Original Article

Journal of Accounting,
Auditing & Finance
The Relative Importance of 1–26
ÓThe Author(s) 2020
Auditor Characteristics Versus Article reuse guidelines:
sagepub.com/journals-permissions
Client Factors in Explaining DOI: 10.1177/0148558X20953059
journals.sagepub.com/home/JAF

Audit Quality

Mara Cameran1 , Domenico Campa2, and Jere R. Francis3,4

Abstract
The U.K. listed firms are used to investigate whether auditor attributes (fixed effects for
audit firms, audit offices, and audit partners) add incrementally to baseline models with
client controls in explaining audit quality. We document that accounting firm fixed effects
add significantly to baseline models. To the extent an accounting firm can standardize its
audits, there should be no differences across engagements. However, we find significant
interoffice differences, and also significant inter-partner differences within offices. R2 analy-
ses, hierarchical linear models, LASSO (least absolute shrinkage and selection operator)
regressions, and R2 decomposition analyses all show that partners are the most important
auditor-related characteristic. To better understand the cause of partner variation, we test
a set of partner demographic variables (in lieu of partner fixed effects), but we find that
they explain little variation, once we control for firm and office differences. We conclude
that partner variation is important in explaining audit quality, but understanding the causes
requires going beyond existing publicly available demographic data.

Keywords
accounting firms, audit offices, engagement partners, earnings quality, going concern
reports, restatements

Introduction
This study investigates the importance of auditor characteristics in explaining audit quality,
after controlling for client factors. The audit quality is assessed through the quality of
audited earnings, proxied by abnormal accruals and financial restatements, and the likeli-
hood an auditor issues a going concern audit report. The construct of earnings quality can
be measured in multiple ways, and we focus on two of the most commonly used metrics:
abnormal accruals and restatements (Dechow et al., 2010). Audit quality is inferred from

1
Bocconi University, Milan, Italy
2
International University of Monaco, INSEEC Research Group, Principality of Monaco
3
Maastricht University, The Netherlands
4
University of Technology Sydney, New South Wales, Australia

Corresponding Author:
Mara Cameran, Department of Accounting, Bocconi University, Via Rontgen 1, 20136 Milan, Italy.
Email: [email protected]
2 Journal of Accounting, Auditing & Finance

higher quality audited earnings (smaller accruals and fewer restatements) and more going
concern audit reports, all else equal. However, also the neutral term ‘‘audit outcomes’’ is
used throughout the article to refer to these empirical observables (DeFond & Zhang, 2014;
Francis, 2011).1
In contrast to prior research, our study adopts a more comprehensive approach to assess
the relative importance of clients and auditors. To do this, we begin with a reduced-form
baseline model of client factors (only) and then conduct a series of econometric analyses to
determine whether models with auditor characteristics add incrementally to the explanatory
power of the reduced-form baseline model. Specifically, we start by introducing sequen-
tially three levels of auditor variables: audit firm fixed effects, engagement office fixed
effects, and individual partner fixed effects. Prior studies have examined in a more piece-
meal manner the effects of audit firm characteristics, typically large versus small firms
(Big 4/non–Big 4), the effects of interoffice differences, and more recently the effects of
individual partner characteristics using publicly observable demographic variables.
However, this body of research has not systematically investigated the incremental effects
of firms, offices, and partners in the kind of nested analysis as in our study. Instead, prior
studies typically focus on just one factor at a time such as certain accounting firm differ-
ences (Lennox & Pittman, 2011), or certain engagement office differences (Francis & Yu,
2009), or the effects of certain partner characteristics (Kallunki et al., 2019; Lo et al.,
2019), while ignoring other auditor characteristics that may also be important.
The research design in our study requires multi-year panel data for accounting firms,
audit offices, and engagement partners. Data on accounting firms and the locations of the
engagement offices that issue audit reports are typically available in most countries and are
obtained from audit reports. However, engagement partner data are relatively new in most
countries. The United Kingdom adopted a partner identification disclosure rule beginning
in 2009, which provides the necessary partner-level panel data, and for this reason our
study uses a U.K. sample of listed firms for the period 2009–2015.2
In audit research there is a limited set of empirically observable auditor characteristics
that has been used to measure accounting firm, audit office, and engagement partner differ-
ences. For this reason, we test auditor characteristics more broadly with a fixed effects
framework. A set of fixed effect variables, such as accounting firm indicator variables, will
generally have greater explanatory power than the use of individual variables using avail-
able data such as accounting firm size or industry expertise. Such variables are typically
incomplete and can omit important characteristics that might not be observable empirically,
which is why fixed effects are a standard econometric control for omitted or unobservable
variables. A good example is cross-country research in financial economics where country-
level fixed effects typically explain more cross-country variance than country-specific vari-
ables such as gross domestic product (GDP) and institutional structures (Doidge et al.,
2004).
Following the standard designs in the auditing literature which takes an audit production
perspective, our research question is whether auditor-related fixed effects add significantly
to baseline models with client controls in explaining audit outcomes.3 We begin by testing
whether accounting firm fixed effects add significantly to the baseline models. This is the
logical starting point for the identification of auditor characteristics because an accounting
firm develops a standardized testing approach and implements a quality control system
designed to ensure compliance with the firm’s policies and a consistent testing approach
across all of its engagements. The first null hypothesis is that there are no interfirm
Cameran et al. 3

differences, that is, accounting firm fixed effects do not add significantly to the reduced-
form baseline model of client factors.
We next test whether there are systematic interoffice differences, incremental to the con-
trol for accounting firm fixed effects. This design determines whether there are ‘‘within
firm’’ differences across an audit firm’s engagement offices. Accounting firms develop
standardized testing approaches and implement quality control systems to ensure consistent
audit quality across engagements. However, accounting firms are organized as networks of
decentralized offices for the delivery of audits to clients, which also means that an account-
ing firm’s testing procedures and control systems are implemented at the decentralized
office level (albeit with oversight). The decentralized office structure means there is the
potential for systematic interoffice variation in the firm’s audits. Thus, the second null
hypothesis is that there are no interoffice differences within firms, after controlling for
accounting firm fixed effects.
Finally, the third null hypothesis is that there are no significant inter-partner differences
on audit quality, after controlling for the effects of accounting firm and office fixed effects.
A partner typically leads an engagement team out of a specific office to service the clients
of that office. The test determines whether there is inter-partner variation within an office
of an accounting firm.4
An F-test analysis of changes in the ‘‘R2-within’’ of our models provides evidence that
accounting firm fixed effects add significantly to the reduced-form baseline model.5 We
also find significant interoffice differences which appear to play a larger role in explaining
audit outcomes than do interfirm differences. Finally, we report significant inter-partner
variation and document that partner effects explain more of the variation in audit quality
than the combined effect of firms and offices.
To assure that these findings are not just driven by the mere introduction of new indepen-
dent variables at each stage, we investigate our research question employing additional
econometric techniques. First, we use hierarchical linear modeling (HLM) that explicitly con-
trols for the nested structure of the accounting firm data in our study, that is, partners are
located in specific offices, and specific offices are part of a specific accounting firm and
obtain consistent results (Chang et al., 2018; Nakagawa & Schielzeth, 2013). Next we use a
penalized regression approach, more precisely LASSO (least absolute shrinkage and selection
operator) regressions, which perform both variable selection and regularization to enhance
the prediction accuracy and interpretability of the statistical model it produces.6 This analysis
indicates once more that audit partners are the most important auditor-related characteristic in
explaining audit quality. We also obtain similar results when analyzing Big 4 clients and
non–Big 4 clients separately under all of the econometrics approaches employed.
The final analysis substitutes publicly available partner demographic variables for part-
ner fixed effects to better understand specific factors that may drive partner differences.
Our analysis examines demographic variables commonly used in previous audit partner lit-
erature such as gender, type of degree, years of experience (which may also proxy for age),
and busyness of the auditor (number of clients) as well as variables not previously investi-
gated such as the quality of the university attended. The amount of variance explained by
the set of specific partner variables in our study is quite small, much less than that
explained by partner fixed effects, and this suggests that other (unknown) partner character-
istics drive the inter-partner differences in our study.7
The results of our study should be of interest to accounting firms and regulators as both
groups are interested in audit quality and gaining a better understanding of the drivers of
audit quality. The findings are strongly suggestive that inter-partner variation is the most
4 Journal of Accounting, Auditing & Finance

important audit factor that affects audit quality. However, our findings also indicate the
insufficiency of using solely publicly available partner demographic variables such as
gender, age, and experience. The point is not to criticize such studies, but rather to empha-
size the need to go beyond such data to understand what it is about partners that matter.
One potentially useful direction to better understand the causes of inter-partner variation
would be to draw on findings in the organizational behavior literature on job performance.
This research finds that the main driver of job performance is a person’s cognitive ability.
The only auditing study of cognitive ability we are aware of is Kallunki et al. (2019) who
find that IQ data from mandatory military service are correlated with audit quality for a
sample of male Swedish auditors. In addition to ability, personality characteristics also
have an important influence on job performance (Schmitt, 2014). Factors in the standard
‘‘big five’’ model of personality traits are openness to experience, conscientious, extraver-
sion, agreeableness, and neuroticism or emotional stability (Barrick & Mount, 1991). It is
also possible that audits are affected by engagement partners’ attitudes to risk and their
risk-taking behavior (Morgeson et al., 2010).
Another aspect of job performance that could be important in the auditing context is the
effect of audit partner leadership behaviors on audit engagements. Studies in the organiza-
tional behavior literature have established an association between leadership and perfor-
mance (e.g., Burke et al., 2006), as well as a link between personality and leadership
(Church et al., 2019). In addition, there is emerging team science research that may be rele-
vant to understanding the performance of partner-led engagement teams (Driskell et al.,
2018; Kozlowski, 2018).
The above discussion makes clear that the way forward in understanding partner effects is
to learn more about audit partners and the organizational structures in which they operate.
There is some emerging research that points in this direction. For example, Bol et al. (2018)
combine an instrument on tacit knowledge with proprietary accounting firm data and docu-
ment that accounting firms place a value on tacit knowledge, as evidenced by higher promot-
ability assessment, better annual assessments, and payment of cash bonuses. Tacit knowledge
is important in building the human capital of individual auditors and audit firms and should
lead to higher quality audits. Grabner et al. (2020) extend this research with proprietary data
from a Big 4 accounting firm, and a survey of its partners, to document that tacit knowledge
of partners is positively associated with the firm’s own internal assessment of audit quality.
In another study, Hardies et al. (2020) use an instrument to measure individual partner per-
sonality traits for a partner sample from a large accounting firm and then link these traits to
auditor skepticism and the auditor’s propensity for skeptical behaviors. Professional skepti-
cism is emphasized in audit standards and is considered by regulators to be a driver of audit
quality. As these studies illustrate, gaining a deeper understanding of audit partner behaviors
will benefit from the support and participation of accounting firms.8
The remainder of the study proceeds as follows. The next section reviews prior literature
and develops the study’s hypotheses. The research design and sample are then discussed,
followed by the empirical results and the study’s conclusions.

