iq and audit quality: do smarter auditors deliver better

IQ and Audit Quality: Do Smarter Auditors Deliver BetterAudits?*

JENNI KALLUNKI, University of Oulu

JUHA-PEKKA KALLUNKI, University of Oulu, Aalto University, Stockholm School ofEconomics

LASSE NIEMI, Aalto University†

HENRIK NILSSON, Stockholm School of Economics

ABSTRACTThis study examines the role of an individual auditor’s cognitive ability in delivering high-qualityaudits. Our results from analyzing archival data from Sweden show that audit partners’ IQ scoresobtained from psychological tests are positively associated with going-concern audit reportingaccuracy and audit fee premiums. We also find some, albeit weak, evidence that audit partners’ IQscores are negatively associated with the income-increasing abnormal accruals of the client. Theseresults suggest that, although audit services are standardized through various control mechanismsand audits are conducted by teams rather than by individual auditors, the cognitive ability of auditpartners responsible for an audit remains important in delivering high-quality audit services.

QI et qualité de l’audit : les auditeurs plus intelligentsproduisent-ils de meilleurs audits?

RÉSUMÉLes auteurs étudient le rôle des capacités cognitives d’un auditeur donné dans la productiond’audits de qualité supérieure. Les résultats de leur analyse de données d’archives suédoisesindiquent que les notes obtenues par les associés d’audit aux tests psychologiques de quotient

*Accepted by Clive Lennox. An earlier version of this paper was presented at the 2017 Contemporary AccountingResearch Conference, generously supported by the Chartered Professional Accountants of Canada. We are grateful totwo anonymous reviewers, Clive Lennox, and Michael Welker for insightful comments that have greatly improved thispaper. We also thank Eli Amir, Daniel Aobdia (the discussant), Jere Francis, Mattias Hamberg, Kathleen Harris, TomasHjelström, Steven Kachelmeier, Patrick Kielty, S. M. Khalid Nainar, Per Olsson, Thomas Omer, Luke Phelps, PetriSahlström, Hanna Setterberg, Heidi Vander Bauwhede, participants at the 2017 Contemporary Accounting ResearchConference, the 2017 University of Florida International Conference on Assurance and Governance, and the 6th EIASMWorkshop on Audit Quality, and seminar participants at Aalto University, the University of Bristol, the University ofExeter, WHU—Otto Beisheim School of Management, CPA Canada—Accounting and Governance Research Centre, theUniversity of Ottawa, and the University of Maastricht for numerous comments and suggestions. We acknowledge Euro-clear Sweden, Finansinspektionen, the Swedish Tax Authorities, and the Swedish Armed Forces for providing the requisitedata, as well as Handelsbanken Research Foundation, the Swedish Research Council (VR), Finnish Savings Banks GroupResearch Foundation, Marcus Wallenberg Research Foundation, OP Group Research Foundation, Tauno Tönning Founda-tion, Suomen Arvopaperimarkkinoiden Edistämissäätiö, Finnish Cultural Foundation, and Academy of Finland (Project#277055) for financial support. This study has been evaluated and approved by the Regional Ethical Review Board inUmeå, Sweden (DNR 08:074 Ö, DNR I3-0449/2009 and 2010-200-32 Ö). All remaining errors are our own. Data availabil-ity: Data sources are described in the article. Data requests should be directed to the administrator of each database.

†Corresponding author.

Contemporary Accounting Research Vol. 36 No. 3 (Fall 2019) pp. 1373–1416 © CAAA

doi:10.1111/1911-3846.12485

intellectuel affichent un lien positif avec l’exactitude des rapports d’audit mettant en doute lacontinuité de l’exploitation et avec la majoration des honoraires d’audit. Les auteurs relèvent égale-ment certaines données qui démontrent, quoique faiblement, que les QI des associés d’audit ont unlien négatif avec les régularisations discrétionnaires anormales du client entraînant une hausse durésultat. Ces constatations semblent indiquer que, même si les services d’audit sont normalisésgrâce à divers mécanismes de contrôle et que les audits sont réalisés par des équipes plutôt que pardes auditeurs à titre individuel, les capacités cognitives des associés d’audit responsables d’unaudit demeurent importantes dans la prestation de services d’audit de qualité supérieure.

1. Introduction

Recently, much archival research has been devoted to individual auditors because of limitationswith using the audit firm as the unit of analysis (Lennox and Wu 2018). This research shows thataudit quality varies within an audit firm, and that much of this variation is attributable to the char-acteristics of the individual auditors in charge of audit engagements (Amir et al. 2014; Cameranet al. 2017; Carey and Simnett 2006; Chen et al. 2010; Chi and Chin 2011; Gul et al. 2013;Knechel et al. 2015). However, it is not fully understood why audit quality varies between indi-vidual auditors.1 Due to the lack of person-level data on auditors, there is no archival evidence onthe role of personal capabilities in the delivery of high-quality audits. Moreover, in many coun-tries the identities of the audit partners in charge of an audit engagement are unknown, becauseaudit reports are signed by audit firms rather than the individual auditors in charge of the auditengagements.

Experimental studies suggest that cognitive ability plays an important role in judgment anddecision making during the audit process (Bonner and Lewis 1990; Gibbins 1984; Nelson andTan 2005; Libby and Luft 1993), but it remains unclear whether audit quality varies withauditors’ IQ in practice. This is because audit firms mitigate the risk of individual auditors deliv-ering low-quality audits by establishing various control mechanisms (Bedard et al. 2008), by pro-moting in-house knowledge-sharing (Dowling 2009), and by organizing audit work as a jointeffort by the audit team comprising several individual auditors (Rich et al. 1997). Thus, an impor-tant, yet largely unsolved, question is whether an auditor’s cognitive ability is associated withaudit quality. Our study contributes to the literature by providing empirical evidence on the asso-ciation between an audit partner’s cognitive ability (IQ) and audit quality as measured by going-concern audit reporting accuracy, audit fees, and abnormal accruals of the client.2

We conduct our study using data from Sweden for two reasons. First, the scores from an IQtest similar to the Armed Forces Qualifications Test (AFQT) used in the United States are avail-able for virtually all Swedish male citizens because military service was compulsory for all malesin Sweden until 2010.3 Swedish male citizens were obliged by law to attend the enlistment test,including comprehensive psychological tests, to assess their cognitive ability. There is muchresearch evidence that validates IQ as a measure of cognitive ability. Specifically, many studiesshow that an individual’s IQ is a powerful predictor of their success in terms of salary and otherincome or job complexity (Beauchamp et al. 2017; Borghans et al. 2008; Hunter and Hunter1984; Hülsheger et al. 2007; Lo 2017; Murray 1998; Salgado et al. 2003). IQ has also beenreported to be correlated with better performance in various decision making situations and with

1. An auditor’s industry expertise and demographic factors such as gender or age have been linked to audit quality(Chin and Chi 2009; Chi and Chin 2011; Gul et al. 2013; Hardies et al. 2016; Ittonen et al. 2015; Zerni 2012).However, these studies do not address the role of an auditor’s cognitive ability in delivering high-quality audits.

2. The American Psychological Association defines cognitive ability (IQ) as “the ability to understand complex ideas,to adapt effectively to the environment, to learn from experience, to engage in various forms of reasoning, to over-come obstacles by taking thought” (Neisser et al. 1996, 77).

3. These tests are based on the methodologies developed in the psychological literature and are conducted by professionalpsychologists.

1374 Contemporary Accounting Research

CAR Vol. 36 No. 3 (Fall 2019)

the probability of having a top position in the corporate world (Adams et al. 2015; Grinblatt et al.2011, 2012, 2015; Keloharju et al. 2016).

The IQ test results from the Swedish military service are extremely reliable due to lack ofincentives to avoid military service by “flunking” the test: virtually all males with sufficient phys-ical and psychological conditions were enlisted (Lindqvist and Vestman 2011). In addition, theprofessional psychologists conducting the IQ test also double-checked each individual’s back-ground information, including school grades, to ensure that the test result was plausible. If the testresult seemed abnormally low, the individual had to retake the test.

Second, during our data period from 2000 to 2009, all public and private Swedish compa-nies, regardless of size, were required to file audited financial statements and the audit partnerresponsible for the audit engagement was required to sign the audit opinion. Combined, thesecharacteristics of the Swedish setting allow us to trace the IQ score for 407 individual male auditpartners and their 31,969 private and 277 public clients during our data period. The average auditpartner in our sample has, on a scale from 1 to 9, an IQ score of 6.82, which is higher than theaverage IQ of the rest of the population, which is 5.0. There is also substantial variation in IQamong the audit partners, suggesting that an audit partner’s IQ could indeed play a role in thequality of their audits.

Our empirical results show that audit quality increases with an audit partner’s IQ score. Spe-cifically, we find that the likelihood of issuing an incorrect going-concern audit report signifi-cantly decreases as the audit partner’s IQ increases. This result holds for Total reporting error (thesum of Type 1 and Type 2 reporting errors) and for Type 1 and Type 2 reporting errors sepa-rately. We also find a significant increase in audit fees in accordance with an audit partner’s IQ,suggesting that clients are willing to pay a fee premium for the services of more capable auditpartners. Finally, we find some, albeit weak, evidence that the client’s income-increasing earningsmanagement decreases as the audit partner’s IQ increases. We measure earnings management byabnormal accruals estimated using the modified Jones (1991) model (Caramanis and Lennox2008). Overall, our results suggest that although an audit is a product of the joint effort of thewhole audit team and is standardized and monitored by various mechanisms of the audit firm andthe audit profession, the judgments and decisions made by the individual auditor in charge of theaudit still play an important role in determining audit quality.

This study is subject to several limitations. Audit quality is a complex concept involvingmany dimensions which are not directly observable for outsiders. Audit quality must therefore bemeasured using proxies that contain measurement error. In an attempt to mitigate such errors, weuse several audit quality proxies frequently used in previous studies—that is, going-concernreporting accuracy, audit fees, and abnormal accruals of the client (Defond and Zhang 2014; Sim-nett et al. 2016). Each of these proxies has unique strengths and weaknesses that are discussed inmore detail in section 3.4

Endogeneity is always a concern in archival studies like this one. In particular, an auditpartner’s IQ could be a proxy for omitted variables correlated with both auditor ability and auditquality, or the direction of causation could run from clients with higher audit quality to the use ofmore capable auditors. For example, clients with high-quality financial reporting or a more com-plex business model, or that are greater in size, may choose smarter auditors. In addition, dataavailability imposes some limitations on the analyses. In particular, because audit fees are notreadily available for privately held companies in Sweden, the audit fee analyses in this paper arebased on publicly listed companies only.

We believe that our results will be of interest to audit firms that select and promote theirauditors. We also believe that the regulatory bodies that supervise the audit profession will be

4. A review article by Defond and Zhang (2014) provides an insightful discussion on the pros and cons of the variousproxies for audit quality.

IQ and Audit Quality 1375


interested in our results. For instance, the PCAOB recently mandated the disclosure of theengagement partner’s name for audits of U.S. listed companies beginning in 2017.

The rest of the paper is organized as follows. In section 2, we discuss whether individualauditor cognitive ability may matter in audit quality. In section 3, we describe the institutionalsetting and research methodology. Empirical results are provided in section 4, and concludingremarks are provided in section 5.

2. Relevant literature

Why an auditor’s cognitive ability may matter for audit quality

An auditor’s cognitive ability may matter for audit quality because auditing is a complex processthat requires many subjective judgments and decisions on the part of the auditor at all stages ofthe audit engagement, from the planning of the audit to the formation of the audit opinion(Hogarth 1991; Knechel 2000). Smart auditors also have incentives to exercise their cognitiveskills to deliver high-quality audits because they are rewarded for high-quality audits and penal-ized for audit failures (Knechel et al. 2013).

Subjective judgments and decision making in auditing

Auditing is a complex process that requires subjective judgments and decisions at all stages ofthe audit engagement, including the planning of the audit, the collection and evaluation of auditevidence, and, finally, the formation of the audit opinion after the audit has been completed(Gibbins 1984; Hogarth 1991; Knechel 2000). Subjective judgments and decisions regarding thenature, extent, and timing of the audit procedures are many, and they determine the success of theaudit work. For instance, auditors assess whether complex accounting transactions are in accor-dance with GAAP. They also assess the economic estimates provided by the client for measuringasset and liability values. These estimates likely involve a great deal of discretion arising fromthe uncertainty of the outcome of future events or from using data that cannot be accumulated ona timely, cost-effective basis (Griffith et al. 2013).

Despite the difficulties in conducting well-controlled experiments in intelligence research(Lo 2017), there are experimental studies that link an auditor’s cognitive ability to audit outcomes(Abdolmohammadi and Shanteau 1992; Bonner and Lewis 1990; Libby and Tan 1994; McKnightand Wright 2011). Bonner and Lewis (1990) model auditor expertise as a function of knowledgeand ability in various auditing tasks. They argue that auditors possess both knowledge and a gen-eral problem-solving ability, which includes the ability to recognize relationships, interpret data,and reason analytically. They find that, although more experienced auditors outperform less expe-rienced auditors on average, knowledge and innate ability provide a better explanation of varia-tion in auditor performance.

Libby and Tan (1994) also report that problem-solving ability affects an auditor’s decisionperformance in unstructured tasks and that knowledge is related to ability through learning. Tanand Libby (1997) show that staff and senior auditors with superior performance evaluations havehigher cognitive ability. They also find that ability does not matter for staff performance evalua-tions if these evaluations mostly reflect work on simple tasks. Abdolmohammadi and Shanteau(1992) report that intelligence is among the most important attributes of an audit partner, therebyhighlighting the importance of auditors’ cognitive skills. Finally, McKnight and Wright (2011)report that both technical knowledge and ability are higher among high-performing auditors thanamong low-performing auditors.

Incentives to deliver high-quality audits

Audit quality is a continuous construct in both theory and practice, ranging from very low to veryhigh audit quality (Defond and Zhang 2014; Francis 2004). Audit failures such as audit reportingerrors obviously occur on the lower end of the quality continuum. Auditors have incentives to



avoid audit failures that expose them and their audit firms to litigation, reputation, and regulationrisks (Defond and Zhang 2014). Litigation risk exposes auditors to financial penalties, whereasreputation risk impairs the ability to attract and retain clients. Regulation risk arises from thethreat of regulatory intervention, which subjects auditors to sanctions that include fines, loss oflicensure, and criminal penalties. These risks are not independent, as litigation and regulatorysanctions are likely to damage the auditor’s reputation as well.

Auditors also have incentives to supply high audit quality because greater auditor compe-tency in delivering high-quality audits is likely to increase the auditor’s reputation capital(Defond and Zhang 2014). Archival studies show that audit quality varies across individual auditpartners and that both audit clients and audit firms reward their auditors for high-quality audits(Amir et al. 2014; Aobdia et al. 2015; Cameran et al. 2017; Carey and Simnett 2006; Chen et al.2010; Chi and Chin 2011; Gul et al. 2013; Knechel et al. 2015). Archival studies also report thatsome of the auditor-level variation in audit quality is attributable to audit partners’ personal char-acteristics (Chu et al. 2017; Gul et al. 2017; Hardies et al. 2016; Li et al. 2017; Sundgren andSvanström 2014), and that audit fees vary across audit partners (Taylor 2011).

Taylor (2011) finds that there are “premium” partners who have fewer clients and shorter ten-ure than “discount” partners within the same audit firm. These “premium” partners can establishtheir reputation and move to more prestigious clients while the “discount” partners cannot estab-lish such a reputation and serve only less prestigious clients. Zerni (2012) finds that the auditor-specific premium is associated with their industry specialization. Goodwin and Wu (2014) findthat the audit fee premium related to an auditor’s industry expertise is mainly a partner-level phe-nomenon. Finally, audit firms seem to reward their partners differently as there is significant vari-ation in partner remuneration even within the same audit firm (Burrows and Black 1998; Knechelet al. 2013).

Why an auditor’s cognitive ability may not matter for audit quality

There are also arguments for why an auditor’s cognitive ability may not matter for audit quality.First, audit firms reduce the potential litigation and reputational losses from low-quality audits bycontrolling the behavior and decision making of individual auditors and by enhancing the skillsand adoption of best practices among auditors (Bedard et al. 2008; Dowling 2009). Second,audits are conducted by teams comprising several individuals, which is likely to reduce the roleof an individual auditor in the production of audits (Rich et al. 1997). Finally, the professionalqualifications of becoming a certified auditor (CPA) serve as entry requirements into the auditprofession, which could prevent an individual with a low IQ from entering the profession.

Risk management, control mechanisms, and knowledge-sharing in audit firms

Audit firms are typically organized in the form of a partnership in which each audit partner is atthe same time both an agent and a principal, sharing the risks (Huddart and Liang 2003). To pro-tect the partnership against excessive risk-taking, audit firms create various monitoring mecha-nisms to control the behavior and decision making of individual audit partners. These controlsrelate to all phases of audit engagement, from client acceptance to audit procedures, and to qualitycontrol reviews of the completed work (Bedard et al. 2008).

Audit firms, moreover, develop their own unique audit approaches and styles to standardizetheir audits (Francis et al. 2014; Lemon et al. 2000). Audit approaches and styles are further rein-forced by in-house knowledge-sharing systems, which make the research, experiences, processes,and working papers of audit teams available for everyone in the audit firm (Chow et al. 2008;Murthy and Kerr 2004; Vera-Munõz et al. 2006). These systems enhance the skills of the mem-bers of audit teams and the adoption of best practices (Dowling 2009). Knowledge-sharing sys-tems also restrict the audit partners’ discretion in their judgment and decision making as eachaudit firm has its own in-house rules for how to interpret auditing and financial reporting stan-dards. Finally, standard setters, such as the International Auditing and Assurance Standards



Board, also require audit firms to establish procedures for systematic reviews to control the qualityof their audit work.

