Gaver and Utke (2019)

THE ACCOUNTING REVIEW American Accounting Association
Vol. 94, No. 3 DOI: 10.2308/accr-52206

May 2019
pp. 113–147
Audit Quality and Specialist Tenure

Jennifer J. Gaver
The University of Georgia
Steven Utke
University of Connecticut
ABSTRACT: We argue that the association between auditor industry specialization and audit quality depends on
how long the auditor has been a specialist. We measure audit quality using absolute discretionary accruals, income-
increasing discretionary accruals, and book-tax differences. Our results, based on a sample of Big 4 audit clients
from 2003–2015, indicate that auditors who have only recently gained the specialist designation produce a level of
audit quality that does not surpass that produced by non-specialist auditors, and is generally lower than the audit
quality produced by seasoned specialists. We estimate that the seasoning process takes two to three years. In
contrast to prior research that finds no effect of specialization after propensity score matching, we find that seasoned
specialists generally produce higher-quality audits than other auditors even after matching. This suggests that the
audit quality effect associated with seasoned industry specialist auditors is not due to differences in client
characteristics.
JEL Classifications: M42.
Data Availability: Data used in this study are available from public sources identified in the text.
Keywords: audit quality; industry specialization; auditor tenure; discretionary accruals; book-tax differences;
auditor changes; propensity score matching.
I. INTRODUCTION
T
he association between auditor industry expertise and audit quality has attracted considerable interest from accounting
researchers. Results from experimental studies suggest that industry expertise enhances auditor judgments regarding
error detection (Solomon, Shields, and Whittington 1999; Owhoso, Messier, and Lynch 2002), risk assessments, and
audit planning (Taylor 2000; Low 2004; Hammersley 2006). Results from archival studies also suggest that industry-specialist
auditors provide higher-quality audits to their clients.1 Compared to clients of non-specialist auditors, clients of specialists have
lower discretionary accruals and higher earnings response coefficients (ERCs) (Balsam, Krishnan, and Yang 2003; Krishnan
2003).
A limitation of archival research on auditor industry specialization is that specialization is not directly observable. Archival
researchers typically use an auditor’s within-industry market share as an indirect measure of specialization, in contrast to the
experience-based measures of industry expertise used in experimental studies. Recent archival work challenges the conclusion
that auditor industry expertise, measured by market share, is associated with audit quality. In particular, Minutti-Meza (2013)
(hereafter, MM) finds no evidence that auditor industry expertise improves audit quality after using propensity score matching
(PSM) to address functional form misspecification arising from differences between characteristics of specialist versus non-
This paper has benefited from helpful comments from Ashley Austin, Ben Ayers, Denny Beresford, Bill Buslepp (discussant), Nathan Goldman, Matt
Hart, Youree Kim, workshop participants at the University of Connecticut, the 2014 AAA Southeast Regional Meeting, and the 2014 AAA Annual
Meeting, and especially Mike Ricci and Quinn Swanquist. We thank Mark L. DeFond (editor) and two anonymous reviewers for their careful and
constructive guidance during the review process. We gratefully acknowledge the support of The University of Georgia’s Terry College of Business and J.
M. Tull School of Accounting. Steven Utke also thanks the University of Connecticut School of Business for financial support.
Editor’s note: Accepted by Mark L. DeFond.
Submitted: August 2015
Accepted: July 2018
Published Online: July 2018
1
Throughout the paper, we refer to audit firms as ‘‘audit firms’’ or ‘‘auditors’’ and clients as ‘‘clients’’ or ‘‘firms.’’
113
114 Gaver and Utke
specialist auditors’ clients. However, DeFond, Erkens, and Zhang (2017a) and Shipman, Swanquist, and Whited (2017) show
that subjective design choices underlying PSM affect the composition of the matched sample and can alter inferences from a
PSM analysis.
In contrast to this recent literature focusing on econometric issues in the relation between auditor characteristics and audit
quality, our paper revisits the relation between audit quality and auditor industry specialization using a more nuanced measure
of specialization than market share alone. Existing market share-based industry specialization measures implicitly assume that
upon achievement of a market share threshold, the auditor immediately implements industry best practices on an audit firm-
wide level. In contrast, we argue that a coordinated approach to industry-wide clients develops over time (Gendron, Cooper,
and Townley 2007) and propose incorporating specialist tenure into existing industry specialization measures. Specialist tenure
refers to an auditor’s experience (length of service) as the market share leader in a given industry. This differs from auditor
tenure, which focuses on an auditor’s experience on a given client. We refer to longer (shorter) tenured specialists as
‘‘seasoned’’ (‘‘unseasoned’’).
The consideration of specialist tenure is important if it is possible for an audit firm to become a market share-based
specialist due to exogenous events unrelated to auditor expertise. For example, mergers and acquisitions by clients, industry
entry by clients, and client growth could make an audit firm the industry market share leader without any action by the audit
firm itself. If it then takes time to hire and develop personnel and implement the audit firm-wide technologies needed to meet
the increased demand for high-quality audits across the auditor’s larger market share in that industry, the newly created
specialist will not immediately function as an industry expert (see Bills, Swanquist, and Whited 2016). In this scenario, newly
created (unseasoned) specialists will function more like non-specialist auditors, producing audits of lower quality compared to
auditors that have had sufficient time to make the necessary investments in personnel and technology to function as experts
(seasoned specialists). That is, by incorporating specialist tenure, we account for the possibility that an audit firm becomes an
industry market share leader following a rapid market share increase unrelated to auditor expertise. On the other hand, if
specialists arise because clients are attracted to the industry expertise previously developed by the audit firm, then specialist
tenure is irrelevant. It is possible that there are some cases where expertise drives the auditor’s market share and others where
market share is a result of exogenous events unrelated to the auditor. Ignoring the latter scenario and viewing all specialists
identically blurs the distinction between specialist and non-specialist auditors and is a potential explanation for the insignificant
results reported by MM.
We begin our analysis by replicating the results of MM. We measure audit quality using absolute discretionary accruals
(ADA), income-increasing discretionary accruals (DA), and book-tax differences (BTD) in our main analyses.2 Our samples,
spanning 2003–2015, consist of up to 25,901 firm-years when measuring audit quality using ADA, 13,863 firm-years when
measuring audit quality using income-increasing DA, and 10,551 firm-years when measuring audit quality using BTD. In all
cases, the firm has a Big 4 auditor in both the current and the preceding year.
We observe a positive association between a market share-based measure of auditor specialization and audit quality in the
full (unmatched) ADA and income-increasing DA samples, but not in the BTD sample. Thus, even without matching, an
industry specialization proxy based only on market share produces somewhat mixed results. Following MM, we then
reexamine the relation between auditor specialization and audit quality after employing PSM. The general pattern of results
(significant in the ADA and income-increasing DA samples, insignificant in the BTD sample) remains after matching, although
the statistical significance is somewhat weaker.
DeFond et al. (2017a) and Shipman et al. (2017) show that PSM results are sensitive to design choices, so we examine the
robustness of our findings across nine separate PSM specifications. In contrast to MM, we find a significant relation between
market share-based industry specialization and audit quality in 44 (89) percent of our PSM samples when measuring audit
quality using either ADA or BTD (using income-increasing DA). The impact of PSM design choices on the relation between
auditor characteristics and audit quality is not the focus of our study (DeFond et al. [2017a] thoroughly address this issue).
However, the results suggest that the relation between market share-based industry specialization and audit quality in the
unmatched sample should not be dismissed as an artifact of differences between specialist and non-specialist auditors’ clients,
as argued by MM.
Our main innovation is to partition industry specialists into experienced (seasoned) and newly created (unseasoned)
specialists, resulting in an industry specialization measure that reflects both market share and seasoning. In order to create our
partition, we must first identify unseasoned specialists. This requires us to investigate the events preceding the creation of
industry specialists, which is critical to our argument that market share dominance can precede the development of industry
expertise. Consistent with specialist tenure likely playing an important role in auditor industry specialization, in almost all
2
We also perform robustness tests that measure audit quality using dichotomous measures (meeting or beating analysts’ forecasts, receiving a going
concern opinion) and the earnings response coefficient.
The Accounting Review

Volume 94, Number 3, 2019
Audit Quality and Specialist Tenure 115
cases, we find that changes in the industry market leader are due to factors external to the auditor, such as industry entry by
clients and client growth, rather than auditor changes. There is little evidence that auditors attain market dominance by
attracting new clients due to their pre-existing industry expertise.
We next reestimate the MM model after partitioning specialists into seasoned and unseasoned categories. We find that
compared to unseasoned specialists, seasoned specialists generally provide higher audit quality. This result holds across our
main audit quality measures (ADA, income-increasing DA, and BTD) and nearly all matching strategies, including strategies
other than PSM. Further, seasoned specialists provide higher audit quality than non-specialists, again across our main audit
quality measures and nearly all matching strategies.3 In terms of economic significance, having a seasoned specialist, rather
than a non-specialist, auditor improves audit quality by about 6 (42) percent using the ADA and income-increasing DA (BTD)
measures. For all three audit quality measures and all matching strategies, the quality of audits produced by unseasoned
specialists is either not statistically different from, or worse than, the quality of audits produced by non-specialist auditors.
Because we find that specialists are generally created by factors external to the auditor, this suggests that auditors likely fall
backward into the industry leadership position and then develop the tools necessary to provide higher-quality audits. Thus,
expertise follows market share dominance, rather than the other way around.
We also investigate the length of time it takes for an auditor to become seasoned. While recent work investigates an auditor’s
learning curve on a specific client (Cameran, Francis, Marra, and Pettinicchio 2015; Cassell, Hansen, Myers, and Seidel 2017),
little evidence exists on the speed with which industry knowledge is created within an audit firm. Our findings suggest that, on
average, seasoning develops during the first two to three years that an auditor is the market share leader, depending on the audit
quality measure. Auditors in their second or third consecutive year as industry leader often produce audit quality that is statistically
indistinguishable from that of seasoned specialists, while auditors in their first one or two years as market share leader often
produce audits of lower quality than seasoned specialists. Thus, the seasoning process for industry specialists appears to take
longer than the relatively quick learning process on a specific client (Cassell et al. 2017). An implication is that auditors in their
initial years as market share leaders have been misclassified as specialists in prior research, blurring the relation between industry
expertise and audit quality. We also find some evidence that audit quality is increasing in a continuous measure of specialist
tenure. Thus, we encourage researchers to consider seasoning when creating specialization measures.
We contribute to the literature on auditor industry expertise in several ways. First, we find that dominant market share is
not a sufficient condition for industry expertise. Researchers must also consider the specialist’s tenure. Our results are also
potentially relevant for researchers outside of auditing who acknowledge the limitations of market share-based measures of
expertise (e.g., Bao and Edmans 2011). Second, we question recent research suggesting that the relation between industry
specialization and audit quality is due to differences in client characteristics. Our analysis suggests that these prior results may
be sensitive to the specific matching technique used, echoing recent studies that suggest that PSM be used with caution
(DeFond, Erkens, and Zhang 2017b; Shipman et al. 2017) or not at all (King and Nielsen 2016).4 Although unseasoned
specialists constitute only about 2 percent of total observations (11 percent of specialist observations), excluding them makes
the effect of specialists on audit quality generally robust to matching. Third, we provide evidence on the process by which
industry expertise is created in an audit firm. We find that it takes two to three years for an auditor to function as an industry
expert, in contrast to the relatively quick learning process for an auditor on an individual client. We also find that auditors often
become industry market share leaders for reasons outside of their control and then invest in industry expertise. This challenges a
common assumption in the literature that clients self-select into industry-specialist auditors (Gul, Fung, and Jaggi 2009).
Finally, we note that, if implemented, mandatory auditor rotation is likely to create unseasoned specialists that may need a
seasoning period to function as true industry experts.
II. HYPOTHESES
We contend that an auditor’s within-industry market share is an incomplete specialization measure because it ignores the
importance of seasoning in developing expertise. However, seasoning alone does not make an industry expert (Ericsson,
Charness, Feltovich, and Hoffman 2006). A critical mass of business, proxied for by market share, must also be present to
induce an auditor to invest in industry-specific personnel and technologies. Thus, we argue that industry expertise is a
multidimensional construct that requires both market share and seasoning.5
3
However, for reasons discussed in Section V, our results vary somewhat with alternative seasoning periods (two or three years, versus the one-year
results discussed here), and are generally weaker when using alternative audit quality proxies (meeting or beating analysts’ forecasts, receiving a going
concern opinion, or the earnings response coefficient) and specialization measures.
4
PSM can also be problematic when partitioning a sample into multiple groups, as required for our study. We discuss and address this issue in Section V.
5
While we are not the first to question the validity of market share-based specialization measures (see Audousset-Coulier, Jeny, and Jiang [2016] for a
review), we are the first to propose within-industry seasoning as a potential improvement to existing industry specialization measures.

116 Gaver and Utke
Gendron et al. (2007) describe the development of audit expertise as an iterative process where good practices are developed,
documented, and subsequently validated over time through the experiences of audit firm personnel.6 An audit firm may decide to
implement these practices on an audit firm-wide level by investing in various decision aids (such as checklists), training sessions
(Bédard 1989; Power 1996), and guidance from a centralized national office (Danos, Eichenseher, and Holt 1989).7 This type of
coordinated approach is more likely when the audit firm has a sufficient volume of clients that could benefit from similar practices,
such as firms in the same industry. Thus, although audit firm personnel develop experience auditing specific clients, leveraging
that expertise throughout the audit firm requires a commitment of resources that is unlikely to be made unless the audit firm has a
large enough market share to justify investing in personnel and technologies specific to an industry.
An important point is that dominant market share in an industry can precede industry investment, as illustrated next. In the
following example, based on an actual specialization change during our sample period, we assume that the home furnishing and
equipment retail industry consists of three relatively large firms and two small firms. Best Buy (BBY) represents 45 percent of
the market, Circuit City (CC) represents 20 percent of the market, Linens-N-Things (LIN) represents 15 percent of the market,
and two small firms (Other1 and Other2) represent the remaining 20 percent of the market.8 In Year 1, Ernst & Young (EY)
audits BBY, KPMG audits CC and LIN, Deloitte audits Other1, and PricewaterhouseCoopers (PwC) audits Other2. In Year 2,
BBY changes auditors from EY to Deloitte, and EY becomes the auditor of LIN. Deloitte is now the market share-based
specialist not only on the audit of BBY, but also on the audit of Other1. We argue that prior to becoming the market share
leader, it is unlikely that Deloitte invested in the industry heavily enough to immediately behave as an industry expert upon
gaining the leadership position. After obtaining the leadership position, Deloitte has the incentive to develop and implement
audit firm-wide industry guidance, which then benefits its audit clients throughout the industry.
Firm (Market Share) Year 1 Year 2
BBY (45%) EY (specialist) Deloitte (specialist)
CC (20%) KPMG KPMG
LIN (15%) KPMG EY
Other1 (10%) Deloitte Deloitte (specialist)
Other2 (10%) PwC PwC
In this example, the industry specialist changed because BBY, the largest firm in the industry, changed auditors. A change
could also result from rapid growth of clients relative to their industry competitors (e.g., BBY and CC could flip positions in the
market), client mergers (e.g., CC could acquire LIN and Others), new firms entering the market, or existing firms leaving the
market.9 Interestingly, in the real-world case, BBY was required to change auditors due to a conflict of interest with EY (Taub
2005). BBY did not self-select to Deloitte, one of the smallest Big 4 auditors in the industry, because of Deloitte’s industry
expertise; if industry expertise drove the auditor selection, then BBY would have selected the auditor with the second-largest
pre-change industry market share, KPMG.
As illustrated in this example, we argue that if the achievement of a dominant market share is the result of exogenous
events, then a dominant market share may not indicate an industry expert; the auditor also needs time to establish audit firm-
wide technologies to support the higher volume of business in the industry. Investments made to support the higher volume of
business will, over time, result in higher-quality audits.10 We, therefore, predict that audits performed by seasoned specialists
will be of higher quality than audits performed by either unseasoned specialists or non-specialist auditors. Our hypotheses, in
alternative form, are as follows:
H1: The quality of audits produced by seasoned industry-specialist auditors is higher than the quality of audits produced
by unseasoned industry-specialist auditors.
6
While we focus on auditors, researchers in industrial organization also examine the development and transfer of knowledge within firms (Tsai 2001;
van Wijk, Jansen, and Lyles 2008; Phelps, Heidl, and Wadhwa 2012).
7
A Big 4 senior manager confirmed that auditors distribute industry knowledge using these methods, plus email alerts.
8
This example is simplified for expositional purposes. Having a small number of firms in an industry is not a necessary condition for the creation of
unseasoned specialists. Over our sample period, unseasoned specialists serve as auditors on clients in diverse industries, some comprised of only a few
firms and others having over 450 firms.
9
In the real-world case, BBY changed auditors and LIN exited the industry because it was taken private. In Section III, we examine the causes of the
specialist changes in our sample. Most are due to exogenous events unrelated to the auditor. We argue that new specialists created by exogenous events
may not immediately behave as industry experts, but instead will develop expertise over time. We acknowledge that some exogenous events might be
more likely to create incentives for the auditor to invest in industry expertise than others. For example, changes that require an auditor to increase the scope
of its industry operations to serve new or larger clients likely provide greater incentives for investment than market share changes due solely to client
divestitures or industry exits. However, parsing these differences is beyond the scope of our study and is a potentially fruitful area for future research.
10
We acknowledge that our example does not account for small firms (that are unable to shift the industry-specialist designation themselves) selecting
into specialist auditors after a large industry change creates a new specialist. To the extent that this takes time to occur, it is consistent with our
hypotheses suggesting that specialization develops over time.