Hypothesis Development and Methods

Testing Framework Using Nested Models
Auditing standards and legal/regulatory requirements in most countries require accounting
firms to have firm-wide policies to control and monitor the quality of audits (Bedard et al.,
Cameran et al. 5

2008). The Public Company Accounting Oversight Board (PCAOB) regulates audit quality
for listed companies in the United States through its Statements on Quality Control,
denoted QC in the PCAOB standards. In countries that have adopted standards of the
International Accounting and Auditing Assurance Board, such as the U.K. sample we use,
the quality control standards are qualitatively similar to the broad principles of the PCAOB
and are set out in International Standard on Quality Control 1.9
Auditing standards and audit regulations set a minimum standard for accounting firms,
but there is considerable evidence of systematic interfirm differences in audits. The main
findings from the past literature suggest that larger firms (Big 4) and auditors with greater
industry expertise do better audits (e.g., Balsam et al., 2003; Cahan & Sun, 2015; Craswell
et al., 1995). Despite the importance of accounting firm characteristics documented in the
prior literature, extant research has not asked the explicit question we investigate, which is
whether an expanded model with accounting firm fixed effects adds significantly to a
reduced-form baseline model of client controls. Moreover, there is a very limited number
of specific observable accounting firm attributes and characteristics that have been tested,
and for this reason we believe accounting firm fixed effects can provide a better assessment
of the degree of interfirm variation. This discussion leads to the first hypothesis, stated in
null form as follows:

Hypothesis 1 (H1): Accounting firm fixed effects do not add significantly to a base-
line model of client controls in explaining audit quality.

To test H1 for the existence of systematic interfirm differences, we empirically estimate

the baseline model in Equation (1) for observable proxies for audit quality, denoted as Y:

Y = vector of client controls: ð1Þ

Next, we then estimate an expanded model in Equation (2) which allows for the possibility
of significant interfirm differences (non-zero fixed effect coefficients):

Y = vector of client controls + accounting firm fixed effects: ð2Þ

An F test determines whether the expanded model with accounting firm fixed effects
adds significantly to the reduced-form model in Equation (1). If accounting firm differences
do not exist, then the fixed effect coefficients are zero and the model in Equation (2) col-
lapses to the reduced-form model in Equation (1).10
The second research question asks whether there are interoffice differences within an
accounting firm that affect audit outcomes. Existing studies document that office character-
istics such as office size and the office’s city-specific industry expertise are important in
explaining audit quality (e.g., Choi et al., 2010; Ferguson et al., 2003; Francis et al., 2014;
Francis & Yu, 2009). However, these studies have typically used restricted samples of Big
4 auditors and do not explicitly control for accounting firm fixed effects which means that
it is possible that the interoffice effects are simply capturing the effect of omitted account-
ing firm variables (e.g., Choi et al., 2010; Francis et al., 2017; Francis & Michas, 2013).
This discussion leads to the second hypothesis, which is stated in null form as follows:
6 Journal of Accounting, Auditing & Finance

Hypothesis 2 (H2): An expanded model with audit office fixed effects does not add
significantly to a reduced-form model with client controls and accounting firm
fixed effects in explaining audit quality.

We formally test H2 by assessing whether the expanded model in Equation (3) with
office fixed effects adds to the model in Equation (2). By controlling for accounting firm
fixed effects, H2 tests whether there are interoffice differences within an accounting firm:

Y = vector of client controls + accounting firm fixed effects + office fixed effects: ð3Þ

An F test determines whether the expanded model in Equation (3) with audit office fixed
effects adds significantly to the reduced-form nested model in Equation (2). If an account-
ing firm’s control system achieves consistency across all of its engagements, the coeffi-
cients of audit office fixed effects will be zero and we would be unable to reject the null
hypothesis in H2.
The third research question asks whether there are inter-partner differences, after con-
trolling for both interfirm and interoffice differences. Recent studies find evidence that
partner characteristics may have an effect on audit outcomes (Cahan & Sun, 2015;
Kallunki et al., 2019; Knechel et al., 2015; Lo et al., 2019). Audit partner research has
emerged in recent years due in part to the growing availability of partner data in many
countries (Lennox & Wu, 2018). These studies find evidence that publicly available partner
demographic data such as gender, age, experience, workload or busyness, industry exper-
tise, risk preferences, and partner tenure with the client are associated with audit quality.11
However, given the significant findings in the literature on the effects of both accounting
firms and practice offices, it is clearly necessary to control for these effects in order to iso-
late the incremental effect (if any) of audit partners in explaining audit quality. However,
the extant partner literature does not typically control for systematic interfirm or interoffice
differences, which makes it difficult to make unambiguous inferences about the idiosyn-
cratic effect of audit partners.12 This leads to the third hypothesis, stated in null form as
follows:

Hypothesis 3 (H3): An expanded model with engagement partner fixed effects does
not add significantly to a reduced-form model with client controls plus accounting
firm and audit office fixed effects in explaining audit quality.

We formally test H3 by assessing whether the expanded model in Equation (4) with
partner fixed effects adds to the reduced-form nested model in Equation (3) reported above
that includes client controls plus accounting firm and audit office fixed effects:

Y = vector of client controls + accounting firm fixed effects + office fixed effects
+ engagement partner fixed effects: ð4Þ

By controlling for both accounting and office fixed effects, H3 explicitly tests whether
there are inter-partner differences within an office of an accounting firm. An F test deter-
mines whether the expanded model in Equation (4) adds significantly to the reduced-form
nested model in Equation (3). This testing approach rules out that omitted variables for
accounting firms and/or engagement offices explain cross-sectional differences in audit out-
comes by individual signing partners.
Cameran et al. 7

Sample and Variables

The study uses U.K. panel data for listed firms from two primary databases: FAME and
DataStream. The time frame is the 7-year period from 2009, that is, the first year the U.K.
regulation required disclosure of the audit engagement partner, through the 2015 fiscal
year. The variables needed for the empirical models (financial accounting data, accounting
firm name, and audit signing partner) come from FAME.13 If a company has missing
values in the FAME database, we use data from DataStream, if available. DataStream is
also used to identify restatements, as that information is not available on FAME. The name
and location of the audit firm, engagement office, and signing partner are obtained from
the audit report in the client’s annual report and publicly available online. When we could
not find the online report, we use FAME as it also provides the original annual reports of
companies, including the audit report.
To construct the sample, an accounting firm must have at least two unique offices, with
the further requirement that an office must have at least two unique audit partners. These
are the minimal conditions for estimating audit office fixed effects and individual partner
fixed effects. The initial sample is 879 unique partners for 1,091 unique clients. Once we
exclude firms missing the necessary financial data as well as those accounting firms and
offices that did not meet the above screens, the final sample is 5,411 client firm-year obser-
vations, consisting of 1,025 unique clients that are audited by 15 accounting firms with 135
unique offices and 665 unique signing partners. In the subsample of Big 4 clients, there are
3,579 firm-year observations, consisting of 628 unique clients that are audited by 70 unique
offices and 450 unique signing partners. The non–Big 4 subsample has 1,832 firm-year
observations, 397 unique clients audited by 11 accounting firms with 65 unique offices and
215 unique signing partners.14
Table 1 has detailed information on the accounting firms in the study. Big 4 (non–Big
4) firms have a median of 200 (26) listed clients in 17 (3) engagement offices, with a
median of 5 (3) unique signing partners per office, and a median of 2 (2) clients per signing
partner.

Empirical Models
The main research design uses a fixed effects framework adapted from the managerial style
analysis in Bertrand and Schoar (2003). As explained above, our first statistical approach
uses a series of nested models to assess the sequential and incremental fixed effects of
accounting firms, offices, and engagement partners in explaining audit quality. The depen-
dent variables in the study are observable audit outcomes from which audit quality is
inferred: the quality of audited earnings (EQ) and the auditor’s going concern reporting
(GC). Better audits are inferred from higher quality audited earnings and more going con-
cern reports, all else equal.
The empirical models in the study, including fixed effects for accounting firms, audit
offices, and engagement partners, are specified as follows:

EQ = a + b1 Sizeit + b2 Levit + b3 DLevit + b4 Cfoit + b5 Roait + b6 Lossit + b7 Retit

+ b8 Newfinit + Sgk Yeark + Skl Industryl + Slm AccountingFirmm ð5Þ
+ Sdn Officen + Sxo Partnero + eit :
8 Journal of Accounting, Auditing & Finance

Table 1. Descriptive Statistics for U.K. Accounting Firms, Audit Offices, and Engagement Partners
(2009–2015).