An audit is a joint effort by an audit team

Since audits are conducted by teams, not individuals, the outcome of an audit depends on thework of the audit team. Although audit partners are involved in all important audit decisions andtheir views affect the judgments of the rest of the audit team (Peytcheva and Gillett 2011; Wilks2002), the work of others still affects the decisions made by audit partners. Specifically, juniormembers of an audit team conduct most of the detailed audit fieldwork, and senior members areresponsible for coordinating this work and for reviewing the notes prepared by the juniors. Auditengagement partners plan and manage the audit engagement, review senior members’ work, andare responsible for audit quality and the client relationship. Finally, audit review partners reviewthe overall audit work and assess the audit engagement risks. Their views are likely to affectengagement partners’ decisions.

Organizing audit work through teams, where individuals at different hierarchical levels ofaudit firms review the work of their subordinates, is an important quality control mechanism(Nelson and Tan 2005; Ramsay 1994). From the risk management point of view, engagementquality-review partners are a critical organizational layer in audit firms because their primaryobjective is to evaluate the overall quality of the audit engagement and the performance of theengagement partners and their team (Epps and Messier 2007). To maintain their independencefrom those who conduct the audit, engagement quality reviewers are not allowed to belong to theaudit team. The interaction between the engagement quality reviewers and the engagement part-ner is often centered on resolving difficult and complex accounting issues for the client (Embyand Favere-Marchesi 2010).

Professional and personal qualifications for becoming an audit partner

There are specific entry requirements for the audit profession that could prevent low-IQ individ-uals from entering this profession. In particular, an academic degree and passing a demandingexam are required to become a CPA. Current partners in an audit firm are most likely to inviteonly the most promising CPAs to join the partnership. Thus, audit partners pass several screen-ings during their careers, and only the most capable auditors are likely to pass these screenings.Consequently, most audit partners are likely to have a high IQ and, therefore, provide high auditquality.

3. Institutional setting and research methodology

Institutional setting in Sweden

Audit regulation

As a member of the European Union (EU), Sweden follows the EU Directives and the Interna-tional Standards on Auditing (ISAs). The EU regulation on auditing is implemented through theAccounting Act, the Auditing Act, and the Company Act, supplemented by the auditing standardsof the Professional Institute for Certified Auditors and other Accounting Professionals (FAR).FAR is a member of the International Federation of Accountants (IFAC) and has adopted theISAs and the IFAC’s Code of Ethics.

In Sweden, both audit firms and their partners in charge of the audits of listed firms are sub-ject to quality control inspections every three years. If their clientele comprises only private firms,this quality control is conducted every six years. There was no mandatory audit partner or firmrotation in Sweden during the sample period. Sweden has a two-tier system of auditor qualifica-tions: approved and authorized auditors. The requirements to become an authorized auditor aremore demanding. To become an authorized auditor, one must have a master’s degree in account-ing and practice of at least five years and must pass a demanding professional competence



examination. To become an approved auditor, one must have a bachelor’s degree and practice forat least three years. Approved auditors who have passed a professional competence examinationthat is less demanding than that of authorized auditors are also allowed to audit all companiesirrespective of their size. The audit certification is valid for five years, and the auditor mustreapply to the supervisory board for a license renewal.

Disclosures on audits

Swedish legislation requires the auditor-in-charge to sign the audit report, and, if the report is signedby more than one auditor, the auditor-in-charge must be specified. The audit partner in charge ofthe audit engagement is legally responsible for the audit quality. Marketing of individual auditorsplays an important role for success in the auditing procurement process (Zerni 2012). For instance,competitive bids disclosed to potential clients typically contain detailed descriptions of each auditteam member and their planned tasks. Competitive bids also often include the curriculum vitae ofthe engagement partner and other key members of the audit team, suggesting that audits are notperceived as uniform among audit partners within audit firms (Fiolleau et al. 2010). Potential cli-ents often request this information, indicating that providing an auditor’s detailed curriculum vitaeis an important signal that the engagement team has the relevant competence to carry out theaudit work.

Data sources and sample construction

We use multiple sources to construct the data set for our empirical analyses. We begin by obtain-ing the information on the identity of individual audit partners from Finansinspektionen (theSwedish Financial Supervisory Authority). The data include all audit partners who have acted asa lead or deputy auditor for at least one listed Swedish company during the sample period from2000 to 2009. We obtain the IQ scores for these audit partners from the cognitive ability test datamaintained by the Swedish Armed Forces.5 The resulting sample contains the IQ scores for407 male auditors, for whom we then retrieved financial statement data for their listed clientsfrom COMPUSTAT Global Vantage, and for their privately held clients from the database pro-vided by Sweden’s leading business and credit agency, UC AB.6 The data include financial state-ments, audit reports, and information on bankruptcies. We exclude finance and insurance industriesdue to their unique financial reports and regulations.

We conduct three sets of empirical analyses to explore the association between auditors’ IQand audit quality. The number of observations varies across these three analyses because of thelimitations arising from the calculations of different audit quality proxies. The going-concernreporting accuracy analyses are based on a sample of 31,969 privately held clients (120,942client-year observations) and 257 publicly listed clients (1,070 client-year observations) auditedby 407 unique audit partners in 54 unique audit firms (5 Big N firms and 49 non-Big N firms).7

There are 823 going-concern reports issued by 243 unique audit partners, 665 Type 1 reportingerrors made by 222 unique audit partners, and 581 Type 2 reporting errors made by 213 uniqueaudit partners. More details are given in Table 3.

The audit fee analyses are based on a sample of 277 publicly listed clients (1,197 client-yearobservations) audited by 286 unique audit partners in 30 unique audit firms (5 Big N firms and

5. IQ scores are not available for women and non-Swedes because they are not obligated to serve in the SwedishArmed Forces.

6. Our sample selection criteria work against finding a relation between auditors’ IQ and audit quality because all theauditors in our sample are authorized auditors who have experience of auditing listed companies, and the majorityof them work for Big N audit firms.

7. In untabulated analyses, we have restricted the sample to include only financially distressed clients. Financial distressis defined as having above median value of the variable PROBZjt or, alternatively, having a negative net income (thedummy variable LOSSjt equals one). These results are essentially similar to those reported in the paper and are avail-able from the authors upon request.



25 non-Big N firms). We include only publicly listed companies in the audit fee analyses becausethe data on audit fees are not readily available in any database for Swedish privately held compa-nies. Finally, the abnormal accruals analyses are based on a sample of 26,673 privately held clients(104,243 client-year observations) and 226 publicly listed clients (926 client-year observations)audited by 406 unique audit partners in 53 unique audit firms (4 Big N firms and 49 non-BigN firms).

Measurement of variables

Audit quality proxies

Going-concern reporting error. We use going-concern reporting errors as an inverse measureof audit quality because issuing an incorrect going-concern audit report (issuing a going-concernaudit report to a client that does not subsequently go bankrupt or, alternatively, not issuing agoing-concern audit report to a client that subsequently goes bankrupt) can be regarded as anindication of low audit quality (DeFond and Zhang 2014; Lennox 1999; Knechel and Vanstraelen2007; Knechel et al. 2015).8 The advantage of using going-concern audit reports in measuringaudit quality is that they are a very direct measure of audit quality because the audit report is theauditor’s responsibility and directly under their influence and control (Defond and Zhang 2014).They, moreover, involve relatively low measurement error. The disadvantage is that the egregiousnature of going-concern audit reports means that they do not capture more subtle compromises inaudit quality and are relatively rare events (Defond and Zhang 2014).

We construct the following three dummy variables to measure going-concern audit reportingerrors. The dummy variable for Type 1 reporting error, TYPE_1_ERRORjt, equals one if an auditpartner issues a going-concern audit report to a client that subsequently does not file for bank-ruptcy within 12 months of the audit partner’s report, and zero otherwise. The dummy variablefor Type 2 reporting error, TYPE_2_ERRORjt, equals one if an audit partner does not issue agoing-concern audit report to a client that subsequently files for bankruptcy within 12 months ofthe audit partner’s report, and zero otherwise. The dummy variable for Total reporting error (thesum of Type 1 and Type 2 errors), TOTAL_ERRORjt, equals one if either of the dummy variablesTYPE_1_ERRORjt or TYPE_2_ERRORjt equals one, and zero otherwise. In cases with successivegoing-concern reports, we also include the subsequent going-concern report(s) to allow for learn-ing effects. Specifically, if the first-time going-concern report turned out to be erroneous, an audi-tor may be more accurate the next year. In untabulated analyses, we have replicated both theunivariate and multivariate analyses reported in Tables 3 and 4 by using only first-time going-concern reports, and obtain essentially similar results.

Audit fees. We use audit fees to proxy for audit quality because they are expected to measurethe auditor’s effort level, which is an input to the audit process that is intuitively related to auditquality (Defond and Zhang 2014). The advantage of using audit fees as a measure of audit qualityis that they are continuous and, thus, able to capture more subtle variations in audit quality. Onelimitation is that, in addition to capturing audit effort, audit fees also capture risk premium andimproved audit efficiency, meaning that an increase in fees cannot unambiguously be interpretedas an increase in audit quality. Another limitation is that audit fees capture the joint outcome ofboth supply and demand factors (Defond and Zhang 2014). To measure audit fees, we construct avariable ln(AUDITFEESjt), which is the natural logarithm of the audit fees paid by a client.

8. The term correct/incorrect audit going-concern report is a simplification as an auditor’s task is to assess future finan-cial distress or bankruptcy using present, not future, information. Thus, clean/going-concern audit reports might becorrect in the light of present information even though they turn out to be erroneous when it comes to predictingbankruptcy. We thank a reviewer for pointing this out.



Abnormal accruals. Our third proxy for audit quality is the client’s financial reporting quality asmeasured by abnormal accruals (Caramanis and Lennox 2008; Dechow et al. 2010; DeFond andZhang 2014; Francis et al. 2014). Financial reporting quality is an intuitively appealing proxy foraudit quality because financial statements are a joint product of both the client and the auditor(DeFond and Zhang 2014). While abnormal accruals do not directly identify GAAP violations,they are expected to detect within-GAAP earnings management, which likely impairs financialreporting quality by misleading investors. However, abnormal accruals are a less direct proxy foraudit quality than, for example, going-concern audit reports because the auditor’s influence on theclient’s financial reporting quality is likely to be more limited. Abnormal accruals also tend tohave high levels of measurement error and even bias (DeFond and Zhang 2014).

To measure abnormal accruals, we construct a variable |DAjt|, which is the absolute value ofthe residual from the following modified cross-sectional Jones (1991) model with an intercept, assuggested by Kothari et al. (2005):

TACjt=TAjt−1 = α0 + α1 1=TAjt−1� �

+ α2 ΔREVjt −ΔRECjt

� �=TAjt−1

� �+ α3 PPEjt=TAjt−1

� �+ εjt , ð1Þ

where j denotes the client and t denotes the year. The variables in model (1) are defined as fol-lows: TACjt is the total accruals for client j in year t (measured as the change in accounts receiv-able plus the change in inventory plus the change in accrued assets minus the change in accountspayable minus the change in accrued liabilities minus depreciation and amortization expense),ΔREVjt is the change in revenues for client j in year t, ΔRECjt is the change in net accountsreceivable for client j on year t, PPEjt is property, plant, and equipment for client j in year t, andTAjt−1 is the total assets for client j in year t−1. We estimate model (1) for each 2-digit SICindustry-year with at least 30 client-year observations.

IQ score

An audit partner’s cognitive ability measure (IQ score) is obtained from psychometric testsadministrated by the Swedish Armed Forces. Military service in Sweden was compulsory until2010. According to the Swedish Act on Liability for Total Defense Service, all males with Swedishcitizenship had to attend the enlistment test at around the age of 18. The enlistment procedurespanned two days involving tests of medical status, physical fitness, cognitive ability, and an inter-view with a psychologist. The enlistment procedure aimed at estimating an individual’s ability tofulfill military services and choosing suitable individuals for different services. The test of cognitiveability consisted of four different subtests (logical, verbal, spatial, and technical), which were gradedon a scale from one to nine. The results of these tests were then transformed to a discrete variableof general cognitive ability that also ranges from one (low ability) to nine (high ability) with anaverage of five, which corresponds to an IQ score of about 100. Carlstedt (2000) and Beauchampet al. (2017) discuss the history of psychometric testing in the Swedish Armed Forces and provideevidence that the measure of cognitive ability is a good measure of general intelligence (Spearman1904).9 Deary (2001) also reports that IQ scores from cognitive ability tests have validity that isalmost unequaled in psychology.

To measure an audit partner’s IQ, we construct a variable IQi, which is equal to an auditpartner’s IQ score ranging from one (the lowest IQ) to nine (the highest IQ). In untabulated ana-lyses, we have reestimated all our regression models by excluding audit partners with very lowIQ scores (IQ scores lower than four) and obtain results that are similar to those reported in this

9. The measure is similar to the AFQT score used in the United States, which measures the aptitude and trainability ofindividuals eligible for enlistment. The four areas used to compute the AFQT score are word knowledge, paragraphcomprehension, arithmetic reasoning, and mathematics knowledge. The AFQT score is frequently employed as ageneral indicator of cognitive skills and learning aptitude (Levine and Rubinstein 2017).



paper, with one exception. Namely, the estimated coefficient for the variable IQi is insignificantin the income-decreasing abnormal accruals regression reported in column (3) of Table 8.

Control variables

Auditor-specific variables. We include in our empirical analyses various auditor-specific variablesto control for auditor characteristics that are likely to affect audit quality as measured by going-concern reporting accuracy, audit fees, and abnormal accruals. First, we control for an auditor’sreputational capital with the variable PROPLISTCLit, which is equal to the proportion of publiclylisted clients of an audit partner (Moizer 1997). We expect that partners with more listed clientshave stronger incentives to provide better audit quality because of their greater reputational capi-tal concerns.10 We also control for auditor industry specialization with the dummy variableINDSPECit, which is equal to one if an audit partner has been classified as a specialist (i.e., theaudit partner’s audited assets for a given industry belong to the highest quartile of its distribu-tion), and zero otherwise (Balsam et al. 2003; Bedard and Biggs 1991; Krishnan 2003; Zerni2012). We expect an audit partner’s industry specialization to improve audit quality. FollowingSundgren and Svanström (2014), we control for the length of an auditor’s career (ln(1 +CAR-EERit)) and the number of clients (ln(1 +NAUDITSit)), but do not have specific predictions forthese variables. We further control for an auditor’s risk preferences by including the dummy vari-able CRIMEi, which is equal to one if an audit partner has been convicted or suspected of acrime, and zero otherwise. We expect that audit partners who have been convicted or suspectedof a crime charge higher audit fees and allow their clients more discretion in financial reporting(Amir et al. 2014), but we do not have a specific prediction for the variable CRIMEi in the going-concern reporting accuracy analyses. Finally, we control for the length of auditor tenure(ln(1 + TENUREijt)) (DeAngelo 1981; Knechel and Vanstraelen 2007; Johnson et al. 2002) andthe relative size of the client (INFLUENCEijt) (Francis and Yu 2009; Li 2009; Ye et al. 2011).We expect the relative size of the client to decrease audit quality, but do not have a predictionfor the length of auditor tenure.

Client-specific variables. Following previous studies (Balsam et al. 2003; Becker et al. 1998;Francis and Yu 2009; Frankel et al. 2002; Johnson et al. 2002; Myers et al. 2003; Reichelt andWang 2010), we include in our empirical analyses a set of control variables for client characteris-tics that are likely to affect audit quality. We include the natural logarithm of total assets (ln(SIZEjt)) and client age (ln(CLIENTAGEjt)) (Becker et al. 1998; Myers et al. 2003). To control forfinancial leverage and liquidity (Balsam et al. 2003; Becker et al. 1998; Francis and Yu 2009;Reichelt and Wang 2010), we include the ratio of total debt to total assets (LEVERAGEjt), theratio of current assets to current liabilities (CURRENTjt), the interest coverage ratio (ln(1 +COV-ERAGEjt)), and the probability of financial distress based on Shumway’s (2001) estimates ofZmijewski’s (1984) financial distress prediction model (ln(PROBZjt)). We further include the ratioof inventory to total assets (INVENTjt) since this account requires extensive auditor judgment(Hay et al. 2006; Simunic 1980). We control for accounting performance including a continuousmeasure of profitability (ROAjt), and the incidence of losses with a dummy variable for negativenet income (LOSSjt) (Francis and Yu 2009). In the going-concern reporting accuracy and abnor-mal accruals analyses, we include a dummy variable for publicly listed clients (LISTEDjt). In thegoing-concern reporting accuracy analyses, we further include a dummy variable that captures thecases in which the client’s total equity is less than half of the common stock due to negativeretained earnings arising from substantial accounting losses (EQ_HALFjt), and a dummy if the cli-ent is a subsidiary of another company (SUBSIDjt) (Knechel et al. 2015). In the audit fee

10. A similar but opposite phenomenon called the contagion effect has been reported both at the audit partner level(Li et al. 2017) and office level (Francis and Michas 2013).



analyses, we include a dummy if the client is audited by two audit firms (JOINTAUDITjt), theproportion of sales generated by foreign operations (FOREIGNSALESjt), a dummy variable forexceptional or extraordinary items (EXCEPTIONjt), and price-to-book ratio (PBjt) (Hay et al.2006; Zerni 2012).

Finally, we include in all our empirical analyses year (YEAR_y), industry (2-digit SIC)(IND_sj), credit rating (CREDITRATE_sj), and audit firm (AUDITFIRM_sj) fixed effects. In theaudit fee and abnormal accruals analyses, we also report results including client fixed effects(CLIENT_sj) instead of industry fixed effects. Because the IQ score of a given audit partner istime-invariant, these results are based on a reduced sample that excludes clients that have notchanged their audit partner during the sample period. We do not include client fixed effects in thegoing-concern reporting accuracy analyses because of the binary nature of the dependent vari-ables (Knechel et al. 2015). Because of the lack of a sufficient number of observations for non-Big N audit firms, we include dummy variables for each of the Big N audit firms and combinenon-Big N audit firms into one dummy variable. All variables are defined in the Appendix.