H2: The quality of audits produced by seasoned industry-specialist auditors is higher than the quality of audits produced
by non-specialist auditors.
H3: The quality of audits produced by unseasoned industry-specialist auditors does not differ from the quality of audits
produced by non-specialist auditors.
Several factors work against finding these results. Our distinction between seasoned and unseasoned specialists assumes
that unseasoned specialists have not had sufficient time to make industry-specific investments that will result in higher-quality
audits for all industry clients. However, individual audit personnel within the firm may have ample experience conducting
audits of other clients in the industry, or the auditor could hire new personnel with industry experience. If these experiences can
be quickly leveraged to produce quality audits on a larger scale, then we are less likely to find a significant difference in audit
quality between seasoned and unseasoned specialists (H1).11 Likewise, non-specialist audit firms may choose to institutionalize
best practices from an industry even if they are not the market share leader in that industry, reducing the likelihood that H2 is
supported. Further, if an auditor achieves a dominant market share by gradually developing expertise and increasing its
presence in an industry, then the auditor may immediately behave as a specialist, reducing the likelihood that we find support
for H1 or H3. Finally, our sample includes only Big 4 audit firms. If Big 4 auditors are experts in all industries (as is often
claimed in their marketing materials), then we will find no differences in audit quality among Big 4 auditors using any measure
of industry specialization, including measures that distinguish unseasoned from seasoned specialists.12
Prior studies (Geiger and Raghunandan 2002; Johnson, Khurana, and Reynolds 2002; J. Myers, L. Myers, and Omer 2003;
Carcello and Nagy 2004) find that audit quality is lower when auditors are new to their clients. Therefore, an alternative
explanation for a lower level of audit quality produced by unseasoned specialists is that, compared to seasoned specialists,
unseasoned specialists are simply less familiar with their clients. However, many newly designated (unseasoned) specialists are
not new to their clients.13 In our example, Deloitte, an unseasoned specialist, is new to BBY, but not to Other1. As discussed in
the next section, we include an auditor tenure variable in our model to control for its association with audit quality.
III. RESEARCH DESIGN
Regression Model and Variables

Our hypotheses relate audit quality to auditor industry expertise. As recommended by DeFond and Zhang (2014), we use
multiple measures of audit quality in our tests: discretionary accruals (absolute and income-increasing) and book-tax
differences. We focus on continuous measures that are available for a large number of firms because our hypotheses posit a
nuanced relation between auditor expertise and audit quality, and our sample consists of firms that, as Big 4 audit clients,
generally have a high level of audit quality. Alternative measures, such as restatements and going concern opinions, are likely
to be relatively rare among these firms, resulting in low-power tests (DeFond and Zhang 2014).14 However, in supplemental
tests discussed in Section V, we also examine earnings response coefficients (ERC), meeting or beating analysts’ earnings
forecasts, and going concern opinions.
Discretionary accruals are a common audit quality proxy, and book-tax differences are incrementally useful to accruals in
identifying earnings management, fraud, and restatements (Phillips, Pincus, and Rego 2003; Ettredge, Sun, Lee, and
Anandarajan 2008; Badertscher, Phillips, Pincus, and Rego 2009; Chi, Pincus, and Teoh 2014). We estimate discretionary
accruals using the performance-adjusted Jones model (Kothari, Leone, and Wasley 2005), and measure book-tax differences as
11
A newly created specialist faces an increased scale of production and may not have enough personnel with relevant industry expertise to serve both the
new (or larger) clients and the existing clients in the industry. In this case, professionals from other industries will be assigned to the audits, decreasing
the average level of industry expertise across all of the auditor’s clients in the industry (similar to the findings of Bills et al. 2016).
12
See, for example, Ernst & Young’s list of Global Industry Centers, in which they have ‘‘invested . . . for sharing industry-focused knowledge and
experience’’ (available at: https://www.ey.com/us/en/industries). This list covers nearly all industries.
13
While auditor tenure relates to repetition on a particular client, specialist tenure (seasoning) relates to repetition as the leader in a particular industry. If
adding one, or a few, large client(s) causes an auditor to become a specialist, then that auditor is unseasoned not only for the new client(s), but also for
all of its continuing (longer-tenured) clients in the industry. It is, therefore, possible to be an unseasoned specialist on a client with which the auditor has
a long-term relationship. In our ADA sample (described in Section IV), unseasoned specialists are more likely than seasoned specialists to be in their
first year with a client (3.7 percent of unseasoned observations involve new clients versus 1.4 percent of seasoned observations). Both seasoned and
unseasoned auditors have relatively long auditor-client relationships, with average auditor tenures of 14.21 and 14.51 years, respectively. In untabulated
tests, discussed in Section V, we find that our results are robust to examining seasoned and unseasoned specialists excluding auditor changes, ruling out
the possibility that auditors that are new to their clients drive our results.
14
Consistent with most prior research, we define audit quality based on audit outputs, reasoning that audit failure occurs when financial statements are
misstated (DeFond and Zhang 2014). The Public Company Accounting Oversight Board (PCAOB) and some recent research (Bell, Causholli, and
Knechel 2015) define audit quality using audit inputs. However, audit inputs are generally unobservable and are not necessarily correlated with the
quality of audit outputs (Bell et al. 2015, footnote 4).

118 Gaver and Utke
grossed-up deferred tax expense scaled by pre-tax income (Chi et al. 2014). Following prior research, we assume that audit
quality is negatively associated with the absolute value of firms’ discretionary accruals (ADA), income-increasing discretionary
accruals (DA), and book-tax differences (BTD).15 Appendix A provides variable definitions.
We employ a model adapted from MM that regresses audit quality (AQ, measured as ADA, income-increasing DA, or BTD)
on a dichotomous market share-based measure of auditor industry specialization, control variables, and year fixed effects.16
That is, before examining the effect of seasoning, we examine the effect of industry specialization using an existing market
share-based specialization proxy. We classify an auditor as a specialist (SPECIALIST) when its U.S. national-level market share
is the highest in a given industry-year and also more than 10 percent higher than the next-largest competitor, with industry
based on two-digit SIC codes (Reichelt and Wang 2010).17 This model is presented in Equation (1).
AQit ¼ a0 þ b1 SPECIALISTit þ b2 LOGMKTit þ b3 LEVit þ b4 ROALit þ b5 ROAit þ b6 LOSSit þ b7 CFOit þ b8 BTMit

þ b9 ABS AC LAGit þ b10 GROWTHit þ b11 ALTMANit þ b12 STDEARNit þ b13 TENUREit þ bt YEAR FEt þ mit
ð1Þ
If market share-based industry-specialist auditors provide higher audit quality than non-specialist auditors, then b1 will be
negative.
In Equation (1), auditors are classified as industry experts based only on market share. However, we argue that a critical
mass of business in an industry is a necessary, but not sufficient, condition to establish auditor expertise and that ‘‘time in
grade’’ as the industry leader is also needed. To test this assertion, we replace SPECIALIST in Equation (1) with two variables:
SEASONED and UNSEASONED. UNSEASONED takes on the value of 1 when the auditor is in its first year of being classified
as a specialist in a given industry, and is 0 otherwise.18 SEASONED takes on the value of 1 when the auditor is classified as a
specialist, and UNSEASONED is coded 0. A 0 coding on both UNSEASONED and SEASONED indicates that the auditor is not
an industry specialist. The revised model is given in Equation (2).
AQit ¼ a0 þ b1 SEASONEDit þ b2 UNSEASONEDit þ b3 LOGMKTit þ b4 LEVit þ b5 ROALit þ b6 ROAit þ b7 LOSSit

þ b8 CFOit þ b9 BTMit þ b10 ABS AC LAGit þ b11 GROWTHit þ b12 ALTMANit þ b13 STDEARNit þ b14 TENUREit
þ bt YEAR FEt þ mit
ð2Þ
If seasoned specialists outperform non-specialists (H2), then the coefficient on SEASONED (b1) in Equation (2) will be
negative. If seasoned specialists outperform unseasoned specialists (H1), then the coefficient on UNSEASONED (b2) will
exceed the coefficient on SEASONED (b1) in Equation (2). Finally, if the quality of audits produced by unseasoned industry
specialists does not differ from the quality of audits produced by non-specialist auditors (H3), then the coefficient on
UNSEASONED (b2) in Equation (2) will be statistically indistinguishable from zero.19
15
We focus on income-increasing, as opposed to income-decreasing or signed discretionary accruals, consistent with prior research (Becker, DeFond,
Jiambalvo, and Subramanyam 1998).
16
The model includes controls for the size (LOGMKT), profitability (ROA, ROAL, CFO, and LOSS), risk (LEV, ALTMAN, and STDEARN), and
growth profile (BTM and GROWTH) of the client and the length of the client’s relationship with the auditor (TENURE). The absolute value of
lagged accruals (ABS_AC_LAG) controls for scale effects in the audit quality variables that are not addressed by other variables in the model. We
do not include MM’s BIG4 control variable because all of our sample firms are Big 4 clients. All other control variables correspond to MM’s
model. Variable definitions are provided in Appendix A. Consistent with MM and Reichelt and Wang (2010), we do not include industry fixed
effects in the accruals models because accruals are estimated by industry. We include industry fixed effects (two-digit SIC) in the book-tax
differences model.
17
Palmrose (1986) and Neal and Riley (2004) define the minimum market share for specialization as 1.2 times the inverse of the number of Big N
auditors. In our study, this is 30 percent (1.2/4 ¼ 0.30). In order to clearly distinguish between firms audited by specialist and non-specialist auditors, we
exclude observations if their auditor is classified as a specialist under the 30 percent rule, but not under our main specialist definition (industry market
share leader by at least 10 percent). Except for the BTD tests, our inferences are unchanged if we include these firm-years in our analysis. Regarding the
BTD tests, analysis suggests that specialists measured under the alternative definition are particularly good at constraining BTD. As such, including the
30 percent specialists as non-specialists in the BTD robustness test adds measurement error to the non-specialists category, as these auditors behave as
true specialists.
18
Although we initially define unseasoned specialists as auditors in their first year as market share leader, the seasoning process could take more than one
year. In Section V, we investigate the length of the seasoning period.
19
In general, prior research does not differentiate audit quality within specialist auditors. An exception is Cahan, Jeter, and Naiker (2011), who find that
clients of specialist auditors with a small number of large clients have lower discretionary accruals compared to clients of specialist auditors with a large
number of small clients. However, as discussed in more detail below, client size is correlated with discretionary accruals and is, therefore, a
confounding factor relative to the auditor specialist-audit quality relation. Our paper differentiates specialists using an attribute (seasoning) other than
client size. We also employ multiple matching techniques to control for functional form misspecification in the relation between auditor specialization
and audit quality.

TABLE 1
Reasons for Industry Specialist Auditor Changes, By Year
Panel A: Reason for Changes by Year

Number of
Industries Auditor Auditor Merger
with Change to Change to by Unseasoned
Unseasoned Unseasoned Unseasoned Auditor Divestiture Industry
Year Auditors Auditor—Major Auditor—Minor Client by Client Entry
2003 8 1 0 1 0 1
2004 9 0 0 0 1 3
2005 8 3 0 1 0 1
2006 9 2 0 0 0 1
2007 4 0 1 2 0 0
2008 1 0 0 0 0 0
2009 6 1 0 1 0 3
2010 5 0 0 1 0 0
2011 7 1 1 0 1 2
2012 8 1 0 0 1 1
2013 3 0 1 1 1 0
2014 7 3 0 0 0 1
2015 1 0 0 0 0 0
2003–2015 76 12 3 7 4 13
Frequency 11.01% 2.75% 6.42% 3.67% 11.93%
(continued on next page)
Reasons for Specialist Changes

Unseasoned specialists are created when auditors’ within-industry market shares change. Our study of specialist tenure,
which requires identifying unseasoned specialists, presents an opportunity to examine the events preceding market share
changes. To our knowledge, ours is the first study to investigate and document this phenomenon. Our analysis encompasses all
specialist changes occurring over our sample period (2003–2015). We first calculate U.S. national-level industry specialization
using firms with the minimum required data in Compustat: U.S. headquarters location, sales, SIC, and auditor. We then identify
76 industry-years with an unseasoned specialist. For each of these industries, we obtain the industry and auditor composition in
the year of the specialist change, as well as the prior year, and then assess the reasons for the change. Based on our review, we
identify nine main events associated with specialist changes. Table 1 presents our results, broken out by year.
In order for there to be a relationship between specialist tenure and audit quality, it must be possible for market share
dominance to precede the development of industry expertise. This occurs when a new specialist arises from industry events
unrelated to the auditor. We find that in most cases (64 out of 76 industry-years; see Table 1), the unseasoned specialist auditor
became the market leader due to exogenous events, such as client growth, industry entry, and mergers. The three most common
drivers of specialist changes are client entry to or exit from the industry and client sales growth. About one-third of specialist
changes are caused by multiple factors. Cases where auditors became specialists solely because clients switched to the new
specialist are relatively rare (eight out of 76 industry-years). In four additional cases, an auditor change large enough to affect
the specialist designation occurred, along with other industry events affecting the specialist designation, most commonly, client
sales growth and industry entry. Finally, in three cases, the new specialist was created as a result of a minor auditor change, too
small to change the specialist on its own, occurring in combination with some other event. Even in these cases, we found no
clear evidence that clients changed auditors because the new auditor was the industry specialist. In fact, in industries where a
new specialist was created, we found nine cases of large clients switching to a non-specialist auditor, suggesting that clients
switching auditors do not herd toward the new specialist.20 Thus, we find little evidence that audit firms first develop industry
20
One reason that firms may be reluctant to herd to the industry market leader is fear that proprietary information could leak to a competitor via a common
auditor (e.g., Coke and Pepsi never share an auditor). However, very few firms disclose a reason for changing auditors. When an explanation for an
auditor change is offered by our sample firms, it is most often a simple statement that the audit is periodically put out to bid as a good governance
practice.

120 Gaver and Utke
TABLE 1 (continued)
Panel B: Reason for Changes by Year (continued)
Major Percent
Auditor of
Change Unrelated Industries Industries
Former Client to with with
Industry Client Sales Unseasoned Multiple Multiple
Year Exit Acquired Growth Auditor Causes Causes
2003 4 0 3 1 3 37.50%
2004 3 1 3 1 2 22.22%
2005 1 1 3 0 2 25.00%
2006 3 1 5 0 3 33.33%
2007 0 1 2 1 2 50.00%
2008 1 0 0 0 0 0.00%
2009 0 0 3 1 2 33.33%
2010 2 0 3 0 1 20.00%
2011 2 0 4 1 2 28.57%
2012 2 1 4 2 3 37.50%
2013 1 0 1 1 1 33.33%
2014 1 3 1 1 2 28.57%
2015 1 0 0 0 0 0.00%
2003–2015 21 8 32 9 23 30.26%
Frequency 19.27% 7.34% 29.36% 8.26%
We review all auditor specialist changes over the period from 2003 to 2015. The sample consists of all firms with the minimum required Compustat
variables: U.S. headquarters location, sales, SIC, and auditor. We identify nine major reasons for the new specialist obtaining the market leadership role.
Merger by Unseasoned Auditor Client indicates that the new industry-specialist’s client grew through acquisition, increasing the new specialist’s market
share above the threshold. Divestiture by Client indicates that some firm in the industry experienced a divestiture that allowed the new specialist to capture
the necessary market share. Industry Entry indicates that some new firm joined the industry (e.g., via initial public offering [IPO]), altering market share in
a way that creates a new specialist. Relatedly, Industry Exit indicates that some firm previously in the industry exited (e.g., via bankruptcy or
privatization). Former Client Acquired indicates that a client not associated with the new specialist was acquired. Similar to Divestiture by Client, this
indicates that the overall industry size decreased, pushing the new specialist above the market share-based specialization threshold. Auditor Change to
Unseasoned Auditor—Major indicates a change to the new specialist that, in itself, is large enough to alter the specialist designation. The Minor version
indicates a change to the new specialist that was not large enough itself to change specialization, but changed specialization when combined with other
events. Client Sales Growth refers to sales growth or declines by firms in the industry that, combined, are large enough to change the specialist designation.
Major Auditor Change Unrelated to Unseasoned Auditor indicates that a client changed auditors, resulting in a large enough market share change to trigger
the creation of a new specialist, even though the new specialist was not involved in the auditor change. Industries with Multiple Causes indicates cases
where an industry experienced more than one of the events detailed here. Note that the sum across columns exceeds the 76 total industry-years
experiencing a specialist change because of the occurrence of multiple causes in some industries.
expertise and then attract clients based on this expertise. This provides initial evidence that seasoning likely affects industry-
specialist audit quality; if an auditor finds itself in the position of market share leader before developing the tools to provide
higher-quality audits, then the audit quality produced by new specialists will differ from that of seasoned specialists.
IV. DATA
Sample
To investigate the association between industry specialization and audit quality, we identify 68,002 firm-year observations
from Compustat during the 2003 through 2015 period with a Big 4 auditor in the current and prior year. We limit the sample to
Big 4 clients in order to avoid any self-selection bias that could arise from auditor choice (Francis 2011) and because only Big 4
auditors tend to be industry specialists (DeFond and Zhang 2014). Further, the Big 4 audit over 95 percent of firms, by market
value, over our sample period. We begin the sample in 2003 to exclude observations influenced by the Enron scandal or the
collapse of Arthur Andersen. We also exclude firms audited by Arthur Andersen in the prior year because Blouin, Grein, and
Rountree (2007) report that some firms retained their Arthur Andersen team under the new auditor, making it difficult to
attribute the reporting quality of the client to the characteristics of the current auditor. As shown in Table 2, of the 68,002 firm-
year observations with Big 4 auditors in the current and prior year, 25,901 (13,863) have sufficient data to estimate Equations