Variable M Q1 Median Q3 SD
Panel A: Big 4 clients
Number of engagement offices per accounting firm 17.750 12.500 17.000 23.000 6.291
Number of listed clients per accounting firm 183.750 146.000 200.000 221.500 54.243
Number of listed clients per engagement office 11.592 3.000 5.000 9.000 21.446
Number of signing partners per accounting firm 124.500 90.000 124.500 138.000 32.588
Number of signing auditors per engagement office 8.648 3.000 5.000 8.000 13.627
Number of unique listed clients per signing partner 3.009 1.000 2.000 4.000 2.241
Panel B: non–Big 4 clients
Number of engagement offices per accounting firm 6.000 2.000 3.000 9.000 6.277
Number of listed clients per accounting firm 42.182 5.000 26.000 53.000 52.579
Number of listed clients per engagement office 7.788 2.000 4.000 9.000 11.826
Number of signing partners per accounting firm 21.182 4.000 11.000 27.000 24.807
Number of signing partners per engagement office 4.470 2.000 3.000 5.000 5.359
Number of unique listed clients per signing partner 3.431 1.000 2.000 4.000 3.303

GC = a + b1 Sizeit + b2 Levit + b3 DLevit + b4 Cfoit + b5 Roait + b6 Lossit

+ b7 Retit + b8 Newfinit + b9 Investmentit + b10 Bankruptcy Riskit
+ b11 Ageit + b12 National MKT Leadit + b13 Local MKT Leadit
ð6Þ
+ b14 Aud Off Sizeit + b15 Client Influenceit + Sgk Yeark
+ Skl Industryl + Slm AccountingFirmm + Sdn Officen + Sxo Partnero + eit :

Variables are defined in the appendix.

The quality of audited earnings (EQ) in Equation (5) is tested with two variables: (a)
performance-adjusted abnormal working capital accruals (ABWC), with controls for the
firm’s contemporaneous performance using return on assets (Kothari et al., 2005)15 and (b)
restatements (RESTATE), which is an indicator variable coded 1 if a firm-year financial
statement is subsequently restated. The going concern model is estimated with an indicator
variable coded 1 if the auditor issues a going concern report (GC).
All empirical models in the study include a set of client control variables for other fac-
tors that can affect earnings quality and going concern reporting. All models have common
control variables for firm size, company leverage, operating cash flow, accounting return
on assets (profitability), stock market returns, the presence of losses, new external financ-
ing, and dummy variables for year and industry (Dechow et al., 1995; Gul et al., 2013;
Johnson et al., 2002; Kothari et al., 2005).
The going concern model includes additional firm-level controls for the age of the firm
(younger firms are riskier), the level of investments (a proxy for liquidity), and client bank-
ruptcy risk (Zmijewski, 1984). We also control for auditor characteristics based on prior
research. The auditor’s national-level and office-level industry leadership controls for sys-
tematic differences in audits conducted by industry experts (Krishnan, 2003). Audit office
size is included to control for its impact on audit quality (Choi et al., 2010; Francis & Yu,
2009). Finally, a variable that measures the relative size of each client to an office is
included to take into account the relative importance of a client to an office, and the
Cameran et al. 9

incentives and risks based on relative client size (Carcello & Li, 2013). As a sensitivity
analysis, we drop these auditor controls because they may cause the understatement of
accounting firm and/or engagement office fixed effects. However, the results are not
affected by their inclusion or exclusion.
All models are estimated using ordinary least squares (OLS) regression with fixed
effects and, given the use of panel data, we use robust standard errors that are clustered by
unique audit client to correct for heteroskedasticity and serial dependence associated to
panel data. The restatement and going concern models have a dichotomous-dependent vari-
able but are estimated using linear probability models (OLS), in line with other studies that
have used fixed effect approach (Gul et al., 2013).16 After we add incrementally the audit
firm, audit office, and partner fixed effects, we measure the increase in the R2-within of the
models and perform an F test.
To assure our findings are not simply the mechanical result of the inclusion of additional
independent variables at each stage, we adopt two additional econometric approaches.
First, we repeat the tests described above for Equations (1) through (4) using HLM which
explicitly controls for the nested structure of the underlying data through random effects
for accounting firms, audit offices, and engagement partners that are unexplained by OLS
(Nakagawa & Schielzeth, 2013). These models are reported later in the article and the
results are consistent with the OLS estimations. Furthermore, we use a penalized regression
approach, in particular a LASSO regression model, to assess the value that individual audit
firm, office, and partner fixed effects potentially add to a baseline model, using Equation
(1) as the starting point. The results, explained and reported later in the article, provide cor-
roborative support to our findings from OLS and HLM.17

Results
Descriptive Statistics
Table 2 reports detailed descriptive statistics for all variables separately for the Big 4 and
non–Big 4 subsamples. For Big 4 (non–Big 4) clients, the mean of ABWC is 6.6 (11.2)
percent of lagged total assets. There are 31% of Big 4 client-years with subsequent restate-
ments, compared with 24.6% for non–Big 4 clients. The restatement rates are similar to
other U.K. studies (Campa & Donnelly, 2016). Going concern reports are issued for 5.6%
of Big 4 client-years, and 13.1% of non–Big 4 client-years. Control variables are compara-
ble with other U.K. studies (Carcello & Li, 2013). Big 4 clients are larger, more highly
leveraged, have lower bankruptcy risk, and report fewer losses compared with non–Big 4
clients.18

Pooled Sample OLS Results

The reduced-form baseline models in Equation (1) are estimated with client controls but
without any auditor-related fixed effects. Untabulated results indicate that all models are
significant at p \ .01, with an R2-within ranging between 2% and 12%. Results for individ-
ual variables are consistent with prior research.
H1 tests whether accounting firm fixed effects add significantly to the reduced-form
baseline model with client controls. The incremental effect of accounting firm fixed effects
is reported in subpanel 1 of Table 3. The F-ratios range in value from 2.46 to 42.02 and
are significant at p \ .01 for all three models, which rejects the null hypothesis that the set
10 Journal of Accounting, Auditing & Finance

Table 2. Firm-Year Descriptive Statistics for the U.K. Big 4 and Non–Big 4 Samples (2009–2015).

Variables N M Q1 Median Q3 SD
Panel A: Big 4 clients
ABWC 3,579 0.066 0.015 0.036 0.077 0.098
RESTATE 3,579 0.310 0.000 0.000 1.000 0.463
GC 3,579 0.056 0.000 0.000 0.000 0.230
Size 3,579 12.646 11.241 12.630 14.070 2.069
Lev 3,579 0.442 0.276 0.432 0.574 0.250
DLev 3,579 –0.003 –0.037 –0.004 0.029 0.115
Cfo 3,579 0.057 0.025 0.080 0.137 0.184
Roa 3,579 –0.008 –0.004 0.040 0.081 0.250
Loss 3,579 0.262 0.000 0.000 1.000 0.440
Ret 3,579 0.172 –0.152 0.116 0.402 0.533
Newfin 3,579 0.921 1.000 1.000 1.000 0.270
Investment 3,579 0.253 0.086 0.160 0.313 0.251
Bankruptcy_Risk 3,579 –1.790 –2.891 –2.016 –1.098 2.086
Age 3,579 22.231 8.000 12.000 25.000 24.620
Nation_MKT_Lead 3,579 0.357 0.000 0.000 1.000 0.479
Local_MKT_Lead 3,579 0.054 0.000 0.000 0.000 0.226
Aud_Off_Size 3,579 19.116 16.854 19.254 21.596 2.714
Client_Influence 3,579 0.117 0.014 0.051 0.147 0.174
Inventory 3,579 0.087 0.003 0.031 0.135 0.119
AccRec 3,579 0.124 0.020 0.094 0.182 0.125
Subsid 3,579 5.770 2.646 4.359 6.928 5.108
BusySeason 3,579 0.704 0.000 1.000 1.000 0.457
Panel B: non–Big 4 clients
ABWC 1,832 0.112 0.024 0.060 0.122 0.158
RESTATE 1,832 0.246 0.000 0.000 0.000 0.431
GC 1,832 0.131 0.000 0.000 0.000 0.338
Size 1,832 9.713 8.743 9.860 10.772 1.615
Lev 1,832 0.404 0.163 0.327 0.522 0.361
DLev 1,832 0.006 –0.045 –0.002 0.050 0.211
Cfo 1,832 –0.081 –0.111 0.013 0.094 0.341
Roa 1,832 –0.204 –0.214 –0.008 0.056 0.537
Loss 1,832 0.520 0.000 1.000 1.000 0.500
Ret 1,832 0.121 –0.335 0.000 0.407 0.663
Newfin 1,832 0.849 1.000 1.000 1.000 0.358
Investment 1,832 0.291 0.079 0.191 0.413 0.277
Bankruptcy_Risk 1,832 –1.042 –3.191 –2.163 –0.764 4.151
Age 1,832 22.411 7.000 13.000 25.000 25.571
Nation_MKT_Lead 1,832 0.001 0.000 0.000 0.000 0.033
Local_MKT_Lead 1,832 0.005 0.000 0.000 0.000 0.070
Aud_Off_Size 1,832 14.857 13.633 14.631 17.118 1.981
Client_Influence 1,832 0.182 0.039 0.090 0.202 0.243
Inventory 1,832 0.065 0.000 0.005 0.078 0.116
AccRec 1,832 0.127 0.003 0.089 0.204 0.138
Subsid 1,832 2.484 1.414 2.449 3.162 1.780
BusySeason 1,832 0.628 0.000 1.000 1.000 0.483

Note. ABWC = performance-adjusted abnormal working capital accruals; RESTATE = restatement; GC = going
concern.
Cameran et al. 11

Table 3. Sequential Testing of the Incremental Significance of Accounting Firms, Engagement Offices,
and Partner Fixed Effects Over Baseline Models.

