Description
We use a unique and confidential database of 15,392 tax audits performed by the Croatian
Tax Administration during the 2002–2006 period to examine the impact of task complexity,
auditor experience, and auditor effort on audit performance. We provide external validation
to prior experimental and analytical research showing that task complexity
decreases while auditor experience and effort increase audit performance. We also extend
this literature by examining the roles of task complexity and experience in moderating the
impact of the effort on audit performance. We find that task complexity mitigates, while
experience enhances the positive relationship between auditor effort and performance.
However, we also find that auditor experience reinforces the positive effect of auditor
effort on performance to a greater degree when complexity is high.
An empirical investigation of the impact of audit and auditor
characteristics on auditor performance
Walid Alissa
a
, Vedran Capkun
a,?
, Thomas Jeanjean
b
, Nadja Suca
c,d
a
HEC Paris, Department of Accounting and Management Control, 1, rue de la liberation, 78351 Jouy-en-Josas, France
b
ESSEC Business School, Department of Accounting and Management Control, Avenue Bernard Hirsch, 95021 Cergy-Pontoise, France
c
Ministry of Finance, Tax Administration, Trg Franje Tudjmana 4, 21000 Split, Croatia
d
University of Split, Faculty of Economics, Matice Hrvatske 31, 21000 Split, Croatia
a b s t r a c t
We use a unique and con?dential database of 15,392 tax audits performed by the Croatian
Tax Administration during the 2002–2006 period to examine the impact of task complex-
ity, auditor experience, and auditor effort on audit performance. We provide external val-
idation to prior experimental and analytical research showing that task complexity
decreases while auditor experience and effort increase audit performance. We also extend
this literature by examining the roles of task complexity and experience in moderating the
impact of the effort on audit performance. We ?nd that task complexity mitigates, while
experience enhances the positive relationship between auditor effort and performance.
However, we also ?nd that auditor experience reinforces the positive effect of auditor
effort on performance to a greater degree when complexity is high. Taken together, our
?ndings provide new evidence on how audit and auditor characteristics impact audit per-
formance, and new insight into how task complexity and auditor experience separately and
jointly moderate the impact of auditor effort on performance.
Ó 2014 Elsevier Ltd. All rights reserved.
Introduction
Audit performance is determined not only by the
inherent complexity of the ?rm or the business unit
audited, but also by audit task and auditor characteristics.
A large body of experimental and theoretical research in
psychology and auditing shows that audit performance
increases with effort (Dye, 1995; Kanfer & Ackerman,
1989; Yeo & Neal, 2004), decreases in task complexity
(Bonner, 1994; Simnett, 1996; Simnett & Trotman, 1989;
Tan, Ng, & Mak, 2002), and increases in auditor experience
(Libby & Frederick, 1990; Lim & Tan, 2010; Simnett, 1996).
In this paper, we offer external validation to these ?ndings
by providing supporting empirical archival evidence. We
also extend this literature by examining the role of task
complexity and experience in moderating the impact of
effort on audit performance.
Due to increasing regulation, the number and complex-
ity of audit tasks is rising, requiring more effort and
knowledge on the part of auditors in internal, ?nancial,
and tax audits alike. This, in turn, puts more pressure on
internal audit departments, audit ?rms, and tax authorities
to understand and manage the design of their audits and
audit teams in order to maximize performance. Given that
audit tasks vary in complexity, and, unavoidably, auditors
vary in knowledge and ability, maximum performance will
be achieved only if the combination of task complexity and
audit team characteristics is such so as to allow for the
highest output per unit of effort (dedicated capacity). Con-
sequently, the optimal allocation of effort becomes one of
the most important drivers of performance, especially in
the environment where the capacity is constrained, suchhttp://dx.doi.org/10.1016/j.aos.2014.06.003
0361-3682/Ó 2014 Elsevier Ltd. All rights reserved.
?
Corresponding author. Tel.: +33 139679611.
E-mail addresses: [email protected] (W. Alissa), [email protected] (V. Capkun),
[email protected] (T. Jeanjean), [email protected] (N. Suca).
Accounting, Organizations and Society 39 (2014) 495–510
Contents lists available at ScienceDirect
Accounting, Organizations and Society
j our nal homepage: www. el sevi er. com/ l ocat e/ aos
as internal audit departments, audit ?rms and tax
authorities.
Past research predominantly uses experiments to
analyze these questions. Experiments provide a high
degree of internal validity (Cook & Campbell, 1976;
Kerlinger, 1986; Libby, Bloom?eld, & Nelson, 2002), but it
is dif?cult to generalize their ?ndings to a real-world
setting (McGrath, 1981; Sacket & Larson, 1990). Contrary
to experiments, archival studies provide the needed
external validity, but come with a lower degree of internal
validity (Scandura & Williams, 2000). This trade-off yields
the need for triangulation of research methods, i.e.
employing both research methods to gain on external
and internal validity (Allee, Bhattacharya, Black, &
Christensen, 2007; Birnberg, 2011; Libby et al., 2002;
McGrath, 1981; Scandura & Williams, 2000). Libby et al.
(2002) and Birnberg (2011) recognize the need for cross-
validation of evidence in ?nancial accounting and behav-
ioral accounting research, dominated by archival and
experimental methods, respectively.
In this study we provide external validation to experi-
mental studies testing the impact of task complexity,
experience, and effort on audit performance. Using a
sample of 15,392 tax audits of 2002 and 2003 corporate
tax ?lings performed by the Croatian Tax Administration
during the 2002–2006 period, we show, consistent with
past experimental evidence, that auditor experience and
effort increase, while task complexity decreases audit
performance.
1
In addition to providing external validation of existing
?ndings, we extend this literature by analyzing the role
of task complexity and auditor experience in the auditor
effort-performance relationship. We ?nd that task com-
plexity mitigates the positive relationship between auditor
effort and audit performance, consistent with the psychol-
ogy literature argument that more complex tasks require
more cognitive effort (Locke & Latham, 1990; Yeo & Neal,
2008) and that there is a diminishing marginal impact of
effort on performance (Kanfer & Ackerman, 1989). We also
?nd that auditor experience enhances the impact of audi-
tor effort on performance. This result is in line with the
argument that effort has a positive impact on performance
only if the individual has some experience with the task
(Yeo & Neal, 2004).
While the ?ndings described above imply that assigning
a highly experienced auditor to a low complexity task
would lead to the highest impact of effort on performance,
this is not the case. Our results indicate that more experi-
ence increases the impact of auditor effort on audit perfor-
mance more when complexity is high. This suggests that
assigning a highly experienced auditor to a more complex
task has a bigger marginal impact on performance than
assigning a highly experienced auditor to a low complexity
task, and consequently maximizes total performance. In
other words, the skills of highly experienced auditors are
wasted on low complexity tasks. These results are
consistent with Tan and Kao (1999) who ?nd that (effort-
inducing) high accountability results in better performance
in highly complex tasks when auditor knowledge and abil-
ity are high. Our results are also consistent with the goal
setting theory arguing that effort and performance depend
on whether the goals are not only attainable, but also chal-
lenging (Fried & Slowik, 2004).
Given that we use archival data, our study differs from
past research in several ways. On the one hand, compared
to an experimental setting, our study has two disadvan-
tages. First, the scope of our analysis is limited by data
availability. For example, while other auditor characteris-
tics, like ability and knowledge, would be useful constructs
to explore, only data on auditor experience is available in
our dataset. Second, our measures are only proxies for con-
structs that can be more accurately measured in an exper-
imental setting. For example, while task complexity can be
manipulated in an experiment, our proxy for task complex-
ity relies on a combination of existing task characteristics
faced by auditors, available in our dataset.
On the other hand, our archival data offers advantages
drawn from the setting we analyze. First, our measures of
auditor experience, effort, and audit performance are those
of a two-person audit team.
2
Given that auditors predomi-
nantly work in teams, this setting enhances the external
validity of our tests. Indeed, in most experimental studies,
auditor characteristics, behavior, and results are measured
and used individually (see Birnberg, 2011). Birnberg (2011)
points at the disadvantages of that approach by arguing that
results from the behavior of individuals cannot be general-
ized to the behavior of those same individuals when they
are acting as a part of a group. Second, we measure audit
performance by the tax adjustment resulting from a tax
audit. Consequently, our study also contributes to the tax
audit literature by further explaining the determinants of
corporate tax audit outcomes. To the best of our knowledge,
previous literature on tax compliance is silent on the effect
of audit and auditor characteristics on tax adjustments. Past
research documents ?rm speci?c factors that in?uence tax
compliance (Chan & Mo, 2000; Mills, 1998; Murray, 1995).
We provide additional insight on audit and audit team char-
acteristics that are associated with tax adjustments.
A related stream of research in accounting uses archival
data to examine the impact of experience and complexity
on performance of analysts (see e.g., Clement, 1999;
Clement, Koonce, & Lopez, 2007; Jacob, Lys, & Neale,
1999). While there are similarities in the two streams of
research (one on analysts and the other on auditors) there
remain important differences between them. Analysts dif-
fer from auditors when it comes to the types of tasks they
perform, their incentives, and the environment they func-
tion in. For example, analyst performance is measured in
this literature as their forecast accuracy. However, given
their role to provide guidance to the market participants,
1
We proxy for audit performance with the amount of tax-adjustments
detected during an audit, scaled by average annual sales. Tax adjustments
can be viewed as a performance outcome in a tax audit since the Croatian
Tax Administration uses tax adjustments to evaluate its own performance,
the performance of its regional of?ces, as well as the performance of its
auditors.
2
As we discuss in the sample and variables construction section, there
are few cases where a team might have more than two auditors. In these
cases, one or both members from the original two-member team could not
continue the mission for whatever reason.
496 W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510
analysts have incentives to and do sacri?ce accuracy for
informativeness (e.g., Louis, Sun, & Urcan, 2013). This could
lead to different predictions and results on the impact of
experience and complexity on performance. In addition,
our study contributes to the literature by providing
evidence on separate and joint moderating effects of com-
plexity and experience on the relationship between effort
and performance, something that has not been explored
in the literature on analysts.
The remainder of the paper proceeds as follows. In the
next section, we present an overview of past literature
and our hypotheses; in Section ‘Institutional background’,
we provide institutional background; in Section ‘Sample
and variable construction’ we present our sample and
provide some descriptive statistics; in Section ‘Research
design’, we describeour researchdesign; inSection‘Results’,
we present our results; and in Section ‘Additional ?ndings
and robustness checks’, we discuss some additional ?ndings
and robustness checks. Section ‘Conclusion’ presents
conclusions.
Literature review and hypotheses development
Empirical archival evidence on the impact of task com-
plexity, auditor effort, and auditor experience, as well as
their interactions on audit performance is rather scant. In
order to build our hypotheses, we thus rely on the theoret-
ical and experimental literature in auditing, management,
and psychology. Given that we provide empirical archival
support to the existing evidence on the direct impact of
auditor experience, effort, and task complexity on audit
performance, we hypothesize only on the moderating role
of auditor experience and task complexity in the auditor
effort-performance relationship (see Fig. 1, Panel A).
Our measure of audit performance is based on
irregularities detected during the audit, similar to mea-
sures based on tasks that subjects perform in experiments
(e.g., Tan & Kao, 1999). Prior research in ?nancial audit
have used number of errors or mistakes caught (e.g.,
Asare & McDaniel, 1996), audit adjustments (e.g.,
Abdolmohammadi & Wright, 1987), and the level of abnor-
mal accruals (e.g., Francis & Yu, 2009; Lennox, Francis, &
Wang, 2014) to measure the performance of an auditor.
Prior research in tax audit have for the most part used
tax audit adjustments to measure tax audit outcomes
(e.g., Chan & Mo, 2000; Mills, 1998; Mills & Sansing,
2000). We follow prior literature and use the tax adjust-
ment detected during the audit as our measure of audit
performance. The advantage of using this measure is that
it is used to evaluate audit performance in the real-world
setting we are analyzing.
The impact of effort on performance has not been stud-
ied extensively in either the accounting or the psychology
literatures. This is attributed to (1) the dif?culty to de?ne
and measure effort as a construct and (2) the lack of pub-
licly available data. Yeo and Neal (2004) argue that effort
is an ‘‘invisible, internal, hypothetical construct that is
not directly observable’’, and consequently dif?cult to
measure and de?ne. Past literature proxies for effort using
either time on task (Caramanis & Lennox, 2008; Christen,
Iyer, & Soberman, 2006; Cloyd, 1997; Cloyd, Pratt, &
Stock, 1996; Fisher & Ford, 1998), or self-reported mea-
sures (Brown & Leigh, 1996). We use time on task as a
measure of auditor effort.
The theoretical and experimental psychology literature
provides arguments and evidence in support of a positive
relationship between effort and performance (for a review
of this literature see, Kanfer & Ackerman, 1989; Weingart,
Fig. 1. Moderators of the relationship between auditor effort and performance.
W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510 497
1992; Yeo & Neal, 2004). Analytical audit research suggests
that achieving a higher audit quality requires an increase
in direct costs to auditors (Dye, 1995; Hillegeist, 1999).
Direct cost is arguably correlated with effort (de?ned as
time on task), as Dye (1995) describes it as the sum ‘‘of
the costs incurred while performing the various tests of
account balances, control tests, analytical reviews, etc.,
described in audit manuals.’’ Hirst (1994) provides experi-
mental evidence that auditors adjust their effort in
response to executives’ incentives to manage earnings.
Caramanis and Lennox (2008) provide empirical archival
evidence in support of this claim.
Our ?rst hypothesis deals with the moderating role of
task complexity in the auditor effort-performance relation-
ship. Prior literature de?nes task complexity as either
objective or subjective (see Bonner, 1994; Campbell,
1988; Maynard & Hakel, 1997). Objective task complexity
is a function of objective task characteristics, while subjec-
tive task complexity is a function of both the task and the
individual’s characteristics. In this paper we analyze the
impact of objective task complexity on audit performance.
Our complexity measure is consistent with Wood’s (1986)
de?nition of complexity. Wood (1986) de?nes complexity
as the sum of information cues to be processed across
the ‘‘number of acts in a subtask’’, and across ‘‘subtasks
in a task’’. Wood (1986) labels this ‘‘Component Complex-
ity’’, which represents a major part in his measure of ‘‘Total
Complexity’’. Applied to our measure of complexity, taxes
are subtasks, and the audit is a task composed of one or
multiple subtasks. In other words, we de?ne one audit as
one task. Consequently, in addition to the type of taxes
audited, the number of taxes serves as a part of the mea-
sure for the complexity of the task.
The limited evidence from experiments yields con?ict-
ing ?ndings onwhether the impact of auditor effort onaudi-
tor performance decreases or increases with an increase in
task complexity. Consistent with the former, in an experi-
ment Mohd-Sanusi and Mohd-Iskandar (2007) ?nd an
increase in audit judgment performance associated with
performance-incentive induced effort only when the task
is less complex, with no such relationship for high complex-
ity tasks. Consistent with the latter, Tan and Kao (1999) ?nd
that (effort-inducing) high accountability results no perfor-
mance improvements in low complex tasks.
Building on Libby (1981, 1985), Bonner (1994) argues
that a change in task characteristics, such as objective task
complexity, can affect auditor judgment, especially when
audit tasks are highly complex. Bonner (1994) further
argues that increases in objective task complexity lead to
decreases in three components of auditor judgment
performance: proper use of knowledge, consistent use
of knowledge, and other direct or unspeci?ed effects.
Other experimental and case study research generally
supports these ?ndings (Asare & McDaniel, 1996; Pratt &
Jiambalvo, 1981; Simnett, 1996; Simnett & Trotman,
1989; Tan et al., 2002).
A more complex task requires more cognitive effort
(Yeo & Neal, 2008) and a higher level of skill (Bonner,
1994). Consequently, per unit of time, a more complex task
will likely result in lower performance (Maynard & Hakel,
1997), as the ability and availability of individuals capable
and willing to exert the necessary cognitive effort will be
limited. Fried and Slowik (2004) argue that the ability of
individuals to successfully pursue multiple complex goals
is limited because they ‘‘heavily tax their cognitive
resources.’’ In addition, a more complex task will typically
require more effort (Locke & Latham, 1990). Given that
there is a diminishing marginal impact of effort on perfor-
mance for every additional unit of effort (Kanfer &
Ackerman, 1989), the relationship between effort and per-
formance should be weaker for high complexity tasks.
Based on the above discussion, we hypothesize as follows
(see Fig. 1, Panel B (a) for a diagram of the hypothesized
relationship):
H1: As task complexity increases, the impact of auditor
effort on audit performance decreases.
Based on the goal setting literature, some might argue
the opposite (i.e., that an increase in task complexity leads
to an increase in the impact of auditor effort on audit per-
formance). This research is based on the premise that more
challenging goals can serve as a motivating factor, espe-
cially when tasks are complex (Campbell, 1988; Maynard
& Hakel, 1997). However, in order for this effect to exist,
goals should be not only challenging but also achievable
(Fried & Slowik, 2004). This would require systematic
and perfect matching between individuals’ capabilities
and demands of the tasks they perform, something that
does not exist in our setting.
Next, we hypothesize on the role of experience in the
effort-performance relationship. Auditor experience has
been studied extensively in the audit literature. These
studies use measures of seniority and years of audit expe-
rience as a proxy for experience (Abdolmohammadi &
Wright, 1987; Lim & Tan, 2010; Simnett, 1996). We use
seniority as a measure of auditor experience.
3
Past research suggests the existence of a positive rela-
tionship between experience and performance. In an
experiment, Abdolmohammadi and Wright (1987) ?nd
that subjects’ experience impacts their audit judgment,
while Simnett (1996) ?nds that auditor experience has a
positive impact on the predictive accuracy of auditors.
Finally, Lim and Tan (2010), ?nd that auditor tenure
improves audit quality. While these studies make use of
experience, others analyze measures correlated with expe-
rience, such as knowledge (Tan & Kao, 1999; Tan et al.,
2002), skill (Bonner, 1991, 1994), and problem-solving
ability (Tan & Kao, 1999). The correlation between these
constructs and experience has been long established in
the literature. For example, studies have shown that expe-
rienced auditors have more and better organized knowl-
edge (Bedard, 1989; Libby & Luft, 1993).
Studies using knowledge, skill, and problem solving
ability all ?nd the same positive impact on audit perfor-
mance (Bonner, 1991, 1994; Tan & Kao, 1999; Tan et al.,
2002). However, Libby and Frederick (1990) show that
experience leads to better performance by auditors
3
In our setting, this dimension is captured through the presence of both
junior auditors and more experienced auditors.
498 W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510
through increases in knowledge. This implies that experi-
ence acts indirectly to increase performance, making
experience an imperfect proxy for knowledge. This biases
against us ?nding a positive relationship between experi-
ence and performance.
On the moderating role of experience, the psychology
literature suggests that experience ampli?es the positive
impact of effort on performance. Yeo and Neal (2004) ?nd
evidence consistent with this relationship and argue that
effort will have no impact on performance if the task is suf-
?ciently novel and the experience of the individual is low.
By contrast, effort will have a positive impact on perfor-
mance if the individual has more prior experience with
the task. Finally, because more experienced auditors spend,
on average, less time in an audit, their marginal return will
be potentially higher.
Contrary to the above arguments, some prior research,
based on resource allocation theory, argues that the rate
of increase in the relationship between effort and perfor-
mance should slow down with practice. Furthermore, the
relationship between effort and performance should
decrease and eventually become insigni?cant (Kanfer &
Ackerman, 1989). If the task has constant information pro-
cessing demands, it will eventually become automated
with increase in practice. However, in a (tax) audit setting,
the tasks involve inconsistent information processing
demands and are highly complex (Bonner, 1994), making
it less likely for these tasks to become automated. Conse-
quently, in our setting, it is unlikely that the relationship
between effort and performance will become insigni?cant
as experience increases.
Based on the above discussion, we hypothesize as fol-
lows (see Fig. 1, Panel B (b) for a diagram of the hypothe-
sized relationship):
H2: As auditor experience increases, the impact of auditor
effort on audit performance increases.