Model specifications

To examine the association between an audit partner’s IQ and going-concern reporting accuracy,we estimate the following logistic regression model separately for Total reporting error, Type1 reporting error, and Type 2 reporting error11:

Pr GC_ERRORjt = 1� �

= F α+ βIQi + γ0X +Fixed effects + εjt

� �, ð2Þ

where i denotes the audit partner, j denotes the client, and t denotes the year. The dependent vari-able, GC_ERRORjt, is either a dummy variable for Total reporting error (TOTAL_ERRORjt), Type1 reporting error (TYPE_1_ERRORjt), or Type 2 reporting error (TYPE2_ERRORjt). The vector Xincludes a comprehensive set of both auditor- and client-specific control variables. The modelalso includes year, industry (2-digit SIC code), credit rating, and audit firm fixed effects. All coef-ficient z-statistics are reported using robust standard errors clustered at the client level in Totalerror and Type 1 error analyses and at the audit firm level in Type 2 error analysis.12 All variablesin model (2) are as described above and in the Appendix.

We estimate the following OLS regression model to examine the association between anaudit partner’s IQ and audit fees13:

ln AUDITFEESjt� �

= α+ βIQi + γ0X + Fixed effects + εjt , ð3Þ

where i denotes the audit partner, j denotes the client, and t denotes the year. All variables inmodel (3) are as described above and in the Appendix. The vector X includes a set of auditor-and client-specific control variables. The model also includes year, industry (2-digit SIC code)(or alternatively, client), credit rating, and audit firm fixed effects. All coefficient t-statistics arereported using robust standard errors clustered at the client level.

11. We have repeated all our logistic regression analyses using Firth’s (1993) penalized maximum likelihood estimationmethod, which is a method to analyze rare events with logistic regression similar to the method in King and Zeng(2001). The untabulated results of these regressions are similar to those reported in the paper.

12. In Type 2 error analysis, we cluster standard errors at the audit firm level rather than at the client level because thereis only one observation for each client in this analysis.

13. We estimate model (3) using publicly listed companies only, because audit fee data are not readily available in anydatabase for Swedish privately held companies. If multiple partners are involved in auditing a listed client, we cal-culate the average IQ of all the audit partners involved in that audit engagement (see Amir et al. 2014).



To test the association between an audit partner’s IQ and abnormal accruals of the client, weestimate the following OLS regression model for all abnormal accruals and separately for positive(income-increasing) and negative (income-decreasing) abnormal accruals:

jDAjt j = α+ βIQi + γ0X + Fixed effects + εjt, ð4Þ

where i denotes the audit partner, j denotes the client, and t denotes the year. The vector X includesa comprehensive set of auditor- and client-specific control variables. The model also includes year,industry (2-digit SIC code) (or alternatively, client), credit rating, and audit firm fixed effects. Allcoefficient t-statistics are reported using robust standard errors clustered at the client level.

4. Results

Descriptive statistics

We report audit partners’ IQ score distribution in Figure 1 and audit partners’ characteristics inTable 1. Panel A of Table 1 reports descriptive statistics for the variables that capture auditpartners’ personal characteristics. The mean value of audit partners’ IQ equals 6.8, which isclearly higher than the mean value of 5.0 in the rest of the population. This is not surprising,since audit partners must have a master’s degree in accounting and must have passed a challeng-ing CPA exam, and it is also likely that only the most capable auditors are invited as partners.Although a majority of the audit partners have an IQ score that is higher than the population aver-age, there is also substantial variation in their IQ. The average audit partner in the samplereceived CPA certification 17.9 years ago (CAREERit) and has 67.7 clients, including both pri-vately held and publicly listed clients (NAUDITSit). Roughly one-third of the audit partners areindustry specialists (INDSPECit), and 16.1 percent have been convicted or suspected of a crime(CRIMEi). Private clients clearly dominate listed clients in the auditors’ client portfolio (PRO-PLISTCLit). Finally, 73.2 percent of the audit partners work in Big N audit firms (BIGNit).

Panel B of Table 1 reports the mean values of the auditor characteristics for different levelsof an audit partner’s IQ. As for the dummy variables INDSPECit, CRIMEi, and BIGNit, the

Figure 1 IQ score distribution for audit partners

0%

5%

10%

15%

20%

25%

30%

35%

1 2 3 4 5 6 7 8 9

srentrap tidua fo noitcarF

IQ score

Notes: The IQ scores range from one (the lowest IQ) to nine (the highest IQ) with a population average offive. The sample includes all 407 male audit partners who have acted as a lead or deputy auditor for at leastone publicly listed Swedish company during the sample period from 2000 to 2009.



numbers reported in the table show, for each level of the audit partners’ IQ, the proportions ofindustry specialist, criminally convicted, and Big N audit firm partners, respectively. Theseresults show that the audit partners with the highest IQ score have significantly fewer criminalconvictions and have significantly more listed clients than the audit partners with the lowest IQscore. The other variables that capture auditor characteristics do not exhibit any significant differ-ences between the lowest- and highest-IQ audit partners. For example, the results show that theproportion of Big N auditors is not significantly different between the lowest- and highest-IQaudit partners.

Table 2 reports descriptive statistics for the variables used in the analyses of going-concernreporting accuracy (panel A), audit fee (panel B), and abnormal accruals (panel C). Panel A ofTable 2 shows that Type 1 reporting errors are quite rare events (0.5 percent), compared to Type2 reporting errors (78.6 percent). This suggests that the threshold for issuing a going-concernaudit report in Sweden is very high, compared to the more litigious auditing environment in theUnited States (Francis 2004). The proportions of both Type 1 and Type 2 reporting errors are

TABLE 1Audit partners’ characteristics

Panel A: Descriptive statistics for personal characteristics variables

Mean Median SD Min Max N

IQi 6.82 7.00 1.26 2.00 9.00 407PROPLISTCLit 0.08 0.03 0.12 0.00 1.00 407INDSPECit 0.36 0.30 0.30 0.00 1.00 407CAREERit 17.90 18.50 7.65 0.00 35.50 407NAUDITSit 67.75 52.00 54.23 4.50 307.40 407CRIMEi 0.16 0.00 0.37 0.00 1.00 407BIGNit 0.73 1.00 0.44 0.00 1.00 407

Panel B: Means by audit partners’ IQ

1–4(Lowest) 5 6 7 8

9(Highest) Diff:

Lowest − HighestN = 13 N = 43 N = 102 N = 134 N = 72 N = 43

PROPLISTCLit 0.03 0.08 0.07 0.07 0.09 0.10 −0.07***INDSPECit 0.41 0.30 0.42 0.35 0.34 0.35 0.06CAREERit 18.81 17.03 16.70 18.55 19.21 17.11 1.69NAUDITSit 86.77 63.82 70.28 68.30 62.78 66.52 20.25CRIMEi 0.39 0.19 0.16 0.13 0.17 0.16 0.22*BIGNit 0.62 0.77 0.77 0.72 0.69 0.77 −0.15

Notes: This table presents descriptive statistics for variables that capture audit partner characteristics (panel A)and the mean values of the characteristics variables by an audit partner’s IQ score (panel B). See the Appendixfor variable definitions. In panel B, we test whether the mean values of personal characteristics variables aresignificantly different between audit partners with the lowest IQ scores (IQ scores from one to four) and thehighest IQ scores (IQ score of nine) by using a two-tailed t-test. As for the dummy variables INDSPECit,CRIMINALit, and BIGNit, the numbers reported in the table show, for each level of audit partners’ IQ, theproportions of industry specialist, criminally convicted, and Big N audit firm partners, respectively. Thesample includes all 407 male audit partners who have acted as a lead or deputy auditor for at least one publiclylisted Swedish company during the sample period from 2000 to 2009. * and *** indicate significance at the10 and 1 percent two-tailed levels, respectively.



TABLE 2Descriptive statistics for the variables used in the analyses

Variable Mean Median SD Min MaxN

(client-years)

Panel A: Going-concern reporting accuracy analyses

TOTAL_ERRORjt 0.010 0.000 0.101 0.000 1.000 122,012TYPE_1_ERRORjt 0.005 0.000 0.074 0.000 1.000 121,273TYPE_2_ERRORjt 0.786 1.000 0.410 0.000 1.000 739IQi 6.791 7.000 1.273 2.000 9.000 122,012PROPLISTCLit 0.036 0.011 0.073 0.000 1.000 122,012INDSPECit 0.505 1.000 0.499 0.000 1.000 122,012ln(1 +CAREERit) 2.929 3.045 0.423 0.000 3.689 122,012ln(1 +NAUDITSit) 4.789 4.868 0.627 1.099 6.234 122,012CRIMEi 0.179 0.000 0.382 0.000 1.000 122,012ln(1 +TENUREijt) 1.623 1.609 0.523 0.693 2.398 122,012INFLUENCEijt 0.018 0.012 0.018 0.002 0.514 122,012ln(SIZEjt) 16.038 15.774 1.807 8.987 21.591 122,012ln(CLIENTAGEjt) 8.379 8.500 0.994 4.369 10.626 122,012LEVERAGEjt 0.610 0.644 0.262 0.034 0.994 122,012CURRENTjt 2.536 1.502 3.962 0.041 29.931 122,012ln(1 +COVERAGEjt) 11.655 11.673 0.242 0.000 12.822 122,012ln(PROBZjt) 0.104 0.066 0.128 0.000 1.000 122,012INVENTjt 0.126 0.004 0.201 0.000 0.849 122,012ROAjt 0.088 0.078 0.224 −0.857 0.762 122,012LOSSjt 0.253 0.000 0.435 0.000 1.000 122,012EQ_HALFjt 0.005 0.000 0.071 0.000 1.000 122,012SUBSIDjt 0.392 0.000 0.488 0.000 1.000 122,012LISTEDjt 0.009 0.000 0.093 0.000 1.000 122,012407 unique audit partners32,226 unique clients

Panel B: Audit fee analyses

ln(AUDITFEESjt) 14.104 13.978 1.370 11.313 18.050 1,197IQi 6.835 7.000 0.944 4.000 9.000 1,197PROPLISTCLit 0.198 0.154 0.181 0.009 1.000 1,197INDSPECit 0.507 0.500 0.388 0.000 1.000 1,197ln(1 +CAREERit) 3.002 3.045 0.302 1.099 3.611 1,197ln(1 +NAUDITSit) 3.372 3.401 0.593 1.099 5.215 1,197CRIMEi 0.186 0.000 0.291 0.000 1.000 1,197ln(1 +TENUREijt) 1.570 1.609 0.393 0.693 2.398 1,197INFLUENCEijt 0.043 0.027 0.059 0.006 0.613 1,197ln(SIZEjt) 20.536 20.279 1.900 16.884 26.054 1,197ln(CLIENTAGEjt) 8.969 8.828 0.906 5.727 10.580 1,197LEVERAGEjt 0.171 0.149 0.155 0.000 0.583 1,197CURRENTjt 2.318 1.780 1.875 0.543 11.865 1,197ln(1 +COVERAGEjt) 8.584 8.598 2.899 0.000 15.441 1,197ln(PROBZjt) −3.771 −3.957 1.500 −8.250 14.115 1,197INVENTjt 0.125 0.097 0.126 0.000 0.491 1,197ROAjt −0.030 0.041 0.233 −0.973 0.318 1,197LOSSjt 0.353 0.000 0.478 0.000 1.000 1,197JOINTAUDITjt 0.106 0.000 0.308 0.000 1.000 1,197

(The table is continued on the next page.)



similar to those reported in previous studies using data from Sweden (Sundgren and Svanström2014). Untabulated Spearman correlations between the independent variables in models (2)–(4) weregenerally low, with few exceptions. We have reestimated models (2)–(4) after excluding all inde-pendent variables with correlation coefficients greater than 0.5 with another independent variable,and the results were similar to those reported in the paper.

Audit partner’s IQ and going-concern reporting accuracy

Table 3 reports the results of the univariate analysis of exploring the association between auditpartners’ IQ and going-concern reporting accuracy. Specifically, we test whether the mean IQscores are significantly different between those audit partners who have issued an incorrect

TABLE 2 (continued)

Variable Mean Median SD Min MaxN

(client-years)

Panel B: Audit fee analyses

FOREIGNSALESjt 0.229 0.000 0.331 0.000 1.000 1,197EXCEPTIONjt 0.364 0.000 0.481 0.000 1.000 1,197PBjt 2.776 1.951 2.751 0.352 21.724 1,197286 unique audit partners277 unique clients

Panel C: Abnormal accruals analyses

|DAjt| 0.245 0.148 0.287 0.002 1.609 105,169IQi 6.784 7.000 1.274 2.000 9.000 105,169PROPLISTCLit 0.036 0.012 0.073 0.000 1.000 105,169INDSPECit 0.507 1.000 0.499 0.000 1.000 105,169ln(1 +CAREERit) 2.943 3.045 0.412 0.000 3.689 105,169ln(1 +NAUDITSit) 4.786 4.868 0.627 1.099 5.971 105,169CRIMEi 0.180 0.000 0.383 0.000 1.000 105,169ln(1 + TENUREijt) 1.713 1.792 0.467 0.693 2.398 105,169INFLUENCEijt 0.018 0.013 0.018 0.004 0.514 105,169ln(SIZEjt) 16.138 15.887 1.793 13.232 21.591 105,169ln(CLIENTAGEjt) 8.517 8.566 0.867 5.951 10.626 105,169LEVERAGEjt 0.599 0.631 0.261 0.034 0.994 105,169CURRENTjt 2.580 1.528 3.995 0.041 29.931 105,169ln(1 +COVERAGEjt) 11.674 11.673 0.012 10.839 12.822 105,169ln(PROBZjt) 0.097 0.063 0.116 0.000 0.999 105,169INVENTjt 0.129 0.007 0.202 0.000 0.849 105,169ROAjt 0.089 0.079 0.209 −0.857 0.762 105,169LOSSjt 0.247 0.000 0.431 0.000 1.000 105,169LISTEDjt 0.009 0.000 0.093 0.000 1.000 105,169406 unique audit partners26,899 unique clients

Notes: This table reports descriptive statistics for the variables used in the going-concern reporting accuracy(panel A), audit fee (panel B), and abnormal accruals (panel C) analyses. The sample in panels A and C (panelB) includes privately held and publicly listed clients (publicly listed clients) of all male audit partners whohave acted as a lead or deputy auditor for at least one publicly listed Swedish company during the sampleperiod from 2000 to 2009. See the Appendix for variable definitions. The variables SIZEjt and AUDITFEESjtare in SEK. All continuous variables are winsorized to the 1st and 99th percentiles of their distributions.



going-concern audit report and those who have not issued such a report. Panel A of Table 3reports the results for the full sample, and panel B reports the results for a reduced sample, whichexcludes clients in extreme financial distress, defined as those that belong to the 10th decile ofthe variable PROBZjt. We report the results for this reduced sample because clients in extremefinancial distress are very likely to fail, and consequently, while there may be some room forjudgment, the auditor often has no choice but to issue a going-concern audit report.

The results reported in Table 3 show that the IQ of an average audit partner who has issued anincorrect going-concern audit report is significantly lower than that of an audit partner who has notissued such a report, suggesting that audit partners’ IQ is associated with improved going-concern

TABLE 3Univariate analysis of audit partner’s IQ and going-concern reporting accuracy

Going-concern reporting error

Yes NoDifference: Yes − NoMean IQ Mean IQ

Panel A: Full sample

Total reporting error 6.69 6.79 −0.10***(2.63)

N (# of unique audit partners) 1,246 (287) 120,766 (407)

Type 1 reporting error 6.64 6.79 −0.15***(2.99)

N (# of unique audit partners) 665 (222) 120,608 (407)

Type 2 reporting error 6.75 6.95 −0.20*(1.84)

N (# of unique audit partners) 581 (213) 158 (104)

Panel B: Reduced sample

Total reporting error 6.67 6.79 −0.12***(2.65)


Type 1 reporting error 6.58 6.79 −0.21***(3.24)


Type 2 reporting error 6.74 6.97 −0.23**(1.98)

N (# of unique audit partners) 548 (208) 119 (87)

Notes: This table reports the univariate results for the association between an audit partner’s IQ and going-concern reporting accuracy for both the full sample (panel A) and the reduced sample excluding clients inextreme financial distress, defined as those belonging to the 10th decile of the variable PROBZ (panel B).Specifically, we test whether the mean IQ scores are significantly different between those audit partners whohave issued an incorrect going-concern audit report and those who have not using a two-tailed t-test. Weconduct this analysis separately for Total reporting error (the sum of Type 1 and Type 2 reporting errors), andType 1 and Type 2 reporting errors. The IQ score obtains values from one (the lowest IQ) to nine (the highestIQ). The sample includes 32,226 privately held and publicly listed clients of all 407 male audit partners whohave acted as a lead or deputy auditor for at least one publicly listed Swedish company during the sampleperiod from 2000 to 2009. See the Appendix for variable definitions. The t-statistics are presented inparentheses. *, **, and *** indicate significance at the 10, 5, and 1 percent two-tailed levels, respectively.