(A) (B) (C)

Dependent variable ABWC RESTATE GC
R2-within 3.08% 11.82% 2.28%
Obs. 5,411 5,411 5,411
1. Incremental significance of audit firm fixed effects (incremental to baseline models)
DR2-withinAF 0.99% 0.31% 0.58%
[F-Stat] [2.89***] [42.02***] [2.46***]
2. Incremental significance of engagement office fixed effects (incremental to audit firm fixed effects)
DR2-withinEO 4.11% 2.26% 3.51%
[F-Stat] [351.21***] [1,743.05***] [2,415.05***]
3. Incremental significance of individual partner fixed effects (incremental to audit firm and office fixed
effects)
DR2-withinIP 13.55% 12.36% 13.62%
[F-Stat] [4.2 3 109***] [1 3 109***] [6.4 3 105***]
Total R2-within with audit firm, engagement office, and individual partner fixed effects plus controls
R2-within 21.73% 26.75% 19.99%

Note. All models are estimated using ordinary least squares with fixed effects and robust standard errors clustered
by unique firm. ABWC = performance-adjusted abnormal working capital accruals; RESTATE = restatement; GC =
going concern; AF = audit firm; EO = engagement office; IP = individual partner.
***Significance at the 1% level.

of fixed effect coefficients is not significantly different from zero. We conclude that
accounting firm fixed effects add significantly to the baseline model, although the changes
in the R2-within of models are small, less than one percentage point.
H2 tests whether audit office fixed effects add significantly to the reduced-form baseline
model with client controls plus accounting firm fixed effects. This is reported in subpanel 2
of Table 3. The F-ratios are quite large and significant at p \ .01 in the three models
which rejects the null hypothesis. The changes in R2-within are still small, though ranging
between 2 and 4 percentage points.
H3 tests whether partner fixed effects add significantly to the reduced-form baseline
model with client controls plus the fixed effects of accounting firms and engagement
offices and is reported in subpanel 3 of Table 3. The F-ratios are very large and are sig-
nificant at p \ .01 in all models, which rejects the null hypothesis. The large F-ratio
values are suggestive that partner fixed effects are relatively more important than either
accounting firm or office fixed effects. The changes in R2-within are around 13 percent-
age points.

Separate Big 4 and Non–Big 4 Model Estimations

Table 4 reports the models in Table 3 estimated separately for the Big 4 and non–Big 4
subsamples. The Big 4 results (Panel A) and non–Big 4 results (Panel B) in Table 4 are
broadly consistent with the full sample findings in Table 3. In Panel A for the Big 4
sample, the F tests fail to reject the null hypothesis that the accounting firm fixed effect
coefficients are not significantly different from zero in any of the three models and
12 Journal of Accounting, Auditing & Finance

Table 4. Big 4 and Non–Big 4 Tests of Incremental Significance of Accounting Firms, Engagement
Offices, and Individual Partner Fixed Effects.

(A) (B) (C)

Dependent variable ABWC RESTATE GC
Panel A: Big 4 clients
R2-within 2.21% 13.09% 3.51%
Obs. 3,579 3,579 3,579
1. Incremental significance of Big 4 audit firm fixed effects (incremental to baseline models)
DR2-withinAF 0.05% 0.02% 0.31%
[F-Stat] [0.51] [0.16] [1.91]
2. Incremental significance of Big 4 engagement office fixed effects (incremental to audit
firm fixed effects)
DR2-withinEO 2.50% 1.85% 3.21%
[F-Stat] [7,105.89***] [7,685.94***] [207.74***]
3. Incremental significance of Big 4 individual partner fixed effects (incremental to audit firm and
office fixed effects)
DR2-withinIP 15.76% 13.02% 12.92%
[F-Stat] [15.501.29***] [1,367.91***] [2.3 3 107***]
Total R2-within with Big 4 audit firm, engagement office, and individual partner fixed effects plus controls
R2-within 20.52% 27.98% 19.95%
Panel B: Non–Big 4 clients
R2-within 5.55% 10.84% 10.28%
Obs. 1,832 1,832 1,832
1. Incremental significance of non–Big 4 audit firm fixed effects (incremental to baseline models)
DR2-withinAF 1.10% 0.55% 0.71%
[F-Stat] [2.40***] [7.14***] [1.29]
2. Incremental significance of non–Big 4 engagement office fixed effects (incremental to audit firm
fixed effects)
DR2-withinEO 2.08% 3.62% 2.42%
[F-Stat] [24.09***] [64.88***] [22.47***]
3. Incremental significance of non–Big 4 individual partner fixed effects (incremental to audit firm
and office fixed effects)
DR2-withinIP 11.64% 11.26% 12.57%
[F-Stat] [1,456.03***] [234.95***] [5,809.95***]
Total R2-within with non–Big 4 audit firm, engagement office, and individual partner fixed effects
plus controls
R2-within 20.37% 26.27% 25.98%

Note. All models are estimated using ordinary least squares with fixed effects robust standard errors clustered by
unique firm. ABWC = performance-adjusted abnormal working capital accruals; RESTATE = restatement; GC =
going concern; AF = audit firm; EO = engagement office; IP = individual partner.
***Significance at the 1% level.

therefore do not add significantly to the reduced-form baseline model without accounting
firm fixed effects. These results suggest there are no systematic differences, on average, in
audit quality across the Big 4 firms. However, results of the incremental tests of audit
offices and engagement partners indicate they add significantly to baseline models and
indicate the presence of interoffice differences within Big 4 firms, and inter-partner differ-
ences within Big 4 offices, as in Table 3 for the full sample estimations.
Cameran et al. 13

Panel B of Table 4 reports the non–Big 4 results and these results mirror the full sample
results in Table 3. F tests for all of the auditor fixed effects are significant (except for audit
firm fixed effects when we investigate GC), with the largest F ratios for the test of partner
fixed effects.

Additional Analyses
HLM. HLM is an alternative estimation approach used when data are nested. In our test set-
ting, auditor data are nested as follows: individual audit partners are nested in a specific
office, and offices are nested within a specific accounting firm. Chang et al. (2018) use
HLM with Taiwanese data and test for the effect of an accounting firm variable (account-
ing firm tenure) and an audit partner variable (lead partner engagement tenure). While the
data structure in Chang et al. (2018) is not a classic nested data set, compared with our
study, they do document that OLS results are not fully consistent with their HLM estima-
tions. Given their findings, and because our data are classically nested, we re-estimate the
models in Equations (1) through (4) using HLM that explicitly recognizes and controls for
the nested structure of the data. Specifically, HLM controls for the variance from random
effects of the nested data that are not explained by OLS methods (Nakagawa & Schielzeth,
2013). The HLM results are reported in Table 5.
Since HLM is a maximum likelihood model, it is a chi-square test that compares
Equations (1) through (4) to determine whether the likelihood ratio of expanded models is
significantly larger than the reduced-form baseline models. Panel A of Table 5 presents the
results for the full sample. The HLM results are consistent with the ordinary least squares
(OLS) evidence in Table 3. All of the chi-square tests are significant at p \ .01, which
indicates the expanded models add significantly to the reduced-form models. Specifically,
in subpanel 1 of Table 5, the chi-square tests indicate there are significantly larger likeli-
hood ratios for the expanded models with accounting firm fixed effects compared with the
reduce-form baseline model with only client controls.
Similarly, in subpanel 2 of Table 5 the chi-square tests show significantly larger likeli-
hood ratios in the expanded model with audit office fixed effects, compared with the
reduced-form model including firm-level controls and accounting firm fixed effects. In sub-
panel 3, we again observe significantly larger likelihood ratios in the expanded model with
engagement partner fixed effects compared with reduced-form models controlling for firm-
level factor as well as audit firm and audit office fixed effects. Note also that the chi-
square values are greatest when adding partner fixed effects (subpanel 3) and are smallest
when adding accounting firm fixed effects in subpanel 1. This is the same pattern observed
in Table 3 for the F ratios and is consistent with partner fixed effects adding relatively
more in explaining audit outcomes than either audit offices or accounting firms. The same
analysis, separately for Big 4 and non–Big 4 clients, is presented in Panels B and C of
Table 5. The results consistently highlight significantly larger likelihood ratios in the
expanded models when engagement partner fixed effects are added to reduced-form models
controlling for firm-level factors, audit firm, and audit office fixed effects.

LASSO regression. To make sure that the increase in the R2-within after the introduction of
partner fixed effects is not just due to the large number of independent variables added at
that stage, we employed a different statistical approach than those reported above, which
focuses on variable selection rather than on the change in the R2-within after the inclusion
14 Journal of Accounting, Auditing & Finance

Table 5. Hierarchical Linear Model Tests of Incremental Significance of Accounting Firms,

Engagement Offices, and Partner Fixed Effects.

(A) (B) (C)

Dependent variable ABWC RESTATE GC
Panel A: Pooled sample
1. Incremental significance of audit firm fixed effects (incremental to baseline models)
Chi-square 55.65*** 55.65*** 32.16***
[p-value] [.000] [.000] [.000]
2. Incremental significance of engagement office fixed effects (incremental to audit firm fixed effects)
Chi-square 236.63*** 236.63*** 198.88***
[p-value] [.000] [.000] [.000]
3. Incremental significance of individual partner fixed effects (incremental to audit firm and office
fixed effects)
Chi-square 1,100.96*** 864.33*** 850.85***
[p-value] [.000] [.000] [.000]
Panel B: Big 4 clients
1. Incremental significance of audit firm fixed effects (incremental to baseline models)
Chi-square 1.84 0.66 11.38**
[p-value] [.399] [.882] [.023]
2. Incremental significance of engagement office fixed effects (incremental to audit firm fixed effects)
Chi-square 178.88*** 90.60*** 130.78***
[p-value] [.000] [.008] [.000]
3. Incremental significance of individual partner fixed effects (incremental to audit firm and office
fixed effects)
Chi-square 565.60*** 589.30*** 526.66***
[p-value] [.000] [.000] [.000]
Panel C: Non–Big 4 clients
1. Incremental significance of audit firm fixed effects (incremental to baseline models)
Chi-square 21.44** 11.45 14.51
[p-value] [.011] [.323] [.151]
2. Incremental significance of engagement office fixed effects (incremental to audit firm fixed effects)
Chi-square 41.36 76.45*** 50.53
[p-value] [.368] [.000] [.146]
3. Incremental significance of individual partner fixed effects (incremental to audit firm and office
fixed effects)
Chi-square 250.25*** 260.59*** 287.87***
[p-value] [.000] [.000] [.000]

Note. ABWC = performance-adjusted abnormal working capital accruals; RESTATE = restatement; GC = going
concern.
**,***Significance at the 5% and 1% level, respectively.

of new variables. In particular, we use a penalized regression model, called LASSO regres-
sion, which selects the best set of explanatory variables from a range of potential covariates
(James et al., 2013).19 More specifically, this approach reduces the number of variables in
a regression model, keeping only those that have a strong effect on the dependent variable.
In our research setting that includes audit firm, audit office, and audit partner fixed effects,
we implement a penalized regression model that keeps only those auditor-related indicators
that contribute more significantly to explain the audit outcome variables. Accordingly,
using the LASSO regression command provided in STATA 16, we estimate a LASSO
Cameran et al. 15

Table 6. Category of the Auditor-Related Variables Selected by a Least Absolute Shrinkage and
Selection Operator Regression (in Addition to Firm-Level Controls).