Finally, we hypothesize on the combined effects of com-
plexity and experience on the relationship between effort
and audit performance. A simple extension of H1 and H2
implies that the combination of high experience and low
complexity should yield the highest performance per unit
of effort invested. However, the relationship is more com-
plicated than this. According to Maynard and Hakel (1997),
both cognitive ability and experience will have a greater
impact on performance when the task is more complex.
To put it differently, a more experienced auditor’s knowl-
edge and ability would be wasted on a simple task. This
is consistent with the goal setting literature that argues
that both effort and performance will depend on whether
the goals are challenging and attainable (Fried & Slowik,
2004). Similarly, Tan and Kao (1999) argue and ?nd that
(effort-inducing) high accountability (also a proxy for
motivation) results in better performance in highly com-
plex tasks when both auditor knowledge and ability are
high. Consistent with the belief that experience has a posi-
tive impact on the effort-performance relationship, we
argue that the effort of a more experienced auditor will
have a stronger impact on performance when the task is
more complex. Therefore, we hypothesize as follows:
H3: Experienced auditor’s effort has a greater impact on
performance when task complexity is high.
Institutional background
To test our predictions, we use a unique and con?den-
tial dataset provided by the Croatian Tax Administration.
Republic of Croatia is a developing European country with
a code law legal system (German origin), a low level of
legal enforcement, a low level of investor protection, and
a bank-based economy.
4
The Croatian Tax Administration
is an administrative organization within the Ministry of
Finance. It is divided into 20 regional of?ces located in 20
Croatian counties. The central of?ce is located in Zagreb
(the capital city). Each regional of?ce has one or more tax
audit divisions.
5
Regional of?ces perform tax audits of all
?rms in their territory, regardless of whether they are incor-
porated in their county. Once a regional of?ce begins a tax
audit, even if the ?rm is incorporated in another county, it
will keep the case until its ?nal resolution.
The Croatian tax system consists of national taxes,
county taxes, city or municipal taxes, joint taxes, and taxes
on games of chance. Most ?rms are subject to value added
tax (VAT) and corporate income tax. Private individuals
pay personal income tax with the portion related to their
salaries being collected monthly by ?rms and paid directly
into the country budget.
The Tax Administration employed on average 778 tax
auditors in 2002/2003. Typically, the Tax Administration
employs business school or law school graduates. Future
tax auditors are trained on the job and are ?rst assigned,
for a one- or two-year period, to the Tax Assessment and
Contributions Division of their regional of?ce (in charge
of receiving and processing tax ?lings and payments). They
are subsequently transferred to the Tax Audit Division of
their regional of?ce. Tax auditors in principle should not
hold a signi?cant interest in any Croatian ?rm; if they do,
the control rights must be transferred to a third party. To
avoid con?icts of interest, auditors are not allowed to audit
?rms owned or governed by relatives or any person with
whom they have close ties. To avoid wrongdoing by any
of the parties in the tax audit process, auditors are not
allowed to perform audits alone. Tax auditors have broad
authority that includes full access to all documents and
facilities of the audited ?rm as well as the authority to
close facilities and con?scate documents and goods.
The regional of?ce, independently or at the request of
the central of?ce, initiates a tax audit. The decision to ini-
tiate a tax audit is made by the head of the regional of?ce
and/or the head of the tax audit division of the regional
of?ce. This is done either as part of a regular audit plan
established at the beginning of each year or on an ad hoc
4
Mills (1998) argues that ?rms cannot costlessly maximize ?nancial
reporting bene?ts and minimize taxes at the same time. Our sample
consists of almost only private ?rms, with lower or no incentives to
maximize ?nancial reporting bene?ts (Cloyd et al., 1996). Croatian ?rms
produce ?nancial statements primarily for tax purposes and to a limited
extent for communication with lenders.
5http://www.porezna-uprava.hr
W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510 499
basis, following a request. From within the tax administra-
tion, two cases of ad hoc audit initiations are possible: (1)
the tax assessment and contributions division requests a
tax audit of a ?rm, or (2) a tax auditor, during an audit,
requests that the audit be extended to one or more other
?rms. From outside the tax administration, two cases of
ad hoc audit initiations are possible: (1) an (anonymous)
whistleblower reports tax fraud, or (2) other administra-
tions of the Ministry of Finance or other ministries of the
Republic or foreign governments request a tax audit. Note
that from our conversations with the tax auditors we infer
that the number of ad hoc audit initiations is ‘‘insigni?-
cant’’. Additionally, tax auditors do not (systematically)
have access to the information about the type of audit
initiation unless necessary for conducting the audit. This
alleviates the concern that in our sample tax auditors use
audit initiation type to generate expectations about the
tax adjustment, which would raise concerns about reverse
causality.
Sample and variable construction
Sample
The con?dential database provided by the Tax Adminis-
tration includes data on all 80,719 tax audits executed in
the Republic of Croatia in the 2002–2006 period, as well
as all annual corporate income tax ?lings (140,135 for
2002 and 135,749 for 2003) and VAT ?lings (71,781 for
2002 and 73,853 for 2003). To ensure con?dentiality and
anonymity, both the ?rms audited and the auditors have
been anonymized.
In order to perform our analysis, we need a ?rm to be
subject to both income tax and VAT, and to have ?led
income tax and VAT declarations in 2002 and/or 2003. This
reduces our sample of corporate tax ?lings to 67,812
observations.
6
Note that Croatian ?rms are not required to
keep detailed accounting and tax data beyond six years after
the tax ?ling for all taxes except personal income tax. There-
fore, analyzing audits performed in the 2002–2006 period
guarantees that we capture most audits for tax ?lings in ?s-
cal years 2002 and 2003.
7
In addition, because tax ?lings are
available for years 2002 and 2003 only, we restrict our sam-
ple to only those tax audits that include the entire years or
parts of 2002 and/or 2003. This reduces the number of
observations to 16,125 audits. We exclude all tax audits that
were performed more than once, as they were coded as a
single audit, which could bias our ?ndings (579 audits).
Finally, we exclude observations for which all data necessary
to perform our analysis are not available (154 audits). Our
?nal sample consists of 15,392 tax audits.
Income tax ?lings contain data on the ?rm’s industry,
place of incorporation, sales, expenses, accounting net
income, and detailed reconciliation between accounting
and taxable income. The corporate income tax rate was
35% in both 2002 and 2003 regardless of the level of tax-
able income. VAT ?lings contain data on domestic and for-
eign sales, sales subject to VAT, purchases giving right to
VAT reimbursement, imports, and the VAT due. The VAT
rate was 22% in both 2002 and 2003. Tax audit data
includes duration (number of hours worked by the audit
team), period audited, type of audit (primarily type of
taxes audited), number of tax auditors on the case, and
tax adjustments per tax and per year.
Table 1
Distribution of audits by year and industry and audited ?rm characteristics.
Fiscal year audited
Audit year 2002 % 2003 % All %
Panel A: Distribution of audits by audit year and ?scal year
2002 5359 66.61 5 0.06 5362 34.85
2003 1889 23.48 5703 64.55 6663 43.31
2004 628 7.81 2378 26.92 2601 16.90
2005 118 1.47 613 6.94 624 4.06
2006 51 0.63 136 1.54 142 0.92
Total 8045 100.00 8835 100.00 15,392 100.00
Industry Number of audits %
Panel B: Distribution of audits by industry
Agriculture 519 3.37
Fishing 49 0.32
Mining 49 0.32
Manufacturing 2843 18.47
Electrical energy, gas 99 0.64
Construction 844 5.48
Retail and wholesale 6548 42.54
Hotels and restaurants 1837 11.93
Transport 815 5.29
Financial services 51 0.33
Real estate 966 6.28
Education 35 0.23
Health services 30 0.19
Household services 707 4.59
Total 15,392 100
In thousands of HRK Sales BTD (%) ForeignSales
Panel C: Firm ?nancial characteristics
N 15,392 15,392 15,392
Mean 167.00 0.17 0.27
S.D. 1048.96 0.42 0.44
Min 0.00 À0.02 0.00
p25 0.81 0.00 0.00
Median 3.50 0.00 0.00
p75 20.55 0.07 1.00
Max 14351.83 1.66 1.00
This table presents distribution of audits by year and industry and audi-
ted ?rm characteristics. The sample contains 15,392 audits of 2002 and
2003 tax records performed by the Croatian Tax Administration in the
2002–2006 period (9950 unique ?rms). All data are provided by the Tax
Administration of the Republic of Croatia. Panel A shows the distribution
of audits by audit year and ?scal year audited. Panel B shows the distri-
bution of audits by industry. Panel C shows ?nancial characteristics of
?rms audited. Audit Year is the year in which tax audit began. Fiscal Year
Audited is the ?scal year of the tax records audited. Industry is the Cro-
atian national industry classi?cation. Sales is the mean of 2002 and 2003
annual sales of ?rms audited. BTD is the mean of the 2002 and 2003
annual book-tax difference of ?rms audited. ForeignSales is a binary var-
iable that equals one if the audited ?rm had foreign sales in either 2002 or
2003.
6
Not all ?rms are subject to VAT, and they become subject to VAT after
their revenues exceed a threshold. Banks and insurance companies are not
subject to VAT but are subject to income tax, and individuals may become
subject to VAT and are seldom subject to corporate income tax.
7
Excluding from our sample audits performed in 2005 and 2006, to
avoid selection concerns associated with later audits, also yields statisti-
cally signi?cant results in all our tests.
500 W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510
Table 1 provides the distribution of audits by year and
industry, as well as ?rm characteristics. Panel A of Table 1
presents the distribution of the years the audit started (i.e.,
2002, 2003, . . ., or 2006) and the ?scal year audited (i.e.,
2002, or 2003). Most audits were performed in 2003 and
2002 (43.31% and 34.85%, respectively). Out of all the
2002 and 2003 ?scal year ?lings audits, 16.90% were con-
ducted in 2004. The number of audits for the 2002 and
2003 ?lings naturally declines over the 2002–2006 period,
as the Tax Administration rarely audits tax ?lings older
than ?ve years and dedicates its capacity to more recent
tax ?lings.
8
In addition to auditing past years’ annual tax ?l-
ings, the Tax Administration systematically audits same-
year ?lings. These audits typically include a combination
of VAT monthly ?lings, personal income tax monthly ?lings,
and one-day audits of sales (revenue audits). The Tax
Administration audited a similar number of ?rms’ ?lings
for 2002 and 2003 (8045 and 8835, respectively).
Panel B of Table 1 presents the distribution of audited
?rms by industry. The largest number of audits was per-
formed in the retail and wholesale industries (6548, or
42.54%), followed by manufacturing (2843, or 18.47%)
and hotels and restaurants (1837, or 11.93%). This re?ects
the overall Croatian economy, which is based primarily
on services and tourism.
9
Panel C of Table 1 presents the relevant ?nancial infor-
mation for the sample of audited ?rms. To proxy for the
size of the audited ?rm, we use FirmSize, measured as the
natural log of the mean of 2002 and 2003 annual sales.
Averaging the sales limits the impact both of sales varia-
tions (which can be signi?cant in our sample, as it consists
of mainly small and medium enterprises) and of new ?rms
entering the market.
10
In cases of absence of 2002 or 2003
sales data, we use the data from the year for which it is
available. The mean ?rm size (Sales) of the ?rms in our sam-
ple is HRK167.00 million (median of HRK3.50 million) rang-
ing from less than HRK10,000 to more than HRK14 billion.
11
In the same vein as sales, we also average the book-tax dif-
ference (BTD), scaled by sales, of 2002 and 2003. The mean
BTD represents 16.61% of sales, with a median of 0.00%. It
ranges from À1.72% to 165.85% of sales. In our sample,
26.60% (4094) of the audited ?rms have foreign sales. We
use an indicator variable to proxy for foreign sales,
ForeignSales, where it equals one if the ?rm reports foreign
sales in either 2002 or 2003, and zero otherwise.
Variable construction
Table 2 presents the construction of our variables of
interest: task complexity, effort, and experience. Panel A
of Table 2 presents the distribution of audits by the type
and number of taxes audited (N:Taxes) of taxes audited.
We categorize taxes audited by the Tax Administration
into ?ve categories: VAT, corporate income tax, personal
income tax, revenue, and other. An audit of VAT can
include one or several monthly and/or annual VAT ?lings.
An audit of corporate income tax can include one or several
annual ?lings because income tax declarations are ?led
annually. An audit of personal income tax includes an audit
of personal income tax, social charges, and related local
taxes (all salary-related items). They can be audited either
based on monthly ?lings or based on annual ?lings, the lat-
ter case being prevalent. An audit of revenue is a special
type of audit that can have an impact on several taxes. Rev-
enue audits typically last for a day but can be extended to a
longer time period. Not declaring revenues can have an
impact on both VAT and corporate income tax. Finally,
‘‘other’’ audits can include longer revenue audits, audits
of speci?c transactions (typically at the request of the Cus-
toms Administration or the Ministry of Interior), audits
extended to other ?rms, and audits of other taxes (e.g.,
gaming taxes).
Most audits in our sample involve auditing one type of
the above taxes (10,172 audits, or 66.09%), followed by
audits of three types of taxes (2548, or 16.55%), and then
audits of two types of taxes (2398, or 15.58%). VAT was
audited in 56.01% of cases, followed by revenue (35.69%)
and personal income tax (23.92%). The table also highlights
the fact that audits involving one or more taxes primarily
include VAT and/or revenue, while those with three or
more taxes almost always include VAT, corporate income
tax, and personal income tax. We asked several tax audi-
tors about the complexity of the audits by category.
According to all interviewed tax auditors, the most com-
plex audit is what they refer to as a ‘‘complete’’ audit
including VAT, corporate income tax, and personal income
tax. We thus code audits including at least VAT, corporate
income tax, and personal income tax as complex audits, and
other audits as simple.
12
Note that, mechanically, a complex
audit implies period audited of one year or more, given that
it includes auditing corporate income tax ?lings, ?led on an
annual basis. Out of the total number of audits (15,392),
1995 (12.96%) are complex and 13,397 (87.04%) are simple
audits. Our complexity measure is a combination of qualita-
tive (type of taxes) and quantitative (number of taxes being
three or more) characteristics, consistent with the Wood
(1986) de?nition of complexity.
The Tax Administration can decide to audit from one
day to several years of accounting and tax records. We
label this audit period TaskSize. Panel B of Table 2 indi-
cates that TaskSize ranges from 1 day to 2544 days (more
than seven years), with a mean of 236 days and a median
of 31 days.
13
Complex audits typically exceed one year of
audit period (median equals 546 days), whereas simple
audits typically include audit periods of less than a year
8
Note that ?ve audits started in 2002 but the year audited was 2003.
Three of those ?ve audits started on December 31, 2002, while the two
remaining cases include both the audit for 2002 and 2003.
9
See, for example, the 2005 Statistical Information, Republic of Croatia
Central Bureau of Statistics.
10
Our results are not sensitive to this choice of size proxy.
11
The exchange rate for the Croatian Kuna (HRK) on January 1, 2003, was
7.45 HRK per Euro and 7.11 per US$, respectively, according to the Croatian
Central Bank.
12
Our results are not sensitive to this de?nition of complex audits. They
remain statistically signi?cant if we de?ne complex audits as those that
include at least VAT and corporate income tax, or those that include at least
personal income tax and corporate income tax.
13
Note that both TaskSize and Effort (de?ned in the coming paragraphs)
are used in our analysis as logged variables. They are presented in the table
in ‘days’ for descriptive purposes.
W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510 501
(median equals 30 days). For brevity, we do not tabulate
this data.
Croatian tax audits are performed by two-person
audit teams.
14
We use information on their seniority to
proxy for auditor experience, Experience. We de?ne Experi-
ence as a binary variable that equals one if none of the
auditors in an audit team is a junior auditor, and zero
otherwise. To distinguish between audit teams with a
junior auditor and those without a junior auditor we turn
to tax administration records. There are two possible audit
records. If the tax administration recorded two auditors
performing an audit, we treat this audit as being per-
formed by two senior auditors. If, however, the record
shows only one auditor performing the audit, we treat
the audit as being performed by one junior and one senior
auditor. The reason is as follows: if the records indicate
only one auditor performing the audit, the second auditor
is a junior auditor who at the moment of performing the
audit does not have an auditor identi?cation number,
and cannot be recorded in the system as having performed
the audit.
15
In our setting, there are no audits performed
by two junior auditors.
In Croatia, the Tax Administration records the number
of hours worked per audit and per auditor. We use the total
number of audit hours worked per audit by auditors, con-
verted into days, as a proxy for auditor effort, Effort, consis-
tent with prior literature (Caramanis & Lennox, 2008).
Given that the hours of a junior auditor are not recorded
in the system, and given that there is virtually no differ-
ence between the ?rst and the second auditor in time
Table 2
Audit and auditor characteristics.
Tax type N.Taxes Total
1 2 3 4 5
Panel A: Distribution of audits by number of taxes and tax type
All audits 10,172 2398 2548 265 9 15,392
VAT 4048 1758 2541 265 9 8621
(39.80%) (73.31%) (99.73%) (100.00%) (100.00%) (56.01%)
Corporate income tax 92 491 1875 239 9 2706
(0.90%) (20.48%) (73.59%) (90.19%) (100.00%) (17.58%)
Personal income tax 898 574 1943 258 9 3682
(8.83%) (23.94%) (76.26%) (97.36%) (100.00%) (23.92%)
Revenue 3691 1071 591 132 9 5494
(36.29%) (44.66%) (23.19%) (49.81%) (100.00%) (35.69%)
Other 1443 902 694 166 9 3214
(14.19%) (37.61%) (27.24%) (62.64%) (100.00%) (20.88%)
Complex tasks 0 0 1754 232 9 1995
(0.00%) (0.00%) (68.84%) (87.55%) (100.00%) (12.96%)
Simple tasks 10,172 2398 794 33 0 13,391
(100.00%) (100.00%) (31.16%) (12.45%) (0.00%) (87.00%)
TaskSize (Days) Effort (Days) Experience BigOf?ce
Panel B: Other audit and auditor characteristics
N 15,392 15,392 15,392 15,392
Mean 236.24 11.85 0.72 0.27
S.D. 316.49 18.84 0.45 0.44
Min 1.00 0.25 0.00 0.00
p25 1.00 1.25 0.00 0.00
Median 31.00 5.00 1.00 0.00
p75 365.00 15.50 1.00 1.00
Max 2544.00 385.50 1.00 1.00
This table presents audit and auditor characteristics for the sample observations. The sample contains 15,392 audits of 2002 and 2003 tax record audits
performed by the Croatian Tax Administration in the 2002–2006 period. All data are provided by the Tax Administration of the Republic of Croatia. N:Taxes
is the number of taxes covered in an audit. Complex is a binary variable that equals one if an audit includes at least VAT, corporate income tax, and personal
income tax; and zero otherwise. Simple audits are all other audits. The percentages in Panel A represent the number of types for a given audit relative to the
total number of all audits in the given column. TaskSize is the number of days of tax records that were audited (i.e., period audited). Effort is the duration of
an audit in auditor days, taking into account all auditors involved. Experience is a binary variable that equals zero if one of the auditors is a junior auditor,
and one otherwise. BigOf?ce is a binary variable that equals one if the regional of?ce is Split or Zagreb, and zero otherwise.
14
While it is possible to have more than two auditors performing an
audit, this is uncommon. In our sample, the number of auditors per audit
ranges from two to eight (untabulated), with the vast majority of audits
(14,866, or 97%) having two auditors. The three percent of audits for which
more than two auditors were present are mostly cases with three auditors
(406, or 2.64% of all cases). These cases are mainly those for which one of
the two initial auditors was replaced during the audit. Consequently, we
include these cases in our analysis. Excluding these cases from our analysis
or controlling for the number of auditors does not change our inferences.
15
We are treating the experience between two auditors in a team and
across experienced audit teams as ?xed. Even further, based in our design
choice, we could have the case where the combined years of experience in
an experienced audit team to be less than or equal a non-experienced audit
team. In those cases, we are less likely to detect differences between audit
performances between the teams, a bias that will work against us.
