency Variables Examined in this Study Independent Variables Management Client Ownership Leverage Size

19

Study "Levels" Tests" Palmrose (1984) Simunic and Stein (1987) Healy and Lys (1986) Eichenseher and Shields (1989) Francis and Wilson (1988)

Dep. Var.

Securities Issues

Name-brand Expertise Name-brand Name-brand Name-brand Name-brand Size

0 0

0 0
e

n/a n/a
n/a n/a n/a
0

+ +
0

n/a
+e

0"

+
0

0 0 0 0
n/a

0 0

"Changes" Tests'' ^ ' Francis and Name-brand Wilson (1988) Size Johnson and Lys (1990) Size

0 n/a

. / -

-1-

Number of tests that found POSITIVE significance NEGATIVE significance NO significance Total times tested

1 1 6 8

2 2 5 9

6 1 4 11

3 0 2 5

/+ Indicates variable was found significant at p < .05. 0 Indicates variable was not found significant. "These studies measured "levels" of their independent variables at a point in time. ""These studies measured "changes" in their independent variables over time. "Francis and Wilson (1988) had both tests of changes and tests of levels, and are, therefore, included in both categories. ''This is a change variable included in their model. 'These results were significant, but the direction is contrary to that hypothesized. 'Asset growth was positive before the switch and negative afterward.

national firm is that used by Who Audits America. Expertise. Shockley and Holt (1983) find that bank loan officers appear to use industry expertise (as measured by industry mar-

ket share) to assess audit firm credibility. Audit firms with expertise may provide greater assurance that financial statement breaches will be detected because the audit firm has a disproportionate amount of rep-

20 utation at stake in the industry in which it specializes.* The industry expertise surrogate for quality used in this study classifies a firm as an "expert" if its market share in the client's industry is ten percent or greater. Client firms are then coded one if they switched from a non-expert to an expert, zero if they experienced no change in expertise, and minus one if they switched from an expert to a non-expert. Independence. The characteristic that is traditionally held to be the strongest indicator of the audit firm's willingness to report a breach is the auditor's perceived independence (see Mautz and Sharaf (1961) or any standard auditing text). The argument is that the larger a specific client firm's fees are in relation to the total fees earned by the audit firm, the less willing the audit firm will be to disclose a breach for fear of losing the client. A measure which considers the size of the client firm relative to the size of the audit firm has not been used in previous research. Since fee data are not available, the size of the client relative to the audit firm's total client base is estimated using data on client revenues. The variable is computed as the difference between the ratio of the switching client firm's revenues to the total revenues of the clients of the old auditor, minus the same ratio for the new auditor. Notationally, J where INDEPj = measure of change in independence for observation i.
'Audit firms with industry expertise may also provide lower cost audits, at the margin, than non-experts due to efficiencies that are gained from specialized knowledge in the industry. However, acquiring and maintaining expertise is costly and, therefore, audit firms may demand a return on their expertise which equals or exceeds the savings from more efficient auditing. For example, Simunic and Stein (1990) demonstrate that a cost of industry concentration is less diversification in the auditor's portfolio of clients, which increases audit risk at the portfolio level. They point out that this may explain why no individual audit firm services 100 percent of an industry.

Auditing, Spring 1992

R; = revenues of client in observation i. TRioid = total revenues of the predecessor auditor in observation i. TRinew = total revenues of the successor auditor in observation i. The audit firm's total client revenues are found in Who Audits America and is the same variable used in the size proxy. This ratio is bounded by one and minus one, with positive numbers indicating a switch to a larger audit firm. Combined Measure. Size, name-brand, expertise, and independence are used to capture the same underlying constructthe auditor's ability to alleviate agency conflicts. They are, however, expected to be imperfect measures of audit quality. Therefore, considering the measures as a group may provide more information on quality than considering them individually. For purposes of performing hypotheses tests, if each of the auditor characteristics is a noisy measure of audit quality, combining them should increase the power of the tests by reducing noise in the dependent variable. The technique of principal components analysis is used in this study to combine the four variables.' Principal components analysis is a factor analysis technique used to extract common factors from a set of variables (Cooley and Lohnes, 1971). This is achieved by performing an eigenvalue analysis on the correlation matrix of the variables of interest to determine the linear combination of the variables that will account for the maximum amount of variance. The common factor is used in the hy-

(1)

'An alternative approach would be to consider each of the measures as measures of different but complementary concepts. In that case, it would be more appropriate to use an equally weighted combination of the measures that would maintain the unique components of each of the variables. This alternative was explored by testing a model which equally weighted the dependent variables (after standardizing their variance). The results were essentially identical to the principal component results.