TABLE 2
Sample Selection
Audit Quality Proxy
Income-
Data Restrictions ADA and DA Increasing DA BTD
Compustat firm-year observations 2003–2015 118,105 118,105 118,105
Less:
Firm-years not audited by Big 4 in the current year (44,861) (44,861) (44,861)
Firm-years not audited by Big 4 in the prior year (5,242) (5,242) (5,242)
Total Audited by Big 4 in Current and Prior Year 68,002 68,002 68,002
Less:
Firm-years with missing dataa (42,101) (42,101) (57,451)
Firm-years with non-positive DA NA (12,038) NA
Total for Unmatched Regressions 25,901 13,863 10,551
Less:
Firm-years not matchedb (13,748) (7,348) (4,600)
Total for Propensity Score Matched Regressions 12,153 6,515 5,951
Plus:
Control firm-years matched to multiple treatment firm-years for main propensity 5,799 3,273 3,628
score matched samplec
Full Propensity Score Matched Sample 17,952 9,788 9,579
a
Observations are deleted when there are insufficient data to calculate the audit quality proxy or to estimate the propensity score model using the specialist
status of the auditor, control variables, and year fixed effects.
b
Observations are matched by propensity score, within common support, with replacement, using a caliper distance of 0.03 and a one-to-three match. The
propensity of choosing a specialist auditor is predicted using a logistic regression of the auditor’s specialist status on variables related to the client’s level
of earnings quality and year and industry fixed effects.
c
Matching with replacement allows a single control firm-year to be matched to several treatment firms. Thus, the control firm-year can appear in the
sample more than once. We use WLS to address the fact that not all observations in our regressions are unique (Hill and Reiter 2006).
(1) and (2) when absolute discretionary accruals (income-increasing DA) are used to measure audit quality. Likewise, 10,551
firm-year observations have sufficient data to estimate Equations (1) and (2) when audit quality is measured using book-tax
differences (BTD).
Following MM, we examine the effect of industry specialization on audit quality using unmatched samples, as well as
samples matched on the propensity of choosing a specialist auditor. We estimate a propensity score using a logistic
regression where the dependent variable is the specialist indicator variable and the independent variables are the control
variables in Equation (1), including industry and year indicator variables. Appendix A presents our logistic propensity score
model. Roberts and Whited (2013) suggest matching with replacement to reduce bias and selecting multiple matches within a
reasonable range (caliper) to increase precision, so we match with replacement and allow each treatment firm to match with
up to three control firms. We use weighted least squares (WLS) regressions to account for the fact that when matching with
replacement, not all of our observations are unique (Dehejia and Wahba 2002; Hill and Reiter 2006; Stuart 2010; DeFond et
al. 2017a; Shipman et al. 2017).21 We cluster standard errors by firm to further address the presence of repeated firms
(deHaan, Hodge, and Shevlin 2013; DeFond et al. 2017a). After matching, we are left in Table 2 with 12,153 unique
21
Ordinary least squares (OLS) has no mechanism for addressing repeated (non-unique) observations. Thus, WLS is required. Observations are matched
by propensity score, within common support, using a caliper distance of 0.03. We weight observations as described in Hill and Reiter (2006, 2232). The
propensity score matching and weighted least squares codes used in this paper are available on Steven Utke’s SSRN page at: https://ssrn.com/
abstract¼2964990. Code for calculating specialist tenure is also available at: https://ssrn.com/abstract¼3212035. We thank Marcelo Coca-Perraillon for
assisting with SAS PSM code (see Coca-Perraillon 2006), and William Thomas for making code available online (available at: http://www.biostat.umn.
edu/%7Ewill/6470stuff/Class25-12/PSmatching.sas).

122 Gaver and Utke
observations in the ADA sample, 6,515 unique observations in the income-increasing DA sample, and 5,951 unique
observations in the BTD sample. The full PSM samples, which include non-unique control firms that are matched to multiple
treatment firms, consist of 17,952, 9,788, and 9,579 observations for the ADA, income-increasing DA, and BTD samples,
respectively.
Although the use of PSM facilitates comparison to MM’s findings, PSM has drawbacks. Broadly, PSM does not ensure
that matched firms are similar (King, Nielsen, Coberley, Pope, and Wells 2011; King and Nielsen 2016; DeFond et al. 2017a;
McMullin and Schonberger 2018).22 As a result, PSM often selects random matches and is unable to resolve the functional
form misspecification (i.e., covariate imbalance) that it is intended to resolve (King et al. 2011; King and Nielsen 2016). To
illustrate, we review recently published papers in the top three accounting journals that use PSM as their primary method of
analysis.23 Across all studies, 17 percent of covariates exhibit potential imbalance after PSM, indicated by a statistical
difference in means and/or medians.24 Further, PSM involves matching treatment firms to a subset of firms identified as control
firms, and discarding the remaining firms. This can result in low-power tests. An additional limitation of PSM that is
particularly relevant for our study is that PSM cannot account for covariate imbalance across multiple groups. This is not a
problem for MM, who makes one comparison of audit quality between specialist and non-specialist auditors. Our research
design, on the other hand, requires us to make three comparisons of audit quality: between seasoned and unseasoned specialists;
between seasoned specialists and non-specialist auditors; and between unseasoned specialists and non-specialists. PSM is
limited in its ability to address multiple groups, and the fact that PSM discards data makes it difficult to perform PSM in the
relatively smaller subgroups that arise when making comparisons within multiple groups.
Recognizing the limitations of PSM for our research setting, we employ two non-PSM matching strategies. The first,
entropy balancing, is a covariate balancing method introduced by Hainmueller (2012) and implemented in Hainmueller and Xu
(2013). Unlike PSM, entropy balancing is an ‘‘equal percent bias reducing’’ matching method, which ensures that covariate
imbalance improves after matching. Entropy balancing achieves this by using an iterative process to reweight control sample
observations until the means of the control sample covariates approximately equal those in the treatment sample. In contrast to
PSM, entropy balancing does not discard observations, which increases power, and does not generate random matches (King et
al. 2011).25 King and Nielsen (2016) also suggest that matching on specific firm characteristics can be a preferable strategy
compared to PSM. Accordingly, our second alternative to PSM is simply to limit the sample to large multinational firms, which
likely require auditors with similar capabilities for addressing complex international issues. Limiting the sample to these large
firms also addresses the concern that the inclusion of small firms can bias results (Bamber, Christensen, and Gaver 2000;
Givoly, Hayn, and Lourie 2016; Hou, Xue, and Zhang 2018).
In summary, our primary tests use the unmatched sample, one-to-three PSM with replacement, entropy balancing, and a
sample matched on firm characteristics. We refer to the results from these tests as our ‘‘main results.’’26 In supplementary
analysis, we also investigate the impact of alternative PSM design choices, as suggested by Shipman et al. (2017), varying the
number of treatment to control firms (one-to-one, one-to-two, one-to-five, and one-to-ten) and matching with and without
replacement.27 PSM performance tends to deteriorate as the number of covariates increases (DeFond et al. 2017a), so we also
use a simplified PSM model based only on size, industry, and year. Finally, we perform robustness checks on our samples that
match on specific firm characteristics, examining large firms and multinational firms separately.28
22
For example, assume that we estimate a propensity score using ROA, GROWTH, and LEV. For simplicity, assume that the estimated beta coefficients
are all 1. Firm T has a propensity score of 0.25, with ROA of 0.10, GROWTH of 0.05, and LEV of 0.10. Firm C also has a propensity score of 0.25 with
ROA of 0.20, GROWTH of 0.20, and LEV of 0.25. Despite identical propensity scores, these firms are clearly very different (King and Nielsen 2016).
Examining covariate balance in a PSM sample only assures covariate balance, on average, across all firms, not match by match.
23
We thank Jonathan Shipman, Quinn Swanquist, and Rob Whited for providing this list of 27 propensity score matching papers reviewed in Shipman et
al. (2017).
24
The potential imbalance ranges from zero to 67 percent of covariates. However, many studies do not report balance statistics, or only test imbalance at
the mean or median, but not both. While determining imbalance is subjective (Shipman et al. 2017), we use an objective test based on statistical
differences to facilitate our review.
25
McMullin and Schonberger (2018) provide an excellent discussion and application of entropy balancing in an accounting setting. We implement
entropy balancing as described in their paper.
26
In our main results, we perform entropy balancing in our entire sample in order to facilitate comparisons to MM. However, entropy balancing in the full
sample does not resolve the inability to balance covariates across multiple groups, a relevant issue in our setting. In additional robustness tests reported
in Section V, we split our sample to focus only on the two groups that we compare in each of our hypotheses, and drop the third group, creating three
separate subsamples (seasoned versus unseasoned, seasoned versus non-specialist, and unseasoned versus non-specialist). We then perform entropy
balancing within each subsample. Our main results are robust to this approach.
27
This includes MM’s sampling choice of one-to-one matching without replacement.
28
Entropy balancing leaves little discretion in design choices (McMullin and Schonberger 2018), limiting the usefulness of robustness checks. We use the
default settings in Hainmueller and Xu’s (2013) ‘‘ebalance’’ Stata macro for our entropy balancing. We thank Jens Hainmueller, Jeff McMullin, Bryce
Schonberger, and Yiqing Xu for making entropy balancing code available online.

Descriptive Statistics and Univariate Results

We present descriptive statistics in Table 3, Table 4, and Table 5. Panels A and B of Table 3 report means, medians, and
standard deviations for variables in the unmatched (Panel A) and matched (Panel B) ADA samples.29 In both panels, about 20
percent of firm-year observations involve a seasoned specialist auditor, 2.5 percent involve an unseasoned specialist auditor,
and the remaining 77.5 percent have a non-specialist auditor. This means that approximately 11 percent of specialist auditors
are in their first year as the industry leader (2.5 percent/(2.5 percent þ 20 percent)). Table 5, which uses the unmatched ADA
sample, shows that the representation of each specialist auditor type varies over the sample years. The highest percentage of
observations with an unseasoned industry specialist occurs in 2011 (5.37 percent) and the lowest occurs in 2015 (0 percent).
We observe similar results in the BTD sample (untabulated).
Panels C and D of Table 3 present means and medians for all variables in the ADA sample after partitioning observations
by auditor specialization. Panel C reports findings for the unmatched sample and Panel D reports findings for the matched
sample. Univariate tests of differences in means and medians provide initial insight into our hypotheses.30 H1 is supported
when measuring audit quality using absolute discretionary accruals (ADA). That is, the mean and median values of ADA are
significantly higher (audit quality is worse) when the auditor is an unseasoned specialist compared to when the auditor is a
seasoned specialist. This is true in both the unmatched (Panel C) and matched (Panel D) samples. Likewise, H2 is supported
when audit quality is measured as ADA. Specifically, both mean and median values of ADA are significantly lower (audit
quality is better) when the auditor is a seasoned specialist compared to cases when the auditor is a non-specialist. This is true in
both unmatched (Panel C) and matched (Panel D) samples. In the unmatched sample, mean and median values of ADA do not
differ significantly between firm-years with non-specialist auditors compared to those with unseasoned specialist auditors. In
the matched sample, unseasoned specialists provide lower audit quality than non-specialist auditors when audit quality is
measured as ADA.
Table 4, where BTD is the measure of audit quality, reports means and medians after partitioning observations by auditor
specialization.31 Panel A reports findings for the unmatched sample and Panel B reports findings for the matched sample.
Univariate tests of differences in means and medians provide consistent support for all of our hypotheses in both matched and
unmatched samples. Mean and median values of BTD are significantly higher (audit quality is worse) when the auditor is an
unseasoned specialist, compared to when the auditor is a seasoned specialist, in both the unmatched (Panel A) and matched
(Panel B) samples, consistent with H1. Likewise, in both unmatched and matched samples, mean and median values of BTD
are significantly higher when the auditor is a non-specialist, compared to cases where there is a seasoned specialist auditor,
supporting H2. Finally, there is no significant difference in mean and median values of BTD between unseasoned specialists
and non-specialist auditors, in both unmatched and matched samples, consistent with H3.
Table 6 presents Pearson and Spearman correlations based on the unmatched ADA sample. Correlations involving absolute
discretionary accruals (ADA) or book-tax differences (BTD) suggest that audit quality is higher when the auditor is a specialist
(SPECIALIST). Further, consistent with our hypotheses, absolute discretionary accruals (ADA) and book-tax differences (BTD)
are negatively correlated with the presence of seasoned specialists (SEASONED), but not unseasoned specialists
(UNSEASONED). Table 6 also reports significant correlations between audit quality variables and the control variables.
These variables are taken into account in our multivariate tests, which are discussed in the next section.
V. MULTIVARIATE RESULTS
Replication of Minutti-Meza (2013)

Following MM, we first examine the effect of auditor industry specialization in our unmatched samples.32 Column (1) in Table
7 presents the results. Panels A and B, C and D, and E and F of Table 7 present results for the ADA, income-increasing DA, and BTD
29
The statistics for DA in Table 3 and Table 6 are based on the larger ADA sample, rather than the subsample of firms with income-increasing DA. Thus,
DA in these tables includes both income-increasing and income-decreasing DA. We focus on income-increasing DA in our multivariate analyses and
defer discussion of results to Section V.
30
Table 3 (Table 4), Panels C and D (A and B) also provide covariate balance details. While assessment of covariate balance is subjective (Shipman et
al. 2017), we generally find improved covariate balance in our matched samples. Because we use nine PSM samples, three samples matched on
specific firm characteristics, and one entropy balanced sample, we believe it is unlikely that potential imbalance in any one sample drives our results.
We also address potential covariate imbalance by including appropriate control variables in the second-stage audit quality model (Shipman et al.
2017).
31
For the sake of brevity, we do not report descriptive statistics for the full BTD sample.
32
We cannot provide an exact replication of MM because, as discussed in Sections III and IV, we have a different sample period (2003–2015), we limit
our sample to clients of Big 4 auditors, and we use alternative measures of audit quality that are more appropriate for our research questions.

124 Gaver and Utke
TABLE 3
Descriptive Statistics
Absolute Discretionary Accruals (ADA) Sample
Panel A: Unmatched Sampleb

Variablea n Mean Median Std. Dev.
ADA 25,901 0.0608 0.0376 0.0730
DA 25,901 0.0004 0.0043 0.0889
SPECIALIST 25,901 0.1740 0.0000 0.3791
SEASONED 25,901 0.1550 0.0000 0.3619
UNSEASONED 25,901 0.0190 0.0000 0.1364
SEASONED2 25,901 0.1359 0.0000 0.3427
UNSEASONED2 25,901 0.0381 0.0000 0.1915
SEASONED3 25,901 0.1204 0.0000 0.3254
UNSEASONED3 25,901 0.0536 0.0000 0.2252
AU_CHANGE 25,901 0.0298 0.0000 0.1699
LOGMKT 25,901 7.0985 7.1175 2.0651
LEV 25,901 0.2920 0.2586 0.2201
ROAL 25,901 0.0012 0.0368 0.1702
ROA 25,901 0.0037 0.0368 0.1590
LOSS 25,901 0.2793 0.0000 0.4486
CFO 25,901 0.0722 0.0867 0.1352
BTM 25,901 0.5210 0.4513 0.7472
ABS_AC_LAG 25,901 0.0874 0.0620 0.0898
GROWTH 25,901 0.1341 0.0728 0.3711
ALTMAN 25,901 59.7805 3.7499 305.5100
STDEARN 25,901 223.1195 39.1577 565.9071
TENURE 25,901 0.7467 1.0000 0.4349
samples, respectively.33 For brevity, we only report coefficients on our variable of interest, SPECIALIST. Coefficients on the control
variables are generally consistent with prior research. If industry specialists provide higher audit quality than non-specialist auditors,
then the coefficients on SPECIALIST will be reliably negative. The results are consistent with these predictions in the ADA (Panels A
and B of Table 7) and income-increasing DA (Panels C and D of Table 7) samples, but not in the BTD (Panels E and F of Table 7)
sample. Thus, even without matching, an industry specialization proxy based only on market share produces somewhat mixed
results. This supports our argument that a purely market share-based proxy is a noisy measure of industry specialization.
MM argues that market share-based measures such as SPECIALIST result in a biased test of the relation between auditor
industry expertise and audit quality. Acquiring a large client increases an auditor’s market share. Client size is correlated with
financial reporting quality, which is the basis for our audit quality measures. Thus, client size is correlated with both the
variable of interest, (market share-based) SPECIALIST, and the dependent variable, audit quality (measured by ADA, income-
increasing DA, and BTD), resulting in model misspecification. MM uses PSM to address this misspecification, matching clients
of specialist and non-specialist auditors on relevant observable dimensions other than the treatment and outcome variables.
The results of reestimating Equation (1) with our main PSM sample (one-to-three with replacement) are reported in column (2)
of Table 7.34 Again, Panels A and B, C and D, and E and F of Table 7 present results for the ADA, income-increasing DA, and BTD
samples, respectively. The coefficient on SPECIALIST in Panels E–F (BTD) is insignificant, while the coefficients in Panels A–B and
C–D (ADA and income-increasing DA) remain significant, although the statistical significance of the result weakens relative to the
results for the unmatched sample reported in column (1). This is largely consistent with MM, who concludes that ‘‘[a]fter matching
33
As discussed earlier, the only signed DA measure examined in our main regression results is income-increasing DA, following Becker et al. (1998).
However, we also performed all analyses presented in Tables 7 through 9 using signed DA as the dependent variable. Results are largely consistent with,
although somewhat weaker than, the ADA and income-increasing DA results. Regarding income-decreasing DA, we note that prior literature (Reichelt and
Wang 2010) finds no evidence that SPECIALIST auditors, using our definition, constrain income-decreasing accruals. Consistent with their results, we find
no evidence that industry specialists in our sample constrain income-decreasing accruals, and virtually no significant differences across any groups of
specialists after performing our partition on seasoning. These results (untabulated) are consistent across matched and unmatched samples.
34
We use WLS when control firms are selected with replacement, and OLS when control firms are selected without replacement (Stuart 2010).