(A) (B) (C)

Dependent variable ABWC RESTATE GC
Panel A: Pooled sample
Audit firms 3% 0% 8%
Engagement offices 24% 0% 52%
Individual partners 73% 100% 40%
Panel B: Big 4 clients
Audit firms 0% 0% 0%
Engagement offices 0% 0% 17%
Individual partners 100% 100% 83%
Panel C: Non–Big 4 clients
Audit firms 0% 0% 17%
Engagement offices 0% 0% 66%
Individual partners 100% 100% 17%

Note. ABWC = performance-adjusted abnormal working capital accruals; RESTATE = restatement; GC = going
concern.

regression for all audit outcomes, imposing that the client-level controls must always be
kept as the independent variables of the model, that is, our Equation (1), and setting,
instead, all auditor-related fixed effects as part of the pool of covariates from which the
LASSO regression has to select from and include in a potential extended model. The result
of this selection is reported in Table 6.
Panel A of Table 6 reports the results for the full sample. Column A of Table 6 indicates
that for abnormal accruals, 73% of the auditor-related fixed effects retained in the LASSO
regression are partner fixed effects. The results for restatements are even stronger since the
LASSO regression excludes all audit firm and office fixed effects, keeping only partner
fixed effects. Results are weaker when we examine going concern reporting since a
LASSO regression keep all types of auditor-related fixed effects in a potential extended
model, the majority of them being office effects. Partner fixed effects, however, still repre-
sent 40% of the auditor-related fixed effects selected by the LASSO regression.
Panels B and C of Table 6 report separately the results for Big 4 and non–Big 4 clients.
Results for abnormal accruals and restatements are similar between Big 4 and non–Big 4
clients and highlight the importance of engagement partners in explaining audit quality. In
these subsample analyses, only partner fixed effects are selected by a LASSO regression.
For going concern reporting, we observe some differences between Big 4 and non–Big 4
clients. For Big 4 clients, 83% of the auditor-related fixed effects kept by a LASSO regres-
sion are linked to individual partners. In contrast, for non–Big 4 clients the majority of the
auditor-related fixed selected by the LASSO regression are audit offices. Overall, the
LASSO regressions indicate that partners are consistently the strongest predictor of audit
outcomes, compared with audit firms and offices.

Analysis of partner demographic variables. Given that partner fixed effects are the domi-
nant auditor factor in explaining audit quality in Tables 3 through 6, to better understand
what drives the partner differences, we extend the analysis by exploring a set of specific
partner demographic variables based on publicly available data commonly used in partner
16 Journal of Accounting, Auditing & Finance

Table 7. Incremental Significance of the Set of Auditor Characteristics Over and Above Accounting
Firm and Office Fixed Effects, Versus the Incremental Significance of Partner Fixed Effects.

(A) (B) (C)

Dependent variable ABWC RESTATE GC
Panel A: Pooled sample (n = 4,658)
DR2-within from individual partner characteristics 0.25% 0.05% 0.08%
[F-Stat] [1.01] [0.48] [0.48]
DR2-within from partner fixed effects 13.27% 11.76% 17.82%
[F-Stat] [3,893.15***] [3.5 3 108***] [1.9 3 108***]
Panel B: Big 4 clients (n = 3,070)
DR2-within from individual partner characteristics 0.29% 0.08% 0.09%
[F-Stat] [1.23] [0.53] [0.44]
DR2-within from partner fixed effects 14.11% 11.90% 18.09%
[F-Stat] [1.1 3 106***] [18,478.65***] [1.7 3 108***]
Panel C: Non–Big 4 clients (n = 1,588)
DR2-within from individual partner characteristics 0.61% 0.00% 0.90%
[F-Stat] [0.86] [1.21] [2.16*]
DR2-within from partner fixed effects 11.88% 11.24% 13.17%
[F-Stat] [274.07***] [420.74***] [3.6 3 105***]

Note. All models are estimated using ordinary least squares with fixed effects and robust standard errors clustered
by unique firm). ABWC = performance-adjusted abnormal working capital accruals; RESTATE = restatement;
GC = going concern.
*,***Significance at the 10% and 1% level, respectively.

studies (e.g., Cahan & Sun, 2015). To do this, we collect the following publicly available
partner information: gender, quality of the university attended, type of degree, and the year
they became a Chartered Accountant (to calculate years of experience). These data come
from membership records of the U.K. professional bodies (i.e., Institute of Chartered
Accountants in England and Wales, Institute of Chartered Accountants of Scotland,
Chartered Accountants of Ireland, Association of Chartered Certified Accountants),
Bloomberg, the website ‘‘Company House’’ (available at beta.companieshouse.gov.uk),
websites of the accounting firms, and LinkedIn. We also use the listed company data in the
sample to measure partner busyness (number of listed clients). Missing data results in a
reduced sample of 544 partners compared with the full sample of 665: 365 partners in Big
4 firms, and 179 in non–Big 4 accounting firms.
Using these characteristics, we added to the baseline models the variables for partner
gender, rank of the university attended, whether a partner has a major in accounting, as
well as variables that measure partner experience and busyness and we test whether they
add whether they increase their explanatory power relative to baseline models. This analy-
sis is reported in Table 7.
The sample in Table 7 is smaller than that used in Tables 3 and 4 due to the lack of
demographic data for some partners. The results show that the set of partner variables does
not add significantly to reduced-form baseline models with client controls plus fixed effects
for accounting firms and audit offices. The F tests are all insignificant at the .10 level for
the pooled sample and for the separate Big and non–Big 4 samples. Using this smaller
sample, for the sake of comparability, we also test the incremental effect of partner fixed
effects, rather than partners’ individual characteristics, and the F tests are all significant
Cameran et al. 17

and consistent with the full sample estimations in Tables 3 and 4. We conclude that a set
of partner demographic variables does not add significantly to models over and above the
reduced models with controls for the fixed effects of accounting firms and offices.20
The partner literature does not typically control for either accounting firm or audit office
fixed effects. While our study does not include every partner characteristics in the existing
partner literature, for the set of publicly available partner variables we examine, the evi-
dence is suggestive the partner variables may not have much explanatory power over and
above the fixed effects of accounting firms and audit offices. Accordingly, the findings
raise questions about the interpretation and importance of partner variables in prior studies,
and it is unclear how incrementally significant the demographic variables are in these
studies.21
In conclusion, we have evidence that partner fixed effects explain far more than individ-
ual audit partner variables. This is also consistent with other fields such as financial eco-
nomics which shows that country fixed effects have far greater explanatory power than the
country-specific variables (Doidge et al., 2004). In the econometrics literature, fixed effects
are a classic control for omitted variables. In our context, the implication is that inter-partner
variation is largely driven by factors other publicly observable demographic variables such as
age, gender, and experience. At this stage, we do not know what the significant drivers of the
inter-partner variation are, but our study is important because it points to the need for more
in-depth research that moves beyond publicly available demographic data.22

R2 decomposition analysis. To further assess the relative importance of client factors versus
auditor characteristics in explaining audit outcomes, we undertake a variance decomposi-
tion analysis, derived from the theoretical work of Shapley (1953) and Owen (1977), based
on iterative combinations of sets of regressors to measure the incremental contribution of
each regressor (or set of regressors in our case) in explaining model R-squares.
Decomposition analysis is viewed as superior to R2 comparisons to understand the impor-
tance of sets of regressors relative to each other (e.g., see Huettner & Sunder, 2012; Sastre
& Trannoy, 2002).23 The evidence (not tabulated) indicates that, for the full sample, the
client control variables in the baseline model (1) are the dominant set of regressors and
account for 62%–67% of the explained variance in the three models. Audit partner fixed
effects are the next most important set of regressors, representing from 24% to 29% of the
explained variance in the three models. Engagement office fixed effects are the third most
important set of regressors, and account for 6%–9% of the explained variance. The least
important set of regressors are accounting firm fixed effects which account from 1% to 4%
of explained variance in most of the models. Similar evidence is observed if we separate
Big 4 and non–Big 4 clients.

Other robustness tests. A potential concern is the relatively small sample size of some
partner-year observations. For this reason, we impose an additional requirement that audit
partners must have at least two unique clients, and partners must have also a minimum of
10 client-year observations to be in the sample. This process increases the number of
partner-year observations, but it reduces the overall sample size to 3,443 firm-year observa-
tions, compared with 5,411 observations for the full sample. For this reduced sample, the
results are consistent with those reported in Table 3 and 4. Based on these additional tests
we have no reason to believe partner fixed effects are misestimated due to partner-year
sample sizes.
18 Journal of Accounting, Auditing & Finance

Fixed effects are a strong control for unobserved time-invariant firm characteristics, but
it also means that auditor-related fixed effect coefficients will have statistically significant
cross-sectional variation only to the extent there are auditor changes of some type in the
data, that is, when a client changes accounting firms and/or audit offices, or if the engage-
ment partner changes. In other words, if a client has exactly the same accounting firm,
same engagement office, and same audit partner over the entire sample period, then the
auditor-related fixed effects will have no explanatory power over and above client fixed
effects. In this respect, the test is less generalizable than tests without client fixed effects.24
Therefore, we re-estimate the models in Table 3 without imposing client fixed effects. The
results (not tabulated) are consistent with Table 3, although the changes in R2 are much
larger for the auditor characteristics, especially for audit office and engagement partner
fixed effects. However, we conclude that the overall tenor of the results in Table 3 are
robust to controlling for, and not controlling for, client fixed effects.
We also estimate an additional LASSO regression where we have included in the set of
potential covariates the personal auditor characteristics, in addition to audit firm, audit
office, and audit partner fixed effects. The LASSO regression continues to prioritize partner
fixed effects and the results of this analysis are in line with those reported in Table 6.
Finally, London is the largest city-specific audit market in the United Kingdom, and to
rule out the results are driven by a London effect we conduct two tests. First, we drop
those observations in London, and second we add a London indicator variable to all
models. Results from these two tests are consistent with the tabled results and indicate that
a London effect does not drive the results.