DeFond

21 various actions (Jensen and Meckling 1976). Debt agreements commonly include covenants based on accounting information which limit these wealth transfers, thereby reducing the residual loss. Auditing increases the reliability of the accounting information used to verify covenant compliance. As the amount of debt increases, the potential amount of the wealth transfer increases, resulting in a greater incentive to transfer wealth away from debtholders and a greater demand for monitoring. This leads to the second hypothesis: Hypothesis 2: Client firms tend to switch to higher (lower) quality audit firms in anticipation of, or as a result of, increases (decreases) in leverage. Again, table 1 indicates that prior results with respect to this variable have been mixed and generally do not support the hypothesis. Leverage is measured here as the change in the ratio of long-term debt to total assets. Accruals. Compensation schemes attempt to reward managers on their marginal product. Since marginal product is not perfectly observable, compensation usually is based upon some variable that is expected to be correlated with marginal product. A frequently used variable is accounting income. Even when compensation is not explicitly based upon income it is expected to be an implicit factor in determining annual raises and the manager's worth in the labor market. The determination of income necessarily involves judgment and discretion, which gives the manager an opportunity to manipulate income. Healy (1985) suggests that short-term accruals (accounts receivable, accounts payable, and inventory) are income determinants vulnerable to management manipulation. The susceptibility of these accounts to misstatement also is evidenced by the relatively large amount of time devoted to these accounts in the audit process. Therefore, the relatively larger these accounts, the greater the vulnerability to

potheses tests as a measure of the auditor's ability to alleviate agency conflicts. Agency Conflict Hypotheses The agency theory literature identifies two aspects of the agency relationship that, in combination, create the agency problem: (1) the divergence in preferences of the manager and owner with respect to the manager's actions, and (2) the imperfect observability of the manager's actions by the owner. When the divergence in preferences increases or the observability of the manager's actions decreases, the residual loss increases. The variables that proxy for the severity of the agency conflict, therefore, are chosen for their hypothesized effect on these two features of the agency problem. Management Ownership. Jensen and Meckling (1976) demonstrate that as the manager's ownership increases, the cost to the manager of consuming perquisites also increases, because the manager bears a larger share of the costs of his or her actions. This implies that the greater the ownership interest of the manager, the more closely aligned his or her preferences are with those of the outside owners and leads to the first (alternative) hypothesis: Hypothesis 1: Client firms tend to switch to higher (lower) quality audit firms in anticipation of, or as a result of, decreases (increases) in the percentage of management ownership. Table 1 indicates that prior results with respect to this variable have been mixed and generally do not support the hypothesis. Consistent with Simunic and Stein (1987) and Eichenseher and Shields (1989) the management ownership variable used in this study is measured as the change in the percentage of stock owned by the managers and directors as disclosed in the firm's 10-K or proxy statement. Leverage. Managers have opportunities to transfer wealth from debtholders by taking

22 manipulation and the greater the demand for monitoring. This yields the third hypothesis: Hypothesis 3: Client firms tend to switch to higher (lower) quality audit firms in anticipation of, or as a result of, increases (decreases) in the relative size of short-term accruals. Previous studies have not examined this variable. Simiilar to Healy (1985), short-term accruals are measured as current assets plus current liabilities minus cash, notes receivable, and notes payable. Control Variables Growth. An alternative hypothesis for observing a significant correlation between the above independent variables and the dependent variable chosen to characterize audit firms is that changes in the chosen independent variables are correlated with changes in client firm size (or growth). A measure of client firm growth is included to control for this possible alternative hypothesis. All prior studies in this area used measures of firm size, with some support for this variable (see table 1). The size variable is measured here as the percentage growth in client firm assets."* Securities Issues. Carpenter and Strawser (1971) argue that changing from a smaller to a larger auditor increases the marketability of new securities. That is, larger auditors lend greater credibility to financial statements used in assessing the value of the securities. This line of thought is examined more completely by Titman and Trueman (1986). They argue that changing auditors provides a signal to the market regarding information about the firm that can be derived from the firm's accounting data.