TABLE4

Regressionanalysisof

auditpartner’sIQ

andgoing-concernreportingaccuracy

Pred.

sign

(1)Dependent

variable=TOTAL_E

RRORjt

(2)Dependent

variable=TYPE_1_E

RRORjt

(3)Dependent

variable=TYPE_2_E

RRORjt

Fullsample

Reduced

sample

Fullsample

Reduced

sample

Fullsample

Reduced

sample

IQi

−−0.093***

−0.130***

−0.089***

−0.142***

−0.102

−0.170***

(−2.93)

(−3.32)

(−2.47)

(−3.18)

(−1.12)

(−2.39)

PROPLISTCLit

−0.495

0.824

0.427

1.187*

0.420

−1.462

(0.85)

(1.12)

(0.60)

(1.63)

(0.08)

(−0.23)

INDSP

ECit

−−0.001

0.119

0.016

0.189*

0.072

0.093

(−0.01)

(1.05)

(0.16)

(1.47)

(0.24)

(0.29)

ln(1+CAREERit)

?−0.022

0.038

0.034

−0.002

0.076

0.296

(−0.23)

(0.32)

(0.32)

(−0.01)

(0.22)

(1.05)

ln(NAUDITS it)

?−0.184

−0.197

−0.212

−0.266

−0.420

−0.615

(−1.41)

(−1.39)

(−1.39)

(−1.42)

(−0.76)

(−1.59)

CRIM

Ei

?−0.118

−0.085

−0.106

−0.092

−0.264

0.129

(−1.08)

(−0.64)

(−0.88)

(−0.60)

(−0.50)

(0.17)

ln(TENUREijt)

?0.065

0.026

0.059

0.045

−0.591**

−0.657**

(0.68)

(0.23)

(0.54)

(0.33)

(−2.10)

(−1.98)

INFLUENCEijt

+−2.474

0.561

−2.834

−0.883

−44.256*

−51.281**

(−0.58)

(0.12)

(−0.54)

(−0.14)

(−1.33)

(−2.05)

ln(SIZEjt)

−0.078***

0.108***

0.077**

0.123***

0.171

0.123

(2.59)

(2.87)

(2.25)

(3.06)

(0.99)

(0.46)

ln(CLIENTAGEjt)

−−0.036

0.031

0.019

0.058

0.136

0.158

(−0.75)

(0.50)

(0.35)

(0.85)

(1.28)

(0.91)

LEVERAGEjt

+1.291***

1.089***

0.756***

0.384

−1.232

−1.042

(5.53)

(3.55)

(2.74)

(0.79)

(−0.84)

(−0.52)

CURRENTjt

−−0.014

−0.008

−0.010

−0.005

0.046

0.019

(−0.99)

(−0.47)

(−0.50)

(−0.27)

(0.40)

(0.15)

ln(1+COVERAGEjt)

−−1.902***

−2.594***

−0.206

−0.196*

2.283***

2.188***

(−37.32)

(−30.48)

(−0.91)

(−1.55)

(4.70)

(2.66)

ln(PROBZjt)

+−2.607***

−3.574***

0.846**

0.699

−0.108

−0.924

(−5.65)

(−4.60)

(1.88)

(0.36)

(−0.16)

(−0.95)

(The

tableiscontinuedon

thenext

page.)



TABLE4(contin

ued)

Pred.

sign

(1)Dependent

variable=TOTAL_E

RRORjt

(2)Dependent

variable=TYPE_1_E

RRORjt

(3)Dependent

variable=TYPE_2_E

RRORjt

Fullsample

Reduced

sample

Fullsample

Reduced

sample

Fullsample

Reduced

sample

INVENTjt

+−0.668***

−0.974***

−0.989***

−1.264***

0.291*

0.284

(−2.62)

(−3.02)

(−3.32)

(−3.18)

(1.50)

(0.69)

ROAjt

−−0.921***

−0.868**

−0.049

−0.080

0.259

−0.732***

(−3.30)

(−2.17)

(−0.15)

(−0.18)

(0.97)

(−3.14)

LOSS

jt+

0.730***

0.214*

0.531***

0.308**

−0.610***

−0.595**

(7.73)

(1.48)

(4.71)

(1.96)

(−2.46)

(−2.07)

EQ_H

ALFjt

+1.753***

−0.943**

3.825***

4.068***

−3.488***

−3.504***

(4.45)

(−1.71)

(23.36)

(11.76)

(−6.14)

(−5.86)

SUBSIDjt

?−0.605***

−0.804***

−0.821***

−1.018***

0.007

0.131

(−6.69)

(−6.81)

(−7.60)

(−7.02)

(0.02)

(0.51)

LISTEDjt

?−0.725*

−1.881***

−1.660**

−11.774***

9.301***

8.020***

(−1.73)

(−4.52)

(−2.30)

(−41.72)

(9.02)

(4.79)

Intercept

Yes

Yes

Yes

Yes

Yes

Yes

Yearfixedeffects

Yes

Yes

Yes

Yes

Yes

Yes

Creditratin

gfixedeffects

Yes

Yes

Yes

Yes

Yes

Yes

Auditfirm

fixedeffects

Yes

Yes

Yes

Yes

Yes

Yes

Industry

fixedeffects

Yes

Yes

Yes

Yes

Yes

Yes

Clustered

standard

errors

Client

Client

Client

Client

Auditfirm

Auditfirm

PseudoR2(%

)43.86

54.24

22.61

17.98

60.02

57.80

LRratio

(χ2)

5,904.12***

5,694.73***

1,816.94***

901.16***

362.43***

289.13

***

N(#

oferrors)

122,012(1,246)

108,981(933)

121,273(665)

108,314(385)

739(581)

667(548)

Notes:Thistablereportsthelogisticregression

results

fortheassociationbetweenan

auditpartner’sIQ

andgoing-concernreportingaccuracy

forboth

thefullsample

andthereducedsampleexcludingclientsin

extrem

efinancialdistress

(definedas

thosebelongingto

the10th

decileof

thevariablePROBZjt).Colum

n(1)reportsthe

results

forTotalreportingerror.Colum

ns(2)and(3)reporttheresults

forType1andType2reportingerrors,respectively.

See

theAppendixforvariabledefinitio

ns.

The

sampleincludes

32,226

privatelyheld

andpublicly

listedclientsof

all407maleauditpartners

who

have

actedas

alead

ordeputy

auditorforatleastonepublicly

listedSwedishcompany

during

thesampleperiod

from

2000

to2009.A

llcontinuous

variablesarewinsorizedto

the1stand99th

percentiles

oftheirdistributio

ns.T

hez-statisticsfrom

robuststandard

errors

clusteredattheclient

(colum

ns(1)and(2))or

auditfirm

(colum

n(3))levelarepresentedin

parentheses.*,

**,and

***indicate

significanceatthe10,5

,and

1percentone-tailed(two-tailed)

levelsforvariableswith

(with

out)apredictedsign,respectively.



reporting accuracy. This result holds for all three reporting error types (i.e., Total, Type 1, and Type2 reporting errors). The results for the full sample and the reduced sample are similar.

Table 4 reports the results of estimating the going-concern reporting accuracy model (model(2)) for both the full sample and the reduced sample separately for Total reporting error (column (1))and Type 1 (column (2)) and Type 2 (column (3)) reporting errors. The results show that the esti-mated coefficients for IQi are significantly negative for all three reporting error types, with theexception of Type 2 reporting error for the full sample. These results suggest that, after controllingfor relevant auditor and client characteristics, going-concern reporting accuracy increases with anaudit partner’s IQ.14

The results for the control variables in Table 4 show that the estimated coefficients for theauditor-specific control variables are generally insignificant. As for the client-specific controlvariables, Total reporting errors are more likely to occur when the client is larger (ln(SIZEjt)),more financially leveraged (LEVERAGEjt), has lower interest coverage ratio (ln(1 +COVERA-GEjt)) or probability of financial distress (ln(PROBZjt)), equity capital lower than half of sharecapital (EQ_HALFjt), less inventory (INVENTjt), or weaker accounting performance (ROAjt andLOSSjt).

We next provide insight into the economic significance of these results by calculating marginalchanges in the probability that an audit partner issues an incorrect going-concern audit report as aresult of changing the levels of the explanatory variables in model (2). The marginal effects showthe marginal changes in the probability for a unit increase in each explanatory variable while hold-ing the other explanatory variables at their mean values. In Table 4, the untabulated marginal effectsof IQi for the full sample (reduced sample) are −0.10 (−0.05), −0.05 (−0.05), and − 0.9 (−1.3) per-cent for Total error, Type 1 error, and Type 2 error, respectively. The untabulated odds ratios forthe variable IQ using the full sample (reduced sample) are 0.911 (0.878), 0.915 (0.868), and 0.903(0.878) for Total error, Type 1 error, and Type 2 error, respectively.

Audit partner’s IQ and audit fees

Table 5 reports the results of the univariate analysis of examining the association between auditpartners’ IQ and audit fees (in million SEK). Specifically, we first sort audit fees into quartiles,where Quartile 1 (Quartile 4) includes the lowest (highest) fees, and then calculate the mean IQscore of audit partners in each fee quartile. Column (1) reports the results for the full sample andcolumn (2) reports the results for the reduced sample, which excludes clients that have not chan-ged their audit partner during the sample period. We report the results for the reduced samplebecause excluding clients that have not changed their audit partner during the sample periodallows us to include client fixed effects in the regression model. The results reported in Table 5show that audit partners’ mean IQ score increases monotonically across the audit fee quartiles.The mean IQ score is 6.58 (6.60) in Quartile 1 for the full (reduced) sample and 7.16 (7.15) inQuartile 4 for the full (reduced) sample, the difference being statistically significant at the 1 per-cent level.

Table 6 reports the results of estimating the audit fee model (model (3)) for both the full sam-ple (column (1)) and the reduced sample including client fixed effects (column (2)). These resultsshow that the estimated coefficient of IQi is significantly positive both in the full sample and in

14. We have also repeated the univariate and multivariate analyses presented in Tables 3 and 4 for the Type 2 reportingerror by using a two-year-ahead (24-month) failure period instead of a one-year-ahead (12-month) failure period.The untabulated results of these analyses show that an audit partner’s IQ is significantly positively associated withgoing-concern reporting accuracy for Total and Type 1 errors for the full and reduced samples, but that there is nosignificant association between an audit partner’s IQ and Type 2 errors. The weaker results for Type 2 errors aremost likely due to the fact that in Sweden the auditor’s task is to evaluate whether there are material uncertaintiesabout the company’s ability to continue as a going concern for a time span of typically 12 months from the fiscalyear-end (RS 570, §18).



the reduced sample, thereby providing consistent evidence that clients are willing to pay a feepremium for audit services of high-IQ audit partners. This effect is economically significant, as aone-unit increase in an audit partner’s IQ increases audit fees by 6.61 (5.23) percent for the fullsample (reduced sample), holding the control variables constant at their mean values. The explan-atory power of both model specifications reported in Table 6 is comparable to that reported inprevious studies, and the signs of the coefficients for the control variables are as expected. Theresults for the auditor-specific control variables show that clients of audit partners with criminalconvictions (CRIMEi) pay higher fees, and influential clients pay lower fees. As for the client-specific control variables, the estimated coefficients of ln(SIZEjt) and EXCEPTIONjt are signifi-cantly positive, and those of CURRENTjt and ln(1 +COVERAGEjt) are significantly negative forboth the full sample and the reduced sample. The rest of the control variables do not exhibit anyconsistently significant relation to audit fees.

Audit partner’s IQ and abnormal accruals

Table 7 reports the results of the univariate analysis of exploring the association betweenaudit partners’ IQ and abnormal accruals. Specifically, we first divide both the full sampleand the reduced sample, which excludes clients that have not changed their audit partner dur-ing the sample period, into five quintiles based on the absolute value of abnormal accruals(the variable |DAjt|), where Quintile 1 (Quintile 5) includes the lowest (highest) abnormalaccruals. We then calculate audit partners’ mean IQ score in each abnormal accruals quintile.Column (1) reports the results for the absolute value of abnormal accruals when both income-increasing and income-decreasing abnormal accruals are included. Columns (2) and (3) reportthe results for the absolute values of income-increasing and income-decreasing abnormalaccruals, respectively.

The results reported in Table 7 show that, for the full sample of income-increasing abnormalaccruals, audit partners’ mean IQ score decreases from Quintile 1 to Quintile 5, the average

TABLE 5Univariate analysis of audit partner’s IQ and audit fees

Audit fee quartile

(1)Full sample

(2)Reduced sample

N Mean audit fees Mean IQ N Mean audit fees Mean IQ

Quartile 1 (lowest) 299 0.293 6.58 223 0.356 6.60Quartile 2 299 0.805 6.63 220 0.959 6.67Quartile 3 300 1.862 6.97 229 2.254 6.98Quartile 4 (highest) 299 13.259 7.16 222 16.419 7.15Difference: Q1 – Q4 −12.966 −0.58*** −16.063 −0.55***

(−7.83) (−6.71)

Notes: This table reports the univariate results for the association between an audit partner’s IQ and auditfees for both the full sample (column (1)) and the reduced sample excluding clients that have not changedtheir audit partner during the sample period (column (2)). Audit fees are sorted into quartiles, where Quartile1 (Quartile 4) includes the lowest (highest) fees. We test whether the mean IQ scores are significantlydifferent between Quartiles 1 and 4 by using a (two-tailed) t-test. The IQ score obtains values from one(the lowest IQ) to nine (the highest IQ). The sample includes 277 publicly listed clients of all 286 male auditpartners who have acted as a lead or deputy auditor for at least one publicly listed Swedish company duringthe sample period from 2000 to 2009. Audit fees are in million SEK. The t-statistics are presented inparentheses. *** indicates significance at the 1 percent two-tailed level.



TABLE 6Regression analysis of audit partner’s IQ and audit fees

Dependent variable = ln(AUDITFEESjt)

Pred. sign(1)

Full sample(2)

Reduced sample

IQi + 0.064*** 0.051*(2.40) (1.63)

PROPLISTCLit + 0.112 −0.076(1.04) (−0.76)

INDSPECit + −0.109** 0.035(−2.12) (0.97)

ln(1 +CAREERit) ? 0.053 −0.100*(0.66) (−1.56)

ln(1 +NAUDITSit) ? −0.163*** −0.038(−2.95) (−0.84)

CRIMEi + 0.167** 0.175**(2.16) (1.85)

ln(1 + TENUREijt) ? −0.020 0.055*(−0.32) (1.34)

INFLUENCEijt − −0.539 −0.654**(−1.23) (−2.25)

ln(SIZEjt) + 0.599*** 0.356***(25.87) (8.65)

ln(CLIENTAGEjt) + 0.027 0.101(0.75) (0.60)

LEVERAGEjt + −0.122 0.641***(−0.64) (3.08)

CURRENTjt − −0.083*** −0.066***(−5.27) (−3.31)

ln(1 +COVERAGEjt) − −0.039*** −0.019**(−2.95) (−1.94)

ln(PROBZjt) + 0.032 −0.090***(0.72) (−3.64)

INVENTjt + −0.011 0.431*(−0.04) (1.35)

ROAjt − −0.089 −0.657***(−0.33) (−3.49)

LOSSjt + −0.002 0.018(−0.03) (0.43)

JOINTAUDITjt + 0.100 0.232***(1.28) (2.40)

FOREIGNSALESjt + 0.127* 0.067(1.52) (0.89)

EXCEPTIONjt + 0.079*** 0.073***(2.52) (3.33)

PBjt + 0.010 −0.012*(1.21) (−1.61)

Intercept Yes YesYear fixed effects Yes YesCredit rating fixed effects Yes YesAudit firm fixed effects Yes YesIndustry fixed effects Yes No

(The table is continued on the next page.)



IQ scores being significantly different at the 10 percent level. However, this result does not holdfor the reduced sample. We do not observe any statistically significant differences in auditpartners’ average IQ between the abnormal accruals quintile groups when both income-increasingand income-decreasing abnormal accruals are included or when only income-decreasing abnormalaccruals are included.

Table 8 reports the results of estimating the abnormal accruals model (model (4)) for boththe full sample and the reduced sample, including client fixed effects. These results show thatthe estimated coefficient of IQi is significantly negative for the absolute value of income-increasing accruals in the full sample, but it is insignificant in the reduced sample. The estimatedcoefficient of IQi for the absolute value of income-decreasing abnormal accruals is significantlypositive in the full sample but insignificant in the reduced sample. The significantly positivecoefficient of IQ for the absolute value of income-decreasing accruals is consistent with Lennoxet al. (2016), who find that audit adjustments result in smaller income-increasing and largerincome-decreasing accruals and that these adjustments help to increase earnings quality. Theestimated coefficient of IQi is not significant for the absolute value of abnormal accruals whenboth income-increasing and income-decreasing abnormal accruals are included. All the auditor-and client-specific control variables in Table 8 are generally significantly related to the absolutevalue of abnormal accruals when both income-increasing and income-decreasing abnormalaccruals are included or when only income-increasing or income-decreasing abnormal accrualsare included.

In sum, we find some evidence that auditors’ IQ scores are negatively associated with income-increasing earnings management of their client. We find only very weak evidence that auditors’ IQis associated with clients’ income-decreasing earnings management. However, given that theseresults only hold in the full sample of clients, and not in the reduced sample, these results should beinterpreted as suggestive. Economic significance of these results is also weak, as a one-unit increasein an audit partner’s IQ decreases (increases) the absolute value of income-increasing accruals byonly 0.2 (0.1) percent for the full sample, holding the control variables constant at their meanvalues.

TABLE 6 (continued)

Dependent variable = ln(AUDITFEESjt)

Pred. sign(1)

Full sample(2)

Reduced sample

Client fixed effects No YesClustered standard errors Client ClientAdjusted R2 (%) 88.52 96.47N 1,197 894

Notes: This table reports the OLS regression results for the association between an audit partner’s IQ andaudit fees for both the full sample (column (1)) and the reduced sample excluding clients that have notchanged their audit partner during the sample period (column (2)). See the Appendix for variable definitions.The sample includes 277 publicly listed clients of all 286 male audit partners who have acted as a lead ordeputy auditor for at least one publicly listed Swedish company during the sample period from 2000 to2009. All continuous variables are winsorized to the 1st and 99th percentiles of their distributions. Thet-statistics from robust standard errors clustered at the client level are presented in parentheses. *, **, and*** indicate significance at the 10, 5, and 1 percent one-tailed (two-tailed) levels for variables with (without)a predicted sign, respectively.