TABLE 3 (continued)
Panel B: Matched Samplec
Variablea n Mean Median Std. Dev.
ADA 17,952 0.0514 0.0332 0.0604
DA 17,952 0.0015 0.0030 0.0733
SPECIALIST 17,952 0.2500 0.0000 0.4330
SEASONED 17,952 0.2228 0.0000 0.4161
UNSEASONED 17,952 0.0272 0.0000 0.1626
LOGMKT 17,952 7.2704 7.3103 1.9651
LEV 17,952 0.3035 0.2785 0.2092
ROAL 17,952 0.0239 0.0380 0.1208
ROA 17,952 0.0258 0.0379 0.1119
LOSS 17,952 0.2176 0.0000 0.4126
CFO 17,952 0.0856 0.0856 0.0945
BTM 17,952 0.5427 0.4765 0.7611
ABS_AC_LAG 17,952 0.0762 0.0579 0.0704
GROWTH 17,952 0.1022 0.0667 0.2562
ALTMAN 17,952 31.8034 3.5414 156.5163
STDEARN 17,952 226.7203 41.1148 583.7800
TENURE 17,952 0.8147 1.0000 0.3886
Panel C: Unmatched Sample Partitioned by Auditor Specializationb

All
Specialists Non- Non-
Non- (Seasoned Specialists Specialists Seasoned
Non- Specialists and versus Seasoned versus Unseasoned versus
Specialists versus Unseasoned) Seasoned Specialists Unseasoned Specialists Unseasoned
Variablea n ¼ 21,395 Specialists n ¼ 4,506 Specialists n ¼ 4,015 Specialists n ¼ 491 Specialists
ADA Mean 0.063 *** 0.050 *** 0.049 0.060 ***
Median 0.039 *** 0.031 *** 0.030 0.039 ***
DA Mean 0.000 0.001 0.001 0.004
Median 0.004 0.005 0.005 0.007
LOGMKT Mean 7.053 *** 7.317 *** 7.332 7.190
Median 7.061 *** 7.361 *** 7.379 7.202
LEV Mean 0.290 *** 0.303 *** 0.304 0.292
Median 0.253 *** 0.283 *** 0.286 0.257
ROAL Mean -0.003 *** 0.023 *** 0.023 *** 0.020
Median 0.036 *** 0.038 *** 0.038 *** 0.047 *
ROA Mean -0.001 *** 0.024 *** 0.023 *** 0.031
Median 0.036 *** 0.038 *** 0.037 *** 0.048 ***
LOSS Mean 0.292 *** 0.218 *** 0.218 *** 0.220
Median 0.000 *** 0.000 *** 0.000 *** 0.000
CFO Mean 0.070 *** 0.084 *** 0.084 *** 0.087
Median 0.087 * 0.084 0.084 ** 0.089
BTM Mean 0.529 *** 0.482 *** 0.487 *** 0.444
Median 0.452 * 0.449 0.453 ** 0.419 **
ABS_AC_LAG Mean 0.090 *** 0.075 *** 0.074 ** 0.081 *
Median 0.064 *** 0.056 *** 0.055 ** 0.057
GROWTH Mean 0.141 *** 0.100 *** 0.096 0.135 ***
Median 0.075 *** 0.064 *** 0.061 0.078 ***
ALTMAN Mean 64.168 *** 38.950 *** 38.452 * 43.019
Median 3.760 ** 3.700 *** 3.618 *** 4.244 ***
STDEARN Mean 224.741 215.421 218.361 191.376
Median 38.822 40.556 39.955 42.100
TENURE Mean 0.732 *** 0.815 *** 0.819 *** 0.784 *
Median 1.000 *** 1.000 *** 1.000 ** 1.000 *

126 Gaver and Utke
TABLE 3 (continued)
Panel D: Matched Sample Partitioned by Auditor Specializationc
All
Specialists and versus Seasoned versus Unseasoned versus
Non-Specialists versus Unseasoned) Seasoned Specialists Unseasoned Specialists Unseasoned
ADA Mean 0.052 ** 0.050 *** 0.048 ** 0.059 ***
Median 0.034 *** 0.031 *** 0.030 *** 0.039 ***
DA Mean 0.002 0.001 0.001 0.005
Median 0.002 0.005 0.005 0.007
LOGMKT Mean 7.252 ** 7.326 ** 7.342 7.199
Median 7.288 * 7.365 ** 7.387 7.215
LEV Mean 0.304 0.302 0.303 0.292
Median 0.277 0.282 0.286 0.257
ROAL Mean 0.024 0.024 0.025 0.023
Median 0.038 ** 0.038 0.038 ** 0.047 *
ROA Mean 0.026 0.026 0.025 0.033
Median 0.038 * 0.038 0.037 *** 0.048 ***
LOSS Mean 0.218 0.218 0.218 0.219
Median 0.000 0.000 0.000 0.000
CFO Mean 0.086 0.086 0.085 0.088
Median 0.086 0.085 0.084 0.089
BTM Mean 0.565 *** 0.476 *** 0.481 *** 0.436
Median 0.487 *** 0.449 *** 0.452 *** 0.419 **
ABS_AC_LAG Mean 0.077 ** 0.074 *** 0.073 0.080 *
Median 0.058 *** 0.055 *** 0.055 0.057
GROWTH Mean 0.105 *** 0.093 *** 0.089 * 0.126 ***
Median 0.068 ** 0.064 *** 0.061 ** 0.078 ***
ALTMAN Mean 32.166 30.714 30.243 34.582
Median 3.510 3.703 3.620 *** 4.243 ***
STDEARN Mean 228.953 220.022 223.230 193.733
Median 41.171 40.556 40.028 42.073
TENURE Mean 0.814 0.816 0.820 0.785 *
Median 1.000 1.000 1.000 1.000 *
***, **, * Indicate significant differences in means or medians between the indicated groups at the 1 percent, 5 percent, or 10 percent level, respectively,
using two-tailed tests.
a
Variable definitions are provided in Appendix A. Variables are winsorized at 1 percent and 99 percent.
b
Panels A and C present results for the unmatched ADA sample of 25,901 firm-year observations that have Big 4 auditors in the current and in the prior
year and meet certain data requirements for the years 2003–2015.
c
Panels B and D present results for the propensity score matched ADA sample of 17,952 firm-year observations. Observations are matched by propensity score,
within common support, with replacement, using a caliper distance of 0.03 and a one-to-three match. The propensity of choosing a specialist auditor is predicted
using a logistic regression of the auditor’s specialist status on variables related to the client’s level of earnings quality and year and industry fixed effects.
clients of specialist and non-specialist auditors on a number of dimensions . . . there is no evidence of differences in commonly used
audit-quality proxies between these two groups of auditors . . . the auditor’s within-industry market share is not a reliable indicator of
audit quality’’ (Minutti-Meza 2013, 779–780). Columns (5) through (12) report coefficients using alternative PSM specifications, in
addition to our main PSM results reported in column (2). When audit quality is measured using ADA (Panels A and B of Table 7) or
BTD (Panels E and F of Table 7), about 56 percent (5/9) of the coefficients on SPECIALIST are insignificant, consistent with MM.
However, only 11 percent (1/9) of results are consistent with MM in the income-increasing DA (Panels C and D of Table 7) sample.
We note that many of the alternative PSM design choices examined in columns (5) through (12) lead to results more consistent with
MM than those of our main PSM design choice.
Columns (3) and (4) of Table 7 present results using non-PSM matching methods. Results are consistent with the
unmatched sample. That is, specialization improves audit quality when measured with ADA and income-increasing DA (Panels
A and C of Table 7), but not BTD (Panels E and F of Table 7). This is echoed in columns (13) and (14) (Panels B, D, and F of
Table 7), which present results using additional firm characteristic matching approaches.

TABLE 4
Descriptive Statistics
Book-Tax Differences (BTD) Sample
Panel A: Unmatched Sample Partitioned by Auditor Specializationb

All
Non- Specialists and versus Seasoned versus Unseasoned versus
BTD Mean 0.077 ** 0.044 *** 0.032 0.121 **
Median 0.039 *** 0.028 *** 0.026 0.042 *
LOGMKT Mean 7.202 *** 7.630 *** 7.649 *** 7.510
Median 7.179 *** 7.576 *** 7.573 *** 7.604
LEV Mean 0.258 0.259 0.257 0.268
Median 0.227 0.234 0.232 * 0.240
ROAL Mean 0.055 *** 0.065 *** 0.064 *** 0.069
Median 0.056 *** 0.063 *** 0.064 *** 0.061
ROA Mean 0.074 ** 0.076 0.076 ** 0.080
Median 0.060 *** 0.065 *** 0.065 *** 0.068
LOSS Mean 0.011 0.009 0.009 0.009
Median 0.000 0.000 0.000 0.000
CFO Mean 0.118 0.117 0.117 0.120
Median 0.109 0.113 0.112 0.117
BTM Mean 0.475 ** 0.456 ** 0.454 0.468
Median 0.414 ** 0.395 *** 0.390 0.420
ABS_AC_LAG Mean 0.074 *** 0.064 *** 0.063 0.068
Median 0.057 *** 0.052 *** 0.051 0.052
GROWTH Mean 0.128 *** 0.104 *** 0.101 0.122 *
Median 0.087 *** 0.075 *** 0.074 0.085 **
ALTMAN Mean 84.242 *** 55.226 *** 57.132 ** 43.126
Median 5.346 5.574 * 5.706 4.846 *
STDEARN Mean 122.529 *** 175.228 *** 177.457 ** 161.072
Median 29.192 *** 37.564 *** 37.051 *** 41.393
TENURE Mean 0.790 *** 0.841 *** 0.847 0.806 *
Median 1.000 *** 1.000 *** 1.000 1.000 *
Tests of H1
Main Test of H1
An explanation for the insignificant coefficient on SPECIALIST in some specifications discussed previously is that SPECIALIST
is a noisy proxy for industry expertise. Specifically, if the quality of audits produced by unseasoned specialists does not differ from
the quality of audits produced by non-specialist auditors, then SPECIALIST contains ‘‘false positives’’ where an auditor is coded as an
industry expert when, in fact, it is not. We address this problem by separating unseasoned specialists from seasoned specialists.
Table 8, Panels A and B, C and D, and E and F present the results of estimating Equation (2), where SEASONED and
UNSEASONED replace SPECIALIST. If, as predicted by H1, seasoned specialists outperform unseasoned specialists, then
the coefficient on UNSEASONED (b2) will exceed the coefficient on SEASONED (b1). That is, the difference (Difference)
between these coefficients (b1 b2) will be reliably negative. Row three (Difference) of columns (1) through (4) (Panels A,
C, and E of Table 7) presents our main results. H1 is generally supported in all four columns across all three audit quality
measures (ADA, income-increasing DA, and BTD).35 This provides evidence that duration of experience as the dominant
35
In order to rule out the possibility that unseasoned auditors that are new to their clients drive our results, we remove observations involving auditor
changes from both SEASONED and UNSEASONED and create a separate variable for auditor changes. Our main results for H1, H2, and H3 are robust
to this modification.

128 Gaver and Utke
TABLE 4 (continued)
Panel B: Matched Sample Partitioned by Auditor Specializationc
All Non- Non-
Non- Specialists Specialists Specialists Seasoned
Non- Specialists (Seasoned and versus Seasoned versus Unseasoned versus
BTD Mean 0.073 * 0.047 *** 0.036 0.123 ***
Median 0.046 *** 0.028 *** 0.025 0.042 **
LOGMKT Mean 7.522 *** 7.643 *** 7.661 7.524
Median 7.423 *** 7.591 *** 7.583 7.604
LEV Mean 0.268 ** 0.259 ** 0.257 0.269
Median 0.234 0.235 * 0.234 0.244
ROAL Mean 0.066 0.066 0.065 0.070
Median 0.060 * 0.063 * 0.064 0.060
ROA Mean 0.074 * 0.076 0.076 * 0.080
Median 0.061 *** 0.065 *** 0.065 ** 0.068
LOSS Mean 0.008 0.009 0.009 0.009
Median 0.000 0.000 0.000 0.000
CFO Mean 0.116 0.117 0.117 0.120
Median 0.105 *** 0.113 ** 0.112 0.117
BTM Mean 0.488 *** 0.456 *** 0.454 0.472
Median 0.426 *** 0.394 *** 0.388 0.422 *
ABS_AC_LAG Mean 0.063 0.063 0.063 0.066
Median 0.051 0.052 0.051 0.053
GROWTH Mean 0.107 0.101 ** 0.098 0.120 **
Median 0.077 0.075 ** 0.074 0.085 **
ALTMAN Mean 44.245 41.660 43.149 32.236
Median 5.200 *** 5.560 *** 5.700 4.798 **
STDEARN Mean 162.023 * 182.782 ** 184.663 170.878
Median 36.772 38.017 37.477 41.460
TENURE Mean 0.841 0.845 0.849 0.820
Median 1.000 1.000 1.000 1.000
***, **, * Indicate significant differences in means or medians between the indicated groups at the 1 percent, 5 percent, or 10 percent level, respectively,
using two-tailed tests.
a
Variable definitions are provided in Appendix A. Variables are winsorized at 1 percent and 99 percent.
b
Panel A presents results for the unmatched BTD sample of 10,551 firm-year observations that have Big 4 auditors in the current and in the prior year and
meet certain data requirements for the years 2003–2015.
c
Panel B presents results for the propensity score matched BTD sample of 9,579 firm-year observations. Observations are matched by propensity score,
within common support, with replacement, using a caliper distance of 0.03 and a one-to-three match. The propensity of choosing a specialist auditor is
predicted using a logistic regression of the auditor’s specialist status on variables related to the client’s level of earnings quality and year and industry
fixed effects.
industry auditor is associated with audit quality. Importantly, this experience effect is apparent even after matching on client
attributes.36
We assess the robustness of this finding by varying the matching approach, as well as the length of the seasoning period.
Columns (5) through (14) in Table 8 show the results of estimating Equation (2) using different matching strategies. The tenor
of our findings is unchanged: Difference is generally negative across sampling approaches for all three audit quality measures.
Row six reports the estimated Difference after redefining unseasoned auditors as those in their first two years as industry experts
(UNSEASONED2). Likewise, row nine reports the estimated Difference when auditors in their first three years as industry
experts are considered unseasoned (UNSEASONED3). In general, lengthening the seasoning period does not alter our
conclusions across the two accruals-based audit quality proxies. However, Difference is generally insignificant in the BTD
36
Our accruals sample includes some non-U.S. firms, which, by definition, cannot have specialist auditors. Removing these firms from the sample does
not change our inferences.