Auditing, Spring 1992

Francis and Wilson (1988) and Johnson and Lys (1990) both include variables for public issues and find that issues in the years subsequent to the auditor change are positively correlated with the change in auditor size and name-brand (see table 1). Since the sample used in this study has only three public issues, the control variable used here is designed to control for all security issues that might involve "marketing" the firm's securities to outsiders. That is, it also controls for private placements and exchanges of stock or debt in acquisitions. METHODOLOGY Cross Temporal Measures Francis and Wilson (1988) performed multivariate tests looking for an association between changes in agency conflict variables and changes in auditor size and namebrand. They measured the changes in agency conflict variables over a period prior to the change in auditor. This focus on the conflicts prior to the change in auditor suggests an assumption that managers react to changes in agency conflicts. However, managers may also be anticipating agency conflict changes when they switch auditors. For example, an increase in audit quality may be made in anticipation of an increase in leverage (which increases agency conflicts) in order to obtain better credit terms." Similarly, if a decrease in management ownership (an increase in agency conflicts) is accompanied by a public stock offering, audit quality may be increased ex-ante in order to gain the confidence of outside investors. Johnson and Lys (1990) examined changes in various

'"However, because some companies in the sample experienced unusually high growth over the years tested, several extreme values occurred in the measurement of this variable. Therefore, growth was truncated at 100 percent (a doubling in size) so as not to give undue weight to these high-growth companies.

"A rationale for increasing audit quality in the period subsequent to an increase in leverage is found in Johnson and Lys (1990, 287-288). They note that "large accounting firms can better achieve efficient diversification of audit risks, and can therefore render services to risky clients at a lower fee than would be charged by firms with limited diversification." They attribute this observation to Eichenseher and Shields (1983) and use it to motivate a univariate analysis that includes the change in leverage before the auditor change.

DeFond

23 systematic. This may be the result of a failing firm's desire to either (1) find a cheaper (lower quality) auditor in order to save money, or (2) seek the consulting services available from Big Eight auditors. Due to the lack of a theory that differentiates these conflicting choices, all firms in bankruptcy at any time during the periods considered are eliminated from the sample. Firms that switch more than once during a two-year period on either side of the selected switch date are excluded from the sample in order to eliminate firms that may be switching auditors simply to capture the short-term advantages of low-balling. Firms that are 20 percent or greater owned subsidiaries of other firms also are excluded.''' Twenty percent is the ownership threshold at which the results of subsidiaries must be included in the financial statements of the parent and, hence, the point at which the parent's auditors must rely on the subsidiary's audited financial statements. In these cases, a change in the subsidiary's auditor may be due to the influence of the parent. In addition, firms that switched due to mergers also are excluded." Measuring the pre-merger agency conflict variables is problematic since there are two sets of financial statements from which to create the variables. Firms that received adverse or disclaimer opinions are eliminated as well, since their financial information is deemed unreliable. The data selection involved a two step procedure. First, a sample was chosen that included all of the switches in the 10th edition of Who Audits America (June 1983) that involved a change in auditor name-brand {e.g., from a non-Big Eight to a Big Eight auditor). This was done to assure that the
'"The exceptions are firms that are 20 percent or greater owned by firms that appear to be stock ownership vehicles, such as family owned trusts and stock brokerage firms that hold stock in street name. "One firm was also excluded because the new auditor's opinion revealed that the change was due to an audit firm merger. Due to lack of information on small audit firm acquisitions, the sample may still include some firms that changed due to audit firm mergers. This is a source of noise in the data.