TABLE7

Univariateanalysisof

auditpartner’sIQ

andabnorm

alaccruals

Abnormalaccrualsquintile

(1)

All

(2)

Income-increasing

(3)

Income-decreasing

Mean|DA|

MeanIQ

MeanIQ

MeanIQ

Fullsample

Reduced

sample

Fullsample

Reduced

sample

Fullsample

Reduced

sample

Quintile

1(low

est)

0.023

6.80

6.77

6.82

6.80

6.78

6.73

N=21,033

N=21,033

N=2,285

N=10,757

N=1,182

N=10,276

N=1,103

Quintile

20.075

6.77

6.77

6.79

6.79

6.75

6.75

N=21,034

N=21,034

N=2,286

N=10,758

N=1,182

N=10,276

N=1,104

Quintile

30.149

6.78

6.72

6.78

6.71

6.79

6.74

N=21,034

N=21,034

N=2,285

N=10,758

N=1,182

N=10,277

N=1,103

Quintile

40.276

6.78

6.80

6.79

6.78

6.77

6.81

N=21,034

N=21,034

N=2,286

N=10,758

N=1,182

N=10,276

N=1,104

Quintile

5(highest)

0.705

6.79

6.74

6.78

6.73

6.79

6.76

N=21,034

N=21,034

N=2,285

N=10,757

N=1,182

N=10,276

N=1,103

Difference:

Q1–Q5

−0.682

0.01

0.03

0.04

*0.07

−0.01

−0.03

(0.69)

(0.67)

(1.88)

(1.25)

(−0.81)

(−0.49)

Notes:Thistablereportstheunivariateresults


auditpartner’sIQ

andabnorm

alaccrualsforboth

thefullsampleandthereducedsample

excludingclientsthat

have

notchangedtheirauditordu

ring

thesampleperiod.C

olum

n(1)reportstheresults

fortheabsolutevalueof

abnorm

alaccrualswhenboth

positiv

e(incom

e-increasing)andnegativ

e(incom

e-decreasing)abnorm

alaccrualsareincluded.C

olum



income-

increasing

andincome-decreasing

abnorm

alaccruals,respectively.

Ineach

column,

abnorm

alaccrualsaresorted

into

quintiles,w

here

Quintile

1(Q

uintile

5)includes

thelowest(highest)abnorm

alaccruals.W

etestwhether

themeanIQ

scores

aresignificantly

differentbetweenQuintiles1and5by

usingatwo-tailedt-test.T

heIQ

scoreobtainsvalues

from

one(the

lowestIQ

score)

tonine

(the

highestIQ

score).T

hesampleincludes

26,899

privatelyheld

andpublicly

listedclientsof

all406male

auditpartnerswho

have

actedas

alead

ordeputy

auditorforat

leastonepublicly


during

thesampleperiod

from

2000

to2009.S

eethe

Appendixforvariabledefinitio

ns.T

het-statisticsarepresentedin

parentheses.*indicatessignificanceat

the10

percenttwo-tailedlevel.



TABLE8

Regressionanalysisof

auditpartner’sIQ

andabnorm

alaccruals

(1)

All

(2)

Income-increasing

(3)

Income-decreasing

Dependent

variable=|DAjt|

Dependent

variable=|DAjt|

Dependent

variable=|DAjt|

Predicted

sign

Fullsample

Reduced

sample

Fullsample

Reduced

sample

Fullsample

Reduced

sample

IQi

−−0.0004

0.003

−0.002**

−0.001

0.001*

−0.000

(−0.62

)(0.78)

(−1.86)

(−0.14)

(1.47)

(−0.05)

PROPLISTCLit

−0.056***

0.057*

0.053***

0.136**

0.061***

−0.002

(4.17)

(1.54)

(2.66)

(2.13)

(3.50)

(−0.03)

INDSP

ECit

−−0.004**

0.024***

−0.006**

0.014

−0.001

0.023**

(−2.08)

(3.07)

(−2.37)

(1.16)

(−0.51)

(1.95)

ln(1+CAREERit)

?0.001

0.030***

0.001

0.043***

0.000

0.014

(0.60)

(3.17)

(0.44)

(2.82)

(0.08)

(0.98)

ln(1+NAUDITS it)

?0.006**

−0.005

0.005

−0.011

0.007**

0.001

(2.42)

(−0.59)

(1.39)

(−0.69)

(2.09)

(0.12)

CRIM

Ei

+0.005**

0.002

0.006**

0.016

0.004

−0.013

(2.13)

(0.22)

(2.01)

(0.95)

(1.14)

(−0.86)

ln(1+TENUREijt)

?−0.029***

−0.041***

−0.035***

−0.053***

−0.022***

−0.027***

(−12.59)

(−5.90)

(−10.31)

(−4.86)

(−7.16)

(−2.66)

INFLUENCEijt

+0.180**

−0.016

0.209**

−0.152

0.170**

0.055

(2.28)

(−0.08)

(1.75)

(−0.44)

(1.68)

(0.25)

ln(SIZEjt)

−−0.011***

0.043***

−0.004***

0.074***

−0.018***

−0.005

(−17.78)

(4.67)

(−4.36)

(4.49)

(−21.96)

(0.25)

ln(FIRMAGEjt)

−−0.016***

−0.067***

−0.025***

−0.082**

−0.006***

−0.021

(−13.51)

(−2.89)

(−14.93)

(−2.32)

(−3.51)

(−0.41)

LEVERAGEjt

+0.039***

0.030

0.052***

0.025

0.034***

0.108**

(7.22)

(0.93)

(6.74)

(0.44)

(4.67)

(2.32)

CURRENTjt

−0.004***

0.003**

0.006***

0.006***

0.003***

−0.000

(15.05

)(1.99)

(13.61)

(2.85)

(6.66)

(−0.08)

ln(1+COVERAGEjt)

−−0.07

6−0.249

−0.099

0.059

−0.067

−0.666*

(−1.11)

(−0.83)

(−0.91)

(0.12)

(−0.84)

(−1.41)

(The

tableiscontinuedon

thenext

page.)



TABLE8(contin

ued)

(1)

All

(2)

Income-increasing

(3)

Income-decreasing

Dependent

variable=|DAjt|

Dependent

variable=|DAjt|

Dependent

variable=|DAjt|

Predicted

sign

Fullsample

Reduced

sample

Fullsample

Reduced

sample

Fullsample

Reduced

sample

ln(PROBZjt)

+0.196***

0.012

0.135***

−0.175**

0.224***

0.070

(13.17)

(0.23)

(5.67)

(−1.75)

(11.97)

(0.90)

INVENTjt

+0.048***

−0.027

0.168***

0.216**

−0.104***

−0.307***

(7.35)

(−0.38)

(17.64)

(1.94)

(−12.93)

(−3.45)

ROAjt

?0.13

1***

0.056*

0.174***

0.135**

0.090***

−0.000

(16.89)

(1.71)

(14.81)

(2.46)

(8.94)

(−0.01)

LOSS

jt?

0.01

5***

0.012

0.013***

0.018

0.020***

0.009

(5.72)

(1.27)

(3.46)

(1.17)

(5.68)

(0.65)

LISTEDjt

−−0.039***

−0.494*

−0.048***

−0.167

−0.022**

−0.226

(−4.60)

(−5.57)

(−3.88)

(−1.14)

(−1.93)

(−1.89)

Intercept

Yes

Yes

Yes

Yes

Yes

Yes

Yearfixedeffects

Yes

Yes

Yes

Yes

Yes

Yes

Creditratin

gfixedeffects

Yes

Yes

Yes

Yes

Yes

Yes

Auditfirm

fixedeffects

Yes

Yes

Yes

Yes

Yes

Yes

Industry

fixedeffects

Yes

No

Yes

No

Yes

No

Client

fixedeffects

No

Yes

No

Yes

No

Yes

Clustered

standard

errors

Client

Client

Client

Client

Client

Client

AdjustedR2(%

)7.08

16.78

9.46

19.25

8.12

16.80

N105,169

11,427

53,788

5,910

51,381

5,517

Notes:ThistablereportstheOLSregression

results


auditpartner’sIQ

andabnorm

alaccrualsforboth

thefullsampleandthereduced

sampleexcludingclientsthathave

notchangedtheirauditorduring

thesampleperiod.C

olum

n(1)reportstheresults


abnorm

alaccrualswhen

both

positiv

e(incom

e-increasing)andnegativ

e(incom

e-decreasing)abnorm

alaccrualsareincluded.C

olum



income-increasing

andincome-decreasing

abnorm

alaccruals,respectively.

See

theAppendixforvariabledefinitio

ns.T

hesampleincludes

26,899

privatelyheld

and

publicly

listedclientsof

all406maleauditpartnerswho

have

actedas

alead

ordeputy

auditorforatleastonepublicly


during

thesample

period

from

2000

to2009.T

het-statisticsfrom

robuststandard

errors

clusteredattheclient

levelarepresentedin

parentheses.*,

**,and

***indicate

significanceat

the10,5

,and

1percentone-tailed(two-tailed)

levelsforvariableswith

(with

out)apredictedsign.



5. Conclusions

A recent boom in archival research on individual audit partners has been motivated by limitationswith using an entire audit firm as the unit of analysis (Lennox and Wu 2018). Our paper contributesto this emerging literature by exploring whether an audit partner’s cognitive ability (IQ) is related toaudit quality. Although experimental research suggests that cognitive ability plays an important rolein judgment and decision making in the audit process (Bonner and Lewis 1990; Gibbins 1984;Nelson and Tan 2005), it is not obvious that an auditor’s cognitive ability matters for audit quality. Inparticular, audit regulation and audit firms’ own control and risk mechanisms and knowledge-sharingsystems may result in the standardization of audits and thus reduce variation in the quality of audits.Audits are, moreover, conducted by teams, not by individuals. Experiments do not usually take intoaccount firm-level quality control systems or the fact that audits are conducted by a group of individ-uals, and consequently the question remains whether differences in auditors’ cognitive ability couldexplain the documented variation in the quality of audit services within the same audit firm.

To the best of our knowledge, this is the first archival study on the role of auditors’ cognitiveability in the quality of audit services. Using Swedish audit partners’ IQ scores from psychologicaltests combined with data on their complete client portfolios, we show that an audit partner’s IQ ispositively related to going-concern reporting accuracy. We also find that an audit partner’s IQ ispositively related to audit fees, suggesting that clients are willing to pay a fee premium for the auditservices of smarter audit partners. Finally, we find some, albeit weak, evidence that an auditpartner’s IQ is negatively related to the income-increasing abnormal accruals of the client.

This study is subject to limitations. First, audit quality is a complex concept involving many dimen-sions that must be measured by using proxies that contain measurement error. In an attempt to triangu-late our findings, we use several proxies that capture different dimensions of audit quality. Second, theuse of archival data raises endogeneity concerns. In particular, an audit partner’s IQ could proxy for cor-related omitted variables, or the direction of causation could run from clients with higher audit qualityto the use of more capable auditors. Third, data availability issues impose some limitations on the ana-lyses because audit fees are not readily available for Swedish privately held companies.

This study provides an important and interesting insight into the nature of audit services. Weshow that, in spite of the various mechanisms that aim to standardize audit quality, the intellectualability of the audit partner still matters for the quality of the audit services delivered to the client.Therefore, these findings may be of interest to those developing and making decisions on mecha-nisms that control and regulate the quality of audit services.

Appendix

Variable definitions

Variable Description Data source

Panel A: Audit quality proxies

TYPE_1_ERRORjt Dummy variable equal to one if an audit partner issuesa going-concern audit report to a client thatsubsequently does not file for bankruptcy within12 months of the audit partner’s report, and zerootherwise

UC AB

TYPE_2_ERRORjt Dummy variable equal to one if an audit partner doesnot issue a going-concern audit report to a client thatsubsequently files for bankruptcy within 12 months ofthe audit partner’s report, and zero otherwise

UC AB

(The Appendix is continued on the next page.)



Appendix (continued)

TOTAL_ERRORjt Dummy variable equal to one if TYPE_1_ERROR orTYPE_2_ERROR equals one, and zero otherwise

UC AB

ln(AUDITFEESjt) The natural logarithm of audit fees. Audit fees arestated in SEK

COMPUSTAT GlobalVantage

DAjt Abnormal accruals of a client, defined as the residualfrom the modified cross-sectional Jones (1991) modelestimated with an intercept (Kothari et al. 2005)

UC AB

|DAjt| The absolute value of the variable DA UC AB

Panel B: Audit partner’s IQ

IQi Audit partners’ IQ score ranging from one (the lowestIQ) to nine (the highest IQ)

Swedish Armed Forces

Panel C: Auditor-specific control variables

PROPLISTCLit The proportion of publicly listed clients of an auditpartner

UC AB; COMPUSTATGlobal Vantage

INDSPECit Dummy variable equal to one if an audit partner isclassified as an industry specialist according to thecriteria that the audit partner’s audited assets for agiven industry belong to the highest quartile of itsdistribution, and zero otherwise


ln(1 +CAREERit) The natural logarithm of one plus the number of yearssince an audit partner received professionalcertification

Professional Institute forCertified Auditors andother AccountingProfessionals

ln(1 +NAUDITSit) The natural logarithm of one plus the total number ofclients of an audit partner


CRIMEi Dummy variable equal to one if an audit partner hasbeen convicted or suspected of a crime, and zerootherwise

Swedish National Councilfor Crime Prevention;Swedish NationalPolice Board

ln(1 + TENUREijt) The natural logarithm of one plus the number of yearsan audit partner has audited a client


INFLUENCEijt The natural logarithm of the proportion of a client’ssales of the sales of all clients of an audit partner


BIGNit Dummy variable equal to one if an audit partner worksfor one of the Big N firms (PWC, Deloitte, KPMG,Ernst &Young, or Arthur Andersen), and zerootherwise


Panel D: Client-specific control variables

ln(SIZEjt) The natural logarithm of total assets of a client. Totalassets are stated in SEK


ln(CLIENTAGEjt) The age of a client UC AB; COMPUSTATGlobal Vantage

LEVERAGEjt The ratio of total debt to total assets of a client UC AB; COMPUSTATGlobal Vantage

CURRENTjt The ratio of current assets to current liabilities of aclient


(The Appendix is continued on the next page.)



References

Abdolmohammadi, M. J., and U. J. Shanteau. 1992. Personal attributes of expert auditors. OrganizationalBehavior and Human Decision Processes 53 (2): 158–72.

Adams, R., M. Keloharju, and S. Knupfer. 2015. Are CEOs born leaders? Lessons from traits of a millionindividuals. Working paper, Harvard Business School.

Amir, E., J. P. Kallunki, and H. Nilsson. 2014. The association between individual audit partners’ risk pref-erences and the composition of their client portfolios. Review of Accounting Studies 19 (1): 103–33.

Appendix (continued)

ln(1 +COVERAGEjt) The natural logarithm of one plus the interest coverageratio of a client


ln(PROBZjt) The natural logarithm of the probability of financialdistress of a client based on Shumway’s (2001)estimates of Zmijewski’s (1984) financial distressprediction model


INVENTjt The ratio of inventory to total assets of a client UC AB; COMPUSTATGlobal Vantage

ROAjt The return on assets ratio of a client UC AB; COMPUSTATGlobal Vantage

LOSSjt Dummy variable equal to one if a client’s net income isnegative, and zero otherwise


EQ_HALFjt Dummy variable equal to one if a client’s total equity isless than half of the common stock due to negativeretained earnings arising from substantial accountinglosses, and zero otherwise

UC AB

SUBSIDjt Dummy variable equal to one if a client is a subsidiaryof another company, and zero otherwise

UC AB

LISTEDjt Dummy variable equal to one if a client is publiclylisted, and zero otherwise


JOINTAUDITjt Dummy variable equal to one if a client is audited bytwo audit firms, and zero otherwise


FOREIGNSALESjt The proportion of a client’s sales generated by foreignoperations


EXCEPTIONjt Dummy variable equal to one if a client reportsexceptional or extraordinary items, and zerootherwise


PBjt A client’s price-to-book ratio COMPUSTAT GlobalVantage

YEAR_y The nine yearly dummy variables for the years 2000–2008


CLIENT_sj Dummy variables for each client UC AB; COMPUSTATGlobal Vantage

IND_sj Dummy variables for each industry (2-digit SIC) UC AB; COMPUSTATGlobal Vantage

AUDITFIRM_sj Dummy variables for different audit firms. We includedummy variables for each of the Big N firms andcombine non-Big N audit firms into one dummyvariable because of the lack of sufficient number ofobservations for non-Big N firms


CREDITRATE_sj The four dummy variables for the credit ratings fromone to four (five being the reference group) issued fora client by the credit rating agency UC AB

UC AB



Aobdia, D., C. J. Lin, and R. Petacchi. 2015. Capital market consequences of audit partner quality. TheAccounting Review 90 (6): 2143–76.

Balsam, S., J. Krishnan, and J. S. Yang. 2003. Auditor industry specialization and earnings quality. Auditing:A Journal of Practice & Theory 22 (2): 71–97.

Beauchamp, J., D. Cesarini, and M. Johannesson. 2017. The psychometric and empirical properties of mea-sures of risk preferences. Journal of Risk and Uncertainty 54 (3): 203–37.

Becker, C., M. DeFond, J. Jiambalvo, and K. R. Subramanyam. 1998. The effect of audit quality on earningsmanagement. Contemporary Accounting Research 15 (1): 1–24.

Bedard, J. C., and S. Biggs. 1991. The effect of domain-specific experience on evaluation of management rep-resentations in analytical procedures. Auditing: A Journal of Practice & Theory 10 (Supplement): 77–90.

Bedard, J. C., D. R. Deis, M. B. Curtis, and J. G. Jenkins. 2008. Risk monitoring and control in audit firms:A research synthesis. Auditing: A Journal of Practice & Theory 27 (1): 187–218.

Bonner, S. E., and B. L. Lewis. 1990. Determinants of auditor expertise. Journal of Accounting Research 28 (Sup-plement): 1–20.

Borghans, L., A. L. Duckworth, J. J. Heckman, and B. Ter Weel. 2008. The economics and psychology ofpersonality traits. Journal of Human Resources 43 (4): 972–1059.

Burrows, G., and C. Black. 1998. Profit sharing in Australian Big 6 accounting firms: An exploratory study.Accounting, Organizations, & Society 23 (5–6): 517–30.

Cameran, M., D. Campa, and J. R. Francis 2017. How important is partner variation in explaining auditquality? Working paper, European Accounting Association, Valencia, Spain.