TABLE 5
Specialist Auditor Type by Yeara
Unseasoned Seasoned
Number of Non-Specialist Specialist Specialist
Year Observations Auditor (%) Auditor (%) Auditor (%)
2003 2,448 80.60% 1.80% 17.61%
2004 2,247 82.60% 1.60% 15.80%
2005 2,097 83.36% 3.67% 12.97%
2006 2,092 82.17% 2.25% 15.58%
2007 2,031 83.26% 0.98% 15.76%
2008 1,933 83.14% 0.05% 16.81%
2009 1,759 84.42% 1.36% 14.21%
2010 1,791 85.99% 0.78% 13.23%
2011 1,806 81.95% 5.37% 12.68%
2012 1,921 82.04% 2.13% 15.83%
2013 1,873 81.74% 3.10% 15.16%
2014 2,000 81.20% 1.60% 17.20%
2015 1,903 82.24% 0.00% 17.76%
2003–2015 25,901 82.60% 1.90% 15.50%
a
The sample consists of 25,901 firm-year observations that have Big 4 auditors in the current and in the prior year and meet data requirements for the ADA
sample for the years 2003–2015.
sample when the seasoning period is greater than one year (Panel C). We return to this point below in our examination of the
length of the seasoning period.37
Alternative Test of H1
An alternative approach for testing H1 is to restrict the sample to specialist clients, and directly compare unseasoned
specialists to seasoned specialists using Equation (3).
AQit ¼ a0 þ b1 UNSEASONEDit þ b2 LOGMKTit þ b3 LEVit þ b4 ROALit þ b5 ROAit þ b6 LOSSit þ b7 CFOit þ b8 BTMit
ð3Þ
An advantage of this approach is that it does not require the design choices associated with matching. H1 is supported if b1 is
positive.
The results of estimating Equation (3) support the inferences from Table 8. We find that when audit quality is measured
using ADA or income-increasing DA, b1 is positive for seasoning periods of one, two, and three years (untabulated). For BTD,
b1 is significant for a one-year seasoning period, but insignificant for seasoning periods of two or three years (untabulated),
echoing the results in Table 8. On balance, these results support H1.
Length of the Seasoning Period
To further investigate the length of the seasoning process, we return to Equation (2) and replace UNSEASONED with three
dummy variables that switch on when an auditor is in their first, second, or third year as a specialist (UNSEASONED1_Dum,
UNSEASONED2_Dum, and UNSEASONED3_Dum, respectively). SEASONED3 is coded 1 when the auditor has been the
industry leader for more than three years. This results in Equation (4).
AQit ¼ a0 þ b1 SEASONED3it þ b2 UNSEASONED1 Dumit þ b3 UNSEASONED2 Dumit þ b4 UNSEASONED3 Dumit
þ b5 LOGMKTit þ b6 LEVit þ b7 ROALit þ b8 ROAit þ b9 LOSSit þ b10 CFOit þ b11 BTMit þ b12 ABS AC LAGit
þ b13 GROWTHit þ b14 ALTMANit þ b15 STDEARNit þ b16 TENUREit þ bt YEAR FEt þ mit
ð4Þ
37
Lengthening the seasoning period increases the percentage of observations that are classified as unseasoned specialists in each regression. For the
unmatched ADA sample, 1.90 percent of observations are classified as unseasoned specialists under the one-year seasoning period (UNSEASONED),
compared to 3.81 percent for the two-year period (UNSEASONED2) and 5.36 percent for the three-year period (UNSEASONED3).

130 Gaver and Utke
TABLE 6
Correlations
Unmatched Samplea,b
Panel A: Correlation Variables ADA to ROAL

Variablec (1) (2) (3) (4) (5) (6) (7) (8) (9)
(1) ADA 0.101 0.065 0.068 0.070 0.002 0.268 0.026 0.285
(2) DA 0.036 0.069 0.004 0.002 0.006 0.032 0.025 0.047
(3) BTD 0.029 0.052 0.021 0.029 0.015 0.031 0.070 0.013
(4) SPECIALIST 0.077 0.001 0.032 0.933 0.303 0.049 0.023 0.058
(5) SEASONED 0.082 0.002 0.038 0.933 0.060 0.048 0.024 0.054
(6) UNSEASONED 0.002 0.006 0.008 0.303 0.060 0.006 0.000 0.015
(7) LOGMKT 0.251 0.010 0.052 0.049 0.050 0.006 0.044 0.380
(8) LEV 0.045 0.012 0.103 0.040 0.040 0.005 0.038 0.089
(9) ROAL 0.138 0.045 0.074 0.040 0.033 0.022 0.403 0.130
(10) ROA 0.138 0.170 0.089 0.038 0.029 0.028 0.432 0.161 0.679
(11) LOSS 0.230 0.184 0.028 0.062 0.058 0.018 0.421 0.063 0.523
(12) CFO 0.150 0.330 0.037 0.012 0.008 0.012 0.369 0.078 0.556
(13) BTM 0.064 0.067 0.094 0.012 0.007 0.015 0.178 0.140 0.098
(14) ABS_AC_LAG 0.154 0.134 0.069 0.060 0.059 0.012 0.191 0.036 0.286
(15) GROWTH 0.049 0.014 0.020 0.032 0.037 0.010 0.066 0.018 0.066
(16) ALTMAN 0.003 0.011 0.140 0.015 0.022 0.018 0.134 0.767 0.367
(17) STDEARN 0.130 0.000 0.008 0.002 0.003 0.001 0.688 0.154 0.111
(18) TENURE 0.086 0.010 0.016 0.072 0.071 0.012 0.148 0.018 0.099
Panel B: Correlation Variables ROA to TENURE

Variablec (10) (11) (12) (13) (14) (15) (16) (17) (18)
(1) ADA 0.324 0.255 0.243 0.075 0.285 0.160 0.064 0.062 0.102
(2) DA 0.276 0.220 0.246 0.053 0.071 0.005 0.009 0.017 0.017
(3) BTD 0.053 0.034 0.060 0.019 0.048 0.022 0.036 0.010 0.002
(4) SPECIALIST 0.059 0.062 0.040 0.024 0.065 0.042 0.031 0.006 0.072
(5) SEASONED 0.053 0.058 0.036 0.020 0.064 0.044 0.030 0.004 0.071
(6) UNSEASONED 0.024 0.018 0.015 0.014 0.010 0.000 0.008 0.008 0.012
(7) LOGMKT 0.418 0.423 0.372 0.070 0.245 0.009 0.037 0.439 0.143
(8) LEV 0.106 0.111 0.064 0.244 0.075 0.027 0.245 0.002 0.029
(9) ROAL 0.715 0.491 0.696 0.113 0.475 0.143 0.015 0.054 0.104
(10) ROA 0.681 0.779 0.069 0.279 0.027 0.013 0.079 0.094
(11) LOSS 0.777 0.496 0.003 0.261 0.010 0.007 0.051 0.089
(12) CFO 0.646 0.495 0.024 0.163 0.056 0.003 0.086 0.084
(13) BTM 0.185 0.018 0.179 0.122 0.045 0.038 0.007 0.004
(14) ABS_AC_LAG 0.176 0.222 0.070 0.111 0.043 0.041 0.030 0.097
(15) GROWTH 0.213 0.158 0.138 0.146 0.014 0.068 0.081 0.139
(16) ALTMAN 0.455 0.283 0.316 0.182 0.066 0.178 0.052 0.037
(17) STDEARN 0.127 0.116 0.139 0.007 0.006 0.136 0.186 0.035
(18) TENURE 0.079 0.089 0.065 0.011 0.081 0.109 0.028 0.121
a
This sample consists of 25,901 firm-year observations (10,551 for BTD) that have Big 4 auditors in the current and prior years and meet certain data
requirements for the years 2003–2015.
b
Pearson (Spearman) correlations are above (below) the diagonal. Bold coefficients are significant at the 5 percent level using two-tailed tests.
c
Variables are defined in Appendix A.
Table 9 reports our main results from estimating Equation (3) using the three alternative measures of audit quality (ADA
and income-increasing DA in Panel A of Table 7, BTD in Panel B of Table 7). The estimated differences in the coefficients on
the seasoned and unseasoned dummy variables are reported in rows 5 through 7 of Table 9. A significantly negative difference
means that an unseasoned specialist provides lower audit quality than a seasoned specialist. As in Table 8, the results vary
somewhat with the audit quality measure. In general, however, they suggest that the seasoning process extends beyond the first

TABLE 7
Estimated Coefficients and p-values from the Regressions of Audit Quality Measures on Auditor Industry
Specialization and Control Variables
Panel A: Absolute Discretionary Accruals

ADAit ¼ a0 þ b1 SPECIALISTit þ b2 LOGMKTit þ b3 LEVit þ b4 ROALit þ b5 ROAit þ b6 LOSSit þ b7 CFOit þ b8 BTMit
Main Results Alternative Matching Results

(2)
(1) Main (3) (4) (5) (6) (7)
Unmatched Matched Entropy MNCs with 1:1 with 1:2 with 1:5 with
Sample Sample Balanced MVE . 500 Replacement Replacement Replacement
Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff.
Variable Pred. (p-value) (p-value) (p-value) (p-value) (p-value) (p-value) (p-value)
SPECIALIST 0.005*** 0.002* 0.006*** 0.003** 0.003** 0.002 0.001
(0.000) (0.088) (0.000) (0.011) (0.015) (0.103) (0.185)
Control Variables Yes Yes Yes Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes
n 25,901 12,153 25,901 10,379 7,910 10,340 14,735
Adjusted R2 0.191 0.153 0.148 0.132 0.158 0.151 0.142
Panel B: Absolute Discretionary Accruals (continued)

Alternative Matching Results
(8) (9) (10) (11) (12)
1:10 with 1:1 without 1:2 without 1:3 without Size-Industry- (13) (14)
Replacement Replacement Replacement Replacement Year MVE . $500 MNCs
SPECIALIST 0.001 0.002** 0.001 0.002** 0.001 0.005*** 0.003***
(0.155) (0.030) (0.199) (0.046) (0.123) (0.000) (0.003)
n 18,393 8,924 12,188 14,198 12,281 17,371 14,145
Adjusted R2 0.143 0.163 0.151 0.159 0.154 0.149 0.168
***, **, * Indicate significance at the 1 percent, 5 percent, and 10 percent levels, respectively.
Huber-White robust standard errors are clustered by firm and are used to control for heteroscedasticity and serial correlation. When predictions are made,
p-values are one-tailed. We present results for a number of different matching techniques in columns (2)–(14). Where applicable, the number after the
colon in the column heading represents the number of control firms. We match with replacement (with Replacement) and without replacement (without
Replacement), perform a reduced match using only size, industry, and year (1:3 with replacement), and perform exact matches with firms with market
value of equity (MVE) greater than $500 million and with multinational firms (MNCs). Regressions for samples matched with replacement are estimated
using WLS (Hill and Reiter 2006).
year. Results using BTD as the measure of audit quality (columns (9) to (12)) indicate that seasoning is complete in the second
year as market share leader; this explains the insignificant Difference in Table 8 for two- and three- year seasoning periods
when using BTD to measure audit quality. Results using income-increasing DA (columns (5) to (8)) indicate that the auditor
does not perform as a true expert until the third year. Results using ADA (columns (1) to (4)) suggest that even after three years,
unseasoned auditors do not perform as true experts. However, in the ADA sample, the size and statistical significance of the
difference in audit quality between seasoned and unseasoned auditors is lower after two or three years compared to the initial
year of industry specialization.

132 Gaver and Utke
TABLE 7 (continued)
Panel C: Income-Increasing (Positive) Discretionary Accruals
DAit ¼ a0 þ b1 SPECIALISTit þ b2 LOGMKTit þ b3 LEVit þ b4 ROALit þ b5 ROAit þ b6 LOSSit þ b7 CFOit þ b8 BTMit

(2)
(1) Main (3) (4) (5) (6) (7)
SPECIALIST 0.005*** 0.002** 0.005*** 0.003*** 0.002* 0.002* 0.002*
(0.000) (0.047) (0.000) (0.006) (0.088) (0.050) (0.067)
n 13,863 6,515 13,863 5,637 4,254 5,532 7,965
Adjusted R2 0.569 0.518 0.527 0.467 0.514 0.532 0.504
Panel D: Income-Increasing (Positive) Discretionary Accruals (continued)

(8) (9) (10) (11) (12)
SPECIALIST 0.002* 0.003** 0.002* 0.002** 0.002 0.005*** 0.005***
(0.058) (0.019) (0.066) (0.024) (0.102) (0.000) (0.000)
n 9,952 4,762 6,362 7,497 6,637 9,439 7,549
Adjusted R2 0.504 0.513 0.530 0.529 0.533 0.497 0.532
An alternative approach for assessing the length of the seasoning period is to use a continuous measure of specialist tenure,
rather than the indicator variables described above.38 In the unmatched ADA and income-increasing DA samples, we find a
negative coefficient on the continuous specialist tenure variable (untabulated), indicating that audit quality is positively related
to specialist tenure. In the PSM (entropy balanced) sample, this result is apparent in the ADA and BTD (ADA and income-
increasing DA) samples. We suggest that audit quality is unlikely to increase linearly with specialist tenure, and this could
explain these mixed results.
38
We note that a concurrent working paper, Barnes (2015), examines specialist tenure using a continuous measure of industry specialization and
unmatched samples.

TABLE 7 (continued)
Panel E: Book-Tax Differences
BTDit ¼ a0 þ b1 SPECIALISTit þ b2 LOGMKTit þ b3 LEVit þ b4 ROALit þ b5 ROAit þ b6 LOSSit þ b7 CFOit þ b8 BTMit

(2)
(1) Main (3) (4) (5) (6) (7)
SPECIALIST 0.020 0.022 0.019 0.003 0.017 0.023 0.021
(0.132) (0.109) (0.144) (0.444) (0.200) (0.104) (0.118)
Industry Fixed Effects Yes Yes Yes Yes Yes Yes Yes
n 10,551 5,951 10,551 6,112 4,097 5,149 6,931
Adjusted R2 0.069 0.080 0.067 0.043 0.085 0.081 0.072
Panel F: Book-Tax Differences (continued)

(8) (9) (10) (11) (12)
SPECIALIST 0.022* 0.027* 0.024* 0.021 0.033** 0.015 0.020
(0.096) (0.065) (0.091) (0.126) (0.042) (0.199) (0.165)
n 8,266 4,572 5,970 6,799 6,024 7,882 7,515
Adjusted R2 0.067 0.076 0.071 0.076 0.072 0.092 0.036
using WLS (Hill and Reiter 2006). Industry fixed effects are included by two-digit SIC in the BTD regressions.
Tests of H2
If seasoned specialists outperform non-specialists, as predicted by H2, then the coefficient on SEASONED (b1) in Equation (2)
will be negative. In general, the results in Table 8 support H2, although they are sensitive to the audit quality measure, the matching
strategy, and the seasoning period. When audit quality is measured as ADA (Panels A and B in Table 8) and income-increasing DA
(Panels C and D in Table 8), b1 is reliably negative for all seasoning periods using the matched and unmatched samples.39 When
39
Again, the main results in the accruals sample are generally robust to excluding non-U.S. firms, although the statistical significance of SEASONED
weakens slightly in some of the PSM samples.

134 Gaver and Utke
TABLE 8
Estimated Coefficients and p-values from the Regressions of Audit Quality Measures on Auditor Industry
Specialization and Control Variables
Where Specialists are Classified as Seasoned or Unseasoned
Panel A: Absolute Discretionary Accruals
ADAit ¼ a0 þ b1 SEASONEDit þ b2 UNSEASONEDit þ b3 LOGMKTit þ b4 LEVit þ b5 ROALit þ b6 ROAit þ b7 LOSSit
þ b8 CFOit þ b9 BTMit þ b10 ABS AC LAGit þ b11 GROWTHit þ b12 ALTMANit þ b13 STDEARNit
þ b14 TENUREit þ bt YEAR FEt þ mit

(2)
(1) Main (3) (4) (5) (6) (7)
Primary Analysis
SEASONED 0.006*** 0.003** 0.007*** 0.004*** 0.004*** 0.003** 0.002**
(0.000) (0.014) (0.000) (0.003) (0.002) (0.020) (0.041)
UNSEASONED ? 0.004 0.007** 0.003 0.002 0.006* 0.007** 0.007**
(0.208) (0.013) (0.263) (0.534) (0.054) (0.020) (0.011)
Difference 0.010*** 0.010*** 0.010*** 0.005** 0.010*** 0.009*** 0.009***
(0.000) (0.000) (0.000) (0.027) (0.000) (0.001) (0.001)
Alternative Seasoning Periods
SEASONED2 0.007*** 0.003*** 0.007*** 0.004*** 0.004*** 0.003** 0.003**
(0.000) (0.008) (0.000) (0.002) (0.001) (0.012) (0.022)
UNSEASONED2 ? 0.001 0.004* 0.000 0.000 0.002 0.003 0.004**
(0.816) (0.085) (0.918) (0.864) (0.341) (0.109) (0.047)
Difference 0.008*** 0.007*** 0.007*** 0.004** 0.007*** 0.007*** 0.007***
(0.001) (0.001) (0.001) (0.033) (0.002) (0.002) (0.001)
SEASONED3 0.008*** 0.004*** 0.008*** 0.005*** 0.005*** 0.004*** 0.003***
(0.000) (0.002) (0.000) (0.000) (0.000) (0.003) (0.005)
UNSEASONED3 ? 0.000 0.003* 0.001 0.001 0.002 0.003* 0.004**
(0.972) (0.061) (0.743) (0.799) (0.327) (0.075) (0.026)
Difference 0.008*** 0.007*** 0.007*** 0.005*** 0.007*** 0.007*** 0.007***
(0.000) (0.000) (0.000) (0.009) (0.000) (0.000) (0.000)
n 25,901 12,153 25,901 10,379 7,910 10,340 14,735
Adjusted R2 0.191 0.153 0.148 0.132 0.158 0.151 0.142
BTD is the audit quality measure (Panels E and F in Table 8), b1 is negative for most samples and seasoning periods. If we gauge
economic significance as the average b1 across all 14 specifications, SEASONED auditors reduce ADA by 6.17 percent (0.004/
0.061), income-increasing DA by 6.43 percent (0.004/0.057), and BTD by 42.42 percent (0.029/0.069). We note that BTD has a
much higher standard deviation than the other measures, resulting in larger, but less precise, estimates.
In sum, we find that SEASONED specialists generally provide higher audit quality than non-specialists across all of the
audit quality measures in unmatched samples. The same is true for most of the matched samples. This is despite the fact that
PSM likely reduces the power of our tests (Shipman et al. 2017) and may be unnecessary in a sample of Big 4 clients with fairly
homogeneous characteristics (Francis 2011). Entropy balancing, which is not subject to the random matching or power
reduction problems of PSM, generally provides stronger support for our hypotheses than PSM. Our finding of an industry
specialization effect in matched samples after separating unseasoned from seasoned specialists suggests that the MM result is