Conclusion
Prior research has explored the effects of audit firm characteristics on audit quality, such as
accounting firm size and the accounting firm’s industry expertise. Other studies have exam-
ined the effects of audit office characteristics and individual partner demographic variables
on audit outcomes. These studies typically focus on just one auditor characteristic and ignore
other auditor factors that might also affect audit quality. In contrast, our study takes what we
view as a more systematic and comprehensive approach to studying auditor effects.
To isolate the idiosyncratic effects of audit firms, offices, and partners, we use a series
of econometric tests to investigate whether auditor attributes, measured with fixed effects,
add incrementally to baseline models with client controls. We find that, overall, interfirm
differences, interoffice differences, and inter-partner differences are incrementally signifi-
cant. However, all of our econometric approaches consistently show that the inter-partner
differences are by far the most important auditor factor in explaining audit quality.
Given the importance of inter-partner differences on audits, we further analyze partner
demographic characteristics and find that when we use such variables in the models in lieu
of partner fixed effects, the incremental importance of partner variables (over and above the
effects of accounting firms and offices) is small and mainly insignificant. Thus, our analysis
concludes that the type of publicly available partner demographic data that are being used in
the partner research literature may not have much explanatory power in explaining audit
quality, once there is control for the fixed effects of accounting firms and audit offices.
Our findings will be of interest to accounting firms and regulators in the United
Kingdom, and to similar countries in the European Union, Australia, Canada, United
States, and other countries that have a well-developed accounting profession and similar
institutions and regulatory oversight. While our results should apply to other developed
Cameran et al. 19

countries with similar regulations, accounting professions, and audit markets, we acknowl-
edge there could be something unique about the United Kingdom that may limit the study’s
generalizability.
The results in our study are strongly suggestive that inter-partner variation is the single
most important auditor characteristic in explaining audit quality. However, it remains for
future research to identify exactly what are the most important partner characteristics that
explain the inter-partner variation we observe in our study. At the beginning of the article
we suggested the organizational behavior literature on job performance might provide some
useful directions in advancing our understanding of audit partner differences. This will
require access to accounting professions in accounting firms to go beyond the limited
understanding that appears to be the case with publicly available demographic partner data.

Appendix
Variable Definitions
Primary dependent variables (in alphabetical order).
ABWC: Absolute value of performance-adjusted ABWC (Kothari et al., 2005), where
ABWC = CA 2 ECA.
This variable is computed by first estimating the regression coefficients from the follow-
ing model:

1
CAt = b1 + b2 D Sales + b3 ROAt1 + et ,
TAt1

where CA is the net income before extraordinary items + depreciation and amortization—
operating cash flows, scaled by beginning of year total assets; ROA is the net income
before extraordinary items, divided by total assets.
The coefficient estimates are then used to measure expected current accruals (ECA),
controlling for performance:

^1 1 ^ 2 ½DSales  DAR + b
^ 3 ROAt1 ,
ECAt = b +b
TAt1

where AR is the accounts receivable.

GC: Indicator variable that takes 1 if the client receives a going concern report modification.
RESTATE: Indicator variable that takes 1 in the presence of restated annual reports.

Control variables (in alphabetical order).

Age: The number of years from company’s foundation to the year of the observation.
Aud_Off_Size: The logged total assets of clients audited by the audit office in the current
fiscal year.
Bankruptcy_risk: A measure of client’s risk based on the Zmijewski (1984) model.
Cfo: Operating cash flow divided by total assets.
Client_Influence: The natural logarithm of total client-firm assets scaled by the sum of
logged assets for all firms audited by the same audit office.
Investment: Total investment assets of the client (cash, cash equivalent, short-term, and
long-term investment) scaled by total assets.
20 Journal of Accounting, Auditing & Finance

Lev: Total debt divided by total assets.

Local_MKT_Lead: Indicator variable that takes 1 if the sum of the assets held by clients of
an audit office in a given industry is higher than the sum of the assets held by the clients of
any other audit office.
Loss: Indicator variable that takes 1 in the case of losses.
Nation_MKT_Lead: An indicator variable that takes 1 if the sum of the assets held by cli-
ents of an accounting firm in a given industry is higher than the sum of the assets held by
the clients of any other accounting firm.
Newfin: Indicator variable that takes 1 if the company issued new debt or equity during the
year.
Ret: Total annual market return for the company.
Roa: Net income divided by total assets.
Size: Natural logarithm of total assets.
DLev: Leverage in year t less leverage in year t 2 1.

Acknowledgments
We would like to acknowledge the support and encouragement of the Editor, Bharat Sarath, the
Associate Editor, Kannan Raghunandan, as well as an anonymous reviewer for her or his important
comments and suggestions. Furthermore, we thank Joe Gerakos, Nick Hallman, Inder Khurana, Mike
Minnis, Marleen Willekens, workshop participants at University of Technology Sydney and
University of Missouri, and conference participants at the inaugural research conference of the
Foundation for Audit Research (Nyenrode University, Netherlands), EIASM Workshop on Audit
Quality European Accounting Association Annual Congress, and the American Accounting
Association Annual Meeting.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/
or publication of this article.

Funding
The author(s) received no financial support for the research, authorship, and/or publication of this
article.

ORCID iD
Mara Cameran https://orcid.org/0000-0001-8480-9242
Notes
1. Dechow et al. (2010) point out that one particular measure of earning quality might be the most
appropriate metric to use (relative to others) in a specific research setting. Sarath (2016) makes a
similar argument about audit quality and suggests that no single measure or proxy is likely to be
relevant to all stakeholders and decision makers. For this reason, we follow prior research and
use multiple measures to proxy for the quality of earnings and audits (DeFond & Zhang, 2014).
2. The European Union (EU) Eighth Directive, issued in 2006, requires the audit engagement part-
ner to be identified in the audit report. The United Kingdom implemented the Eighth Directive
in 2009 and was one of the first EU countries to do so.
3. The Editor-in-Chief notes this design could also be used to investigate how client factors incre-
mentally affect the quality of audited earnings, after controlling for auditor characteristics.
Cameran et al. 21

4. Gul et al. (2013) is the only study we are aware of that estimates models with fixed effects (FEs)
for firms, offices, and partners. Their study examines small, mainly single office accounting
firms in China where it is common for audit reports to be signed by non-partners. However, Gul
et al. (2013) investigate a different research question and use a different testing approach.
Instead of a series of logically nested models to sequentially test the incremental effects of audit
firms, offices, and individuals, they conduct horse race tests to determine which of the three sets
of auditor FEs makes the greatest contribution to the total explanatory power of models. To do
this they estimate a full model (including FEs for audit firms, offices, and partners) and then
determine the reduction in R2 when dropping FEs for audit firms, then repeating the process by
comparing the full model to a reduced model without the audit office FEs, and finally comparing
the full model to a reduced model without partner FEs.
5. In panel data where there are longitudinal observations for the same firms, FEs represent the
firm-specific means. In panel data analysis, the term FEs estimator, also called the within estima-
tor, refers to an estimator for the coefficients in the regression model including those FEs (one
time-invariant intercept for each firm). In our study, we use FE (or within estimators) because
we are interested in analyzing the effect of variables that vary over time. Accordingly, the use of
FE explores the relationship between our predictors and the outcome variables within each
entity. FE remove the effect of those time-invariant characteristics so we can assess the net
effect of the predictors on our outcome variables (Torres-Reyna, 2007). Since we use the within
estimator, we investigate our research question through the analysis of the change in the
R2-within which measures how much of the variation in the dependent variable within entities is
captured by our models (i.e., how well our explanatory variables account for changes in the audit
outcomes within each of the entities over time). The within R2 is estimated under the assumption
that individual firms are fixed quantities, so their effects are removed from the model. We thank
the referee for suggesting this analysis.
6. We thank the referee for suggesting this approach.
7. Some current working papers on auditor-client matching suggest that matching effects can par-
tially explain audit engagement outcomes (e.g., Bills & Jensen, 2010; Cook et al., 2020; Li et al.,
2018). However, the focus in our study is not on matching and selection effects in the auditor
choice decision. We thank the referee for this insight and acknowledge that the matching condi-
tions between auditor and client may play some (unknown) role in explaining our results,
although the use of FEs in the paper is a standard econometric procedure to control for unknown
client factors that may systematically affect observable outcomes.
8. A good example of how collaboration between audit scholars and accounting firms can work
successfully is the initiative of the Foundation for Auditing Research in the Netherlands (https://
foundationforauditingresearch.org/en/). The Foundation was formed in 2015 by the largest Dutch
accounting firms and has been active in facilitating researcher access to accounting firms’ propri-
etary archival data, as well as access to accounting professionals for surveys and experiments.
To date, the Foundation has supported 19 research projects, including the above cited studies by
Grabner et al. (2020) and Hardies et al. (2020).
9. Quality Control for Firms that Perform Audits and Reviews of Financial Statement, and Other
Assurance and Related Services Engagements, available at: https://www.ifac.org/system/files/
downloads/ISQC_1_standalone_2009_Handbook.pdf.
10. Following Bertrand and Schoar (2003), F tests determine whether the FE variables are incremen-
tally significant in the model estimations. However, because Fee et al. (2013) question the valid-
ity of the F tests in Bertrand and Schoar (2003), we conduct counter-factual tests to establish
F-test validity in our setting as explained in the endnote 17.
11. Audit partner studies include Ke et al. (2014), Chen et al. (2016), Carey and Simnett (2006),
Chin and Chi (2009), Goodwin and Wu (2014, 2016), Hardies et al. (2015), Knechel et al.
(2015), Sundgren and Svanstrom (2014), Wang et al. (2015), and Porumb et al. (2017).
22 Journal of Accounting, Auditing & Finance