502 W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510
spent on the task (an untabulated mean and median of
zero), in order to adjust Effort for the presence of the junior
auditor, we assign to a junior auditor the same number of
days assigned to the senior auditor in the system. Effort
ranges from 0.25 days to 385.50 days, with a mean of
11.85 days and a median of 5 days. Out of our total number
of audits, 72% (11,079 cases) are audits performed by more
experienced auditors (senior auditors) and 28% (4313
cases) are audits performed by teams where one of the
two auditors is a junior auditor.
The Tax Administration operates from twenty regional
of?ces that vary in terms of number of auditors, number
of audits performed, area covered, and size of ?rms in their
county. To explore the differences in size of the regional
of?ces, we isolate the two largest of?ces and label them
as a BigOf?ce (Francis & Yu, 2009). We single out Split
and Zagreb as big regional of?ces for three reasons. First,
the Split regional of?ce is the one with the largest number
of audits performed in our sample (2878 audits, or 18.70%
of the total) and covers the largest geographical area. Sec-
ond, being the capital city of?ce, the Zagreb regional of?ce
covers the largest number of ?rms and the largest number
of large ?rms, followed by the Split regional of?ce. Third,
the number of tax auditors is the highest in Zagreb (203
in 2002 and 191 in 2003), followed by Split (112 in 2002
and 100 in 2003). About one-quarter of all the audits in
our sample (4080) are performed by the two big audit
of?ces.
We use the natural log of tax adjustment scaled by
average annual sales detected during the audit as our mea-
sure of audit performance because the Croatian tax admin-
istration evaluates its auditors, regional of?ces, and its
overall performance based on the amounts of tax adjust-
ments detected during audits. The tax adjustment is scaled
by sales to account for size differences across ?rms. The tax
adjustment is integrated in the internal reporting and per-
formance measurement system of the Tax Administration.
From interviews with Croatian tax auditors we can infer
that pressure is exercised on auditors to increase tax
adjustments, as reports of tax adjustments are drawn per
auditor systematically. This is used by the central of?ce
to evaluate regional of?ces’ performance.
We present the descriptive statistics concerning the
audit outcomes in Table 3. Panel A of Table 3 presents
the percentage of audits ‘with’ and ‘without’ a tax adjust-
ment. About seventy-four percent (11,403 audits) of audits
resulted in no tax adjustments, whereas the remaining
3983 audits resulted in tax adjustments. Panel B of Table 3
presents tax adjustments scaled by sales (winsorized at
2.5%) and the log of tax adjustment scaled by sales (TaxAdj)
that we use as our dependent variable in the remainder of
the study. Note that tax adjustments are never negative,
i.e. the Tax Administration never recorded a case where
taxes were reduced for a ?rm. The mean tax adjustment
represents 15.69% of sales (a median of 1.31%) and ranges
from 0.01% to 242.38%. Table 4 presents the correlations
between the variables used in the analysis. The table
indicates that TaxAdj is positively correlated with Effort
and Experience and negatively correlated with Complexity,
consistent with our predictions. The table also indicates
that more experienced auditors exert less effort and more
complex tasks require more effort. An interesting correla-
tion, however, is the negative correlation between Experi-
ence and Complexity, which suggests that more senior
auditors do not automatically get assigned to high complex
audit cases.
Research design
To test our hypotheses, we estimate an OLS regression
model with TaxAdj, our measure of audit performance as
dependent variable, and measures of task complexity,
auditor effort, and auditor experience, as well as other
known control variables fromthe literature as independent
variables.
We estimate the following regression models:
TaxAdj
it
¼ a
0
þa
1
Complexity þa
2
Effort
þa
3
Experience þa
4
Complexity ÂEffort
þb
1
FirmSize þb
2
BTD þb
3
ForeignSales
þb
4
BigOffice þb
5
MillsLambda þdIndustry
þdAudit Year þe ð1Þ
TaxAdj
it
¼ a
0
þa
1
Complexity þa
2
Effort
þa
3
Experience þa
5
Experience ÂEffort
þb
1
FirmSize þb
2
BTD þb
3
ForeignSales
þb
4
BigOffice þb
5
MillsLambda þdIndustry
þdAudit Year þe ð2Þ
where Complexity is a binary variable that equals one if the
audit includes at least VAT, corporate income tax, and
Table 3
Audit outcome.
Number of audits %
Panel A: Distribution of audits by audit outcome
No tax adjustment detected 11,409 74.12
Tax adjustment detected 3983 25.88
Total 15,392 100
Tax adjustment
(% of sales,)
TaxAdj
Panel B: Tax adjustments
N 3983 3983
Mean 15.69 À4.21
S.D. 44.58 2.60
Min 0.01 À20.26
p25 0.29 À5.84
Median 1.31 À4.33
p75 7.20 À2.63
Max 242.38 11.55
This table presents audit outcome descriptive statistics. The sample
contains 15,392 audits of 2002 and 2003 tax records performed by the
Croatian Tax Administration in the 2002–2006 period. All data are pro-
vided by the Tax Administration of the Republic of Croatia. No tax
adjustment detected indicates that no tax adjustment was detected in the
course of the audit. Tax adjustment detected indicates there was a tax
adjustment detected in the course of the audit. Tax Adjustment in Panel B
is the tax adjustment, for the 3983 audits that had an adjustment, scaled
by the mean of 2002 and 2003 annual sales. TaxAdj is the natural log of
the tax adjustment scaled by the mean of 2002 and 2003 annual sales.
W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510 503
personal income tax; and zero otherwise.
16
Effort is the
natural log of the total number of days worked by all
auditors on the case. Experience is a binary variable that
equals zero if one of the auditors was a junior auditor, and
one otherwise. Consistent with H1, we expect the coef?cient
associated with Complexity ÂEffort (a
4
in Model 1) to be
negative and statistically signi?cant. With respect to H2,
we expect the coef?cient associated with the ExperienceÂ
Effort (a
5
in Model 2) to be positive and statistically
signi?cant.
In addition to our variables of interest, we include a
number of control variables in our regression models that
are identi?ed in the prior literature to be correlated with
TaxAdj and that could be correlated with our variables of
interest. We include FirmSize; BTD; ForeignSales (Mills,
1998), BigOf?ce (Francis & Yu, 2009), Industry, and Audit
Year dummies as control variables in our regression
models. Caramanis and Lennox (2008) show ?rm size to
be positively correlated with audit length. We thus control
for ?rm size by including the natural log of the average
sales of 2002 and 2003, FirmSize. Consistent with Mills
and Sansing (2000), we expect FirmSize to be negatively
correlated with TaxAdj. Mills (1998) shows book-tax differ-
ences to be positively correlated with tax adjustment. BTD
equals book-income minus tax-income, scaled by sales.
Consistent with Mills (1998), we expect BTD to be posi-
tively correlated with TaxAdj.
Firms with foreign operations will have more items to
audit, which could imply longer audits (i.e., more effort).
Chan and Mo (2000) ?nd a negative association, while
Mills (1998) and Mills and Sansing (2000) ?nd either no
association or a positive association between export-
oriented ?rms and tax adjustment. To control for this
effect, we include an indicator variable (ForeignSales) in
our regression model that takes the value of one if the ?rm
had foreign sales in either 2002 or 2003, and zero other-
wise, but we do not form expectations on the association
between this variable and TaxAdj.
Tax auditors in larger of?ces are more likely to detect
material problems because a large of?ce has more inspec-
tors, and these inspectors have exposure to more ?rms,
which leads to greater collective human capital in the
of?ce. Caramanis and Lennox (2008) and Francis and Yu
(2009) show that big audit ?rms and of?ces are associated
with better audit performance. To control for the size of
the of?ce, we introduce a variable, BigOf?ce, which equals
one if the regional of?ce is Split or Zagreb, and zero other-
wise. We expect BigOf?ce to be positively correlated with
TaxAdj.
From interviews with tax auditors and from prior liter-
ature, we cannot form directional predictions on if and
how Experience or Complexity may be correlated with any
of our control variables. The reasoning is as follows: the
Tax Administration does not follow a system to determine
which ?rms to audit; the decision to audit a ?rm is made
by 20 regional of?ces independently. In addition, some
audits are performed following the ?ling of anonymous
reports, or at the request of the Customs Administration
or the Ministry of Interior. Nevertheless, we control for
the above-cited variables to account for potential biases
that could still exist and because past literature shows
these variables to be correlated with auditor effort.
Because our measure of audit performance is the tax
adjustment, we must take into account a potential
selection bias in our sample, as only cases for which tax
adjustments (3983 audits) exist are included in our sam-
ple. Similar to Mills (1998), we use a two-stage Heckman
(1979) procedure that takes into account the probability
of tax adjustments being detected. In the ?rst stage, we
estimate the probability of an adjustment to be proposed
by the tax authorities and, using the parameters from this
Table 4
Correlations.
TaxAdj Complexity Effort Experience FirmSize BTD ForeignSales BigOf?ce
TaxAdj 1
Complexity À0:065
???
1
Effort 0.240
???
0.133
???
1
Experience 0.070
???
À0:078
???
À0:086
???
1
Size À0:515
???
0.082
???
0.252
???
À0:008 1
BTD 0.224
???
À0.087
???
À0.048
???
0.018 À0:265
???
1
ForeignSales À0:219
???
0.031
??
0.113
???
0.029
?
0.354
???
À0:101
???
1
BigOf?ce 0.115
???
À0:098
???
0.096
???
0.277
???
0.044
???
0.004 0.019 1
This table presents the correlations between the variables used in the analysis. The sample contains 3983 audits with tax adjustment detected from 2002 to
2003 tax records performed by the Croatian Tax Administration in the 2002–2006 period. All data are provided by the Tax Administration of the Republic of
Croatia. TaxAdj is the natural log of the tax adjustment scaled by the mean of 2002 and 2003 annual sales. Complexityis a binary variable that equals one if an
audit includes at least VAT, corporate income tax, and personal income tax; and zero otherwise. Effort is the natural log of the duration of an audit
(measured in auditor-days). Experience is a binary variable equal to zero if one of the auditors is a junior auditor, and one otherwise. FirmSize is the natural
log of the mean of 2002 and 2003 annual sales of ?rms audited. BTD is the mean of the 2002 and 2003 annual book-tax difference of ?rms audited.
ForeignSales is a binary variable equal to one if the audited ?rm had foreign sales in either 2002 or 2003, and zero otherwise. BigOf?ce is a binary variable
equal to one if the regional of?ce is Split or Zagreb, and zero otherwise.
?
Two tailed statistical signi?cance at the 10 percent level.
??
Two tailed statistical signi?cance at the 5 percent level.
???
Two tailed statistical signi?cance at the 1 percent level.
16
In a different analysis, we replace Complexity with dummy variables for
each of the tax types to take into account the possibility that individual tax
characteristics are driving our results. Our results remain statistically
signi?cant.
504 W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510
model, we compute the inverse Mills ratio for all observa-
tions. In the second stage, we include the inverse Mills
ratio as an additional control variable and estimate the
amount of the tax adjustment.
17
The dependent variable in the ?rst stage is coded one if
a tax adjustment exists, and zero otherwise. It is a function
of TaskSize (i.e., period audited), TaskType (a VAT dummy),
N:Taxes, Effort, Experience; FirmSize; BTD; ForeignSales,
BigOf?ce, Loss; Industry, and AuditYear indicator variables.
Note that following Lennox et al. (2012), several variables
in the ?rst stage regression model are not included in
our second stage models. Also note that we use
TaskSize; TaskType, and N:Taxes instead of the Complexity
variable in the ?rst stage, as we aim to better specify the
model, while testing our hypotheses in the second stage
model, thus using a single proxy for audit complexity.
Using Complexity in the ?rst stage regression model instead
of the three above-mentioned variables does not change
our inferences.
Finally, to test our third hypothesis, H3, we estimate the
following regression model:
TaxAdj
it
¼ a
0
þa
1
Experience þa
2
Effort
HTC
þa
3
Experience ÂEffort
HTC
þa
4
Effort
LTC
þa
5
Experience ÂEffort
LTC
þþa
6
FirmSize
þa
7
BTD þa
8
ForeignSales þa
9
BigOffice
þa
10
MillsLambda þdIndustry
þdAudit Year þe ð3Þ
First, we split the sample into two subsamples based on
Effort, Effort when the task is complex (Effort
HTC
) and Effort
when the task is simple (Effort
LTC
). In other words, Effort
HTC
equals Effort when Complexity equals one, and zero
otherwise, and Effort
LTC
equals Effort when Complexity
equals zero, and zero otherwise. Then, we interact both
of these new variables with Experience. This yields two
interactions: (1) (Experience ÂEffort
HTC
), which represents
the impact of auditor experience on the auditor effort-tax
adjustment relationship when task complexity is high,
and (2) (Experience ÂEffort
LTC
), which represents the
impact of auditor experience on the auditor effort-tax
adjustment relationship when task complexity is low. To
compare the relative importance of experience on the
effort-task complexity relationship, when complexity is
high vs. when complexity is low, we test the difference
between the coef?cient associated with (ExperienceÂ
Effort
HTC
) and the coef?cient associated with (ExperienceÂ
Effort
LTC
). Consistent with our Hypothesis (3) we expect
this difference to be positive, indicating that higher auditor
experience increases the positive impact of auditor effort
on audit performance more when task complexity is high.
Results
In Table 5, column 1, we present a version of Model 1
with main effects only. Consistent with prior research,
the coef?cient associated with Complexity is negative and
statistically signi?cant, while the coef?cients associated
with Effort and Experience are positive and statistically sig-
ni?cant. Holding everything else constant, a more complex
audit will yield a tax adjustment that is 23.66% lower than
a less complex audit.
18
Holding everything else constant, a
10% increase in auditor effort (approximately one day at
the mean and half a day at the median) increases the tax
adjustment by 16.99%. An audit involving a highly experi-
enced auditor yields a tax adjustment that is 81.12% higher
compared to an audit involving an inexperienced auditor,
holding everything else constant. All of our control variables
have the predicted signs, consistent with prior literature.
FirmSize is negatively correlated with TaxAdj, while BTD
and BigOf?ce are positively correlated with TaxAdj. Foreign-
Sales are negatively correlated with TaxAdj, consistent with
Chan and Mo (2000).
Column 2 of Table 5 presents the results from running
Model 1, where we test whether task complexity reduces
the impact of auditor effort on audit performance (H1).
Our results are consistent with this hypothesis. The coef?-
cient associated with Effort is positive and statistically
signi?cant, while the coef?cient associated with the Com-
plexity  Effort interaction is negative and statistically sig-
ni?cant. The sum of the two coef?cients is positive and
statistically signi?cant. For a 10% increase in auditor effort,
the tax adjustment increases by 17.79% when Complexity is
low, and by 16.02% when Complexity is high. This translates
into an impact of Effort on TaxAdj that is 11.06% (17.79% vs.
16.02%) stronger when Complexity is low. All of our control
variables have the predicted signs.
Column 3 of Table 5 presents the results of whether
auditor experience increases the impact of auditor effort
on audit performance (Model 2, H2). Both the coef?cient
associated with Effort, and the coef?cient associated with
the Experience  Effort interaction are positive and statisti-
cally signi?cant, validating H2. The impact of auditor expe-
rience on the auditor effort-tax adjustment relationship is
economically signi?cant. In cases with low auditor experi-
ence, a 10% increase in auditor effort will yield a 15.95%
increase in tax adjustment, whereas in cases with high
auditor experience, it yields an 18.27% increase in tax
adjustment. This translates into a 14.55% stronger impact
of auditor effort on tax adjustment when auditor experi-
ence is high. All control variables have the predicted signs
and are statistically signi?cant.
Finally, in Table 6 we show estimates of Model 3, which
tests whether the effort of a more experienced auditor is
17
Our study could also suffer from a second potential selection bias, as
?rms may not be randomly audited by the tax authorities. To take into
account this possibility, as a robustness check, we run a probit model that
explains the probability of being audited as a function of the ?rm size,
book-tax differences, existence of foreign operations, auditors belonging to
a big of?ce, reporting losses, and industry and year ?xed effects. From this
model, we incorporate a second inverse mills ratio in our main analysis. Our
results (untabulated) remain statistically signi?cant, and this second
inverse mills ratio is not statistically signi?cant in all our models, raising
questions about the existence of this selection bias.
18
To calculate the percentage change in the dependent variable when the
independent variable is an indicator variable, we use [exp (coef?cient)-1].
When the independent variable is continuous and logged, we use
[1.10
ˆ
(coef?cient)-1] to obtain a percentage change in the dependent
variable for a 10% change in the independent variable.
W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510 505
more bene?cial when exerted on a more complex
audit (H3). Both interactions (Experience ÂEffort
HTC
and
Experience ÂEffort
LTC
) are positive and signi?cant. The
incremental effect of high auditor experience when task
complexity is high is 2.40% (16.03% of the impact of
Effort
HTC
), compared to the 1.70% (11.35% of the impact of
Effort
LTC
) when task complexity is low. The difference
between the coef?cients associated with the two interac-
tions is statistically signi?cant. As in the previous regres-
sion models, all control variables have the predicted
signs and are statistically signi?cant.
In Figs. 2 and 3 we plot our results. In Fig. 2, Panel A, we
illustrate our results from Table 5, column 2. We plot the
values of audit performance (TaxAdj) as a function of
auditor effort (Effort) in complex versus simple tasks
(Complexity). The plot shows higher audit performance in
a simple task than in a complex task. The slopes of the lines
representing simple and complex tasks also show that an
increase in auditor effort increases audit performance
more (less) in a simple (complex) task, consistent with
our ?rst hypothesis (H1).
In Panel B of Fig. 2 we plot the relationship between
Effort and TaxAdj by using the estimates from Table 5, col-
umn 3, i.e. for auditors with high versus low experience
(Experience). The plot shows higher audit performance
when auditors have more experience. It also shows that
audit effort increases performance more (less) when audi-
tors have more (less) experience, consistent with our sec-
ond hypothesis (H2).
In Fig. 3 we compare the differential impact of effort for
auditors with a high versus low experience (Experience x
Effort) on performance (TaxAdj) in a complex versus simple
task (Complexity). This ?gure is based on Table 6 estimates.
Consistent with our third hypothesis (H3), in a complex
task, the effort of a more experienced auditor has a higher
impact on performance than in a simple task.
Additional ?ndings and robustness checks
Additional ?ndings
In untabulated results, we estimate our regression
model (1), but interact the Complexity and Experience
variables. Our results show that auditor experience
mitigates the negative impact of task complexity on audit
performance. The coef?cient associated with Complexity
is negative and statistically signi?cant, whereas the one
associated with the Complexity  Experience interaction is
positive and statistically signi?cant. The sum of the two
coef?cients is negative and statistically signi?cant. A com-
plex task will lower tax adjustment by 38.43% compared to
a simple task when auditor experience is low, and by
15.04% when auditor experience is high. This translates
into a negative impact of task complexity being only one
third as large when experience is high. In general, all prior
studies ?nd that auditor skill, knowledge, and problem-
solving ability mitigate the impact of task complexity on
audit performance (Abdolmohammadi & Wright, 1987;
Bonner, 1994; Chang, Ho, & Liao, 1997; Tan & Kao, 1999;
Tan et al., 2002). Consistent with these prior studies, we
?nd that experience, as a related construct (see Section ‘Lit-
erature review and hypotheses development’), mitigates
the negative relation between task complexity and audit
performance.
Our results also indicate that experience has a bigger
positive impact on audit performance when a task is
complex than when it is simple. The coef?cient associated
with the Experience variable, the coef?cient of ComplexityÂ
Experience and the sum of the two coef?cients are all
positive and statistically signi?cant. A more experienced
auditor will perform 62.09% better (have a 62.09% higher
Table 5
The impact of tax complexity, auditor effort, and auditor experience on tax
adjustment.
(1) (2) (3)
Constant 1.666
???
1.439
???
2.011
???
(3.759) (3.111) (4.377)
Complexity À0.27
???
0.24 À0.27
???
(4.023) (0.807) (3.949)
Effort 1.646
???
1.718
???
1.553
???
(14.562) (14.252) (13.202)
Experience 0.594
???
0.611
???
À0.043
(7.891) (8.017) (0.194)
Complexity ÂEffort À0.159
?
(1.753)
Experience ÂEffort 0.208
???