TABLE 8 (continued)
Panel B: Absolute Discretionary Accruals (continued)
(8) (9) (10) (11) (12)
Primary Analysis
SEASONED 0.002** 0.003*** 0.002** 0.003*** 0.002** 0.006*** 0.004***
(0.029) (0.004) (0.044) (0.006) (0.024) (0.000) (0.000)
UNSEASONED ? 0.007*** 0.006** 0.007** 0.006** 0.007** 0.002 0.003
(0.009) (0.032) (0.011) (0.036) (0.013) (0.458) (0.376)
Difference 0.010*** 0.010*** 0.009*** 0.009*** 0.010*** 0.008*** 0.007***
(0.000) (0.000) (0.001) (0.001) (0.000) (0.001) (0.010)
SEASONED2 0.003** 0.004*** 0.002** 0.004*** 0.003** 0.007*** 0.005***
(0.015) (0.002) (0.028) (0.002) (0.012) (0.000) (0.000)
UNSEASONED2 ? 0.004** 0.003 0.004* 0.003 0.004* 0.000 0.001
(0.045) (0.245) (0.068) (0.123) (0.067) (0.893) (0.720)
Difference 0.007*** 0.006*** 0.006*** 0.007*** 0.007*** 0.007*** 0.006**
(0.001) (0.002) (0.003) (0.002) (0.001) (0.000) (0.010)
SEASONED3 0.003*** 0.005*** 0.003*** 0.004*** 0.004*** 0.008*** 0.006***
(0.003) (0.001) (0.010) (0.001) (0.003) (0.000) (0.000)
UNSEASONED3 ? 0.004** 0.002 0.004* 0.003 0.004* 0.000 0.000
(0.027) (0.218) (0.056) (0.137) (0.054) (0.844) (0.912)
Difference 0.007*** 0.007*** 0.007*** 0.007*** 0.007*** 0.007*** 0.006***
(0.000) (0.000) (0.001) (0.000) (0.000) (0.000) (0.005)
n 18,393 8,924 12,188 14,198 12,281 17,371 14,145
Adjusted R2 0.143 0.163 0.151 0.159 0.154 0.149 0.168
attributable to mixing seasoned specialists with unseasoned specialists, who act more like non-specialists. We contend that
testing the association between auditor industry expertise and audit quality requires consideration of specialist tenure. However,
we acknowledge that our tests are not robust across all audit quality proxies, as discussed below in our analysis of alternative
audit quality measures.
Analysis of H3
If the quality of audits produced by unseasoned specialists does not differ from the quality of audits produced by non-
specialists, as stated in H3, then the coefficients on UNSEASONED (b2) in Equation (2) will be statistically indistinguishable
from zero. In general, we find little evidence in Panels A and B, C and D, and E and F of Table 8 that unseasoned specialist
audit quality differs from that of non-specialists. There are a few significantly positive values of b2 in Panels A and B (ADA)
and E (BTD) in Table 8, indicating that audit quality is lower when the auditor is an unseasoned specialist rather than a non-
specialist. Thus, audit quality is not statistically better, and possibly worse, when the auditor is an unseasoned specialist rather

136 Gaver and Utke
TABLE 8 (continued)
Panel C: Income-Increasing (Positive) Discretionary Accruals
DAit ¼ a0 þ b1 SEASONEDit þ b2 UNSEASONEDit þ b3 LOGMKTit þ b4 LEVit þ b5 ROALit þ b6 ROAit þ b7 LOSSit
þ b8 CFOit þ b9 BTMit þ b10 ABS AC LAGit þ b11 GROWTHit þ b12 ALTMANit þ b13 STDEARNit þ b14 TENUREit

(2)
(1) Main (3) (4) (5) (6) (7)
Primary Analysis
SEASONED 0.006*** 0.003** 0.006*** 0.004*** 0.003** 0.003** 0.002**
(0.000) (0.017) (0.000) (0.002) (0.041) (0.021) (0.024)
UNSEASONED ? 0.002 0.003 0.001 0.000 0.003 0.003 0.003
(0.527) (0.319) (0.674) (0.980) (0.320) (0.405) (0.231)
Difference 0.008*** 0.006** 0.007** 0.004 0.006** 0.005** 0.006**
(0.006) (0.031) (0.011) (0.115) (0.033) (0.043) (0.024)
SEASONED2 0.007*** 0.003*** 0.007*** 0.004*** 0.003** 0.003** 0.003**
(0.000) (0.009) (0.000) (0.001) (0.024) (0.012) (0.011)
UNSEASONED2 ? 0.000 0.002 0.000 0.000 0.002 0.001 0.002
(0.969) (0.361) (0.891) (0.884) (0.404) (0.480) (0.216)
Difference 0.007*** 0.005*** 0.006*** 0.004** 0.005** 0.005** 0.005***
(0.002) (0.009) (0.004) (0.038) (0.013) (0.015) (0.005)
SEASONED3 0.006*** 0.003** 0.006*** 0.004*** 0.003** 0.003** 0.003**
(0.000) (0.013) (0.000) (0.004) (0.027) (0.016) (0.017)
UNSEASONED3 ? 0.002 0.000 0.002 0.002 0.000 0.000 0.001
(0.174) (0.876) (0.162) (0.255) (0.810) (0.995) (0.711)
Difference 0.004** 0.003** 0.004** 0.002 0.004** 0.003* 0.003**
(0.024) (0.037) (0.024) (0.194) (0.034) (0.050) (0.031)
n 13,863 6,515 13,863 5,637 4,254 5,532 7,965
Adjusted R2 0.569 0.518 0.527 0.467 0.514 0.532 0.504
than a non-specialist. Worse audit quality could occur if a newly created industry specialist does not have the wherewithal to
meet the increased scale demanded by an expanded client base (Bills et al. 2016).
Alternative Definition of Specialist

Neal and Riley (2004) urge researchers to carefully consider their choice of specialization measure in studies of auditor
industry expertise. Accordingly, we examine the sensitivity of our results to an alternative definition of specialization. Palmrose
(1986) and Neal and Riley (2004) specify the minimum market share for specialization as 1.2 times the inverse of the number
of Big N auditors. In our study, this is 30 percent (1.2/4 ¼ 0.30). We reestimate our main results from Equations (1) and (2)
using the 30 percent measure of specialization instead of our original definition (industry market share leader by at least 10
percent). The results (untabulated) are generally consistent with our hypotheses, although somewhat weaker overall. First,
across both matched and unmatched samples and for all three audit quality measures (ADA, income-increasing DA, and BTD),
the coefficient on SPECIALIST (Equation (1)) is generally negative. The coefficient on SEASONED (Equation (2)) is also
negative, supporting H2. However, the coefficient on UNSEASONED (Equation (2)) also tends to be negative, especially at
longer seasoning periods. Thus, we generally find evidence consistent with H3 (no statistical difference between the audit

TABLE 8 (continued)
Panel D: Income-Increasing (Positive) Discretionary Accruals (continued)
(8) (9) (10) (11) (12)
Primary Analysis
SEASONED 0.002** 0.003*** 0.002** 0.003*** 0.002** 0.006*** 0.006***
(0.020) (0.007) (0.022) (0.005) (0.035) (0.000) (0.000)
UNSEASONED ? 0.003 0.002 0.003 0.004 0.004 0.001 0.002
(0.270) (0.496) (0.253) (0.220) (0.164) (0.695) (0.616)
Difference 0.006** 0.005** 0.006** 0.007** 0.007** 0.007** 0.007**
(0.029) (0.041) (0.027) (0.014) (0.016) (0.012) (0.011)
SEASONED2 0.003*** 0.004*** 0.003** 0.004*** 0.003** 0.007*** 0.006***
(0.009) (0.004) (0.011) (0.003) (0.016) (0.000) (0.000)
UNSEASONED2 ? 0.002 0.001 0.002 0.002 0.003 0.000 0.000
(0.242) (0.663) (0.290) (0.351) (0.195) (0.823) (0.895)
Difference 0.005*** 0.005** 0.005*** 0.005*** 0.006*** 0.007*** 0.006***
(0.006) (0.016) (0.009) (0.006) (0.006) (0.001) (0.005)
SEASONED3 0.003** 0.004*** 0.003** 0.003*** 0.003** 0.007*** 0.006***
(0.015) (0.006) (0.021) (0.009) (0.028) (0.000) (0.000)
UNSEASONED3 ? 0.000 0.001 0.000 0.000 0.001 0.002 0.003
(0.774) (0.770) (0.874) (0.801) (0.688) (0.250) (0.110)
Difference 0.003** 0.003** 0.003* 0.003* 0.003** 0.005*** 0.003
(0.038) (0.048) (0.054) (0.072) (0.042) (0.007) (0.106)
n 9,952 4,762 6,362 7,497 6,637 9,439 7,549
Adjusted R2 0.504 0.513 0.530 0.529 0.533 0.497 0.532
quality produced by unseasoned specialists compared to non-specialists) only for a one-year seasoning period. This is
consistent with auditors learning over time, as discussed above. H1 (seasoned specialists provide higher audit quality than
unseasoned specialists) is only supported when measuring audit quality using income-increasing DA. There are two
explanations for these somewhat weaker results. First, unlike our main specialization measure, which ensures that an auditor is
a market leader by at least 10 percent, the 30 percent rule may be too stringent for some industries and too liberal for others if
auditors group just above and below the 30 percent threshold in certain industries. Second, and relatedly, seasoning may occur
faster when measured by the 30 percent threshold if some auditors operate close to the specialist level just before achieving the
formal specialist designation.
Further analysis sheds light on why changing the specialist definition weakens results. We find that unseasoned specialists
defined using our primary specialization measure have larger mean increases in market share over the prior period compared to
unseasoned specialists defined using the 30 percent measure (16.7 percent versus 14.3 percent). Auditors experiencing rapid
industry market share growth are less likely to have an adequate supply of industry experts to meet staffing needs in the short
term and, therefore, are more likely to function as non-specialist auditors (Bills et al. 2016). In contrast, as discussed in Section

138 Gaver and Utke
TABLE 8 (continued)
Panel E: Book-Tax Differences
BTDit ¼ a0 þ b1 SEASONEDit þ b2 UNSEASONEDit þ b3 LOGMKTit þ b4 LEVit þ b5 ROALit þ b6 ROAit þ b7 LOSSit
þ b8 CFOit þ b9 BTMit þ b10 ABS AC LAGit þ b11 GROWTHit þ b12 ALTMANit þ b13 STDEARNit
þ b14 TENUREit þ bt YEAR FEt þ mit

(2)
(1) Main (3) (4) (5) (6) (7)
Primary Analysis:
SEASONED 0.027* 0.030* 0.030* 0.019 0.024 0.031** 0.028*
(0.077) (0.054) (0.059) (0.189) (0.121) (0.050) (0.067)
UNSEASONED ? 0.020 0.031 0.031 0.079** 0.029 0.030 0.024
(0.576) (0.400) (0.372) (0.041) (0.419) (0.413) (0.497)
Difference 0.047 0.061* 0.061* 0.099*** 0.053* 0.061* 0.052*
(0.106) (0.053) (0.050) (0.009) (0.070) (0.050) (0.085)
Alternative Seasoning Periods:
SEASONED2 0.025 0.031* 0.029* 0.011 0.025 0.031* 0.026*
(0.108) (0.066) (0.074) (0.326) (0.126) (0.063) (0.094)
UNSEASONED2 ? 0.007 0.002 0.003 0.015 0.006 0.001 0.004
(0.795) (0.948) (0.916) (0.618) (0.827) (0.981) (0.867)
Difference 0.018 0.033 0.032 0.026 0.031 0.030 0.022
(0.268) (0.139) (0.135) (0.230) (0.139) (0.148) (0.232)
SEASONED3 0.029* 0.038** 0.034* 0.018 0.032* 0.037** 0.032*
(0.089) (0.041) (0.056) (0.240) (0.080) (0.042) (0.066)
UNSEASONED3 ? 0.006 0.005 0.001 0.018 0.009 0.001 0.001
(0.804) (0.851) (0.976) (0.503) (0.716) (0.970) (0.977)
Difference 0.023 0.042* 0.035 0.036 0.042* 0.038* 0.031
(0.209) (0.073) (0.109) (0.136) (0.069) (0.090) (0.140)
n 10,551 5,951 10,551 6,112 4,097 5,149 6,931
Adjusted R2 0.069 0.080 0.067 0.043 0.085 0.081 0.072
II, auditors experiencing slower market share growth have more time to marshal needed resources for the increase in scale, and
are more likely to have been operating close to the critical mass of business threshold prior to earning the specialist designation.
This suggests that first-year specialists based on our primary definition of specialization are more likely to be truly unseasoned
than first-year specialists based on the 30 percent measure. We believe that the results using our primary specialization measure
are stronger because that measure allows a more powerful test of our hypotheses.
Alternative Audit Quality Measures

In this section, we examine our hypotheses using three additional audit quality measures. First, consistent with MM, we
use measures that capture the likelihood of (1) meeting or beating analysts’ forecasts, or (2) receiving a going concern opinion.
Because these are binary variables, we reestimate Equations (1) and (2) as logistic regressions. We supplement the meet or beat
regression with the natural logarithm of the firm’s analyst following and the standard deviation of analysts’ forecasts, following
MM. We estimate the going concern model in distressed firms (Reichelt and Wang 2010). We include industry fixed effects in
both models, following MM.

TABLE 8 (continued)
Panel F: Book-Tax Differences (continued)
(8) (9) (10) (11) (12)
Primary Analysis:
SEASONED 0.030* 0.036** 0.030* 0.027* 0.039** 0.028* 0.031*
(0.052) (0.027) (0.061) (0.084) (0.029) (0.068) (0.078)
UNSEASONED ? 0.025 0.028 0.007 0.012 0.002 0.055 0.036
(0.485) (0.448) (0.847) (0.755) (0.965) (0.130) (0.374)
Difference 0.054* 0.064** 0.037 0.038 0.040 0.083** 0.067*
(0.075) (0.044) (0.166) (0.166) (0.156) (0.013) (0.062)
Alternative Seasoning Periods:
SEASONED2 0.030* 0.040** 0.029* 0.028* 0.038** 0.025 0.021
(0.064) (0.022) (0.079) (0.096) (0.042) (0.105) (0.193)
UNSEASONED2 ? 0.001 0.007 0.012 0.005 0.020 0.009 0.018
(0.983) (0.805) (0.665) (0.855) (0.480) (0.744) (0.562)
Difference 0.030 0.047* 0.017 0.023 0.018 0.034 0.003
(0.163) (0.051) (0.286) (0.235) (0.285) (0.128) (0.468)
SEASONED3 0.035* 0.045** 0.032* 0.032* 0.042** 0.036** 0.018
(0.051) (0.017) (0.075) (0.081) (0.035) (0.048) (0.236)
UNSEASONED3 ? 0.001 0.001 0.013 0.005 0.017 0.015 0.022
(0.967) (0.959) (0.615) (0.854) (0.496) (0.542) (0.421)
Difference 0.034 0.046* 0.019 0.027 0.025 0.050** 0.004
(0.124) (0.051) (0.260) (0.190) (0.205) (0.037) (0.454)
n 8,266 4,572 5,970 6,799 6,024 7,882 7,515
Adjusted R2 0.067 0.076 0.071 0.076 0.072 0.092 0.036
using WLS (Hill and Reiter 2006). Industry fixed effects are included by two-digit SIC in the BTD regression.
When audit quality is measured as the likelihood of meeting or beating analysts’ forecasts, the coefficient on SEASONED3
(specialists with more than three years of seasoning) is generally negative (p-values 0.13), and there is a detectable difference
in the audit quality produced by seasoned and unseasoned specialists. In the subsample consisting of specialists only, we find
some evidence that seasoned specialists provide higher-quality audits than unseasoned specialists. Overall, while weaker than
our main results, tests using the likelihood of meeting or beating estimates are generally consistent with our three hypotheses.40
This is distinct from MM, who finds no evidence of a specialist effect on the likelihood of meeting or beating estimates when
seasoned and unseasoned specialists are not separated.
Analyses using going concern opinions yield no evidence that specialists (regardless of tenure) have an effect on audit
quality. However, going concern opinions may not be an appropriate audit quality proxy for our setting. First, going concern
40
As discussed in Section III, the weak results could be a result of using a dichotomous, rather than continuous, audit quality measure in our nuanced
setting.