Caramanis, C., and C. Lennox. 2008. Audit effort and earnings management. Journal of Accounting andEconomics 45 (1): 116–38.

Carey, P., and R. Simnett. 2006. Audit partner tenure and audit quality. The Accounting Review 81 (3):653–76.

Carlstedt, B. 2000. Cognitive abilities: Aspects of structure, process and measurement. PhD dissertation,Gothenburg University, Gothenburg.

Chen, S., S. Y. Sun, and D. Wu. 2010. Client importance, institutional improvements, and audit quality inChina: An office and individual auditor level analysis. The Accounting Review 85 (1): 127–58.

Chi, H. Y., and C. L. Chin. 2011. Firm versus partner measures of auditor industry expertise and effects onauditor quality. Auditing: A Journal of Practice & Theory 30 (2): 201–29.

Chin, C., and H. Chi. 2009. Reducing restatements with increased industry expertise. ContemporaryAccounting Research 26 (3): 729–65.

Chow, C. W., J. L. Ho, and S. C. Vera-Munõz. 2008. Exploring the extent and determinants of knowledgesharing in audit engagements. Asia-Pacific Journal of Accounting & Economics 15 (2): 141–60.

Chu, J., A. Florou, and P. Pope. 2017. Does accounting education add value in auditing? Evidence from theUK. Working paper, European Accounting Association, Valencia, Spain.

DeAngelo, L. E. 1981. Auditor independence, “low balling,” and disclosure regulation. Journal of Account-ing and Economics 3 (2): 113–27.

Deary, I. J. 2001. Intelligence: A very short introduction. New York: Oxford University Press.Dechow, P., W. Ge, and C. Schrand. 2010. Understanding earnings quality: A review of the proxies, their

determinants and their consequences. Journal of Accounting and Economics 50 (2–3): 344–401.DeFond, M. L., and J. Zhang. 2014. A review of archival auditing research. Journal of Accounting and Eco-

nomics 58 (2–3): 275–326.Dowling, C. 2009. Appropriate audit support system use: The influence of auditor, audit team, and firm fac-

tors. The Accounting Review 84 (3): 771–810.Emby, C., and M. Favere-Marchesi. 2010. Review partners and engagement partners: The interaction process

in engagement quality review. Auditing: A Journal of Practice & Theory 29 (2): 215–32.Epps, K. K., and W. F. Messier. 2007. Engagement quality reviews: A comparison of audit firm practices.

Auditing: A Journal of Practice & Theory 26 (2): 167–81.Fiolleau, K., K. Hoang, K. Jamal, and S. Sunder. 2010. Engaging auditors: Field investigation of a courtship.

Working paper, University of Alberta and Yale University.



Firth, D. 1993. Bias reduction of maximum likelihood estimates. Biometrika 80 (1): 27–38.Francis, J. R. 2004. What do we know about audit quality? British Accounting Review 6 (4): 345–68.Francis, J. R., and P. N. Michas. 2013. The contagion effect of low-quality audits. The Accounting Review

88 (2): 521–52.Francis, J. R., M. L. Pinnuck, and O. Watanabe. 2014. Auditor style and financial statement comparability.

The Accounting Review 89 (2): 605–33.Francis, J. R., and M. D. Yu. 2009. Big 4 office size and audit quality. The Accounting Review 84 (5):

1521–52.Frankel, R. M., M. F. Johnson, and K. K. Nelson. 2002. The relation between auditors’ fees for nonaudit ser-

vices and earnings management. The Accounting Review 77 (Supplement): 71–105.Gibbins, M. 1984. Propositions about the psychology of professional judgment in public accounting. Journal

of Accounting Research 22 (1): 103–25.Goodwin, J., and D. Wu. 2014. Is the effect of industry expertise on audit pricing an office-level or a

partner-level phenomenon? Review of Accounting Studies 19 (4): 1532–78.Griffith, E. E., J. S. Hammersley, and K. Kadous. 2013. Auditing complex estimates: Process, problems, and

preliminary recommendations for improving auditor performance. Working paper, University ofWisconsin—Madison.

Grinblatt, M., S. Ikäheimo, M. Keloharju, and S. Knüpfer. 2015. IQ and mutual fund choice. ManagementScience 62 (4): 924–44.

Grinblatt, M., M. Keloharju, and J. T. Linnainmaa. 2011. IQ and stock market participation. Journal ofFinance 66 (6): 2121–64.

Grinblatt, M., M. Keloharju, and J. T. Linnainmaa. 2012. IQ, trading behavior, and performance. Journal ofFinancial Economics 104 (2): 339–62.

Gul, F. A., M. Ma, and K. M. Lai. 2017. Busy auditors, partner-client tenure and audit quality: Evidencefrom an emerging market. Journal of International Accounting Research 16 (1): 83–105.

Gul, F. A., D. Wu, and Z. Yang. 2013. Do individual auditors affect audit quality? Evidence from archivaldata. The Accounting Review 88 (6): 1993–2023.

Hardies, K., D. Breesch, and J. Branson. 2016. Do fe(male) auditors impair audit quality? Evidence fromgoing-concern opinions. European Accounting Review 25 (1): 7–34.

Hay, D., W. R. Knechel, and N. Wong. 2006. Audit fees: a meta-analysis of the effects of supply anddemand attributes. Contemporary Accounting Research 23 (1): 141–91.

Hogarth, R. M. 1991. A perspective on cognitive research in accounting. Accounting Review 66 (2): 277–90.Huddart, S., and P. Liang. 2003. Accounting in partnerships. American Economic Review 93 (2): 410–4.Hülsheger, U., G. Maier, and T. Stumpp. 2007. Validity of general mental ability for the prediction of job performance

and training success in Germany: A meta-analysis. International Journal of Selection and Assessment15 (1): 3–18.

Hunter, J., and F. Hunter. 1984. Validity and utility of alternative predictors of job performance. Psychologi-cal Bulletin 96 (1): 72–98.

Ittonen, K., K. Johnstone, and E.-R. Myllymäki. 2015. Audit partner public-client specialisation and clientabnormal accruals. European Accounting Review 24 (3): 607–33.

Johnson, V., I. Khurana, and J. K. Reynolds. 2002. Audit firm tenure and the quality of accounting earnings.Contemporary Accounting Research 19 (4): 637–60.

Jones, J. J. 1991. Earnings management during import relief investigation. Journal of Accounting Research29 (2): 193–228.

Keloharju, M., S. Knüpfer, and J. Tåg. 2016. Equal opportunity? Gender gaps in CEO appointments andexecutive pay. Working paper, Harvard Business School.

King, G., and L. Zeng. 2001. Logistic regression in rare events data. Political Analysis 9 (2): 137–63.Knechel, W. R. 2000. Behavioral research in auditing and its impact on audit education. Issues in Account-

ing Education 15 (4): 695–712.Knechel, W. R., L. Niemi, and M. Zerni. 2013. Empirical evidence on the implicit determinants of compen-

sation in Big 4 audit partnerships. Journal of Accounting Research 51 (2): 349–87.



Knechel, W. R., and A. Vanstraelen. 2007. The relationship between auditor tenure and audit quality impliedby going concern opinions. Auditing: A Journal of Practice & Theory 26 (1): 113–31.

Knechel, W. R., A. Vanstraelen, and M. Zerni. 2015. Does the identity of engagement partners matter? Ananalysis of audit partner reporting decisions. Contemporary Accounting Research 32 (4): 1443–78.

Kothari, S. P., A. J. Leone, and C. E. Wasley. 2005. Performance matched discretionary accruals. Journal ofAccounting and Economics 39 (1): 163–97.

Krishnan, G. 2003. Does Big 6 auditor industry expertise constrain earnings management? Accounting Hori-zon (Supplement): 1–16.

Lemon, W. M., K. W. Tatum, and W. S. Turley. 2000. Developments in the audit methodologies of largeaccounting firms. London: ABG Professional Information.

Lennox, C. 1999. Are large auditors more accurate than small auditors? Accounting and Business Research29 (3): 217–27.

Lennox, C., and X. Wu. 2018. A review of the archival literature on audit partners. Accounting Horizons32 (2): 1–35.

Lennox, C., X. Wu, and T. Zhang. 2016. The effect of audit adjustments on earnings quality: Evidence fromChina. Journal of Accounting and Economics 61 (2–3): 545–62.

Levine, R., and Y. Rubinstein. 2017. Smart and illicit: Who becomes an entrepreneur and do they earnmore? Quarterly Journal of Economics 132 (2): 963–1018.

Li, C. 2009. Does client importance affect auditor independence at the office level? Empirical evidence fromgoing-concern opinions. Contemporary Accounting Research 26 (1): 201–30.

Li, L., B. Qi, G. Tian, and G. Zhang. 2017. The contagion effect of low-quality audits at the level of individ-ual auditors. The Accounting Review 92 (1): 137–63.

Libby, R., and J. Luft. 1993. Determinants of judgment performance in accounting settings: Ability, knowl-edge, motivation, and environment. Accounting, Organizations and Society 18 (5): 425–50.

Libby, R., and H.-T. Tan. 1994. Modeling the determinants of audit expertise. Accounting, Organizationsand Society 19 (8): 701–16.

Lindqvist, E., and R. Vestman. 2011. Labor market returns to cognitive and noncognitive ability: Evidencefrom the Swedish enlistment. American Economic Journal: Applied Econometrics 3 (1): 101–28.

Lo, C. 2017. Is there a relationship between high IQ scores and positive life outcomes? A critical review.Psychology 8 (4): 627–35.

McKnight, C. A., and W. F. Wright. 2011. Characteristics of relatively high-performance auditors. Auditing:A Journal of Practice & Theory 30 (1): 191–206.

Moizer, P. 1997. Auditor reputation: The international empirical evidence. International Journal of Auditing1 (1): 61–74.

Murray, C. 1998. Income inequality and IQ. Washington: AEI Press.Murthy, U. S., and D. S. Kerr. 2004. Comparing audit team effectiveness via alternative modes of computer-

mediated communication. Auditing: A Journal of Practice & Theory 23 (1): 141–52.Myers, J. N., L. A. Myers, and T. C. Omer. 2003. Exploring the term of the auditor-client relationship and

the quality of earnings: A case for mandatory auditor rotation? The Accounting Review 78 (3): 779–99.Neisser, U., G. Boodoo, T. J. Bouchard, W. A. Boykin, N. Brody, and S. Urbina. 1996. Intelligence:

Knowns and unknowns. American Psychologist 51 (2): 77–101.Nelson, M., and H.-T. Tan. 2005. Judgment and decision making research auditing: A task, person, and

interpersonal interaction perspective. Auditing: A Journal of Practice & Theory 24 (s-1): 41–71.Peytcheva, M., and P. R. Gillett. 2011. How partners’ views influence auditor judgment. Auditing: A Journal

of Practice & Theory 30 (4): 285–301.Ramsay, R. J. 1994. Senior/manager differences in audit workpaper review performance. Journal of

Accounting Research 32 (1): 127–35.Reichelt, K. J., and D. Wang. 2010. National and office-specific measures of auditor industry expertise and

effects on audit quality. Journal of Accounting Research 48 (3): 647–86.Rich, J. S., I. Solomon, and K. T. Trotman. 1997. Multi-auditor judgment/decision making research:

A decade later. Journal of Accounting Literature 16: 86–126.



Salgado, J. F., N. Anderson, S. Moscoso, C. Bertua, and F. De Fruyt. 2003. International validity generalization ofGMA and cognitive abilities: A European community meta-analysis. Personnel Psychology 56 (3): 573–605.

Shumway, T. 2001. Forecasting bankruptcy more accurately: A simple hazard model. Journal of Business74 (1): 101–24.

Simnett, R., E. Carson, and A. Vanstraelen. 2016. International archival auditing and assurance research: Trends,methodological issues and opportunities. Auditing: A Journal of Practice & Theory 35 (3): 1–32.

Simunic, D. A. 1980. The pricing of audit services: Theory and evidence. Journal of Accounting Research18 (1): 161–90.

Spearman, C. 1904. General intelligence, objectively determined and measured. American Journal of Psy-chology 15 (2): 201–93.

Sundgren, S., and T. Svanström. 2014. Auditor-in-charge characteristics and going concern reporting.Contemporary Accounting Research 31 (2): 531–50.

Tan, H.-T., and R. Libby. 1997. Tacit managerial versus technical knowledge as determinants of audit exper-tise in the field. Journal of Accounting Research 35 (1): 97–113.

Taylor, S. D. 2011. Does audit fee homogeneity exist? Premiums and discounts attributable to individualpartners. Auditing: A Journal of Practice & Theory 30 (4): 249–72.

Vera-Munõz, S. C., J. L. Ho, and C. W. Chow. 2006. Enhancing knowledge sharing in public accountingfirms. Accounting Horizons 20 (2): 133–55.

Wilks, T. F. 2002. Predecisional distortion of evidence as a consequence of real-time audit review. TheAccounting Review 77 (1): 51–71.

Ye, P., E. Carson, and R. Simnett. 2011. Threats to auditor independence: The impact of relationship andeconomic bonds. Auditing: A Journal of Practice & Theory 30 (1): 121–48.

Zerni, M. 2012. Audit partner specialization and audit fees: Some evidence from Sweden. ContemporaryAccounting Research 29 (1): 312–40.

Zmijewski, M. E. 1984. Methodological issues related to the estimation of financial distress predictionmodels. Journal of Accounting Research 22 (Supplement): 59–82.

Why Shouldn’t Higher-IQ Audit Partners Deliver Better Audits?A Discussion of “IQ and Audit Quality: Do Smarter Auditors

Deliver Better Audits?”*

DANIEL AOBDIA, Northwestern University†

1. Introduction

Whether individual audit partners matter in audit firms has become a topic of extensive academicresearch (see Lennox and Wu 2018 for a review of the literature). This research is timely given thatseveral regulators, including the PCAOB in the United States, recently mandated disclosure of the

*Accepted by Clive Lennox. An earlier version of this discussion was presented at the 2017 Contemporary AccountingResearch Conference, generously supported by the Chartered Professional Accountants of Canada. I am thankful to CliveLennox, Juha-Pekka Kallunki, Robert Magee, Lasse Niemi, Nemit Shroff, Beverly Walther, and the 2017 ContemporaryAccounting Research conference participants for helpful comments and suggestions. I was a Senior Economic ResearchFellow at the Center of Economic Analysis of the PCAOB between September 2014 and September 2016. This discus-sion paper was written after my affiliation with the PCAOB ended. The views expressed in this paper are mine and donot necessarily reflect the views of the Board, individual Board members, or staff of the PCAOB.

†Corresponding author.



name of the engagement partner, beginning with audit reports issued after January 31, 2017. Sev-eral papers find evidence consistent with individual engagement partners’ influencing audit quality(Gul et al. 2013; Aobdia et al. 2015; Knechel et al. 2015; Cameran et al. 2017). Most of thesepapers use a fixed effects design derived from Bertrand and Schoar (2003), which cannot answerwhy and how a partner’s attributes have an influence on audit quality. Gul et al. (2013, table 6) findthat partners’ observable characteristics, such as education level, gender, and age, explain only upto three percent of individual partner fixed effects. Given the interest of audit regulators and aca-demics to understand the determinants of audit quality, more research on individual partners’ char-acteristics is needed to determine which of them significantly influence audit outcomes.1

Prior literature focuses on several personal characteristics of individual audit partners, suchas experience or expertise, age, education, and gender. Kallunki, Kallunki, Niemi, and Nilsson(2019, hereafter KKNN) contribute to this literature by using a unique data set of Swedish malepartners’ cognitive abilities (i.e., intelligence quotient, or IQ). Their evidence is important becauseit is intuitive that the IQ of partners should have an influence on their job performance, a part ofwhich is measured using audit quality. However, whether partner differences in IQ explain asmall or large part of inter-partner variation in audit quality remains an empirical question. Over-all, my interpretation of KKNN’s evidence is that the economic significance of the role of partnerIQ on audit quality appears more limited than priors might imply. However, higher-IQ partnersmight be better at selling audit and nonaudit services to their clients.

KKNN identify a setting where they have a measure of all Swedish male partners’ IQs. TheSwedish National Service Administration measures IQ during the equivalent of the Armed ForcesQualifications Test in the United States. Importantly, all male Swedish citizens were obliged by lawto take this test around the age of 18, because military service in Sweden was compulsory until2010. Thus, with the exception of female and non-Swedish partners, the data in KKNN, based onthe IQ of 407 audit partners, are comprehensive and do not rely on voluntary participation, thus mit-igating selection bias concerns.2 KKNN identify significant variation in IQ across audit partners.Fourteen percent of the partners have an IQ that is at the average or below the average of the entireSwedish population. The largest proportion of partners, 33 percent, has a measured IQ between the77th and 89th percentile of the theoretical average IQ distribution. This sweet spot in terms ofpartners’ most prevalent abilities might be helpful for accounting programs around the world to tar-get prospective students who are more likely to remain in public accounting in the future.3

KKNN focus on three different dependent variables, each with a different sample. In two largesamples mostly composed of privately held companies, they find a negative association between part-ner IQ and the propensity to make mistakes in issuing going-concern opinions, as well as weak evi-dence of a negative association between partner IQ and income-increasing abnormal accruals. Thelatter association is present only when KKNN do not control for time-invariant client characteristics inthe form of company fixed effects. In a smaller sample of publicly listed companies, they find a posi-tive association between partner IQ and audit fees. Overall, KKNN conclude, with some caveats, thatengagement partners and their cognitive abilities matter in conducting high-quality audits.

Although KKNN is a welcome and intuitive first step in the archival literature toward under-standing whether IQ matters from an audit quality standpoint, their evidence is only a first step.

1. Consistent with this view, DeFond and Zhang (2014, 304) call for more research about “individual auditor charac-teristics, such as professional skepticism, personality traits, gender, the complex audit team interactions, and thesocio-economic characteristics.”

2. Selection bias concerns could still be present given that IQ might influence the likelihood of becoming an auditpartner in the first place.