(3.009)
FirmSize À0.825
???
À0.828
???
À0.832
???
(43.457) (43.191) (43.361)
BTD 0.478
???
0.480
???
0.469
???
(7.223) (7.196) (7.036)
ForeignSales À0.268
???
À0.273
???
À0.279
???
(3.661) (3.701) (3.784)
BigOf?ce 0.540
???
0.538
???
0.564
???
(7.787) (7.704) (8.042)
MillsLambda 0.946
???
1.032
???
1.025
???
(4.680) (4.960) (5.033)
Industry FE Included Included Included
Audit year FE Included Included Included
N 15,392 15,392 15,392
Uncensored N 3983 3983 3983
Wald v
2
3666.59 3625.58 3640.78
Prob. > v
2
0.00 0.00 0.00
This table presents results from estimating an OLS model for tax adjust-
ments. The sample contains 15,392 audits of 2002 and 2003 tax records
performed by the Croatian Tax Administration in the 2002–2006 period.
All data are provided by the Tax Administration of the Republic of Croatia.
The dependent variable, TaxAdj, is the natural log of the tax adjustment
scaled by the average of sales in the years 2002 and 2003 (see text).
Complexityis a binary variable that equals one if an audit includes at least
VAT, corporate income tax, and personal income tax; and zero otherwise.
Effort is the natural log of the duration of an audit (measured in auditor-
days). Experience is a binary variable equal to zero if one of the auditors is
a junior auditor, and one otherwise. FirmSize is the natural log of the mean
of 2002 and 2003 annual sales of ?rms audited. BTD is the mean of the
2002 and 2003 annual book-tax difference of ?rms audited. ForeignSales is
a binary variable equal to one if the audited ?rm had foreign sales in
either 2002 or 2003, and zero otherwise. BigOf?ce is a binary variable
equal to one if the regional of?ce is Split or Zagreb, and zero otherwise.
MillsLambda is the inverse Mills ratio resulting from the ?rst stage
regression with a binary variable equal to one if irregularities were
detected and zero otherwise as the dependent variable, and the inde-
pendent variables from Model (1) plus a TaskSize; TaskType (a VAT
dummy), and N:Taxes in place of Complexity, and a Loss dummy variable
that equals one if the ?rm had a loss and zero otherwise.
?
Two tailed statistical signi?cance at the 10 percent level, respectively.
**
Two tailed statistical signi?cance at the 5 percent level.
???
Two tailed statistical signi?cance at the 1 percent level.
506 W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510
tax adjustment) than a less experienced auditor when a
task is simple, and 123.67% better when a task is complex.
Robustness checks
Endogeneity
Our unique database and setting allow us to draw
conclusions from the archival data without threats of
endogeneity between audit performance and audit/auditor
characteristics. In our setting, seniority of auditors, task
complexity, and auditor effort are not decided based on
the expected tax adjustment. Auditors are assigned to
tasks based on their availability, in an effort to maximize
the use of the Tax Administration’s capacity. Task complex-
ity is decided based on the constrained capacity, the need
to use capacity throughout the year, and the need to audit
different taxes across multiple periods at any time during
the year. An example of this would be an audit of a
monthly VAT ?ling compared to an audit of several taxes
based on an annual tax ?ling. Auditor effort is decided by
the auditor, and not the supervisor choosing the ?rm to
be audited. Finally, the same ?rms are not likely to be
audited systematically every year, and even if this is the
case, the same taxes are not likely to be audited every
year.
19
The preceding arguments suggest that the Tax
Administration has little information to form expectations
on the amount of tax adjustment. Note that the Croatian
Tax Administration does not use a formal model to select
?rms to audit.
That being said, absent other information, the auditors
may still form expectations about the potential tax adjust-
ments based on ?rm’s tax ?lings, which would lead to the
endogeneity (reverse causality) problem, as they would
adjust their effort as a function of the expected tax adjust-
ment. Therefore, we searched for a setting where auditors
are unlikely to form expectation with respect to our
dependent variable. We use the book-tax-difference (the
difference between book income and taxable income) as
a proxy for tax adjustment expectations.
20
Mills (1998)
?nds that book-tax-differences are the main driver of and
positively correlated with the tax adjustments detected
during a tax audit. We conduct a test in which we divide
our sample into ?rms with positive book-tax-difference
amounts (those more likely to have a higher tax adjustment)
and ?rms with no or negative book-tax-difference amounts
(those less likely to have a tax adjustment). In the sample of
?rms with no or negative book-tax-difference amounts our
results are unchanged compared to our main results. When
comparing the two subsamples, results are similar. The
untabulated ?ndings of this test suggest that our main
results are unlikely to be driven by auditors forming expec-
tations about tax adjustments.
Alternative measure for Complexity
To take into account the possibility that our complexity
measure captures something other than complexity, we
run two robustness checks where we isolate only the dif-
ferences in characteristics of the task. First, we estimate
our regression models with N:Taxes (the number of taxes
audited) and/or TaskSize (period audited) as additional
control variables. Controlling for N:Taxes and/or TaskSize
keeps the number of taxes and the period audited constant,
leaving only differences in tax types in our Complexity
Table 6
Joint moderating role of task complexity and experience on auditor effort-
performance relationship.
TaxAdj
Constant 1.959
???
(4.29)
Experience À0.020
(0.09)
Effort
HTC
1.465
???
(12.023)
Experience ÂEffort
HTC
0.249
???
(3.418)
Effort
LTC
1.597
???
(13.738)
Experience ÂEffort
LTC
0.177
??
(2.492)
FirmSize À0.834
???
(43.605)
BTD 0.471
???
(7.063)
ForeignSales À0.275
???
(3.733)
BigOf?ce 0.554
???
(7.887)
MillsLambda 1.041
???
(5.176)
Industry FE Included
Audit Year FE Included
N 15,392
Uncensored N 3983
Wald v
2
3641.59
Prob. > v
2
0.00
This table presents results from estimating an OLS model for tax adjust-
ments. The sample contains 15,392 audits of 2002 and 2003 tax records
performed by the Croatian Tax Administration in the 2002–2006 period.
All data are provided by the Tax Administration of the Republic of Croatia.
The dependent variable, TaxAdj, is the natural log of the tax adjustment
scaled by the average of sales in the years 2002 and 2003 (see text). Effort
is the natural log of the duration of an audit (measured in auditor-days).
Effort
HTC
equals Effort when Complexity equals one, and zero otherwise.
Effort
LTC
equals Effort when Complexity equals zero, and zero otherwise.
Complexity is a binary variable that equals one if an audit includes at least
VAT, corporate income tax, and personal income tax; and zero otherwise
Experience is a binary variable that equals zero if one of the auditors is a
junior auditor, and one otherwise. FirmSize is the natural log of the mean
of 2002 and 2003 annual sales of ?rms audited. BTD is the mean of the
2002 and 2003 annual book-tax difference of ?rms audited. ForeignSales is
a binary variable that equals one if the audited ?rm had foreign sales in
either 2002 or 2003, and zero otherwise. BigOf?ce is a binary variable that
equals one if the regional of?ce is Split or Zagreb, and zero otherwise.
MillsLambda is the inverse Mills ratio resulting from the ?rst stage
regression with a binary variable equal to one if irregularities were
detected and zero otherwise as the dependent variable, and the inde-
pendent variables from Model (1) plus TaskSize; TaskType (a VAT dummy),
and N:Taxes in place of the Complexity and Loss dummy variable that
equals one if the ?rm had a loss and zero otherwise.
?
Two tailed statistical signi?cance at the 10 percent levels, respectively.
??
Two tailed statistical signi?cance at the 5 percent level.
???
Two tailed statistical signi?cance at the 1 percent level.
19
These arguments are based on informal conversations we have had
with the Croatian tax auditors.
20
Another potential proxy for the expectation of the amount of tax
adjustment would be prior year tax adjustments. However, in our setting,
?rms are not systematically audited every year and even if this occurs, the
audits are not likely to cover the same taxes.
W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510 507
0
5
1
0
1
5
2
0
P
e
r
f
o
r
m
a
n
c
e
0 2 4 6 8 10
Effort
Simple Task Complex Task
Panel A: Task Complexity
0
5
1
0
1
5
2
0
P
e
r
f
o
r
m
a
n
c
e
0 2 4 6 8 10
Effort
Low Experience High Experience
Panel B: Auditor Experience
Fig. 2. Task complexity and auditor experience impact on the effort-performance relationship.
2
2
.
5
3
3
.
5
4
4
.
5
P
e
r
f
o
r
m
a
n
c
e
0 2 4 6 8 10
Experience x Effort
Complex Task Simple Task
Fig. 3. Task complexity and auditor experience joint impact on the effort-performance relationship.
508 W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510
variable. Our results remain statistically signi?cant. Sec-
ond, we limit the sample to only those audits with one
tax audited. We asked tax auditors if they had one tax to
audit which one would be the most complex. Their answer
was Corporate Income Tax, followed by VAT. Given that
our reduced sample of 10,172 observations contains only
92 Corporate Income Tax observations (45 in the sample
with tax adjustments), we use VAT as a proxy for a com-
plex task. Our results also remain statistically signi?cant.
Conclusion
Using a sample of 15,392 tax audits performed on 2002
and 2003 corporate tax ?lings performed by the Croatian
Tax Administration during the 2002–2006 period, we
provide external validation of the prior experimental liter-
ature results that auditor experience and effort increase,
while task complexity decreases audit performance. We
extend this literature by examining the impact of task
complexity and auditor experience on the relationship
between auditor effort and performance, proxied for by
the amount of tax adjustments scaled by average annual
sales. Based on arguments from the psychology literature
suggesting that more complex tasks require more cogni-
tive effort (Locke & Latham, 1990; Yeo & Neal, 2008), and
that there is a diminishing marginal impact of effort on
performance (Kanfer & Ackerman, 1989), we ?nd that
audit task complexity mitigates the positive relationship
between auditor effort and audit performance. We also
?nd that auditor experience enhances the impact of audi-
tor effort on performance, in line with the argument that
effort has a positive impact on performance only if the
individual has some experience with the task (Yeo &
Neal, 2004). Finally, we show that high experience
increases the impact of auditor effort on audit performance
more when complexity is high, consistent with the goal
setting theory arguing that effort and performance will
depend on whether the goals are challenging and attain-
able (Fried & Slowik, 2004).
Compared to prior (theoretical and experimental)
literature, our study offers advantages, but also has certain
limitations. Our study provides a real-life setting, offering
external validation, allowing us to analyze the perfor-
mance and the characteristics of an audit team, as opposed
to an individual as it is done in most experimental studies
Birnberg (2011). Due to the real-life setting, we are
however limited by the choice of variables, as well as the
quality of the proxies used. As we measure audit perfor-
mance by the tax adjustment resulting from a tax audit,
our study also contributes to the tax audit literature by
further explaining the determinants of a corporate tax
audit outcome. We provide additional insight on audit
and audit team characteristics that are associated with
tax adjustments.
Acknowledgements
The authors gratefully acknowledge ?nancial support
from HEC Paris and Fondation HEC, Investissements
d’Avenir (ANR-11-IDEX-0003/Labex Ecodec/ANR-11-
LABX-0047). Walid Alissa and Vedran Capkun are members
of GREGHEC, Unit CNRS UMR 2959. The authors are grate-
ful for the data provided by the Tax Administration of Min-
istry of Finance of Republic of Croatia. We are grateful for
the comments received at the University of Innsbruck, IE
Business School, and IPAG Business School as well as at
the 2012 American Accounting Association annual meet-
ing. We are especially thankful to Florian Hoos, Clive Len-
nox, Cedric Lesage, Martin Messner and Steve Salterio for
their valuable input.
References
Abdolmohammadi, M., & Wright, A. (1987). An examination of the effects
of experience and task complexity on audit judgments. The Accounting
Review, 62(1), 1–13.
Allee, K., Bhattacharya, N., Black, E., & Christensen, T. (2007). Pro forma
disclosure and investor sophistication: External validation of
experimental evidence using archival data. Accounting, Organizations
and Society, 32, 201–222.
Asare, S., & McDaniel, L. (1996). The effects of familiarity with the
preparer and task complexity on the effectiveness of the audit review
process. The Accounting Review, 71(2), 139–159.
Bedard, J. (1989). Expertise in auditing: Myth or reality? Accounting,
Organizations and Society, 14(1–2), 113–131.
Birnberg, J. (2011). A proposed framework for behavioral accounting
research. Behavioral Research in Accounting, 23, 1–43.
Bonner, S. (1991). Is experience necessary in cue measurement? The case
of auditing tasks. Contemporary Accounting Research, 8(1), 253–269.
Bonner, S. (1994). A model of the effects of audit task complexity.
Accounting, Organizations and Society, 19(3), 213–234.
Brown, S., & Leigh, T. (1996). A new look at psychological climate and its
relationship to job involvement, effort, and performance. Journal of
Applied Psychology, 81(4), 358–368.
Campbell, D. (1988). Task complexity: A review and analysis. The Academy
of Management Review, 13(1), 40–52.
Caramanis, C., & Lennox, C. (2008). Audit effort and earnings
management. Journal of Accounting and Economics, 45(1), 116–138.
Chang, J., Ho, J., & Liao, W. (1997). The effects of justi?cation, task
complexity and experience/training on problem-solving performance.
Behavior Research in Accounting Supplement, 98–116.
Chan, K., & Mo, P. (2000). Tax holidays and tax noncompliance: An
empirical study of corporate tax audits in china’s developing
economy. The Accounting Review, 75(4), 469–484.
Christen, M., Iyer, G., & Soberman, D. (2006). Job satisfaction, job
performance, and effort: A reexamination using agency theory.
Journal of Marketing, 70(1), 137–150.
Clement, M. (1999). Analyst forecast accuracy: Do ability, resources, and
portfolio complexity matter? Journal of Accounting and Economics, 27,
285–303.
Clement, M., Koonce, L., & Lopez, T. (2007). The role of task-speci?c
forecasting experience and innate ability in understanding analyst
forecasting performance. Journal of Accounting and Economics, 44(3),
378–398.
Cloyd, C. (1997). Performance in tax research tax: The joint effect of
knowledge and accountability. The Accounting Review, 72, 111–131.
Cloyd, B., Pratt, J., & Stock, T. (1996). The use of ?nancial accounting
choice to support aggressive tax positions: Public and private ?rms.
Journal of Accounting Research, 34(1), 23–43.
Cook, T., & Campbell, D. (1976). The design and conduct of quasi-
experiments and true experiments in ?eld settings. Handbook of
industrial and organisational psychology. Chicago, IL: Rand-McNally.
Dye, R. (1995). Incorporation and the audit market. Journal of Accounting
and Economics, 19(1), 75–114.
Fisher, S., & Ford, J. (1998). Differential effects of learner effort and goal
orientation on two learning outcomes. Personnel Psychology, 51(2),
397–420.
Francis, J., & Yu, M. (2009). Big 4 of?ce size and audit quality. The
Accounting Review, 84(5), 1521–1552.
Fried, Y., & Slowik, L. (2004). Enriching goal-setting theory with time: An
integrated approach. The Academy of Management Review, 29(3),
404–422.
Heckman, J. (1979). Sample selection bias as a speci?cation error.
Econometrica, 47(1), 153–161.
Hillegeist, S. (1999). Financial reporting and auditing under alternative
damage apportionment rules. The Accounting Review, 74(3), 347–369.
W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510 509
Hirst, D. E. (1994). Auditor sensitivity to earnings management.
Contemporary Accounting Research, 11(1), 405–422.
Jacob, J., Lys, T., & Neale, M. (1999). Expertise in forecasting performance
of security analysts. Journal of Accounting and Economics, 28(1), 51–82.
Kanfer, R., & Ackerman, P. (1989). Motivation and cognitive abilities: An
integrative/aptitude-treatment interaction approach to skill
acquisition. Journal of Applied Psychology, 74(4), 657–690.
Kerlinger, F. (1986). Foundations of behavioural research. New York, NY:
Holt, Rinehart, and Winston.
Lennox, C. S., Francis, J. R., & Wang, Z. (2012). Selection models in
accounting research. The Accounting Review, 87(2), 589–616.
Libby, R. (1981). Accounting and human information processing: Theory and
applications. Contemporary topics in accounting series. Englewood
Cliffs, NJ: Prentice-Hall.
Libby, R. (1985). Availability and the generation of hypotheses in
analytical review. Journal of Accounting Research, 23(2), 648–667.
Libby, R., Bloom?eld, R., & Nelson, M. (2002). Experimental research in
?nancial accounting. Accounting, Organizations and Society, 27,
775–810.
Libby, R., & Frederick, D. (1990). Experience and the ability to explain
audit ?ndings. Journal of Accounting Research, 28(2), 348–367.
Libby, R., & Luft, J. (1993). Determinants of judgement performance in
accounting setting: Ability, knowledge, motivation, and environment.
Accounting, Organizations and Society, 18(5), 425–450.
Lim, C., & Tan, H. (2010). Does auditor tenure improve audit quality?
Moderating effects of industry specialization and fee dependence.
Contemporary Accounting Research, 27, 923–957.
Locke, E., & Latham, G. (1990). A theory of goal setting and task
performance. New York, NY: Prentice Hall.
Louis, H., Sun, A., & Urcan, O. (2013). Do analysts sacri?ce forecast
accuracy for informativeness? Management Science, 59, 1688–1708.
Maynard, D., & Hakel, M. (1997). Effects of objective and subjective task
complexity on performance. Human Performance, 10(4), 303–330.
McGrath, J. (1981). Dilemmatics: The study of research choices and
dilemmas. American Behavioral Scientist, 25(2), 179–210.
Mills, L. (1998). Book-tax differences and internal revenue service
adjustments. Journal of Accounting Research, 36(2), 343–356.
Mills, L., & Sansing, R. (2000). Strategic tax and ?nancial reporting
decisions: Theory and evidence. Contemporary Accounting Research,
17(1), 85–106.
Mohd-Sanusi, Z., & Mohd-Iskandar, T. (2007). Audit judgment
performance: Assessing the effect of performance incentives, effort
and task complexity. Managerial Auditing Journal, 22(1), 34–52.
Murray, M. (1995). Sales tax compliance and audit selection. National Tax
Journal, 48, 515–530.
Pratt, J., & Jiambalvo, J. (1981). Relationship between leader behaviors and
audit team performance. Accounting, Organizations and Society, 6,
133–142.
Sacket, P., & Larson, J. (1990). Research strategies and tactics in industrial
and organisational psychology. In Hough Dunette (Ed.), Handbook of
industrial and organisational psychology. Palo Alto, CA: Consulting
Psychologist Press.
Scandura, T., & Williams, E. (2000). Research methodology in
management: Current practices, trends, and implications for future
research. Academy of Management Journal, 43, 1248–1264.
Simnett, R. (1996). The effect of information selection, information
processing and task complexity on predictive accuracy of auditors.
Accounting, Organizations and Society, 21(7), 699–719.
Simnett, R., & Trotman, K. (1989). Auditor versus model: Information
choice and information processing. The Accounting Review, 64(3),
514–528.
Tan, H., & Kao, A. (1999). Accountability effects on auditors’ performance:
The in?uence of knowledge, problem-solving ability, and task
complexity. Journal of Accounting Research, 37(1), 209–223.
Tan, H., Ng, T., & Mak, B. (2002). The effects of task complexity on
auditors’ performance: The impact of accountability and knowledge.
Auditing: A journal of Practice & Theory, 21(2), 81–95.
Weingart, L. (1992). Impact of group goals, task component complexity,
effort, and planning on group performance. Journal of Applied
Psychology, 77(5), 682–693.
Wood, R. (1986). Task complexity: De?nition of the construct.
Organizational Behavior and Human Decision Processes, 37, 60–82.
Yeo, G., & Neal, A. (2004). A multilevel analysis of effort, practice, and
performance: Effects; of ability, conscientiousness, and goal
orientation. Journal of Applied Psychology, 89(2), 231.
Yeo, G., & Neal, A. (2008). Subjective cognitive effort: A model of states,
traits, and time. Journal of Applied Psychology, 93(3), 617–631.