140
TABLE 9
Estimated Coefficients and p-values from the Regressions of Audit Quality Measures on Auditor Industry Specialization and Control Variables
Where Specialists are Classified as Seasoned or Unseasoned Separately for One, Two, or Three Years
AQit ¼ a0 þ b1 SEASONED3it þ b2 UNSEASONED1 Dumit þ b3 UNSEASONED2 Dumit þ b4 UNSEASONED3 Dumit þ b5 LOGMKTit þ b6 LEVit þ b7 ROALit
þ b8 ROAit þ b9 LOSSit þ b10 CFOit þ b11 BTMit þ b12 ABS AC LAGit þ b13 GROWTHit þ b14 ALTMANit þ b15 STDEARNit þ b16 TENUREit
Panel A: Regressions Where Specialists Are Classified as Seasoned or Unseasoned Separately for One, Two, or Three Years
Dependent Variable: ADA Dependent Variable: Income-Increasing DA
(2) (6)
(1) Main (3) (4) (5) Main (7) (8)
Unmatched Matched Entropy MNCs with Unmatched Matched Entropy MNCs with
Sample Sample Balanced MVE . 500 Sample Sample Balanced MVE . 500
Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff.
Variable Pred. (p-value) (p-value) (p-value) (p-value) (p-value) (p-value) (p-value) (p-value)
SEASONED3 0.008*** 0.004*** 0.008*** 0.005*** 0.006*** 0.003** 0.006*** 0.004***
(0.000) (0.002) (0.000) (0.000) (0.000) (0.013) (0.000) (0.004)
UNSEASONED1_Dum ? 0.004 0.007** 0.003 0.002 0.002 0.003 0.001 0.000
(0.210) (0.014) (0.269) (0.536) (0.526) (0.320) (0.670) (0.978)
UNSEASONED2_Dum ? 0.003 0.000 0.004 0.001 0.002 0.001 0.002 0.001
(0.332) (0.937) (0.171) (0.742) (0.500) (0.753) (0.469) (0.774)
UNSEASONED3_Dum ? 0.001 0.003 0.001 0.001 0.008*** 0.004 0.008*** 0.006**
(0.654) (0.327) (0.649) (0.806) (0.002) (0.189) (0.006) (0.022)
b1 b2 0.011*** 0.011*** 0.011*** 0.007** 0.008*** 0.006** 0.008*** 0.004
(0.000) (0.000) (0.000) (0.012) (0.005) (0.025) (0.009) (0.113)
b1 b3 0.005** 0.004* 0.004* 0.004 0.005** 0.004* 0.005** 0.003
(0.045) (0.063) (0.074) (0.112) (0.040) (0.060) (0.044) (0.111)
b1 b4 0.006** 0.007** 0.007** 0.006* 0.002 0.000 0.001 0.002
(0.031) (0.015) (0.027) (0.051) (0.240) (0.437) (0.343) (0.229)
Control Variables Yes Yes Yes Yes Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes
Industry Fixed Effects No No No No No No No No
n 25,901 12,153 25,901 10,379 13,863 6,515 13,863 5,637
Adjusted R2 0.191 0.154 0.150 0.133 0.570 0.518 0.528 0.467

Gaver and Utke

TABLE 9 (continued)
Panel B: Regressions Where Specialists Are Classified as Seasoned or Unseasoned Separately for One, Two, or Three Years (continued)
Dependent Variable: BTD
(10)
(9) Main (11) (12)
Unmatched Matched Entropy MNCs with

Sample Sample Balanced MVE . 500
Coeff. Coeff. Coeff. Coeff.

Variable Pred. (p-value) (p-value) (p-value) (p-value)
SEASONED3 0.030* 0.038** 0.035* 0.019
Audit Quality and Specialist Tenure
(0.087) (0.041) (0.051) (0.233)

UNSEASONED1_Dum ? 0.020 0.032 0.032 0.079**
(0.575) (0.385) (0.359) (0.042)
UNSEASONED2_Dum ? 0.034 0.027 0.027 0.057
(0.327) (0.432) (0.447) (0.160)
UNSEASONED3_Dum ? 0.005 0.009 0.007 0.024
(0.914) (0.823) (0.861) (0.593)
b1 b2 0.049 0.070** 0.067** 0.098**
(0.108) (0.043) (0.044) (0.015)
b1 b3 0.005 0.011 0.009 0.038
(0.449) (0.382) (0.408) (0.197)
b1 b4 0.025 0.047 0.028 0.042
(0.572) (0.274) (0.261) (0.179)
Control Variables Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes
Industry Fixed Effects Yes Yes Yes Yes
n 10,551 5,951 10,551 6,112
Adjusted R2 0.069 0.080 0.067 0.043
Huber-White robust standard errors are clustered by firm and are used to control for heteroscedasticity and serial correlation. When predictions are made, p-values are one-tailed. The matched sample
regression is estimated using WLS (Hill and Reiter 2006). Industry fixed effects are not included in the accruals models because the dependent variable is industry-adjusted (Minutti-Meza 2013). For
book-tax differences, industry fixed effects are included by two-digit SIC.
141
142 Gaver and Utke
opinions are rare and issued to distressed clients, resulting in low statistical power (DeFond and Zhang 2014). This is a
particular concern for our study of clients of Big 4 auditors, which likely include fewer distressed firms compared to samples in
prior studies that include non-Big 4 clients. Second, Butler, Leone, and Willenborg (2004) find no evidence that going concern
opinions are associated with earnings management, raising questions about their validity as a proxy for audit quality. Chu,
Fogel-Yaari, and Zhang (2016) also suggest that going concern reports are frequently issued in error, and audit quality models
using going concern opinions yield inconsistent estimates, further undermining the validity of going concern opinions as an
audit quality proxy.
Our final measure of audit quality is the ERC. ERCs are widely used in the audit quality literature and are estimable for a
wide range of firms (Teoh and Wong 1993; Balsam et al. 2003; Lim and Tan 2008). They differ from our other measures of
audit quality in that they are based on market perceptions and, thus, are an indirect audit quality measure.
We regress the cumulative abnormal return around annual earnings announcement (CAR) on unexpected earnings (UE),
SPECIALIST, the interaction between unexpected earnings and the specialist variable (SPECIALIST UE), and control
variables from prior auditor specialization studies (Balsam et al. 2003; Lim and Tan 2008). This model is presented in Equation
(5a). If specialists produce higher-quality audits, then we expect a stronger market reaction to the earnings surprises of their
clients, indicated by a positive coefficient on SPECIALIST UE:
CARit ¼ a0 þ b1 UEit þ b2 SPECIALISTit þ b3 SPECIALIST UEit þ b4 BTMit þ b5 BTM UEit þ b6 VOLATILITYit
þ b7 VOLATILITY UEit þ b8 LEVit þ b9 LEV UEit þ b10 LOGMKTit þ b11 LOGMKT UEit þ b12 LOSSit
þ b13 LOSS UEit þ bt YEAR FEt þ bk YEAR FE UEt þ mit
ð5aÞ
When we estimate Equation (5a), we observe a positive coefficient on SPECIALIST UE only in the unmatched and entropy
balanced samples.
We then we replace SPECIALIST in Equation (5a) with two variables: SEASONED and UNSEASONED. Both variables are
interacted with UE. This model is presented in Equation (5b):
CARit ¼ a0 þ b1 UEit þ b2 SEASONEDit þ b3 SEASONED UEit þ b4 UNSEASONEDit þ b5 UNSEASONED UEit
þ b6 BTMit þ b7 BTM UEit þ b8 VOLATILITYit þ b9 VOLATILITY UEit þ b10 LEVit þ b11 LEV UEit
þ b12 LOGMKTit þ b13 LOGMKT UEit þ b14 LOSSit þ b15 LOSS UEit þ bt YEAR FEt þ bk YEAR FE
UEt þ mit
ð5bÞ
If seasoned specialists outperform non-specialists (H2), then the coefficient on SEASONED UE (b3) will be positive. If
seasoned specialists outperform unseasoned specialists (H1), then the coefficient on SEASONED UE (b3) will exceed the
coefficient on UNSEASONED UE (b5 ). Finally, if the quality of audits produced by unseasoned industry specialists does not
differ from the quality of audits produced by non-specialist auditors (H3), then the coefficient on UNSEASONED UE (b5 )
will be indistinguishable from zero. When we estimate Equation (5b), we observe a positive coefficient on SEASONED UE in
the unmatched and entropy balanced samples. This provides limited support for H2. On the other hand, we find little evidence
to support H1 or H3 when audit quality is gauged by the earnings response coefficient.
PSM When Using Partitioning Variables

As discussed earlier, a concern with using PSM in our setting is that we perform comparisons across multiple groups, but
PSM cannot account for covariate imbalance across each group. One approach to address this issue is to perform PSM
separately for each group. However, the issues inherent in PSM (e.g., sample size reduction) are especially problematic when
splitting samples into smaller groups. Given the limitations of PSM, we take two approaches to address possible covariate
imbalance across our partitions (untabulated). First, as discussed above in our alternative test of H1, we test H1 (seasoned
specialists provide higher-quality audits than unseasoned specialists) in the subsample of specialists only. As shown in Tables 3
and 4, there are relatively few differences in covariates between seasoned and unseasoned specialists. Results of this analysis
are consistent with our main results. Second, we split our sample to focus only on the two groups that we compare in each of
our hypotheses, and drop the third group, creating three separate subsamples (seasoned versus unseasoned, seasoned versus
non-specialist, and unseasoned versus non-specialist). We then perform entropy balancing within each subsample. Entropy
balancing ensures covariate balance and avoids discarding data and random matching inherent in PSM. The results of these
analyses are consistent with those presented in our main analyses. The implication is that differences in client characteristics are
unlikely to drive the relation between industry specialization and audit quality.

VI. CONCLUSION
Our study provides archival evidence that auditor industry expertise is generally associated with audit quality. However, a
dominant market share by itself does not make an industry expert. Auditors who find themselves in the dominant industry
position for the first time produce a level of audit quality that is indistinguishable from, or worse than, that produced by non-
specialist auditors, and lower than the audit quality produced by seasoned specialists. This pattern of results generally holds
even after using a multitude of matching techniques, which calls into question prior research attributing the specialization effect
to differences in client characteristics (MM). Our evidence suggests that the seasoning process takes up to three years, at which
time, unseasoned specialists produce audits of a similar quality to seasoned specialists. However, we caution that our results are
not robust to all audit quality measures.
To our knowledge, ours is one of the first papers to differentiate audit quality within specialist auditors and to estimate the
speed with which knowledge is created and assimilated by an audit firm. We also provide initial evidence on what causes
auditors to become specialists, finding that industry market share leaders are often created from exogenous events and
subsequently begin to act as specialists over time. Thus, expertise follows market share dominance, rather than the other way
around. This challenges a common assumption in the literature that clients self-select into industry specialist auditors (Gul et al.
2009).
Our approach differs somewhat from earlier city-level specialization studies because our sample is limited to clients of
the Big 4, which, compared to smaller auditors, have a more centralized national focus. Thus, our conclusions relate to Big
4 national industry specialists; examining industry specialization at the city level is beyond the scope of our paper. We look
to future research to provide additional evidence on these important topics using city-level specialization (e.g., Barnes
2015).
REFERENCES
Altman, E. I. 1983. Corporate Financial Distress: A Complete Guide to Predicting, Avoiding, and Dealing with Bankruptcy. New York,
NY: Wiley.
Altman, E. I. 2013. Predicting financial distress of companies: Revisiting the Z-Score and ZETA models. In Handbook of Research
Methods and Applications in Empirical Finance, edited by Bell, A. R., C. Brooks, and M. Prokopczuk, 428–456. Cheltenham,
U.K.: Edward Elgar.
Audousset-Coulier, S., A. Jeny, and L. Jiang. 2016. The validity of auditor industry specialization measures. Auditing: A Journal of
Practice & Theory 35 (1): 139–161. https://doi.org/10.2308/ajpt-51176
Badertscher, B. A., J. D. Phillips, M. Pincus, and S. O. Rego. 2009. Earnings management strategies and the trade-off between tax benefits
and detection risk: To conform or not to conform? The Accounting Review 84 (1): 63–97. https://doi.org/10.2308/accr.2009.84.1.63
Balsam, S., J. Krishnan, and J. S. Yang. 2003. Auditor industry specialization and earnings quality. Auditing: A Journal of Practice &
Theory 22 (2): 71–97. https://doi.org/10.2308/aud.2003.22.2.71
Bamber, L. S., T. E. Christensen, and K. M. Gaver. 2000. Do we really ‘‘know’’ what we think we know? A case study of seminal
research and its subsequent overgeneralization. Accounting, Organizations and Society 25 (2): 103–129. https://doi.org/10.1016/
S0361-3682(99)00027-6
Bao, J., and A. Edmans. 2011. Do investment banks matter for M&A returns? Review of Financial Studies 24 (7): 2286–2315. https://doi.
org/10.1093/rfs/hhr014
Barnes, M. 2015. The Consideration of Prior Experience in Auditor Industry Specialization. Doctoral dissertation, The University of
Oklahoma.
Becker, C. L., M. L. DeFond, J. Jiambalvo, and K. R. Subramanyam. 1998. The effect of audit quality on earnings management.
Contemporary Accounting Research 15 (1): 1–24. https://doi.org/10.1111/j.1911-3846.1998.tb00547.x
Bédard, J. 1989. Expertise in auditing: Myth or reality? Accounting, Organizations and Society 14 (1/2): 113–131. https://doi.org/10.
1016/0361-3682(89)90037-8
Bell, T. B., M. Causholli, and W. R. Knechel. 2015. Audit firm tenure, non-audit services, and internal assessments of audit quality.
Journal of Accounting Research 53 (3): 461–509. https://doi.org/10.1111/1475-679X.12078
Bills, K. L., Q. T. Swanquist, and R. L. Whited. 2016. Growing pains: Audit quality and office growth. Contemporary Accounting
Research 33 (1): 288–313. https://doi.org/10.1111/1911-3846.12122
Blouin, J., B. M. Grein, and B. R. Rountree. 2007. An analysis of forced auditor change: The case of former Arthur Andersen clients. The
Accounting Review 82 (3): 621–650. https://doi.org/10.2308/accr.2007.82.3.621
Butler, M., A. J. Leone, and M. Willenborg. 2004. An empirical analysis of auditor reporting and its association with abnormal accruals.
Journal of Accounting and Economics 37 (2): 139–165. https://doi.org/10.1016/j.jacceco.2003.06.004
Cahan, S. F., D. C. Jeter, and V. Naiker. 2011. Are all industry specialist auditors the same? Auditing: A Journal of Practice & Theory 30
(4): 191–222. https://doi.org/10.2308/ajpt-10181