3. As a caveat, this assertion makes the assumption that audit firms need their lower-level staff members to have simi-lar IQs to their engagement partners. Higher IQs could be more (less) desirable for lower-level staff, but less (more)so for engagement partners. In general, more research on the cognitive abilities of lower-level audit firm staff mem-bers could be helpful to understand the long-term needs of the accounting profession.

Discussion of “IQ and Audit Quality” 1405


The discussion presented here emphasizes how this line of research can be extended in the future.Perhaps due to size restrictions in some of their samples, KKNN do not explore under which cir-cumstances partner IQ is more likely to matter. However, the role of a partner’s cognitive abilitiesis likely to differ depending on audit firm- and client-specific circumstances—for example, if thepartner works for a larger audit firm or works for a larger client with an extensive engagementteam and has less involvement in daily audit activities. The role of a partner’s IQ also likelychanges depending on the partner’s objective function, which is not necessarily to reach the high-est audit quality possible, but perhaps to keep clients satisfied, increase audit and nonaudit fees,and attract new clients while maintaining an adequate level of audit quality. IQ is also likely toinfluence a partner’s career decisions, such as acquiring knowledge and expertise, thus suggestinga relation between a partner’s IQ and expertise characteristics, such as industry specialization.

KKNN’s evidence mostly applies to the audits of very small private firms in Sweden. In theirfirst sample of distressed companies, the average number of companies per partner, 119, is veryhigh, and implies that a partner signs an audit every two days. Partners in these companies mightrely less on large engagement teams, so this setting might be more powerful to find results. Undersuch circumstances, whether all of KKNN’s results can be generalized to publicly traded compa-nies outside of Sweden is unclear. An audit of a publicly traded company is likely to involvemany more audit firm resources and audit firm support, such as larger engagement teams, specificaudit methodologies and the use of specialists, especially for the largest audit firms. Thus, an indi-vidual partner’s IQ might matter less in such a large structure. Given that the work of an auditpartner represents only a small portion of an audit in the typical setting of publicly traded compa-nies, future research also needs to focus on other elements that could influence audit quality. Forexample, the overall quality control systems of an audit firm, such as a firm’s methodology, andthe characteristics and expertise of lower-level engagement team members might matter as muchas the characteristics of the engagement partner.

The remainder of this discussion proceeds as follows. First, I discuss prior research and howKKNN fits into the literature and complements the findings of prior studies. Second, I discuss theempirics of KKNN. Third, I present concluding remarks and offer future research opportunities.

2. How does KKNN fit in prior literature?

IQ and job performance proxied by audit quality

A long strand of literature, beginning in industrial psychology, provides experimental evidencethat IQ predicts job performance (e.g., Hunter 1986; Richardson and Norgate 2015). This predic-tive power derives from several channels. A higher-IQ individual is able to acquire more relevantknowledge and to use this knowledge more effectively when facing specific problem-solvingtasks (e.g., Hunter 1986; Bonner and Lewis 1990). Consistent with these ideas, prior experimentalliterature finds positive correlations between IQ, job knowledge, and job performance.4 However,the magnitude of these correlations is still subject to debate. Several studies find that the correla-tions are very high, whereas more recent studies argue that the correlations could be more modest(e.g., Richardson and Norgate 2015).

The evidence of the role of IQ on performance is more limited in archival settings, due to thelack of reliable IQ data. One exception is a series of papers in the finance literature that use com-prehensive IQ and stock trading data for males in Finland. The main findings are that higher-IQindividuals are more likely to participate in the stock market (Grinblatt et al. 2011). Further,higher-IQ investors exhibit superior market timing, stock-picking skills, and trade execution andare less subject to behavioral biases (Grinblatt et al. 2012). The underlying premise in these

4. As discussed in KKNN, the experimental audit literature also finds positive associations between IQ and job performancein the auditing context (e.g., Bonner and Lewis 1990; Libby and Tan 1994).



studies is that individuals with higher cognitive abilities are better able to process complex infor-mation and make better decisions as a result.

In light of the cited literature, the priors that IQ influences audit quality should be reasonablyhigh. More than a question of direction, the authors perhaps should have stressed the question ofmagnitude in the paper. In particular, is IQ an important determinant of audit quality, perhaps themain one, or just one factor among many? This question cannot be answered based on the empiri-cal analyses and tables in the paper. For example, not known is by how much the explanatorypower of the regressions increases when IQ is included as an explanatory variable, and correla-tions between IQ and audit quality are not reported. Some insights on this issue can be obtainedfrom several of the empirical analyses. Based on table 3, the average IQ of partners who make anerror in the issuance of a going-concern opinion is 6.69 out of 9, compared with an average IQ of6.79 out of 9 for partners who do not make an error. While the difference is statistically signifi-cant, the economic significance of this result appears reasonably low given that the potentialrange of partner IQ goes from one to nine and that significant variation exists in the data, at leastfor IQs between four and nine.5 In contrast, based on column (2) of table 6, an increase of partnerIQ of one (less than one standard deviation) is associated with a reasonably large increase in auditfees of 6.6 percent.6 For discretionary accruals, the evidence is generally insignificant. Focusingon income-increasing discretionary accruals for the full sample in table 7, the difference in partnerIQ between the lowest and highest quintiles of discretionary accruals is only 0.04, despite beingmarginally statistically significant. So, with the exception of audit fees, which might be influencedby a partner’s abilities to sell rather than increasing actual audit quality, the influence of IQ onaudit quality appears to be small. This result is potentially surprising in light of initial priors andmight have deserved further examination in the paper.7

Analytical studies of moral hazard and audit quality also implicitly point to a link between apartner’s abilities and audit quality. A partner’s IQ could directly affect their incentives if the cost ofproviding effort differs based on an individual’s innate characteristics. Audit firms are typically orga-nized as partnerships, and partners make decentralized decisions in performing various tasks (Liu andSimunic 2005). The analytical literature highlights incentives at the audit firm level to conduct high-quality audits, because of reputation and litigation costs (e.g., DeAngelo 1981; Dye 1993). However,individual audit partners have incentives to shirk when their effort is unobservable (Holmstrom 1982;Balachandran and Ramakrishnan 1987). The main tension in these studies is that providing effort iscostly. The analytical literature, with a few exceptions, such as Huddart and Liang (2003), generallyassumes that the cost of effort is the same for all individuals in the partnership. In practice, the cost ofproviding effort is likely to differ based on an individual’s innate characteristics. The Appendix pro-vides a simple analytical framework based on this intuition, adapted from Holmstrom (1982) andHuddart and Liang (2003). In the model, N individuals form a partnership and equally share the pro-ceeds, which are the sum of audit fees less expected litigation costs, which decrease as each partner’sindividual effort increases.8 Providing effort is costly, and the cost of effort differs based on anindividual’s IQ. The main finding is that, assuming that the cost of providing effort is lower when theperson has higher ability, individual effort and output will be higher for individuals with higher IQ, inequilibrium. The incremental change in effort between low- and high-IQ individuals will be lower

5. The economic significance appears larger for the marginal effects in the multivariate analyses of table 4, especiallyfor Type 2 errors.

6. This number is computed as e0.064−1.7. Because the analysis on audit fees is restricted to publicly traded corporations, it might not be directly comparable

to the other analyses that are conducted on much larger samples mostly composed of private companies. Ideally,KKNN should also have reported the results of their other analyses on the sample of publicly traded corporationsfor comparability.

8. The allocation rule is based on the proceeds, because effort cannot be contracted upon. For simplicity I assume asymmetric allocation rule that implies equal sharing.



when N, the number of partners in the partnership, increases.9 Thus, the role of IQ on audit perfor-mance is likely to depend on the partnership’s size. While plausible, the assumption that individualswith higher IQ have a lower cost of providing effort is not necessarily true, especially if higher-IQindividuals are bored more easily by auditing tasks. Under this scenario, the comparative staticswould reverse. Some non-linearities could also exist in the relationship between IQ and the cost ofproviding effort and, therefore, audit quality. Further, consistent with Dye (1993), an audit firm maybe required, to avoid litigation costs, to provide a level of assurance that meets only current auditingstandards, in which case, in equilibrium, higher-ability partners are likely to exercise lower effort thanlower-ability partners do, so that they reach the same level of assurance. Under this possibility, littleto no relation between IQ and audit quality would be expected.10

A potential explanation for weak effects of IQ on audit quality

Despite strong priors that IQ should be positively associated with audit quality, high performancein auditing might not be the same as providing high-quality audits, as hinted at in the analyticaldiscussion above. Individual auditors have incentives to conduct more profitable audits, growtheir clientele, and retain valuable clients. These objectives could serve as alternative measures ofperformance and provide individual auditors incentives to neglect audit quality or lower theirindependence (Reynolds and Francis 2000). Higher-IQ auditors would be better at meeting suchobjectives, in which case the relationship between IQ and audit quality could become insignifi-cant if higher-IQ partners strive to reach only an adequate level of audit quality (Dye 1993), oreven turn opposite if they are willing to compromise their independence more often to retainimportant clients.

Overall, whether a partner’s IQ is associated with audit quality likely depends on thepartner’s objective function. Knechel et al. (2013), using the same Swedish setting as KKNN, findthat a Big 4 partner’s compensation is tied to both rainmaking and technical abilities (Huddart2013), which suggests that higher-IQ partners might focus more on selling audit or nonaudit ser-vices instead of providing higher audit quality services. Consistent with this idea, in an earlierversion of their paper, KKNN report a positive association between partner IQ and nonaudit fees,which might stem from higher-IQ auditors being better at selling services to their clients, insteadof just providing higher-quality audits.11 The difference in economic magnitude between theresults on audit fees and other measures of audit quality (going-concern opinions and discretion-ary accruals) also suggests that higher-IQ partners might charge more to their clients due to fac-tors other than just higher audit quality. More focus on partner’s rainmaking abilities might havebeen helpful.

Also potentially worthwhile would have been for KKNN to explore differences in the rela-tionship between IQ and audit quality for each of the Big 4 firms. Each of these firms appears tohave adopted, at least in Sweden, a different compensation structure, with some firms more ori-ented toward revenue generation opportunities and others more toward audit quality (Knechel

9. A natural question is why higher-ability partners would consider forming a partnership with lower-ability partners.It is relevant even if partners are unable to observe the other partners’ innate abilities, because they still have toshare the profits of the partnership with them. Prior literature models this possibility using two elements:(i) risk-averse partners do not like facing a risky output, in which case the reduction in risk by forming a partnershipcan compensate for the cost of sharing the partnership profits with lower-ability partners (e.g., Huddart and Liang2003) and (ii) synergies increase in larger partnerships (Huddart and Liang 2005), in which case the synergies cancompensate for the cost of sharing the partnership profits with lower-ability partners. Adding these features to themodel would not change the main result that individual effort and output will be higher for individuals with ahigher IQ, but doing so may alter the result on incremental change in effort between low- and high-IQ partnersbased on the size of the partnership.

10. Under this scenario, higher-IQ partners might try to increase the number of clients compared with lower-IQpartners.

11. This association might also be consistent with higher-IQ partners providing higher-quality nonaudit services. Disen-tangling these explanations is not possible in the absence of measures of audit and nonaudit services profitability.



et al. 2013), implying a different definition of high performance depending on the firm. Thus, par-titioning the sample by audit firm might have helped determine the role of IQ on audit qualityand audit fees, depending on a partner’s objective function.

Potential moderating effects of the relation between IQ and audit quality

The role of a partner’s IQ on audit quality is also likely to differ depending on the audit firm andthe engagement considered. Larger audit firms provide more resources and support to theirengagement partners to conduct higher-quality audits, such as extensive methodologies, comput-erized audit systems, and specialist services. Their quality control systems are also likely to bemore extensive, implying greater monitoring of their partners (e.g., Huddart and Liang 2005).Industry expert auditors are also likely to have more resources available. Thus, the role of an indi-vidual audit partner’s IQ on audit quality could be lower in larger and industry-specialist firms.Similarly, engagement teams are likely to be larger on bigger clients, in which case the role of aspecific individual, even if it is the engagement partner, is probably more diluted.

Thus, conducting several cross-sectional tests based on auditor size, industry specialization,client size, a partner’s rainmaking abilities, prior knowledge and industry specialization, and otherauditor, client, and partner characteristics would have been instructive for KKNN. The relation-ship between IQ and audit quality is likely to vary with auditor size, industry specialization, andclient size. All of these might dilute a partner’s influence in an audit. Finding results consistentwith this idea could have strengthened the study. Results not consistent with this idea would alsohave helped to further understanding of the role of audit firms’ quality control systems and teamsand their influence on audit quality.

The authors could also have conducted several analyses to examine the relation between IQand the acquisition of knowledge, such as a partner’s decision to become an industry specialist.Existing research finds that higher-IQ individuals more effectively acquire and apply prior knowl-edge (Hunter 1986; Bonner and Lewis 1990). Thus, the role of IQ could be enhanced by apartner’s industry or client expertise or seniority.

KKNN could also have explored non-linearities in the relationship between IQ and auditquality, as partners in the highest IQ categories might be more eager to leave public accountingand be courted by their clients more often.

3. Empirical setting, results, and limitations of KKNN

Specificities of the Swedish setting

KKNN take advantage of the Swedish setting, which gives researchers access to unique data toanswer several interesting research questions in auditing and other topics. As such, the paper isrelated to Knechel et al. (2013), who provide evidence on partners’ salaries at the Big 4, and Kne-chel et al. (2015), who focus on the identity of the engagement partner, mostly for private firms.All three papers use the same Swedish setting.

Despite the data advantage, several particularities of this setting might limit the generalizabil-ity of the findings to publicly traded companies outside of Sweden.12 First, the number of pub-licly traded corporations is very limited. Thus, two out of KKNN’s three samples mostly focuson small private firms, whose audits appear to be very different from the audits of publicly tradedcorporations (Kinney 2015). This difference shows in the number of engagements audited by thepartners in Sweden. According to KKNN’s table 2, the average number of engagements for eachpartner for each year ranges from 28 for the publicly traded corporations sample to 119 for the(mostly) private firms sample. In other words, an audit partner in Sweden signs an audit opinion

12. Many European countries have similar audit requirements to Sweden’s for small private companies and limitednumbers of publicly traded corporations (see, e.g., Minnis and Shroff 2017 for more details). Thus, the results inKKNN may apply to these countries.



every two to nine business days, a frequency that seems much higher than in many other coun-tries. This suggests that either audits are plain vanilla and are conducted by one person (the engage-ment partner) or they are conducted by teams and partners who just sign off on them withoutspending any significant time. Alternatively, the sample might include numerous subsidiaries thatare part of the same audit of a larger corporation (based on table 2, 39 percent of the client-yearobservations in the going-concern sample are for subsidiaries). Given that the audit of a subsidiaryis unlikely to be independent from the audit of the entire corporation, the authors could have con-sidered a different unit of analysis—that is, the parent company. Even the number of publicly tradedcorporations audited by each partner seems high. Based on their panel B of table 1, the averagenumber of public clients of a partner ranges from 2.6 to 6.6 depending on a partner’s IQ (the totalnumber of audits ranges from 62.8 to 86.8). As a comparison point, the average number of publiccorporation engagements in the United States reported in Aobdia and Petacchi (2017) is approxi-mately 2.

Second, audit regulations, such as those regarding auditor independence requirements, appearmore limited in Sweden than in other countries, such as China, Taiwan, and the United States.13

Engagement partner rotation is not mandatory in Sweden. As a result, it appears that most clientskeep the same partner during the entire 2000–2009 sample period. Out of 105,169 company-years, representing 26,899 companies, KKNN are able to identify only 11,427 company-years forfirms that experience a partner rotation (including non-rotation years). This sample reductionmight explain why the results on income-increasing accruals become insignificant when theauthors restrict the analysis to this subsample and incorporate company fixed effects. This specifi-cation could have provided stronger evidence of the influence of partner IQ on audit quality bycontrolling for time-invariant company-specific characteristics. Further, audit firms are allowed tomarket a much broader range of nonaudit services to their audit clients than in other countries,which could increase incentives for partners to cross-sell nonaudit services to their clients andcould explain the positive associations between partner IQ and audit fees in the paper, as well aswith nonaudit fees in an earlier version of the paper.

Distribution of IQ scores

A noteworthy finding in KKNN is that significant variation exists in IQ scores among male auditpartners in Sweden. KKNN report that the average IQ score for their partners is 6.8, comparedwith the average of 5 for the theoretical distribution of the Swedish population. KKNN do notconduct a comparison of partners’ IQ distribution with the theoretical distribution of the Swedishpopulation. Figure 1 does so.

Partners in Sweden generally have a higher IQ than the average population.14 While 40 per-cent of the theoretical distribution should have an IQ score between one and four, only three per-cent of the partners have IQs in this range. While some lower-IQ individuals become partners,even at the Big N audit firms, these individuals represent the exception rather than the rule.15

13. These limited requirements could be due to the fact that most companies in Sweden are non-public. Whether non-public companies should have their disclosures and audits regulated is unclear (Minnis and Shroff 2017).

14. Based on Figure 1, the distribution of partners’ IQs has first order stochastic dominance over the distribution of theaverage Swedish population.

15. An individual’s IQ could be measured with error, which could explain why some audit partners showed lowIQ. KKNN indicate that if the army test result seemed abnormally low, the individual had to retake the test. But thiscould create upward bias in the overall IQ distribution in case the original errors are random and only errors towardthe lower end are corrected. Further, even though individuals could not avoid military service in Sweden because oflow IQ, IQ is still a predictor of the type of service to which conscripts were enlisted (Lindqvist and Vestman2011). Thus, some individuals may have tried to game the test to enlist in their preferred service (which could havehappened if an individual had geographical preferences, for example). Another concern is that IQ, measured herefor the partners when they were approximately 18 years old, does not necessarily remain stable over time. Recentresearch based on taxi drivers in London, England, shows that the brain can change based on the type of work per-formed (Woollett and Maguire 2011).