client firm attributes over periods both before and after the auditor switch. While they did not look specifically at changes in the agency conflict variables examined here, their study does indicate that managers may change auditors both in reaction to and in anticipation of changes in firm operations and activities.'^ If some agency conflict changes are anticipated and some unanticipated (by management), auditor changes will be driven by agency conflict changes that occur both before and after the auditor switch. There is, however, no clear way to identify the anticipated versus unanticipated changes in the agency conflict surrogates measured here.'^ Therefore, in an effort to capture the causal changes expected to be found both before and after the auditor switch, the agency conflict variables are measured over the period from two years before to two years after the specific year in which the auditor change occurred. In addition, this four-year period is disaggregated into the sub-periods before and after the switch, which provides a test of the sensitivity of the results to the time period tested. Controlling for Non-Agency Cost Switches and Sample Selection In addition to agency-cost motivated changes, other factors may induce clients to change auditors. Therefore, to reduce noise and avoid the need to proxy for non-agency cost variables, several categories of switching firms are excluded from the tests. Schwartz and Menon (1985) find that firms in bankruptcy switch auditors more frequently than other firms and that the change in monitoring level (as measured by the Big Eight vs. non-Big Eight dichotomy) is un-

'^While not considered in their multivariate tests, Johnson and Lys (1990) did perform univariate tests of changes in leverage in the period prior to the change in auditor. '^Were it possible to identify predictable versus unpredictable changes in these variables, the most powerful test would be to include only the unpredictable agency confiict changes before the switch and only the predictable changes after the switch.

24 sample included a reasonable number of firms that actually experienced quality changes. Who Audits America (June 1983) contains a listing of the 1,861 publicly traded companies that switched auditors between 1979 and the early part of 1983. Using the selection criteria noted in the previous paragraph, 101 companies were found in this step. Since name-brand differences are expected to be correlated with size, expertise, and independence, this selection process also is expected to result in a reasonably large amount of variance in these measures. Next, the remaining client companies in the 10th edition of Who Audits America (June 1983) were randomly sampled until 30 firms with no change in auditor name-brand were found that met the sample restrictions noted above. These 30 firms were added to assure that there is no bias in the test results due to truncation of the middle range of the distribution of changes in quality.'* Thus, the final sample consists of 131 auditor switches. For tests of the sub-periods before and after the switch, one of the firms has incomplete information; therefore, only 130 firms are used for these tests.'^ A breakdown of the
"In this case, the middle range of the distribution of changes in quality is the firms that have no change in quality. Theoretically, excluding the middle range of the distribution of the dependent variable can result in biased regression coefficients. Sensitivity analysis on the data used in this study, however, indicated that the results were qualitatively invariant to the inclusion or exclusion of these thirty firms. '^Of the 1,730 switches listed in Who Audits America (June 1983) that are excluded from the sample, 656 are excluded because they did not change quality with respect to the brand-name criterion; 718 are excluded because either (1) the old or new auditor was not disclosed in Who Audits America (1982), (2) the switching entities were not corporations (e.g., they were limited partnerships), (3) the switches were erroneously classified {i.e., they had not switched auditors), or (4) financial information was not available from Disclosure or Bechtel. The remaining 356 are excluded due to either (1) the firm had not been in existence for at least two years prior or subsequent to the switch, (2) more than one switch occurred during the periods examined, (3) the firm merged within two years of the swtich, (4) 20 percent or more of the firm was owned by another corporation, (5) the firm went bankrupt during the periods examined, or (6) the firm received an adverse or disclaimer audit opinion during the periods examined.

Auditing, Spring 1992

TABLE 2 Summary of Switch Types Big Eight/Second Tier/Local Classifications* +2

-H
3 8 6 5 0 0 22

0 2 4 8 9 6 1 30

-1 6 7 7 4 0 0 24

-2 3 2 6 8 1 0 20

Total 18 34 38 33 7 1 131

* where + 2 = from local firm to Big Eight firm +1 = from local firm to second tier firm, or from second tier firm to Big Eight firm 0 = no change in classification - 1 = from Big Eight firm to second tier firm, or from second tier firm to local firm - 2 = from Big Eight firm to local firm

sample by year using the name-brand classification is shown in table 2. Proxies and 10-Ks were obtained from the Disclosure, Inc. and Bechtel Information Services microfiche data bases. These data bases contain all publicly available information filed with the SEC during the required periods. Statistical Tests The three hypotheses are jointly tested by estimating the significance of the estimated coefficients in the following model: COMB = bo + b,MGT -I- bjLEV + bjACRL -I- b4GR0WTH + bjISSUE -I- e where COMB = the principal components linear combination of the four audit firm measures, MGT = the change in mangement ownership percentage, LEV = the change in ratio of longterm debt to assets. (2)