510 W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510
doc_395447565.pdf
We use a unique and confidential database of 15,392 tax audits performed by the Croatian
Tax Administration during the 2002–2006 period to examine the impact of task complexity,
auditor experience, and auditor effort on audit performance. We provide external validation
to prior experimental and analytical research showing that task complexity
decreases while auditor experience and effort increase audit performance. We also extend
this literature by examining the roles of task complexity and experience in moderating the
impact of the effort on audit performance. We find that task complexity mitigates, while
experience enhances the positive relationship between auditor effort and performance.
However, we also find that auditor experience reinforces the positive effect of auditor
effort on performance to a greater degree when complexity is high.
An empirical investigation of the impact of audit and auditor
characteristics on auditor performance
Walid Alissa
a
, Vedran Capkun
a,?
, Thomas Jeanjean
b
, Nadja Suca
c,d
a
HEC Paris, Department of Accounting and Management Control, 1, rue de la liberation, 78351 Jouy-en-Josas, France
b
ESSEC Business School, Department of Accounting and Management Control, Avenue Bernard Hirsch, 95021 Cergy-Pontoise, France
c
Ministry of Finance, Tax Administration, Trg Franje Tudjmana 4, 21000 Split, Croatia
d
University of Split, Faculty of Economics, Matice Hrvatske 31, 21000 Split, Croatia
a b s t r a c t
We use a unique and con?dential database of 15,392 tax audits performed by the Croatian
Tax Administration during the 2002–2006 period to examine the impact of task complex-
ity, auditor experience, and auditor effort on audit performance. We provide external val-
idation to prior experimental and analytical research showing that task complexity
decreases while auditor experience and effort increase audit performance. We also extend
this literature by examining the roles of task complexity and experience in moderating the
impact of the effort on audit performance. We ?nd that task complexity mitigates, while
experience enhances the positive relationship between auditor effort and performance.
However, we also ?nd that auditor experience reinforces the positive effect of auditor
effort on performance to a greater degree when complexity is high. Taken together, our
?ndings provide new evidence on how audit and auditor characteristics impact audit per-
formance, and new insight into how task complexity and auditor experience separately and
jointly moderate the impact of auditor effort on performance.
Ó 2014 Elsevier Ltd. All rights reserved.
Introduction
Audit performance is determined not only by the
inherent complexity of the ?rm or the business unit
audited, but also by audit task and auditor characteristics.
A large body of experimental and theoretical research in
psychology and auditing shows that audit performance
increases with effort (Dye, 1995; Kanfer & Ackerman,
1989; Yeo & Neal, 2004), decreases in task complexity
(Bonner, 1994; Simnett, 1996; Simnett & Trotman, 1989;
Tan, Ng, & Mak, 2002), and increases in auditor experience
(Libby & Frederick, 1990; Lim & Tan, 2010; Simnett, 1996).
In this paper, we offer external validation to these ?ndings
by providing supporting empirical archival evidence. We
also extend this literature by examining the role of task
complexity and experience in moderating the impact of
effort on audit performance.
Due to increasing regulation, the number and complex-
ity of audit tasks is rising, requiring more effort and
knowledge on the part of auditors in internal, ?nancial,
and tax audits alike. This, in turn, puts more pressure on
internal audit departments, audit ?rms, and tax authorities
to understand and manage the design of their audits and
audit teams in order to maximize performance. Given that
audit tasks vary in complexity, and, unavoidably, auditors
vary in knowledge and ability, maximum performance will
be achieved only if the combination of task complexity and
audit team characteristics is such so as to allow for the
highest output per unit of effort (dedicated capacity). Con-
sequently, the optimal allocation of effort becomes one of
the most important drivers of performance, especially in
the environment where the capacity is constrained, suchhttp://dx.doi.org/10.1016/j.aos.2014.06.003
0361-3682/Ó 2014 Elsevier Ltd. All rights reserved.
?
Corresponding author. Tel.: +33 139679611.
E-mail addresses: [email protected] (W. Alissa), [email protected] (V. Capkun),
[email protected] (T. Jeanjean), [email protected] (N. Suca).
Accounting, Organizations and Society 39 (2014) 495–510
Contents lists available at ScienceDirect
Accounting, Organizations and Society
j our nal homepage: www. el sevi er. com/ l ocat e/ aos
as internal audit departments, audit ?rms and tax
authorities.
Past research predominantly uses experiments to
analyze these questions. Experiments provide a high
degree of internal validity (Cook & Campbell, 1976;
Kerlinger, 1986; Libby, Bloom?eld, & Nelson, 2002), but it
is dif?cult to generalize their ?ndings to a real-world
setting (McGrath, 1981; Sacket & Larson, 1990). Contrary
to experiments, archival studies provide the needed
external validity, but come with a lower degree of internal
validity (Scandura & Williams, 2000). This trade-off yields
the need for triangulation of research methods, i.e.
employing both research methods to gain on external
and internal validity (Allee, Bhattacharya, Black, &
Christensen, 2007; Birnberg, 2011; Libby et al., 2002;
McGrath, 1981; Scandura & Williams, 2000). Libby et al.
(2002) and Birnberg (2011) recognize the need for cross-
validation of evidence in ?nancial accounting and behav-
ioral accounting research, dominated by archival and
experimental methods, respectively.
In this study we provide external validation to experi-
mental studies testing the impact of task complexity,
experience, and effort on audit performance. Using a
sample of 15,392 tax audits of 2002 and 2003 corporate
tax ?lings performed by the Croatian Tax Administration
during the 2002–2006 period, we show, consistent with
past experimental evidence, that auditor experience and
effort increase, while task complexity decreases audit
performance.
1
In addition to providing external validation of existing
?ndings, we extend this literature by analyzing the role
of task complexity and auditor experience in the auditor
effort-performance relationship. We ?nd that task com-
plexity mitigates the positive relationship between auditor
effort and audit performance, consistent with the psychol-
ogy literature argument that more complex tasks require
more cognitive effort (Locke & Latham, 1990; Yeo & Neal,
2008) and that there is a diminishing marginal impact of
effort on performance (Kanfer & Ackerman, 1989). We also
?nd that auditor experience enhances the impact of audi-
tor effort on performance. This result is in line with the
argument that effort has a positive impact on performance
only if the individual has some experience with the task
(Yeo & Neal, 2004).
While the ?ndings described above imply that assigning
a highly experienced auditor to a low complexity task
would lead to the highest impact of effort on performance,
this is not the case. Our results indicate that more experi-
ence increases the impact of auditor effort on audit perfor-
mance more when complexity is high. This suggests that
assigning a highly experienced auditor to a more complex
task has a bigger marginal impact on performance than
assigning a highly experienced auditor to a low complexity
task, and consequently maximizes total performance. In
other words, the skills of highly experienced auditors are
wasted on low complexity tasks. These results are
consistent with Tan and Kao (1999) who ?nd that (effort-
inducing) high accountability results in better performance
in highly complex tasks when auditor knowledge and abil-
ity are high. Our results are also consistent with the goal
setting theory arguing that effort and performance depend
on whether the goals are not only attainable, but also chal-
lenging (Fried & Slowik, 2004).
Given that we use archival data, our study differs from
past research in several ways. On the one hand, compared
to an experimental setting, our study has two disadvan-
tages. First, the scope of our analysis is limited by data
availability. For example, while other auditor characteris-
tics, like ability and knowledge, would be useful constructs
to explore, only data on auditor experience is available in
our dataset. Second, our measures are only proxies for con-
structs that can be more accurately measured in an exper-
imental setting. For example, while task complexity can be
manipulated in an experiment, our proxy for task complex-
ity relies on a combination of existing task characteristics
faced by auditors, available in our dataset.
On the other hand, our archival data offers advantages
drawn from the setting we analyze. First, our measures of
auditor experience, effort, and audit performance are those
of a two-person audit team.
2
Given that auditors predomi-
nantly work in teams, this setting enhances the external
validity of our tests. Indeed, in most experimental studies,
auditor characteristics, behavior, and results are measured
and used individually (see Birnberg, 2011). Birnberg (2011)
points at the disadvantages of that approach by arguing that
results from the behavior of individuals cannot be general-
ized to the behavior of those same individuals when they
are acting as a part of a group. Second, we measure audit
performance by the tax adjustment resulting from a tax
audit. Consequently, our study also contributes to the tax
audit literature by further explaining the determinants of
corporate tax audit outcomes. To the best of our knowledge,
previous literature on tax compliance is silent on the effect
of audit and auditor characteristics on tax adjustments. Past
research documents ?rm speci?c factors that in?uence tax
compliance (Chan & Mo, 2000; Mills, 1998; Murray, 1995).
We provide additional insight on audit and audit team char-
acteristics that are associated with tax adjustments.
A related stream of research in accounting uses archival
data to examine the impact of experience and complexity
on performance of analysts (see e.g., Clement, 1999;
Clement, Koonce, & Lopez, 2007; Jacob, Lys, & Neale,
1999). While there are similarities in the two streams of
research (one on analysts and the other on auditors) there
remain important differences between them. Analysts dif-
fer from auditors when it comes to the types of tasks they
perform, their incentives, and the environment they func-
tion in. For example, analyst performance is measured in
this literature as their forecast accuracy. However, given
their role to provide guidance to the market participants,
1
We proxy for audit performance with the amount of tax-adjustments
detected during an audit, scaled by average annual sales. Tax adjustments
can be viewed as a performance outcome in a tax audit since the Croatian
Tax Administration uses tax adjustments to evaluate its own performance,
the performance of its regional of?ces, as well as the performance of its
auditors.
2
As we discuss in the sample and variables construction section, there
are few cases where a team might have more than two auditors. In these
cases, one or both members from the original two-member team could not
continue the mission for whatever reason.
496 W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510
analysts have incentives to and do sacri?ce accuracy for
informativeness (e.g., Louis, Sun, & Urcan, 2013). This could
lead to different predictions and results on the impact of
experience and complexity on performance. In addition,
our study contributes to the literature by providing
evidence on separate and joint moderating effects of com-
plexity and experience on the relationship between effort
and performance, something that has not been explored
in the literature on analysts.
The remainder of the paper proceeds as follows. In the
next section, we present an overview of past literature
and our hypotheses; in Section ‘Institutional background’,
we provide institutional background; in Section ‘Sample
and variable construction’ we present our sample and
provide some descriptive statistics; in Section ‘Research
design’, we describeour researchdesign; inSection‘Results’,
we present our results; and in Section ‘Additional ?ndings
and robustness checks’, we discuss some additional ?ndings
and robustness checks. Section ‘Conclusion’ presents
conclusions.
Literature review and hypotheses development
Empirical archival evidence on the impact of task com-
plexity, auditor effort, and auditor experience, as well as
their interactions on audit performance is rather scant. In
order to build our hypotheses, we thus rely on the theoret-
ical and experimental literature in auditing, management,
and psychology. Given that we provide empirical archival
support to the existing evidence on the direct impact of
auditor experience, effort, and task complexity on audit
performance, we hypothesize only on the moderating role
of auditor experience and task complexity in the auditor
effort-performance relationship (see Fig. 1, Panel A).
Our measure of audit performance is based on
irregularities detected during the audit, similar to mea-
sures based on tasks that subjects perform in experiments
(e.g., Tan & Kao, 1999). Prior research in ?nancial audit
have used number of errors or mistakes caught (e.g.,
Asare & McDaniel, 1996), audit adjustments (e.g.,
Abdolmohammadi & Wright, 1987), and the level of abnor-
mal accruals (e.g., Francis & Yu, 2009; Lennox, Francis, &
Wang, 2014) to measure the performance of an auditor.
Prior research in tax audit have for the most part used
tax audit adjustments to measure tax audit outcomes
(e.g., Chan & Mo, 2000; Mills, 1998; Mills & Sansing,
2000). We follow prior literature and use the tax adjust-
ment detected during the audit as our measure of audit
performance. The advantage of using this measure is that
it is used to evaluate audit performance in the real-world
setting we are analyzing.
The impact of effort on performance has not been stud-
ied extensively in either the accounting or the psychology
literatures. This is attributed to (1) the dif?culty to de?ne
and measure effort as a construct and (2) the lack of pub-
licly available data. Yeo and Neal (2004) argue that effort
is an ‘‘invisible, internal, hypothetical construct that is
not directly observable’’, and consequently dif?cult to
measure and de?ne. Past literature proxies for effort using
either time on task (Caramanis & Lennox, 2008; Christen,
Iyer, & Soberman, 2006; Cloyd, 1997; Cloyd, Pratt, &
Stock, 1996; Fisher & Ford, 1998), or self-reported mea-
sures (Brown & Leigh, 1996). We use time on task as a
measure of auditor effort.
The theoretical and experimental psychology literature
provides arguments and evidence in support of a positive
relationship between effort and performance (for a review
of this literature see, Kanfer & Ackerman, 1989; Weingart,
Fig. 1. Moderators of the relationship between auditor effort and performance.
W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510 497
1992; Yeo & Neal, 2004). Analytical audit research suggests
that achieving a higher audit quality requires an increase
in direct costs to auditors (Dye, 1995; Hillegeist, 1999).
Direct cost is arguably correlated with effort (de?ned as
time on task), as Dye (1995) describes it as the sum ‘‘of
the costs incurred while performing the various tests of
account balances, control tests, analytical reviews, etc.,
described in audit manuals.’’ Hirst (1994) provides experi-
mental evidence that auditors adjust their effort in
response to executives’ incentives to manage earnings.
Caramanis and Lennox (2008) provide empirical archival
evidence in support of this claim.
Our ?rst hypothesis deals with the moderating role of
task complexity in the auditor effort-performance relation-
ship. Prior literature de?nes task complexity as either
objective or subjective (see Bonner, 1994; Campbell,
1988; Maynard & Hakel, 1997). Objective task complexity
is a function of objective task characteristics, while subjec-
tive task complexity is a function of both the task and the
individual’s characteristics. In this paper we analyze the
impact of objective task complexity on audit performance.
Our complexity measure is consistent with Wood’s (1986)
de?nition of complexity. Wood (1986) de?nes complexity
as the sum of information cues to be processed across
the ‘‘number of acts in a subtask’’, and across ‘‘subtasks
in a task’’. Wood (1986) labels this ‘‘Component Complex-
ity’’, which represents a major part in his measure of ‘‘Total
Complexity’’. Applied to our measure of complexity, taxes
are subtasks, and the audit is a task composed of one or
multiple subtasks. In other words, we de?ne one audit as
one task. Consequently, in addition to the type of taxes
audited, the number of taxes serves as a part of the mea-
sure for the complexity of the task.
The limited evidence from experiments yields con?ict-
ing ?ndings onwhether the impact of auditor effort onaudi-
tor performance decreases or increases with an increase in
task complexity. Consistent with the former, in an experi-
ment Mohd-Sanusi and Mohd-Iskandar (2007) ?nd an
increase in audit judgment performance associated with
performance-incentive induced effort only when the task
is less complex, with no such relationship for high complex-
ity tasks. Consistent with the latter, Tan and Kao (1999) ?nd
that (effort-inducing) high accountability results no perfor-
mance improvements in low complex tasks.
Building on Libby (1981, 1985), Bonner (1994) argues
that a change in task characteristics, such as objective task
complexity, can affect auditor judgment, especially when
audit tasks are highly complex. Bonner (1994) further
argues that increases in objective task complexity lead to
decreases in three components of auditor judgment
performance: proper use of knowledge, consistent use
of knowledge, and other direct or unspeci?ed effects.
Other experimental and case study research generally
supports these ?ndings (Asare & McDaniel, 1996; Pratt &
Jiambalvo, 1981; Simnett, 1996; Simnett & Trotman,
1989; Tan et al., 2002).
A more complex task requires more cognitive effort
(Yeo & Neal, 2008) and a higher level of skill (Bonner,
1994). Consequently, per unit of time, a more complex task
will likely result in lower performance (Maynard & Hakel,
1997), as the ability and availability of individuals capable
and willing to exert the necessary cognitive effort will be
limited. Fried and Slowik (2004) argue that the ability of
individuals to successfully pursue multiple complex goals
is limited because they ‘‘heavily tax their cognitive
resources.’’ In addition, a more complex task will typically
require more effort (Locke & Latham, 1990). Given that
there is a diminishing marginal impact of effort on perfor-
mance for every additional unit of effort (Kanfer &
Ackerman, 1989), the relationship between effort and per-
formance should be weaker for high complexity tasks.
Based on the above discussion, we hypothesize as follows
(see Fig. 1, Panel B (a) for a diagram of the hypothesized
relationship):
H1: As task complexity increases, the impact of auditor
effort on audit performance decreases.
Based on the goal setting literature, some might argue
the opposite (i.e., that an increase in task complexity leads
to an increase in the impact of auditor effort on audit per-
formance). This research is based on the premise that more
challenging goals can serve as a motivating factor, espe-
cially when tasks are complex (Campbell, 1988; Maynard
& Hakel, 1997). However, in order for this effect to exist,
goals should be not only challenging but also achievable
(Fried & Slowik, 2004). This would require systematic
and perfect matching between individuals’ capabilities
and demands of the tasks they perform, something that
does not exist in our setting.
Next, we hypothesize on the role of experience in the
effort-performance relationship. Auditor experience has
been studied extensively in the audit literature. These
studies use measures of seniority and years of audit expe-
rience as a proxy for experience (Abdolmohammadi &
Wright, 1987; Lim & Tan, 2010; Simnett, 1996). We use
seniority as a measure of auditor experience.
3
Past research suggests the existence of a positive rela-
tionship between experience and performance. In an
experiment, Abdolmohammadi and Wright (1987) ?nd
that subjects’ experience impacts their audit judgment,
while Simnett (1996) ?nds that auditor experience has a
positive impact on the predictive accuracy of auditors.
Finally, Lim and Tan (2010), ?nd that auditor tenure
improves audit quality. While these studies make use of
experience, others analyze measures correlated with expe-
rience, such as knowledge (Tan & Kao, 1999; Tan et al.,
2002), skill (Bonner, 1991, 1994), and problem-solving
ability (Tan & Kao, 1999). The correlation between these
constructs and experience has been long established in
the literature. For example, studies have shown that expe-
rienced auditors have more and better organized knowl-
edge (Bedard, 1989; Libby & Luft, 1993).
Studies using knowledge, skill, and problem solving
ability all ?nd the same positive impact on audit perfor-
mance (Bonner, 1991, 1994; Tan & Kao, 1999; Tan et al.,
2002). However, Libby and Frederick (1990) show that
experience leads to better performance by auditors
3
In our setting, this dimension is captured through the presence of both
junior auditors and more experienced auditors.
498 W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510
through increases in knowledge. This implies that experi-
ence acts indirectly to increase performance, making
experience an imperfect proxy for knowledge. This biases
against us ?nding a positive relationship between experi-
ence and performance.
On the moderating role of experience, the psychology
literature suggests that experience ampli?es the positive
impact of effort on performance. Yeo and Neal (2004) ?nd
evidence consistent with this relationship and argue that
effort will have no impact on performance if the task is suf-
?ciently novel and the experience of the individual is low.
By contrast, effort will have a positive impact on perfor-
mance if the individual has more prior experience with
the task. Finally, because more experienced auditors spend,
on average, less time in an audit, their marginal return will
be potentially higher.
Contrary to the above arguments, some prior research,
based on resource allocation theory, argues that the rate
of increase in the relationship between effort and perfor-
mance should slow down with practice. Furthermore, the
relationship between effort and performance should
decrease and eventually become insigni?cant (Kanfer &
Ackerman, 1989). If the task has constant information pro-
cessing demands, it will eventually become automated
with increase in practice. However, in a (tax) audit setting,
the tasks involve inconsistent information processing
demands and are highly complex (Bonner, 1994), making
it less likely for these tasks to become automated. Conse-
quently, in our setting, it is unlikely that the relationship
between effort and performance will become insigni?cant
as experience increases.
Based on the above discussion, we hypothesize as fol-
lows (see Fig. 1, Panel B (b) for a diagram of the hypothe-
sized relationship):
H2: As auditor experience increases, the impact of auditor
effort on audit performance increases.