144 Gaver and Utke
Cameran, M., J. R. Francis, A. Marra, and A. Pettinicchio. 2015. Are there adverse consequences of mandatory auditor rotation? Evidence
from the Italian experience. Auditing: A Journal of Practice & Theory 34 (1): 1–24. https://doi.org/10.2308/ajpt-50663
Carcello, J. V., and A. L. Nagy. 2004. Audit firm tenure and fraudulent financial reporting. Auditing: A Journal of Practice & Theory 23
(2): 55–69. https://doi.org/10.2308/aud.2004.23.2.55
Cassell, C. A., J. C. Hansen, L. A. Myers, and T. A. Seidel. 2017. Does the timing of auditor changes affect audit quality? Evidence from
the initial year of the audit engagement. Journal of Accounting, Auditing & Finance (forthcoming). https://doi.org/10.1177/
0148558X17726241
Chi, S. S., M. Pincus, and S. H. Teoh. 2014. Mispricing of book-tax differences and the trading behavior of short sellers and insiders. The
Accounting Review 89 (2): 511–543. https://doi.org/10.2308/accr-50644
Chu, L., H. Fogel-Yaari, and P. Zhang. 2016. Is an Auditor’s Propensity to Issue Going Concern Audit Reports a Valid Proxy for Audit
Quality? Working paper, University of Toronto.
Coca-Perraillon, M. 2006. Matching with Propensity Scores to Reduce Bias in Observational Studies. NorthEast SAS Users Group
(NESUG). Available at: https://www.lexjansen.com/nesug/nesug06/an/da13.pdf
Danos, P., J. W. Eichenseher, and D. L. Holt. 1989. Specialized knowledge and its communication in auditing. Contemporary Accounting
Research 6 (1): 91–109. https://doi.org/10.1111/j.1911-3846.1989.tb00746.x
Davis, L. R., B. S. Soo, and G. M. Trompeter. 2009. Auditor tenure and the ability to meet or beat earnings forecasts. Contemporary
Accounting Research 26 (2): 517–548. https://doi.org/10.1506/car.26.2.8
DeFond, M., and J. Zhang. 2014. A review of archival auditing research. Journal of Accounting and Economics 58 (2/3): 275–326.
https://doi.org/10.1016/j.jacceco.2014.09.002
DeFond, M., D. H. Erkens, and J. Zhang. 2017a. Do client characteristics really drive the Big N audit quality effect? New evidence from
propensity score matching. Management Science 63 (11): 3628–3649. https://doi.org/10.1287/mnsc.2016.2528
DeFond, M., D. H. Erkens, and J. Zhang. 2017b. The Big N effect persists after matching on client characteristics: A response to
Lawrence, Minutti-Meza, and Zhang 2017. Management Science 63 (11): 3652–3653.
deHaan, E., F. Hodge, and T. Shevlin. 2013. Does voluntary adoption of a clawback provision improve financial reporting quality?
Contemporary Accounting Research 30 (3): 1027–1062. https://doi.org/10.1111/j.1911-3846.2012.01183.x
Dehejia, R. H., and S. Wahba. 2002. Propensity-score matching methods for nonexperimental causal studies. Review of Economics and
Statistics 84 (1): 151–161. https://doi.org/10.1162/003465302317331982
Ericsson, K. A., N. Charness, P. J. Feltovich, and R. R. Hoffman. 2006. The Cambridge Handbook of Expertise and Expert Performance.
Cambridge, U.K.: Cambridge University Press.
Ettredge, M. L., L. Sun, P. Lee, and A. A. Anandarajan. 2008. Is earnings fraud associated with higher deferred tax and/or book minus tax
levels? Auditing: A Journal of Practice & Theory 27 (1): 1–33. https://doi.org/10.2308/aud.2008.27.1.1
Francis, J. R. 2011. A framework for understanding and researching audit quality. Auditing: A Journal of Practice & Theory 30 (2): 125–
152. https://doi.org/10.2308/ajpt-50006
Geiger, M. A., and K. Raghunandan. 2002. Auditor tenure and audit reporting failures. Auditing: A Journal of Practice & Theory 21 (1):
67–78. https://doi.org/10.2308/aud.2002.21.1.67
Gendron, Y., D. J. Cooper, and B. Townley. 2007. The construction of auditing expertise in measuring government performance.
Accounting, Organizations and Society 32 (1/2): 101–129. https://doi.org/10.1016/j.aos.2006.03.005
Givoly, D., C. Hayn, and B. Lourie. 2016. Does the Tail Wag the Dog? Small-Firm-Bias in Capital Market Research. Working paper,
The Pennsylvania State University, University of California, Los Angeles, and University of California, Irvine.
Gul, F. A., S. Y. K. Fung, and B. Jaggi. 2009. Earnings quality: Some evidence on the role of auditor tenure and auditors’ industry
expertise. Journal of Accounting and Economics 47 (3): 265–287. https://doi.org/10.1016/j.jacceco.2009.03.001
Hainmueller, J. 2012. Entropy balancing for causal effects: A multivariate reweighting method to produce balanced samples in
observational studies. Political Analysis 20 (1): 25–46. https://doi.org/10.1093/pan/mpr025
Hainmueller, J., and Y. Xu. 2013. Ebalance: A Stata package for entropy balancing. Journal of Statistical Software 54 (7): 1–18. https://
doi.org/10.18637/jss.v054.i07
Hammersley, J. S. 2006. Pattern identification and industry-specialist auditors. The Accounting Review 81 (2): 309–336. https://doi.org/
10.2308/accr.2006.81.2.309
Hill, J., and J. P. Reiter. 2006. Interval estimation for treatment effects using propensity score matching. Statistics in Medicine 25 (13):
2230–2256. https://doi.org/10.1002/sim.2277
Hou, K., C. Xue, and L. Zhang. 2018. Replicating anomalies. Review of Financial Studies (forthcoming).
Johnson, V. E., I. K. Khurana, and J. K. Reynolds. 2002. Audit-firm tenure and the quality of financial reports. Contemporary Accounting
Research 19 (4): 637–660. https://doi.org/10.1506/LLTH-JXQV-8CEW-8MXD
King, G., and R. Nielsen. 2016. Why Propensity Scores Should Not Be Used for Matching. Working paper, Harvard University and
Massachusetts Institute of Technology.
King, G., R. Nielsen, C. Coberley, J. E. Pope, and A. Wells. 2011. Comparative Effectiveness of Matching Methods for Causal Inference.
Working paper, Harvard University and Healthways, Inc.
Kothari, S. P., A. J. Leone, and C. E. Wasley. 2005. Performance matched discretionary accrual measures. Journal of Accounting and
Economics 39 (1): 163–197. https://doi.org/10.1016/j.jacceco.2004.11.002

Krishnan, G. V. 2003. Does Big 6 auditor industry expertise constrain earnings management? Accounting Horizons 17 (Supplement): 1–
16. https://doi.org/10.2308/acch.2003.17.s-1.1
Lim, C.-Y., and H.-T. Tan. 2008. Non-audit service fees and audit quality: The impact of auditor specialization. Journal of Accounting
Research 46 (1): 199–246. https://doi.org/10.1111/j.1475-679X.2007.00266.x
Low, K.-Y. 2004. The effects of industry specialization on audit risk assessments and audit-planning decisions. The Accounting Review 79
(1): 201–219. https://doi.org/10.2308/accr.2004.79.1.201
McMullin, J. L., and B. Schonberger. 2018. Entropy-Balanced Discretionary Accruals. Working paper, Indiana University and University
of Rochester.
Minutti-Meza, M. 2013. Does auditor industry specialization improve audit quality? Journal of Accounting Research 51 (4): 779–817.
https://doi.org/10.1111/1475-679X.12017
Myers, J. N., L. A. Myers, and T. C. Omer. 2003. Exploring the term of the auditor-client relationship and the quality of earnings: A case
for mandatory auditor rotation? The Accounting Review 78 (3): 779–799. https://doi.org/10.2308/accr.2003.78.3.779
Neal, T. L., and R. R. Riley, Jr. 2004. Auditor industry specialist research design. Auditing: A Journal of Practice & Theory 23 (2): 169–
177. https://doi.org/10.2308/aud.2004.23.2.169
Owhoso, V. E., W. F. Messier, and J. G. Lynch. 2002. Error detection by industry specialized teams during sequential audit review.
Journal of Accounting Research 40 (3): 883–900. https://doi.org/10.1111/1475-679X.00075
Palmrose, Z.-V. 1986. Audit fees and auditor size: Further evidence. Journal of Accounting Research 24 (1): 97–110. https://doi.org/10.
2307/2490806
Phelps, C., R. Heidl, and A. Wadhwa. 2012. Knowledge, networks, and knowledge networks: A review and research agenda. Journal of
Management 38 (4): 1115–1166. https://doi.org/10.1177/0149206311432640
Phillips, J., M. Pincus, and S. O. Rego. 2003. Earnings management: New evidence based on deferred tax expense. The Accounting
Review 78 (2): 491–521. https://doi.org/10.2308/accr.2003.78.2.491
Power, M. 1996. Making things auditable. Accounting, Organizations and Society 21 (2/3): 289–315. https://doi.org/10.1016/0361-
3682(95)00004-6
Reichelt, K. J., and D. Wang. 2010. National and office-specific measures of auditor industry expertise and effects on audit quality.
Journal of Accounting Research 48 (3): 647–686. https://doi.org/10.1111/j.1475-679X.2009.00363.x
Roberts, M. R., and T. M. Whited. 2013. Endogeneity in empirical corporate finance. In Handbook of the Economics of Finance, Volume
2A, edited by Constantinides, G. M., M. Harris, and R. M. Stulz, 493–572. Amsterdam, The Netherlands: Elsevier.
Shipman, J. E., Q. T. Swanquist, and R. L. Whited. 2017. Propensity score matching in accounting research. The Accounting Review 92
(1): 213–244. https://doi.org/10.2308/accr-51449
Solomon, I., M. D. Shields, and O. R. Whittington. 1999. What do industry-specialist auditors know? Journal of Accounting Research 37
(1): 191–208. https://doi.org/10.2307/2491403
Stuart, E. A. 2010. Matching methods for causal inference: A review and a look forward. Statistical Science 25 (1): 1–21. https://doi.org/
10.1214/09-STS313
Taub, S. 2005. Best Buy parts with Ernst & Young. CFO.com (January 5). Available at: http://ww2.cfo.com/accounting-tax/2005/01/best-
buy-parts-with-ernst-young/
Taylor, M. H. 2000. The effects of industry specialization on auditors’ inherent risk assessments and confidence judgments.
Contemporary Accounting Research 17 (4): 693–712. https://doi.org/10.1506/3LDH-AV52-0F4W-H4BB
Teoh, S. H., and T. J. Wong. 1993. Perceived auditor quality and the earnings response coefficient. The Accounting Review 68 (2): 346–
366.
Tsai, W. 2001. Knowledge transfer in intraorganizational networks: Effects of network position and absorptive capacity on business unit
innovation and performance. Academy of Management Journal 44 (5): 996–1004.
Utke, S. 2018. Miscodings in Compustat’s auditor variable: Issues, identification, and correction. Advances in Accounting 43: 56–59.
https://doi.org/10.1016/j.adiac.2018.09.002.
van Wijk, R., J. J. P. Jansen, and M. A. Lyles. 2008. Inter- and intra-organizational knowledge transfer: A meta-analytic review and
assessment of its antecedents and consequences. Journal of Management Studies 45 (4): 830–853. https://doi.org/10.1111/j.1467-
6486.2008.00771.x
APPENDIX A
Variable Definitions and Propensity Score Model
Dependent Variables
AQit represents one of the following three audit quality proxies: ADA, income-increasing DA, or BTD.
ADAit is the absolute value of the residual (e it) from Equation (A1), based on Kothari, Leone, and Wasley (2005), for firm i
in year t. We estimate the regression annually for each industry based on two-digit SIC codes, requiring at least ten

146 Gaver and Utke
observations per industry and excluding firms where the absolute value of total accruals scaled by total assets exceeds
1, following Kothari et al. (2005).
ACit ¼ a þ b0 ð1=ATit1 Þ þ b1 DREVit þ b2 PPEit þ b3 ROAit1 þ eit ðA1Þ
where:
ACit is total accruals for firm i in year t, defined as net income from continuing operations minus operating cash flow
scaled by total assets at the end of year t1;
ATit1 is total assets for firm i at the end of year t1;
DREVit is the change in revenue for firm i at the end of year t scaled by total assets at the end of year t1;
PPEit is net property, plant, and equipment for firm i at the end of year t scaled by total assets at the end of year t1;
and
ROAit1 is net income for firm i in year t1 scaled by total assets at the start of year t1.
DAit is the signed value of the residual from Equation (A1). When using DA as our audit quality proxy in our regressions,
we limit our sample to firms with income-increasing DA (DA . 0).
BTDit is the grossed-up deferred tax expense scaled by pre-tax income (Chi et al. 2014). Deferred tax expense equals
deferred federal tax plus deferred foreign tax. If missing, then deferred tax equals Compustat deferred tax. If still
missing, then deferred tax equals total tax expense minus current tax expense. To gross-up deferred tax expense, we
divide deferred tax expense by the U.S. tax rate, and multiply by 1 minus the rate. We exclude foreign firms, utilities,
financials, and loss firms.
CARit is the size-adjusted three-day abnormal return around the annual earnings announcement (Lim and Tan 2008).
Variables of Interest
SPECIALISTit takes on the value of 1 when an auditor has a market share that is the highest in a given industry and also
more than 10 percent higher than the next-largest competitor during the year, and is 0 otherwise. Each auditor’s market
share is defined as the sum of sales in an industry-year for each auditor, divided by the sum of sales across all auditors
in the industry-year. We measure specialization at the U.S. national level, where U.S. auditors are determined by the
client’s headquarters location. We correct Compustat’s auditor variable as described in Utke (2018).
SEASONEDit takes on the value of 1 when the auditor is classified as a specialist and UNSEASONED is coded 0;
otherwise, SEASONED is 0.
UNSEASONEDit takes on the value of 1 when the auditor is in its first year of being classified as a specialist in a given
industry, and is 0 otherwise.
SEASONED2it takes on the value of 1 when the auditor is classified as a specialist and UNSEASONED2 is coded 0;
otherwise, SEASONED2 is 0.
UNSEASONED2it takes on the value of 1 when the auditor is in its first or second year of being classified as a specialist in a
given industry, and is 0 otherwise.
SEASONED3it takes on the value of 1 when the auditor is classified as a specialist and UNSEASONED3 is coded 0;
otherwise, SEASONED3 is 0.
UNSEASONED3it takes on the value of 1 when the auditor is in its first, second, or third year of being classified as a
specialist in a given industry, and is 0 otherwise.
UNSEASONED1_Dumit takes on the value of 1 when the auditor is in its first year of being classified as a specialist in a
UNSEASONED2_Dumit takes on the value of 1 when the auditor is in its second year of being classified as a specialist in a
UNSEASONED3_Dumit takes on the value of 1 when the auditor is in its third year of being classified as a specialist in a
Control Variables
LOGMKTit is the natural logarithm of the market value of equity for firm i in year t.
LEVit is total debt for firm i in year t divided by average total assets in year t.
ROAit is net income for firm i in year t divided by average total assets in year t.
ROALit is net income for firm i in year t1 divided by average total assets in year t1.
LOSSit takes on the value of 1 if net income is negative for firm i in year t, and is 0 otherwise.
CFOit is operating cash flow for firm i in year t divided by average total assets in year t.

BTMit is the book value of equity divided by the market value of equity for firm i in year t.
ABS_AC_LAGit is the absolute value of total accruals for firm i in year t1 divided by average total assets in year t1.
GROWTHit is sales growth from the prior year for firm i in year t.
ALTMANit is the Altman (1983) financial distress score, as clarified by Altman (2013), for firm i in year t.
STDEARNit is the standard deviation of income before extraordinary items for firm i over the prior four years (t1 to t4).
TENUREit takes on the value of 1 if the client has had the same auditor for five or more years, and is 0 otherwise. Our
TENURE variable is defined slightly differently than in MM. MM sets his tenure variable to 1 if a client has the same
auditor for more than two years, and reports that auditor tenure exceeds two years in 99.3 percent of his sample
observations. Our definition allows for variation in the variable and is more consistent with prior literature (Davis,
Soo, and Trompeter 2009).
YEAR_FEt are year fixed effects.
UEit is the actual year-end earnings minus the most recent median earnings forecast, scaled by stock price two days before
the earnings announcement (Balsam et al. 2003).
VOLATILITYit is the standard deviation of daily stock returns over the 90-day window ending seven days prior to the
earnings announcement (Lim and Tan 2008).
Propensity Score Matching Model

Observations are matched by propensity score, within common support, using a caliper distance of 0.03. In alternative
specifications, we vary the number of treatment to control firms (one-to-one, one-to-two, one-to-three, one-to-five, and one-to-
ten) and match with and without replacement. Following MM, the propensity of choosing a specialist auditor is predicted using
a logistic regression of the auditor’s specialist status on variables related to the client’s level of earnings quality and year and
industry fixed effects (Equation (A2)) as follows:
SPECIALISTit ¼ a0 þ b1 LOGATit þ b2 LEVit þ b3 ROALit þ b4 ROAit þ b5 LOSSit þ b6 CFOit þ b7 BTMit
þ b8 ABS AC LAGit þ b9 GROWTHit þ b10 ALTMANit þ b11 STDEARNit þ b12 TENUREit þ bt YEAR FEt
þ br IND FEi þ mit
ðA2Þ
LOGATit is the natural logarithm of firm i’s total assets in year t. Consistent with MM, we use the natural logarithm of total
assets in the matching model instead of LOGMKTit because LOGATit generally results in better matches.
IND_FEi are two-digit SIC industry fixed effects.


Gaver and Utke (2019)

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

Gaver and Utke (2019)

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Gaver and Utke (2019)

Uploaded by

Copyright:

Available Formats

THE ACCOUNTING REVIEW American Accounting Association

Vol. 94, No. 3 DOI: 10.2308/accr-52206

Audit Quality and Specialist Tenure

The Accounting Review

The Accounting Review

The Accounting Review

III. RESEARCH DESIGN

Regression Model and Variables

The Accounting Review

AQit ¼ a0 þ b1 SPECIALISTit þ b2 LOGMKTit þ b3 LEVit þ b4 ROALit þ b5 ROAit þ b6 LOSSit þ b7 CFOit þ b8 BTMit

AQit ¼ a0 þ b1 SEASONEDit þ b2 UNSEASONEDit þ b3 LOGMKTit þ b4 LEVit þ b5 ROALit þ b6 ROAit þ b7 LOSSit

The Accounting Review

Panel A: Reason for Changes by Year

Reasons for Specialist Changes

The Accounting Review

The Accounting Review

Total for Unmatched Regressions 25,901 13,863 10,551

Total for Propensity Score Matched Regressions 12,153 6,515 5,951

The Accounting Review

The Accounting Review

Descriptive Statistics and Univariate Results

Replication of Minutti-Meza (2013)

The Accounting Review

Panel A: Unmatched Sampleb

The Accounting Review

Panel C: Unmatched Sample Partitioned by Auditor Specializationb

The Accounting Review

The Accounting Review

Panel A: Unmatched Sample Partitioned by Auditor Specializationb

The Accounting Review

The Accounting Review

The Accounting Review

Panel A: Correlation Variables ADA to ROAL

Panel B: Correlation Variables ROA to TENURE

The Accounting Review

Panel A: Absolute Discretionary Accruals

Main Results Alternative Matching Results

Panel B: Absolute Discretionary Accruals (continued)

(continued on next page)

The Accounting Review

Main Results Alternative Matching Results

Panel D: Income-Increasing (Positive) Discretionary Accruals (continued)

(continued on next page)

The Accounting Review

Main Results Alternative Matching Results

Panel F: Book-Tax Differences (continued)

The Accounting Review

Main Results Alternative Matching Results

The Accounting Review

(continued on next page)

The Accounting Review

Main Results Alternative Matching Results

Alternative Definition of Specialist

The Accounting Review

(continued on next page)

The Accounting Review

Main Results Alternative Matching Results

Alternative Audit Quality Measures