DeFond

25 Fjq = the factor score coefficient for observation i and audit quality variable q, Sjq = the standardized measure of the change in audit quality for observation i and audit quality variable q. Table 3 gives descriptive statistics of the variables included in the test of the hypotheses (referred to in the table as the "spanned period") and the variables included in the sensitivity analysis for the subperiods before and after the switch. Pairwise Pearson correlation coefficients were examined for the independent variables. While 16 of the 46 correlation coefficients are significant at the .05 level or less, the magnitudes of the coefficients are generally small (all are .31 or less). Thus, the degree of multicollinearity is expected to be relatively low. Regression Results Table 4 shows the results of the OLS regressions for the "spanned" test (of changes in the agency variables from two years before to two years after the auditor switch) and "sub-periods" test (of changes in the agency variables over the sub-periods before and after the switch). The spanned test supports hypothesis 1 (that changes in management ownership are inversely correlated with changes in audit quality) and hypothesis 2 (that changes in leverage are positively correlated with changes in audit quality) at p < .05. Hypothesis 3 (that changes in short-term accruals are positively correlated with changes in audit quality) is not supported, although the coefficient on this variable is significant in the opposite direction of that hypothesized, using a one-tail test. The negative sign indicates that a change to a higher (lower) quality auditor is associated with a decrease (increase) in shortterm accruals. There are several possible reasons for the results with respect to accruals. For example, if a propensity for higher accruals indicates an increase in the

ACRL = the change in the ratio of short-term accruals to total assets, GROWTH = the percentage change in total assets, and ISSUE = the ratio of the proceeds from securities issues during the year subsequent to the switch divided by the book value of assets at the beginning of that year. The hypotheses predict that the coefficient b| will be negative and that hj., hj,, h^, and b5 will be positive. RESULTS Descriptive Statistics Pairwise Pearson correlation coefficients were examined for the four audit quality measures and the coefficients ranged from .572 to .959. This high degree of correlation is not surprising given the construction of the variables.'* The factor score coefficients generated by the principal components procedure were .220 for auditor size, .289 for name-brand, .293 for expertise, and .295 for independence. The procedure produced only one significant factor, with an eigenvalue of 3.284, indicating that 82 percent of the variance is explained by the resulting factor. The distribution of the scores indicates that each variable is relatively important in contributing to the explanation of the overall variance. Notationally, the combined dependent variable is formed as follows: (3) where COMB I = the combined audit quality measure for observation i.

"While the principal components procedure requires correlation between the variables in order to extract a common factor, sensitivity analysis is performed in a later section to assure that the results are not "driven" by the inclusion of any one variable.

26 TABLE 3

Auditing, Spring 1992

Mean, Median and Standard Deviation of Independent Variables" Mean AGENCY VARIABLES MANAGEMENT OWNERSHIP MGTspanned period MGTbefore MGTafter LEVERAGE LEVspanned period LEVbefore LEVafter ACCRUALS ACRLspanned period ACRLbefore ACRLafter CONTROL VARIABLES GROWTHspanned period GROWTHbefore GROWTHafter ISSUE"spanned period ISSUEbefore & after
Med

Std Dev

-0.026 -0.024 -0.046 -0.015 -0.008 -0.007 -0.172 -0.017 0.001 0.347 0.258 0.171 0.104 0.105

-0.010 0.000 0.000 -0.000 -0.005 -0.002 -0.010 0.007 -0.003 0.384 0.207 0.105 0.005 0.005

0.129 0.110 0.097 0.185 0.140 0.167 0.244 0.209 0.201 0.549 0.403 0.442 0.291 0.292

"The sample size is 131 for the period spanning the switch and 130 for the before and after periods. 'ISSUE is always measured with respect to securities issued in the year after the switch. The difference in measures between the spanned period and the before and after periods is due to the reduced sample in the periods before and after the switch.

riskiness of the client, this association could result from differences in risk tolerances across auditors based upon quality." Specifically, the difference would indicate that lower quality auditors may be more risk tolerant and hence more willing to take on clients with increasing accruals. The control variables for growth and securities issues are positive, as predicted, and significant at p < .01 and p < .10, respectively. The overall model is significant at p < .01, and a test of the marginal explanatory power of the three agency variables over the control variables considered alone (not presented) indicates that they add significantly to the model at p < .05. '"Differences in risk tolerances across auditors are discussed in Simunic and Stein (1990).
The adjusted R-square for the model with just the