Finally, we hypothesize on the combined effects of com-
plexity and experience on the relationship between effort
and audit performance. A simple extension of H1 and H2
implies that the combination of high experience and low
complexity should yield the highest performance per unit
of effort invested. However, the relationship is more com-
plicated than this. According to Maynard and Hakel (1997),
both cognitive ability and experience will have a greater
impact on performance when the task is more complex.
To put it differently, a more experienced auditor’s knowl-
edge and ability would be wasted on a simple task. This
is consistent with the goal setting literature that argues
that both effort and performance will depend on whether
the goals are challenging and attainable (Fried & Slowik,
2004). Similarly, Tan and Kao (1999) argue and ?nd that
(effort-inducing) high accountability (also a proxy for
motivation) results in better performance in highly com-
plex tasks when both auditor knowledge and ability are
high. Consistent with the belief that experience has a posi-
tive impact on the effort-performance relationship, we
argue that the effort of a more experienced auditor will
have a stronger impact on performance when the task is
more complex. Therefore, we hypothesize as follows:
H3: Experienced auditor’s effort has a greater impact on
performance when task complexity is high.
Institutional background
To test our predictions, we use a unique and con?den-
tial dataset provided by the Croatian Tax Administration.
Republic of Croatia is a developing European country with
a code law legal system (German origin), a low level of
legal enforcement, a low level of investor protection, and
a bank-based economy.
4
The Croatian Tax Administration
is an administrative organization within the Ministry of
Finance. It is divided into 20 regional of?ces located in 20
Croatian counties. The central of?ce is located in Zagreb
(the capital city). Each regional of?ce has one or more tax
audit divisions.
5
Regional of?ces perform tax audits of all
?rms in their territory, regardless of whether they are incor-
porated in their county. Once a regional of?ce begins a tax
audit, even if the ?rm is incorporated in another county, it
will keep the case until its ?nal resolution.
The Croatian tax system consists of national taxes,
county taxes, city or municipal taxes, joint taxes, and taxes
on games of chance. Most ?rms are subject to value added
tax (VAT) and corporate income tax. Private individuals
pay personal income tax with the portion related to their
salaries being collected monthly by ?rms and paid directly
into the country budget.
The Tax Administration employed on average 778 tax
auditors in 2002/2003. Typically, the Tax Administration
employs business school or law school graduates. Future
tax auditors are trained on the job and are ?rst assigned,
for a one- or two-year period, to the Tax Assessment and
Contributions Division of their regional of?ce (in charge
of receiving and processing tax ?lings and payments). They
are subsequently transferred to the Tax Audit Division of
their regional of?ce. Tax auditors in principle should not
hold a signi?cant interest in any Croatian ?rm; if they do,
the control rights must be transferred to a third party. To
avoid con?icts of interest, auditors are not allowed to audit
?rms owned or governed by relatives or any person with
whom they have close ties. To avoid wrongdoing by any
of the parties in the tax audit process, auditors are not
allowed to perform audits alone. Tax auditors have broad
authority that includes full access to all documents and
facilities of the audited ?rm as well as the authority to
close facilities and con?scate documents and goods.
The regional of?ce, independently or at the request of
the central of?ce, initiates a tax audit. The decision to ini-
tiate a tax audit is made by the head of the regional of?ce
and/or the head of the tax audit division of the regional
of?ce. This is done either as part of a regular audit plan
established at the beginning of each year or on an ad hoc
4
Mills (1998) argues that ?rms cannot costlessly maximize ?nancial
reporting bene?ts and minimize taxes at the same time. Our sample
consists of almost only private ?rms, with lower or no incentives to
maximize ?nancial reporting bene?ts (Cloyd et al., 1996). Croatian ?rms
produce ?nancial statements primarily for tax purposes and to a limited
extent for communication with lenders.
5http://www.porezna-uprava.hr
W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510 499
basis, following a request. From within the tax administra-
tion, two cases of ad hoc audit initiations are possible: (1)
the tax assessment and contributions division requests a
tax audit of a ?rm, or (2) a tax auditor, during an audit,
requests that the audit be extended to one or more other
?rms. From outside the tax administration, two cases of
ad hoc audit initiations are possible: (1) an (anonymous)
whistleblower reports tax fraud, or (2) other administra-
tions of the Ministry of Finance or other ministries of the
Republic or foreign governments request a tax audit. Note
that from our conversations with the tax auditors we infer
that the number of ad hoc audit initiations is ‘‘insigni?-
cant’’. Additionally, tax auditors do not (systematically)
have access to the information about the type of audit
initiation unless necessary for conducting the audit. This
alleviates the concern that in our sample tax auditors use
audit initiation type to generate expectations about the
tax adjustment, which would raise concerns about reverse
causality.
Sample and variable construction
Sample
The con?dential database provided by the Tax Adminis-
tration includes data on all 80,719 tax audits executed in
the Republic of Croatia in the 2002–2006 period, as well
as all annual corporate income tax ?lings (140,135 for
2002 and 135,749 for 2003) and VAT ?lings (71,781 for
2002 and 73,853 for 2003). To ensure con?dentiality and
anonymity, both the ?rms audited and the auditors have
been anonymized.
In order to perform our analysis, we need a ?rm to be
subject to both income tax and VAT, and to have ?led
income tax and VAT declarations in 2002 and/or 2003. This
reduces our sample of corporate tax ?lings to 67,812
observations.
6
Note that Croatian ?rms are not required to
keep detailed accounting and tax data beyond six years after
the tax ?ling for all taxes except personal income tax. There-
fore, analyzing audits performed in the 2002–2006 period
guarantees that we capture most audits for tax ?lings in ?s-
cal years 2002 and 2003.
7
In addition, because tax ?lings are
available for years 2002 and 2003 only, we restrict our sam-
ple to only those tax audits that include the entire years or
parts of 2002 and/or 2003. This reduces the number of
observations to 16,125 audits. We exclude all tax audits that
were performed more than once, as they were coded as a
single audit, which could bias our ?ndings (579 audits).
Finally, we exclude observations for which all data necessary
to perform our analysis are not available (154 audits). Our
?nal sample consists of 15,392 tax audits.
Income tax ?lings contain data on the ?rm’s industry,
place of incorporation, sales, expenses, accounting net
income, and detailed reconciliation between accounting
and taxable income. The corporate income tax rate was
35% in both 2002 and 2003 regardless of the level of tax-
able income. VAT ?lings contain data on domestic and for-
eign sales, sales subject to VAT, purchases giving right to
VAT reimbursement, imports, and the VAT due. The VAT
rate was 22% in both 2002 and 2003. Tax audit data
includes duration (number of hours worked by the audit
team), period audited, type of audit (primarily type
taxes audited), number of tax auditors on the case, and
tax adjustments per tax and per year.
Table 1
Distribution of audits by year and industry and audited ?rm characteristics.
Fiscal year audited
Audit year 2002 % 2003 % All %
Panel A: Distribution of audits by audit year and ?scal year
2002 5359 66.61 5 0.06 5362 34.85
2003 1889 23.48 5703 64.55 6663 43.31
2004 628 7.81 2378 26.92 2601 16.90
2005 118 1.47 613 6.94 624 4.06
2006 51 0.63 136 1.54 142 0.92
Total 8045 100.00 8835 100.00 15,392 100.00
Industry Number of audits %
Panel B: Distribution of audits by industry
Agriculture 519 3.37
Fishing 49 0.32
Mining 49 0.32
Manufacturing 2843 18.47
Electrical energy, gas 99 0.64
Construction 844 5.48
Retail and wholesale 6548 42.54
Hotels and restaurants 1837 11.93
Transport 815 5.29
Financial services 51 0.33
Real estate 966 6.28
Education 35 0.23
Health services 30 0.19
Household services 707 4.59
Total 15,392 100
In thousands of HRK Sales BTD (%) ForeignSales
Panel C: Firm ?nancial characteristics
N 15,392 15,392 15,392
Mean 167.00 0.17 0.27
S.D. 1048.96 0.42 0.44
Min 0.00 À0.02 0.00
p25 0.81 0.00 0.00
Median 3.50 0.00 0.00
p75 20.55 0.07 1.00
Max 14351.83 1.66 1.00
This table presents distribution of audits by year and industry and audi-
ted ?rm characteristics. The sample contains 15,392 audits of 2002 and
2003 tax records performed by the Croatian Tax Administration in the
2002–2006 period (9950 unique ?rms). All data are provided by the Tax
Administration of the Republic of Croatia. Panel A shows the distribution
of audits by audit year and ?scal year audited. Panel B shows the distri-
bution of audits by industry. Panel C shows ?nancial characteristics of
?rms audited. Audit Year is the year in which tax audit began. Fiscal Year
Audited is the ?scal year of the tax records audited. Industry is the Cro-
atian national industry classi?cation. Sales is the mean of 2002 and 2003
annual sales of ?rms audited. BTD is the mean of the 2002 and 2003
annual book-tax difference of ?rms audited. ForeignSales is a binary var-
iable that equals one if the audited ?rm had foreign sales in either 2002 or
2003.
6
Not all ?rms are subject to VAT, and they become subject to VAT after
their revenues exceed a threshold. Banks and insurance companies are not
subject to VAT but are subject to income tax, and individuals may become
subject to VAT and are seldom subject to corporate income tax.
7
Excluding from our sample audits performed in 2005 and 2006, to
avoid selection concerns associated with later audits, also yields statisti-
cally signi?cant results in all our tests.
500 W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510
Table 1 provides the distribution of audits by year and
industry, as well as ?rm characteristics. Panel A of Table 1
presents the distribution of the years the audit started (i.e.,
2002, 2003, . . ., or 2006) and the ?scal year audited (i.e.,
2002, or 2003). Most audits were performed in 2003 and
2002 (43.31% and 34.85%, respectively). Out of all the
2002 and 2003 ?scal year ?lings audits, 16.90% were con-
ducted in 2004. The number of audits for the 2002 and
2003 ?lings naturally declines over the 2002–2006 period,
as the Tax Administration rarely audits tax ?lings older
than ?ve years and dedicates its capacity to more recent
tax ?lings.
8
In addition to auditing past years’ annual tax ?l-
ings, the Tax Administration systematically audits same-
year ?lings. These audits typically include a combination
of VAT monthly ?lings, personal income tax monthly ?lings,
and one-day audits of sales (revenue audits). The Tax
Administration audited a similar number of ?rms’ ?lings
for 2002 and 2003 (8045 and 8835, respectively).
Panel B of Table 1 presents the distribution of audited
?rms by industry. The largest number of audits was per-
formed in the retail and wholesale industries (6548, or
42.54%), followed by manufacturing (2843, or 18.47%)
and hotels and restaurants (1837, or 11.93%). This re?ects
the overall Croatian economy, which is based primarily
on services and tourism.
9
Panel C of Table 1 presents the relevant ?nancial infor-
mation for the sample of audited ?rms. To proxy for the
size of the audited ?rm, we use FirmSize, measured as the
natural log of the mean of 2002 and 2003 annual sales.
Averaging the sales limits the impact both of sales varia-
tions (which can be signi?cant in our sample, as it consists
of mainly small and medium enterprises) and of new ?rms
entering the market.
10
In cases of absence of 2002 or 2003
sales data, we use the data from the year for which it is
available. The mean ?rm size (Sales) of the ?rms in our sam-
ple is HRK167.00 million (median of HRK3.50 million) rang-
ing from less than HRK10,000 to more than HRK14 billion.
11
In the same vein as sales, we also average the book-tax dif-
ference (BTD), scaled by sales, of 2002 and 2003. The mean
BTD represents 16.61% of sales, with a median of 0.00%. It
ranges from À1.72% to 165.85% of sales. In our sample,
26.60% (4094) of the audited ?rms have foreign sales. We
use an indicator variable to proxy for foreign sales,
ForeignSales, where it equals one if the ?rm reports foreign
sales in either 2002 or 2003, and zero otherwise.
Variable construction
Table 2 presents the construction of our variables of
interest: task complexity, effort, and experience. Panel A
of Table 2 presents the distribution of audits by the type
and number of taxes audited (N:Taxes) of taxes audited.
We categorize taxes audited by the Tax Administration
into ?ve categories: VAT, corporate income tax, personal
income tax, revenue, and other. An audit of VAT can
include one or several monthly and/or annual VAT ?lings.
An audit of corporate income tax can include one or several
annual ?lings because income tax declarations are ?led
annually. An audit of personal income tax includes an audit
of personal income tax, social charges, and related local
taxes (all salary-related items). They can be audited either
based on monthly ?lings or based on annual ?lings, the lat-
ter case being prevalent. An audit of revenue is a special
type of audit that can have an impact on several taxes. Rev-
enue audits typically last for a day but can be extended to a
longer time period. Not declaring revenues can have an
impact on both VAT and corporate income tax. Finally,
‘‘other’’ audits can include longer revenue audits, audits
of speci?c transactions (typically at the request of the Cus-
toms Administration or the Ministry of Interior), audits
extended to other ?rms, and audits of other taxes (e.g.,
gaming taxes).
Most audits in our sample involve auditing one type of
the above taxes (10,172 audits, or 66.09%), followed by
audits of three types of taxes (2548, or 16.55%), and then
audits of two types of taxes (2398, or 15.58%). VAT was
audited in 56.01% of cases, followed by revenue (35.69%)
and personal income tax (23.92%). The table also highlights
the fact that audits involving one or more taxes primarily
include VAT and/or revenue, while those with three or
more taxes almost always include VAT, corporate income
tax, and personal income tax. We asked several tax audi-
tors about the complexity of the audits by category.
According to all interviewed tax auditors, the most com-
plex audit is what they refer to as a ‘‘complete’’ audit
including VAT, corporate income tax, and personal income
tax. We thus code audits including at least VAT, corporate
income tax, and personal income tax as complex audits, and
other audits as simple.
12
Note that, mechanically, a complex
audit implies period audited of one year or more, given that
it includes auditing corporate income tax ?lings, ?led on an
annual basis. Out of the total number of audits (15,392),
1995 (12.96%) are complex and 13,397 (87.04%) are simple
audits. Our complexity measure is a combination of qualita-
tive (type of taxes) and quantitative (number of taxes being
three or more) characteristics, consistent with the Wood
(1986) de?nition of complexity.
The Tax Administration can decide to audit from one
day to several years of accounting and tax records. We
label this audit period TaskSize. Panel B of Table 2 indi-
cates that TaskSize ranges from 1 day to 2544 days (more
than seven years), with a mean of 236 days and a median
of 31 days.
13
Complex audits typically exceed one year of
audit period (median equals 546 days), whereas simple
audits typically include audit periods of less than a year
8
Note that ?ve audits started in 2002 but the year audited was 2003.
Three of those ?ve audits started on December 31, 2002, while the two
remaining cases include both the audit for 2002 and 2003.
9
See, for example, the 2005 Statistical Information, Republic of Croatia
Central Bureau of Statistics.
10
Our results are not sensitive to this choice of size proxy.
11
The exchange rate for the Croatian Kuna (HRK) on January 1, 2003, was
7.45 HRK per Euro and 7.11 per US$, respectively, according to the Croatian
Central Bank.
12
Our results are not sensitive to this de?nition of complex audits. They
remain statistically signi?cant if we de?ne complex audits as those that
include at least VAT and corporate income tax, or those that include at least
personal income tax and corporate income tax.
13
Note that both TaskSize and Effort (de?ned in the coming paragraphs)
are used in our analysis as logged variables. They are presented in the table
in ‘days’ for descriptive purposes.
W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510 501
(median equals 30 days). For brevity, we do not tabulate
this data.
Croatian tax audits are performed by two-person
audit teams.
14
We use information on their seniority to
proxy for auditor experience, Experience. We de?ne Experi-
ence as a binary variable that equals one if none of the
auditors in an audit team is a junior auditor, and zero
otherwise. To distinguish between audit teams with a
junior auditor and those without a junior auditor we turn
to tax administration records. There are two possible audit
records. If the tax administration recorded two auditors
performing an audit, we treat this audit as being per-
formed by two senior auditors. If, however, the record
shows only one auditor performing the audit, we treat
the audit as being performed by one junior and one senior
auditor. The reason is as follows: if the records indicate
only one auditor performing the audit, the second auditor
is a junior auditor who at the moment of performing the
audit does not have an auditor identi?cation number,
and cannot be recorded in the system as having performed
the audit.
15
In our setting, there are no audits performed
by two junior auditors.
In Croatia, the Tax Administration records the number
of hours worked per audit and per auditor. We use the total
number of audit hours worked per audit by auditors, con-
verted into days, as a proxy for auditor effort, Effort, consis-
tent with prior literature (Caramanis & Lennox, 2008).
Given that the hours of a junior auditor are not recorded
in the system, and given that there is virtually no differ-
ence between the ?rst and the second auditor in time
Table 2
Audit and auditor characteristics.
Tax type N.Taxes Total
1 2 3 4 5
Panel A: Distribution of audits by number of taxes and tax type
All audits 10,172 2398 2548 265 9 15,392
VAT 4048 1758 2541 265 9 8621
(39.80%) (73.31%) (99.73%) (100.00%) (100.00%) (56.01%)
Corporate income tax 92 491 1875 239 9 2706
(0.90%) (20.48%) (73.59%) (90.19%) (100.00%) (17.58%)
Personal income tax 898 574 1943 258 9 3682
(8.83%) (23.94%) (76.26%) (97.36%) (100.00%) (23.92%)
Revenue 3691 1071 591 132 9 5494
(36.29%) (44.66%) (23.19%) (49.81%) (100.00%) (35.69%)
Other 1443 902 694 166 9 3214
(14.19%) (37.61%) (27.24%) (62.64%) (100.00%) (20.88%)
Complex tasks 0 0 1754 232 9 1995
(0.00%) (0.00%) (68.84%) (87.55%) (100.00%) (12.96%)
Simple tasks 10,172 2398 794 33 0 13,391
(100.00%) (100.00%) (31.16%) (12.45%) (0.00%) (87.00%)
TaskSize (Days) Effort (Days) Experience BigOf?ce
Panel B: Other audit and auditor characteristics
N 15,392 15,392 15,392 15,392
Mean 236.24 11.85 0.72 0.27
S.D. 316.49 18.84 0.45 0.44
Min 1.00 0.25 0.00 0.00
p25 1.00 1.25 0.00 0.00
Median 31.00 5.00 1.00 0.00
p75 365.00 15.50 1.00 1.00
Max 2544.00 385.50 1.00 1.00
This table presents audit and auditor characteristics for the sample observations. The sample contains 15,392 audits of 2002 and 2003 tax record audits
performed by the Croatian Tax Administration in the 2002–2006 period. All data are provided by the Tax Administration of the Republic of Croatia. N:Taxes
is the number of taxes covered in an audit. Complex is a binary variable that equals one if an audit includes at least VAT, corporate income tax, and personal
income tax; and zero otherwise. Simple audits are all other audits. The percentages in Panel A represent the number of types for a given audit relative to the
total number of all audits in the given column. TaskSize is the number of days of tax records that were audited (i.e., period audited). Effort is the duration of
an audit in auditor days, taking into account all auditors involved. Experience is a binary variable that equals zero if one of the auditors is a junior auditor,
and one otherwise. BigOf?ce is a binary variable that equals one if the regional of?ce is Split or Zagreb, and zero otherwise.
14
While it is possible to have more than two auditors performing an
audit, this is uncommon. In our sample, the number of auditors per audit
ranges from two to eight (untabulated), with the vast majority of audits
(14,866, or 97%) having two auditors. The three percent of audits for which
more than two auditors were present are mostly cases with three auditors
(406, or 2.64% of all cases). These cases are mainly those for which one of
the two initial auditors was replaced during the audit. Consequently, we
include these cases in our analysis. Excluding these cases from our analysis
or controlling for the number of auditors does not change our inferences.
15
We are treating the experience between two auditors in a team and
across experienced audit teams as ?xed. Even further, based in our design
choice, we could have the case where the combined years of experience in
an experienced audit team to be less than or equal a non-experienced audit
team. In those cases, we are less likely to detect differences between audit
performances between the teams, a bias that will work against us.