12. Aobdia et al. (2015) do not study audit quality, per se, but is one of the few studies other than
Gul et al. (2013) to use an FE research design to control for accounting firm effects in their
study of Taiwanese listed companies, and the market reaction to audit partner disclosures.
However, Aobdia et al. (2015) do not control for audit office effects which may be an important
omission given the robust evidence in the literature on the importance of interoffice variations in
audit quality. In addition, like China, audit reports in Taiwan are signed by two individuals
which makes it difficult to unambiguously measure partner effects.
13. The FAME database is administered by Bureau Van Dijk and has comprehensive information on
companies in the United Kingdom and Ireland.
14. The 11 non–Big 4 accounting firms that meet the sample screens are: Baker Tilly Audit, BDO,
Crowe Clark Whitehill, Grant Thornton, Hazlewoods, Mazars, Menzies, Nexia Smith and
Williamson, PKF Littlejohn, PKF UK (acquired by BDO on April 19, 2013), and Price Bailey.
The total number of partners is lower than the sum of partners working in Big 4 firms and those
employed in non–Big 4 firms because we have seven cases of partners switching to different
types of audit firms over the time period investigated.
15. Accruals models have long been criticized on econometric grounds (Dechow et al., 1995).
Despite these concerns, it is also the case that abnormal accruals are correlated with accruals,
and Dechow and Dichev (2002) show that the level of accruals is a good proxy for earnings qual-
ity based on their model of accrual estimation errors. Large abnormal accruals are also associated
with SEC enforcements (Dechow et al., 1996). In an audit context, large abnormal accruals have
been shown to be associated with negative PCAOB inspection reports, which suggests low qual-
ity audits (Aobdia, 2019; Gunny & Zhang, 2013). So, despite the econometric issues, there is
empirical evidence linking abnormal accruals to both low-quality earnings and low-quality
audits. However, given these concerns, we use measures other than accruals and triangulate the
results (DeFond & Zhang, 2014).
16. As indicated by Gul et al. (2013, p. 2001), while ‘‘it is theoretically appealing to estimate a
logistic model for a dichotomous dependent variable, we apply OLS because the ‘complete or
quasi-complete separation’ problem in the logistic FE model occurs in our data since there are
clients that never take a value of 1’’ for restatement and going concerns thus ‘‘it is impossible to
compute the maximum likelihood values of the FE coefficients for such auditors. Nevertheless,
for dichotomous dependent variables, OLS coefficient estimates remain unbiased, especially in
large samples (Wooldridge, 2010).’’ In addition, Greene (2004) suggests that linear models can
accommodate a large number of industry- and year-FEs with fewer estimation biases than non-
linear models. This is exactly our case. However, to be on the safe side, we estimate a reduced
form of model (5) with our dichotomous proxy for EQ and a reduced form of model (6) using a
logistic regression (i.e., in order to have a model that converges we estimate it separately for Big
4 clients and non–Big 4 clients) and the main results hold. In addition, since we also employ
hierarchical linear modeling (HLM) which is estimated as a maximum likelihood model (rather
than OLS), this also gives confidence that the models for restatements and going concern opi-
nions, that are estimated using linear probability models, are robust.
17. To rule out the risk that partner effects are a statistical artifact, we randomly reassigned auditor
partners to audit engagements (analogous to the random assignment of managers to companies in
Fee et al., 2013). We do this random assignment 50 times and re-estimate our empirical models
each time, expecting that this random assignment will result in insignificant F tests. Our expecta-
tions are confirmed as the F tests are unexpectedly significant (p \ .10) only five times for
ABWC and four times for GC, and never for RESTATE. These results are within the bounds of
sampling error and rule out that partner effects, in our test setting, are a statistical artifact.
18. Untabulated data indicate that the pooled sample mean of ABWC is 8.1% of lagged total assets,
28.9% of firm-year observations have subsequent restatements (RESTATE) and 8.1% of firm-
years have going concern audit reports (GC). All untabulated data/tests are available from the
authors upon request.
Cameran et al. 23

19. The shrinkage involves fitting a model with all the predictors, but where some of the estimated
coefficients move toward zero relative to classical estimates (James et al., 2013). As conse-
quence, the variance model reduces and by estimating some of the coefficients to be zero, vari-
able selection is also performed. The most common techniques to accomplish this task are ridge
regression and LASSO (least absolute shrinkage and selection operator; e.g., Park & Casella,
2008). We used the latter since it overcomes the disadvantage of ridge regression’s inability to
reduce the number of predictors in the final model. LASSO shrinks parameter estimates toward
zero and, in some cases, equate parameters to be exactly zero, thus allowing the exclusion of
some of the variables from the model. The estimation of a final model, which includes only a
subset of the predictors, also improves model interpretability (Pereira et al., 2016).
20. To further assess partners, we compare to the R2-within of a baseline model which includes client
characteristics and the partner demographic variables, and the incremental FEs of audit firms,
offices and partners. The results (not tabulated) indicate that the incremental effect of including
partner FEs is similar to Table 3. More precisely, the F test from the inclusion of partner effects is
always significant at p \ .01 and the R2-within, for the whole sample, goes from 8.86% of the
baseline model to 22.17% after the inclusion of partner FE (an increase of 13.31%) for abnormal
accruals; it goes from 14.74% in the baseline model to 26.57% after the inclusion of partner effects
(an increase of 11.83%) for restatements; it goes from 6.37% in the baseline model to 24.43%
after the inclusion of partner effects (an increase of 18.06%) for going concern reporting. We
obtain similar evidence if we split the sample between Big 4 and non–Big 4 clients. Thus, the
inclusion of the partner demographic variables in baseline model has no effect on the incremental
significance of partner FEs, and further shows that the set of partner demographic variables have
little explanatory power. We thank the referee for suggesting this analysis.
21. A recent paper by Kallunki et al. (2019) finds that Swedish IQ metrics are associated with audit
quality. IQ is, arguably, a compelling and direct measure of innate differences in partner ability
that might be associated with audit quality and, indeed, it is statistically significant. However,
the economic magnitude of the intelligence effect on audit quality is quite limited. Moreover, the
variation in IQ in the sample is very narrow with a small interquartile range.
22. The referee suggested examining a subsample of partners that change offices or firms to better
separate partner effects from audit firm and audit office effects. Of the 665 unique partners in
the sample, there are only 48 partners that meet this condition for a reduced sample of n = 474
firm-year observations. We re-estimate the models for this reduced sample. The primary results
are robust in both the OLS and HLM model estimations for all test variables, and in most cases
they are even stronger. The LASSO results are also robust for abnormal accruals, but not for the
restatements and going concern opinions, most likely due to low power because of the signifi-
cantly reduced instances of restatements/going concerns. Overall, these reduced sample results
reinforce the full sample findings that audit partner variation is more important than either firms
or offices. We thank the referee for suggesting the analysis.
23. Following Huettner and Sunder (2012), the R2 decomposition using the Shapley-Owen value pro-
vides an attractive diagnostic tool for identifying important (groups of) explanatory variables in a
given regression model and it is the best method that should be used for this purpose. Indeed, the
Shapley value and its axiomatic foundations can be generalized. In addition, the Owen value
constitutes such a generalization where an a priori grouping of the regressor variables is taken
into account, which accommodates many empirical analyses in practice. For more information
about the variance decomposition method, see Huettner and Sunder (2012).
24. In the Big 4 sample, only 13% of clients have the same accounting firm, office, and partner for
the period investigated; and for the non–Big 4 sample, 27% have the same accounting firm,
office, and partner for the period investigated. For the Big 4 (non–Big 4) unique clients in the
sample panel, there are 110 (71) changes in accounting firm, 248 (123) changes in audit office,
and 596 (282) changes in audit partner over the seven-year sample period.
24 Journal of Accounting, Auditing & Finance

References
Aobdia, D. (2019). Do practitioner assessments agree with academic proxies for audit quality?
Evidence from PCAOB and internal inspections. Journal of Accounting and Economics, 67(1),
144–174.
Aobdia, D., Lin, C., & Petacchi, R. (2015). Capital market consequences of audit partner quality. The
Accounting Review, 90(6), 2143–2176.
Balsam, S., Krishnan, J., & Yang, J. (2003). Auditor industry specialization and earnings quality.
Auditing: A Journal of Practice and Theory, 22(2), 71–97.
Barrick, M., & Mount, M. (1991). The Big Five personality dimensions and job performance: A
meta-analysis. Journal of Personnel Psychology, 44(1), 1–26.
Bedard, J., Deis, D., Curtis, M., & Jenkins, J. (2008). Risk monitoring and control in audit firms: A
research synthesis. Auditing: A Journal of Practice & Theory, 27(1), 187–218.
Bertrand, M., & Schoar, A. (2003). Managing with style: The effect of managers on firm policies.
The Quarterly Journal of Economics, 118(4), 1169–1208.
Bills, J. L., & Jensen, L. (2010). Auditor-client pairing: A positive assortative matching market.
Journal of Accounting Research, 13(1), 1–38.
Bol, J., Estep, C., Moers, F., & Peecher, M. (2018). The role of tacit knowledge in auditor experience
and human capital development. Journal of Accounting Research, 56(4), 1205–1252.
Burke, C., Stagl, K., Klein, C., Goodwin, G., Salas, E., & Halpin, S. (2006). What type of leadership
behaviors are functional in teams? A meta-analysis. The Leadership Quarterly, 17(3), 288–307.
Cahan, S. F., & Sun, J. (2015). The effect of audit experience on audit fees and audit quality. Journal
of Accounting, Auditing & Finance, 30(1), 78–100.
Campa, D., & Donnelly, R. (2016). Non-audit services provided to audit clients, independence of
mind and independence in appearance: Latest evidence from large UK listed companies.
Accounting and Business Research, 46(4), 422–449.
Carcello, J., & Li, C. (2013). Costs and benefits of requiring an engagement partner signature: Recent
experience in the United Kingdom. The Accounting Review, 88(5), 1511–1546.
Carey, P., & Simnett, R. (2006). Audit partner tenure and audit quality. The Accounting Review,
81(3), 653–676.
Chang, Y., Lin, Y., Liu, L., Shiue, M., & Wheatley, C. (2018). The use of hierarchical linear model-
ing to address lack of-independence in empirical auditing research. European Accounting Review,
27(1), 185–196.
Chen, F., Peng, S., Xue, S., Yang, Z., & Ye, F. (2016). Do audit clients successfully engage in opin-
ion shopping? Partner-level evidence. Journal of Accounting Research, 54(1), 79–112.
Chin, C., & Chi, H. (2009). Reducing restatements with increased industry expertise. Contemporary
Accounting Research, 26(3), 729–765.
Choi, J., Kim, C., Kim, J., & Zang, Y. (2010). Audit office size, audit quality, and audit pricing.
Auditing: A Journal of Practice and Theory, 29(1), 73–97.
Church, B. K., Dai, N. T., Kuang, X., & Liu, X. (2019). The role of auditor narcissism in auditor-
client negotiations: Evidence from China. Contemporary Accounting Research. Advance online
publication. https://doi.org/10.1111/1911-3846.12565
Cook, J. A., Kowaleski, Z. T., Minnis, M., Sutherland, A., & Zehms, K. M. (2020). Auditors are
known by the companies they keep. Journal of Accounting and Economics, 70(1), https://doi.org/
10.1016/j.jacceco.2020.101314
Craswell, A., Francis, J., & Taylor, S. (1995). Auditor brand name reputations and industry specializa-
tions. Journal of Accounting and Economics, 20(3), 297–322.
Dechow, P. M., & Dichev, I. (2002). The quality of accruals and earnings: The role of accrual estima-
tion errors. The Accounting Review, 7(Suppl.), 35–59.
Dechow, P. M., Ge, W., & Schrand, C. M. (2010). Understanding earnings quality: A review of the
proxies, their determinants and their consequences. Journal of Accounting and Economics, 50(2–
3), 344–401.
Cameran et al. 25