The results in the test of the sub-periods indicate that the change in management ownership is significant at p < .10 in the period after the switch, but not significant in the period before the switch. The change in leverage is significant at p < .10 in both the period before and after the switch. The change in accruals is not significant in the hypothesized direction in either sub-period, DISCUSSION AND ADDITIONAL ANALYSIS Audit Quality Measure principal components analysis is used in ^^^^ ^^^^y ^^ ^^^^j^ ^ ^^j^^le which is a linear combination of several audit quality measures. An analysis of the sensitivity of jj^^ ^g^^j^^ j ^ jj^j^ measure was performed
, . , n- ui

control variables is .131 compared to .191 for the entire model.

by regressmg the agency conflict variables on each of the individual measures used to

DeFond TABLE 4 OLS Results with Combined Audit Firm Measure as the Dependent Variable COMB = bo + b.MGT + bzLEV + bjACRL + b4GROWTH -I- bjISSUE + e Period Spanning the Auditor Change Coeff. AGENCY VARIABLES MANAGEMENT OWNERSHIP (-) MGTspanned period MGTbefore MGTafter LEVERAGE(+) LEVspanned period LEVbefore LEVafter ACCRUALS(+) ACRLspanned period ACRLbefore ACRLafter CONTROL VARIABLES GROWTH (+) GROWTHspanned period GROWTHbefore GROWTHafter ISSUE (+) Adjusted R^ F Stat.
* = p < .10 (one-tail) ** = p < .05 (one-tail) *** = p < .01 (one-tail)

27

Variables

Sub-Periods Before & After Coeff. T-Stat.

T-Stat.

-0.012

-1.830** 0.004 0.015 -0.515 -1.642*

0.843

1.885* 1.042 0.764 1.602* 1.476*

-0.634

-1.904 0.736 0.443 -1.736 -1.019

0.422

2.711***
0.426 0.246

1.955** 1.233 1.226

0.425

1.513* 0.191 7.134***

0.374

0.168 3.901***

form the combined measure {i.e., namebrand, size, expertise, and independence). The results (see tables 5 and 6) indicate that while the signs of the coefficients are the same as those in table 4, the strength of the significance of the coefficients varies with the choice of the dependent variable used in the regression. For the period spanning before and after the switch, three of the four models (size, independence, and namebrand) find both management ownership and leverage to be significant. Of the models that look at the sub-periods, only two models (independence and name-brand) indicate

significant effects for both management and leverage. This suggests that the associations detected in previous research are somewhat sensitive to the choice of the surrogate used to characterize the auditor's ability to alleviate agency conflicts.^'

^'This observation is partially supported by the results of Francis and Wilson (1988). That paper found a large disparity between their regressions that used size versus name-brand as the dependent variable. Their differences, however, may have been driven by differences in the samples used for each of the tests.

28 TABLE 5

Auditing, Spring 1992

OLS Results with Auditor Size and Independence as the Dependent Variable DEP = bo + b.MGT -I- bjLEV + bjACRL + b4GROWTH + bjISSUE + e
Variables Coeff.

Size
Coeff.

Independence Coeff. Coeff.

(t-stat.)
AGENCY VARIABLES MANAGEMENT OWNERSHIP MGTspanned period MGTbefore MGTafter LEVERAGE (-^) LEVspanned period LEVbefore LEVafter ACCRUALS (-I-) ACRLspanned period ACRLbefore ACRLafter CONTROL VARIABLES GROWTH (-I-) GROWTHspanned period GROWTHbefore GROWTHafter ISSUE (-^) Adjusted R^ Model F statistic * = p < . 10 (one-tail) ** = p < .05 (one-tail) *** = p < .01 (one-tail)

(t-stat.)

(t-stat.)

(t-stat.)