502 W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510
spent on the task (an untabulated mean and median of
zero), in order to adjust Effort for the presence of the junior
auditor, we assign to a junior auditor the same number of
days assigned to the senior auditor in the system. Effort
ranges from 0.25 days to 385.50 days, with a mean of
11.85 days and a median of 5 days. Out of our total number
of audits, 72% (11,079 cases) are audits performed by more
experienced auditors (senior auditors) and 28% (4313
cases) are audits performed by teams where one of the
two auditors is a junior auditor.
The Tax Administration operates from twenty regional
of?ces that vary in terms of number of auditors, number
of audits performed, area covered, and size of ?rms in their
county. To explore the differences in size of the regional
of?ces, we isolate the two largest of?ces and label them
as a BigOf?ce (Francis & Yu, 2009). We single out Split
and Zagreb as big regional of?ces for three reasons. First,
the Split regional of?ce is the one with the largest number
of audits performed in our sample (2878 audits, or 18.70%
of the total) and covers the largest geographical area. Sec-
ond, being the capital city of?ce, the Zagreb regional of?ce
covers the largest number of ?rms and the largest number
of large ?rms, followed by the Split regional of?ce. Third,
the number of tax auditors is the highest in Zagreb (203
in 2002 and 191 in 2003), followed by Split (112 in 2002
and 100 in 2003). About one-quarter of all the audits in
our sample (4080) are performed by the two big audit
of?ces.
We use the natural log of tax adjustment scaled by
average annual sales detected during the audit as our mea-
sure of audit performance because the Croatian tax admin-
istration evaluates its auditors, regional of?ces, and its
overall performance based on the amounts of tax adjust-
ments detected during audits. The tax adjustment is scaled
by sales to account for size differences across ?rms. The tax
adjustment is integrated in the internal reporting and per-
formance measurement system of the Tax Administration.
From interviews with Croatian tax auditors we can infer
that pressure is exercised on auditors to increase tax
adjustments, as reports of tax adjustments are drawn per
auditor systematically. This is used by the central of?ce
to evaluate regional of?ces’ performance.
We present the descriptive statistics concerning the
audit outcomes in Table 3. Panel A of Table 3 presents
the percentage of audits ‘with’ and ‘without’ a tax adjust-
ment. About seventy-four percent (11,403 audits) of audits
resulted in no tax adjustments, whereas the remaining
3983 audits resulted in tax adjustments. Panel B of Table 3
presents tax adjustments scaled by sales (winsorized at
2.5%) and the log of tax adjustment scaled by sales (TaxAdj)
that we use as our dependent variable in the remainder of
the study. Note that tax adjustments are never negative,
i.e. the Tax Administration never recorded a case where
taxes were reduced for a ?rm. The mean tax adjustment
represents 15.69% of sales (a median of 1.31%) and ranges
from 0.01% to 242.38%. Table 4 presents the correlations
between the variables used in the analysis. The table
indicates that TaxAdj is positively correlated with Effort
and Experience and negatively correlated with Complexity,
consistent with our predictions. The table also indicates
that more experienced auditors exert less effort and more
complex tasks require more effort. An interesting correla-
tion, however, is the negative correlation between Experi-
ence and Complexity, which suggests that more senior
auditors do not automatically get assigned to high complex
audit cases.
Research design
To test our hypotheses, we estimate an OLS regression
model with TaxAdj, our measure of audit performance as
dependent variable, and measures of task complexity,
auditor effort, and auditor experience, as well as other
known control variables fromthe literature as independent
variables.
We estimate the following regression models:
TaxAdj
it
¼ a
0
þa
1
Complexity þa
2
Effort
þa
3
Experience þa
4
Complexity ÂEffort
þb
1
FirmSize þb
2
BTD þb
3
ForeignSales
þb
4
BigOffice þb
5
MillsLambda þdIndustry
þdAudit Year þe ð1Þ
TaxAdj
it
¼ a
0
þa
1
Complexity þa
2
Effort
þa
3
Experience þa
5
Experience ÂEffort
þb
1
FirmSize þb
2
BTD þb
3
ForeignSales
þb
4
BigOffice þb
5
MillsLambda þdIndustry
þdAudit Year þe ð2Þ
where Complexity is a binary variable that equals one if the
audit includes at least VAT, corporate income tax, and
Table 3
Audit outcome.
Number of audits %
Panel A: Distribution of audits by audit outcome
No tax adjustment detected 11,409 74.12
Tax adjustment detected 3983 25.88
Total 15,392 100
Tax adjustment
(% of sales,)
TaxAdj
Panel B: Tax adjustments
N 3983 3983
Mean 15.69 À4.21
S.D. 44.58 2.60
Min 0.01 À20.26
p25 0.29 À5.84
Median 1.31 À4.33
p75 7.20 À2.63
Max 242.38 11.55
This table presents audit outcome descriptive statistics. The sample
contains 15,392 audits of 2002 and 2003 tax records performed by the
Croatian Tax Administration in the 2002–2006 period. All data are pro-
vided by the Tax Administration of the Republic of Croatia. No tax
adjustment detected indicates that no tax adjustment was detected in the
course of the audit. Tax adjustment detected indicates there was a tax
adjustment detected in the course of the audit. Tax Adjustment in Panel B
is the tax adjustment, for the 3983 audits that had an adjustment, scaled
by the mean of 2002 and 2003 annual sales. TaxAdj is the natural log of
the tax adjustment scaled by the mean of 2002 and 2003 annual sales.
W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510 503
personal income tax; and zero otherwise.
16
Effort is the
natural log of the total number of days worked by all
auditors on the case. Experience is a binary variable that
equals zero if one of the auditors was a junior auditor, and
one otherwise. Consistent with H1, we expect the coef?cient
associated with Complexity ÂEffort (a
4
in Model 1) to be
negative and statistically signi?cant. With respect to H2,
we expect the coef?cient associated with the ExperienceÂ
Effort (a
5
in Model 2) to be positive and statistically
signi?cant.
In addition to our variables of interest, we include a
number of control variables in our regression models that
are identi?ed in the prior literature to be correlated with
TaxAdj and that could be correlated with our variables of
interest. We include FirmSize; BTD; ForeignSales (Mills,
1998), BigOf?ce (Francis & Yu, 2009), Industry, and Audit
Year dummies as control variables in our regression
models. Caramanis and Lennox (2008) show ?rm size to
be positively correlated with audit length. We thus control
for ?rm size by including the natural log of the average
sales of 2002 and 2003, FirmSize. Consistent with Mills
and Sansing (2000), we expect FirmSize to be negatively
correlated with TaxAdj. Mills (1998) shows book-tax differ-
ences to be positively correlated with tax adjustment. BTD
equals book-income minus tax-income, scaled by sales.
Consistent with Mills (1998), we expect BTD to be posi-
tively correlated with TaxAdj.
Firms with foreign operations will have more items to
audit, which could imply longer audits (i.e., more effort).
Chan and Mo (2000) ?nd a negative association, while
Mills (1998) and Mills and Sansing (2000) ?nd either no
association or a positive association between export-
oriented ?rms and tax adjustment. To control for this
effect, we include an indicator variable (ForeignSales) in
our regression model that takes the value of one if the ?rm
had foreign sales in either 2002 or 2003, and zero other-
wise, but we do not form expectations on the association
between this variable and TaxAdj.
Tax auditors in larger of?ces are more likely to detect
material problems because a large of?ce has more inspec-
tors, and these inspectors have exposure to more ?rms,
which leads to greater collective human capital in the
of?ce. Caramanis and Lennox (2008) and Francis and Yu
(2009) show that big audit ?rms and of?ces are associated
with better audit performance. To control for the size of
the of?ce, we introduce a variable, BigOf?ce, which equals
one if the regional of?ce is Split or Zagreb, and zero other-
wise. We expect BigOf?ce to be positively correlated with
TaxAdj.
From interviews with tax auditors and from prior liter-
ature, we cannot form directional predictions on if and
how Experience or Complexity may be correlated with any
of our control variables. The reasoning is as follows: the
Tax Administration does not follow a system to determine
which ?rms to audit; the decision to audit a ?rm is made
by 20 regional of?ces independently. In addition, some
audits are performed following the ?ling of anonymous
reports, or at the request of the Customs Administration
or the Ministry of Interior. Nevertheless, we control for
the above-cited variables to account for potential biases
that could still exist and because past literature shows
these variables to be correlated with auditor effort.
Because our measure of audit performance is the tax
adjustment, we must take into account a potential
selection bias in our sample, as only cases for which tax
adjustments (3983 audits) exist are included in our sam-
ple. Similar to Mills (1998), we use a two-stage Heckman
(1979) procedure that takes into account the probability
of tax adjustments being detected. In the ?rst stage, we
estimate the probability of an adjustment to be proposed
by the tax authorities and, using the parameters from this
Table 4
Correlations.
TaxAdj Complexity Effort Experience FirmSize BTD ForeignSales BigOf?ce
TaxAdj 1
Complexity À0:065
???
1
Effort 0.240
???
0.133
???
1
Experience 0.070
???
À0:078
???
À0:086
???
1
Size À0:515
???
0.082
???
0.252
???
À0:008 1
BTD 0.224
???
À0.087
???
À0.048
???
0.018 À0:265
???
1
ForeignSales À0:219
???
0.031
??
0.113
???
0.029
?
0.354
???
À0:101
???
1
BigOf?ce 0.115
???
À0:098
???
0.096
???
0.277
???
0.044
???
0.004 0.019 1
This table presents the correlations between the variables used in the analysis. The sample contains 3983 audits with tax adjustment detected from 2002 to
2003 tax records performed by the Croatian Tax Administration in the 2002–2006 period. All data are provided by the Tax Administration of the Republic of
Croatia. TaxAdj is the natural log of the tax adjustment scaled by the mean of 2002 and 2003 annual sales. Complexityis a binary variable that equals one if an
audit includes at least VAT, corporate income tax, and personal income tax; and zero otherwise. Effort is the natural log of the duration of an audit
(measured in auditor-days). Experience is a binary variable equal to zero if one of the auditors is a junior auditor, and one otherwise. FirmSize is the natural
log of the mean of 2002 and 2003 annual sales of ?rms audited. BTD is the mean of the 2002 and 2003 annual book-tax difference of ?rms audited.
ForeignSales is a binary variable equal to one if the audited ?rm had foreign sales in either 2002 or 2003, and zero otherwise. BigOf?ce is a binary variable
equal to one if the regional of?ce is Split or Zagreb, and zero otherwise.
?
Two tailed statistical signi?cance at the 10 percent level.
??
Two tailed statistical signi?cance at the 5 percent level.
???
Two tailed statistical signi?cance at the 1 percent level.
16
In a different analysis, we replace Complexity with dummy variables for
each of the tax types to take into account the possibility that individual tax
characteristics are driving our results. Our results remain statistically
signi?cant.
504 W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510
model, we compute the inverse Mills ratio for all observa-
tions. In the second stage, we include the inverse Mills
ratio as an additional control variable and estimate the
amount of the tax adjustment.
17
The dependent variable in the ?rst stage is coded one if
a tax adjustment exists, and zero otherwise. It is a function
of TaskSize (i.e., period audited), TaskType (a VAT dummy),
N:Taxes, Effort, Experience; FirmSize; BTD; ForeignSales,
BigOf?ce, Loss; Industry, and AuditYear indicator variables.
Note that following Lennox et al. (2012), several variables
in the ?rst stage regression model are not included in
our second stage models. Also note that we use
TaskSize; TaskType, and N:Taxes instead of the Complexity
variable in the ?rst stage, as we aim to better specify the
model, while testing our hypotheses in the second stage
model, thus using a single proxy for audit complexity.
Using Complexity in the ?rst stage regression model instead
of the three above-mentioned variables does not change
our inferences.
Finally, to test our third hypothesis, H3, we estimate the
following regression model:
TaxAdj
it
¼ a
0
þa
1
Experience þa
2
Effort
HTC
þa
3
Experience ÂEffort
HTC
þa
4
Effort
LTC
þa
5
Experience ÂEffort
LTC
þþa
6
FirmSize
þa
7
BTD þa
8
ForeignSales þa
9
BigOffice
þa
10
MillsLambda þdIndustry
þdAudit Year þe ð3Þ
First, we split the sample into two subsamples based on
Effort, Effort when the task is complex (Effort
HTC
) and Effort
when the task is simple (Effort
LTC
). In other words, Effort
HTC
equals Effort when Complexity equals one, and zero
otherwise, and Effort
LTC
equals Effort when Complexity
equals zero, and zero otherwise. Then, we interact both
of these new variables with Experience. This yields two
interactions: (1) (Experience ÂEffort
HTC
), which represents
the impact of auditor experience on the auditor effort-tax
adjustment relationship when task complexity is high,
and (2) (Experience ÂEffort
LTC
), which represents the
impact of auditor experience on the auditor effort-tax
adjustment relationship when task complexity is low. To
compare the relative importance of experience on the
effort-task complexity relationship, when complexity is
high vs. when complexity is low, we test the difference
between the coef?cient associated with (ExperienceÂ
Effort
HTC
) and the coef?cient associated with (ExperienceÂ
Effort
LTC
). Consistent with our Hypothesis (3) we expect
this difference to be positive, indicating that higher auditor
experience increases the positive impact of auditor effort
on audit performance more when task complexity is high.
Results
In Table 5, column 1, we present a version of Model 1
with main effects only. Consistent with prior research,
the coef?cient associated with Complexity is negative and
statistically signi?cant, while the coef?cients associated
with Effort and Experience are positive and statistically sig-
ni?cant. Holding everything else constant, a more complex
audit will yield a tax adjustment that is 23.66% lower than
a less complex audit.
18
Holding everything else constant, a
10% increase in auditor effort (approximately one day at
the mean and half a day at the median) increases the tax
adjustment by 16.99%. An audit involving a highly experi-
enced auditor yields a tax adjustment that is 81.12% higher
compared to an audit involving an inexperienced auditor,
holding everything else constant. All of our control variables
have the predicted signs, consistent with prior literature.
FirmSize is negatively correlated with TaxAdj, while BTD
and BigOf?ce are positively correlated with TaxAdj. Foreign-
Sales are negatively correlated with TaxAdj, consistent with
Chan and Mo (2000).
Column 2 of Table 5 presents the results from running
Model 1, where we test whether task complexity reduces
the impact of auditor effort on audit performance (H1).
Our results are consistent with this hypothesis. The coef?-
cient associated with Effort is positive and statistically
signi?cant, while the coef?cient associated with the Com-
plexity  Effort interaction is negative and statistically sig-
ni?cant. The sum of the two coef?cients is positive and
statistically signi?cant. For a 10% increase in auditor effort,
the tax adjustment increases by 17.79% when Complexity is
low, and by 16.02% when Complexity is high. This translates
into an impact of Effort on TaxAdj that is 11.06% (17.79% vs.
16.02%) stronger when Complexity is low. All of our control
variables have the predicted signs.
Column 3 of Table 5 presents the results of whether
auditor experience increases the impact of auditor effort
on audit performance (Model 2, H2). Both the coef?cient
associated with Effort, and the coef?cient associated with
the Experience  Effort interaction are positive and statisti-
cally signi?cant, validating H2. The impact of auditor expe-
rience on the auditor effort-tax adjustment relationship is
economically signi?cant. In cases with low auditor experi-
ence, a 10% increase in auditor effort will yield a 15.95%
increase in tax adjustment, whereas in cases with high
auditor experience, it yields an 18.27% increase in tax
adjustment. This translates into a 14.55% stronger impact
of auditor effort on tax adjustment when auditor experi-
ence is high. All control variables have the predicted signs
and are statistically signi?cant.
Finally, in Table 6 we show estimates of Model 3, which
tests whether the effort of a more experienced auditor is
17
Our study could also suffer from a second potential selection bias, as
?rms may not be randomly audited by the tax authorities. To take into
account this possibility, as a robustness check, we run a probit model that
explains the probability of being audited as a function of the ?rm size,
book-tax differences, existence of foreign operations, auditors belonging to
a big of?ce, reporting losses, and industry and year ?xed effects. From this
model, we incorporate a second inverse mills ratio in our main analysis. Our
results (untabulated) remain statistically signi?cant, and this second
inverse mills ratio is not statistically signi?cant in all our models, raising
questions about the existence of this selection bias.
18
To calculate the percentage change in the dependent variable when the
independent variable is an indicator variable, we use [exp (coef?cient)-1].
When the independent variable is continuous and logged, we use
[1.10
ˆ
(coef?cient)-1] to obtain a percentage change in the dependent
variable for a 10% change in the independent variable.
W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510 505
more bene?cial when exerted on a more complex
audit (H3). Both interactions (Experience ÂEffort
HTC
and
Experience ÂEffort
LTC
) are positive and signi?cant. The
incremental effect of high auditor experience when task
complexity is high is 2.40% (16.03% of the impact of
Effort
HTC
), compared to the 1.70% (11.35% of the impact of
Effort
LTC
) when task complexity is low. The difference
between the coef?cients associated with the two interac-
tions is statistically signi?cant. As in the previous regres-
sion models, all control variables have the predicted
signs and are statistically signi?cant.
In Figs. 2 and 3 we plot our results. In Fig. 2, Panel A, we
illustrate our results from Table 5, column 2. We plot the
values of audit performance (TaxAdj) as a function of
auditor effort (Effort) in complex versus simple tasks
(Complexity). The plot shows higher audit performance in
a simple task than in a complex task. The slopes of the lines
representing simple and complex tasks also show that an
increase in auditor effort increases audit performance
more (less) in a simple (complex) task, consistent with
our ?rst hypothesis (H1).
In Panel B of Fig. 2 we plot the relationship between
Effort and TaxAdj by using the estimates from Table 5, col-
umn 3, i.e. for auditors with high versus low experience
(Experience). The plot shows higher audit performance
when auditors have more experience. It also shows that
audit effort increases performance more (less) when audi-
tors have more (less) experience, consistent with our sec-
ond hypothesis (H2).
In Fig. 3 we compare the differential impact of effort for
auditors with a high versus low experience (Experience x
Effort) on performance (TaxAdj) in a complex versus simple
task (Complexity). This ?gure is based on Table 6 estimates.
Consistent with our third hypothesis (H3), in a complex
task, the effort of a more experienced auditor has a higher
impact on performance than in a simple task.
Additional ?ndings and robustness checks
Additional ?ndings
In untabulated results, we estimate our regression
model (1), but interact the Complexity and Experience
variables. Our results show that auditor experience
mitigates the negative impact of task complexity on audit
performance. The coef?cient associated with Complexity
is negative and statistically signi?cant, whereas the one
associated with the Complexity  Experience interaction is
positive and statistically signi?cant. The sum of the two
coef?cients is negative and statistically signi?cant. A com-
plex task will lower tax adjustment by 38.43% compared to
a simple task when auditor experience is low, and by
15.04% when auditor experience is high. This translates
into a negative impact of task complexity being only one
third as large when experience is high. In general, all prior
studies ?nd that auditor skill, knowledge, and problem-
solving ability mitigate the impact of task complexity on
audit performance (Abdolmohammadi & Wright, 1987;
Bonner, 1994; Chang, Ho, & Liao, 1997; Tan & Kao, 1999;
Tan et al., 2002). Consistent with these prior studies, we
?nd that experience, as a related construct (see Section ‘Lit-
erature review and hypotheses development’), mitigates
the negative relation between task complexity and audit
performance.
Our results also indicate that experience has a bigger
positive impact on audit performance when a task is
complex than when it is simple. The coef?cient associated
with the Experience variable, the coef?cient of ComplexityÂ
Experience and the sum of the two coef?cients are all
positive and statistically signi?cant. A more experienced
auditor will perform 62.09% better (have a 62.09% higher
Table 5
The impact of tax complexity, auditor effort, and auditor experience on tax
adjustment.
(1) (2) (3)
Constant 1.666
???
1.439
???
2.011
???
(3.759) (3.111) (4.377)
Complexity À0.27
???
0.24 À0.27
???
(4.023) (0.807) (3.949)
Effort 1.646
???
1.718
???
1.553
???
(14.562) (14.252) (13.202)
Experience 0.594
???
0.611
???
À0.043
(7.891) (8.017) (0.194)
Complexity ÂEffort À0.159
?
(1.753)
Experience ÂEffort 0.208
???
(3.009)
FirmSize À0.825
???