Dechow, P. M., Sloan, R., & Sweeney, A. (1995). Detecting earnings management. The Accounting
Review, 70(2), 193–225.
Dechow, P. M., Sloan, R., & Sweeney, A. (1996). Causes and consequences of earnings manipulation:
An analysis of firms subject to enforcement actions by the SEC. Contemporary Accounting
Research, 13(1), 1–36.
DeFond, M., & Zhang, J. (2014). A review of archival audit research. Journal of Accounting and
Economics, 59(2–3), 275–326.
Doidge, C., Karolyi, G., & Stulz, R. (2004). Why are foreign firms listed in the U.S. worth more?
Journal of Financial Economics, 71(2), 205–238.
Driskell, J. E., Salas, E., & Driskell, T. (2018). Foundations of teamwork and collaboration. American
Psychologist, 73(4), 334–348.
Fee, C. E., Hadlock, C., & Pierce, J. (2013). Managers with and without style: Evidence using exo-
genous variation. Review of Financial Studies, 26(2), 567–601.
Ferguson, A., Francis, J., & Stokes, D. (2003). The effects of firm-wide and office-level industry
expertise on audit pricing. The Accounting Review, 78(2), 429–448.
Francis, J. (2011). A framework for understanding and researching audit quality. Auditing: A Journal
of Practice and Theory, 20(2), 125–152.
Francis, J., Mehta, M., & Zhao, W. (2017). Audit office reputation shocks from gains and losses of
major industry clients. Contemporary Accounting Research, 34(4), 1922–1974.
Francis, J., & Michas, P. (2013). The contagion effect of low-quality audits. The Accounting Review,
88(2), 521–552.
Francis, J., Michas, P., & Yu, M. (2014). Office size of Big 4 auditors and client restatements.
Contemporary Accounting Research, 30(4), 1626–1661.
Francis, J., & Yu, M. (2009). Big Four office size and audit quality. The Accounting Review, 84(5),
1521–1552.
Goodwin, J., & Wu, D. (2014). What is the relationship between audit partner busyness and audit
quality? Contemporary Accounting Research, 33(1), 341–377.
Goodwin, J., & Wu, D. (2016). Is the effect of industry expertise on audit pricing an office-level or a
partner-level phenomenon? Review of Accounting Studies, 19(4), 1532–1578.
Grabner, I., Künneke, J., & Moers, F. (2020). Tacit knowledge, audit quality, and talent identification
[Working paper]. WU Vienna University.
Greene, W. (2004). The behavior of the fixed effects estimator in nonlinear models. The Econometric
Journal, 7(1), 98–119.
Gul, F., Wu, D., & Yang, Z. (2013). Do individual auditors affect audit quality? Evidence from archi-
val data. The Accounting Review, 88(6), 1993–2023.
Gunny, K., & Zhang, T. (2013). PCAOB inspection reports and audit quality. Journal of Accounting
and Public Policy, 32(2), 136–160.
Hardies, K., Breesch, D., & Branson, J. (2015). The female audit fee premium. Auditing: A Journal
of Practice and Theory, 34(4), 171–195.
Hardies, K., Janssen, J., Vanstraelen, A., & Zehms, K. (2020). Professional skepticism profiles, effects
on audit processes and outcomes, and the moderating role of audit firm culture [Working paper].
Antwerp University.
Huettner, F., & Sunder, M. (2012). Axiomatic arguments for decomposing goodness of fit according
to Shapley and Owen values. Electronic Journal of Statistics, 6, 1239–1250.
James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning with
applications in R. Springer.
Johnson, V., Khurana, I., & Reynolds, J. K. (2002). Audit-firm tenure and the quality of financial
reports. Contemporary Accounting Research, 19(4), 637–660.
Kallunki, J., Mikkonen, J., Niemi, L., & Nilsson, H. (2019). IQ and audit quality: Do smarter auditors
deliver better audits? Contemporary Accounting Research, 36(3), 1373–1416.
Ke, B., Lennox, C., & Xin, Q. (2014). The effect of China’s weak institutional environment on the
quality of Big Four audits. The Accounting Review, 90(4), 1591–1619.
26 Journal of Accounting, Auditing & Finance

Knechel, R., Vanstraelen, A., & Zerni, M. (2015). Does the identity of engagement partners matter?
An analysis of audit partner reporting decisions. Contemporary Accounting Research, 32(4),
1443–1478.
Kothari, S. P., Leone, A., & Wasley, C. (2005). Performance matched discretionary accrual measures.
Journal of Accounting and Economics, 39(1), 163–197.
Kozlowski, S. (2018). Enhancing the effectiveness of work groups and teams: A reflection.
Perspectives on Psychological Science, 13(2), 205–212.
Krishnan, G. (2003). Does Big 6 auditor industry expertise constrain earnings management?
Accounting Horizons, 17(Suppl.), 1–16.
Lennox, C. S., & Pittman, J. A. (2011). Voluntary audits versus mandatory audits. The Accounting
Review, 86(5), 1655–1678.
Lennox, C. S., & Wu, X. (2018). A review of the archival literature on audit partners. Accounting
Horizons, 32(2), 1–35.
Li, K., McNichols, M. F., & Raghunandan, A. (2018). A two-sided matching model of the audit
market for IPO firms [Working paper]. https://ssrn.com/abstract=3117828
Lo, A. K., Lin, K., & Wong, R. (2019). Does availability of audit partners affect audit quality?
Evidence from China. Journal of Accounting, Auditing and Finance.
Morgeson, F., DeRue, D. S., & Karam, E. (2010). Leadership in teams: A functional approach to
understanding leadership structure and processes. Journal of Management, 36(1), 5–39.
Nakagawa, S., & Schielzeth, H. (2013). A general and simple method for obtaining R2 from general-
ized linear mixed-effects models. Methods in Ecology and Evolution, 4(2), 133–142.
Owen, G. (1977). Values of games with priori unions. In R. Heim & O. Moeschlin (Eds.), Essays in
mathematical economics and game theory (pp. 76–88). Springer-Verlag.
Park, T., & Casella, G. (2008). The Bayesian Lasso. Journal of the American Statistical Association,
103(482), 681–686.
Pereira, J. M., Basto, M., & da Silva, A. F. (2016). The logistic Lasso and Ridge regression in pre-
dicting corporate failure. Procedia Economics and Finance, 39, 634–641.
Porumb, V., de Jong, A., Huijgen, C., Marra, T., & van Dalen, J. (2017). Individual auditor style and
audit quality in a high reputation risk setting [Working paper]. https://papers.ssrn.com/sol3/
papers.cfm?abstract_id=3048713
Sarath, B. (2016). Audit quality within adverse selection markets. Asian Review of Accounting, 24(1),
2–18.
Sastre, M., & Trannoy, A. (2002). Shapley inequality decomposition by factor components. Journal
of Economics, 9(Suppl.), 51–89.
Schmitt, N. (2014). Personality and cognitive ability as predictors of effective performance at work.
Annual Review of Organizational Psychology and Organizational Behavior, 2(1), 45–65.
Shapley, L. (1953). A value for n-person games. In H. W. Kuhn & A. W. Tucker (Eds.), Annals of
mathematical studies. Vol. 1: Contributions to the theory of games (pp. 307–317). Princeton
University Press.
Sundgren, S., & Svanstrom, T. (2014). Auditor-in-charge characteristics and going-concern reporting.
Contemporary Accounting Research, 31(2), 531–550.
Torres-Reyna, O. (2007). Panel data analysis fixed and random effects using Stata. Princeton
University. https://www.princeton.edu/~otorres/Panel101.pdf
Wang, Y., Yu, L., & Zhao, Y. (2015). The association between audit-partner quality and engagement
quality: Evidence from financial report misstatements. Auditing: A Journal of Practice and
Theory, 34(3), 81–111.
Wooldridge, J. (2010). Econometric analysis of cross section and panel data (4th ed.). MIT Press.
Zmijewski, M. (1984). Methodological issues related to the estimation of financial distress prediction
models. Journal of Accounting Research, 22(Suppl.), 59–82.