-0.008 (-1.620)* -0.003 (-0.467) -0.009 (-1.272) 0.646 (1.847)** 0.731 (1.443)* 0.590 (1.458)* -0.427 (-1.638) -0.507 (-1.532) -0.313 (-0.922)

-0.007 (-1.638)** -0.002 (-0.395) -0.008 (-1.310)* 0.521 (1.694)** 0.667 (1.497)* .434 (1.219) -0.445 (-1.942) -0.510 (-1.748) -0.348 (-1.164)

0.236 (1.934)** 0.292 (1.718)** 0.094 (0.604) 0.365 (1.548)* .125 3.046***

0.277 (2.581)*** 0.322 (2.146)** 0.109 (0.790) 0.290 (1.394)* .155 3.628***

0.390 (1.774)** .147 5.482***

0.316 (1.633)* .178 6.627***

It is observed, however, that the results sible implication is that the results using the using the name-brand surrogate presented in combined dependent variable are "driven" table 6 are quite similar to the results using by the presence of the name-brand variable. the combined measure in table 4. One pos- This possibility was explored by rerunning

DeFond

29

TABLE 6 Logit Results with Auditor Name-Brand and Expertise as the Dependent Variable DEP = bo + b,MGT + bzLEV + bjACRL + b4GROWTH + bjISSUE -t- e
Variables Name-Brand Coeff Coeff. Expertise Coeff. Coeff.

(t-stat.)
AGENCY VARIABLES MANAGEMENT OWNERSHIP MGTspanned period MGTbefore MGTafter LEVERAGE (-I-) LEVspanned period LEVbefore LEVafter ACCRUALS (-I-) ACRLspanned period ACRLbefore ACRLafter CONTROL VARIABLES GROWTH {+) GROWTHspanned period GROWTHbefore GROWTHafter ISSUE (-I-) Pseudo R^ Model Chi-square * = p < .10 (one-tail) ** = p < .05 (one-tail) *** = p < .01 (one-tail)

(t-stat.)

(t-stat.)

(t-stat.)

-0.024 (-1.826)** -0.010 (-0.661) -0.030 (-1.548)* 2.096 (2.223)** 2.544 (1.924)** 1.928 (1.824)** -1.386 (-1.911) -1.822 (-1.973) -0.979 (-1.111)

-0.018 (-1.283) -0.004 (-0.250) -0.039 (-1.933)** 0.985 (0.998) 0.950 (0.679) 1.088 (0.967) -1.288 (-1.730) -1.364 (-1.445) -1.168 (-1.244)

0.819 (2.572)*** 0.796 (1.757)** 0.535 (1.300)* 1.021 (1.257) .040 34.42

0.835 (2.394)*** 0.599 (1.280) 0.715 (1.636)* 0.111 (0.179) .008 20.13**

1.088 (1.387)* .057 33.47***

0.276 (0.472) .034 18.61***

the analysis using a principal components measure of the dependent variable that excluded the name-brand measure of quality. This yielded factor score coefficients of .31826 for expertise, .39528 for auditor size.

and .39448 for independence. The results of the regressions using this new combined measure of audit quality are qualitatively identical to the results in table 4 that used the combined measure that included the

30 name-brand surrogate. Thus, it does not appear that the name-brand measure was solely responsible for the results in table 4. It does, however, indicate that the name-brand measure is a good substitute for the more complex measure that combines several quality measures. Agency Cost Measures The primary analysis suggests that, to the extent that agency costs are captured by leverage and ownership, changes in agency costs are associated with auditor changes. One implication of this finding is that events are occurring that result in changes in leverage and ownership around the time of the auditor change. This was investigated by looking for evidence of activity that affected ownership and debt in the statement of changes in financial position in the year of the auditor change and the subsequent year. This examination revealed that there was a high frequency of both debt and stock issued in connection with activities such as bond conversions, new bank borrowings, the exercise of options and public offerings. Specifically, 72 percent of the sample firms that switched to an auditor with a higher name-brand reputation issued additional debt and 30 percent issued additional stock for these purposes. This compares to 43 percent and 20 percent, respectively, for the firms switching to a lower name-brand auditor. These findings indicate that events occurred around the time of the auditor changes that would plausibly trigger management to switch auditors. Measurement Period

Auditing, Spring 1992

The effects of using a spanned period versus sub-periods before and after the switch can be seen by comparing the two models in table 4. Significant associations occur both before and after the auditor switch, consistent with management both anticipating and reacting to agency conflict changes. In addition, tests (not presented here) indicate that the explanatory power of the agency variables is marginally significant at p < .05 over the control variables in the spanned model, but this is not the case in the subperiod model. This indicates that the results of studies attempting to link audit quality and agency conflicts are dependent upon the time period over which the agency variables are measured. CONCLUSION The results of this study lend support to the hypotheses that changes in management ownership and leverage are associated with changes in the auditor's ability to alleviate agency conflicts, independently of changes in firm growth and securities issues. The results also suggest that the association between audit quality and agency conflicts is sensitive to both the time period over which changes in agency conflicts are measured (before and after auditor switching) and the choice of the measure used to proxy for audit quality. Finally, use of a more complex measure that combined multiple measures of audit quality yielded similar results to that obtained by the name-brand proxy.