À0.828
???
À0.832
???
(43.457) (43.191) (43.361)
BTD 0.478
???
0.480
???
0.469
???
(7.223) (7.196) (7.036)
ForeignSales À0.268
???
À0.273
???
À0.279
???
(3.661) (3.701) (3.784)
BigOf?ce 0.540
???
0.538
???
0.564
???
(7.787) (7.704) (8.042)
MillsLambda 0.946
???
1.032
???
1.025
???
(4.680) (4.960) (5.033)
Industry FE Included Included Included
Audit year FE Included Included Included
N 15,392 15,392 15,392
Uncensored N 3983 3983 3983
Wald v
2
3666.59 3625.58 3640.78
Prob. > v
2
0.00 0.00 0.00
This table presents results from estimating an OLS model for tax adjust-
ments. The sample contains 15,392 audits of 2002 and 2003 tax records
performed by the Croatian Tax Administration in the 2002–2006 period.
All data are provided by the Tax Administration of the Republic of Croatia.
The dependent variable, TaxAdj, is the natural log of the tax adjustment
scaled by the average of sales in the years 2002 and 2003 (see text).
Complexityis a binary variable that equals one if an audit includes at least
VAT, corporate income tax, and personal income tax; and zero otherwise.
Effort is the natural log of the duration of an audit (measured in auditor-
days). Experience is a binary variable equal to zero if one of the auditors is
a junior auditor, and one otherwise. FirmSize is the natural log of the mean
of 2002 and 2003 annual sales of ?rms audited. BTD is the mean of the
2002 and 2003 annual book-tax difference of ?rms audited. ForeignSales is
a binary variable equal to one if the audited ?rm had foreign sales in
either 2002 or 2003, and zero otherwise. BigOf?ce is a binary variable
equal to one if the regional of?ce is Split or Zagreb, and zero otherwise.
MillsLambda is the inverse Mills ratio resulting from the ?rst stage
regression with a binary variable equal to one if irregularities were
detected and zero otherwise as the dependent variable, and the inde-
pendent variables from Model (1) plus a TaskSize; TaskType (a VAT
dummy), and N:Taxes in place of Complexity, and a Loss dummy variable
that equals one if the ?rm had a loss and zero otherwise.
?
Two tailed statistical signi?cance at the 10 percent level, respectively.
**
Two tailed statistical signi?cance at the 5 percent level.
???
Two tailed statistical signi?cance at the 1 percent level.
506 W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510
tax adjustment) than a less experienced auditor when a
task is simple, and 123.67% better when a task is complex.
Robustness checks
Endogeneity
Our unique database and setting allow us to draw
conclusions from the archival data without threats of
endogeneity between audit performance and audit/auditor
characteristics. In our setting, seniority of auditors, task
complexity, and auditor effort are not decided based on
the expected tax adjustment. Auditors are assigned to
tasks based on their availability, in an effort to maximize
the use of the Tax Administration’s capacity. Task complex-
ity is decided based on the constrained capacity, the need
to use capacity throughout the year, and the need to audit
different taxes across multiple periods at any time during
the year. An example of this would be an audit of a
monthly VAT ?ling compared to an audit of several taxes
based on an annual tax ?ling. Auditor effort is decided by
the auditor, and not the supervisor choosing the ?rm to
be audited. Finally, the same ?rms are not likely to be
audited systematically every year, and even if this is the
case, the same taxes are not likely to be audited every
year.
19
The preceding arguments suggest that the Tax
Administration has little information to form expectations
on the amount of tax adjustment. Note that the Croatian
Tax Administration does not use a formal model to select
?rms to audit.
That being said, absent other information, the auditors
may still form expectations about the potential tax adjust-
ments based on ?rm’s tax ?lings, which would lead to the
endogeneity (reverse causality) problem, as they would
adjust their effort as a function of the expected tax adjust-
ment. Therefore, we searched for a setting where auditors
are unlikely to form expectation with respect to our
dependent variable. We use the book-tax-difference (the
difference between book income and taxable income) as
a proxy for tax adjustment expectations.
20
Mills (1998)
?nds that book-tax-differences are the main driver of and
positively correlated with the tax adjustments detected
during a tax audit. We conduct a test in which we divide
our sample into ?rms with positive book-tax-difference
amounts (those more likely to have a higher tax adjustment)
and ?rms with no or negative book-tax-difference amounts
(those less likely to have a tax adjustment). In the sample of
?rms with no or negative book-tax-difference amounts our
results are unchanged compared to our main results. When
comparing the two subsamples, results are similar. The
untabulated ?ndings of this test suggest that our main
results are unlikely to be driven by auditors forming expec-
tations about tax adjustments.
Alternative measure for Complexity
To take into account the possibility that our complexity
measure captures something other than complexity, we
run two robustness checks where we isolate only the dif-
ferences in characteristics of the task. First, we estimate
our regression models with N:Taxes (the number of taxes
audited) and/or TaskSize (period audited) as additional
control variables. Controlling for N:Taxes and/or TaskSize
keeps the number of taxes and the period audited constant,
leaving only differences in tax types in our Complexity
Table 6
Joint moderating role of task complexity and experience on auditor effort-
performance relationship.
TaxAdj
Constant 1.959
???
(4.29)
Experience À0.020
(0.09)
Effort
HTC
1.465
???
(12.023)
Experience ÂEffort
HTC
0.249
???
(3.418)
Effort
LTC
1.597
???
(13.738)
Experience ÂEffort
LTC
0.177
??
(2.492)
FirmSize À0.834
???
(43.605)
BTD 0.471
???
(7.063)
ForeignSales À0.275
???
(3.733)
BigOf?ce 0.554
???
(7.887)
MillsLambda 1.041
???
(5.176)
Industry FE Included
Audit Year FE Included
N 15,392
Uncensored N 3983
Wald v
2
3641.59
Prob. > v
2
0.00
This table presents results from estimating an OLS model for tax adjust-
ments. The sample contains 15,392 audits of 2002 and 2003 tax records
performed by the Croatian Tax Administration in the 2002–2006 period.
All data are provided by the Tax Administration of the Republic of Croatia.
The dependent variable, TaxAdj, is the natural log of the tax adjustment
scaled by the average of sales in the years 2002 and 2003 (see text). Effort
is the natural log of the duration of an audit (measured in auditor-days).
Effort
HTC
equals Effort when Complexity equals one, and zero otherwise.
Effort
LTC
equals Effort when Complexity equals zero, and zero otherwise.
Complexity is a binary variable that equals one if an audit includes at least
VAT, corporate income tax, and personal income tax; and zero otherwise
Experience is a binary variable that equals zero if one of the auditors is a
junior auditor, and one otherwise. FirmSize is the natural log of the mean
of 2002 and 2003 annual sales of ?rms audited. BTD is the mean of the
2002 and 2003 annual book-tax difference of ?rms audited. ForeignSales is
a binary variable that equals one if the audited ?rm had foreign sales in
either 2002 or 2003, and zero otherwise. BigOf?ce is a binary variable that
equals one if the regional of?ce is Split or Zagreb, and zero otherwise.
MillsLambda is the inverse Mills ratio resulting from the ?rst stage
regression with a binary variable equal to one if irregularities were
detected and zero otherwise as the dependent variable, and the inde-
pendent variables from Model (1) plus TaskSize; TaskType (a VAT dummy),
and N:Taxes in place of the Complexity and Loss dummy variable that
equals one if the ?rm had a loss and zero otherwise.
?
Two tailed statistical signi?cance at the 10 percent levels, respectively.
??
Two tailed statistical signi?cance at the 5 percent level.
???
Two tailed statistical signi?cance at the 1 percent level.
19
These arguments are based on informal conversations we have had
with the Croatian tax auditors.
20
Another potential proxy for the expectation of the amount of tax
adjustment would be prior year tax adjustments. However, in our setting,
?rms are not systematically audited every year and even if this occurs, the
audits are not likely to cover the same taxes.
W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510 507
0
5
1
0
1
5
2
0
P
e
r
f
o
r
m
a
n
c
e
0 2 4 6 8 10
Effort
Simple Task Complex Task
Panel A: Task Complexity
0
5
1
0
1
5
2
0
P
e
r
f
o
r
m
a
n
c
e
0 2 4 6 8 10
Effort
Low Experience High Experience
Panel B: Auditor Experience
Fig. 2. Task complexity and auditor experience impact on the effort-performance relationship.
2
2
.
5
3
3
.
5
4
4
.
5
P
e
r
f
o
r
m
a
n
c
e
0 2 4 6 8 10
Experience x Effort
Complex Task Simple Task
Fig. 3. Task complexity and auditor experience joint impact on the effort-performance relationship.
508 W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510
variable. Our results remain statistically signi?cant. Sec-
ond, we limit the sample to only those audits with one
tax audited. We asked tax auditors if they had one tax to
audit which one would be the most complex. Their answer
was Corporate Income Tax, followed by VAT. Given that
our reduced sample of 10,172 observations contains only
92 Corporate Income Tax observations (45 in the sample
with tax adjustments), we use VAT as a proxy for a com-
plex task. Our results also remain statistically signi?cant.
Conclusion
Using a sample of 15,392 tax audits performed on 2002
and 2003 corporate tax ?lings performed by the Croatian
Tax Administration during the 2002–2006 period, we
provide external validation of the prior experimental liter-
ature results that auditor experience and effort increase,
while task complexity decreases audit performance. We
extend this literature by examining the impact of task
complexity and auditor experience on the relationship
between auditor effort and performance, proxied for by
the amount of tax adjustments scaled by average annual
sales. Based on arguments from the psychology literature
suggesting that more complex tasks require more cogni-
tive effort (Locke & Latham, 1990; Yeo & Neal, 2008), and
that there is a diminishing marginal impact of effort on
performance (Kanfer & Ackerman, 1989), we ?nd that
audit task complexity mitigates the positive relationship
between auditor effort and audit performance. We also
?nd that auditor experience enhances the impact of audi-
tor effort on performance, in line with the argument that
effort has a positive impact on performance only if the
individual has some experience with the task (Yeo &
Neal, 2004). Finally, we show that high experience
increases the impact of auditor effort on audit performance
more when complexity is high, consistent with the goal
setting theory arguing that effort and performance will
depend on whether the goals are challenging and attain-
able (Fried & Slowik, 2004).
Compared to prior (theoretical and experimental)
literature, our study offers advantages, but also has certain
limitations. Our study provides a real-life setting, offering
external validation, allowing us to analyze the perfor-
mance and the characteristics of an audit team, as opposed
to an individual as it is done in most experimental studies
Birnberg (2011). Due to the real-life setting, we are
however limited by the choice of variables, as well as the
quality of the proxies used. As we measure audit perfor-
mance by the tax adjustment resulting from a tax audit,
our study also contributes to the tax audit literature by
further explaining the determinants of a corporate tax
audit outcome. We provide additional insight on audit
and audit team characteristics that are associated with
tax adjustments.
Acknowledgements
The authors gratefully acknowledge ?nancial support
from HEC Paris and Fondation HEC, Investissements
d’Avenir (ANR-11-IDEX-0003/Labex Ecodec/ANR-11-
LABX-0047). Walid Alissa and Vedran Capkun are members
of GREGHEC, Unit CNRS UMR 2959. The authors are grate-
ful for the data provided by the Tax Administration of Min-
istry of Finance of Republic of Croatia. We are grateful for
the comments received at the University of Innsbruck, IE
Business School, and IPAG Business School as well as at
the 2012 American Accounting Association annual meet-
ing. We are especially thankful to Florian Hoos, Clive Len-
nox, Cedric Lesage, Martin Messner and Steve Salterio for
their valuable input.
References
Abdolmohammadi, M., & Wright, A. (1987). An examination of the effects
of experience and task complexity on audit judgments. The Accounting
Review, 62(1), 1–13.
Allee, K., Bhattacharya, N., Black, E., & Christensen, T. (2007). Pro forma
disclosure and investor sophistication: External validation of
experimental evidence using archival data. Accounting, Organizations
and Society, 32, 201–222.
Asare, S., & McDaniel, L. (1996). The effects of familiarity with the
preparer and task complexity on the effectiveness of the audit review
process. The Accounting Review, 71(2), 139–159.
Bedard, J. (1989). Expertise in auditing: Myth or reality? Accounting,
Organizations and Society, 14(1–2), 113–131.
Birnberg, J. (2011). A proposed framework for behavioral accounting
research. Behavioral Research in Accounting, 23, 1–43.
Bonner, S. (1991). Is experience necessary in cue measurement? The case
of auditing tasks. Contemporary Accounting Research, 8(1), 253–269.
Bonner, S. (1994). A model of the effects of audit task complexity.
Accounting, Organizations and Society, 19(3), 213–234.
Brown, S., & Leigh, T. (1996). A new look at psychological climate and its
relationship to job involvement, effort, and performance. Journal of
Applied Psychology, 81(4), 358–368.
Campbell, D. (1988). Task complexity: A review and analysis. The Academy
of Management Review, 13(1), 40–52.
Caramanis, C., & Lennox, C. (2008). Audit effort and earnings
management. Journal of Accounting and Economics, 45(1), 116–138.
Chang, J., Ho, J., & Liao, W. (1997). The effects of justi?cation, task
complexity and experience/training on problem-solving performance.
Behavior Research in Accounting Supplement, 98–116.
Chan, K., & Mo, P. (2000). Tax holidays and tax noncompliance: An
empirical study of corporate tax audits in china’s developing
economy. The Accounting Review, 75(4), 469–484.
Christen, M., Iyer, G., & Soberman, D. (2006). Job satisfaction, job
performance, and effort: A reexamination using agency theory.
Journal of Marketing, 70(1), 137–150.
Clement, M. (1999). Analyst forecast accuracy: Do ability, resources, and
portfolio complexity matter? Journal of Accounting and Economics, 27,
285–303.
Clement, M., Koonce, L., & Lopez, T. (2007). The role of task-speci?c
forecasting experience and innate ability in understanding analyst
forecasting performance. Journal of Accounting and Economics, 44(3),
378–398.
Cloyd, C. (1997). Performance in tax research tax: The joint effect of
knowledge and accountability. The Accounting Review, 72, 111–131.
Cloyd, B., Pratt, J., & Stock, T. (1996). The use of ?nancial accounting
choice to support aggressive tax positions: Public and private ?rms.
Journal of Accounting Research, 34(1), 23–43.
Cook, T., & Campbell, D. (1976). The design and conduct of quasi-
experiments and true experiments in ?eld settings. Handbook of
industrial and organisational psychology. Chicago, IL: Rand-McNally.
Dye, R. (1995). Incorporation and the audit market. Journal of Accounting
and Economics, 19(1), 75–114.
Fisher, S., & Ford, J. (1998). Differential effects of learner effort and goal
orientation on two learning outcomes. Personnel Psychology, 51(2),
397–420.
Francis, J., & Yu, M. (2009). Big 4 of?ce size and audit quality. The
Accounting Review, 84(5), 1521–1552.
Fried, Y., & Slowik, L. (2004). Enriching goal-setting theory with time: An
integrated approach. The Academy of Management Review, 29(3),
404–422.
Heckman, J. (1979). Sample selection bias as a speci?cation error.
Econometrica, 47(1), 153–161.
Hillegeist, S. (1999). Financial reporting and auditing under alternative
damage apportionment rules. The Accounting Review, 74(3), 347–369.
W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510 509
Hirst, D. E. (1994). Auditor sensitivity to earnings management.
Contemporary Accounting Research, 11(1), 405–422.
Jacob, J., Lys, T., & Neale, M. (1999). Expertise in forecasting performance
of security analysts. Journal of Accounting and Economics, 28(1), 51–82.
Kanfer, R., & Ackerman, P. (1989). Motivation and cognitive abilities: An
integrative/aptitude-treatment interaction approach to skill
acquisition. Journal of Applied Psychology, 74(4), 657–690.
Kerlinger, F. (1986). Foundations of behavioural research. New York, NY:
Holt, Rinehart, and Winston.
Lennox, C. S., Francis, J. R., & Wang, Z. (2012). Selection models in
accounting research. The Accounting Review, 87(2), 589–616.
Libby, R. (1981). Accounting and human information processing: Theory and
applications. Contemporary topics in accounting series. Englewood
Cliffs, NJ: Prentice-Hall.
Libby, R. (1985). Availability and the generation of hypotheses in
analytical review. Journal of Accounting Research, 23(2), 648–667.
Libby, R., Bloom?eld, R., & Nelson, M. (2002). Experimental research in
?nancial accounting. Accounting, Organizations and Society, 27,
775–810.
Libby, R., & Frederick, D. (1990). Experience and the ability to explain
audit ?ndings. Journal of Accounting Research, 28(2), 348–367.
Libby, R., & Luft, J. (1993). Determinants of judgement performance in
accounting setting: Ability, knowledge, motivation, and environment.
Accounting, Organizations and Society, 18(5), 425–450.
Lim, C., & Tan, H. (2010). Does auditor tenure improve audit quality?
Moderating effects of industry specialization and fee dependence.
Contemporary Accounting Research, 27, 923–957.
Locke, E., & Latham, G. (1990). A theory of goal setting and task
performance. New York, NY: Prentice Hall.
Louis, H., Sun, A., & Urcan, O. (2013). Do analysts sacri?ce forecast
accuracy for informativeness? Management Science, 59, 1688–1708.
Maynard, D., & Hakel, M. (1997). Effects of objective and subjective task
complexity on performance. Human Performance, 10(4), 303–330.
McGrath, J. (1981). Dilemmatics: The study of research choices and
dilemmas. American Behavioral Scientist, 25(2), 179–210.
Mills, L. (1998). Book-tax differences and internal revenue service
adjustments. Journal of Accounting Research, 36(2), 343–356.
Mills, L., & Sansing, R. (2000). Strategic tax and ?nancial reporting
decisions: Theory and evidence. Contemporary Accounting Research,
17(1), 85–106.
Mohd-Sanusi, Z., & Mohd-Iskandar, T. (2007). Audit judgment
performance: Assessing the effect of performance incentives, effort
and task complexity. Managerial Auditing Journal, 22(1), 34–52.
Murray, M. (1995). Sales tax compliance and audit selection. National Tax
Journal, 48, 515–530.
Pratt, J., & Jiambalvo, J. (1981). Relationship between leader behaviors and
audit team performance. Accounting, Organizations and Society, 6,
133–142.
Sacket, P., & Larson, J. (1990). Research strategies and tactics in industrial
and organisational psychology. In Hough Dunette (Ed.), Handbook of
industrial and organisational psychology. Palo Alto, CA: Consulting
Psychologist Press.
Scandura, T., & Williams, E. (2000). Research methodology in
management: Current practices, trends, and implications for future
research. Academy of Management Journal, 43, 1248–1264.
Simnett, R. (1996). The effect of information selection, information
processing and task complexity on predictive accuracy of auditors.
Accounting, Organizations and Society, 21(7), 699–719.
Simnett, R., & Trotman, K. (1989). Auditor versus model: Information
choice and information processing. The Accounting Review, 64(3),
514–528.
Tan, H., & Kao, A. (1999). Accountability effects on auditors’ performance:
The in?uence of knowledge, problem-solving ability, and task
complexity. Journal of Accounting Research, 37(1), 209–223.
Tan, H., Ng, T., & Mak, B. (2002). The effects of task complexity on
auditors’ performance: The impact of accountability and knowledge.
Auditing: A journal of Practice & Theory, 21(2), 81–95.
Weingart, L. (1992). Impact of group goals, task component complexity,
effort, and planning on group performance. Journal of Applied
Psychology, 77(5), 682–693.
Wood, R. (1986). Task complexity: De?nition of the construct.
Organizational Behavior and Human Decision Processes, 37, 60–82.
Yeo, G., & Neal, A. (2004). A multilevel analysis of effort, practice, and
performance: Effects; of ability, conscientiousness, and goal
orientation. Journal of Applied Psychology, 89(2), 231.
Yeo, G., & Neal, A. (2008). Subjective cognitive effort: A model of states,
traits, and time. Journal of Applied Psychology, 93(3), 617–631.
510 W. Alissa et al. / Accounting, Organizations and Society 39 (2014) 495–510
doc_395447565.pdf