Overall, IQs of seven, eight, and nine are overrepresented for engagement partners in Swedencompared with the overall population, but the partner IQ distribution peaks at seven, with 33 per-cent of partners in this category, most likely because the total population with an IQ of eight ornine is limited. This suggests that a sweet spot exists in terms of audit partners recruiting, atseven, which corresponds to an IQ range between the 77th and 89th percentiles of the theoreticalIQ distribution.

A useful comparison would be to determine whether academic accounting programs recruit, onaverage, students with an IQ of seven, given that this category eventually becomes overrepresentedin the distribution of engagement partners. The issue is important given concerns repeatedly raisedover time that the United States has a shortage of accounting professionals (e.g., Gonzalez 2017;Monga 2017). A mismatch between the type of students who enroll in accounting programs andindividuals who stay in the accounting profession could have long-term consequences in the supplyof accounting professionals and can contribute to the shortage, in case mismatched students aremore likely to leave the accounting profession. While comparing a partner’s typical IQ and the typi-cal accounting student IQ is not feasible, partly because audit firms might require different types ofIQ at different seniority levels, and Sweden is likely different from other countries, measurable testscores, such as the General Record Examination and the Graduate Management Admission Test(GMAT), can be studied, assuming that such test outcomes are a proxy for cognitive ability and thatthe distributions of these tests are representative of the general population. The University of Texasat Austin’s master of professional accounting program enrolled students between 2015 and 2017with average GMAT scores of 660, which corresponds to the 78th percentile of the GMAT scoresdistribution.16 This number is in line with the distribution percentiles implied by an IQ of 7. In

Figure 1 Swedish partners and theoretical population IQ distributions

Notes: This figure presents the IQ distributions for male partners in Sweden and the theoretical average population.The theoretical average population distribution is based on the Stanine distribution, which discretely partitions thenormal distribution into nine intervals, with a mean of five (see Grinblatt et al. 2011 for more details).

16. See https://www.mccombs.utexas.edu/~/media/Files/MSB/MPA%20Responsive/Fact-Sheet.pdf and https://www.mba.com/us/the-gmat-exam/gmat-exam-scores/your-score-report/what-percentile-rankings-mean.aspx.



https://www.mccombs.utexas.edu/~/media/Files/MSB/MPA%20Responsive/Fact-Sheet.pdf

https://www.mba.com/us/the-gmat-exam/gmat-exam-scores/your-score-report/what-percentile-rankings-mean.aspx

https://www.mba.com/us/the-gmat-exam/gmat-exam-scores/your-score-report/what-percentile-rankings-mean.aspx

general, more research on the cognitive abilities of lower-level audit firm staff members could behelpful to understand the long-term needs of the accounting profession.

Empirical analyses in KKNN

KKNN use three measures of audit quality. The first one is based on errors in the issuance ofgoing-concern opinions; the second, on audit fees; and the third, on discretionary accruals. Whilethese three measures have been extensively used in prior literature and have the potential to mea-sure audit quality (e.g., DeFond and Zhang 2014), they are not devoid of limitations. The decisionto issue a going-concern opinion represents only a limited portion of the audit, which implies thatthe decision to issue such an opinion could be of high quality but not the remainder of the audit,or vice versa (see, e.g., Aobdia 2019; Bowler 2017). Further, the number of going-concern opin-ions and bankruptcies appears relatively limited in KKNN’s sample. A going-concern opinionalso is issued when substantial doubt exists that the entity will be able to continue as a going con-cern, which does not imply that the company will always go bankrupt. As such, the definitionautomatically implies that some Type 1 errors will arise, which is one of the measures of auditquality used in the analysis. Nevertheless, an advantage of looking at issuance of going-concernopinions is that the partner, not lower-level audit team members, usually makes the decision.Alternative measures of audit quality are generally more representative of an entire audit team’seffort.

Audit fees can proxy for audit effort and quality. Consistent with this view, prior studies find ahigh correlation between audit hours and audit fees, as well as a positive association between audithours, audit fees, and financial reporting quality (e.g., Caramanis and Lennox 2008; Lobo and Zhao2013; Aobdia 2019). However, audit fees could also measure a partner’s ability to sell audit ser-vices. In a prior version of the paper presented at the Contemporary Accounting Research confer-ence, the authors report that both audit and nonaudit fees have similar associations with partnerIQ. This could be consistent with higher-IQ partners being able to sell more services.

Finally, accruals might be a noisy measure of audit quality (e.g., DeFond and Zhang 2014;Aobdia 2019). Interpreting the results with an accrual measure of audit quality thus could be diffi-cult, especially given that the results become insignificant when KKNN focus on partner switchesand incorporate client fixed effects in their regressions. Ideally, KKNN should have also used ameasure of restatements as a measure of audit quality, but whether an available measure exists inSweden in the context of auditing of (mostly) private firms is unclear.

Similar to many audit partner studies, KKNN’s analyses are unable to disentangle higher-IQpartners’ choosing their clients better from conducting better audit work. This is because partnersare not randomly assigned to their clients. The task is particularly difficult in Sweden given thatthe number of partner rotations appears limited, due to the absence of mandatory rotation require-ments. Selection issues may explain why some of the coefficients on the control variables appearcounterintuitive. For example, in the reduced sample, industry-specialist partners issue moregoing-concern opinions to clients who eventually do not declare bankruptcy. Partners suspectedof a crime charge an 18 percent premium in audit fees, but industry-specialist partners charge a10 percent discount. The effect of industry-specialist partners also reverses in the analysis of dis-cretionary accruals when client fixed effects are included in the regressions.

Finally, partner IQ could influence several partner-level characteristics. The decisions to tar-get and audit specific types of clients, such as public companies, acquire knowledge and expertisein specific areas, and engage in a crime could be related to a partner’s IQ. Consistent with thisview, prior research finds that higher-IQ individuals are able to acquire more relevant knowledgethat they apply more effectively, which suggests that the decision to specialize in a particularindustry is linked to a partner’s IQ (Hunter 1986; Bonner and Lewis 1990). The authors couldhave applied path and mediation analysis to determine the overall role of IQ on audit quality, aswell as direct and indirect roles depending on the decision to acquire knowledge, specialize, andservice specific types of clients.



4. Concluding remarks and opportunities for future research

KKNN provide intuitive archival evidence about whether a partner’s IQ influences audit quality.KKNN’s research fits into a growing literature on audit partners (see Lennox and Wu 2018 for areview), which strives to answer whether individuals matter in audit firms. Auditing is a knowledge-intensive industry that requires highly qualified people. While human capital is one of the mostimportant assets in the auditing industry, “the fact remains that we know very little about the peoplewho conduct audits” (Francis 2011, 134). Further, even though audit engagement teams are reason-ably large, the engagement partner could have a disproportionate impact on the quality of the workcompleted. PCAOB standard AS 1201 states that “[t]he engagement partner is responsible for theengagement and its performance. Accordingly, the engagement partner is responsible for propersupervision of the work of engagement team members and compliance with PCAOB standards.”

While KKNN is an important first step in understanding in an archival setting the role of anauditor’s cognitive abilities on audit quality, their evidence is only a first step. Ideally, KKNN’sanalyses should be extended to different countries, larger samples of publicly traded corporations,and different measures of audit quality or job performance, such as the retention of existing cli-ents or acquisition of new clients. This would allow for more specific tests to understand underwhich conditions a partner’s IQ matters the most. One interesting area to pursue is on differencesin the role of engagement partners depending on the audit regulation environment. An engage-ment partner’s objective function might differ depending on the existence of a regulator and onthe regulator’s activities, if such a regulator exists. Public disclosure of a partner’s name is alsolikely to influence the role of an engagement partner due to reputation concerns.

Future research may also have the opportunity to explore the role of other individuals inaudit firms below the engagement partner. Even though individual engagement partners appear toinfluence audit quality, audit teams are often very large, and partner hours are generally small incomparison with total engagement hours (Aobdia 2018). This might explain why the economicsignificance in KKNN sometimes appears limited and suggests that exploring the role of otherindividuals in audit teams, including their expertise and work, would be fruitful. The main issueis data availability. While the name of the engagement partner has been made public in manyjurisdictions, much less information is available about lower-level team members. The availabilityof non-public data, perhaps through the PCAOB fellowship program, might help bridge this gapin the literature. Along the lines of this paper, an intriguing angle would be to focus on the cogni-tive abilities of lower-level team members, to determine whether the attributes that make a partnersuccessful and able to improve audit quality are the same for lower-level team members. Thismight help audit firms refine their recruiting and personnel retention strategies and accountingprograms improve their approaches to targeting prospective students.

Further, for the largest auditors, significant resources, such as firm methodologies, quality con-trol systems, and access to firm specialists, are available to engagement teams. These resources havethe potential to compensate for an underperforming engagement partner. Consistent with this idea,in 2017 the PCAOB fined Grant Thornton $1.5 million in an enforcement action for “assigning twopartners from its Philadelphia office, with known audit quality concerns, to serve as engagementpartners on two separate fiscal year end 2013 issuer audits, without providing them sufficient sup-port or monitoring” (emphasis added, PCAOB 2017). Understanding the role of these resources onthe work of engagement teams and how they can enable or limit audit quality would be helpful todetermine what matters in audits. Another issue of interest is the matching between engagementpartners and individual clients, given that, as mentioned in the Grant Thornton enforcement action,the assignment of a partner with known quality issues contributed to several misstatements.

Appendix

This Appendix presents a simple analytical framework of moral hazard in a partnership, based onHolmstrom (1982) and Huddart and Liang (2003). The aim is to understand how the provision of



effort differs based on a partner’s IQ. The model purposely excludes complexities such as why agroup of partners of unequal characteristics would decide to form a partnership. This can be mod-eled by introducing additional elements that would not change the basic intuition of the firstresult, such as a stochastic output, risk aversion, and synergistic elements in the partnerships (seeHuddart and Liang 2003, 2005 for more details).17

N individuals form a partnership and equally share the proceeds, which are the sum of auditfees less expected litigation costs. Each partner’s individual effort is not contractible. Only thetotal proceeds x are. I consider a symmetric equilibrium which implies that x is equally shared

between the partners. Formally, the total output x is equal to x=Pi=N

i= 0 F− Pxi

� �, where F is the

audit fee charged to each client, assumed to be the same, and P/xi are the expected litigationcosts. P is a constant, and xi is the individual effort of each partner i. Thus, expected litigationcosts are a decreasing function of an individual partner’s effort.

The cost of providing effort differs based on the type of partner, which can be high IQ (H) orlow IQ (L). Formally, the cost of effort when partner i exercises effort xi is given by C xið Þ= a:θk:x2i ,where k is the type of partner (H or L) and a is a constant.

In equilibrium, partner i adjusts their effort xi to maximize utility net of effort. Thus, the part-

ner optimizes maxxi1N

Pj=Nj= 0 F− P

xj

� �−a �θk � x2i .

Taking the first order condition, partner i’s effort in equilibrium is equal to xi = P2Naθk

� �13. This

result is comparable to Holmstrom (1982) and indicates that when the partnership size increases,a partner will not provide as much effort as if they were working alone. This is because all theother partners reap the rewards of a specific partner’s effort, whereas this partner bears the entirecost of their own effort.

This equilibrium effort yields two observations, relevant for this discussion.

1. If θL > θH, which indicates that, for a given effort, the cost of providing this effort islarger for a lower-IQ partner than a higher-IQ partner, then in equilibrium the higher-IQpartner will provide more effort than the lower-IQ partner.

2. The difference in effort provision between a high- and low-IQ partner decreases as thepartnership size, N, increases.

References

Aobdia, D. 2018. The impact of the PCAOB individual engagement inspection process—Preliminary evi-dence. The Accounting Review 93 (4): 53–80.

Aobdia, D. 2019. Do practitioner assessments agree with academic proxies for audit quality? Evidence fromPCAOB and internal inspections. Journal of Accounting and Economics 67 (1): 144–74.

Aobdia, D., and R. Petacchi. 2017. Consequences of low-quality audits for engagement partners. Workingpaper, PCAOB.

Aobdia, D., C.-J. Lin, and R. Petacchi. 2015. Capital market consequences of audit partner quality. TheAccounting Review 90 (6): 2143–76.

Balachandran, B. V., and R. T. S. Ramakrishnan. 1987. A theory of audit partnerships: Audit firm size andfees. Journal of Accounting Research 25 (1): 111–26.

Bertrand, M., and A. Schoar. 2003. Managing with style: The effect of managers on firm policies. QuarterlyJournal of Economics 118 (4): 1169–208.

Bonner, S. E., and B. L. Lewis. 1990. Determinants of auditor expertise. Journal of Accounting Research28 (Supplement): 1–20.

17. The second result would change depending on some of the modeling assumptions. What is unlikely to change isthat the difference in effort provision in equilibrium between a high- and low-IQ partner changes based on the part-nership size.



Bowler, B. 2017. Are going concern opinions associated with lower audit impact? Working paper, Univer-sity of Florida.

Cameran, M., D. Campa, and J. R. Francis. 2017. How important is partner variation in explaining auditquality? Working paper, European Accounting Association.

Caramanis, C., and C. Lennox. 2008. Audit effort and earnings management. Journal of Accounting andEconomics 45 (1): 116–38.

DeAngelo, L. 1981. Auditor size and audit quality. Journal of Accounting and Economics 3 (3): 183–99.DeFond, M. L., and J. Zhang. 2014. A review of archival auditing research. Journal of Accounting and Eco-

nomics 58 (2–3): 275–326.Dye, R. 1993. Auditing standards, legal liability, and auditor wealth. Journal of Political Economy 101 (5):

887–914.Francis, J. R. 2011. A framework for understanding and researching audit quality. Auditing: A Journal of

Practice & Theory 30 (2): 125–52.Gonzalez, A. 2017. The state of accounting recruitment and talent shortages in 2017. Going Concern. http://

goingconcern.com/accounting-recruitment-talent-shortage/, retrieved July 30, 2018.Grinblatt, M., M. Keloharju, and J. T. Linnainmaa. 2011. IQ and stock market participation. Journal of

Finance 66 (6): 2121–64.Grinblatt, M., M. Keloharju, and J. T. Linnainmaa. 2012. IQ, trading behavior, and performance. Journal of

Financial Economics 104 (2): 339–62.Gul, F. A., D. Wu, and Z. Yang. 2013. Do individual auditors affect audit quality? Evidence from archival

data. The Accounting Review 88 (6): 1993–2023.Holmstrom, B. 1982. Moral hazard in teams. Bell Journal of Economics 13 (2): 324–40.Huddart, S. 2013. Discussion of empirical evidence on the implicit determinants of compensation in Big

4 audit partnerships. Journal of Accounting Research 51 (2): 389–97.Huddart, S., and P. J. Liang. 2003. Accounting in partnerships. American Economic Review Papers and Pro-

ceedings 93 (2): 410–14.Huddart, S., and P. J. Liang. 2005. Profit sharing and monitoring in partnerships. Journal of Accounting and

Economics 40 (1–3): 153–87.Hunter, J. E. 1986. Cognitive ability, cognitive aptitudes, job knowledge, and job performance. Journal of

Vocational Behavior 29 (3): 340–62.Kallunki, J., J.-P. Kallunki, L. Niemi, and H. Nilsson. 2019. IQ and audit quality: Do smarter auditors

deliver better audits? Contemporary Accounting Research, this issue.Kinney, W. R. 2015. Discussion of “Does the identity of engagement partner matter? An analysis of audit

partner reporting decisions.” Contemporary Accounting Research 32 (4): 1479–88.Knechel, W. R., L. Niemi, and M. Zerni. 2013. Empirical evidence on the implicit determinants of compen-

sation in Big 4 audit partnerships. Journal of Accounting Research 51 (2): 349–87.Knechel, W. R., A. Vanstraelen, and M. Zerni. 2015. Does the identity of engagement partners matter?

An analysis of audit partner reporting decisions. Contemporary Accounting Research 32 (4):1443–78.

Lennox, C., and X. Wu. 2018. A review of the archival literature on audit partners. Accounting Horizons32 (2): 1–35.

Libby, R., and H.-T. Tan. 1994. Modeling the determinants of audit expertise. Accounting, Organizationsand Society 19 (8): 701–16.

Lindqvist, E., and R. Vestman. 2011. The labor market returns to cognitive and noncognitive ability: Evi-dence from the Swedish enlistment. American Economic Journal: Applied Economics 3 (1): 101–28.

Liu, X., and D. A. Simunic. 2005. Profit sharing in an auditing oligopoly. The Accounting Review 80 (2):677–702.

Lobo, G. J., and Y. Zhao. 2013. Relation between audit effort and financial report misstatements: Evidencefrom quarterly and annual restatements. The Accounting Review 88 (4): 1385–412.

Minnis, M., and N. Shroff. 2017. Why regulate private firm disclosure and auditing? Accounting and Busi-ness Research 47 (5): 1–30.



http://goingconcern.com/accounting-recruitment-talent-shortage/

http://goingconcern.com/accounting-recruitment-talent-shortage/

Monga, V. 2017. As regulations change, companies grapple with accountant shortage.Wall Street Journal,March 6,https://www.wsj.com/articles/as-regulations-change-companies-grapple-with-accountant-shortage-1488812401,retrieved July 30, 2018.

PCAOB. 2017. Order instituting disciplinary proceedings, making findings, and imposing sanctions in thematter of Grant Thornton LLP, respondent. PCAOB Release No. 105-2017-054, December 19.

Reynolds, J., and J. Francis. 2000. Does size matter? The influence of large clients on office-level auditorreporting decisions. Journal of Accounting and Economics 30 (3): 375–400.

Richardson, K., and S. H. Norgate. 2015. Does IQ really predict job performance? Applied DevelopmentalScience 19 (3): 153–69.

Woollett, K., and E. A. Maguire. 2011. Acquiring “the knowledge” of London’s layout drives structuralbrain changes. Current Biology 21 (24): 2109–14.



https://www.wsj.com/articles/as-regulations-change-companies-grapple-with-accountant-shortage-1488812401

iq and audit quality: do smarter auditors deliver better

Documents