REFERENCES
Carpenter, C. G., and R. Strawser. 1971. Displacetnent of auditors when clients go public. Journal of Accountancy (June): 5558. Cooley, W., and Lohnes. 1971. Multivariate Data Analysis. John Wiley and Sons, Inc. DeAngelo, L. 1981a. Atiditor independence, 'low balling' and disclosure regulation. Journal of Accounting and Economics (August): 113-127. . 1981b. Auditor size and audit quality. Journal of Accounting and Economics (Decetnber): 183-199. Dopuch, N., and D. Simunic. 1982. The competition in auditing: an assessment. Fourth Symposium on Auditing Research, 401-450. Urbana: University of Illinois.

DeFond Eichenseher, J. W., and D. Shields. 1989. Corporate capital structure and auditor 'fit'. Advances in Accounting (Supplement 1): 39-56. , and . 1983. The correlates of CPA-firm change for publicly-held corporations. Auditing: A Journal of Practice & Theory, 39-56. Francis, J., and E. Wilson. 1988. Auditor changes: a test of theories relating to agency costs and auditor differentiation. The Accounting Review (October): 663-682. Healy, P. 1985. The impact of bonus schemes on the selection of accounting principles. Journal of Accounting and Economics (April): 85-107. , and T. Lys. 1986. Auditor changes following Big Eight takeovers of non-Big Eight audit firms. Journal of Accounting and Public Policy (Winter): 251-265. Jensen, M., and W. Meckling. 1976. Theory of the firm; managerial behavior, agency costs and ownership structure. Journal of Financial Economics (October): 305 360. Jensen, R. E. 1982. The competition in auditing: An assessmenta discussion. Fourth Symposium on Auditing Research, 452-466. Urbana: University of Illinois. Johnson, W. B., and T. Lys. 1990. The market for audit services: evidence from voluntary auditor changes. Journal of Accounting and Economics (January): 2 8 1 308. Leddy, T. 1982. The competition in auditing: An assessmenta discussion. Fourth Symposium on Auditing Research, 467-482. Urbana: University of Illinois. Mautz, R. K., and H. Sharaf. 1961. The Philosophy of Auditing. American Accounting Association, Sarasota. Palmrose, Z. 1984. The demand for quality-differentiated audit services in an agency setting: an empirical investigation. Sixth Illinois Auditing Research Symposium. Schwartz, K., and K. Menon. 1985. Auditor switches by failing firms. The Accounting Review (April): 248-261. Shockley, A., and R. N. Holt. 1983. A behavioral investigation of supplier differentiation in the market for audit services. Journal of Accounting Research (Autumn): 545-564. Simunic, D., and M. Stein. 1990. Audit risk in a client portfolio context. Contemporary Accounting Research (Spring): 329343. , and . 1987. Product differentiation in auditing: auditor choice in the market for unseasoned new issues. Canadian Certified General Accountants' Research Foundation, Vancouver, Canada. Smith, D. B. 1986. Auditor 'subject to' opinions, disclaimers, and auditor changes. Auditing: A Journal of Practice & Theory (Fall): 95-108. , and D. R. Nichols. 1982. A market test of investor reaction to disagreements. Journal of Accounting and Economics (October): 109120. Titman, S., and B. Trueman. 1986. Information quality and the valuation of new issues. Journal of Accounting and Economics (June): 159172. Watts, R., and J. Zimmerman. 1980. The markets for independence and independent auditors. Unpublished manuscript. University of Rochester, Rochester, N.Y. Who Audits America. 1979 through 1983. Spencer Phelps Harris, Editor. Who Audits the World. 1983. V. B. Bavishi and H. E. Wyman, University of Connecticut.

31